基于MPSO-LM算法的车用电池组SOC预测

SOC gauge of battery for EV based on MPSO-LM algorithms

在对车用氢镍电池组进行了不同工况和温度下的充放电实验,获取了大量能真实反映电池动态行为和特征的实验数据的基础上,建立了一个Back-propagation神经网络的车用动力电池组的仿真模型,实现对电池SOC的预测。为提高BP算法的训练速度和估算精度,设计了一种将改进粒子群算法(MPSO)与Leyenberg-Marquardt(LM)算法组合使用的混合算法(MPSO-LM)用于优化训练BP神经网络。仿真结果表明,所提议的MPSO-LM算法比BP算法更有效,具有较快的收敛速度和较高的预测精度。测试结果中97%数据达到5%的误差或更小。

A back-propagation artificial neural network for SOC gauge was developed based on testing of Ni-MH battery packs for electric vehicle under different driving conditions and temperatures as well as a mass of data records were obtained from laboratory, which captured properly the dynamic behavior and characteristics of the battery pack. In order to improve the performance of BP network, the neural network structure optimization was carried out using modified particle swarm optimization combined with Levenberg-Marquardt algorithm. The technique provides more consistent and fast solution in obtaining the local minimal as compared to other method. The simulation results show that the proposed method matches more than 97 percent of the data sets with sum square error of 5 percent or less.