Document6

632 IEEE TRANSACTIONS ON SMART GRID, VOL. 5, NO. 2, MARCH 2014

Location-Based Forecasting of Vehicular ChargingLoad on the Distribution System

Nima Ghiasnezhad Omran, Student Member, IEEE, and Shaahin Filizadeh, Senior Member, IEEE

Abstract—This paper presents a procedure for location-basedforecasting of the potential vehicular charging load at off-homecharging stations. A location-specific fuzzy decisionmaking systemis proposed to characterize the charging behavior and determinethe probability of charging by means of a three dimensional inputobtained from a real-world driving dataset. The obtained chargingand parking probability figures are then used in prediction of thelocal aggregated charging demand. The flexibility and usefulnessof the developed procedure is exemplified in the case studies of twomajor shopping centers.

Index Terms—Fuzzy decision making, location-based predic-tion, off-home charging station, vehicular charging demand.

I. INTRODUCTION

V EHICULAR charging load on the grid is expected to riseas plug-in (hybrid) electric vehicles (PEVs) experience

rapid improvements, which will lead to their increased perfor-mance, lower price and hence further public acceptance [1]–[3].In the context of the future smart grid, grid-connected vehicularstorage systems are envisioned not only to require energy, butalso to be potentially contributing to the supply of energy duringperiods of high demand [3]–[5].Estimation of the aggregated vehicular demand on a specific

power system helps in the planning of generation to keep pacewith the growth of the load as PEVs gain larger market shares[5]–[7]. It must, however, be noted that vehicular charging maylead to severe issues at the distribution level even at low marketpenetration levels and prior to manifesting any major problemson the generation or transmission level [5]–[9]. Thus the dis-tributed nature of the charging load and its localized impact onthe assets of the grid must be evaluated in the planning of thedistribution system.In order to investigate the intensity of these potential issues, a

reliable prediction of the profile of the charging demand on thedistribution network is essential. Conventional load forecastingmethods in power systems for different time horizons employvarious statistical and artificial intelligence techniques (e.g.,

Manuscript received March 12, 2013; revised June 21, 2013 and August 21,2013; accepted September 17, 2013. Date of publication December 05, 2013;date of current version February 14, 2014. Financial support for this work wasprovided by the Natural Sciences and Engineering Research Council (NSERC)of Canada. Paper no. TSG-00212-2013.The authors are with the Department of Electrical and Computer Engi-

neering, University of Manitoba, Winnipeg, MB, R3T 5V6, Canada (e-mail:[email protected]; [email protected]).Color versions of one or more of the figures in this paper are available online

at http://ieeexplore.ieee.org.Digital Object Identifier 10.1109/TSG.2013.2282773

time series, multiple regression, expert systems, fuzzy logic,neural network, etc.) [10]–[15]. These methods mostly usehistorical data along with influential parameters on the demandprofile (e.g., climatic conditions, day-type, land usage, etc.).However, due to lack of actual comprehensive historical data,efforts for prediction of the charging demand of the upcomingvehicular load have been mostly done through defining optimalcharging scenarios from the utility and the customers’ points ofview [16]–[19]. Generally, realizing temporal and spatial be-havior of the charging demand on the power system depends onsuch factors as the convenience and cost of charging, adequateparking time, availability of charging stations and their rating,and most importantly the charging need of PEVs [20], [21]. Itis also reasonable to assume that the society will not developradically different transportation habits based on the chargingrequirements of the PEVs. In fact, the PEV technologies andthe utility planners may aim to capture the existing drivingtraits of the society and enhance their services in that direction.Therefore, in the absence of actual measured data of the

charging demand use of real-world driving data in the areaof interest provides a realistic picture of driving habits andcharacteristics [21]–[25]. Layout of roads and traffic patternsinfluence not only the driving traits but also vehicular energyconsumption, which together contribute to timing and distri-bution of charging demand. Local driving data also providesinformation about potential parking locations where futurePEVs may be connected to the network for charging.While PEVs are not widely adopted in a given jurisdiction,

it is reasonable to assume that home plugging will be thedominant (or in most cases perhaps the only) mode of charging[26]. As the penetration-level of PEVs increases additionaloff-home charging stations should be gradually assigned instrategic locations in order to fulfill the charging demand ofon-road PEVs. Places where a large number of parking eventsoccur (e.g., shopping centers) will obviously be more subjectedto vehicular charging, so they can be considered as main can-didates for future charging stations and necessary fortificationsof the network.This paper investigates the potential charging demand in

such high-density off-home parking locations, while consid-ering home charging as the most favorable charging mode.Using real-world driving data, a location-based fuzzy deci-sion-making algorithm (Sections II and III) is implementedto predict the probability of charging for a given parkingevent. Model sensitivity is then analyzed (Section IV), andthe aggregated potential charging load for two major shoppingcenters in the city of Winnipeg [27] is investigated as casestudy (Sections V and VI).

1949-3053 © 2013 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

GHIASNEZHAD OMRAN AND FILIZADEH: LOCATION-BASED FORECASTING OF VEHICULAR CHARGING LOAD ON THE DISTRIBUTION SYSTEM 633

Fig. 1. Flowchart of the location-based vehicular charging load forecasting procedure.

