Integrating particle swarm optimization with convolutional and long short-term memory neural networks for real vehicle data-based lithium-ion battery health estimation

The State of Health (SOH) of batteries is a crucial determinant for the driving range and safety in electric vehicles (EVs). The emergence of big data technologies has recently made it possible to estimate the SOH of batteries online. However, existing methods often fall short in practical EV settings because traditional theoretical model- based solutions do not take into account driving behaviors and the complexity of environmental influences. To overcome these obstacles, this study introduces a novel approach for estimating the SOH of power batteries. This method utilizes real-world EV operational and environmental temperature data from a national big data alliance of new energy vehicles. It employs an integrated technique combining Particle Swarm Optimization with Convolutional Neural Networks and Long Short-Term Memory networks (PSO-CNN-LSTM). This approach includes extracting various health indicators from historical data, calculating equivalent internal resistance and capacity across different temperatures through methods that incorporate a forgetting factor and ampere-hour counting. SOH estimates for each charge and discharge cycle are made based on internal resistance and capacity values, and a cumulative SOH estimate is derived from these individual estimates' errors. By employing diverse models to understand the battery's degradation pattern and forecast future deterioration, this method has been tested with actual data, demonstrating high accuracy and broad applicability in predicting battery health in real-world EV scenarios, with a maximum estimation error of about 3 %.