Research On State Of Charge Estimation Of Lithium Batteries Based On Data-Driven

With the development of society,the country is paying more and more attention to the requirements of environmental protection,calling for low-carbon travel.New energy vehicles have the characteristics of low-carbon and energy-saving,and have more advantages compared to gasoline vehicles.It is precisely this that they are becoming increasingly popular among the mainstream consumer group.The Battery Management System(BMS),as a key component of new energy vehicles,is used to convey the current state of the battery to the driver.The State of Charge(SOC)is the core element,and accurate estimation of it is a prerequisite for the safe and stable operation of the vehicle.Therefore,this article conducts research on the SOC of batteries.Firstly,introduce the basic concepts of lithium batteries,including their classification and characteristics,and analyze the working principle of lithium batteries to sort out the key parameters of the batteries.Afterwards,a testing platform was established to test the charging and discharging characteristics,temperature characteristics,open circuit voltage characteristics,and different operating conditions of the lithium battery.The operating conditions tested in this article were the Urban Dynamometer Driving Schedule(UDDS)in the United States,the New European Driving Cycle(NEDC),and the New European Driving Cycle(NEDC)Dynamic Stress Test(DST)and Beijing Bus Dynamic Stress Test(BBDST)for pure electric buses.Secondly,the Back Propagation(BP)neural network is selected as the foundation to improve the BP neural network.Considering that the traditional optimization method is based on the steepest descent method,this thesis uses the Levenberg Marquardt(LM)algorithm instead of the steepest descent method,and adds momentum term to the algorithm to update the weight of the BP neural network.The results show that: The improved algorithm has significantly improved the estimation performance compared to the standard BP neural network.However,in order to obtain more accurate estimation results,an ensemble algorithm was used for optimization.The improved BP neural network was used as a base learner and combined with the Adaboost algorithm.After testing,it was found that the momentum LM-BP-Adaboost algorithm has relatively stable estimation performance.Finally,considering that each algorithm has its own characteristics in estimating battery SOC,in order to complement the advantages of the algorithms,the Kalman Filter(KF)algorithm and battery model are introduced.After comparing several filtering algorithms,this thesis first uses the Extended Kalman Filter(EKF)to estimate battery SOC,and then uses the Momentum LM-BP-Adaboost algorithm to compensate for its error,Finally,the corrected results were output and their effectiveness was tested under various operating conditions.The results showed that the accuracy of this algorithm was significantly improved compared to using only the EKF algorithm.