You are on page 1of 20

Model Validation: FIDVR Event

Prepared for: Bonneville Power Administration Prepared by Alex Borden Bernard Lesieutre

1. Introduction
Fault-Induced Delayed Voltage Recovery (FIDVR) events are characterized by a depressed voltage for 5 to 30 seconds following a fault. These events are of concern because they show a temporary loss of voltage control in an area, and they pose a risk of cascading to a larger area, especially if another unexpected event occurs while the voltage is depressed. These events have been studied and are believed to be initiated by the stall of low-inertia induction motors during the fault. Motors at risk of stalling include compressor-driving loads such as air conditioners. The subject of FIDVR events, their causes and solutions is covered in a NERC white paper on this topic [NERC]. To understand and anticipate the possibility of such events and assess their potential to cascade further, plausible simulation models are needed. After past simulation models failed to capture FIDVR events, the WECC Modeling and Validation Working Group developed an improved load model, in part, to be able to represent FIDVR dynamics. The WECC composite load model has this capability. To use the model, suitable parameters need be chosen to match events of interest. This report focuses on load modeling efforts to validate models that are used to simulate FIDVR events. The approach here uses the industry-standard simulation program PSLF and models obtained from WECC. The simulation results are compared to recorded event data, and parameters are adjusted to better fit the model to data. The process is automated in the MATLAB environment. In the study, we examine the impact of adjusting parameters of the WECC Composite Load Model. In the following sections we examine two FIDVR data sets. The first set of data only includes a voltage measurement from the transmission interconnection. The second data set includes voltage, active power and reactive power measurements. As the results of this report show, it is possible to represent the data from FIDVR events by adjusting the load model parameters.

2. FIDVR Data 1
The system data provided for this model included two representations for the network, an 8-bus reduced model, and a 92-bus detailed model. The purpose of this is to examine how well a reducedorder model will replicate the event, and to compare how the resulting load model parameters depend on the detail of the representation. The WECC composite load model is used in these studies. It is represented below in Figure 1.


Figure 1. WECC Composite Load Model (CMPLDW)

The WECC Composite Load Model allows for precision modeling of distribution feeders and loads. The total load is divided into fractions of electronic, static and motor. The user is allowed to specify parameters of four groups of motors. Each of the four motor groups can be customized to have unique parameters of inertias, tripping and reclosing thresholds, can be 3-phase or 1-phase airconditioning units and more. The CMPLDW has anywhere between 118-130 parameters to be specified depending on how many of the four motor groups are 1-phase versus 3-phase. Some of the parameters have default values given in the help manual, however many parameters have no default value (i.e. fractions of load types, voltage trip and reclose levels, etc). There is therefore no default set of parameters that can

be used as a starting point as some parameters must be implicitly defined by the engineer modeling the system. We started investigations using the parameters provided to us with the model. The reduced network is shown in Figure 2. This is a one-line diagram of the simplified system model used to study the FIDVR event. There is one composite load at bus 6. At time 0.8 seconds a fault is applied at that bus. At time 0.85 seconds the fault is removed and one of the two lines connecting buses 5 and 6 is also opened. The simulation ends at a final time of 50 seconds.
G 1 SVD D

6 SVD D Figure 2. Network Model 1: One-line diagram

A measured voltage profile of the FIDVR event at the load bus 6 is shown in Figure 3. We highlight important features of the measurement including the initial steady-state voltage, the fault, the slow voltage recovery and the steady-state final voltage.

Figure 3. Time (sec) vs. Voltage Actual Measurement

Existing model parameter sets used in simulation did not accurately match the recorded measurement of the FIDVR event. In Figure 4, we show the results of one of the (better) parameter sets that were provided. The initial load model parameters are listed in Table 1.

