Research On The Prediction Of The Remaining Using Life Of Lithium-ion Batteries Based On Multi-kernel Relevance Vector Machines

In order to ensure the safe and stable operation of lithium-ion batteries,it is necessary to perform fault prediction and health management.Among them,the remaining useful life(RUL)prediction is the core function of the battery management system.Based on the multi-kernel relevance vector machine model,this paper studies the RUL prediction of lithium-ion batteries.The main research work includes the following aspects:(1)In order to solve the problem that direct degradation parameters such as capacity or impedance are difficult to measure,the initial indirect aging feature set of lithium-ion batteries is constructed in this topic.Starting from the data of charging current,charging voltage and discharging voltage,the regular changes under different cycles are studied.Through observation and analysis,the basic features of two charging processes were extracted respectively,namely,the three isobaric rise charging time features representing the initial,middle and end of battery charging,and the three isobaric discharge voltage difference features representing the initial,middle and end of battery discharge were analyzed,so as to construct the initial indirect aging feature set.(2)In order to solve the problem of lack of good generality of initial indirect aging indexes,a framework of fusion aging factors based on initial indirect aging characteristics of lithium-ion batteries was designed.Firstly,Box-Cox linear transformation was used to extract8 groups of initial aging features for nonlinear transformation,and a progressive linear model between initial aging features and capacity was obtained to improve the correlation between initial features and capacity.Then,the typical correlation is used to reduce dimension fusion of the initial feature method after nonlinear transformation,and the typical variable with the greatest correlation with capacity is obtained.Finally,the fusion aging factor which can replace the direct aging index is obtained as the input of the prediction model.(3)A new fusion kernel function was designed to represent the capacity degradation trend of lithium-ion batteries,which improved the prediction performance of the relevance vector machine model.By analyzing the principle of relevance vector machine and the degradation process of lithium-ion battery,the linear kernel captures the monotone decline trend in the degradation process,the Gaussian kernel captures the capacity rebound phenomenon and the polynomial kernel increases the general performance of the prediction model,and then the fusion kernel function is constructed.In the form of linear combination,the fusion kernel function combines the advantages of the three kernel functions to simulate the capacity decay process of lithium-ion battery,so as to improve the problems of the relevance vector machine model and improve the prediction accuracy of the model.(4)An improved Gray Wolf constrained optimization algorithm(IGWCO)was designed to optimize the fusion kernel function combination coefficient and kernel parameters.In order to solve the problem that the gray Wolf optimization algorithm cannot be applied to constraint problems,the nonlinear multi-stage penalty function method is introduced to improve the gray Wolf optimization algorithm,so that it can be applied to constraint optimization problems.At the same time,in order to improve the poor diversity of the initial population and easy to fall into the local optimal problem,a better point set initialization method and a further improved algorithm of nonlinear convergence factor were introduced,so as to construct IGWCO algorithm to optimize the kernel parameters and combination coefficients.Finally,an improved multi-kernel relevance vector machine(IMKRVM)model was constructed to predict the RUL of Li-ion batteries.Based on the above research,the fusion aging factor and IMKRVM prediction models which can accurately describe the degradation characteristics of lithium-ion batteries are established in this project,and the high precision estimation of RUL is realized.Experimental verification shows that under various temperature conditions and different working conditions,the RUL estimation error is less than 20 cycles,and the probabilistic prediction results can be obtained.The reliability of the model is high,which provides an effective basis for the safety management of lithium-ion batteries.