Upload
jungho-park
View
101
Download
3
Embed Size (px)
Citation preview
Seoul National University
Seoul National University System Health & Risk Management
2017/2/25 ‐ 1 ‐
Correlation metric
Seoul National University System Health & Risk Management
Jungho Park*
Seoul National University ‐ 2 ‐
Validation metric
Validation metric : a mathematical operator that measures the difference between a
system response quantity (SRQ) obtained from a simulation result and one obtained
from experimental measurement.(verification and validation in scientific computing)
Figure reference : Verification, validation, and predictive capability in computational engineering and physics, Oberkampf et al. ,Applied mechanics(2004)
Seoul National University ‐ 3 ‐
Validation metric 의종류
1. Classical hypothesis testing
‐ 평균및분산에대한가설을세우고, 얻어진실험결과로부터가설검정실시
‐ 장점 : 모델의적합도여부를결정가능
‐ 단점 : 실험의개수가적을때는이용불가능
Liu, Yu, et al. "Toward a better understanding of model validation metrics."Transactions of the ASME‐R‐Journal of Mechanical Design 133.7 (2011): 071005.
Seoul National University ‐ 4 ‐
Validation metric 의종류
2. Bayes factor
‐ Bayesian hypothesis testing 에서유래
‐ Null, alternative 가설의 posterior distribution 의비에의해결정
Liu, Yu, et al. "Toward a better understanding of model validation metrics."Transactions of the ASME‐R‐Journal of Mechanical Design 133.7 (2011): 071005.
B=bayes factor
Seoul National University ‐ 5 ‐
Validation metric 의종류
3. Frequentist’s metric
‐ Hypothesis 로부터모델의적합도를 ‘yes ‘ or ‘no’를결정하기보다는실험과시
뮬레이션값의차이를정량화
Liu, Yu, et al. "Toward a better understanding of model validation metrics."Transactions of the ASME‐R‐Journal of Mechanical Design 133.7 (2011): 071005.
tane estimated predictionerrors estimated s dard devidationN numberof physicalobservation
Estimated error in the predictive model with a confidence level of 100(1‐ α)% that the true error is in the interval =
e
Seoul National University ‐ 6 ‐
Validation metric 의종류
4. Area metric
‐ Mean, variance 같은moment 가아닌시험, 시뮬레이션분포의전체적모양을
비교
‐ 시험, 시뮬레이션개수가적을때사용가능
‐ U‐pooling method 와함께자주쓰임
Liu, Yu, et al. "Toward a better understanding of model validation metrics."Transactions of the ASME‐R‐Journal of Mechanical Design 133.7 (2011): 071005.
Seoul National University ‐ 7 ‐
Error metric(or correlation metric )의종류
1. Vector norms
2. Average residual and Its Standard Deviation
3. Coefficient of correlation and cross relation
Limitations: Not able to distinguish error due to phase from error due to magnitude
Limitations: Positive and negative differences at various point may cancel out
2
11
( )
1
N
i
N
R RS
N
( )i iRi a b
Limitations: Sensitive to phase differenceNot able to distinguish error due to phase from error due to magnitude
1 1 1
2 2 2 2
1 1 1 1
( )( )
( ) ( ) ( ) ( )
N n N n N n
i i n i i ni i i
N n N n N n N n
i i i n i ni i i i
N n a b a bn
N n a a N n b b
*Comparing Time Histories for Validation of Simulation Models: Error Measures and Metrics, H.Sarin, M.Kokkolaras, G.Hulbert, P.Papalambro, S.Barbat, R.‐J.Yang, Journal of Dynamic Systems, Measurement, and Control(2010)
Seoul National University ‐ 8 ‐
Error metric(or correlation metric )의종류
4. Sprague and Geers metric
5. Russel’s error measure
1&G
1 cos ( ),ABS
AA BB
P
& 1,
2 2& & &S G S G S GC M P
2 2
1 1 1, , ,
N N N
i i i ii i i
AA BB AB
a b a b
N N N
Characteristics: Phase error portion consideredLimitations: lumped the entire information into , ,
Magnitude :
Phase :
Total :
10( ) log (1 )AA BBR AA BB
AA BB
M sign
Characteristics: Phase error portion consideredLimitations: lumped the entire information into , ,
No magnitude error
*Comparing Time Histories for Validation of Simulation Models: Error Measures and Metrics, H.Sarin, M.Kokkolaras, G.Hulbert, P.Papalambro, S.Barbat, R.‐J.Yang, Journal of Dynamic Systems, Measurement, and Control(2010)
Seoul National University ‐ 9 ‐
Error metric(or correlation metric )의종류
6. Normalized Integral Square Error(NISE)
7. Dynamic Time Warping
2 ∗ 2∗
2 ∗ 1 ∗
12
Phase : Magnitude: Shape :
Total :
Characteristics: Shape error portion consideredLimitations: Magnitude portion can be negative. (which mean magnitude portion can decrease overall error)
Characteristics: Algorithm for measuring discrepancy between time history
*Comparing Time Histories for Validation of Simulation Models: Error Measures and Metrics, H.Sarin, M.Kokkolaras, G.Hulbert, P.Papalambro, S.Barbat, R.‐J.Yang, Journal of Dynamic Systems, Measurement, and Control(2010)
Seoul National University ‐ 10 ‐
Error metric(or correlation metric )의종류
8. Weighted Integrated Factor (WIFac)
1
max , ⋅ 1 max 0, ⋅max ,
max ,0 1
1∑∑
Seoul National University ‐ 11 ‐
Correlation metric and validation metric
Validation metric : a mathematical operator that measures the difference between a
system response quantity (SRQ) obtained from a simulation result and one obtained
from experimental measurement.(verification and validation in scientific computing)
0 5 10 15 20 25 300
50
100
150
200
250
300
350Simulation
Time(ms)
Res
ulta
nt A
cc(g
)0 0.2 0.4 0.6 0.8 1
0
10
20
30
40
50
60
70
80
90
WIFac
Den
sity
Exp :Sim :
Time(s)Ac
c(g)
0 0.005 0.01 0.015 0.02 0.025 0.030
20
40
60
80
100
120
140
160
180
Meanof exp.
:
0 5 10 15 20 25 300
50
100
150
200
250
300
350Simulation
Time(ms)
Res
ulta
nt A
cc(g
)
0 5 10 15 20 25 300
50
100
150
200
250
300
350Simulation
Time(ms)
Res
ulta
nt A
cc(g
)
0 5 10 15 20 25 300
50
100
150
200
250
300
350Simulation
Time(ms)
Res
ulta
nt A
cc(g
)
+3σ +1.5σ
‐3σ ‐1.5σ
WIFac
Derivation of WIFac for Simulation(0.5275, 0.5133,0.5293,0.5183)
0 5 10 15 20 25 300
50
100
150
200
250
300
350Experiment
Time(ms)
Res
ulta
nt A
cc(g
)
HIC 324HIC 565HIC 347HIC 290
Derivation of WIFac for Experiment(0.7738, 0.6186, 0.7648, 0.7308)
WIFac is not Validation metric, but Area metric is
0 0.5 10
0.5
1Funi
FuCDF
Um = 0.2641