Remaining Useful Life Prediction For Lithium-ion Battery Based On Sparse Gaussian Process Regression

With the development of social economics and energy storage technology,lithium-ion battery is widely used in various fields of the whole society because of its high stable voltage,longevity,light weight,environment-protecting and other advantages.But the performance of lithium-ion battery will degrade under continuous charging and discharging.It is of great significance to predict the Remaining Useful Life(RUL)of battery based on its history data.In this thesis,Gaussian Process Regression(GPR)is applied to prediction.GPR is one of the data-driven approaches which is able to describe the probability meanings of predictions.But GPR sometimes runs into computational problems.To solve the problems,two new algorithms are proposed,which speed up both training and prediction.And one of them is able to process newly acquired data and update prediction model.The new algorithms have practical significance to improve performance of GPR,which also make it easier to predict the RUL of lithium-ion battery.The thesis emphasized on several aspects as follows:(1)Since GPR cannot feasibly be applied to big and growing data sets,this thesis proposes a new sparse GPR algorithm called Two-step Gaussian Process Regression(TGPR)which optimizes regular GPR algorithm by using the inducing inputs,speeding up both training and prediction.Then TGPR is applied to two models,the experimental results compared with regular GPR algorithm show that TGPR is faster and more accurate.(2)Since GPR runs into high time complexity problem when using newly acquired data to update previous prediction model,this thesis proposes a new GPR algorithm based on TPGR called Fast-update Two-step Gaussian Process Regression(FTGP),which calculates a posterior distribution to incorporate new data and to update the prediction model faster.Then FTGP is applied to two models,the experimental results compared with regular GPR algorithm show that FTGP gets an excellent performance to process newly acquired data and update prediction model.(3)Both TGPR and FTGP are applied to predict the RUL of lithium-ion battery.The experimental results show that,TGPR makes it faster to predict the RUL while FTGP makes it possible to process newly acquired battery data and update prediction model.It also shows that both of new algorithms make a precise prediction of the RUL of lithium-ion battery.