Research ArticleDistributed Fault Detection for Wireless Sensor Networks Basedon Support Vector Regression

Yong Cheng1 Qiuyue Liu2 Jun Wang2 Shaohua Wan 3 and Tariq Umer 4

1 Jiangsu Key Laboratory of AgriculturalMeteorology NanjingUniversity of Information Science amp Technology Nanjing 210044 China2Department of Computer amp Software Nanjing University of Information Science amp Technology Nanjing 210044 China3School of Information and Safety Engineering Zhongnan University of Economics and Law Wuhan 430073 China4COMSATS University Islamabad Wah Campus Pakistan

Correspondence should be addressed to ShaohuaWan shaohuawanieeeorg

Received 27 June 2018 Revised 25 September 2018 Accepted 30 September 2018 Published 21 October 2018

Guest Editor Huaming Wu

Copyright copy 2018 Yong Cheng et al This is an open access article distributed under the Creative Commons Attribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

Because the existing approaches for diagnosing sensor networks lead to low precision and high complexity a new fault detectionmechanism based on support vector regression and neighbor coordination is proposed in this work According to the redundantinformation about meteorological elements collected by a multisensor a fault prediction model is built using a support vectorregression algorithm and it achieves residual sequences Then the node status is identified by mutual testing among reliableneighbor nodes Simulations show that when the sensor fault probability in wireless sensor networks is 40 the detection accuracyof the proposed algorithm is over 87 and the false alarm ratio is below 7 The detection accuracy is increased by up to 13 incontrast to other algorithms This algorithm not only reduces the communication to sensor nodes but also has a high detectionaccuracy and a low false alarm ratio The proposed algorithm is suitable for fault detection in meteorological sensor networks withlow node densities and high failure ratios

1 Introduction

Wireless sensor networks (WSNs) consist of a large numberof sensor nodes which are small and low-costWSNs are usedto sense physical conditions collect and process informationabout the objects in the coverage area and send informationto the observer for further processing and analysis [1ndash4] Todate WSNs have been widely used in many critical fieldssuch as environmental surveillance emergency navigationtraffic monitoring and industrial control [5 6] Becauseof the limited computing capacity and energy of wirelessmeteorological sensor network nodes as well as the complexcharacteristics and real-time data of meteorological informa-tion unexpected faults appear in nodes after long run times[7] Therefore detecting sensor nodes with faulty readingswhich can greatly improve the performance of the wirelesssensor network is necessary

Rajasegarar et al proposed a distributed nonparamet-ric anomaly detection algorithm that identifies anomalous

measurements at nodes based on data clustering [8] Theyuse a hyperspherical clustering algorithm and the k-nearestneighbor scheme to collaboratively detect anomalies in wire-less sensor network data A localized fault identificationalgorithm in wireless sensor networks is studied by Ding etal [9] this is a distributed fault detection algorithm whereeach sensor node compares its own sensed data with themedian of neighboring data to identify its own fault statusThe performance of localized diagnosis algorithm is limiteddue to the uneven nature of the nodes in wireless sensornetworks Daniel et al proposed a classification based votingmethod for anomaly detection [10] that uses five differentclassifiers to detect anomalies with reliable estimations toreplace the affected measurements This method fails in caseswhere a large dataset is considered

The literature also indicates that the fault recognition rateof current fault detection algorithms inWSNs decreases withhigh failure rates Many fault detection methods exploit thedifference between the readings of neighbor nodes because a

HindawiWireless Communications and Mobile ComputingVolume 2018 Article ID 4349795 8 pageshttpsdoiorg10115520184349795

2 Wireless Communications and Mobile Computing

sensor nodersquos readings are similar to a neighboring nodersquos datain many applications However most readings cause excessoverhead This paper proposes a new approach distributedfault detection for wireless sensor networks based on supportvector regression which can improve the accuracy of thefault diagnosis algorithm It uses weather data from sensorsin WSNs as the data set and is combined with the neighborcollaboration method to createset up a new support vectormachine model forecast system The proposed algorithmcan save the overhead due to the frequent interactionsbetween nodes It makes better use of the characteristics of ameteorological sensor network to improve the fault diagnosisaccuracy which is more suitable for sparseWSNs even whenthe failure rate is very high The contributions of this paperare summarized as follows

(i) Aiming to solve the problems of the low accuracy andhigh complexity of the existing fault diagnosis meth-ods used for wireless meteorological sensor networksa fault diagnosis method based on support vectorregression and neighbor cooperation is proposed

(ii) The proposed model uses the residuals generated bythe redundant spatiotemporal meteorological valuesbetween different sensor nodes and neighbor coop-eration methods to improve the accuracy of faultdiagnosis algorithms and save the overhead generateddue to the frequent interactions between nodes

(iii) The experimental results show that the accuracy ofthe proposed algorithm is greater than 87 and thatthe accuracy of fault detection is improved by nearly13The proposed algorithm has high fault detectionaccuracy and a low false alarm rate and it reduces thenode traffic It is more applicable for wireless weathersensor networks with a sparse node distribution andhigh failure rate

The remainder of this article is organized as followsSection 2 presents related works on fault detection forWSNsSection 3 introduces distributed fault detection algorithmsbased on support vector regression andmultisensor coopera-tion Section 4 presents the experimental settings and evalu-ation metrics and theoretically analyzes the performance ofthe presented methods Finally this article is concluded inSection 5

2 Related Work

The fault detection of wireless sensor networks can bedivided into centralized detection and distributed detectionaccording to the tasks In general for a low informationflow centralized fault detection technology is simple toimplement and easy to use and it can effectively locatethe failure node this technology is suitable for small-scale meteorological sensor networks However its short-comings are even more obvious its main problems arethe central node bottleneck near-central node heat issuesdelays wireless channel congestion and poor system scal-ability Therefore for meteorological sensing networks withlarge networks and limited available resources to reduce

the system energy consumption and ensure the systemreliability the distributed detection method is generallyconsidered

The distributed fault detection algorithm mainly adoptsthe idea of local decision-making The node compares thelocal collected weather information and the informationcollected from neighboring nodes the algorithm finallydetermines whether have a fault The distributed fault detec-tion method does not need to send all of the information tothe central node Instead each node in the sensor networkcompletes the fault detection task in the network eitherindependently or partially In the distributed fault detectionalgorithm the node can performmore decisions locally thusfurther reducing the amount of traffic generated by the datasent to the central node This also balances the amount ofmessage interaction in the network and reduces both theenergy consumption and network congestion furthermorethe life of the entire network is extended

Distributed fault detection is applicable to most meteoro-logical sensor networks and is the future trend of fault detec-tion [11ndash16] Reference [17] proposed a typical distributedfault detection (DFD) algorithm The DFD algorithm com-putes the similarity of the data sensed by neighboring nodesconcurrently to determine the initial state of the nodeThe state and neighbor nodes test each other to determinewhether a node is faulty and spreads the diagnosed resultto its neighboring nodes but the DFD method must causethe node to communicate with neighboring node three timesand then determine the state of the node resulting in alarge amount of data communication therefore the DFDalgorithm must consume large amounts of energy

In recent years many scholars have proposed a numberof improved distributed fault detection methods based on theDFD algorithm [18ndash27] References [28ndash32] show that whenthe number of neighboring network nodes is small and theprobability of node failure is large the DFD algorithm per-formance will decrease sharply because the DFD algorithmjudges too harshly if the node is normal condition Therefore[32] improved theDFDalgorithmandmodified the judgmentconditions of the final state of DFD algorithm References[33 34] proposed a distribution adaptive sensor network faultdetection mechanism based on the DFD algorithm Theyfurther introduced the concept of the reliability level basedon the credibility of the neighboring node which selects thetrusted node and compares the data from the local nodeto determine the node status References [35 36] proposeda fault detection mechanism in which historic data andneighbors cooperate and fuse First a root node is selectedIf the root node is normal it is used as a reference node whenthe network node status is judged if the root node is a faultthe root node is reselected from the remaining nodes and thestate is determined until the reference node is found Thenthe reference node is used to make a decision on the state ofthe neighbor node If the neighboring node is normal it isused as a reference nodeThe neighbors are then determinedif the neighbor fails it cannot be used as a reference node Tocomplete the determination of the status of all network nodesthe fault detection algorithm is used again on the nodes thatstill have an unknown status

Wireless Communications and Mobile Computing 3

In addition to the fault detection method based on thevoting mechanism of neighbor nodes Min D proposed amedian-based fault detection algorithm that used the datacollected by neighboring nodes to sort the data sequence andobtains intermediate data from the sequence The obtainedvalue compares the value with the data collected by thenode If the difference exceeds the threshold the nodefails Reference [30] proposed a method for distributingthe detection tasks for the typical clustering structure of asensor network The fault detection method in each unitcluster uses the periodic data exchange between cluster headsand neighboring cluster head nodes until cluster head faultdetection is completedWithin a cluster the cluster head nodeis responsible for periodically broadcasting detecting andlocating faulty nodes

Currently most fault detection technologies used inmeteorological sensor networks are based on the distributedmethodThe distributed fault detection algorithm distributesthe computational cost over all of the nodes in the networkand the decentralized features make it self-organized so thecharacteristics of improving resource efficiency and facili-tating implementation are very suitable for the applicationrequirements of a meteorological sensor network Howeveron the one hand the distributed method based on the votingstrategy often has a large amount of additional communi-cation overhead on the other hand in a network with alow node distribution density and a high failure rate theperformance of the traditional distributed methods is greatlyreduced In a large-scale meteorological sensor networkthe traditional distributed fault detection method has theproblem that its detection performance decreases sharply andits energy consumption increases

The SVM classifier has gained popularity due to itsoptimum solution and its simple numerical comparison fordata classification Several SVM-based approaches have beenproposed [37ndash40] for anomaly detection in WSNs Based onthe above analysis the fault detection rate of the traditionalfault detection algorithm decreases rapidly when the faultdetection rate is high When implementing the traditionaldistributed algorithm the high performance of the detectionis achieved by using multiple methods of communicationbetween neighbor nodes With a large amount of extraoverhead a distributed node fault detection algorithm basedon support vector machine regression prediction modelis proposed Through the meteorological sensors carriedon the nodes of the wireless weather sensor network themeteorological elements are collected to construct a supportvector machine regression algorithm prediction model andresiduals are generated using redundant information con-cerning the time and space of meteorological element valuesbetween different sensors in a node In combination withneighbor cooperation methods the accuracy of the faultdetection algorithm is improved and the overhead generatedby the frequent interaction between nodes is saved Thusthe characteristics of meteorological sensor networks arebetter used to improve the fault detection accuracy thusmaking the algorithm more suitable for wireless weathersensor networks with sparse nodes and high sensor failurerates

3 Distributed Fault Detection Based onSupport Vector Regression

31 Support Vector Regression SVR (Support Vector Regres-sion) was originally introduced under linear and divisibleconditions and was developed as an effectual way to solveprediction problems [41] Consider a set of training data(119909119894 119910119894) (119894 = 1 2 119899) which are historical perceptiondata from sensors where n is the total number of datapoints The original inputs are first mapped into a high-dimensional feature space by nonlinear mapping 120601 and thelinear regression function is produced

119910 = 119908 sdot 120601 (119909) + 119887 (1)

where the dimension w is the dimension of the feature spaceThe resolutions of w and b are transformed into a convexityquadratic programming problem

min119908119887120577

12 1199082 + 119862

119899

sum119894=1

(120577119894 + 120577lowast119894 )

119910119894 minus 119908 sdot 120601 (119909119894) minus 119887 le 120576 + 120577119894

119904119905 119908 sdot 120601 (119909119894) + 119887 minus 119910119894 le 120576 + 120577119894

120577119894 120577lowast119894 ge 0 119894 = 1 2 119899

(2)

SVR can solve small sample problems and has good gen-eralization ability when using the principles of structuralrisk minimization The constant Cgt0 is a punishment coef-ficient 120577119894 and 120577lowast119894 are the slack variables Meanwhile theLagrangian multipliers 120572119894 and 120572lowast119894 are introduced to analyzea quadratic programming (QP) problem with linear stateinequality constraints Then the above optimization problemis transformed into its dual form

max 119871 (119908 119887)

= minus12

119899

sum119894=1

119899

sum119895=1

(120572119894 minus 120572lowast119894 ) (120572119895 minus 120572

lowast119895 )119870 (119909119894 119909119895)

minus 120576119899

sum119894=1

(120572119894 + 120572lowast119894 ) +

119899

sum119894=1

119910119894 (120572119894 minus 120572lowast119894 )

119904119905119899

sum119894=1

119910119894 (120572119894 minus 120572lowast119894 ) = 0

0 le 120572119894 120572lowast119894 le 119862

(3)

where 119870(119909119894 119909119895) is the kernel function [42] By using thekernel function idea this theory can change a problem innonlinearity space into one in linearity space to reduce thealgorithm complexity The regressive function is denoted asfollows

119891 (119909) =119899

sum119909119894isin119878119881

(120572119894 minus 120572lowast119894 )119870 (119909119894 119909) + 119887 (4)

where SV is the support vector set

4 Wireless Communications and Mobile Computing

32 Multisensor Cooperation A node in a WSN which hasmultiple sensors collects different types of environmentalinformation such as temperature humidity light and carbondioxide concentration Some of the types of environmentalinformation from the same node have close correlations Inaddition to the node collection of meteorological elementsvoltage is an important parameter in fault diagnosisThenodevoltage has an obvious effect on the temperature and othermeteorological elements Therefore the proposed algorithmproduces multiple estimates by setting up a multiple SVRforecast modelThese estimates are compared with the resid-ual sequences which come from meteorological elementsFor temperature humidity and voltage we build two SVRprediction models 1198781198811198771 and 1198781198811198772 with three types of datawhich have a redundant relationship in time and space 1198781198811198771builds the SVR prediction model with a sample consistingof temperature and humidity The current moment for k isset and the sample consisting of past continuous data is asfollows

119909119896 = (119879119896minus1 119879119896minus2 119879119896minus119897 119867119896 119867119896minus1 119867119896minus119897+1) (5)

where 119879119896minus1 119879119896minus2 119879119896minus119897 are the temperature from momentk-1 to k-l and119867119896 119867119896minus1 119867119896minus119897+1 are denoted as the humid-ity from moment k to k-l+1 The temperature values frommoment k-2 to k-l and the humidity values from k-1 to k-l+1 are used as the input samples the temperature value atk-1 is used as the output sample The estimated value 119879119896 from1198781198811198771 is a function of temperature and humidity at the pastmoment

