Research On Failure Prediction Method Of Lithium Ion Battery Based On Integrated Learning And Improved RVM

Lithium-ion batteries have become popular in various devices because of their high energy density,long service life,and stable electrochemical performance compared with other types of batteries.At the same time,the development of hybrid electric vehicles and military drones has received more and more attention.Both technologies rely on lithium-ion batteries.In this case,lithium-ion battery failure will lead to uncontrolled operation,downtime,and even Catastrophic system failure.However,during the recycling process,the performance of the battery is gradually degraded,which shortens the battery life and even poses a safety hazard.Therefore,real-time monitoring and failure prediction of lithium-ion battery health is necessary.This paper mainly studies the failure prediction method of lithium ion battery based on correlation vector machine,and specific content includes:1).The structure and working principle of lithium-ion battery are analyzed in detail,and several main performance indexes of lithium battery are expounded.The performance test and charge-discharge cycle experiment of lithium battery are carried out through independent experimental platform.The effect of discharge rate on battery performance and battery aging are studied.The effect on the discharge voltage and actual capacity.2)Traditionally,the data-driven method based on monitoring battery parameters for degradation state recognition has poor generalization performance,ignoring the difference in individual battery degradation processes and resulting in lack of adaptive ability to different batteries.Aiming at the above problems,this paper selects the battery capacity to characterize its health status,and studies a dynamic integrated online capacity estimation method based on the improved Relevance Vector Machine(RVM).First,after analyzing the battery life state characteristic parameters,the isostatic drop discharge time is selected as the indirect health factor of the lithium battery.Secondly,the Bat algorithm(BA)is introduced to optimize the RVM model parameters,establish the relationship model between indirect health factor and capacity,and construct a wavelet kernel function to further improve the estimation accuracy of the model.Then,consider the difference between lithium batteries.Degradation process,using the degradation data of different batteries in the same batch of batteries,respectively,training the above-mentioned BA-RVM-based model as a sub-model of the integrated model and using the firstorder partial correlation coefficient weighting algorithm to continuously update the weight of the submodel by using online data to realize Online estimation of capacity;finally,the correctness of the dynamic integrated BA-RVM estimation method based on NASA battery data verification.The results show that compared with the traditional single data-driven model method,the estimation method studied in this paper is accurate and effective,with higher precision,more generalization ability and self-adaptive ability,and the estimation result has probabilistic output.3)Aiming at the nonlinear degradation process of lithium-ion battery and the poor performance of traditional RVM model in long-term prediction,this paper proposes an improved bat algorithm(IBB)optimized online learning RVM battery RUL prediction model.Firstly,the phase space reconstruction technology is used to process the historical capacity data of the battery.The method based on the mean entropy is used to determine the optimal embedding dimension and the training data set of the model is constructed.Secondly,an online learning related vector machine model is proposed for the lithium battery RUL.Online prediction is performed,and the model can be automatically updated by using online prediction results.The kernel parameters of the updated model are automatically optimized by the improved bat algorithm.Finally,based on NASA experimental data and independent experimental data,the correctness of the method is verified.Experimental results show that the method has higher prediction accuracy in long-term prediction.