| Lithium ion battery is an important part of energy storage system,which is widely used in many industrial fields.Lithium ion batteries are equipped with battery management system(BMS),which can monitor the battery condition and intelligent maintenance,so as to ensure the stable and healthy operation of the battery.State of charge(SOC)estimation,state of Health(SOH)estimation and remaining useful life(rul)prediction are the core functions of BMS system.However,these state parameters can not be measured by sensors,but can only be estimated with measurable observations combined with mathematical algorithms.The mathematical method can be divided into model-based and data-driven estimation methods,which have their own advantages,but the single method tend to show obvious shortcomings.It is necessary to integrate the two methods to further enhance the stability and nonlinear mapping ability of the estimation method.In addition,there is a complex coupling relationship among the state parameters,which affect each other in the whole cycle operation of the battery.In order to realize the comprehensive evaluation and real-time management of the battery,it is necessary to jointly estimate the SOC,SOH and RUL in the whole life cycle of the battery.Therefore,this paper studies the separate and joint state estimation methods of lithium-ion battery based on hybrid method of model and data-driven.The main research work includes the following contents:(1)The separate estimation method of SOC is studied.Aiming at the problems that the conventional unscented Kalman filter(UKF)lacks adaptive robustness,finds it hard to determine noise variance and is dependent on the accuracy of RC model,an adaptive unscented Kalman filter(AUKF)algorithm is proposed by introducing convergence criteria,adjusting measurement noise variance and process noise variance as well as modifying Kalman gain adaptively.And the convergence speed,estimation accuracy and stability of SOC estimation based on AUKF are improved.Moreover,a more practical linear interpolation method is proposed to model the changes of RC parameters with temperature and SOC,so as to improve the adaptability of RC model.(2)The separate estimation method of SOH is studied.For the scenarios of full charging and partial charging of the battery,the estimation method based on ICA and Box-Cox transform and the estimation method based on voltage segment and error compensation model are designed respectively.The former requires a complete charging voltage curve,which has fast calculation time,high estimation accuracy and strong robustness;The latter only needs the charging voltage segment,where the least squares support vector machine(LSSVM)is used to dynamically compensate the fitting error of empirical degradation model,which improves the estimation accuracy and meanwhile,requires less training samples.(3)The SOH-SOH-RUL joint estimation method during the whole life of the battery is studied.Combining the advantages of each separate estimation method and fully considering the coupling effect of each state parameter,a SOH-SOH-RUL joint estimation framework is proposed.In each charge-discharge cycle,the framework uses the charging voltage segment for health feature extraction and parameter identification in the charge stage of the cycle.The extracted health features are combined with LSSVM and Gaussian process regression(GPR)to estimate the SOH and RUL of the current cycle,and the identified parameters are combined with AUKF algorithm to estimate the SOC in the discharge stage of the cycle.The framework combines the stability of the model-based method and the learning ability of the data-driven method,which can jointly realize the long-term stability estimation of SOC,SOH and RUL within the same framework.And the modeling process does not involve the historical operation data of the battery to be tested,so it is highly practical.Experimental results show that the proposed method has good estimation accuracy and stability. |
PDF Full Text Request