Dynamic Equivalent Modeling for Microgrids Based on LSTM Recurrent Neural Network

Abstract

An equivalent modeling method of microgrid based on LSTM recurrent neural network suggests in this paper. The advantage of neural network in dealing with nonlinear systems is utilized to solve the equivalent modeling problem of grid-connected microgrids. The equivalent model of microgrid based on LSTM recurrent neural network is established by collecting the current and power data of the common coupling point(PCC) between microgrid and distribution grid. According to the modeling requirement, design a neural network structure with 3 inputs and 2 outputs. In the process of training LSTM recurrent neural network, the current and power of the PCC are taken as the input and output of the network respectively, and the switching power of the microgrid and the distribution grid is taken as the evaluation index of the accuracy of the equivalent model. The proposed LSTM microgrid equivalent model is proved to be applicable and accurate by constructing a microgrid model comprising distributed generation systems in PSCAD4.5.