Robustness-Improved Method for Measurement-Based Equivalent Modeling of Active Distribution Network

Abstract

Accurate equivalent model can be efficient to analyze the dynamic properties of active distribution network (ADN) as well as assess their impacts on stabilities of interconnected power system. However, due to the stochastic nature of renewable resources and time-varying configurations of load conditions, traditional ADN equivalent model may not be robust enough to different operation conditions. To overcome the limitations, this article proposed a robustness-improved method for dynamic equivalent modeling of ADN. To sketch out the most representative operation conditions of ADN, two-step clustering method with fisher discriminant analysis are used for grouping of operation conditions featured characteristic data sets. With the key parameter based identification technique applied, the multiple solution issue in parameter identification process could be effectively avoided. To further enhance the robustness of equivalent model, long short-term memory neural network is adopted to generalize the identified parameters. The performance of proposed modeling method is comprehensively evaluated by an actual ADN based verification cases.