Figure 4. Initial CMPLDW Parameters Simulation Table 1. Initial Load Model Parameters

Bss 0 Rfdr 0.04 Xfdr 0.05 fb 0.75 Xxf 0.08 Tfixhs 1 Tfixls 1 ltc 1 Tmin 0.9 Tmax 1.1 step 0.00625 Vmin 1 Vmax 1.02 Tdel 30 Ttap 5 Rcmp 0 Xcmp 0 FmA 0.15 FmB 0.1 FmC 0.05 FmD 0.3 Fel 0.1 Pfel 1 Vd1 0.75

Vd2 0.65 Frcel 0.25 pfs -0.99 P1e 2 P1c 0.54546 P2e 1 P2c 0.45454 Pfrq 0 Q1e 2 Q1c -0.5 Q2e 1 Q2c 1.5 Qfrq -1 MtypA 3 MtypB 3 MtypC 3 MtypD 1 LfmA 0.7 LFmB 0.8 LFmC 0.8 LFmD 1 RaA 0.04 RaB 0.03 RaC 0.03

CompPF D LsA LsB LsC VstallD LpA LpB LpC RstallD LppA LppB LppC XstallD TpoA TpoB TpoC TstallD TppoA TppoB TppoC FrstD HA HB HC

0.97 1.8 1.8 1.8 0.6 0.1 0.16 0.16 0.1 0.083 0.12 0.12 0.1 0.092 0.1 0.1 0.02 0.002 0.0026 0.0026 0 0.05 1 0.1

VrstD EtrqA EtrqB EtrqC TrstD Ftr1A Ftr1B Ftr1C FuvrD Vtr1A Vtr1B Vtr1C Vtr1D Ttr1A Ttr1B Ttr1C Ttr1D Vtr2A Vtr2B Vtr2C Vtr2D Ttr2A Ttr2B Ttr2C

0.9 0 2 2 0.4 0.2 0.2 0.2 0.2 0.75 0.7 0.7 0.65 999 0.02 0.02 0.02 0.5 0.7 0.7 0.9 0.02 0.02 0.02

Ttr2D 5 Ftr2A 0.5 Ftr2B 0.3 Ftr2C 0.3 Vc1offD 0.4 Vrc1A 0.9 Vrc1B 0.9 Vrc1C 0.9 Vc2offD 0.4 Trc1A 9999 Trc1B 0.02 Trc1C 0.02 Vc1onD 0.45 Vrc2A 0.8 Vrc2B 0.85 Vrc2C 0.85 Vc2onD 0.45 Trc2A 0.02 Trc2B 99999 Trc2C 99999 TthD 15 Th1tD 0.7 Th2tD 1.9 TvD 0.025

The initial simulation in Figure 4 does not match key features of interest. The most notable inaccuracy is the voltage recovery period in simulation is much faster than the actual measurements. This period is when air-conditioning units are consecutively locking up and tripping offline. Of the four motor groups, three of the four groups (A, B & C) are 3-phase motors and the remaining group (D) represents 1-phase air conditioning units. We began manually adjusting certain parameters to correct the simulations. The most important parameter was the thermal time constant Tth of the 1-phase air conditioners of motor group D. This parameter was increased to allow for a slower voltage recovery period. This one change dramatically improved the fit of the simulation. Similarly the thermal protection trip start level and trip completion level parameters Th1t and Th2t were slightly changed as well. Very few other parameters required changing. Fine-tuning adjustments were made to motor load fractions, voltage protection trip and reclose levels and time to trip and reclose. The resulting voltage is shown in Figure 5, and the corresponding parameters shown in Table 2. Changes to the base parameters are color-highlighted in the table. The larger changes are bolder in color, and minor changes are duller.

Figure 5. Improved CMPLDW Parameters Simulation

Table 2. Improved Load Model Parameters

Bss 0 Rfdr 0.04 Xfdr 0.05 fb 0.75 Xxf 0.08 Tfixhs 1 Tfixls 1 ltc 1 Tmin 0.9 Tmax 1.1 step 0.00625 Vmin 1 Vmax 1.02 Tdel 30 Ttap 5 Rcmp 0 Xcmp 0 FmA 0.17 FmB 0.1 FmC 0.05 FmD 0.23 Fel 0.1 Pfel 1 Vd1 0.75

Vd2 0.65 Frcel 0.25 pfs -0.99 P1e 2 P1c 0.54546 P2e 1 P2c 0.45454 Pfrq 0 Q1e 2 Q1c -0.5 Q2e 1 Q2c 1.5 Qfrq -1 MtypA 3 MtypB 3 MtypC 3 MtypD 1 LfmA 0.7 LFmB 0.8 LFmC 0.8 LFmD 1 RaA 0.04 RaB 0.03 RaC 0.03