II. LOCATION-BASED VEHICULAR LOAD PREDICTION

Location-based study of vehicular load aims to predict the po-tential load demand due to PEV charging at off-home parkinglocations. Whether or not drivers decide to charge their vehiclesin a specific place depends on various parameters; however, hu-mans’ decision making process for cases like this often does notinvolve precise computations or analyses. Indeed, drivers usu-ally evaluate the situation using their own experience, conve-nience, and other factors, most importantly economic consider-ations.The essence of a fuzzy inference engine is to quantify a com-

plex process, e.g., charging decision making, through reason-able assumptions that capture the experience of the expert (i.e.,the driver). A fuzzy inference system was proposed in [25] tosimulate a driver’s decision-making process for PEV charging.It used the present state-of-charge (SOC) and parking duration(PD) as its inputs to predict the probability of charging, inde-pendent from the actual place of parking. However, in the loca-tion-based prediction method presented in this paper, the actualdriving distance to home (DTH) is also used as an additionalinput. This is due to the fact that most drivers would prefer tocharge their vehicles at home rather than at other locations. Sev-eral factors contribute to this preference including longer du-ration of vehicle down-time at home (e.g., overnight), and po-tentially less expensive electricity. The developed fuzzy infer-ence system takes this preference into account but also includesprovisions for the case when there is doubt whether the currentstate of charge of the battery is adequate for the drive to home,through proper rules.Fig. 1 illustrates the flowchart of the proposed procedure

for location-based vehicular load prediction. It starts with ananalysis of a real-world driving dataset for the area of interest,and extracts characteristics of recorded parking events. Then,a fuzzy decision-making engine uses the extracted statisticalresults to predict the average probability of charging for everyhour at the location of interest. This, together with local parkingcharacteristics and market information, determines the expectedcharging demand. The following sections present details of theprocedure.

III. LOCATION-BASED FUZZY DECISION-MAKING UNIT

The three input parameters of SOC, PD, and DTH need to beexpressed with linguistic terms and their corresponding mem-bership functions. Once the inputs are fuzzified, the rules of thefuzzy system are applied to generate the respective outputs. Theoutputs are then combined and defuzzified to yield the output ofthe fuzzy inference engine, which is the probability of charging.The following subsections present details of the membershipfunctions assigned to each input and output variable [28], [29].

A. State of Charge (SOC)

The SOC is the most readily available input to the driver. Itindicates the amount of stored electrical energy that is presentlyavailable. Fig. 2(a) shows membership functions created for theSOC. Three linguistic terms, i.e., Low, Medium, and High, aredefined to cover the whole working area of the battery storage.The on-board vehicle controller only operates the battery withinlower and upper bounds, to ensure its longevity. In the studypresented here a range of [15,85]% is considered (correspondingto a 70% depth of discharge); that is why the Low and Highmembership functions shown in Fig. 2(a) attain a value of 1.0below 15% and over 85% SOC, respectively.

B. Parking Duration (PD)

Parking duration is a variable that represents the antic-ipated length of the parking event. Note that most driversdo not have an exact length of time for parking at the onsetof a parking event. Therefore, it is assumed that they ex-press the parking duration with three linguistic terms (Short,Average, Long) as shown in Fig. 2(b). For example, themembership function labeled “Average” fully encompasses

all parking events rangingfrom around 45 minutes to around 3 hours. Therefore, anyparking with a duration in this interval will be equally treated;this makes the model less sensitive to the deviation between theactual length of the parking event and its anticipated durationby the driver.The linguistic terms for the PD input are highly affected by

the available level of charging. This is because the amount ofcharge that can be replenished over the period of charging is


Fig. 2. Membership functions of the input variables. (a) state-of-charge (SOC);(b) parking duration (PD); (c) normalized driving distance to home (DTH).

directly determined by ratings of the charger. For example aparking event whose duration may be considered “Short” (andhence not worthwhile) for regular charging may indeed be con-sidered “Average” if fast-charging facilities exist. The member-ship functions in Fig. 2(b) are designed with the assumption oflevel-1 charging (120-V, 15-A).

C. Actual Driving Distance to Home (DTH)

By experience a prudent driver of a plug-in vehicle mayhave a reasonable estimation of how much SOC is required todrive the vehicle to a convenient charging location, i.e., homein this study. However, knowledge of the required SOC maybe a higher-than-normal expectation for an average driver andinvolves several uncertainties. On the other hand, an estimateof the driving distance to home is doable for most drivers.Therefore, in the fuzzy inference system described here theactual driving distance to home is used as a representative ofthe required SOC to drive home.The DTH assigned to a parking event includes the mileage

of all the subsequent trips during the day before arriving homefor overnight parking. Indeed, the DTH is a factor that enhancesthe precision of the charging behavior prediction significantly:it shows the experience of the driver about daily trips, and their

TABLE IRULES OF THE FUZZY SYSTEM

timing and mileage (i.e., the expert knowledge). It also con-siders the location of each parking event, thereby making thedecision-making process location-specific.The DTH input is categorized under three linguistic terms of

Short, Average, and Long [Fig. 2(c)]. Note that the battery ca-pacity plays an important role in the definition of the range forthese linguistic terms. A battery with a higher capacity allows alonger all-electric range. To avoid development of membershipfunctions for the DTH for vehicles with different battery capac-ities, a normalized figure (based on the maximum all-electricrange) for the DTH is used.In reality the all-electric range of a vehicle will depend on

such factors as the age of the battery, driving patterns of thedriver, and traffic, among other things (see Section III-E for fur-ther discussion); thus the actual all-electric range will likely beless than nominal.

