Research On Lithium Battery SOC Estimation Method Based On Adaptive Learning

With the popularity of automobiles,energy consumption and environmental pollution have become social issues that cannot be ignored.Electric vehicles(EVs)have emerged as a new energy-saving and environmentally friendly means of transportation.Power battery is one of the core components of EVs,the accurate estimation of State of Charge(SOC)directly relates to the performance and driving safety of EVs.Accurate SOC estimation can prevent abnormal working mode such as overcharge and over discharge.The data-driven method is one of the more widely used SOC estimation techniques,which can well fit the mapping relationship between battery external parameters and SOC.However,traditional data-driven models often ignore the impact of environmental changes on model reliability.First,the EVs work in a wide range of ambient temperatures and dynamic changes in operating conditions.It is difficult for traditional data-driven modeling methods to obtain data covering all temperatures and working conditions through experiments.Second,as the aging and decay of power batteries,the relationship between parameters and SOC changes,leading to the failure of the estimation model,and the accuracy of the SOC estimation decreases.Therefore,in view of the above problems,this paper focuses on the research of SOC estimation methods based on adaptive learning,the main work is as follows:(1)Apply ensemble learning to SOC estimation.When the old ensemble model fails,use the new battery data to train a new ensemble model,and then use the genetic algorithm based on mutual information to select sub-models from the old and new ensemble model to form the final prediction model.In this way,the model is dynamically updated.(2)Improve the Gaussian process regression(GPR)to obtain an SOC estimation method based on online Gaussian process regression(Online-GPR).When new battery data is available,the GPR covariance matrix and hyperparameters are updated by reorganizing the training set to achieve adaptive learning of the prediction model.(3)Conducted a lithium battery charge and discharge experiment,collected experimental data and processed data.The concept drift phenomenon is explained from the perspective of data distribution.Then the above two adaptive algorithms are used to solve the concept drift and improve the accuracy of SOC estimation.Finally,the superiority of the two algorithms is verified through comparative experiments.