119879119896 = 119891 (119879119896minus1 119879119896minus2 119879119896minus119897+1 119867119896 119867119896minus1 119867119896minus119897+2) (6)

1198781198811198772 builds the SVR prediction model with a sample thatconsists of the temperature and voltage The current momentfor k is set and the sample consists of past continuous data asfollows

119909119896lowast = (119879119896minus1 119879119896minus2 119879119896minus119897 119881119896 119881119896minus1 119881119896minus119897+1) (7)

where 119881119896 119881119896minus1 119881119896minus119897+1 are denoted as the voltage frommoment k to k-l+1 The estimated value 119879119896

1015840from 1198781198811198772 is a

function of temperature and voltage at the past moment

1198791198961015840 = 119891 (119879119896minus1 119879119896minus2 119879119896minus119897+1 119881119896 119881119896minus1 119881119896minus119897+2) (8)

We produce two estimated values of the temperature 119879119896 and1198791198961015840atmoment k after training the SVR forecasting model and

compare the results with the real data from the temperaturesensor to produce the residual sequence

1198641 = 119879119896 minus 1198791198961015840

1198642 = 119879119896 minus 119879119896

1198643 = 1198791198961015840 minus 119879119896

(9)

The prediction model can accurately output the tempera-ture values 119879119896 and 119879119896

1015840 according the effective history data

collected by the sensors If the data from the temperaturesensors cause an exception at moment k the residual 1198641 and1198642 are above the threshold value 1205791 and 1198643 basically showsno change We update the predictions with the actual data ifthe residual of the parameters from the nodes is less than thethreshold parameters The forecast model prepares the nextprediction

33 Distributed FaultDiagnosis Algorithm InWSNs becauseof the time-space continuum of the environmental factorsand the densely deployed nodes a node has the characteristicof spatial similarity with its nearby nodes that is the nearbynodes are likely to have similar measurements Nodes in themonitoring area will affect the accuracy of the fault diagnosisalgorithm when an event occurs As a result the algorithmgiven in this paper is combined with the fault diagnosisand neighbor coordination methods and introduces thecredibility evaluation mechanism The credibility of nodescomes not only from their own sensory information butalso from the judgment of the sensory information by theneighbor nodes which effectively eliminates the influence ofthe fault node

We consider thatN sensor nodes are randomly placed in aunit square field Without loss of generality we suppose thatthe location of each node is known and that all nodes havethe same communication radius R The average number ofnodes within a transmission range is the node density usedto illustrate the number of neighboring changes Each nodecan communicate with other nodes by one-hop ormultihopsNodes are assumed to be neighboring nodes if they are in eachotherrsquos coverage radius Each node periodically broadcasts itsmeasurements or decision such as temperature humidity airpressure and wind speed to all its neighbors The WSNs aremodelled as the system graph G(VE) where V represents aset of nodes in WSNs and E represents a set of logical linksbetween nodes Let 119889119894119904119905(119904119894 119904119895) denote the distance betweennode 119904119894 and node 119904119895 then E is based on the followingdefinition

119864 = (119904119894 119904119895) isin 1198812 | 119889119894119904119905 (119904119894 119904119895) le 119877 (10)

Thenodeswithin the transmission range of node 119904119894 belongto the neighborhood 119873(119904119894) 119873(119904119894) sub 119873 We apply faultdetection to node 119904119894 at each t timestamp The measurementof node 119904119894 at t time is denoted as 119909119894119905 Assume that theneighborhood 119873(119904119894) of node (119904119894) consists of Negi sensornodes that is 119904119895 isin 119873(119904119894) 119895 = 1 119873119890119892119894

First we use the trust level 120582119895 for the node 119904119895 isin 119873(119904119894)Each node has the same trust level at first and we set120582119895 = 120582119898119886119909 We adjust the trust level of nodes within theneighborhood by using the fault diagnosis mechanism basedon support vector machine regression If the meteorologicalelements are above the threshold value then 120582119895 = 120582119895 minus 1

Otherwise we transmit the predicted value 119909119895119896+1 and the trustlevel 120582119895 to node 119878119894 We denote the weight value for the fault

Wireless Communications and Mobile Computing 5

detection caused by the sensor data of the neighbor node 119904119895as 120596119895

120596119895 =120582119895119878119880119872

(11)

where SUM is the sum of the trust levels of all nodes withinthe neighborhood

For node 119878119894 we use the fault diagnosis mechanism basedon support vector machine regression to get the predictiondata of the next moment We calculate the failure levelindicator by the space-time correlation between nodes inwireless sensor networks

119891119894 =10038171003817100381710038171003817100381710038171003817100381710038171003817119894119896+1 minus

119873119890119892119894

sum119895=1

120596119895 sdot 119895

119896+1

10038171003817100381710038171003817100381710038171003817100381710038171003817(12)

If 119891119894 gt 1205792 the trust level 120582119894 = 0 and node 119904119894 fails Otherwisethe predicted value is updated by the actual value and theforecast model to prepare the next prediction

4 Simulation

We experiment in MATLAB to assess the performance of theproposed approachTheWSNs contain 200 nodes in a squareregion of 30 times 30 units Each sensor is randomly placed in aunit grid The measurements of the nodes in the normal areaare subject to a Gaussian distribution The data from Intellab are used in the experiment [43] including temperaturehumidity and voltage as the experimental data We use thetile radial primary kernel function and set 120582119898119886119909 = 10 1205791 =05 1205792 = 0375 The performance of the proposed DSFD(Distributed SVR Fault Detection) algorithm is evaluated andcompared with the existing DFD algorithm in [17] in termsof the detection accuracy (DA) and false alarm rate (FAR)in the network All experiments are repeated 100 times anddata for analysis are the averaged to ensure the statisticalsignificance of the experiments To assess the effect of faultynode identification two indicators are usually employeddetection accuracy and false alarm rate

41 Detection Accuracy Detection accuracy (DA) refers tothe ratio of the number of correctly identified faulty nodesto the total number of actual fault nodes

119863119860 = |119865||119876|

(13)

where F is the set of fault nodes which the algorithm hasdetected and Q is the set of actual fault nodes

We compare two algorithms in terms of detection accu-racy under different sensor density configurations in Figures1 and 2 respectively When the failure rate is lower than25 the fault detection precisions of the two algorithmsare greater than 91 With an increase in the node failurerate the fault detection precisions of the two algorithmsare decreased but the DSFD algorithm has a higher faultdetection accuracy than does the DFD However we cansee that with a decrease in node density the performance

Det

ectio

n ac

cura

cy

DSFDDFD

01 015 02 025 03 035 04005Sensor fault probability

065

07

075

08

085

09

095

1

Figure 1 Fault sensor detection accuracy when the average degreeis 5

Det

ectio

n ac

cura

cy

DSFDDFD

01 015 02 025 03 035 04005Sensor fault probability

065

07

075

08

085

09

095

1

Figure 2 Fault sensor detection accuracy when the average degreeis 10

of each algorithm improves Taking Figure 1 as an examplewhen the sensor fault probability is higher than 40 thefault detection accuracy of DSFD algorithm is still over 87which is an improvement of 13 over the DFD algorithmThe DFD algorithm first determines the nodersquos initial stateby comparing the data from its neighborhood nodes withitself then the status of the node is determined accordingto the initial state of the node and the adjacent nodes Thismight occur because when the fault rate is high and thenumber of neighbors is large the misdiagnosis rate of DFDis high The DSFD algorithm constructs a support vectormachine regression forecasting model with historical dataand accurately determines the fault node The DSFD addsthe reference objects according to the correlation betweenmultiple sensors on nodes to reduce dependence on theneighbor nodes

6 Wireless Communications and Mobile Computing

False

alar

m ra

te

DSFDDFD

0002004006008

01012014016018

02

01 015 02 025 03 035 04005Sensor fault probability

Figure 3 Fault sensor false alarm rate when the average degree is 5

False

alar

m ra

te

DSFDDFD

0002004006008

01012014016018

02

01 015 02 025 03 035 04005Sensor fault probability

Figure 4 Fault sensor false alarm rate when the average degree is10

411 False Alarm Rate The false alarm rate (FAR) refers tothe ratio of the number of normal nodes that are mistaken asfault nodes to the total number of normal nodes

119865119860119877 = |119865 minus 119876||119873 minus 119876|

(14)

where N is the total number of nodes in the WSNsFigures 3 and 4 show the false alarm rate against the

sensor fault probability for different average number ofneighbors They indicate the performance of each algorithmat densities of 5 and 10 From the two figures we can seethat with an increase in the sensor fault probability the falsealarm rate of each algorithm increases The higher the faultprobability is the higher false alarm rate is As Figure 3 showsthe false alarm rate of DSFD is 147 it is still below 7when the sensor fault probability is 40This occurs becausethe DFD algorithm diagnoses all nodes in the monitoringfield and uses many sampling times by comparing the sensed

data from neighbor nodes Many of the sensor tests of goodsensors are likely faulty so these good sensors are then diag-nosed as faulty sensors However the DSFD algorithm notonly uses the collaborative operation of neighboring peers butalso combines the support vector machine (SVM) regressionalgorithm with the information redundancy between thesensors in the wireless sensor network The proposed DSFDalgorithm avoids the misdiagnosis caused by the number ofneighbor nodes and the incorrect data from neighbor nodesthereby achieving high detection accuracy

5 Conclusion

In this paper we modelled and analyzed a fault diagnosismechanism based on support vector machine regressionamong sensor observations in wireless sensor networksaccording to the redundant information of meteorologicalelements collected by multisensors The fault predictionmodel is built using a support vector regression algorithmto achieve residual sequences The proposed algorithm out-performs previous DFD in terms of faulty sensor detectionaccuracy and false alarm rates The fault detection algorithmachieves high detection accuracy and low false alarm rateswhich are more suitable for sparse WSNs even when thefailure rate is very high

Data Availability

The data used to support the findings of this study areavailable from the corresponding author upon request

Conflicts of Interest

The authors declare that they have no conflicts of interest

Acknowledgments

This work was supported by the National Natural ScienceFoundation of China (61402236 61373064 and 41875184)the CERNET Innovation Project (NGII20160318) and theJiangsu Province ldquoSix Talent Peaks Project in JiangsuProvincerdquo (2015-DZXX-015)

References

[1] XMiao K Liu Y He D Papadias QMa and Y Liu ldquoAgnosticdiagnosis Discovering silent failures in wireless sensor net-worksrdquo IEEE Transactions on Wireless Communications vol 12no 12 pp 6067ndash6075 2013

[2] S Wan Y Zhang and J Chen ldquoOn the construction ofdata aggregation tree with maximizing lifetime in large-scalewireless sensor networksrdquo IEEE Sensors Journal vol 16 no 20pp 7433ndash7440 2016

[3] S Wan and Y Zhang ldquoCoverage hole bypassing in wirelesssensor networksrdquo The Computer Journal vol 60 no 10 pp1536ndash1544 2017

[4] Shaohua Wan ldquoEnergy-efficient adaptive routing and context-aware lifetime maximization in wireless sensor networksrdquo

Wireless Communications and Mobile Computing 7

International Journal of Distributed Sensor Networks vol 2014Article ID 321964 16 pages 2014

[5] J Kong J-H Cui D Wu and M Gerla ldquoBuilding underwaterad-hoc networks and sensor networks for large scale real-timeaquatic applicationsrdquo inProceedings of theMilitary Communica-tions Conference (MILCOM rsquo05) pp 1535ndash1541 October 2005

[6] N Xu S Rangwala K K Chintalapudi et al ldquoA wireless sensornetwork for structural monitoringrdquo in Proceedings of the 2ndInternational Conference on Embedded Networked Sensor Sys-tems (SenSys rsquo04) pp 13ndash24 November 2004 (Catalan)

[7] Z You X ZhaoHWanWNNHung YWang andMGu ldquoAnovel fault diagnosis mechanism for wireless sensor networksrdquoMathematical and Computer Modelling vol 54 no 1-2 pp 330ndash343 2011

[8] S Rajasegarar C Leckie M Palaniswami and J C BezdekldquoDistributed anomaly detection in wireless sensor networksrdquo inProceedings of the 10th IEEE Singapore International Conferenceon Communication systems (ICCS rsquo06) pp 1ndash5 IEEE October2006

[9] M Ding D Chen K Xing and X Cheng ldquoLocalized fault-tolerant event boundary detection in sensor networksrdquo inProceedings of the IEEE 24th Annual Joint Conference of the IEEEComputer and Communications Societies (INFOCOM rsquo05) vol2 pp 902ndash913 March 2005

[10] D I Curiac and C Volosencu ldquoEnsemble based sensinganomaly detection in wireless sensor networksrdquo Expert Systemswith Applications vol 39 no 10 pp 9087ndash9096 2012

[11] M Ur-Rehman N A Malik X Yang Q H Abbasi Z Zhangand N Zhao ldquoA low profile antenna for millimeter-wavebody-centric applicationsrdquo IEEE Transactions on Antennas andPropagation vol 65 no 12 pp 6329ndash6337 2017

[12] C Wang H Lin and H Jiang ldquoCANS Towards congestion-adaptive and small stretch emergency navigation with wirelesssensor networksrdquo IEEE Transactions on Mobile Computing vol15 no 5 pp 1077ndash1089 2016

[13] J Wen B Zhou W H Mow and X-W Chang ldquoAn efficientalgorithm for optimally solving a shortest vector problem incompute-and-forward designrdquo IEEE Transactions on WirelessCommunications vol 15 no 10 pp 6541ndash6555 2016

[14] J Wen J Wang and Q Zhang ldquoNearly optimal bounds fororthogonal least squaresrdquo IEEE Transactions on Signal Process-ing vol 65 no 20 pp 5347ndash5356 2017

[15] J Wen Z Zhou Z Liu M-J Lai and X Tang ldquoSharp sufficientconditions for stable recovery of block sparse signals by blockorthogonal matching pursuitrdquo 2016 httpsarxivorgabs160502894

[16] S Rani S H Ahmed R Talwar and J Malhotra ldquoCan sensorscollect big data an energy-efficient big data gathering algo-rithm for a WSNrdquo IEEE Transactions on Industrial Informaticsvol 13 no 4 pp 1961ndash1968 2017

[17] J Chen S Kher and A Somani ldquoDistributed fault detectionof wireless sensor networksrdquo in Proceedings of the Workshop onDependability Issues in Wireless Ad Hoc Networks and SensorNetworks pp 65ndash72 2006