D. Rule Table and Defuzzification

A set of 25 rules for charging decision-making are devel-oped for a Mamdani-type fuzzy model [29] as shown in Table I.These rules act on the three inputs of the fuzzy system to pro-duce outputs that are then aggregated and defuzzified to yieldthe probability of charging for a specific parking event. Beforedefuzzification, the probability of charging is described usingseven linguistic terms as shown in Fig. 3. The fuzzy “AND” isimplemented using the “min” operator, and the center-of-massoperation is used for defuzzification.The design of the rules in this fuzzy system is done with a

view to maintain reasonability of the assumption from a pru-dent driver’s point-of-view. The main consideration in the de-


Fig. 3. Probability of charging (fuzzy system output).

sign of the rule table is that if the SOC of the battery is less thanthe driver’s estimation of the required SOC to drive the vehiclehome (based on the DTH), the probability of charging increaseswith respect to the parking duration.It should further be noted that when the DTH is “Long”

both the “Medium” and “Low” SOC are treated similarly;this reflects the fact that a “Long” DTH most likely repre-sents a parking event in the earlier hours, and from a driver’spoint-of-view foreseeing all future trips throughout the rest ofthe day might not be possible; therefore an increased chance ofcharging is given to the “Medium” SOC range.

E. Additional Factors

1) Climatic Conditions and Aging: It must be noted thatother factors than the ones considered here may impact adriver’s decision to plug in for charging. Severe climaticconditions (e.g., extreme heat or cold, humidity, etc.) not onlyaffect the driver’s decision making, but may also affect thethree considered inputs. For example, temperature variationsdo impact the chemical reactions within a battery and henceits SOC, and its total capacity (affecting the DTH), and alsothe required heating/cooling energy. Although there are studiesthat aim to approximate such variations [30], [31], the level ofdetail and complexity required by these methods renders theminfeasible for aggregated long-term vehicular load forecasting,which also involve a large number of storage units with dif-ferent properties. Therefore, instead of attempting to augmentthe model to directly include such auxiliary effects, the paperinvestigates the sensitivity of predicted probability of charging(output of the fuzzy model) as well as final forecasted load (forthe presented case study) to uncertainties in its input variablescollectively caused by such factors as seasonal temperaturevariations, humidity, aging, measurement errors, etc. Theseanalyses are shown in Sections IV (for the model) and VI (forthe presented case study).2) Utility Tariffs: The DTH input, which is the preference

factor for the location of charging, accounts for the driver’s de-sire for the least expensive transportation and convenience ofcharging. In the design of the fuzzy rule table two cost effec-tiveness objectives are considered as follows:a) consumption of electricity has preference over gas be-cause of the lower cost of electricity;

b) home-charging has preference over off-home chargingbecause of lower cost (as well as convenience);

Although the convenience of charging is not quantitativelycountable, it is possible to include the effect of electricity cost

variation by only modifying the DTH input without any fur-ther change in the kernel of the developed model. A favorablecharging rate at the time of parking will entice the driver tocharge at an off-home location, and this can be captured bysuitably increasing the DTH input. Conversely a decrease inthe DTH resembles an unfavorable charging rate, and hence adriver’s inclination to charge at home. For example, these canbe due to a utility’s time-variant tariffs for peak and off-peakhours, and/or incentives offered for off-home charging. How-ever, it should be noted that as long as the two said cost objec-tives of the model are valid the change in the DTH is expectedto be small; i.e., when the price increases but stays below theequivalent gas price, or when the price decreases but remainsmore expensive than the convenience threshold over which adriver prefers to charge at home.The investigation of the effect of price variation is mostly

useful when a more accurate forecast of charging demand isrequired for short term, and exact charging costs in each locationand for different time are available.

IV. MODEL PERFORMANCE ASSESSMENT

The developed fuzzymodel is designed to capture the driver’sdecision making process through a reasonable set of rules actingon the three inputs. The adoptedmembership functions and ruleswill certainly have an impact on the output of the fuzzy system,i.e., the predicted probability of charging.Fig. 4 shows the probability of charging for three represen-

tative values of the PD input (30 min, 120 min, and 240 min),while the SOC and the DTH vary within their respective ranges.In all three surfaces, the probability of charging increases grad-ually starting from maximum SOC and minimum DTH. Fur-thermore the figures show an increasing probability of chargingas the PD increases, as noted by the top left corner of eachfigure where the probability of charging increases from 60% for

to essentially 100% for . Theseare reasonable expectations, which are satisfied through the se-lection and composition of the rules.In order to assess the performance of the fuzzy model an anal-

ysis of the sensitivity of its output is undertaken with respect tovariations of the inputs. This analysis quantifies the expecteddeviation of the output when inevitable uncertainties occur inthe estimation of the inputs. These uncertainties may arise dueto factors such as the ones discussed in Section III-E.Table II presents the results of sensitivity analysis for indi-

vidual variations of 20% in each input. For example the tableshows that the average change in the probability of charging(predicted by the fuzzy model) is 5.70% when the SOC inputdecreases by 20%.The table shows that the output is most sensitive to the DTH

input, which encapsulates cost effectiveness and convenienceof charging (home preference factor). The SOC, which capturesthe available charge factor, is the second most influential input;and the least sensitivity belongs to the PD input, which quanti-fies the worthiness of the available parking time for charging.The analysis presented in this section shows that the combi-

nation of the membership functions and the rules do indeed, andas intended, lead to make the probability of charging more re-sponsive to the inputs with a higher importance, i.e., the DTH


Fig. 4. Fuzzy decision making surfaces for three representative PDs. (a); (b) ; (c) .