CompPF D LsA LsB LsC VstallD LpA LpB LpC RstallD LppA LppB LppC XstallD TpoA TpoB TpoC TstallD TppoA TppoB TppoC FrstD HA HB HC

0.97 1.8 1.8 1.8 0.6 0.1 0.16 0.16 0.1 0.083 0.12 0.12 0.1 0.092 0.1 0.1 0.02 0.002 0.0026 0.0026 0 0.05 1 0.1

VrstD EtrqA EtrqB EtrqC TrstD Ftr1A Ftr1B Ftr1C FuvrD Vtr1A Vtr1B Vtr1C Vtr1D Ttr1A Ttr1B Ttr1C Ttr1D Vtr2A Vtr2B Vtr2C Vtr2D Ttr2A Ttr2B Ttr2C

0.9 0 2 2 0.4 0.2 0.2 0.2 0.17 0.75 0.5 0.5 0.65 999 0.02 0.02 0.02 0.5 0.7 0.7 0.9 0.02 0.02 0.02

Ttr2D Ftr2A Ftr2B Ftr2C Vc1offD Vrc1A Vrc1B Vrc1C Vc2offD Trc1A Trc1B Trc1C Vc1onD Vrc2A Vrc2B Vrc2C Vc2onD Trc2A Trc2B Trc2C TthD Th1tD Th2tD TvD

5 0.47 0.3 0.3 0.4 0.9 0.65 0.65 0.4 9999 0.6 0.6 0.45 0.639 0.85 0.85 0.45 0.73 99999 99999 30 0.3 2.05 0.025

Voltage Trip & Reclose Levels

Time to Trip & Reclose (sec)

Thermal Protection Parameters

Motor Fraction Parameters (%)

The effects of parameter changes are plotted below in Figures 6 - 8.

Figure 6. Trip & Reclose Voltage Level and Time Parameter Changes

* It is important to note the voltage trip and reclose levels happen at the distribution level load bus Vld in the CMPLDW model in Figure 1. The actual voltage measurement recorded was taken at the transmission level bus V in the CMPLDW model.

Figure 7. Thermal Protection Parameter Changes (Time vs. Temperature)

Figure 8. Load Fraction and Trip Fractions Changed

The parameter changes made for this data set allowed for a close fit simulation to the measurements. Even with this solution it cannot be absolutely concluded that the values chosen were unique. There are many parameters in the model, and it is likely that another set, or a family of connected parameter values would yield an equal, or nearly equal fit. Next we apply the parameters to a larger model of the system. A comparison of the number of different types of elements in the reduced and more detailed model is given by

# Buses: # Generators: # Composite Loads: # Lines: # Transformers:

8 1 1 5 4

92 1 29 81 21

Using the same parameters from Table 1 and applying them uniformly to all 29 composite loads yielded the simulation shown in Figure 9. As with reduced-order model, the initial load parameter set did not give a simulation that matched the data well.

Figure 9. Initial CMPLDW Parameters Simulation

Nearly the same parameters from Table 2 were used uniformly for all 29 composite loads except for the fine-tuning of 4 values: - Vrc1B & Vrc1C: 0.65 0.6 - Vrc2a: 0.639 0.6 27.5 - TthD: 30 The improved simulation using these parameters is shown in Figure 10.

Figure 10. Improved CMPLDW Parameters Simulation

3. FIDVR Data 2
In this section we report on the fitting of a second set of FIDVR data that includes both a (single) voltage measurement, and recorded active and reactive power measurements at two locations. The one-line diagram for the network model is shown in Figure 11. There are two composite loads at bus 4 and 6. At time 5 seconds the per-unit voltage at generator bus 1 drops from 1.04 to 0.75 until 5.16 seconds when it returns to voltage 1.04 per-unit.

G 1 SVD D

4 G SVD D

Figure 11. One-line diagram for simulation at time 50 seconds

Fitting parameters to this data set proved to be more challenging because there are more measurements and features to fit. Furthermore, there was one bus voltage measurement; however it was not known where the measurement was taken. It was hypothesized to either be at bus 4 or at bus 6. There was also two active power flow and two reactive power flow measurements recorded. These flows were measured on the line connecting bus 2 to 3 and the line connecting bus 2 to 5. It was also unknown if the station connected at bus 4 was switched in or not. Simulations were done with the station both switched in and switched out. It was observed that the station had only little effect on the outcome of the simulations. Since the station had minimal impact on the simulation it was just assumed it was switched in for the simulations in this report.