[18] S Rani S H Ahmed J Malhotra and R Talwar ldquoEnergyefficient chain based routing protocol for underwater wirelesssensor networksrdquo Journal of Network and Computer Applica-tions vol 92 pp 42ndash50 2017

[19] D Li and J Zhang ldquoEfficient implementation to numericallysolve the nonlinear time fractional parabolic problems onunbounded spatial domainrdquo Journal of Computational physicsvol 322 pp 415ndash428 2016

[20] A Munir J Antoon and A Gordon-Ross ldquoModeling andanalysis of fault detection and fault tolerance in wireless sensornetworksrdquoACMTransactions on EmbeddedComputing Systemsvol 14 no 1 article 3 2015

[21] G S Brar S Rani V Chopra R Malhotra H Song and SH Ahmed ldquoEnergy efficient direction-based PDORP routingprotocol for WSNrdquo IEEE Access vol 4 pp 3182ndash3194 2016

[22] ANAlvi SH Bouk SHAhmedMA YaqubM Sarkar andH Song ldquoBEST-MAC Bitmap-Assisted Efficient and ScalableTDMA-Based WSN MAC Protocol for Smart Citiesrdquo IEEEAccess vol 4 pp 312ndash322 2016

[23] S Rani and S H AhmedMulti-Hop Routing in Wireless SensorNetworks An Overview Taxonomy and Research ChallengesSpringer 2015

[24] S Rani R Talwar J Malhotra S H Ahmed M Sarkar and HSong ldquoA novel scheme for an energy efficient internet of thingsbased on wireless sensor networksrdquo Sensors vol 15 no 11 pp28603ndash28626 2015

[25] E Ould-Ahmed-Vall B H Ferri and G F Riley ldquoDistributedfault-tolerance for event detection using heterogeneouswirelesssensor networksrdquo IEEE Transactions on Mobile Computing vol11 no 12 pp 1994ndash2007 2012

[26] S C Chan H C Wu and K M Tsui ldquoRobust recursiveeigendecomposition and subspace-based algorithmswith appli-cation to fault detection in wireless sensor networksrdquo IEEETransactions on Instrumentation and Measurement vol 61 no6 pp 1703ndash1718 2012

[27] R Huang X Qiu and L Rui ldquoSimple random sampling-basedprobe station selection for fault detection in wireless sensornetworksrdquo Sensors vol 11 no 3 pp 3117ndash3134 2011

[28] J Medina-Garcıa T Sanchez-Rodrıguez J Galan A DelgadoF Gomez-Bravo and R Jimenez ldquoA wireless sensor systemfor real-time monitoring and fault detection of motor arraysrdquoSensors vol 17 no 3 p 469 2017

[29] T Muhammed and R A Shaikh ldquoAn analysis of fault detectionstrategies in wireless sensor networksrdquo Journal of Network andComputer Applications vol 78 pp 267ndash287 2017

[30] H Artail A Ajami T Saouma and M Charaf ldquoA faultynode detection scheme for wireless sensor networks that usedata aggregation for transportrdquo Wireless Communications andMobile Computing vol 16 no 14 pp 1956ndash1971 2016

[31] M Panda and P M Khilar ldquoDistributed Byzantine fault detec-tion technique in wireless sensor networks based on hypothesistestingrdquo Computers and Electrical Engineering vol 48 pp 270ndash285 2015

[32] P Jiang ldquoA new method for node fault detection in wirelesssensor networksrdquo Sensors vol 9 no 2 pp 1282ndash1294 2009

[33] K P Sharma and T P Sharma ldquorDFD reactive distributed faultdetection in wireless sensor networksrdquo Wireless Networks vol23 no 4 pp 1145ndash1160 2017

[34] M Bo H Darong and W Shaohua ldquoNTRU implementa-tion of efficient privacy-preserving location-based querying inVANETrdquoWireless Communications and Mobile Computing vol2018 Article ID 7823979 11 pages 2018

[35] Y Yang Z Gao H Zhou and X Qiu ldquoAn uncertainty-based distributed fault detectionmechanism for wireless sensornetworksrdquo Sensors vol 14 no 5 pp 7655ndash7683 2014

[36] D Wang S Wan and N Guizani ldquoContext-based probabilityneural network classifiers realized by genetic optimization formedical decision makingrdquo Multimedia Tools and Applicationsvol 77 no 17 pp 21995ndash22006 2018

Wireless Communications and Mobile Computing 3

In addition to the fault detection method based on thevoting mechanism of neighbor nodes Min D proposed amedian-based fault detection algorithm that used the datacollected by neighboring nodes to sort the data sequence andobtains intermediate data from the sequence The obtainedvalue compares the value with the data collected by thenode If the difference exceeds the threshold the nodefails Reference [30] proposed a method for distributingthe detection tasks for the typical clustering structure of asensor network The fault detection method in each unitcluster uses the periodic data exchange between cluster headsand neighboring cluster head nodes until cluster head faultdetection is completedWithin a cluster the cluster head nodeis responsible for periodically broadcasting detecting andlocating faulty nodes

Currently most fault detection technologies used inmeteorological sensor networks are based on the distributedmethodThe distributed fault detection algorithm distributesthe computational cost over all of the nodes in the networkand the decentralized features make it self-organized so thecharacteristics of improving resource efficiency and facili-tating implementation are very suitable for the applicationrequirements of a meteorological sensor network Howeveron the one hand the distributed method based on the votingstrategy often has a large amount of additional communi-cation overhead on the other hand in a network with alow node distribution density and a high failure rate theperformance of the traditional distributed methods is greatlyreduced In a large-scale meteorological sensor networkthe traditional distributed fault detection method has theproblem that its detection performance decreases sharply andits energy consumption increases

The SVM classifier has gained popularity due to itsoptimum solution and its simple numerical comparison fordata classification Several SVM-based approaches have beenproposed [37ndash40] for anomaly detection in WSNs Based onthe above analysis the fault detection rate of the traditionalfault detection algorithm decreases rapidly when the faultdetection rate is high When implementing the traditionaldistributed algorithm the high performance of the detectionis achieved by using multiple methods of communicationbetween neighbor nodes With a large amount of extraoverhead a distributed node fault detection algorithm basedon support vector machine regression prediction modelis proposed Through the meteorological sensors carriedon the nodes of the wireless weather sensor network themeteorological elements are collected to construct a supportvector machine regression algorithm prediction model andresiduals are generated using redundant information con-cerning the time and space of meteorological element valuesbetween different sensors in a node In combination withneighbor cooperation methods the accuracy of the faultdetection algorithm is improved and the overhead generatedby the frequent interaction between nodes is saved Thusthe characteristics of meteorological sensor networks arebetter used to improve the fault detection accuracy thusmaking the algorithm more suitable for wireless weathersensor networks with sparse nodes and high sensor failurerates

3 Distributed Fault Detection Based onSupport Vector Regression

31 Support Vector Regression SVR (Support Vector Regres-sion) was originally introduced under linear and divisibleconditions and was developed as an effectual way to solveprediction problems [41] Consider a set of training data(119909119894 119910119894) (119894 = 1 2 119899) which are historical perceptiondata from sensors where n is the total number of datapoints The original inputs are first mapped into a high-dimensional feature space by nonlinear mapping 120601 and thelinear regression function is produced

119910 = 119908 sdot 120601 (119909) + 119887 (1)

where the dimension w is the dimension of the feature spaceThe resolutions of w and b are transformed into a convexityquadratic programming problem

min119908119887120577

12 1199082 + 119862

119899

sum119894=1

(120577119894 + 120577lowast119894 )

119910119894 minus 119908 sdot 120601 (119909119894) minus 119887 le 120576 + 120577119894

119904119905 119908 sdot 120601 (119909119894) + 119887 minus 119910119894 le 120576 + 120577119894

120577119894 120577lowast119894 ge 0 119894 = 1 2 119899

(2)

SVR can solve small sample problems and has good gen-eralization ability when using the principles of structuralrisk minimization The constant Cgt0 is a punishment coef-ficient 120577119894 and 120577lowast119894 are the slack variables Meanwhile theLagrangian multipliers 120572119894 and 120572lowast119894 are introduced to analyzea quadratic programming (QP) problem with linear stateinequality constraints Then the above optimization problemis transformed into its dual form

max 119871 (119908 119887)

= minus12

119899

sum119894=1

119899

sum119895=1

(120572119894 minus 120572lowast119894 ) (120572119895 minus 120572

lowast119895 )119870 (119909119894 119909119895)

minus 120576119899

sum119894=1

(120572119894 + 120572lowast119894 ) +

119899

sum119894=1

119910119894 (120572119894 minus 120572lowast119894 )

119904119905119899

sum119894=1

119910119894 (120572119894 minus 120572lowast119894 ) = 0

0 le 120572119894 120572lowast119894 le 119862

(3)

where 119870(119909119894 119909119895) is the kernel function [42] By using thekernel function idea this theory can change a problem innonlinearity space into one in linearity space to reduce thealgorithm complexity The regressive function is denoted asfollows

119891 (119909) =119899

sum119909119894isin119878119881

(120572119894 minus 120572lowast119894 )119870 (119909119894 119909) + 119887 (4)

where SV is the support vector set

4 Wireless Communications and Mobile Computing

32 Multisensor Cooperation A node in a WSN which hasmultiple sensors collects different types of environmentalinformation such as temperature humidity light and carbondioxide concentration Some of the types of environmentalinformation from the same node have close correlations Inaddition to the node collection of meteorological elementsvoltage is an important parameter in fault diagnosisThenodevoltage has an obvious effect on the temperature and othermeteorological elements Therefore the proposed algorithmproduces multiple estimates by setting up a multiple SVRforecast modelThese estimates are compared with the resid-ual sequences which come from meteorological elementsFor temperature humidity and voltage we build two SVRprediction models 1198781198811198771 and 1198781198811198772 with three types of datawhich have a redundant relationship in time and space 1198781198811198771builds the SVR prediction model with a sample consistingof temperature and humidity The current moment for k isset and the sample consisting of past continuous data is asfollows

119909119896 = (119879119896minus1 119879119896minus2 119879119896minus119897 119867119896 119867119896minus1 119867119896minus119897+1) (5)

where 119879119896minus1 119879119896minus2 119879119896minus119897 are the temperature from momentk-1 to k-l and119867119896 119867119896minus1 119867119896minus119897+1 are denoted as the humid-ity from moment k to k-l+1 The temperature values frommoment k-2 to k-l and the humidity values from k-1 to k-l+1 are used as the input samples the temperature value atk-1 is used as the output sample The estimated value 119879119896 from1198781198811198771 is a function of temperature and humidity at the pastmoment

119879119896 = 119891 (119879119896minus1 119879119896minus2 119879119896minus119897+1 119867119896 119867119896minus1 119867119896minus119897+2) (6)

1198781198811198772 builds the SVR prediction model with a sample thatconsists of the temperature and voltage The current momentfor k is set and the sample consists of past continuous data asfollows

119909119896lowast = (119879119896minus1 119879119896minus2 119879119896minus119897 119881119896 119881119896minus1 119881119896minus119897+1) (7)

where 119881119896 119881119896minus1 119881119896minus119897+1 are denoted as the voltage frommoment k to k-l+1 The estimated value 119879119896

1015840from 1198781198811198772 is a

function of temperature and voltage at the past moment

1198791198961015840 = 119891 (119879119896minus1 119879119896minus2 119879119896minus119897+1 119881119896 119881119896minus1 119881119896minus119897+2) (8)

We produce two estimated values of the temperature 119879119896 and1198791198961015840atmoment k after training the SVR forecasting model and

compare the results with the real data from the temperaturesensor to produce the residual sequence

1198641 = 119879119896 minus 1198791198961015840

1198642 = 119879119896 minus 119879119896

1198643 = 1198791198961015840 minus 119879119896

(9)

The prediction model can accurately output the tempera-ture values 119879119896 and 119879119896

1015840 according the effective history data

collected by the sensors If the data from the temperaturesensors cause an exception at moment k the residual 1198641 and1198642 are above the threshold value 1205791 and 1198643 basically showsno change We update the predictions with the actual data ifthe residual of the parameters from the nodes is less than thethreshold parameters The forecast model prepares the nextprediction

33 Distributed FaultDiagnosis Algorithm InWSNs becauseof the time-space continuum of the environmental factorsand the densely deployed nodes a node has the characteristicof spatial similarity with its nearby nodes that is the nearbynodes are likely to have similar measurements Nodes in themonitoring area will affect the accuracy of the fault diagnosisalgorithm when an event occurs As a result the algorithmgiven in this paper is combined with the fault diagnosisand neighbor coordination methods and introduces thecredibility evaluation mechanism The credibility of nodescomes not only from their own sensory information butalso from the judgment of the sensory information by theneighbor nodes which effectively eliminates the influence ofthe fault node

We consider thatN sensor nodes are randomly placed in aunit square field Without loss of generality we suppose thatthe location of each node is known and that all nodes havethe same communication radius R The average number ofnodes within a transmission range is the node density usedto illustrate the number of neighboring changes Each nodecan communicate with other nodes by one-hop ormultihopsNodes are assumed to be neighboring nodes if they are in eachotherrsquos coverage radius Each node periodically broadcasts itsmeasurements or decision such as temperature humidity airpressure and wind speed to all its neighbors The WSNs aremodelled as the system graph G(VE) where V represents aset of nodes in WSNs and E represents a set of logical linksbetween nodes Let 119889119894119904119905(119904119894 119904119895) denote the distance betweennode 119904119894 and node 119904119895 then E is based on the followingdefinition

119864 = (119904119894 119904119895) isin 1198812 | 119889119894119904119905 (119904119894 119904119895) le 119877 (10)

Thenodeswithin the transmission range of node 119904119894 belongto the neighborhood 119873(119904119894) 119873(119904119894) sub 119873 We apply faultdetection to node 119904119894 at each t timestamp The measurementof node 119904119894 at t time is denoted as 119909119894119905 Assume that theneighborhood 119873(119904119894) of node (119904119894) consists of Negi sensornodes that is 119904119895 isin 119873(119904119894) 119895 = 1 119873119890119892119894

First we use the trust level 120582119895 for the node 119904119895 isin 119873(119904119894)Each node has the same trust level at first and we set120582119895 = 120582119898119886119909 We adjust the trust level of nodes within theneighborhood by using the fault diagnosis mechanism basedon support vector machine regression If the meteorologicalelements are above the threshold value then 120582119895 = 120582119895 minus 1

