Remaining Useful Life Prediction Of Lithium-ion Batteries Based On Fusion Algorithm

Prognostics is of great significance as increasing requirements for system reliability.This technology is to predict possible failures of the system to eliminate potential safety hazards and reduce casualties and property losses.Prognostics mainly includes early detection of faults and prediction of remaining useful life.Lithium-ion battery is a widely used energy component,so the accurate estimation of its remaining useful life can be replaced in time before the battery fails,which could effectively guarantee the normal operation of the system and avoid serious accidents.This thesis proposes three fusion prediction algorithms of lithium-ion battery remaining useful life based on dual-exponential degradation model.Firstly,the main reason for battery capacity attenuation is briefly analyzed.The degradation model of the battery is discussed,and the remaining useful life prediction of the battery is converted into the battery capacity prediction.Then,the unscented Kalman filter is applied to the life prediction of lithium-ion battery.On this basis,the fusion idea of predicting the residual correction by data-driven method is proposed.The first fusion algorithm obtains a BP neural network to iteratively predict the future residual trend and restore the predictive update function of the filter.Through the simulation experiment of lithium-ion battery capacity data set,four evaluation indicators are introduced to verify the feasibility of this fusion idea.In order to further improve the prediction performance,the second fusion prediction algorithm is proposed.The K-means clustering algorithm is used to adjust the basis function center of the RBF neural network,and the optimized network is used to predict the filter residual value and update the system state and covariance matrix.Then the state estimation of the unscented Kalman filter is modified.In order to improve the prediction performance of the fusion algorithm and the adaptability to different battery individuals,the historical residual decomposition sequence is reconstructed based on the first fusion algorithm and the reliability of the residual data set is improved.The simulation experiments of totally different data sets illustrate the practicability and effectiveness of the proposed fusion algorithm.