Simultaneously estimating two battery states by combining a long short-term memory network with an adaptive unscented Kalman filter

Accurate state of charge (SOC) and state of energy (SOE) estimations for lithium-ion batteries (LIBs) are of great significance in battery management system (BMS). Especially, the two battery states estimation with SOC and SOE at the same time, can promote the battery life and ensure the system reliability of LIBs. In this work, a novel long short-term memory network combined with an adaptive unscented Kalman filter (LSTM-AUKF) method is proposed to estimate SOC and SOE simultaneously. The proposed LSTM-AUKF method is validated with several dynamic driving schedules (dynamic stress test, US06 test, and federal urban driving schedule) under different temperatures and different initial errors. Experimental results reveal that the proposed method can effectively co-estimate the SOC and SOE for the LIBs with high accuracy and low complexity. The recorded root means square error (RMSE) and mean absolute error (MAE) of SOC are controlled within 0.43% and 0.41% respectively. Meanwhile, the RMSE and MAE of SOE estimation are less than 0.46% and 0.44%, respectively. Furthermore, the proposed LSTM-AUKF has been compared with single LSTM, LSTM combined with an unscented Kalman filter and other methods for SOC and SOE estimation, the results indicate that the proposed method has excellent performance in reducing computation complexity and enhancing estimation accuracy.