Otherwise we transmit the predicted value 119909119895119896+1 and the trustlevel 120582119895 to node 119878119894 We denote the weight value for the fault

Wireless Communications and Mobile Computing 5

detection caused by the sensor data of the neighbor node 119904119895as 120596119895

120596119895 =120582119895119878119880119872

(11)

where SUM is the sum of the trust levels of all nodes withinthe neighborhood

For node 119878119894 we use the fault diagnosis mechanism basedon support vector machine regression to get the predictiondata of the next moment We calculate the failure levelindicator by the space-time correlation between nodes inwireless sensor networks

119891119894 =10038171003817100381710038171003817100381710038171003817100381710038171003817119894119896+1 minus

119873119890119892119894

sum119895=1

120596119895 sdot 119895

119896+1

10038171003817100381710038171003817100381710038171003817100381710038171003817(12)

If 119891119894 gt 1205792 the trust level 120582119894 = 0 and node 119904119894 fails Otherwisethe predicted value is updated by the actual value and theforecast model to prepare the next prediction

4 Simulation

We experiment in MATLAB to assess the performance of theproposed approachTheWSNs contain 200 nodes in a squareregion of 30 times 30 units Each sensor is randomly placed in aunit grid The measurements of the nodes in the normal areaare subject to a Gaussian distribution The data from Intellab are used in the experiment [43] including temperaturehumidity and voltage as the experimental data We use thetile radial primary kernel function and set 120582119898119886119909 = 10 1205791 =05 1205792 = 0375 The performance of the proposed DSFD(Distributed SVR Fault Detection) algorithm is evaluated andcompared with the existing DFD algorithm in [17] in termsof the detection accuracy (DA) and false alarm rate (FAR)in the network All experiments are repeated 100 times anddata for analysis are the averaged to ensure the statisticalsignificance of the experiments To assess the effect of faultynode identification two indicators are usually employeddetection accuracy and false alarm rate

41 Detection Accuracy Detection accuracy (DA) refers tothe ratio of the number of correctly identified faulty nodesto the total number of actual fault nodes

119863119860 = |119865||119876|

(13)

where F is the set of fault nodes which the algorithm hasdetected and Q is the set of actual fault nodes

We compare two algorithms in terms of detection accu-racy under different sensor density configurations in Figures1 and 2 respectively When the failure rate is lower than25 the fault detection precisions of the two algorithmsare greater than 91 With an increase in the node failurerate the fault detection precisions of the two algorithmsare decreased but the DSFD algorithm has a higher faultdetection accuracy than does the DFD However we cansee that with a decrease in node density the performance

Det

ectio

n ac

cura

cy

DSFDDFD

01 015 02 025 03 035 04005Sensor fault probability

065

07

075

08

085

09

095

1

Figure 1 Fault sensor detection accuracy when the average degreeis 5

Det

ectio

n ac

cura

cy

DSFDDFD

01 015 02 025 03 035 04005Sensor fault probability

065

07

075

08

085

09

095

1

Figure 2 Fault sensor detection accuracy when the average degreeis 10

of each algorithm improves Taking Figure 1 as an examplewhen the sensor fault probability is higher than 40 thefault detection accuracy of DSFD algorithm is still over 87which is an improvement of 13 over the DFD algorithmThe DFD algorithm first determines the nodersquos initial stateby comparing the data from its neighborhood nodes withitself then the status of the node is determined accordingto the initial state of the node and the adjacent nodes Thismight occur because when the fault rate is high and thenumber of neighbors is large the misdiagnosis rate of DFDis high The DSFD algorithm constructs a support vectormachine regression forecasting model with historical dataand accurately determines the fault node The DSFD addsthe reference objects according to the correlation betweenmultiple sensors on nodes to reduce dependence on theneighbor nodes

6 Wireless Communications and Mobile Computing

False

alar

m ra

te

DSFDDFD

0002004006008

01012014016018

02

01 015 02 025 03 035 04005Sensor fault probability

Figure 3 Fault sensor false alarm rate when the average degree is 5

False

alar

m ra

te

DSFDDFD

0002004006008

01012014016018

02

01 015 02 025 03 035 04005Sensor fault probability

Figure 4 Fault sensor false alarm rate when the average degree is10

411 False Alarm Rate The false alarm rate (FAR) refers tothe ratio of the number of normal nodes that are mistaken asfault nodes to the total number of normal nodes

119865119860119877 = |119865 minus 119876||119873 minus 119876|

(14)

where N is the total number of nodes in the WSNsFigures 3 and 4 show the false alarm rate against the

sensor fault probability for different average number ofneighbors They indicate the performance of each algorithmat densities of 5 and 10 From the two figures we can seethat with an increase in the sensor fault probability the falsealarm rate of each algorithm increases The higher the faultprobability is the higher false alarm rate is As Figure 3 showsthe false alarm rate of DSFD is 147 it is still below 7when the sensor fault probability is 40This occurs becausethe DFD algorithm diagnoses all nodes in the monitoringfield and uses many sampling times by comparing the sensed

data from neighbor nodes Many of the sensor tests of goodsensors are likely faulty so these good sensors are then diag-nosed as faulty sensors However the DSFD algorithm notonly uses the collaborative operation of neighboring peers butalso combines the support vector machine (SVM) regressionalgorithm with the information redundancy between thesensors in the wireless sensor network The proposed DSFDalgorithm avoids the misdiagnosis caused by the number ofneighbor nodes and the incorrect data from neighbor nodesthereby achieving high detection accuracy

5 Conclusion

In this paper we modelled and analyzed a fault diagnosismechanism based on support vector machine regressionamong sensor observations in wireless sensor networksaccording to the redundant information of meteorologicalelements collected by multisensors The fault predictionmodel is built using a support vector regression algorithmto achieve residual sequences The proposed algorithm out-performs previous DFD in terms of faulty sensor detectionaccuracy and false alarm rates The fault detection algorithmachieves high detection accuracy and low false alarm rateswhich are more suitable for sparse WSNs even when thefailure rate is very high

Data Availability

The data used to support the findings of this study areavailable from the corresponding author upon request

Conflicts of Interest

The authors declare that they have no conflicts of interest

Acknowledgments

This work was supported by the National Natural ScienceFoundation of China (61402236 61373064 and 41875184)the CERNET Innovation Project (NGII20160318) and theJiangsu Province ldquoSix Talent Peaks Project in JiangsuProvincerdquo (2015-DZXX-015)

References

[1] XMiao K Liu Y He D Papadias QMa and Y Liu ldquoAgnosticdiagnosis Discovering silent failures in wireless sensor net-worksrdquo IEEE Transactions on Wireless Communications vol 12no 12 pp 6067ndash6075 2013

[2] S Wan Y Zhang and J Chen ldquoOn the construction ofdata aggregation tree with maximizing lifetime in large-scalewireless sensor networksrdquo IEEE Sensors Journal vol 16 no 20pp 7433ndash7440 2016

[3] S Wan and Y Zhang ldquoCoverage hole bypassing in wirelesssensor networksrdquo The Computer Journal vol 60 no 10 pp1536ndash1544 2017

[4] Shaohua Wan ldquoEnergy-efficient adaptive routing and context-aware lifetime maximization in wireless sensor networksrdquo

Wireless Communications and Mobile Computing 7

International Journal of Distributed Sensor Networks vol 2014Article ID 321964 16 pages 2014

[5] J Kong J-H Cui D Wu and M Gerla ldquoBuilding underwaterad-hoc networks and sensor networks for large scale real-timeaquatic applicationsrdquo inProceedings of theMilitary Communica-tions Conference (MILCOM rsquo05) pp 1535ndash1541 October 2005

[6] N Xu S Rangwala K K Chintalapudi et al ldquoA wireless sensornetwork for structural monitoringrdquo in Proceedings of the 2ndInternational Conference on Embedded Networked Sensor Sys-tems (SenSys rsquo04) pp 13ndash24 November 2004 (Catalan)

[7] Z You X ZhaoHWanWNNHung YWang andMGu ldquoAnovel fault diagnosis mechanism for wireless sensor networksrdquoMathematical and Computer Modelling vol 54 no 1-2 pp 330ndash343 2011

[8] S Rajasegarar C Leckie M Palaniswami and J C BezdekldquoDistributed anomaly detection in wireless sensor networksrdquo inProceedings of the 10th IEEE Singapore International Conferenceon Communication systems (ICCS rsquo06) pp 1ndash5 IEEE October2006

[9] M Ding D Chen K Xing and X Cheng ldquoLocalized fault-tolerant event boundary detection in sensor networksrdquo inProceedings of the IEEE 24th Annual Joint Conference of the IEEEComputer and Communications Societies (INFOCOM rsquo05) vol2 pp 902ndash913 March 2005

[10] D I Curiac and C Volosencu ldquoEnsemble based sensinganomaly detection in wireless sensor networksrdquo Expert Systemswith Applications vol 39 no 10 pp 9087ndash9096 2012

[11] M Ur-Rehman N A Malik X Yang Q H Abbasi Z Zhangand N Zhao ldquoA low profile antenna for millimeter-wavebody-centric applicationsrdquo IEEE Transactions on Antennas andPropagation vol 65 no 12 pp 6329ndash6337 2017

[12] C Wang H Lin and H Jiang ldquoCANS Towards congestion-adaptive and small stretch emergency navigation with wirelesssensor networksrdquo IEEE Transactions on Mobile Computing vol15 no 5 pp 1077ndash1089 2016

[13] J Wen B Zhou W H Mow and X-W Chang ldquoAn efficientalgorithm for optimally solving a shortest vector problem incompute-and-forward designrdquo IEEE Transactions on WirelessCommunications vol 15 no 10 pp 6541ndash6555 2016

[14] J Wen J Wang and Q Zhang ldquoNearly optimal bounds fororthogonal least squaresrdquo IEEE Transactions on Signal Process-ing vol 65 no 20 pp 5347ndash5356 2017

[15] J Wen Z Zhou Z Liu M-J Lai and X Tang ldquoSharp sufficientconditions for stable recovery of block sparse signals by blockorthogonal matching pursuitrdquo 2016 httpsarxivorgabs160502894

[16] S Rani S H Ahmed R Talwar and J Malhotra ldquoCan sensorscollect big data an energy-efficient big data gathering algo-rithm for a WSNrdquo IEEE Transactions on Industrial Informaticsvol 13 no 4 pp 1961ndash1968 2017

[17] J Chen S Kher and A Somani ldquoDistributed fault detectionof wireless sensor networksrdquo in Proceedings of the Workshop onDependability Issues in Wireless Ad Hoc Networks and SensorNetworks pp 65ndash72 2006

[18] S Rani S H Ahmed J Malhotra and R Talwar ldquoEnergyefficient chain based routing protocol for underwater wirelesssensor networksrdquo Journal of Network and Computer Applica-tions vol 92 pp 42ndash50 2017

[19] D Li and J Zhang ldquoEfficient implementation to numericallysolve the nonlinear time fractional parabolic problems onunbounded spatial domainrdquo Journal of Computational physicsvol 322 pp 415ndash428 2016

[20] A Munir J Antoon and A Gordon-Ross ldquoModeling andanalysis of fault detection and fault tolerance in wireless sensornetworksrdquoACMTransactions on EmbeddedComputing Systemsvol 14 no 1 article 3 2015

[21] G S Brar S Rani V Chopra R Malhotra H Song and SH Ahmed ldquoEnergy efficient direction-based PDORP routingprotocol for WSNrdquo IEEE Access vol 4 pp 3182ndash3194 2016

[22] ANAlvi SH Bouk SHAhmedMA YaqubM Sarkar andH Song ldquoBEST-MAC Bitmap-Assisted Efficient and ScalableTDMA-Based WSN MAC Protocol for Smart Citiesrdquo IEEEAccess vol 4 pp 312ndash322 2016

[23] S Rani and S H AhmedMulti-Hop Routing in Wireless SensorNetworks An Overview Taxonomy and Research ChallengesSpringer 2015

[24] S Rani R Talwar J Malhotra S H Ahmed M Sarkar and HSong ldquoA novel scheme for an energy efficient internet of thingsbased on wireless sensor networksrdquo Sensors vol 15 no 11 pp28603ndash28626 2015

[25] E Ould-Ahmed-Vall B H Ferri and G F Riley ldquoDistributedfault-tolerance for event detection using heterogeneouswirelesssensor networksrdquo IEEE Transactions on Mobile Computing vol11 no 12 pp 1994ndash2007 2012

[26] S C Chan H C Wu and K M Tsui ldquoRobust recursiveeigendecomposition and subspace-based algorithmswith appli-cation to fault detection in wireless sensor networksrdquo IEEETransactions on Instrumentation and Measurement vol 61 no6 pp 1703ndash1718 2012

[27] R Huang X Qiu and L Rui ldquoSimple random sampling-basedprobe station selection for fault detection in wireless sensornetworksrdquo Sensors vol 11 no 3 pp 3117ndash3134 2011

[28] J Medina-Garcıa T Sanchez-Rodrıguez J Galan A DelgadoF Gomez-Bravo and R Jimenez ldquoA wireless sensor systemfor real-time monitoring and fault detection of motor arraysrdquoSensors vol 17 no 3 p 469 2017

[29] T Muhammed and R A Shaikh ldquoAn analysis of fault detectionstrategies in wireless sensor networksrdquo Journal of Network andComputer Applications vol 78 pp 267ndash287 2017

[30] H Artail A Ajami T Saouma and M Charaf ldquoA faultynode detection scheme for wireless sensor networks that usedata aggregation for transportrdquo Wireless Communications andMobile Computing vol 16 no 14 pp 1956ndash1971 2016

[31] M Panda and P M Khilar ldquoDistributed Byzantine fault detec-tion technique in wireless sensor networks based on hypothesistestingrdquo Computers and Electrical Engineering vol 48 pp 270ndash285 2015

[32] P Jiang ldquoA new method for node fault detection in wirelesssensor networksrdquo Sensors vol 9 no 2 pp 1282ndash1294 2009

[33] K P Sharma and T P Sharma ldquorDFD reactive distributed faultdetection in wireless sensor networksrdquo Wireless Networks vol23 no 4 pp 1145ndash1160 2017

[34] M Bo H Darong and W Shaohua ldquoNTRU implementa-tion of efficient privacy-preserving location-based querying inVANETrdquoWireless Communications and Mobile Computing vol2018 Article ID 7823979 11 pages 2018

