Parameter Subset Selection for Power System Model Calibration Using Both Sensitivity and Identifiability

In power systems, it is essential to ensure that the dynamic model predictions align with the actual measurements, which is usually achieved through model calibration. Due to parameter sensitivity and identifiability, it is often impossible to estimate all of the model parameters, resulting in the performance of model calibration highly dependent on the selection of the parameters for estimation. In this paper, we propose a methodology for parameter subset selection for model calibration as a combinatorial optimization problem, using the initial guess derived from the parameter sensitivity and identifiability analysis based on the sensitivity matrix. Considering the dependence of unidentifiable parameter sets and sensitivity magnitude, we determine the largest sensitive-identifiable parameter subset that not only has the largest number of parameters satisfying the parameter collinearity constraint but also has the largest sensitivity magnitude. We adopt the variable neighborhood search technique to solve the combinatorial optimization problem, achieving notable efficiency enhancement over the forward selection method. The proposed method is applied to a hydro synchronous generator model. The presented results show that the proposed method is highly time efficient and is able to improve the performance of model calibration. Compared to the forward selection method, our proposed approach has much higher computational efficiency, achieving a speedup of 6,480. When considering the parameter subset identified by our proposed approach for calibration, the performance of the parameter estimation is enhanced in both the accuracy of the estimated parameter values and the consistence between the model outputs and the actual measurements.