Optimal Design Of Battery Management Systems For Electric Vehicles

In recent years,as the environmental problems caused by fossil fuels have become more serious,many countries have made more significant efforts to research and develop new energy sources,and new energy vehicles have emerged.The rise of new energy vehicles can effectively reduce CO2 emissions and slow down environmental degradation.At this stage,the most significant number of new energy vehicles are electric vehicles.With the widespread use of electric vehicle batteries,research into power batteries and battery management systems(BMS)has intensified.As one of the essential functions of the BMS,battery state estimation includes State of Charge(SOC)estimation and State of Health(SOH)estimation.Therefore,in this paper,the traditional SOC and SOH estimation methods are studied,analyzed,and optimized,considering different working situations and data characteristics.The main research elements of this paper are as follows:(1)An analysis of electric vehicle power battery systems,studying the internal structure of lithium-ion batteries,their operating principles,and the main functions of the BMS.It focuses on the battery state estimation function of the BMS and common estimation methods,such as SOC and SOH,and their problems.(2)This paper proposed an improved ampere-hour integration method based on the Long Short-Term Memory(LSTM)network model.The traditional ampere-hour integration method treats the actual battery capacity,which varies with environmental factors and the battery’s usage conditions,as a constant,resulting in significant errors in SOC estimation.Analyze the relevant factors affecting the variation of the actual battery capacity,identify the main influencing factors,and input them into the LSTM neural network to predict the actual battery capacity,which is used to improve the ampere-hour integration method.We used the predicted actual battery capacity to replace the battery’s rated capacity in the traditional ampere-hour integration method and optimize the traditional ampere-hour integration method.The test results show that the error of SOC estimation results of this method is less than 10% of the national standard requirement,which improves the accuracy of SOC estimation.The LSTM neural network has a significant advantage in dealing with timing problems such as currents in this estimation method.It is suitable for algorithm simulation,but the structure could be more complex and conducive to practical applications.(3)A joint SOH-SOC estimation model for lithium-ion batteries based on GWO-BP neural network is proposed to address the problem that the structure of the above method is too complex while improving the ampere-hour integration method.The use of GWOBP neural networks in this model is simple in structure and easy to apply to a microcontroller for practical battery management systems.Based on the analysis of the main influencing factors of SOH,the ONLY-BP neural network,and the GWO-BP neural network were used to predict SOH separately,and the GWO-BP neural network with higher and more stable prediction accuracy was selected for the SOH prediction part of the joint estimation model.The predicted SOH is combined with the ampere-hour integration method to correct for the actual battery capacity.In the validation of this method,the stability of the joint estimation model was verified using a mixture of battery data obtained from the experiments in this paper and data from the NASA battery data.The SOC estimation accuracy was improved,and the SOC error was stabilized within 5%.(4)Design of battery data acquisition tests using a four-channel lithium-ion battery capacity tester at room temperature.This four-channel lithium-ion battery capacity tester with four independent acquisition channels ensures accurate data acquisition.The battery charge cut-off voltage was set to 4.2V,the discharge cut-off voltage to 2.75 V,the discharge multiplier to 0.25 C,and the charge multiplier to 0.5C,and the battery was emptied and then filled for one use cycle.A total of 1278 sets of cell data were collected for experimental validation to verify the accuracy of the improved SOC estimation method.