This network model includes two composite loads. Initially it was assumed that the loads were sufficiently similar to be represented by identical models, however this yielded poor results. Better results were obtained when the two composite loads were represented using different parameters. This doubled the number of parameters to consider while varying the parameter estimation. The initial load model parameters provided inaccurate simulations as shown in Figures 12-14. Since it was not known which of the two buses the voltage measurement was recorded at, both bus voltages were plotted from the simulation in Figure 12. The initial data set is listed in Table 3.

Figure 12. Initial Voltage simulation

Figure 13. Initial Active Power Flows

Figure 14. Initial Reactive Power Flows

Table 3 Initial CMPLDW Parameters

Bss 0 Rfdr 0.0296 Xfdr 0.025 fb 0.748 Xxf 0.08 Tfixhs 1 Tfixls 1 ltc 1 Tmin 0.9 Tmax 1.1 step 0.00625 Vmin 1.025 Vmax 1.04 Tdel 30 Ttap 5 Rcmp 0 Xcmp 0 FmA 0.18 FmB 0.12 FmC 0.08 FmD 0.3 Fel 0.1 Pfel 1 Vd1 0.75

Vd2 0.65 Frcel 0.3 pfs -0.99 P1e 2 P1c 0.54546 P2e 1 P2c 0.45454 Pfrq 0 Q1e 2 Q1c -0.5 Q2e 1 Q2c 1.5 Qfrq -1 MtypA 3 MtypB 3 MtypC 3 MtypD 1 LfmA 0.7 LFmB 0.8 LFmC 0.8 LFmD 1 RaA 0.04 RaB 0.03 RaC 0.03

CompPF D LsA LsB LsC VstallD LpA LpB LpC RstallD LppA LppB LppC XstallD TpoA TpoB TpoC TstallD TppoA TppoB TppoC FrstD HA HB HC

0.97 1.8 1.8 1.8 0.62 0.1 0.16 0.16 0.09 0.083 0.12 0.12 0.09 0.092 0.1 0.1 0.03 0.002 0.0026 0.0026 0 0.05 1 0.1

VrstD EtrqA EtrqB EtrqC TrstD Ftr1A Ftr1B Ftr1C FuvrD Vtr1A Vtr1B Vtr1C Vtr1D Ttr1A Ttr1B Ttr1C Ttr1D Vtr2A Vtr2B Vtr2C Vtr2D Ttr2A Ttr2B Ttr2C

0.9 0 2 2 0.4 0.2 0.25 0.25 0.2 0.7 0.7 0.7 0.6 0.03 0.02 0.02 0.03 0.5 0.8 0.8 0.9 0.02 999 999

Ttr2D 5 Ftr2A 0.75 Ftr2B 1 Ftr2C 1 Vc1offD 0.4 Vrc1A 1 Vrc1B 1 Vrc1C 1 Vc2offD 0.4 Trc1A 9999 Trc1B 9999 Trc1C 9999 Vc1onD 0.45 Vrc2A 0.6 Vrc2B 1 Vrc2C 1 Vc2onD 0.45 Trc2A 0.02 Trc2B 99999 Trc2C 99999 TthD 15 Th1tD 0.7 Th2tD 1.9 TvD 0.02

Initial attempts to manually change parameter were cumbersome and not so successful, so an automated parameter selection process was developed that provided good results. To find the best solution the automated selection algorithm set lower and upper bounds for each parameter along with step sizes of possible parameters values between these bounds. Then, a complete approach, in theory, would enumerate all possible combinations of these parameter values and pick the best solution. The important question is, What is the best solution? There are many metrics that could be introduced to assess the quality of a simulation. We have identified features above that might be used for this purpose. The FIDVR studies have few independent features that characterize their behavior and we found it convenient to apply a least-squared error approach (2-norm) to the active and reactive power flow measurements that are known. We omitted the voltage measurement from the metric as its location is uncertain, however we do plot the results to ensure that the simulated voltages appear consistent with the measurement.

