Research On Prediction Method Of Remaining Useful Life Of Lithium Ion Battery Based On Fusion Type Data Driven

With the increase in the use of lithium-ion batteries in life and production,it has been involved in housing,transportation,and high-tech fields such as aerospace and satellites.The accidents caused by the failure of lithium batteries have occurred from time to time,and the losses are alarming.Therefore,the determination of the health status of lithium-ion batteries and the study of the remaining life prediction problems can effectively ensure the stable operation of various systems of the lithium batteries,and reduce potential safety hazards.Therefore,research on it has great practical significance.In this thesis,lithium-ion battery is taken as the research object,and its characteristics,health status characterization and remaining life prediction methods are studied.The main research work is as follows:(1)First of all,according to the working principle of lithium-ion batteries,the characterization of the health status of lithium-ion batteries and the causes of failure were analyzed.The fitting of lithium-ion battery capacity data obtained from accelerated test of lithium-ion battery life at the University of Maryland Advanced Life Cycle Engineering Research Center was performed.The relationship between lithium-ion battery capacity and depth of discharge is obtained,providing experimental evidence for the establishment of the following capacity degradation model and the study of remaining life.Then based on the above studies,various models for lithium-ion lifetime prediction were further analyzed.Finally,the lithium battery capacity decay model was selected as the basis for particle filter(PF)-based lithium-ion battery life prediction.(2)Data-driven Least Squares Support Vector Machine(LSSVM)algorithm is a new type of machine learning method.This algorithm does not need to rely on accurate physical models when performing regression prediction.With a small sample size,good predictions can still be achieved.However,higher prediction accuracy is based on the selection of appropriate penalty and nuclear parameters.To solve this problem,this thesis introduces Particle Swarm Optimization(PSO)algorithm to optimize the parameters of LSSVM.In order to avoid the PSO algorithm is easy to fall into premature problems,this thesis also proposes an improved algorithm CPSO,which isused to optimize the parameters of LSSVM,and applied to the field of remaining useful life(RUL)prediction of lithium-ion batteries.(3)Although the LSSVM algorithm can accurately predict the RUL of lithium-ion batteries under most application conditions,the lithium-ion battery has many applications and environmental complexity,so due to uncontrollable noise and observation error,it will inevitably affect the prediction accuracy of RUL of the lithium-ion battery,and the prediction results will have hidden risks of high uncertainty.The model-driven PF method can give the probability density distribution of prediction results with certain confidence.This method has a more accurate expression effect for the posterior probability density based on control quantity and observation.At the same time,PF does not need to consider whether the random variable is Gaussian when predicting nonlinear systems.Therefore,the paper proposes a PF algorithm based lithium-ion battery RUL prediction method for the representation of the uncertainty of the prediction results.(4)Although the PF algorithm can express the uncertainty of the prediction result,the PF algorithm introduces the resampling mechanism to solve the defect of the degradation of the particle,and it also causes the problem of particle depletion.Because re-sampling will copy particles with high weights many times,this will make the particles lose diversity.Especially in a small noise environment,all particles will degenerate into one particle after multiple iterations.To solve this problem,this thesis introduces a method based on data-driven LSSVM algorithm and PF algorithm.The LSSVM algorithm is used to train the data obtained by historical measurement,and then the future measurement data is predicted.Then the PF algorithm is integrated to perform RUL prediction,which not only avoids the problem of particle shortage in the PF algorithm,but also enables the prediction result to express uncertainty.