Study On State Of Health Assessment And Remaining Useful Life Prediction Of Lithium-ion Battery

Lithium-ion batteries are widely used in the power system of electric vehicles by reason of their superior performance.But their safety has always been one of the core issues of electric vehicles and has always been concerned by the public.Being able to monitor the state of health(SOH)of lithium batteries in a timely and effective manner is of great help to the real-time driving of electric vehicles.At the same time,the prediction of the remaining useful life(RUL)of batteries can help users to study and judge in advance and effectively formulate battery replacement strategies.Based on these two aspects,this paper takes the NASA power lithium-ion battery aging experimental data as the research object,and does not rely on the traditional battery degradation model,The following work has been done with the idea of data-driven:(1)Starting from the characteristics and aging mechanism of lithium batteries,the internal and external factors affecting the capacity decline process were analyzed and studied.The SOH was defined by capacity,and the characteristic curves of temperature,current and voltage during the battery charging and discharging cycle were drawn.Five groups of characteristic parameters were extracted,including charge time of peak temperature,average discharge temperature,charge time of equal current drop,charge time of equal voltage rise and average discharge voltage.Then,parameters with a coefficient greater than 0.8 are selected through Pearson correlation analysis as health factors that can characterize the battery capacity decay laterally,and used for real-time assessment of battery SOH.(2)A SOH evaluation method based on Particle Swarm Optimization(PSO)for Extreme Learning Machine(ELM)is proposed.ELM requires few parameters to be adjusted,which is easy to calculate and fast to train.However,due to the randomly generated input weight and hidden layer bias,the prediction performance of ELM is affected.In view of the shortcomings of ELM itself and the unstable output of ELM for battery SOH prediction,PSO algorithm is introduced to optimize the parameters of ELM model.Applying the Principal Component Analysis(PCA)to deal with the redundancy between the health factors of the input model,Selecting the first principal component as the input information of the model.In the end,based on the algorithm of PSO which optimize the ELM,the evaluation model is built.According to the experimental results,it reveals that the evaluation error of the proposed method is within 3%,and the accuracy is high.(3)A RUL prediction method based on uncertainty analysis is proposed.The prediction of RUL is essentially the prediction of capacity decline.In order to effectively deal with the fluctuation of capacity local regeneration during battery degradation,the original capacity data is decomposed into several Intrinsic Mode Functions(IMF)and a residual using the Complete Ensemble Empirical Mode Decomposition with Adaptive Noise(CEEMDAN)algorithm,Then using the Long Short-Term Memory(LSTM)neural network sub-model to estimate the residual component with the declining trend of battery capacity,and use the Gaussian Process Regression(GPR)sub-model to fit the high-frequency IMF component caused by local regeneration of capacity,so directly and simultaneously obtain the long-term dependence of capacity and the uncertainty quantification caused by capacity regeneration.Finally,the experimental results show that the proposed method not only has high prediction accuracy,but also can provide the uncertainty expression of the prediction,which solves the one-sided problem of the previous RUL prediction that only has the capacity point estimation,and has good engineering application significance.