State-of-charge estimation for lithium-ion batteries of electric vehicle based on sensor random error compensation

Abstract

Accurate state of charge (SOC) estimation is essential for ensuring the security and efficiency of lithium-ion batteries. Generally, the signal detected by a sensor inevitably has noise and drift. Therefore, the prediction process will indefinitely accumulate this random error, resulting in lower estimation accuracy, which is particularly prominent when the performance of vehicle sensors is poor. To solve this problem, a method combining an adaptive extended Kalman filter (AEKF) with a long short-term memory (LSTM) network is proposed in this paper. The original data with random errors observed in each period is denoised by the AEKF; then, the processed values are served as the input of the LSTM neural network. In this way, it avoids the disadvantage that the LSTM depends too much on the accuracy of training data and retains the advantages of the high robustness of the AEKF and strong nonlinear characteristics of the LSTM network. The proposed method is compared with an LSTM method without pre-processing the input data on the New European Driving Cycle (NEDC) and the latest World Light Vehicle Test Cycle (WLTC) datasets. Subsequently, the experimental results are analysed and compared with other algorithms under various initial noise, temperatures, and unknown initial SOC conditions. The results show that the accuracy of SOC estimation is improved a lot in the case of compensating the random input error by the AEKF in advance in comparison with feeding it into a neural network algorithm directly, converging quickly under inaccurate initial conditions and having strong robustness against noise and other factors. In particular, the root-mean-square error is less than 0.6 %, and the maximum error is less than 1.6 %.