State of Health Estimation Framework of Li-on Battery Based on Improved Gaussian Process Regression for Real Car Data

With the diversification of power batteries in the market, the estimation and management of State of Health (SOH) become more and more important. The building of the models based on experimental data needs much time and many resources, and the models do not have strong versatility. In this paper, a data-driven on-line SOH estimation framework is built by using the measured data of vehicle operating process. In case that the charging data is incomplete and with low precision, an indirect feature (IF) extraction and combination scheme based on Incremental Capacity Analysis (ICA) is built, so as to realize the mapping between IF and SOH based on Gaussian Process Regression (GPR), and Multi-Colony Particle Swarm Optimization (MPSO) is adopted to solve the local optimum of GPR hyper-parameters. Finally, after the algorithm is verified and evaluated by applying NASA (National Aeronautics and Space Administration) dataset and real vehicle data, and when the training amount of operation data is enough, the error of SOH estimation can be controlled within 2%, which shows that the algorithm has good versatility and estimation precision.