[35] Y Yang Z Gao H Zhou and X Qiu ldquoAn uncertainty-based distributed fault detectionmechanism for wireless sensornetworksrdquo Sensors vol 14 no 5 pp 7655ndash7683 2014

[36] D Wang S Wan and N Guizani ldquoContext-based probabilityneural network classifiers realized by genetic optimization formedical decision makingrdquo Multimedia Tools and Applicationsvol 77 no 17 pp 21995ndash22006 2018

4 Wireless Communications and Mobile Computing

32 Multisensor Cooperation A node in a WSN which hasmultiple sensors collects different types of environmentalinformation such as temperature humidity light and carbondioxide concentration Some of the types of environmentalinformation from the same node have close correlations Inaddition to the node collection of meteorological elementsvoltage is an important parameter in fault diagnosisThenodevoltage has an obvious effect on the temperature and othermeteorological elements Therefore the proposed algorithmproduces multiple estimates by setting up a multiple SVRforecast modelThese estimates are compared with the resid-ual sequences which come from meteorological elementsFor temperature humidity and voltage we build two SVRprediction models 1198781198811198771 and 1198781198811198772 with three types of datawhich have a redundant relationship in time and space 1198781198811198771builds the SVR prediction model with a sample consistingof temperature and humidity The current moment for k isset and the sample consisting of past continuous data is asfollows

119909119896 = (119879119896minus1 119879119896minus2 119879119896minus119897 119867119896 119867119896minus1 119867119896minus119897+1) (5)

where 119879119896minus1 119879119896minus2 119879119896minus119897 are the temperature from momentk-1 to k-l and119867119896 119867119896minus1 119867119896minus119897+1 are denoted as the humid-ity from moment k to k-l+1 The temperature values frommoment k-2 to k-l and the humidity values from k-1 to k-l+1 are used as the input samples the temperature value atk-1 is used as the output sample The estimated value 119879119896 from1198781198811198771 is a function of temperature and humidity at the pastmoment

119879119896 = 119891 (119879119896minus1 119879119896minus2 119879119896minus119897+1 119867119896 119867119896minus1 119867119896minus119897+2) (6)

1198781198811198772 builds the SVR prediction model with a sample thatconsists of the temperature and voltage The current momentfor k is set and the sample consists of past continuous data asfollows

119909119896lowast = (119879119896minus1 119879119896minus2 119879119896minus119897 119881119896 119881119896minus1 119881119896minus119897+1) (7)

where 119881119896 119881119896minus1 119881119896minus119897+1 are denoted as the voltage frommoment k to k-l+1 The estimated value 119879119896

1015840from 1198781198811198772 is a

function of temperature and voltage at the past moment

1198791198961015840 = 119891 (119879119896minus1 119879119896minus2 119879119896minus119897+1 119881119896 119881119896minus1 119881119896minus119897+2) (8)

We produce two estimated values of the temperature 119879119896 and1198791198961015840atmoment k after training the SVR forecasting model and

compare the results with the real data from the temperaturesensor to produce the residual sequence

1198641 = 119879119896 minus 1198791198961015840

1198642 = 119879119896 minus 119879119896

1198643 = 1198791198961015840 minus 119879119896

(9)

The prediction model can accurately output the tempera-ture values 119879119896 and 119879119896

1015840 according the effective history data

collected by the sensors If the data from the temperaturesensors cause an exception at moment k the residual 1198641 and1198642 are above the threshold value 1205791 and 1198643 basically showsno change We update the predictions with the actual data ifthe residual of the parameters from the nodes is less than thethreshold parameters The forecast model prepares the nextprediction

33 Distributed FaultDiagnosis Algorithm InWSNs becauseof the time-space continuum of the environmental factorsand the densely deployed nodes a node has the characteristicof spatial similarity with its nearby nodes that is the nearbynodes are likely to have similar measurements Nodes in themonitoring area will affect the accuracy of the fault diagnosisalgorithm when an event occurs As a result the algorithmgiven in this paper is combined with the fault diagnosisand neighbor coordination methods and introduces thecredibility evaluation mechanism The credibility of nodescomes not only from their own sensory information butalso from the judgment of the sensory information by theneighbor nodes which effectively eliminates the influence ofthe fault node

We consider thatN sensor nodes are randomly placed in aunit square field Without loss of generality we suppose thatthe location of each node is known and that all nodes havethe same communication radius R The average number ofnodes within a transmission range is the node density usedto illustrate the number of neighboring changes Each nodecan communicate with other nodes by one-hop ormultihopsNodes are assumed to be neighboring nodes if they are in eachotherrsquos coverage radius Each node periodically broadcasts itsmeasurements or decision such as temperature humidity airpressure and wind speed to all its neighbors The WSNs aremodelled as the system graph G(VE) where V represents aset of nodes in WSNs and E represents a set of logical linksbetween nodes Let 119889119894119904119905(119904119894 119904119895) denote the distance betweennode 119904119894 and node 119904119895 then E is based on the followingdefinition

119864 = (119904119894 119904119895) isin 1198812 | 119889119894119904119905 (119904119894 119904119895) le 119877 (10)

Thenodeswithin the transmission range of node 119904119894 belongto the neighborhood 119873(119904119894) 119873(119904119894) sub 119873 We apply faultdetection to node 119904119894 at each t timestamp The measurementof node 119904119894 at t time is denoted as 119909119894119905 Assume that theneighborhood 119873(119904119894) of node (119904119894) consists of Negi sensornodes that is 119904119895 isin 119873(119904119894) 119895 = 1 119873119890119892119894

First we use the trust level 120582119895 for the node 119904119895 isin 119873(119904119894)Each node has the same trust level at first and we set120582119895 = 120582119898119886119909 We adjust the trust level of nodes within theneighborhood by using the fault diagnosis mechanism basedon support vector machine regression If the meteorologicalelements are above the threshold value then 120582119895 = 120582119895 minus 1

Otherwise we transmit the predicted value 119909119895119896+1 and the trustlevel 120582119895 to node 119878119894 We denote the weight value for the fault

Wireless Communications and Mobile Computing 5

detection caused by the sensor data of the neighbor node 119904119895as 120596119895

120596119895 =120582119895119878119880119872

(11)

where SUM is the sum of the trust levels of all nodes withinthe neighborhood

For node 119878119894 we use the fault diagnosis mechanism basedon support vector machine regression to get the predictiondata of the next moment We calculate the failure levelindicator by the space-time correlation between nodes inwireless sensor networks

119891119894 =10038171003817100381710038171003817100381710038171003817100381710038171003817119894119896+1 minus

119873119890119892119894

sum119895=1

120596119895 sdot 119895

119896+1

10038171003817100381710038171003817100381710038171003817100381710038171003817(12)

If 119891119894 gt 1205792 the trust level 120582119894 = 0 and node 119904119894 fails Otherwisethe predicted value is updated by the actual value and theforecast model to prepare the next prediction

4 Simulation

We experiment in MATLAB to assess the performance of theproposed approachTheWSNs contain 200 nodes in a squareregion of 30 times 30 units Each sensor is randomly placed in aunit grid The measurements of the nodes in the normal areaare subject to a Gaussian distribution The data from Intellab are used in the experiment [43] including temperaturehumidity and voltage as the experimental data We use thetile radial primary kernel function and set 120582119898119886119909 = 10 1205791 =05 1205792 = 0375 The performance of the proposed DSFD(Distributed SVR Fault Detection) algorithm is evaluated andcompared with the existing DFD algorithm in [17] in termsof the detection accuracy (DA) and false alarm rate (FAR)in the network All experiments are repeated 100 times anddata for analysis are the averaged to ensure the statisticalsignificance of the experiments To assess the effect of faultynode identification two indicators are usually employeddetection accuracy and false alarm rate

41 Detection Accuracy Detection accuracy (DA) refers tothe ratio of the number of correctly identified faulty nodesto the total number of actual fault nodes

119863119860 = |119865||119876|

(13)

where F is the set of fault nodes which the algorithm hasdetected and Q is the set of actual fault nodes

We compare two algorithms in terms of detection accu-racy under different sensor density configurations in Figures1 and 2 respectively When the failure rate is lower than25 the fault detection precisions of the two algorithmsare greater than 91 With an increase in the node failurerate the fault detection precisions of the two algorithmsare decreased but the DSFD algorithm has a higher faultdetection accuracy than does the DFD However we cansee that with a decrease in node density the performance

Det

ectio

n ac

cura

cy

DSFDDFD

01 015 02 025 03 035 04005Sensor fault probability

065

07

075

08

085

09

095

1

Figure 1 Fault sensor detection accuracy when the average degreeis 5

Det

ectio

n ac

cura

cy

DSFDDFD

01 015 02 025 03 035 04005Sensor fault probability

065

07

075

08

085

09

095

1

Figure 2 Fault sensor detection accuracy when the average degreeis 10

of each algorithm improves Taking Figure 1 as an examplewhen the sensor fault probability is higher than 40 thefault detection accuracy of DSFD algorithm is still over 87which is an improvement of 13 over the DFD algorithmThe DFD algorithm first determines the nodersquos initial stateby comparing the data from its neighborhood nodes withitself then the status of the node is determined accordingto the initial state of the node and the adjacent nodes Thismight occur because when the fault rate is high and thenumber of neighbors is large the misdiagnosis rate of DFDis high The DSFD algorithm constructs a support vectormachine regression forecasting model with historical dataand accurately determines the fault node The DSFD addsthe reference objects according to the correlation betweenmultiple sensors on nodes to reduce dependence on theneighbor nodes

6 Wireless Communications and Mobile Computing

False

alar

m ra

te

DSFDDFD

0002004006008

01012014016018

02

01 015 02 025 03 035 04005Sensor fault probability

Figure 3 Fault sensor false alarm rate when the average degree is 5

False

alar

m ra

te

DSFDDFD

0002004006008

01012014016018

02

01 015 02 025 03 035 04005Sensor fault probability

Figure 4 Fault sensor false alarm rate when the average degree is10

411 False Alarm Rate The false alarm rate (FAR) refers tothe ratio of the number of normal nodes that are mistaken asfault nodes to the total number of normal nodes

119865119860119877 = |119865 minus 119876||119873 minus 119876|

(14)

where N is the total number of nodes in the WSNsFigures 3 and 4 show the false alarm rate against the

sensor fault probability for different average number ofneighbors They indicate the performance of each algorithmat densities of 5 and 10 From the two figures we can seethat with an increase in the sensor fault probability the falsealarm rate of each algorithm increases The higher the faultprobability is the higher false alarm rate is As Figure 3 showsthe false alarm rate of DSFD is 147 it is still below 7when the sensor fault probability is 40This occurs becausethe DFD algorithm diagnoses all nodes in the monitoringfield and uses many sampling times by comparing the sensed

data from neighbor nodes Many of the sensor tests of goodsensors are likely faulty so these good sensors are then diag-nosed as faulty sensors However the DSFD algorithm notonly uses the collaborative operation of neighboring peers butalso combines the support vector machine (SVM) regressionalgorithm with the information redundancy between thesensors in the wireless sensor network The proposed DSFDalgorithm avoids the misdiagnosis caused by the number ofneighbor nodes and the incorrect data from neighbor nodesthereby achieving high detection accuracy

5 Conclusion

In this paper we modelled and analyzed a fault diagnosismechanism based on support vector machine regressionamong sensor observations in wireless sensor networksaccording to the redundant information of meteorologicalelements collected by multisensors The fault predictionmodel is built using a support vector regression algorithmto achieve residual sequences The proposed algorithm out-performs previous DFD in terms of faulty sensor detectionaccuracy and false alarm rates The fault detection algorithmachieves high detection accuracy and low false alarm rateswhich are more suitable for sparse WSNs even when thefailure rate is very high

Data Availability

The data used to support the findings of this study areavailable from the corresponding author upon request

Conflicts of Interest

The authors declare that they have no conflicts of interest

Acknowledgments

This work was supported by the National Natural ScienceFoundation of China (61402236 61373064 and 41875184)the CERNET Innovation Project (NGII20160318) and theJiangsu Province ldquoSix Talent Peaks Project in JiangsuProvincerdquo (2015-DZXX-015)

References

[1] XMiao K Liu Y He D Papadias QMa and Y Liu ldquoAgnosticdiagnosis Discovering silent failures in wireless sensor net-worksrdquo IEEE Transactions on Wireless Communications vol 12no 12 pp 6067ndash6075 2013

[2] S Wan Y Zhang and J Chen ldquoOn the construction ofdata aggregation tree with maximizing lifetime in large-scalewireless sensor networksrdquo IEEE Sensors Journal vol 16 no 20pp 7433ndash7440 2016

[3] S Wan and Y Zhang ldquoCoverage hole bypassing in wirelesssensor networksrdquo The Computer Journal vol 60 no 10 pp1536ndash1544 2017

[4] Shaohua Wan ldquoEnergy-efficient adaptive routing and context-aware lifetime maximization in wireless sensor networksrdquo

Wireless Communications and Mobile Computing 7

International Journal of Distributed Sensor Networks vol 2014Article ID 321964 16 pages 2014

[5] J Kong J-H Cui D Wu and M Gerla ldquoBuilding underwaterad-hoc networks and sensor networks for large scale real-timeaquatic applicationsrdquo inProceedings of theMilitary Communica-tions Conference (MILCOM rsquo05) pp 1535ndash1541 October 2005

[6] N Xu S Rangwala K K Chintalapudi et al ldquoA wireless sensornetwork for structural monitoringrdquo in Proceedings of the 2ndInternational Conference on Embedded Networked Sensor Sys-tems (SenSys rsquo04) pp 13ndash24 November 2004 (Catalan)

[7] Z You X ZhaoHWanWNNHung YWang andMGu ldquoAnovel fault diagnosis mechanism for wireless sensor networksrdquoMathematical and Computer Modelling vol 54 no 1-2 pp 330ndash343 2011

[8] S Rajasegarar C Leckie M Palaniswami and J C BezdekldquoDistributed anomaly detection in wireless sensor networksrdquo inProceedings of the 10th IEEE Singapore International Conferenceon Communication systems (ICCS rsquo06) pp 1ndash5 IEEE October2006

[9] M Ding D Chen K Xing and X Cheng ldquoLocalized fault-tolerant event boundary detection in sensor networksrdquo inProceedings of the IEEE 24th Annual Joint Conference of the IEEEComputer and Communications Societies (INFOCOM rsquo05) vol2 pp 902ndash913 March 2005

[10] D I Curiac and C Volosencu ldquoEnsemble based sensinganomaly detection in wireless sensor networksrdquo Expert Systemswith Applications vol 39 no 10 pp 9087ndash9096 2012

