State of health estimation and prediction of electric vehicle power battery based on operational vehicle data

With the rapid development of new energy vehicle industry, power battery is an important power source for new energy vehicles. Effective estimation and prediction of power battery health state (SOH) can help companies to effectively estimate and predict the health state of power battery, so as to ensure the safe operation of new energy vehicles. In this paper, we propose a SOH estimation and prediction method based on a long short-term memory network (LSTM) with time series model, and this method uses multi-source features. We extract potential health features from three perspectives and design the LSTM network model to construct a nonlinear mapping relationship between health features and SOH. To better exploit the battery time series information for SOH prediction, we built a time series prediction model containing trend, cycle and holiday models, and used particle swarm algorithm for multi-model optimization. In order to fully exploit the driver usage behavior and time and other information contained in different charge/discharge cycles, where the cycle model is built to include year, month, week, day, etc., SOH prediction can be performed for each future day without changing the original trend of the feature. The final model validation is performed on two vehicle validation datasets. The experimental results show that the model built in this paper outperforms traditional LSTM, GRU, BP and other network models in terms of accuracy of SOH evaluation and prediction.