准抗毁化电源蓄电池SOC预测的GA-BP网络方法

GA-BP network based battery SOC prediction for quasi anti-damage power supply

针对蓄电池荷电状态的预测问题,从蓄电池荷电状态与其可直接测量的外特性参数之间不确定的非线性关系出发,依据BP网络映射功能使其可以以任意精确度逼近非线性函数、遗传算法的良好全局搜索寻找最优能力使其解决BP网络盲目选择初始权值、阈值的问题,并利用数值最优化LM算法训练BP网络使其解决BP网络收敛速度慢和容易陷入局部最小值的问题,提出了一种蓄电池荷电状态预测的遗传算法和BP网络相结合方法。设计了准抗毁化电源蓄电池荷电状态的BP网络和GA-BP网络预测模型。仿真结果表明,预测模型经过训练后,可以通过蓄电池的实时外特性参数预测蓄电池的实时荷电状态;GA-BP网络的收敛速度和预测精确度均优于BP网络。验证了GA-BP网络预测方法的有效性。

In order to predict the State of Charge（SOC） of the battery,considering the uncertain non-lin-ear relationship between the SOC of the battery and external characteristic parameters measured directly,according to the nonlinear mapping ability of Back Propagation（BP） network for approximating nonlinear function in an arbitrary accuracy and the global searching optimum of Genetic Algorithm（GA） for solving the problem of blind selecting initial and threshold value,utilizing Levenberg-Marquardt（LM） algorithm for training the BP network to get rid of low convergence velocity and locking in local minimum value in BP network,a SOC prediction method combined GA algorithm with BP network is presented in this pa-per.The SOC prediction models with BP network and GA-BP network for quasi anti-damage power supply were designed.The simulation results show that all trained models are able to predict the true values of SOC through the actual magnitudes of discharge voltages and currents.The convergence velocity and pre-diction error with GA-BP network are all better than those with BP network.The effectiveness of GA-BP network prediction approach is verified.