Lithium-ion Battery Remaining Useful Life Prediction Integrating LSTM And PF

With the advantages of high energy density,high reliability and long lifetime,lithium-ion batteries are broadly used in electric vehicles.With the repeated charges and discharges,the battery performance continues to degradate which may easily result in system failures and accidents.Remaining useful life(RUL)of lithium-ion battery is an important indicator for evaluating battery performance.Accurate prediction of battery RUL is of great significance for ensuring safety,which has attracted widely attentions and has become a research hotspot.Among them,long short-term memory(LSTM)neural network has achieved good RUL prediction results on experimental data.However,due to that the accumulated errors occur when LSTM is employed to predict the battery RUL and there are a lot of noise data and abnormal values on electric vehicle operating data,the RUL prediction accuracy is seriously reduced.In order to eliminate the abnormal values on electric vehicle operating data and make the data change smoothly,an anomaly detection algorithm based on LSTM is proposed.In order to improve the RUL prediction accuracy,a RUL prediction algorithm integrating LSTM and particle filter(PF)is proposed.The main work of this thesis is as follows:(1)An anomaly detection algorithm based on LSTM(AD-LSTM)is proposed to eliminate the abnormal values on vehicle operating data and make the data change more stable.The LSTM prediction model is employed to fit the changes of battery parameters.The sampling points where the errors between the predicted values and true values exceed the threshold are regarded as abnormal points.And,the error threshold is determined automatically based on the optimization condition maximizing the difference between the error sequences before and after eliminating the least anomalies.Then,the abnormal values are replaced by the predicted values.The experimental results on the true operation dataset(TOD)show that the smoothness of data is significantly improved,and the correlation between the degradation parameter and capacity is enhanced after preprocessing,providing data basis for accurate RUL prediction.(2)A lithium-ion battery RUL prediction algorithm integrating LSTM and PF(LSTM-PF)is proposed to improve the RUL prediction accuracy.The health features strongly related to capacity are extracted from serval parameters based on Spearman correlation analysis and are incorporated into a health indicator based on principal component analysis.The LSTM prediction model is employed to fit the degradation trend of health indicator.The state prediction ability of PF is employed to provide calibration parameters for correctting the accumulated errors of LSTM.The prediction accuracy is imporved by integrating LSTM and PF.The experimental results show that the minimum root mean square error of the relevant RUL prediction methods on the NASA dataset is 0.0186,and the root mean square error of LSTM-PF is 0.01236.The root mean square error of LSTM-PF on the TOD dataset is 2.268641,which is the smallest compared with other methods.Therefore,the LSTM-PF prediction method proposed in this thesis effectively improves the RUL prediction accuracy.(3)A prototype system for lithium-ion battery remaining useful life prediction is designed and implemented.The system has serval functions such as data preprocessing,HI construction,RUL prediction and visualization display.It is convenient for users to analyze the original data and prediction results which are displayed visually.