Research On Remaining Useful Life Prediction Method For Lithium-power Batteries

Owing to the advantages of high energy density,long service life,low self-discharging rate,and no pollution,Li-power batteries(Li PBs)are broadly worked in electric vehicles(EVs),energy storage power plants,and aerospace scenarios.The remaining useful life(RUL)of Li PBs can visualize the current state of battery lifetime and measure the cycle time and safety performance of EVs,which is the most important index.However,the health status of Li PBs is declining in service,characterized by discharging capacity loss and shortened service life.Therefore,in order to warn of battery failure and ensure the safety of EVs,it is necessary to develop a methodology that can accurately forecast the Li PB RUL.Aiming at the real-time and accuracy of Li PBs RUL prediction,this thesis analyzes the aging mechanism of Li PBs and proposes Li PBs RUL prediction method based on hybrid model and small sample dual data-driven.The main research contents are summarized as follows.(1)The analysis of aging mechanisms and characteristics for Li PBs.Considering the aging process of Li PBs,describing the internal material and structure,and then theoretical analyzing the capacity failure mechanism.Based on the high-performance battery test platform built in the laboratory,LPBs degradation experiments are conducted to clarify that the charging and discharging multiplier is directly proportional to the battery aging rate,and the dramatic change of battery operating temperature will accelerate the battery lifespan aging.This provides the foundation for the subsequent research on the effective RUL prediction method for LPBs.(2)Li PBs RUL prediction method based on improved variational modal decomposition with hybrid learning.To address the problem of strong nonlinearity and deep noise in the aging process of Li PBs,this thesis integrates the empirical model approach with data-driven to propose a Li PBs RUL prediction method.The posterior feedback confidence(PFC)method is proposed to prefer the variational modal decomposition(VMD)modal layer,and the Li PB aging data is reconstructed to generate the degradation trend sequence and residual sequences.The prediction model is established based on particle filter and gaussian process regression,and then calculate the LPB RUL.The experimental results confirm that the prediction error is less than 1.5% for the predicted degradation curve,and the RUL prediction error is within 3cycles,which has excellent prediction accuracy and generalizability.(3)Small-sample dual data-driven Li PBs RUL prediction method.Aiming at the limited computational cost of in-vehicle battery management system,this thesis constructs a new BLS-LSTM fusion neural network model,which uses broad learning system(BLS)to create feature nodes and augmentation nodes,and then outlaw and extend the input layer of long short-term memory neural network(LSTMNN)based on the input health factor.This fusion neural network requires only 25% of the aging data set to train the model,and effectively predicts the RUL of LPBs with low computational cost and good robustness.