350 likes | 518 Views
Model Calibration Test Cases. Bernie Lesieutre. WECC MVWG, Salt Lake City, UT, June 18, 2013. Model Calibration Test Cases. Data from 7 simulated events 3 sets used for model calibration 4 sets used for model consistency check
E N D
Model Calibration Test Cases Bernie Lesieutre WECC MVWG, Salt Lake City, UT, June 18, 2013
Model Calibration Test Cases • Data from 7 simulated events • 3sets used for model calibration • 4 sets used for model consistency check + model independently checked by others (Dmitry) using data from other events Discussion of nonuniqueness of parameter sets
Analysis • What are the features of these responses? • Primarily long-term trends, tens of seconds • Some oscillatory responses. • First attempt using simplest Least Squares Approach • Estimate parameter values • Estimate bounds on values • Use PSLF to generate sensitivity model. Iterate.
Event 1 Original Parameters Modified Parameters
Event 2 Original Parameters Modified Parameters
Event 3 Original Parameters Modified Parameters
Event 4 Original Parameters Modified Parameters
Event 5 Original Parameters Modified Parameters
Event 6 Original Parameters Modified Parameters
Event 7 Original Parameters Modified Parameters
Parameter Changes 0.052 (0.003) 0.0 xcomp 3.62 4.42 (0.18) -0.1 0.0 0.1 H 3.5 4.0 4.5 5.0
Parameter Changes 400 443 (175) Ka 200 300 400 500 0.021 (0.003) 0.035 0.79 (0.12) 1.25 Kf 0.01 0.02 0.03 0.04 0.05 Tf 0.5 1.0 1.5
Parameter Changes 12.6 (0.29) 30 T5 10 15 20 25 30 Ks 8.2 (0.18) 15 0 10 20
Parameter Changes 1.82 (0.35) 0.045 (0.0015) 1.27 1.76 (0.22) 0.05 1.08 Tr Tw rperm 0 2 4 6 8 10 0.04 0.05 0.06 0.07 1.0 1.5 2.0 2.5 0.46 (0.09) 0.5 rtemp 0.1 0.2 0.3 0.4 0.5
NOT UNIQUE Analyses of the sensitivity model and of the parameter covariance matrix suggests that certain parameter adjustments will have negligible affect on the simulations. These include: • The value of Ka almost unilaterally • Patterns of changes in at least four different directions with little change.
Not Unique Fit for values of Ka=443, 400, and 250.
Not Unique H 4.42 4.42 Xc 0.052 0.052 Ka 443 400 Kf 0.021 0.021 Tf 0.79 0.78 T5 12.6 12.6 Ks 8.23 8.23 Rp 0.045 0.045 Rt 0.46 0.50 Tr 1.82 1.65 Tw 1.76 1.76
Not Unique H 4.42 4.38 Xc 0.052 0.052 Ka 443 400 Kf 0.021 0.021 Tf 0.79 0.77 T5 12.6 12.6 Ks 8.23 8.25 Rp 0.045 0.045 Rt 0.46 0.52 Tr 1.82 1.67 Tw 1.76 1.27
Not Unique H 4.42 3.62 Xc 0.052 0.052 Ka 443 400 Kf 0.021 0.020 Tf 0.79 0.71 T5 12.6 12.6 Ks 8.23 8.27 Rp 0.045 0.045 Rt 0.46 0.50 Tr 1.82 1.74 Tw 1.76 1.27
Sensitivity Model Singular Values Corresponding singular vectors indicate specific parameter changes that result in similar solutions 25997 Rperm 22203 xcomp, Kf 15942 xcomp, Kf 743 Rtemp, Tr 277 Tf, 246 T5, Ks 173 H 138 Tw 98 T5, Ks 86 Rtemp, Tr 0 .2 Ka Tr Rtemp
NOT UNIQUE H 4.42 4.47 Xc 0.052 0.052 Ka 443 400 Kf 0.021 0.021 Tf 0.79 0.72 T5 12.6 12.6 Ks 8.23 8.26 Rp 0.045 0.045 Rt 0.46 0.40 Tr 1.82 2.02 Tw 1.76 2.00
Training Signal 1 H 4.42 4.47 Xc 0.052 0.046 Ka 443 500 Kf 0.021 0.027 Tf 0.79 0.89 T5 12.6 12.2 Ks 8.23 8.58 Rp 0.045 0.045 Rt 0.46 0.32 Tr 1.82 2.61 Tw 1.76 2.04
Training Signals 1 4 6 H 4.42 4.46 Xc 0.052 0.48 Ka 443 500 Kf 0.021 0.026 Tf 0.79 0.88 T5 12.6 12.3 Ks 8.23 8.42 Rp 0.045 0.045 Rt 0.46 0.32 Tr 1.82 2.60 Tw 1.76 2.08
Conclusion • Finding a decent set of parameters to match data is not difficult in this case. • There are too many parameters to tweak. Confidence that the parameters are “right” is low. • Confidence that the parameters will work accurately in simulations is moderate to high.