State Estimation And Life Prediction Of Lithium-ion Phosphate Battery Under Abnormal Working Conditions

Accurate and reliable state estimation and life prediction of lithium-ion battery have great significance for ensuring the safety,cruising range and energy management.To address the challenge of state estimation and life prediction for electric vehicles under abnormal working conditions,several studies are investigated,including state of charge(SOC)estimation of lithium-ion phosphate batteries under complex working conditions,aging mechanism identification and small sample state of health(SOH)estimation under slight overcharge cycling,aging analysis and nonlinear SOH estimation under high rate discharge cycling,and the multi-scale data-driven approach for remaining useful life(RUL)prediction considering uncertainty.Firstly,with the short-term cycling,the problems of temperature sensitivity and the flat open circuit voltage curve of lithium-ion phosphate batteries are conducted.A novel method based on nonlinear auto-regression with exogenous input(NARX)dynamic neural network and Sage-Husa adaptive Kalman filter is proposed.To address the parameter sensitivity of NARX dynamic neural network,the sine cosine algorithm is adopted to obtain the optimal parameters.Then,the offline training model is established based on the improved NARX,and the adaptive Kalman filter is selected for online estimation,which obtain a closed-loop information fusion estimation strategy.The observed data from 26650 and 18650lithium-ion phosphate batteries at 10 °C,25 °C and 40 °C has been applied to verify the accuracy and robustness of proposed method under four working conditions.The proposed method can achieve strong robustness and high accuracy in SOC estimation with a maximum relative error below 2%.Secondly,with the long-term cycling,the batteries exhibit different aging states,which cause poor charging matching performance.To this end,the slight overcharge phenomenon will occur.The lithium-ion phosphate batteries are investigated under slight overcharge voltage(i.e.,3.65 V,3.75 V and 3.80 V)cycling.Based on the battery fading analysis,incremental capacity analysis and differential voltage analysis,the slight overcharge aging mechanism of the battery is preliminarily identified.Combined with battery disassembly,the electrode material performance is investigated.The aging mechanism of lithium-ion phosphate battery under slight overcharge cycling is revealed based on X-ray diffraction,scanning electron microscope and energy dispersive spectroscopy analysis.The matching relationship between the incremental capacity curve under slight overcharge and the aging of the positive and negative electrodes are explored based on aging mechanism.The potential health indicators are extracted based on the charging capacity increment curve and gray relation analysis is utilized.To reduce redundant information among various features,the principal component analysis is adopted to obtain the syncretic health indicator.The Gaussian process regression is established for SOH estimation under small sample with slight overcharge cycling.The proposed method can achieve high accuracy and strong reliability in SOH estimation.Thirdly,the aging mechanism of lithium-ion phosphate battery is conducted under high rate discharge conditions,the batteries are selected at different discharge rates(i.e.,1 C,2 C and 3 C)and the experimental data is collected.Through the characterization of the external characteristics of battery attenuation and non-destructive analysis,the aging mechanism of high rate discharge cycling is initially identified.Combined with the structure,morphology and element analysis of the electrode material after the battery disassembly,the aging mechanism of lithium-ion phosphate battery is clarified under high rate discharge conditions.Meanwhile,the incremental capacity curve,charging time and voltage are extracted as potential health indicators.To obtain the low-dimensional and low-noise health indicator,a stacked autoencoder neural network is proposed to obtain the fusion health indicator.For SOH estimation with nonlinear and external interference under high rate discharge conditions,the Gaussian mixture regression is established,which shows that the proposed method has accurate and reliable SOH estimation.Finally,capacity regeneration after a static and pause period and uncertainty quantification in different working conditions are investigated for time series future value prediction.A novel multi-scale data-driven approach with optimal health indicator is proposed to obtain accurate and reliable RUL prediction.The optimal charging voltage time interval is obtained based on the charging voltage model as the health indicator for RUL prediction.The variational mode decomposition is selected to multi-scale decompose proposed health indicator as intrinsic mode functions and residual term.A prediction method with uncertainty is established based on Gaussian process regression to characterize capacity regeneration and random disturbance.The Dropout Monte Carlo gated recurrent unit neural network is proposed to establish uncertainty prediction model with global degradation.The prediction results of the intrinsic mode functions and residual item prediction results are integrated to obtain the RUL prediction.The proposed method can obtain high accuracy and strong robustness in RUL prediction with root mean square error limited below 3%.