The objective function takes the form:

It should also be noted that the original data had 1501 time-step measurements between time 0 and 50 seconds. PSLF however simulates 2000+ time-steps between time 0 and 50 seconds. Therefore we found it appropriate to only compare actual to simulated values at the 1501 time-steps we had real data for. This required some interpolation between points in the simulated cases to compare simulated values to the actual measurements. In practice it is not feasible to enumerate all possible combinations of parameter values. Therefore a local steepest descent method by parameter was employed. From the given set of initial parameters, the algorithm would cycle through multiple values of each parameter individually and set that parameter to the value minimizing the objective function. Once that parameter was set it would move on to the next parameter and repeat. When all parameters had been set for the first time, the automation would start all over again with the first parameter. The automation would finally terminate when parameter values and objective function stopped changing. The simulated active and reactive power flows are not continuously differentiable functions. They encounter step changes during faults, trips and recloses, transformer tap changes, etc. If the objective function was continuously differentiable a nonlinear optimization program could be used. Unfortunately this is not the case, so the current automation process at this time involves some user input and guidance around these points. This remains a topic of future research and development. One of the solutions obtained is in this manner is shown below in Figures 15-17. It was simulated with the station switched in, and there were separate parameter values for each composite load. As noted previously, there is a family of parameter values that yield result of equal quality. The corresponding parameter values are shown in Tables 4-5

Figure 15. Improved Voltage Simulations

Figure 16. Improved Active Power Flows

Figure 17. Improved Reactive Power Flows

Table 4 : Improved Bus 6 CMPLDW Parameters

Bss 0 Rfdr 0.0175 Xfdr 0.0835 fb 0.5 Xxf 0.08 Tfixhs 1 Tfixls 1 ltc 1 Tmin 0.9 Tmax 1.1 step 0.00625 Vmin 1 Vmax 1.02 Tdel 30 Ttap 5 Rcmp 0 Xcmp 0 FmA 0.15 FmB 0.0667 FmC 0.1 FmD 0.4

Vd2 0.65 Frcel 0.25 pfs -0.99 P1e 2 P1c 0.54546 P2e 1 P2c 0.45454 Pfrq 0 Q1e 2 Q1c -0.5 Q2e 1 Q2c 1.5 Qfrq -1 MtypA 3 MtypB 3 MtypC 3 MtypD 1 LfmA 0.9667 LFmB 1.1 LFmC 0.7 LFmD 1

CompPF D 0.97 LsA 1.8 LsB 1.8 LsC 1.8 VstallD 0.6 LpA 0.2 LpB 0.2 LpC 0.2 RstallD 0.124 LppA 0.13 LppB 0.18 LppC 0.13 XstallD 0.114 TpoA 0.15 TpoB 0.2 TpoC 0.2 TstallD 0.033 TppoA 0.0026 TppoB 0.0026 TppoC 0.0026 FrstD 0.5

VrstD 0.9 EtrqA 2 EtrqB 2 EtrqC 2 TrstD 0.4 Ftr1A 0.2 Ftr1B 0.1 Ftr1C 0.2333 FuvrD 0.22 Vtr1A 0.7 Vtr1B 0.7 Vtr1C 0.7 Vtr1D 0.65 Ttr1A 999 Ttr1B 0.02 Ttr1C 0.02 Ttr1D 0.02 Vtr2A 0.7 Vtr2B 0.7 Vtr2C 0.7 Vtr2D 0.9

Ttr2D Ftr2A Ftr2B Ftr2C Vc1offD Vrc1A Vrc1B Vrc1C Vc2offD Trc1A Trc1B Trc1C Vc1onD Vrc2A Vrc2B Vrc2C Vc2onD Trc2A Trc2B Trc2C TthD

5 0.5 0.3333 0.1667 0.4 0.9 0.9333 0.9333 0.4 9999 0.01 0.6 0.45 0.9333 0.85 0.85 0.45 0.01 99999 99999 25

