Multi-Step Ahead Battery SOC Estimation Using Data-Driven Prognostics and Health Management

This paper proposes a data-driven multi-step ahead battery state of charge (SOC) forecasting system that can be used for prognostics and health management (PHM) of a battery management system (BMS).Two long short-term memory (LSTM) recurrent neural networks (RNN) are implemented and tested, due to their unique capability to, at the same time, use a high number of past time steps and forecast a horizon of interest at real-time.This online inference is then capable to provide an advisory window in case the BMS needs to take any preventive action.The LSTM models, a stacked-LSTM and a Bidi-LSTM, are compared to a statistical-based algorithm which uses a combination of autoregressive integrated moving average (ARIMA) with a polynomial regressor that fits measured variables into the battery SOC.The three methods are tested and validated against a wealthy battery dataset and results demonstrate the feasibility of using the Bidi-LSTM RNN as multi-step ahead SOC forecast estimator.