Comparison of Equivalent Circuit Model and LSTM-Based Model for Battery Condition Estimation of Energy Storage Systems

To monitor and analyze the state of energy storage batteries, two methods for estimating the state of charge (SoC) are proposed: (1) circuit model-based SoC estimation and (2) machine learning-based SoC estimation. For the circuit model-based method, an equivalent circuit for energy storage battery is modeled using resistors and capacitors, and then an extended Kalman filter is utilized to optimize the battery condition estimation. For machine learning-based method, long short-term memory (LSTM) with historical battery data is used to model battery condition estimation. Both methods show good estimation results. However, LSTM exhibits lower error rates in terms of root mean square error (RMSE). Therefore, for energy storage batteries whose performance cannot be tested, historical data-driven LSTM can more accurately estimate the SoC of the battery.