Fel Pfel Vd1

0.1 1 0.75

RaA RaB RaC

0.09 0.05 0.17

HA HB HC

0.1 0.1 0.1

Ttr2A Ttr2B Ttr2C

0.02 0.02 0.02

Th1tD Th2tD TvD

0.4 2.25 0.025

Table 5 Improved Bus 4 CMPLDW Parameters

Bss 0 Rfdr 0.034 Xfdr 0.01 fb 0.8333 Xxf 0.08 Tfixhs 1 Tfixls 1 ltc 1 Tmin 0.9 Tmax 1.1 step 0.00625 Vmin 1 Vmax 1.02 Tdel 30 Ttap 5 Rcmp 0 Xcmp 0 FmA 0.175 FmB 0.0667 FmC 0.1 FmD 0.4 Fel 0.1 Pfel 1 Vd1 0.75

Vd2 0.65 Frcel 0.25 pfs -0.99 P1e 2 P1c 0.54546 P2e 1 P2c 0.45454 Pfrq 0 Q1e 2 Q1c -0.5 Q2e 1 Q2c 1.5 Qfrq -1 MtypA 3 MtypB 3 MtypC 3 MtypD 1 LfmA 0.7 LFmB 1.1 LFmC 0.7 LFmD 1 RaA 0.01 RaB 0.01 RaC 0.0867

CompPF D LsA LsB LsC VstallD LpA LpB LpC RstallD LppA LppB LppC XstallD TpoA TpoB TpoC TstallD TppoA TppoB TppoC FrstD HA HB HC

0.97 1.8 1.8 1.8 0.6 0.2 0.2 0.2 0.124 0.13 0.18 0.13 0.114 0.15 0.2 0.2 0.033 0.0026 0.0026 0.0026 0.5 0.1 0.1 0.1

VrstD 0.9 EtrqA 2 EtrqB 2 EtrqC 2 TrstD 0.4 Ftr1A 0.2 Ftr1B 0.2333 Ftr1C 0.2333 FuvrD 0.22 Vtr1A 0.7 Vtr1B 0.7 Vtr1C 0.7 Vtr1D 0.65 Ttr1A 999 Ttr1B 0.02 Ttr1C 0.02 Ttr1D 0.02 Vtr2A 0.7 Vtr2B 0.7 Vtr2C 0.7 Vtr2D 0.9 Ttr2A 0.02 Ttr2B 0.02 Ttr2C 0.02

Ttr2D Ftr2A Ftr2B Ftr2C Vc1offD Vrc1A Vrc1B Vrc1C Vc2offD Trc1A Trc1B Trc1C Vc1onD Vrc2A Vrc2B Vrc2C Vc2onD Trc2A Trc2B Trc2C TthD Th1tD Th2tD TvD

5 0.5 0.3333 0.1667 0.4 0.9 1.1 0.9333 0.4 9999 0.1083 0.01 0.45 0.9333 0.85 0.85 0.45 0.01 99999 99999 25 0.4 2.25 0.025

This simulation is a close fit, but it is still not perfect. The largest feature gap is the final steadystate reactive power flow Q2-5. At time 50 seconds the reactive power flowing from bus 2 to bus 5 is -303 MVar. The actual measurements show this reactive power flow around time 50 seconds should be approximately somewhere between -135 and -180 MVar. Therefore the difference between the simulated final Q2-5 and actual measurement is off by about 123 to 168 MVar. Coincidentally the SVD at bus 6 at time 50 seconds is injecting 120 MVar and the shunt at bus 6 is injecting 60 MVar. In total the two are injecting 180 MVar. We ran simulation with and without the SVD, and the final reactive power flow Q2-5 is correct when the SVD is absent, however the initial steady-state value is then increased and is incorrect. Therefore we hypothesize that the SVD model does not accurately represent the action of this device during the event. Either the device is controlled in a different manner in the response to the disturbance, or it may have been switched out after the fault.

4. Conclusions
In this study we have estimated composite load model parameters to match FIDVR event data. We examined two data sets and three different network models. In all cases the simulations were greatly improved through adjustments of load model parameters. While we were successful in better matching simulations to data, we offer some cautionary observations. There are a lot of parameters that can be varied, and it is likely that many parameter sets yield similar results. In the first case we considered we identified parameters to which the simulations were most sensitive and changed those first. In all cases we varied parameters from an initial set a set that was presumably considered plausible. Therefore the solutions are biased by this initial set, which we argue is a sensible and practical approach. Further research is needed to clarify the space of nearby parameter values that yield similar results. Also research is needed to help guide any automated procedure with respect to potential discontinuous actions. The refinement of automated algorithms continues in our on-going work.

You might also like