TABLE IISENSITIVITY ANALYSIS RESULTS OF THE FUZZY MODEL

and the SOC. It further shows that the robustness of the fuzzymodel’s output to the variations of the inputs.

V. DATA ATTRIBUTES AND SUBSYSTEM MODELS

In the following two sections, a set of real-world driving data(for the city of Winnipeg) and the vehicular model used are pre-sented.

A. Recorded Driving Data

Real-world driving data collected from 74 conventional vehi-cles inWinnipeg [27] is used in the analyses shown in this paper.This set includes data from participants from different areas ofthe city with diverse demographic characteristics (income, age,

gender, etc.). Instantaneous latitude, longitude, and speed of thevehicle in each trip as well as their time and date were recorded.As indicated earlier, two major shopping centers in Winnipeg

are considered. All recorded parking events in these two loca-tions are extracted from the dataset and their statistical charac-teristics (e.g., parking time, duration, density, etc.) are derivedand used as attributes of parking events in these shopping cen-ters. The large number of parking events recorded (from dif-ferent participants) in each of the two locations provides confi-dence about the validity of the data as an indicator of the entirepopulation [32].

B. Vehicle Subsystem Models

A backward vehicular model [2] is developed to calculate themechanical energy ( in Joules) used by the vehicle in eachtrip (defined as the distance travelled between any two consec-utive stops). This model is based on Newton’s second law ofmotion and is shown in (1)–(4) below.

(1)

(2)

(3)

and

(4)

where , , and are the instantaneous mechanical power(W), speed (m/s), and propulsion force (N), respectively. ,

, and (all in N) are the aerodynamic drag, rollingforce, and grading resistance, respectively. is the air density

, is the frontal area of the vehicle , is theaerodynamic drag coefficient, is the tailwind speed (m/s),is the road grade, and is the vehicle mass (kg). In the

simulations presented in Section V, the road grade and the windspeed are set to zero.The total required electrical energy from the battery can be

calculated using the consumed mechanical energy, the contri-bution of regenerative braking, and the efficiencies of differentdrive train components, as shown in (5).

(5)

where is the total electrical energy (from the battery) andis the total regenerative energy (mechanical) during a trip.

, , , and are efficiencies of the vehicle transmis-sion system, generator, motor, and regenerative braking systemrespectively. is the amount of energy consumed forheating or cooling (air-conditioning load) the vehicle cabin ina trip, which will depend, at least partly, on the ambient tem-perature . Note that other factors such as a driver’s choice andthe action of the vehicle controller may impact the amount ofheating/cooling power consumed.The change in the electrical energy manifests itself as vari-

ations of the SOC of the battery. It is, therefore, necessary to


determine the battery SOC given the electrical energy transac-tions. A simplified expression [33] for calculating the SOC isgiven a follows in (6).

(6)

where is the state of charge at the beginning of the trip,is the nominal terminal voltage of the battery (V), and is

the capacity of battery (Ah). To account for the losses that occurduring grid charging (ac-dc converter, plug, etc.) an efficiencyfigure of 90% is applied to the drawn power.

VI. SIMULATION RESULTS

The simulation results in this section show the probability ofcharging for both weekday and weekend in the two shoppingcenters. As an example of the load forecasting procedure shownin Fig. 1, the expected vehicular load in one of the locations isshown. The simulation results also include level-1 and level-2charging scenarios.

A. Simulation Setup and Vehicle Specifications

In the simulations presented it is assumed that vehicles leavehome fully charged (with ). This is because thetypically long overnight downtime of the vehicle is adequateto fully charge its battery. It also conforms to the underlyingassumption that home is the preferred location for charging.During the daily trips, the SOC of the battery may decline downto a minimum of 15%. Although charging may be availableto some PEV owners at other places (such as work place) tocreate the worst-case scenario no charging is considered in otheroff-home locations prior to arriving in the locations of interest.Equations (1)–(5) show that specifications of a vehicle have

significant effect on its required energy. In relation with the re-quired energy, the battery capacity determines the variations ofthe SOC. The battery capacity also directly impacts the all-elec-tric range, which is a determining factor in the DTH input to thefuzzy system.In this study three types of plug-in vehicles, namely the

