A data-driven SOE estimation framework for lithium-ion batteries under drive cycle conditions over wide temperature range

Accurately estimating the state of energy (SOE) of a battery is crucial in battery system management to enhance battery operation's reliability and safety. Due to the sensitivity of SOE to temperature and operating conditions, it is challenging to measure SOE directly. To achieve accurate estimation results, accurate and stable estimation models are indispensable. Therefore, the aim of this paper is to propose a data-driven SOE estimation method for lithium-ion batteries, so as to achieve accurate estimation results. In this study, the method describes SOE based on current, terminal voltage, temperature, and state of charge (SOC) curves of different temperatures and multiple driving cycles, especially the differences in discharge curves between these data. Then, based on the fusion of convolutional neural network (CNN) and bilayer gated recurrent unit (BGRU), a model is designed to enhance the capacity of model to extract sequence features for the SOE estimation in various scenarios. Additionally, the sliding window technique is used to segment the input data, creating a multi-temporal input structure that improves the correlation between battery parameters and optimizes the impact of output on SOE. In order to further verify the estimation performance of the model, experiments are compared under different machine learning models. The results show that the proposed model can provide accurate SOE estimation drive cycle conditions over wide temperature range, and battery material conditions. The RMSE and MAE of the model are limited to 0.7 % and 0.6 %, respectively.