Study On Prediction Of Remaining Useful Life Of Lithium-Ion Battery Based On Particle Filter

As the lithium-ion battery related technology become more and more improved,it is increasingly used as an energy storage device in power grid energy storage,electric equipment,electric vehicles,military equipment and other fields.Lithium-ion batteries,which are the core components of all kinds of electrical equipment,are essential for long-term safe and smooth operation.During the use of the battery,its capacity will gradually degrade,and the internal resistance will continuously increase.When the battery decays to a certain degree,it can no longer be operated as a stable power supply device.At this time,it should be replaced immediately,otherwise it will bring disastrous consequences.In the normal operation of the battery,the time for which the battery can continue to work normally is estimated,and the remaining service life is predicted,which can provide the maintainer with timely information to facilitate the necessary maintenance or replacement of the battery at the first time.It is of great significance to avoid the occurrence of adverse consequences,and it is also a very important research field at present.In this paper,lithium cobalt oxide battery is taken as the research object,and the degradation process and mechanism of the battery are analyzed in detail.Profiling the external factors such as the temperature,discharge rate,depth of discharge of the battery and the electrochemical reaction mechanism inside the battery occur at the surface of the electrode and inside the electrolyte.Revealing effects of internal and external factors on battery degradation.Based on the charge and discharge cycle data of lithium cobalt oxide battery published by NASA and CALCE,this paper uses the standard particle filter algorithm（PF）and the improved support vector regression particle filter algorithm（SVM-PF）to predict the remaining cycle life of the above batteries.The prediction of different test data validates the universality of the algorithm.On the other hand,the standard particle filter algorithm and the support vector regression particle filter algorithm were used to predict the same data,and the advantages and disadvantages of the two algorithms in the prediction accuracy were compared,and the experimental basis for the actual vehicle on-line prediction algorithm selection was provided.The research results show that the standard particle filter algorithm is very suitable for predicting the remaining cycle life of the battery,and its prediction accuracy is high.The relative errors of the battery life prediction results for NASAB0005,B0006,and B0007 are 23%,30% and 26% respectively.The relative error of the life prediction results of the CS2₃3,CS2₃4,CS2₃6,and CS2₃7 four batteries based on the CALCE publication are 1.85%,1.96%,2.88%,3.26%.In contrast,support vector regression particle filtering is superior to standard particle filtering in predicting speed and accuracy.For NASAB0005,B0007,and B0018 batteries,the relative errors predicted by the support vector regression particle filter algorithm are 20.51%,24.69% and 8.33% respectively.For the CS2₃3,CS2₃4,CS2₃6,and CS2₃7 batteries released by CALCE,the relative errors of the four battery life predictions using the support vector regression particle filter algorithm are 0.93%,0.98%,1.92% and 2.17% respectively.It is proved that support vector regression particle filter has better prediction performance than standard particle filter,and its prediction accuracy and adaptability to severe fluctuation data are obviously better than standard particle filter algorithm.Therefore,support vector regression particle filter as an emerging algorithms,which can be more widely used in data prediction and related fields.