Toyota Prius plug-in hybrid, the Chevrolet Volt, and the NissanLeaf, are considered. The Prius and the Volt are representativesof PHEVs with light-duty and heavy-duty battery storage,respectively. Nissan Leaf represents EVs with higher capacityof battery storage (in comparison with PHEVs, which have theoption of switching to gas).Some adjustments in the developed decision making proce-

dure are required to meet the special conditions of Nissan Leaf(or other battery electric vehicles, if considered). i.e., it is as-sumed that if there is doubt whether or not the remaining SOCis adequate to drive the vehicle home, the driver will have tocharge the battery. This primarily affects conditions when theSOC is “Low” or “Medium” and the DTH is “Long.”Table III shows the specifications of the three vehicles consid-

ered in the simulation, as well as the efficiencies for drive traincomponents. Although the efficiency of drive train componentsdo vary depending to the operating conditions, assumption ofconstant efficiency figures is commonly made in high-level ve-hicular studies [34], [35], and is therefore adopted here as well.A constant value of 500 W is used to approximately rep-

resent the heating/cooling power or additional electrical loads

TABLE IIIVEHICLE AND DRIVETRAIN SPECIFICATIONS

onboard. Note that an accurate characterization of the actualheating/cooling power requires data that is not reliably quan-tifiable, and also that the is only a small portion of theconsumed energy during a trip and hence the impact of its vari-ations on the SOC are small. The analysis in Section IV showedthat the probability of charging has only modest sensitivity toSOC variations, and hence the use of a constant value will notbe detrimental to the validity of the results shown.

B. Weekday/Weekend Probability of Charging and Load

The SOC, PD, and DTH attributes for every parking eventat the selected shopping centers are given to the fuzzy systemto generate a probability of charging. The value of probabilityfor each parking event contributes to the average probability forthe hour the parking occurs. If parking continues to the nexthour(s), the same probability of charging will be carried to thenext hour(s) as long as the battery is not fully charged.Figs. 5 and 6 show the predicted average probability of

charging in every hour for the two locations during a typicalweekday and weekend, respectively. The value of probabilityin each hour indicates the percentage of the vehicles that areparked at the given hour and will charge. The zero averageprobability for some hours is due to the lack of adequate samplepoints in such hours.As shown in Figs. 5 and 6, the value of probability is mostly

dependent on the type of vehicle (i.e., the capacity of theirbattery storage). Probability of charging for light-duty batterystorage (e.g., Prius plug-in) is much higher than the othertwo. It is due to that in most cases such vehicles arrive withnearly depleted battery. The substantially lower probabilityof charging for the Volt and Leaf at these off-home locationsis an indication that the daily mileage of these large-capacityvehicles (prior to arriving in the shopping centers) is likelyto be much less than their all-electric range; this implies thatthese vehicles will likely receive the main part of their chargingdemand during home plugging.Apart from the probability for charging, for prediction of the

potential peak load (kW) and energy demand (kWh), the numberof the parking events that occur during the day, their duration,and distribution among different hours must also be known. Theprobability of parking curves for shopping center 1 for boththe weekday and weekend are extracted from the dataset andis shown in Fig. 7. The procedure for doing so along with theprobability of parking curves for shopping center 2 is given in[32].As an example of the prediction process, the potential peak

load is calculated at shopping center 1 for both weekday and


Fig. 5. Average charging probability for weekday (level 1 charging); (a) shop-ping center 1, (b) shopping center 2.

Fig. 6. Average charging probability for weekend (level 1 charging); (a) shop-ping center 1, (b) shopping center 2.

weekend. This example shows how the probability of chargingcurves in Figs. 5 and 6 enable calculation of the peak load

Fig. 7. Probability of parking for shopping center 1 for weekday and weekend.

for any given number of parking events, types of vehicles,and market penetration levels of PEVs. In particular, with anassumption of 20% penetration level of PEVs, and an averagedaily number of parking events in shopping center 1 equal to10 000 it is expected that 2000 PEVs will arrive and park inthis shopping center each day. As seen in Fig. 7, at 4 P.M., theprobability of parking is equal to 9% for a weekday, implyingthat of the PEVs are expected to be parkedat this particular hour. It is further assumed that of the PEVs,35% are light-duty (similar to Prius), 35% are heavy-duty(similar to Volt), and 30% are all-electric (similar to Leaf). Thenumber of PEV types as per the assumed breakdown will there-fore be 63 Priuses, 63 Volts and 54 Leafs. The probabilities ofcharging at 4 P.M. for each PEV type is then read from Fig. 5(a)as 47% for the Prius, 20% for the Volt, and 13% for the Leaf.With a nominal power of 1.8 kW for level-1 charging (withunity power factor), this results in an expected charging load of

.Similar calculations can be repeated for other hours.The expected peak charging power as a function of time for

shopping center 1 for both weekday and weekends are shownin Figs. 8 and 9 (the traces labeled “expected”) for the assumedpenetration level and breakdown of PEV types. The procedurecan be easily used to consider other PEVs, other penetrationlevels, or other composition of PEV types according to marketacceptance of specific vehicle types in different jurisdictions.The procedure can also be adopted to analyze the charging loadfor each individual day of the week if adequate samples areavailable.Figs. 8 and 9 each also show two additional traces labeled as

“lower” and “upper” around the “expected” peak power. Thesecurves show the average upper and lower bands of the expectedcharging power for a given level of uncertainty in the input pa-rameters. As mentioned in Sections III-E and IV factors suchas temperature variations, aging, measurement error, etc., maycontribute to uncertainty in the inputs to the fuzzy engine andthereby impact the predicated charging load. In this section si-multaneous random variations in the inputs around their nom-inal values for each parking event are considered and their col-lective impact on the charging load is evaluated. In particular,for every parking event a random change of up to 20% is ap-plied to each input with a directional consideration for the SOCand DTH inputs. In particular two sets of experiments are con-ducted; the first set considers random changes in the positive


Fig. 8. Peak power at shopping center 1 during a weekday (level 1 charging).

