Study On State Of Charge Estimation For Power Lithium Battery Pack

With the rapid development of vehicle industry, extensive use of traditional vehicles not only brings convenience to people, but also causes a series of environment and resource issues. Development of electric vehicles is important means dealing with these issues. Compared with lead acid battery and nickel metal hydride battery, lithium battery has great advantages in energy density and memory effect. As an important parameter of BMS, SOC of lithium battery has important significance to SOH estimation, driving range estimation and energy balance.This research used SAMSUNG brand ICR18650-22 F type battery as subjects. Firstly, battery characteristics is tested in order to obtain it's polarization voltage characteristics, internal resistance characteristics, open circuit voltage characteristics, capacity characteristics and temperature characteristic. Secondly, Electric vehicle is modeled based on ADVISOR2002 in MATLAB 6.5. LEAF as a classical electric vehicle of Nissan Company is selected as vehicle body model. Parameter of battery pack model is from battery characteristics test. Make the model work in test condition and collect data of lithium battery pack. Then Extreme Learning Machine is improved for better performance. Direct data channels are built between input nodes and output nodes in order to improve the accuracy of SOC model. Test results show, improved ELM is excellent than traditional ELM in maximum error and mean square error. Lastly, in the choice of SOC estimation methods, four SOC estimation methods: EKF, UKF, AEKF and AUKF are introduced in this part. AEKF and AUKF with adaptive modular, which don't need to set measurement variance and state variance in the initial stage, have excellent adaptability to time varying noise. Experimental results shows that AEKF and AUKF are more excellent than EKF and UKF in convergence rate and accuracy. AUKF is slightly better than EKF in accuracy.