Life Prediction Of Lithium-ion Batteries Based On Particle Filter And Its Improved Algorithm

The lithium-ion batteries have been widely used in portable electronic devices,electric bicycles, electric vehicles, satellites, aerospace and other fields, and its failure willresult in accidents and huge economic losses, even threatening seriously people’s lifesafety. Therefore, the researches on lithium-ion battery life prediction take effect inensuring the safe operation of the equipment or systems in practical engineering. Particlefilter, as a model-based prognostic approach, has been widely used in real-world usage.However, there are still a number of problems needed to be solved in the process of itsdevelopment.Based on statistical learning theory, a combined prediction method for lithium-ionbattery life prediction using particle filter and support vector regression is implemented inthis paper, which can solve the existent problems in single prediction method based onparticle filter effectively. Firstly, the measuring indicators of battery life are introducedand the main factors leading to lithium-ion battery capacity fade are analyzed in theory.Matlab curve fitting toolbox is used to calculate the parameters of the lithium-ion batterycapacity degradation model. One-step prediction algorithm with measurement updated andmulti-step prediction algorithm without measurement updated based on particle filter arestudied and applied for the lithium-ion battery life prediction. Secondly, considering thedeficiency of prediction method based on particle filter, one-step prediction algorithm withmeasurement updated and iterative multi-step prediction algorithm using SVR particlefilter without measurement updated, which are followed by the analysis and comparisonsof time series multi-step prediction strategies, are implemented and then applied for thelithium-ion battery life prediction. Compared with the prediction method based on particlefilter by simulations, the above methods not only improve the precision of multi-stepprediction, but also possess good tracking performance of battery state in the future time.In the end of this paper, main ideas are summarized and some questions for further studiesare put forward.