Fig. 9. Peak power at shopping center 1 for weekend (level 1 charging).

direction for the SOC and negative direction for the DTH input(e.g., resulting from a chance of charging before or after theevent), yielding the “lower” probability of charging. A secondset of experiments is done with negative changes in the SOC andpositive changes in the DTH (e.g., due to severely cold temper-atures), leading to the “upper” probability of charging. The rel-atively tight placement of the two bands around the “expected”trace is an indication of the model robustness to inevitable vari-ations of the inputs.The predicted vehicular charging power must be added to the

present load profile to obtain the total peak power at the locationof interest. This will be necessary for planning of the distribu-tion network to decide about augmentation of network assets toprepare for the potential vehicular charging demand.

C. Probability of Charging and Demand for Fast Charging

To show the impact of fast charging on the potential loaddemand, the feasibility of level-2 charging (240 V, up to 30A) is also considered at shopping center 1. The membershipfunctions for the PD variable of the fuzzy decision making unit(Section III-B) need to bemodified, as the linguistic terms Short,Average, and Long are defined based on the charge that can bereceived by the battery in a certain time interval. This increasessignificantly with a level-2 charger over the same period of time.All other procedures for calculating the probability of chargingremain unaffected.Fig. 10 shows the probability of charging using a level-2

charger at shopping center 1. The general shape of these proba-bility curves are not drastically different from those for level-1shown in Fig. 5(a) despite the effective change in the PD. This

Fig. 10. Average probability of charging (weekday, shopping center 1, level-2).

Fig. 11. Peak power at shopping center 1 (weekday, level 2).

is due to the fact that the charging decision is mainly depen-dent to the two other factors (i.e., DTH, and SOC), which re-main unchanged. The implications of level-2 charging on thepeak power, however, are significant as seen in Fig. 11, whichshows the peak charging load for the same assumed conditionsof Fig. 8. As seen the “expected” peak power steeply rises to 620kW (as opposed to just under 130 kW in level-1 charging). Thepower remains significantly higher than level-1 charging for theentire duration of time. “Upper” and “lower” bands similar tothe ones in Fig. 8 are also shown.Fig. 12 demonstrates the hourly “average” load demand

curves for the two levels of charging for shopping center 1 andfor the considered composition of PEVs. They are obtained byconsidering the duration of each parking event for as long ascharging continues. These curves are indicators of the expectedenergy demand for every hour. For example, the curve forlevel-2 charging shows an average load of around 400 kW at 1P.M. This indicates that the charging vehicles between 12 noonand 1:00 P.M. are expected to receive 400 kWh of energy.It is observed from Fig. 12 the expected energy demand for

level-2 chargingwill bemuch higher. Additionally, it is seen thatits variations are much steeper. This is due to the fact that PEVsconnected to a level-2 charger will draw large amounts of powerover a short period of time and disconnect when fully charged,which suddenly drops their demand power by a large amount.This is particularly true for light-capacity PEVs (such as thePrius) whose battery can be charged from a level-2 charger inless than half an hour.Analysis of the parking events in shopping center 1 [32]

shows a large number of short-duration parking events (lasting


Fig. 12. Average load demand in shopping center 1 for level-1 and level-2charging (weekday).

less than 1 hour). While level-1 charging may not provide asubstantial amount of charge for such parking events, level-2charging is appreciably more beneficial. The large amount ofenergy received from a level-2 charger will alleviate part of thePEVs’ needs when these vehicles arrive home and are pluggedin overnight. This may in fact assist in reducing the impact ofcoincidental charging on the distribution network in residentialareas.The foregoing analysis of level-2 charging is done with the

assumption that the cost of charging is not affected significantlyby the charging level. In reality level-2 charging is likelymore expensive than level-1 charging. As it was explained inSection III-E2, it is possible to include the effect of the priceof charging in prediction of the probability of charging byadjusting the DTH input. Despite this, the given analysis isstill valid because it shows the potential increase of the loadshould level-2 charging be made available. It can also be usedin deciding whether level-2 charging is economically viable,given the revenue that can be obtained from the sale of theextra energy.

