Remaining Useful Life Prediction Of Lithium-ion Battery With Capacity Recovery Effect

The main purpose of this paper is to solve the problem that the current remaining useful life prediction of lithium-ion battery usually ignored the capacity recovery effect.The multi-mode model built with the distinction in the battery capacity change modes can obtain RUL more accurately.The model parameters were optimized using the intelligent algorithm,and the overall accuracy was improved.Taking the instrument error and model error into account,we used particle filter algorithm to obtain the prediction results with uncertainty expression.The main contents and conclusions are as follows:(1)Summarized the main methods of RUL prediction of lithium-ion battery and analyzed their advantages and disadvantages.Put forward the existing problems in RUL prediction and established a multi-mode model for battery recovery effect during rest.Summarized the working mechanism,failure mechanism and capacity recovery mechanism of lithium ion battery.The error caused by battery capacity recovery effect was not considered in the past RUL prediction.The NASA battery life test data was visualized and analyzed,and three capacity change modes were found which can completely explain the change process of battery capacity.(2)Three kinds of capacity change models were built respectively.It was found that there was a linear relationship between the underlying capacity and the underlying cycle time.The recovered capacity was exponentially dependent on recovery time.The degradation of the surface capacity was proportional to the cumulative cycle time.The fitting effect of the model in NASA#5 and#6 batteries was good.The model was applied to the prediction of#18 battery data and the result was compared with the other 2 widely used models.The results verify the superiority of our model.(3)Considering the difference between local optimum and global optimum,genetic algorithm and crowd search algorithm were used to optimize the parameters of multi-mode model.Both of them obtained better prediction results,but the crowd search algorithm with adjacent search ability was more advantageous in search speed and goodness of fit.(4)Considering the existence of measurement error and model error,particle filter was applied to predict the life of battery.The result showed that the error was only 2 cycles and the probability distribution of RUL was obtained,which was more instructive for battery maintenance.