| At present,electric vehicles are popular because of advantages such as low pollution and low price.Lithium-ion battery is the most widely used power source in electric vehicles.Effective management of the State of Health(SOH)of the battery is related to the safety and operating efficiency of the vehicles.Due to the changeable operating environment of the electric vehicles and the complex electrochemical reaction inside the battery,the battery data has nonlinear,complex and variable characteristics.The SOH is an important parameter that characterizes the performance of the battery.Since the SOH cannot be measured directly and is susceptible to other parameters,it is challenging to accurately estimate the SOH of the battery,which has attracted extensive attention from scholars and engineering technicians and has become a research hotspot.Scholars have put forward many SOH estimation methods.Among them,the Back Propagation(BP)Neural Network has better learning ability and error convergence ability in nonlinear data fitting.It has achieved good SOH estimation results,but there are problems that it is easy to fall into local optimization and weak fitting ability,which affects the estimation stability and is difficult to be applied to real collected data.To this end,a Caputo BP Neural Network based on LAdaBoost is proposed to improve the estimation accuracy;an Extended Kalman Filter algorithm based on innovation is proposed to solve the preprocessing problem of battery data;on this basis,Lithium-ion battery SOH management system is designed and implemented.The main work of the thesis is as follows:(1)A Caputo BP Neural Network Method for Estimating State of Health of Lithium-ion Battery Based on Logic AdaBoost Algorithm(LAdaBoost-CBP)is proposed to achieve accurate estimation of SOH.Use fractional order calculus instead of integer order calculus to solve the local optimal problem of BP Neural Network;introduce an update factor of Logic-based deformation function in the sample weight update formula of AdaBoost to solve the problem of sample degradation;use CBP as weak learners and use LAdaBoost to promote the weak learners to a strong learner,so as to accurately fit the mapping relationship between SOH and battery measurable parameters,so that the SOH can be accurately estimated.Comparing the algorithm proposed in this thesis with related algorithms on NASA battery dataset and GongKuang Database(GKDB),the results show that the root mean square error is reduced by 1.5% ~ 36.8% and 7.4%~23.3%,which is significantly improved the SOH estimation accuracy.(2)A Preprocessing Algorithm of Battery Data Based on Innovation Extended Kalman Filter(IN-EKF)is proposed to eliminate abnormal values that are not eliminated by the Extended Kalman Filtering(EKF),thus provide stable input data for the LAdaBoost-CBP model to ensure the stability of the estimation accuracy.It is proposed to use the orthogonality of innovation to identify the abnormal values of the EKF filtered data,and replace the original abnormal values with the fitting values obtained by the least square method to ensure data continuity and stability.Comparing IN-EKF algorithm with related algorithms on the NASA battery dataset and GKDB database,the results show that the number of abnormal values is reduced by 528 ~ 672 and 2720 compared with the EKF algorithm,abnormal values processing ability is significantly improved;and based on the preprocessed data of IN-EKF algorithm,SOH estimation is performed with higher accuracy,the root mean square error of SOH estimation is reduced by 5% ~ 18.3% and 27.9%,which effectively reduces the influence of abnormal values on estimation accuracy.(3)Based on the Browser / Server model,MySQL database and Java language,a prototype system for Lithium-ion battery SOH management was designed and implemented.The system has functions such as data import,data preprocessing,SOH estimation and result display to meet the needs of SOH accurate estimation. |
PDF Full Text Request