VII. CONCLUSION

A location-based forecasting procedure of vehicular chargingload was presented in the paper. It uses a fuzzy inference systemwith three real-world inputs to emulate a driver’s decision tocharge at an off-home charging location. The main advantageof this approach is bringing the driver’s experience factor to thedecisionmaking process. The fuzzy system produces an averageprobability of charging curve (e.g., Fig. 5), which together witha probability of parking curve (e.g., Fig. 7) can be used to predictthe vehicular charging load due to any perceived combination,number, and type of plug-in vehicles. Central to this is avail-ability of data for driving and parking that best characterize thelocal patterns. The general procedure outlined in Fig. 1 can beeasily adopted for any other location, any day-type or season, aslong as reasonably reliable driving samples are available. Notethat since plug-in vehicles are still in their early stages of en-tering the market, there is no large-volume tagged (measured)data pertinent to the charging behavior of PEV owners. Henceit is not directly possible to fully validate the proposed model’spredictions; however the sensitivity analysis results presented in

the paper demonstrate the ability of the model in realizing theassumptions underlying its design.Utilities can affect the charging demand by proper coordi-

nation of i) offering fast chargers, which change the effect ofthe PD input on the received charge, and ii) regulating tariffs,which in effect change the DTH input. The results such as theones produced in the paper (Figs. 8, 9, 11, and 12) along withother economic and technical considerations can be used in theplanning of the scale, location, and type of charging infrastruc-ture, and whether upgrading network assets will be required inproviding such service.

REFERENCES[1] C. C. Chan, “The state of the art of electric and hybrid vehicles,” Proc.

IEEE, vol. 90, no. 2, pp. 247–275, Feb. 2002.[2] M. Ehsani, Y. Gao, and A. Emadi, Modern Electric, Hybrid Electric,

Fuel Cell Vehicles: Fundamentals, Theory, Design. Boca Raton, FL,USA: CRC, 2004.

[3] W. Su, H. Rahimi-Eichi,W. Zeng, andM. Chow, “A survey on the elec-trification of transportation in a smart grid environment,” IEEE Trans.Ind. Informat., vol. 8, no. 1, pp. 1–10, Feb. 2012.

[4] W. Kempton and J. Tomic, “Vehicle-to-grid power fundamental: Cal-culating capacity and net revenue,” J. Power Sources, vol. 144, no. 1,pp. 268–279, Jun. 2005.

[5] M. D. Galus, M. Zima, and G. Anderson, “On integration of plug-inhybrid electric vehicles into existing power system structures,” EnergyPolicy, vol. 38, pp. 6736–6745, Nov. 2010.

[6] S.W.Hadley and A. A. Tsvetkova, “Potential impacts of plug-in hybridelectric vehicles on regional power generation,” Electr. J., vol. 22, no.10, pp. 56–68, Dec. 2009.

[7] S. Rahman and G. Sharestha, “An investigation into the impact of elec-tric vehicle load on the electric utility system,” IEEE Trans. PowerDel., vol. 8, no. 2, pp. 591–597, Apr. 1993.

[8] J. Taylor, A. Maitra, M. Alexander, D. Brooks, and M. Duvall, “Eval-uation of the impact of plug-in electric vehicle loading in distributionsystem operations,” in Proc. IEEE Power Energy Soc. Gen. Meet., Cal-gary, AB, Canada, Jul. 2009.

[9] R. C. Green, II, L. Wang, and M. Alam, “The impact of plug-in hybridelectric vehicles on distribution networks: A review and outlook,” Re-newable Sustain. Energy Rev., vol. 15, pp. 544–553, Jan. 2011.

[10] A. A. Sallam and O. P. Malik, Electric Distribution Systems.Hoboken, NJ, USA: Wiley, 2011, pp. 33–65.

[11] H. Hahn, S. Meyer-Nieberg, and S. Pickl, “Electric load forecastingmethods: Tools for decision making,” Eur. J. Oper. Res., vol. 199, no.3, pp. 902–907, Dec. 2009.

[12] I. Moghram and S. Rahman, “Analysis and evaluation of five short-term load forecasting techniques,” IEEE Trans. Power Syst., vol. 4, no.4, pp. 1484–1491, Oct. 1989.

[13] D. C. Park, M. A. El-Sharkawi, R. J. Marks, II, L. E. Atlas, and M. J.Damborg, “Electric load forecasting using an artificial neural network,”IEEE Trans. Power Syst., vol. 6, no. 4, pp. 442–449, May 1991.

[14] M. Chow, J. Zhu, and H. Tram, “Application of fuzzy multi-objectivedecision making in spatial load forecasting,” IEEE Trans. Power Syst.,vol. 13, no. 3, pp. 1185–1190, Aug. 1998.

[15] H. S. Hippert, C. E. Pedreira, and R. C. Souza, “Neural networks forshort-term load forecasting: A review and evaluation,” IEEE Trans.Power Syst., vol. 16, no. 1, pp. 44–55, Feb. 2001.

[16] P. Fernández, T. G. S. Román, R. Cossent, C. M. Domingo, and P.Frías, “Assessment of the impact of plug-in electric vehicles on distri-bution networks,” IEEE Trans. Power Syst., vol. 26, no. 1, pp. 206–213,Feb. 2011.

[17] K. Clement-Nyns, E. Haesen, and J. Driesen, “The impact of chargingplug-in hybrid electric vehicles on a residential distribution grid,” IEEETrans. Power Syst., vol. 25, no. 1, pp. 371–380, Feb. 2010.

[18] E. Sortomme, M. M. Hindi, S. D. MacPherson, and S. S. Venkata,“Co-ordinated charging of plug-in hybrid electric vehicles to minimizedistribution system losses,” IEEE Trans. Smart Grid, vol. 2, no. 1, pp.198–205, Mar. 2011.

[19] K. Qian, C. Zhou, M. Allan, and Y. Yuan, “Modeling of load demanddue to EV battery charging in distribution systems,” IEEE Trans.Power Syst., vol. 26, no. 2, pp. 802–810, May 2011.

