Dynamic Equivalent Model of Active Distribution Network Based on Improved Reinforcement Learning Algorithm

Abstract

Modeling of active distribution network (ADN) has received a great change due to the appearance of dynamic phenomena arising from the increasing penetration levels of distributed generation. In order to deal with the problem of time variability of ADN, a dynamic equivalent model based on a load model pool is proposed, which includes direct-drive wind turbines, photovoltaic power generation and comprehensive load. On the aspect of parameter identification, this research improves a conventional reinforcement learning algorithm by introducing an adaptive learning rate and adding new search directions. In order to solve the problem of variation of load components, firstly, a generalized load model parameter library is established taking-into account different load levels, then update weights of every generalized load model in real time. The results on an 11-bus system demonstrate that the improved reinforcement learning algorithm can effectively carry out dynamic equivalent modeling of the ADN integrated with renewable energy sources.