[11] M Ur-Rehman N A Malik X Yang Q H Abbasi Z Zhangand N Zhao ldquoA low profile antenna for millimeter-wavebody-centric applicationsrdquo IEEE Transactions on Antennas andPropagation vol 65 no 12 pp 6329ndash6337 2017

[12] C Wang H Lin and H Jiang ldquoCANS Towards congestion-adaptive and small stretch emergency navigation with wirelesssensor networksrdquo IEEE Transactions on Mobile Computing vol15 no 5 pp 1077ndash1089 2016

[13] J Wen B Zhou W H Mow and X-W Chang ldquoAn efficientalgorithm for optimally solving a shortest vector problem incompute-and-forward designrdquo IEEE Transactions on WirelessCommunications vol 15 no 10 pp 6541ndash6555 2016

[14] J Wen J Wang and Q Zhang ldquoNearly optimal bounds fororthogonal least squaresrdquo IEEE Transactions on Signal Process-ing vol 65 no 20 pp 5347ndash5356 2017

[15] J Wen Z Zhou Z Liu M-J Lai and X Tang ldquoSharp sufficientconditions for stable recovery of block sparse signals by blockorthogonal matching pursuitrdquo 2016 httpsarxivorgabs160502894

[16] S Rani S H Ahmed R Talwar and J Malhotra ldquoCan sensorscollect big data an energy-efficient big data gathering algo-rithm for a WSNrdquo IEEE Transactions on Industrial Informaticsvol 13 no 4 pp 1961ndash1968 2017

[17] J Chen S Kher and A Somani ldquoDistributed fault detectionof wireless sensor networksrdquo in Proceedings of the Workshop onDependability Issues in Wireless Ad Hoc Networks and SensorNetworks pp 65ndash72 2006

[18] S Rani S H Ahmed J Malhotra and R Talwar ldquoEnergyefficient chain based routing protocol for underwater wirelesssensor networksrdquo Journal of Network and Computer Applica-tions vol 92 pp 42ndash50 2017

[19] D Li and J Zhang ldquoEfficient implementation to numericallysolve the nonlinear time fractional parabolic problems onunbounded spatial domainrdquo Journal of Computational physicsvol 322 pp 415ndash428 2016

[20] A Munir J Antoon and A Gordon-Ross ldquoModeling andanalysis of fault detection and fault tolerance in wireless sensornetworksrdquoACMTransactions on EmbeddedComputing Systemsvol 14 no 1 article 3 2015

[21] G S Brar S Rani V Chopra R Malhotra H Song and SH Ahmed ldquoEnergy efficient direction-based PDORP routingprotocol for WSNrdquo IEEE Access vol 4 pp 3182ndash3194 2016

[22] ANAlvi SH Bouk SHAhmedMA YaqubM Sarkar andH Song ldquoBEST-MAC Bitmap-Assisted Efficient and ScalableTDMA-Based WSN MAC Protocol for Smart Citiesrdquo IEEEAccess vol 4 pp 312ndash322 2016

[23] S Rani and S H AhmedMulti-Hop Routing in Wireless SensorNetworks An Overview Taxonomy and Research ChallengesSpringer 2015

[24] S Rani R Talwar J Malhotra S H Ahmed M Sarkar and HSong ldquoA novel scheme for an energy efficient internet of thingsbased on wireless sensor networksrdquo Sensors vol 15 no 11 pp28603ndash28626 2015

[25] E Ould-Ahmed-Vall B H Ferri and G F Riley ldquoDistributedfault-tolerance for event detection using heterogeneouswirelesssensor networksrdquo IEEE Transactions on Mobile Computing vol11 no 12 pp 1994ndash2007 2012

[26] S C Chan H C Wu and K M Tsui ldquoRobust recursiveeigendecomposition and subspace-based algorithmswith appli-cation to fault detection in wireless sensor networksrdquo IEEETransactions on Instrumentation and Measurement vol 61 no6 pp 1703ndash1718 2012

[27] R Huang X Qiu and L Rui ldquoSimple random sampling-basedprobe station selection for fault detection in wireless sensornetworksrdquo Sensors vol 11 no 3 pp 3117ndash3134 2011

[28] J Medina-Garcıa T Sanchez-Rodrıguez J Galan A DelgadoF Gomez-Bravo and R Jimenez ldquoA wireless sensor systemfor real-time monitoring and fault detection of motor arraysrdquoSensors vol 17 no 3 p 469 2017

[29] T Muhammed and R A Shaikh ldquoAn analysis of fault detectionstrategies in wireless sensor networksrdquo Journal of Network andComputer Applications vol 78 pp 267ndash287 2017

[30] H Artail A Ajami T Saouma and M Charaf ldquoA faultynode detection scheme for wireless sensor networks that usedata aggregation for transportrdquo Wireless Communications andMobile Computing vol 16 no 14 pp 1956ndash1971 2016

[31] M Panda and P M Khilar ldquoDistributed Byzantine fault detec-tion technique in wireless sensor networks based on hypothesistestingrdquo Computers and Electrical Engineering vol 48 pp 270ndash285 2015

[32] P Jiang ldquoA new method for node fault detection in wirelesssensor networksrdquo Sensors vol 9 no 2 pp 1282ndash1294 2009

[33] K P Sharma and T P Sharma ldquorDFD reactive distributed faultdetection in wireless sensor networksrdquo Wireless Networks vol23 no 4 pp 1145ndash1160 2017

[34] M Bo H Darong and W Shaohua ldquoNTRU implementa-tion of efficient privacy-preserving location-based querying inVANETrdquoWireless Communications and Mobile Computing vol2018 Article ID 7823979 11 pages 2018

[35] Y Yang Z Gao H Zhou and X Qiu ldquoAn uncertainty-based distributed fault detectionmechanism for wireless sensornetworksrdquo Sensors vol 14 no 5 pp 7655ndash7683 2014

[36] D Wang S Wan and N Guizani ldquoContext-based probabilityneural network classifiers realized by genetic optimization formedical decision makingrdquo Multimedia Tools and Applicationsvol 77 no 17 pp 21995ndash22006 2018

Wireless Communications and Mobile Computing 5

detection caused by the sensor data of the neighbor node 119904119895as 120596119895

120596119895 =120582119895119878119880119872

(11)

where SUM is the sum of the trust levels of all nodes withinthe neighborhood

For node 119878119894 we use the fault diagnosis mechanism basedon support vector machine regression to get the predictiondata of the next moment We calculate the failure levelindicator by the space-time correlation between nodes inwireless sensor networks

119891119894 =10038171003817100381710038171003817100381710038171003817100381710038171003817119894119896+1 minus

119873119890119892119894

sum119895=1

120596119895 sdot 119895

119896+1

10038171003817100381710038171003817100381710038171003817100381710038171003817(12)

If 119891119894 gt 1205792 the trust level 120582119894 = 0 and node 119904119894 fails Otherwisethe predicted value is updated by the actual value and theforecast model to prepare the next prediction

4 Simulation

We experiment in MATLAB to assess the performance of theproposed approachTheWSNs contain 200 nodes in a squareregion of 30 times 30 units Each sensor is randomly placed in aunit grid The measurements of the nodes in the normal areaare subject to a Gaussian distribution The data from Intellab are used in the experiment [43] including temperaturehumidity and voltage as the experimental data We use thetile radial primary kernel function and set 120582119898119886119909 = 10 1205791 =05 1205792 = 0375 The performance of the proposed DSFD(Distributed SVR Fault Detection) algorithm is evaluated andcompared with the existing DFD algorithm in [17] in termsof the detection accuracy (DA) and false alarm rate (FAR)in the network All experiments are repeated 100 times anddata for analysis are the averaged to ensure the statisticalsignificance of the experiments To assess the effect of faultynode identification two indicators are usually employeddetection accuracy and false alarm rate

41 Detection Accuracy Detection accuracy (DA) refers tothe ratio of the number of correctly identified faulty nodesto the total number of actual fault nodes

119863119860 = |119865||119876|

(13)

where F is the set of fault nodes which the algorithm hasdetected and Q is the set of actual fault nodes

We compare two algorithms in terms of detection accu-racy under different sensor density configurations in Figures1 and 2 respectively When the failure rate is lower than25 the fault detection precisions of the two algorithmsare greater than 91 With an increase in the node failurerate the fault detection precisions of the two algorithmsare decreased but the DSFD algorithm has a higher faultdetection accuracy than does the DFD However we cansee that with a decrease in node density the performance

Det

ectio

n ac

cura

cy

DSFDDFD

01 015 02 025 03 035 04005Sensor fault probability

065

07

075

08

085

09

095

1

Figure 1 Fault sensor detection accuracy when the average degreeis 5

Det

ectio

n ac

cura

cy

DSFDDFD

01 015 02 025 03 035 04005Sensor fault probability

065

07

075

08

085

09

095

1

Figure 2 Fault sensor detection accuracy when the average degreeis 10

of each algorithm improves Taking Figure 1 as an examplewhen the sensor fault probability is higher than 40 thefault detection accuracy of DSFD algorithm is still over 87which is an improvement of 13 over the DFD algorithmThe DFD algorithm first determines the nodersquos initial stateby comparing the data from its neighborhood nodes withitself then the status of the node is determined accordingto the initial state of the node and the adjacent nodes Thismight occur because when the fault rate is high and thenumber of neighbors is large the misdiagnosis rate of DFDis high The DSFD algorithm constructs a support vectormachine regression forecasting model with historical dataand accurately determines the fault node The DSFD addsthe reference objects according to the correlation betweenmultiple sensors on nodes to reduce dependence on theneighbor nodes

6 Wireless Communications and Mobile Computing

False

alar

m ra

te

DSFDDFD

0002004006008

01012014016018

02

01 015 02 025 03 035 04005Sensor fault probability

Figure 3 Fault sensor false alarm rate when the average degree is 5

False

alar

m ra

te

DSFDDFD

0002004006008

01012014016018

02

01 015 02 025 03 035 04005Sensor fault probability

Figure 4 Fault sensor false alarm rate when the average degree is10

411 False Alarm Rate The false alarm rate (FAR) refers tothe ratio of the number of normal nodes that are mistaken asfault nodes to the total number of normal nodes

119865119860119877 = |119865 minus 119876||119873 minus 119876|

(14)

where N is the total number of nodes in the WSNsFigures 3 and 4 show the false alarm rate against the

sensor fault probability for different average number ofneighbors They indicate the performance of each algorithmat densities of 5 and 10 From the two figures we can seethat with an increase in the sensor fault probability the falsealarm rate of each algorithm increases The higher the faultprobability is the higher false alarm rate is As Figure 3 showsthe false alarm rate of DSFD is 147 it is still below 7when the sensor fault probability is 40This occurs becausethe DFD algorithm diagnoses all nodes in the monitoringfield and uses many sampling times by comparing the sensed

data from neighbor nodes Many of the sensor tests of goodsensors are likely faulty so these good sensors are then diag-nosed as faulty sensors However the DSFD algorithm notonly uses the collaborative operation of neighboring peers butalso combines the support vector machine (SVM) regressionalgorithm with the information redundancy between thesensors in the wireless sensor network The proposed DSFDalgorithm avoids the misdiagnosis caused by the number ofneighbor nodes and the incorrect data from neighbor nodesthereby achieving high detection accuracy

5 Conclusion

In this paper we modelled and analyzed a fault diagnosismechanism based on support vector machine regressionamong sensor observations in wireless sensor networksaccording to the redundant information of meteorologicalelements collected by multisensors The fault predictionmodel is built using a support vector regression algorithmto achieve residual sequences The proposed algorithm out-performs previous DFD in terms of faulty sensor detectionaccuracy and false alarm rates The fault detection algorithmachieves high detection accuracy and low false alarm rateswhich are more suitable for sparse WSNs even when thefailure rate is very high

Data Availability

The data used to support the findings of this study areavailable from the corresponding author upon request

Conflicts of Interest

The authors declare that they have no conflicts of interest

Acknowledgments

This work was supported by the National Natural ScienceFoundation of China (61402236 61373064 and 41875184)the CERNET Innovation Project (NGII20160318) and theJiangsu Province ldquoSix Talent Peaks Project in JiangsuProvincerdquo (2015-DZXX-015)

References

[1] XMiao K Liu Y He D Papadias QMa and Y Liu ldquoAgnosticdiagnosis Discovering silent failures in wireless sensor net-worksrdquo IEEE Transactions on Wireless Communications vol 12no 12 pp 6067ndash6075 2013

[2] S Wan Y Zhang and J Chen ldquoOn the construction ofdata aggregation tree with maximizing lifetime in large-scalewireless sensor networksrdquo IEEE Sensors Journal vol 16 no 20pp 7433ndash7440 2016

[3] S Wan and Y Zhang ldquoCoverage hole bypassing in wirelesssensor networksrdquo The Computer Journal vol 60 no 10 pp1536ndash1544 2017

[4] Shaohua Wan ldquoEnergy-efficient adaptive routing and context-aware lifetime maximization in wireless sensor networksrdquo

Wireless Communications and Mobile Computing 7

International Journal of Distributed Sensor Networks vol 2014Article ID 321964 16 pages 2014

[5] J Kong J-H Cui D Wu and M Gerla ldquoBuilding underwaterad-hoc networks and sensor networks for large scale real-timeaquatic applicationsrdquo inProceedings of theMilitary Communica-tions Conference (MILCOM rsquo05) pp 1535ndash1541 October 2005

[6] N Xu S Rangwala K K Chintalapudi et al ldquoA wireless sensornetwork for structural monitoringrdquo in Proceedings of the 2ndInternational Conference on Embedded Networked Sensor Sys-tems (SenSys rsquo04) pp 13ndash24 November 2004 (Catalan)

[7] Z You X ZhaoHWanWNNHung YWang andMGu ldquoAnovel fault diagnosis mechanism for wireless sensor networksrdquoMathematical and Computer Modelling vol 54 no 1-2 pp 330ndash343 2011

[8] S Rajasegarar C Leckie M Palaniswami and J C BezdekldquoDistributed anomaly detection in wireless sensor networksrdquo inProceedings of the 10th IEEE Singapore International Conferenceon Communication systems (ICCS rsquo06) pp 1ndash5 IEEE October2006

[9] M Ding D Chen K Xing and X Cheng ldquoLocalized fault-tolerant event boundary detection in sensor networksrdquo inProceedings of the IEEE 24th Annual Joint Conference of the IEEEComputer and Communications Societies (INFOCOM rsquo05) vol2 pp 902ndash913 March 2005

