Equivalent Modeling Based on Long Short-term Memory Neural Network for Virtual Synchronous Generator

The virtual synchronous generator (VSG) is emerging as an effective approach to controlling the grid-connected inverters, which can provide virtual inertia, handling both reactive and active power by mimicking the behavior of the traditional synchronous generator (SG). Traditional VSG modeling methods cannot simulate the dynamic and nonlinear characteristics of VSG effectively, to solve this problem, an equivalent data-driven modeling method of VSG by using long short-term memory (LSTM) network is proposed. Then, a linearized small-signal model of VSG is established in MATLAB/SIMULINK to obtain operational data as the training and testing data. Finally, simulation comparisons between different data-driven models based on various neural networks are performed to verify effectiveness of the proposed model. The results indicate that LSTM-based equivalent model can reflect the dynamic and nonlinear characteristics of the VSG under different operating conditions with high precision and efficiency.