[20] S. Bae and A. Kwasinski, “Spatial and temporal model of electric ve-hicle charging demand,” IEEE Trans. Smart Grid, vol. 3, no. 1, pp.394–403, Mar. 2012.


[21] A. Rautiainen, S. Repo, P. Järventausta, A. Mutanen, K. Vuorilehto,and K. Jalkanen, “Statistical charging load modeling of PHEVs in elec-tricity distribution networks using national travel survey,” IEEE Trans.Smart Grid, vol. 3, no. 4, pp. 1650–1659, Dec. 2012.

[22] Z. Darabi andM. Ferdowsi, “Aggregated impact of plug-in hybrid elec-tric vehicles on electricity demand profile,” IEEE Trans. Sustain. En-ergy, vol. 2, no. 4, pp. 501–508, Oct. 2011.

[23] M. D. Galus, R. A. Waraich, F. Noembrini, K. Steurs, G. Georges, K.Boulouchos, K. W. Axhausen, and G. Anderson, “Integrating powersystems, transport systems and vehicle technology for electric mobilityimpact assessment and efficient control,” IEEE Trans. Smart Grid, vol.3, no. 2, pp. 934–949, Jun. 2012.

[24] D. Steen, L. A. Tuan, O. Carlson, and L. Bertling, “Assessment ofelectric vehicle charging scenarios based on demographic data,” IEEETrans. Smart Grid, vol. 3, no. 3, pp. 1457–1468, Sep. 2012.

[25] S. Shahidinejad, S. Filizadeh, and E. L. Bibeau, “Profile of chargingload on the grid due to plug-in vehicles,” IEEE Trans. Smart Grid, vol.3, no. 1, pp. 135–141, Mar. 2012.

[26] ECOtality, “The EV Project,” Q1-2012 Report.[27] S. Shahidinejad, E. L. Bibeau, and S. Filizadeh, “Winnipeg Driving

Cycle: WPG02,” [Online]. Available: http://mspace.lib.umanitoba.ca/handle/1993/3997

[28] L. A. Zadeh, “The concept of a linguistic variable and its applicationto approximate reasoning-I, II, III,” Inf. Sci., vol. 8;8;9, pp. 199–249,301–357; 43–80, 1975.

[29] E. H. Mamdani, “Application of fuzzy logic to approximate reasoningusing linguistic synthesis,” IEEE Trans. Comput., vol. C-26, no. 12,pp. 1182–1191, Dec. 1977.

[30] V. Agarwal, K. Uthaichana, R. A. DeCarlo, and L. H. Tsoulalas, “De-velopment and validation of a battery model useful for discharging andcharging power control and lifetime estimation,” IEEE Trans. EnergyConvers., vol. 25, no. 3, pp. 821–835, Sep. 2010.

[31] Y. Xing,W.He,M. Pecht, andK. L. Tsui, “State of charge estimation oflithium-ion batteries using the open-circuit voltage at various ambienttemperatures,” Appl. Energy, vol. 113, pp. 106–115, Jan. 2014.

[32] N. Ghiasnezhad and S. Filizadeh, “Characterization of prospectivecharging locations of plug-in vehicles using real-world driving data,”in Proc. IEEE Power Energy Soc. Gen. Meet., Vancouver, BC, Canada,Jul. 2013.

[33] F. U. Syed, M. L. Kuang, J. Czubay, and H. Ying, “Derivation and ex-perimental validation of a power-split hybrid electric vehicle model,”IEEE Trans. Veh. Technol., vol. 55, no. 6, pp. 1731–1747, Nov. 2006.

[34] D.W. Gaoi, C. Mi, and A. Emadi, “Modeling and simulation of electrichybrid vehicles,” Proc. IEEE, vol. 95, no. 4, pp. 729–745, Apr. 2007.

[35] E. Tara, S. Shahidinejad, S. Filizadeh, and E. Bibeau, “Battery storagesizing in a retrofitted plug-in hybrid electric vehicle,” IEEE Trans. Veh.Technol., vol. 59, no. 6, pp. 2786–2794, Jul. 2010.

Nima Ghiasnezhad Omran (S’12) received his B.Sc. in electrical engineeringfrom Tabriz University, Tabriz, Iran, and the M.Sc. degree in electrical engi-neering from K. N. Toosi University of Technology, Tehran, Iran. He joinedthe University of Manitoba, Canada, in 2010 where he is presently a doctoralcandidate in the Department of Electrical and Computer Engineering. His areasof interest include power electronics, power systems planning, distribution net-work control, and smart grids.

Shaahin Filizadeh (S’02–M’05–SM’10) received his B.Sc. and M.Sc. degreesin electrical engineering, from Sharif University of Technology in Tehran, Iran,in 1996 and 1998, respectively. He obtained his Ph.D. from the University ofManitoba, Canada, in 2004.He is currently an Associate Professor with the Department of Electrical and

Computer Engineering of the University of Manitoba. His areas of interest in-clude electromagnetic transient simulation, nonlinear optimization, and powerelectronic applications in power systems and vehicle propulsion.Dr. Filizadeh is a Registered Professional Engineer in the Province of

Manitoba.

Documents

Document6