[10] D I Curiac and C Volosencu ldquoEnsemble based sensinganomaly detection in wireless sensor networksrdquo Expert Systemswith Applications vol 39 no 10 pp 9087ndash9096 2012

[11] M Ur-Rehman N A Malik X Yang Q H Abbasi Z Zhangand N Zhao ldquoA low profile antenna for millimeter-wavebody-centric applicationsrdquo IEEE Transactions on Antennas andPropagation vol 65 no 12 pp 6329ndash6337 2017

[12] C Wang H Lin and H Jiang ldquoCANS Towards congestion-adaptive and small stretch emergency navigation with wirelesssensor networksrdquo IEEE Transactions on Mobile Computing vol15 no 5 pp 1077ndash1089 2016

[13] J Wen B Zhou W H Mow and X-W Chang ldquoAn efficientalgorithm for optimally solving a shortest vector problem incompute-and-forward designrdquo IEEE Transactions on WirelessCommunications vol 15 no 10 pp 6541ndash6555 2016

[14] J Wen J Wang and Q Zhang ldquoNearly optimal bounds fororthogonal least squaresrdquo IEEE Transactions on Signal Process-ing vol 65 no 20 pp 5347ndash5356 2017

[15] J Wen Z Zhou Z Liu M-J Lai and X Tang ldquoSharp sufficientconditions for stable recovery of block sparse signals by blockorthogonal matching pursuitrdquo 2016 httpsarxivorgabs160502894

[16] S Rani S H Ahmed R Talwar and J Malhotra ldquoCan sensorscollect big data an energy-efficient big data gathering algo-rithm for a WSNrdquo IEEE Transactions on Industrial Informaticsvol 13 no 4 pp 1961ndash1968 2017

[17] J Chen S Kher and A Somani ldquoDistributed fault detectionof wireless sensor networksrdquo in Proceedings of the Workshop onDependability Issues in Wireless Ad Hoc Networks and SensorNetworks pp 65ndash72 2006

[18] S Rani S H Ahmed J Malhotra and R Talwar ldquoEnergyefficient chain based routing protocol for underwater wirelesssensor networksrdquo Journal of Network and Computer Applica-tions vol 92 pp 42ndash50 2017

[19] D Li and J Zhang ldquoEfficient implementation to numericallysolve the nonlinear time fractional parabolic problems onunbounded spatial domainrdquo Journal of Computational physicsvol 322 pp 415ndash428 2016

[20] A Munir J Antoon and A Gordon-Ross ldquoModeling andanalysis of fault detection and fault tolerance in wireless sensornetworksrdquoACMTransactions on EmbeddedComputing Systemsvol 14 no 1 article 3 2015

[21] G S Brar S Rani V Chopra R Malhotra H Song and SH Ahmed ldquoEnergy efficient direction-based PDORP routingprotocol for WSNrdquo IEEE Access vol 4 pp 3182ndash3194 2016

[22] ANAlvi SH Bouk SHAhmedMA YaqubM Sarkar andH Song ldquoBEST-MAC Bitmap-Assisted Efficient and ScalableTDMA-Based WSN MAC Protocol for Smart Citiesrdquo IEEEAccess vol 4 pp 312ndash322 2016

[23] S Rani and S H AhmedMulti-Hop Routing in Wireless SensorNetworks An Overview Taxonomy and Research ChallengesSpringer 2015

[24] S Rani R Talwar J Malhotra S H Ahmed M Sarkar and HSong ldquoA novel scheme for an energy efficient internet of thingsbased on wireless sensor networksrdquo Sensors vol 15 no 11 pp28603ndash28626 2015

[25] E Ould-Ahmed-Vall B H Ferri and G F Riley ldquoDistributedfault-tolerance for event detection using heterogeneouswirelesssensor networksrdquo IEEE Transactions on Mobile Computing vol11 no 12 pp 1994ndash2007 2012

[26] S C Chan H C Wu and K M Tsui ldquoRobust recursiveeigendecomposition and subspace-based algorithmswith appli-cation to fault detection in wireless sensor networksrdquo IEEETransactions on Instrumentation and Measurement vol 61 no6 pp 1703ndash1718 2012

[27] R Huang X Qiu and L Rui ldquoSimple random sampling-basedprobe station selection for fault detection in wireless sensornetworksrdquo Sensors vol 11 no 3 pp 3117ndash3134 2011

[28] J Medina-Garcıa T Sanchez-Rodrıguez J Galan A DelgadoF Gomez-Bravo and R Jimenez ldquoA wireless sensor systemfor real-time monitoring and fault detection of motor arraysrdquoSensors vol 17 no 3 p 469 2017

[29] T Muhammed and R A Shaikh ldquoAn analysis of fault detectionstrategies in wireless sensor networksrdquo Journal of Network andComputer Applications vol 78 pp 267ndash287 2017

[30] H Artail A Ajami T Saouma and M Charaf ldquoA faultynode detection scheme for wireless sensor networks that usedata aggregation for transportrdquo Wireless Communications andMobile Computing vol 16 no 14 pp 1956ndash1971 2016

[31] M Panda and P M Khilar ldquoDistributed Byzantine fault detec-tion technique in wireless sensor networks based on hypothesistestingrdquo Computers and Electrical Engineering vol 48 pp 270ndash285 2015

[32] P Jiang ldquoA new method for node fault detection in wirelesssensor networksrdquo Sensors vol 9 no 2 pp 1282ndash1294 2009

[33] K P Sharma and T P Sharma ldquorDFD reactive distributed faultdetection in wireless sensor networksrdquo Wireless Networks vol23 no 4 pp 1145ndash1160 2017

[34] M Bo H Darong and W Shaohua ldquoNTRU implementa-tion of efficient privacy-preserving location-based querying inVANETrdquoWireless Communications and Mobile Computing vol2018 Article ID 7823979 11 pages 2018

[35] Y Yang Z Gao H Zhou and X Qiu ldquoAn uncertainty-based distributed fault detectionmechanism for wireless sensornetworksrdquo Sensors vol 14 no 5 pp 7655ndash7683 2014

[36] D Wang S Wan and N Guizani ldquoContext-based probabilityneural network classifiers realized by genetic optimization formedical decision makingrdquo Multimedia Tools and Applicationsvol 77 no 17 pp 21995ndash22006 2018

6 Wireless Communications and Mobile Computing

False

alar

m ra

te

DSFDDFD

0002004006008

01012014016018

02

01 015 02 025 03 035 04005Sensor fault probability

Figure 3 Fault sensor false alarm rate when the average degree is 5

False

alar

m ra

te

DSFDDFD

0002004006008

01012014016018

02

01 015 02 025 03 035 04005Sensor fault probability

Figure 4 Fault sensor false alarm rate when the average degree is10

411 False Alarm Rate The false alarm rate (FAR) refers tothe ratio of the number of normal nodes that are mistaken asfault nodes to the total number of normal nodes

119865119860119877 = |119865 minus 119876||119873 minus 119876|

(14)

where N is the total number of nodes in the WSNsFigures 3 and 4 show the false alarm rate against the

sensor fault probability for different average number ofneighbors They indicate the performance of each algorithmat densities of 5 and 10 From the two figures we can seethat with an increase in the sensor fault probability the falsealarm rate of each algorithm increases The higher the faultprobability is the higher false alarm rate is As Figure 3 showsthe false alarm rate of DSFD is 147 it is still below 7when the sensor fault probability is 40This occurs becausethe DFD algorithm diagnoses all nodes in the monitoringfield and uses many sampling times by comparing the sensed

data from neighbor nodes Many of the sensor tests of goodsensors are likely faulty so these good sensors are then diag-nosed as faulty sensors However the DSFD algorithm notonly uses the collaborative operation of neighboring peers butalso combines the support vector machine (SVM) regressionalgorithm with the information redundancy between thesensors in the wireless sensor network The proposed DSFDalgorithm avoids the misdiagnosis caused by the number ofneighbor nodes and the incorrect data from neighbor nodesthereby achieving high detection accuracy

5 Conclusion

In this paper we modelled and analyzed a fault diagnosismechanism based on support vector machine regressionamong sensor observations in wireless sensor networksaccording to the redundant information of meteorologicalelements collected by multisensors The fault predictionmodel is built using a support vector regression algorithmto achieve residual sequences The proposed algorithm out-performs previous DFD in terms of faulty sensor detectionaccuracy and false alarm rates The fault detection algorithmachieves high detection accuracy and low false alarm rateswhich are more suitable for sparse WSNs even when thefailure rate is very high

Data Availability

The data used to support the findings of this study areavailable from the corresponding author upon request

Conflicts of Interest

The authors declare that they have no conflicts of interest

Acknowledgments

This work was supported by the National Natural ScienceFoundation of China (61402236 61373064 and 41875184)the CERNET Innovation Project (NGII20160318) and theJiangsu Province ldquoSix Talent Peaks Project in JiangsuProvincerdquo (2015-DZXX-015)

References

[1] XMiao K Liu Y He D Papadias QMa and Y Liu ldquoAgnosticdiagnosis Discovering silent failures in wireless sensor net-worksrdquo IEEE Transactions on Wireless Communications vol 12no 12 pp 6067ndash6075 2013

[2] S Wan Y Zhang and J Chen ldquoOn the construction ofdata aggregation tree with maximizing lifetime in large-scalewireless sensor networksrdquo IEEE Sensors Journal vol 16 no 20pp 7433ndash7440 2016

[3] S Wan and Y Zhang ldquoCoverage hole bypassing in wirelesssensor networksrdquo The Computer Journal vol 60 no 10 pp1536ndash1544 2017

[4] Shaohua Wan ldquoEnergy-efficient adaptive routing and context-aware lifetime maximization in wireless sensor networksrdquo

Wireless Communications and Mobile Computing 7

International Journal of Distributed Sensor Networks vol 2014Article ID 321964 16 pages 2014

[5] J Kong J-H Cui D Wu and M Gerla ldquoBuilding underwaterad-hoc networks and sensor networks for large scale real-timeaquatic applicationsrdquo inProceedings of theMilitary Communica-tions Conference (MILCOM rsquo05) pp 1535ndash1541 October 2005

[6] N Xu S Rangwala K K Chintalapudi et al ldquoA wireless sensornetwork for structural monitoringrdquo in Proceedings of the 2ndInternational Conference on Embedded Networked Sensor Sys-tems (SenSys rsquo04) pp 13ndash24 November 2004 (Catalan)

[7] Z You X ZhaoHWanWNNHung YWang andMGu ldquoAnovel fault diagnosis mechanism for wireless sensor networksrdquoMathematical and Computer Modelling vol 54 no 1-2 pp 330ndash343 2011

[8] S Rajasegarar C Leckie M Palaniswami and J C BezdekldquoDistributed anomaly detection in wireless sensor networksrdquo inProceedings of the 10th IEEE Singapore International Conferenceon Communication systems (ICCS rsquo06) pp 1ndash5 IEEE October2006

[9] M Ding D Chen K Xing and X Cheng ldquoLocalized fault-tolerant event boundary detection in sensor networksrdquo inProceedings of the IEEE 24th Annual Joint Conference of the IEEEComputer and Communications Societies (INFOCOM rsquo05) vol2 pp 902ndash913 March 2005

[10] D I Curiac and C Volosencu ldquoEnsemble based sensinganomaly detection in wireless sensor networksrdquo Expert Systemswith Applications vol 39 no 10 pp 9087ndash9096 2012

[11] M Ur-Rehman N A Malik X Yang Q H Abbasi Z Zhangand N Zhao ldquoA low profile antenna for millimeter-wavebody-centric applicationsrdquo IEEE Transactions on Antennas andPropagation vol 65 no 12 pp 6329ndash6337 2017

[12] C Wang H Lin and H Jiang ldquoCANS Towards congestion-adaptive and small stretch emergency navigation with wirelesssensor networksrdquo IEEE Transactions on Mobile Computing vol15 no 5 pp 1077ndash1089 2016

[13] J Wen B Zhou W H Mow and X-W Chang ldquoAn efficientalgorithm for optimally solving a shortest vector problem incompute-and-forward designrdquo IEEE Transactions on WirelessCommunications vol 15 no 10 pp 6541ndash6555 2016

[14] J Wen J Wang and Q Zhang ldquoNearly optimal bounds fororthogonal least squaresrdquo IEEE Transactions on Signal Process-ing vol 65 no 20 pp 5347ndash5356 2017

[15] J Wen Z Zhou Z Liu M-J Lai and X Tang ldquoSharp sufficientconditions for stable recovery of block sparse signals by blockorthogonal matching pursuitrdquo 2016 httpsarxivorgabs160502894

[16] S Rani S H Ahmed R Talwar and J Malhotra ldquoCan sensorscollect big data an energy-efficient big data gathering algo-rithm for a WSNrdquo IEEE Transactions on Industrial Informaticsvol 13 no 4 pp 1961ndash1968 2017

[17] J Chen S Kher and A Somani ldquoDistributed fault detectionof wireless sensor networksrdquo in Proceedings of the Workshop onDependability Issues in Wireless Ad Hoc Networks and SensorNetworks pp 65ndash72 2006

[18] S Rani S H Ahmed J Malhotra and R Talwar ldquoEnergyefficient chain based routing protocol for underwater wirelesssensor networksrdquo Journal of Network and Computer Applica-tions vol 92 pp 42ndash50 2017

[19] D Li and J Zhang ldquoEfficient implementation to numericallysolve the nonlinear time fractional parabolic problems onunbounded spatial domainrdquo Journal of Computational physicsvol 322 pp 415ndash428 2016

[20] A Munir J Antoon and A Gordon-Ross ldquoModeling andanalysis of fault detection and fault tolerance in wireless sensornetworksrdquoACMTransactions on EmbeddedComputing Systemsvol 14 no 1 article 3 2015

[21] G S Brar S Rani V Chopra R Malhotra H Song and SH Ahmed ldquoEnergy efficient direction-based PDORP routingprotocol for WSNrdquo IEEE Access vol 4 pp 3182ndash3194 2016

[22] ANAlvi SH Bouk SHAhmedMA YaqubM Sarkar andH Song ldquoBEST-MAC Bitmap-Assisted Efficient and ScalableTDMA-Based WSN MAC Protocol for Smart Citiesrdquo IEEEAccess vol 4 pp 312ndash322 2016

[23] S Rani and S H AhmedMulti-Hop Routing in Wireless SensorNetworks An Overview Taxonomy and Research ChallengesSpringer 2015

