A Data-Driven Modeling Method of Virtual Synchronous Generator Based on LSTM Neural Network
Jiangbin Tian, Guohui Zeng, Jinbin Zhao, Xiangchen Zhu, Zhenhua Zhang
Abstract
The virtual synchronous generator (VSG) exhibits high-dimensional complexity, necessitating a tradeoff between accuracy and complexity when constructing small-signal models. To address this issue, this article proposes a data-driven modeling approach based on long short-term memory (LSTM) neural networks. The focus is on mapping relationships between electrical quantities, considering the influence of irrational factors on model accuracy. A detailed data-driven modeling approach for VSG is proposed and verified in this article. Due to the time-series correlation in the electrical data generated during VSG operation, the LSTM algorithm, known for its excellent time-series prediction capabilities, is chosen to construct the VSG data-driven model. Several complex VSG application scenarios are used to validate the effectiveness and accuracy of the proposed modeling approach. In conclusion, the LSTM-based VSG data-driven modeling outperforms small-signal models and recurrent neural network data-driven models in terms of accuracy and stability.