Research On The Online Assessment Of State Of Health Of Lithium-battery In Electric Vehicle

In order to deal with the increasingly significant challenges of environmental pollution caused by traditional fuel vehicle emissions,electric vehicles have gradually attracted wide attention at home and abroad in recent years.T he development of electric vehicles has been adopted as a national strategy by the governments of different countries for accelerating the development of their technological research and development,as well as their industries.Lithium battery merits high energy density,long cycle life,wide temperature adaptability,low self discharge rate and environmental protection,so the lithium battery has gradually become an important focus in the development of energy storage unit of electric vehicle.State of Health(SOH)is a key parameter for fault diagnoses and safety early-warnings in the whole life cycle of lithium batteries in electric vehicles.Accurate evaluation of SOH has important significan ce for the enhancement of the overall performance of lithium batteries.At present,domestic and foreign scholars are paying increasing attention to the study of SOH evaluation of lithium battery,with particular focus on the on-line evaluation method.However,it remains challenging to directly measure the parameters of lithium battery such as: discharge capacity and AC/DC impedance during practical application and so on.Therefore,It is difficult to use these parameters as an input variable in online evaluation model.Based on the data of working voltage,charging current,charging time and battery number that can be directly measured during the charging process of electric vehicle,the thesis studies on-line evaluation of SOH by combining the capacity estimation model of segment data and the life cycle attenuation prediction model.Estimation of lithium battery capacity is generally carried out via the machine learning method with parameters that map directly to the battery capacity.Due to the individual electrochemical differences of lithium batter y,it is difficult to cover all individual battery models completely.To overcome the problems of individual differences in capacity estimation model,a usable capacity estimation model for lithium battery based on EKF-GPR iteration algorithm has been proposed.Different from the capacity estimation models established by other battery data training,this proposed model estimates the available charge capacity of lithium batteries by the initial fully charging data of the battery itself and fragment charging data after the battery is fully charged.Gaussian Process Regression(GPR)model has been first established as the state equation of Extended Kalman Filtering(EKF)to estimate the complete battery voltage curve when charging.An iterative EKF-GPR algorithm has been proposed to improve the accuracy of EKF model parameters based on the initial fully charged data,so as to reduce the model deviation caused by the gradual change of charging curve in the process of lithium battery performance degradation.When the model is applied to the same type of batteries of different production batches,the experimental results show that the accuracy of the iterative EKF-GPR model is better than that of the capacity incremental model.At present,the prediction method,used to predict whole life cycle SOH of lithium batteries,is based on experimental data from the same batch of batteries in the laboratory under the same environment.The prediction results of health performance may differ due to the daily working conditions of electric vehicles and differences in lithium battery performance among different batches of lithium batteries.If all of the collected data is used for the model training,the model 's performance may be adversely affected.It is difficult to forecast SOH of lithium-ion batteries accurately using GPR model based on a single kernel function.Therefore,training data filtering method has been proposed based on the uncertainty of SOH prediction results.The sum of Maternard co-variance function and neural network is set as a new kernel function to build a SOH prediction model of lithium battery with GPR method.The uncertainties in prediction results include dispersion of prediction results near the center of training class data clusters and GPR variance.The evaluation uncertainty from the synthetic result is obtained,and prediction results with the smallest evaluation uncertainty are chosen as SOH prediction results of the lithium battery.Experimental results have shown that evaluation uncertainty is better at evaluating credibility of lithium battery SOH prediction model as compared to GPR prediction variance.At the same time,based on the SOH data of different batches of battery as the training data,the adap tive prediction model adopting the smallest evaluation uncertainty as the training data selection basis is more accurate than the prediction model using all data for training.On-line SOH evaluation of lithium battery packs in electric vehicles can ensure safety during the operation of electric vehicles in real time.At present,the discharge capacity of lithium battery pack is mainly used to evaluate the degree of battery deterioration.However,lithium battery complex discharging conditions during the operation of electric vehicles makes it impossible to estimate the full discharge capacity of the lithium battery.Furthermore,as the capacity of a battery pack is inconsistent with that of the sum of all of battery cells,it is difficult to establish a model with the experimental data from battery cell.Based on data of the operating voltage,charging time,number of charge cycles,and number of the lithium battery pack in electric vehicles,the thesis establishes an online evaluation model of the lithium battery pack SOH based on the available charge capacity.The proposed model is capable of evaluating the lithium battery pack SOH in the electric vehicles immediately after each charging cycle.The difference between available charge capacity and available discharge capacity in SOH evaluation has been quantitatively analyzed,before the mathematical model of using available charging capacity to evaluate SOH is established.An improved algorithm of iterative EKF-GPR has been proposed due to differences between the lithium battery pack and battery cell.Based on the EKF-GPR algorithm,the attenuation change of lithium battery pack in the whole life cycle is adopted as the measurement equation of EKF model to improve the single charging curve for noise measurement.The available charge capacity of lithium battery pack for current charging is estimated,and the SOH of current lithium battery for electric vehicle is evaluated online.The effectiveness of this online SOH evaluation model has been validated by the daily charging data of electric buses in Shenzhen.