[24] S Rani R Talwar J Malhotra S H Ahmed M Sarkar and HSong ldquoA novel scheme for an energy efficient internet of thingsbased on wireless sensor networksrdquo Sensors vol 15 no 11 pp28603ndash28626 2015

[25] E Ould-Ahmed-Vall B H Ferri and G F Riley ldquoDistributedfault-tolerance for event detection using heterogeneouswirelesssensor networksrdquo IEEE Transactions on Mobile Computing vol11 no 12 pp 1994ndash2007 2012

[26] S C Chan H C Wu and K M Tsui ldquoRobust recursiveeigendecomposition and subspace-based algorithmswith appli-cation to fault detection in wireless sensor networksrdquo IEEETransactions on Instrumentation and Measurement vol 61 no6 pp 1703ndash1718 2012

[27] R Huang X Qiu and L Rui ldquoSimple random sampling-basedprobe station selection for fault detection in wireless sensornetworksrdquo Sensors vol 11 no 3 pp 3117ndash3134 2011

[28] J Medina-Garcıa T Sanchez-Rodrıguez J Galan A DelgadoF Gomez-Bravo and R Jimenez ldquoA wireless sensor systemfor real-time monitoring and fault detection of motor arraysrdquoSensors vol 17 no 3 p 469 2017

[29] T Muhammed and R A Shaikh ldquoAn analysis of fault detectionstrategies in wireless sensor networksrdquo Journal of Network andComputer Applications vol 78 pp 267ndash287 2017

[30] H Artail A Ajami T Saouma and M Charaf ldquoA faultynode detection scheme for wireless sensor networks that usedata aggregation for transportrdquo Wireless Communications andMobile Computing vol 16 no 14 pp 1956ndash1971 2016

[31] M Panda and P M Khilar ldquoDistributed Byzantine fault detec-tion technique in wireless sensor networks based on hypothesistestingrdquo Computers and Electrical Engineering vol 48 pp 270ndash285 2015

[32] P Jiang ldquoA new method for node fault detection in wirelesssensor networksrdquo Sensors vol 9 no 2 pp 1282ndash1294 2009

[33] K P Sharma and T P Sharma ldquorDFD reactive distributed faultdetection in wireless sensor networksrdquo Wireless Networks vol23 no 4 pp 1145ndash1160 2017

[34] M Bo H Darong and W Shaohua ldquoNTRU implementa-tion of efficient privacy-preserving location-based querying inVANETrdquoWireless Communications and Mobile Computing vol2018 Article ID 7823979 11 pages 2018

[35] Y Yang Z Gao H Zhou and X Qiu ldquoAn uncertainty-based distributed fault detectionmechanism for wireless sensornetworksrdquo Sensors vol 14 no 5 pp 7655ndash7683 2014

[36] D Wang S Wan and N Guizani ldquoContext-based probabilityneural network classifiers realized by genetic optimization formedical decision makingrdquo Multimedia Tools and Applicationsvol 77 no 17 pp 21995ndash22006 2018

Wireless Communications and Mobile Computing 7

International Journal of Distributed Sensor Networks vol 2014Article ID 321964 16 pages 2014

[5] J Kong J-H Cui D Wu and M Gerla ldquoBuilding underwaterad-hoc networks and sensor networks for large scale real-timeaquatic applicationsrdquo inProceedings of theMilitary Communica-tions Conference (MILCOM rsquo05) pp 1535ndash1541 October 2005

[6] N Xu S Rangwala K K Chintalapudi et al ldquoA wireless sensornetwork for structural monitoringrdquo in Proceedings of the 2ndInternational Conference on Embedded Networked Sensor Sys-tems (SenSys rsquo04) pp 13ndash24 November 2004 (Catalan)

[7] Z You X ZhaoHWanWNNHung YWang andMGu ldquoAnovel fault diagnosis mechanism for wireless sensor networksrdquoMathematical and Computer Modelling vol 54 no 1-2 pp 330ndash343 2011

[8] S Rajasegarar C Leckie M Palaniswami and J C BezdekldquoDistributed anomaly detection in wireless sensor networksrdquo inProceedings of the 10th IEEE Singapore International Conferenceon Communication systems (ICCS rsquo06) pp 1ndash5 IEEE October2006

[9] M Ding D Chen K Xing and X Cheng ldquoLocalized fault-tolerant event boundary detection in sensor networksrdquo inProceedings of the IEEE 24th Annual Joint Conference of the IEEEComputer and Communications Societies (INFOCOM rsquo05) vol2 pp 902ndash913 March 2005

[10] D I Curiac and C Volosencu ldquoEnsemble based sensinganomaly detection in wireless sensor networksrdquo Expert Systemswith Applications vol 39 no 10 pp 9087ndash9096 2012

[11] M Ur-Rehman N A Malik X Yang Q H Abbasi Z Zhangand N Zhao ldquoA low profile antenna for millimeter-wavebody-centric applicationsrdquo IEEE Transactions on Antennas andPropagation vol 65 no 12 pp 6329ndash6337 2017

[12] C Wang H Lin and H Jiang ldquoCANS Towards congestion-adaptive and small stretch emergency navigation with wirelesssensor networksrdquo IEEE Transactions on Mobile Computing vol15 no 5 pp 1077ndash1089 2016

[13] J Wen B Zhou W H Mow and X-W Chang ldquoAn efficientalgorithm for optimally solving a shortest vector problem incompute-and-forward designrdquo IEEE Transactions on WirelessCommunications vol 15 no 10 pp 6541ndash6555 2016

[14] J Wen J Wang and Q Zhang ldquoNearly optimal bounds fororthogonal least squaresrdquo IEEE Transactions on Signal Process-ing vol 65 no 20 pp 5347ndash5356 2017

[15] J Wen Z Zhou Z Liu M-J Lai and X Tang ldquoSharp sufficientconditions for stable recovery of block sparse signals by blockorthogonal matching pursuitrdquo 2016 httpsarxivorgabs160502894

[16] S Rani S H Ahmed R Talwar and J Malhotra ldquoCan sensorscollect big data an energy-efficient big data gathering algo-rithm for a WSNrdquo IEEE Transactions on Industrial Informaticsvol 13 no 4 pp 1961ndash1968 2017

[17] J Chen S Kher and A Somani ldquoDistributed fault detectionof wireless sensor networksrdquo in Proceedings of the Workshop onDependability Issues in Wireless Ad Hoc Networks and SensorNetworks pp 65ndash72 2006

[18] S Rani S H Ahmed J Malhotra and R Talwar ldquoEnergyefficient chain based routing protocol for underwater wirelesssensor networksrdquo Journal of Network and Computer Applica-tions vol 92 pp 42ndash50 2017

[19] D Li and J Zhang ldquoEfficient implementation to numericallysolve the nonlinear time fractional parabolic problems onunbounded spatial domainrdquo Journal of Computational physicsvol 322 pp 415ndash428 2016

[20] A Munir J Antoon and A Gordon-Ross ldquoModeling andanalysis of fault detection and fault tolerance in wireless sensornetworksrdquoACMTransactions on EmbeddedComputing Systemsvol 14 no 1 article 3 2015

[21] G S Brar S Rani V Chopra R Malhotra H Song and SH Ahmed ldquoEnergy efficient direction-based PDORP routingprotocol for WSNrdquo IEEE Access vol 4 pp 3182ndash3194 2016

[22] ANAlvi SH Bouk SHAhmedMA YaqubM Sarkar andH Song ldquoBEST-MAC Bitmap-Assisted Efficient and ScalableTDMA-Based WSN MAC Protocol for Smart Citiesrdquo IEEEAccess vol 4 pp 312ndash322 2016

[23] S Rani and S H AhmedMulti-Hop Routing in Wireless SensorNetworks An Overview Taxonomy and Research ChallengesSpringer 2015

[24] S Rani R Talwar J Malhotra S H Ahmed M Sarkar and HSong ldquoA novel scheme for an energy efficient internet of thingsbased on wireless sensor networksrdquo Sensors vol 15 no 11 pp28603ndash28626 2015

[25] E Ould-Ahmed-Vall B H Ferri and G F Riley ldquoDistributedfault-tolerance for event detection using heterogeneouswirelesssensor networksrdquo IEEE Transactions on Mobile Computing vol11 no 12 pp 1994ndash2007 2012

[26] S C Chan H C Wu and K M Tsui ldquoRobust recursiveeigendecomposition and subspace-based algorithmswith appli-cation to fault detection in wireless sensor networksrdquo IEEETransactions on Instrumentation and Measurement vol 61 no6 pp 1703ndash1718 2012

[27] R Huang X Qiu and L Rui ldquoSimple random sampling-basedprobe station selection for fault detection in wireless sensornetworksrdquo Sensors vol 11 no 3 pp 3117ndash3134 2011

[28] J Medina-Garcıa T Sanchez-Rodrıguez J Galan A DelgadoF Gomez-Bravo and R Jimenez ldquoA wireless sensor systemfor real-time monitoring and fault detection of motor arraysrdquoSensors vol 17 no 3 p 469 2017

[29] T Muhammed and R A Shaikh ldquoAn analysis of fault detectionstrategies in wireless sensor networksrdquo Journal of Network andComputer Applications vol 78 pp 267ndash287 2017

[30] H Artail A Ajami T Saouma and M Charaf ldquoA faultynode detection scheme for wireless sensor networks that usedata aggregation for transportrdquo Wireless Communications andMobile Computing vol 16 no 14 pp 1956ndash1971 2016

[31] M Panda and P M Khilar ldquoDistributed Byzantine fault detec-tion technique in wireless sensor networks based on hypothesistestingrdquo Computers and Electrical Engineering vol 48 pp 270ndash285 2015

[32] P Jiang ldquoA new method for node fault detection in wirelesssensor networksrdquo Sensors vol 9 no 2 pp 1282ndash1294 2009

[33] K P Sharma and T P Sharma ldquorDFD reactive distributed faultdetection in wireless sensor networksrdquo Wireless Networks vol23 no 4 pp 1145ndash1160 2017

[34] M Bo H Darong and W Shaohua ldquoNTRU implementa-tion of efficient privacy-preserving location-based querying inVANETrdquoWireless Communications and Mobile Computing vol2018 Article ID 7823979 11 pages 2018

[35] Y Yang Z Gao H Zhou and X Qiu ldquoAn uncertainty-based distributed fault detectionmechanism for wireless sensornetworksrdquo Sensors vol 14 no 5 pp 7655ndash7683 2014

[36] D Wang S Wan and N Guizani ldquoContext-based probabilityneural network classifiers realized by genetic optimization formedical decision makingrdquo Multimedia Tools and Applicationsvol 77 no 17 pp 21995ndash22006 2018

8 Wireless Communications and Mobile Computing

[37] H Saeedi Emadi and S M Mazinani ldquoA novel anomalydetection algorithm using DBSCAN and SVM in wirelesssensor networksrdquo Wireless Personal Communications vol 98no 2 pp 2025ndash2035 2018

[38] T Qiu A Zhao F Xia W Si and D O Wu ldquoROSE robustnessstrategy for scale-free wireless sensor networksrdquo IEEEACMTransactions on Networking vol 25 no 5 pp 2944ndash2959 2017

[39] T Qiu R Qiao andD OWu ldquoEabs An event-aware backpres-sure scheduling scheme for emergency internet of thingsrdquo IEEETransactions on Mobile Computing no 1 pp 72ndash84 2018

[40] S Rajasegarar C Leckie J C Bezdek and M PalaniswamildquoCentered hyperspherical and hyperellipsoidal one-class sup-port vector machines for anomaly detection in sensor net-worksrdquo IEEE Transactions on Information Forensics and Secu-rity vol 5 no 3 pp 518ndash533 2010

[41] D M J Tax and R P W Duin ldquoSupport vector domaindescriptionrdquo Pattern Recognition Letters vol 20 no 11ndash13 pp1191ndash1199 1999

[42] B Scholkopf A Smola and K-R Muller ldquoNonlinear compo-nent analysis as a kernel eigenvalue problemrdquoNeural Computa-tion vol 10 no 5 pp 1299ndash1319 1998

[43] S Madden ldquoIntel lab datardquo Web page Intel 2004

International Journal of

AerospaceEngineeringHindawiwwwhindawicom Volume 2018

RoboticsJournal of

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

Active and Passive Electronic Components

VLSI Design

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

Shock and Vibration

Hindawiwwwhindawicom Volume 2018

Civil EngineeringAdvances in

Acoustics and VibrationAdvances in

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

Electrical and Computer Engineering

Journal of

Advances inOptoElectronics

Hindawiwwwhindawicom

Volume 2018

Hindawi Publishing Corporation httpwwwhindawicom Volume 2013Hindawiwwwhindawicom

The Scientific World Journal

Volume 2018

Control Scienceand Engineering

Journal of

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom

Journal ofEngineeringVolume 2018

SensorsJournal of

Hindawiwwwhindawicom Volume 2018

International Journal of

RotatingMachinery

Hindawiwwwhindawicom Volume 2018

Modelling ampSimulationin EngineeringHindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

Chemical EngineeringInternational Journal of Antennas and

Propagation

International Journal of

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

Navigation and Observation

International Journal of

Hindawi

wwwhindawicom Volume 2018

Advances in

Multimedia

Submit your manuscripts atwwwhindawicom

International Journal of

AerospaceEngineeringHindawiwwwhindawicom Volume 2018

RoboticsJournal of

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

Active and Passive Electronic Components

VLSI Design

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

Shock and Vibration

Hindawiwwwhindawicom Volume 2018

Civil EngineeringAdvances in

Acoustics and VibrationAdvances in

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

Electrical and Computer Engineering

Journal of

Advances inOptoElectronics

Hindawiwwwhindawicom

Volume 2018

Hindawi Publishing Corporation httpwwwhindawicom Volume 2013Hindawiwwwhindawicom

The Scientific World Journal

Volume 2018

Control Scienceand Engineering

Journal of

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom

Journal ofEngineeringVolume 2018

SensorsJournal of

Hindawiwwwhindawicom Volume 2018

International Journal of

RotatingMachinery

Hindawiwwwhindawicom Volume 2018

Modelling ampSimulationin EngineeringHindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

Chemical EngineeringInternational Journal of Antennas and

Propagation

International Journal of

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

Navigation and Observation

International Journal of

Hindawi

wwwhindawicom Volume 2018

Advances in

Multimedia

Submit your manuscripts atwwwhindawicom