Study On Parameter Identification Of SOC Model From The Battery Used In HEV And Its Intelligent Estimation

With the development of the hybrid electric vehicle, a key role has been played inthe vehicle performance for the battery and its management system. A good batterymanagement should be of some capacities such as monitoring the voltage, charging anddischarging current, temperature which could diagnosis battery’s characteristicparameters and estimating the state of charge and the maximum charging anddischarging current in order to improve the performance and life of the battery. But atpresent, there are the difficulties such that the precision SOC estimation is not high andthe balance performances for different batteries were poor in the process of the studyand develop of the battery management system. It is a primary task how to solve theproblems mentioned above for the experts and engineers. Therefore the SOC modelparameter identification and its intelligence estimation was studied in the paper basedon the “985engineering” project of Hunan university and Key laboratory open fundproject of Jiangsu province. This topic is of theoretical significance and applicationvalue.Therefore, some methods such as Simulink software, support vector machinetechnology, genetic algorithm and simulation experiment were fused and were used tostudy the battery in HEV SOC model parameter identification and its intelligenceestimation. The main innovations and research work are expressed as follows:(1) According to the advantages and disadvantages of battery, the security andperformance of the nickel metal were estimated. The working principle and thecharacteristics were studied and a good theory basis for parameter identification of thenickel metal hybrid battery model had been laid.(2) Auxiliary variable method and least squares technique were used in PNGVmodel to estimate the parameters. Simulation analysis show that, the maximal error is4.2V, the mean error is0.57V, reflecting the real conditions accurately.(3) Genetic algorithm was used to optimize the least square support vectormachine parameters and estimating the battery SOC with the least square supportvector machine. Results show that, the maximal error is2.36%; the mean error is0.48%. The training time obviously reduced and it’s good for online estimating.