| In order to protect the earth’s ecological environment and respond to the call of energy conservation and emission reduction policies,electric vehicles(EVs)are developing rapidly instead of traditional fuel vehicles.Lithium-ion batteries have become the main force of power battery due to its advantages such as high energy density and long cycle life.With the use of EVs,the performance of lithium-ion batteries is irreversibly aging.The prediction of the remaining useful life(RUL)of power battery can provide a strong guarantee for the reliability and stability of electric vehicles.Therefore,the development of electric vehicles faces a key problem,that is,the state of health(SOH)estimation and RUL prediction in the whole life cycle of power batteries.On the one hand,the power battery has some problems in the vehicle stage,such as poor time-varying prediction,inaccurate long-term prediction,and inconsistent battery pack parameters.On the other hand,most of the research is still in the vehicle stage,and there is less research on the state estimation and RUL prediction of retired batteries.The traditional charging and discharging test methods are not suitable for retired batteries.Faced with the above challenges,this paper is based on the whole life cycle of power lithium-ion batteries from the on-board stage to the retirement stage,taking single batteries and battery packs as the research object,and exploring the state estimation and life prediction methods of batteries in different life stages and different research objects.The main research work and innovations of this paper are as follows:(1)In order to solve the problem that the capacity state and model parameters of lithium ion batteries are difficult to update online,an online prediction method of battery residual life based on dual fusion method is proposed.The phase space reconstruction technology is used to deeply explore the battery degradation time series data.The fusion framework refers to the use of data-driven models to provide future observations for model-based filtering algorithms,effectively overcoming the shortcomings of a single method.A dual fusion algorithm is used to realize the synchronous update of battery capacity and data-driven model parameters.The one step predicted value is iterated to the data-driven model to realize the online prediction of the RUL.Taking NASA # 5 battery as an example,when the prediction starting point is 100 cycles,the RUL prediction error of the proposed method in this paper is 0.8%,which is 3.8% lower than that of the fusion method and 6.2% lower than that of the signal method.The simulation results show that the prediction of the proposed dual fusion method is accurate,and the prediction results become more and more accurate with the increase of the prediction starting point.(2)In view of the long-term dependence characteristics that the real degradation process of lithium-ion battery is affected by historical data,a fractional Brownian motion(FBM)model based on dynamic drift parameters is proposed to describe the non-Markovian characteristics of the actual battery degradation process.Aiming at the nonlinear time-varying characteristics of battery capacity degradation process,Kalman filter(KF)algorithm is used to update the drift coefficient of FBM model.The maximum likelihood estimation algorithm is used to solve other fixed parameters in the model.Based on the first hitting time and weak convergence theory,the probability density function of the uncertainty of battery RUL prediction results is derived.Finally,experimental datasets and NASA datasets are used to verify the effectiveness of the proposed method.For the NASA #5 battery datasets,when the prediction starting point is 60 cycles and 80 cycles,the relative errors of RUL prediction results of this method are 2.941% and 2.083% respectively,which is better than many methods such as the BM model and FBM model with fixed drift parameters.(3)The single battery can not meet the power demand of EVs,and the battery pack is composed of hundreds of battery cells in series and parallel.There is inconsistency in cell-to-cell parameters within a battery pack,and battery pack performance depends on the worst performing cell.To solve this problem,this paper proposed a battery modeling considering parameter inconsistency and RUL prediction method based on inconsistent degradation process.A simplified inconsistent model of battery pack is established,and Monte Carlo method is used to simulate the influence of topology structure and parameter inconsistent distribution on the current distribution and capacity distribution of battery pack.Considering the capacity inconsistency of battery pack degradation process,the concept of competitive failure is introduced,and the capacity failure threshold of any single battery is defined to mean battery pack failure.The D-Vine Copula function is used to establish the correlation of the multivariate degradation process in the battery pack,and the probability density function of the battery pack RUL is obtained.The results show that the battery pack RUL prediction results considering the cell-to-cell correlation of capacity degradation are more accurate than the independent degradation process.(4)The SOH and echelon life of retired batteries are important indicators of echelon utilization.The traditional charge and discharge test method will reduce the remaining capacity of the battery and cause a waste of resources.Aiming at the requirement of nondestructive testing of retired batteries,a method of battery SOH estimation and echelon life prediction based on conditional generative adversarial network(C-GAN)is proposed.A fractional electrochemical impedance spectroscopy model is established to characterize the dynamic characteristics of retired battery modules.Simulated annealing-Gauss Newton method is used to identify the parameters of the model to avoid the results falling into local optimization.According to the characteristics of the small sample dataset of retired batteries,the C-GAN is used to enhance the training dataset of the prediction model,and the SOH estimation and echelon life prediction of retired batteries are realized by the gated recurrent unit neural network.This idea of data generation saves the battery test cost,enriches the training data,and improves the prediction accuracy.In summary,this paper is based on the application sequence in the whole life cycle of lithium-ion batteries,in the face of the different challenges of battery state detection and RUL prediction in the on-board stage and the retirement stage,breakthroughs have been made in the online life prediction of single battery,the life prediction of battery non-Markov degradation process,battery pack inconsistent performance and life prediction,and the SOH estimation and life prediction of retired battery modules.The study improves the accuracy of the algorithm in theory and practice.The research results of this paper have guiding significance for improving the utilization value of power batteries and ensuring the safe and reliable operation of batteries in the whole life cycle. |
PDF Full Text Request