Parameter estimation of a composite load model using a hybrid approach

Power system;Load modeling;composite load (CLOD) model;Digital Fault Recorder;Fault;Pseudomedian filter;Measurement based;Minimization

Guo, Siming ; Overbye ; Thomas J.

关键词: Power system;    Load modeling;    composite load (CLOD) model;    Digital Fault Recorder;    Fault;    Pseudomedian filter;    Measurement based;    Minimization;   

Others  :  https://www.ideals.illinois.edu/bitstream/handle/2142/34251/Guo_Siming.pdf?sequence=1&isAllowed=y

美国|英语

来源: The Illinois Digital Environment for Access to Learning and Scholarship

【 摘 要 】

Accurate dynamic load models are needed to ensure meaningful transient simulation results. We investigate the use of a composite load (CLOD) model to represent an aggregate load. Two algorithms for the parameter estimation of the CLOD model are presented and analyzed. The first uses nonlinear optimization and computes a solution with 2.02% error for a validation case, but requires a 49 dB signal-to-noise ratio (SNR) to achieve this. The second method uses least squares and has an error of 11.7% for the same case, but only requires an 18 dB SNR. Higher accuracy is obtained by setting tighter bounds on the parameters used in the simulations and by using more simulations, up to a cap of approximately 100. In order to apply the above algorithms to digital fault recorder data from a real disturbance, we first convert the three-phase sinusoids to a positive sequence dynamic phasor. The resulting signal has significant noise content, which is filtered using a pseudomedian filter due to its edge-preserving quality. As part of this work, a design methodology for a pseudomedian filter is created, which accepts a cutoff frequency f_c and designs a filter with 3 dB of attenuation at f_c and a stopband at 2.75f_c. A 120 Hz cutoff frequency is chosen for this work. However, even after post-processing, the measured and simulated signals are very dissimilar, resulting in both algorithms failing to identify a reasonable load model. We conclude that a measurement-based parameter estimation method is ill-suited for a complex, nonlinear load model.