Deep learning-based state of charge estimation method for real-world vehicles with LFP batteries
Electric vehicles equipped with lithium iron phosphate(LFP)batteries exhibit a flat period in the state of charge(SOC)versus open-circuit voltage curve within the SOC range of 20％to 95％.This makes it challenging to accurately estimate SOC using model-based methods.Data-driven ap-proaches face difficulties in obtaining accurate sample data from the real world and simulating real-world battery system operation in experimental environments.To address this,this paper proposes a deep learning-based SOC estimation method specifically designed for LFP batteries in real-world ve-hicle applications.The method utilizes operational data from real-world LFP battery systems in ve-hicles to accurately label SOC using a reverse ampere-hour integration technique.It introduces a novel CNNGRUM(Convolutional Neural Network-Gated Recurrent Unit)model to predict SOC,combining multiple layers of convolutional neural networks with multiple layers of gated recurrent units.The model leverages four features:current,voltage,temperature,and charge/discharge process ampere-hour integral to estimate SOC for LFP batteries.The proposed method was trained and vali-dated on 20 electric vehicles equipped with LFP batteries in real-world conditions,achieving SOC es-timation with a maximum absolute error of 2.85％,root mean square error(RMSE)of 0.61％,and mean absolute error(MAE)of 0.42％.
lithium iron phosphate batterySOC estimationdeep learningelectric vehicle
万方数据
维普
