A comparative study of machine learning based modeling methods for Lithium-ion battery

Abstract

Abstract A suitable battery model plays an important role in assisting accurate state estimation for power battery used in electric vehicles. This paper compares the applications of four commonly used machine learning methods (decision tree, k-nearest neighbour, support vector machine and neural network) in lithium-ion battery modeling. The adaptability on working condition, temperature and degradation of above four modeling methods are analysed in detail. Results show that neural network performs best when working condition changes. All the models basically have the same performance on adaptability to temperature. The battery dynamic characteristics change significantly in the aging process and it is necessary to include battery test data under different degradation levels into training sets as to obtain a model that can predict the voltage response accurately in various aging states.