遗传算法优化神经网络整包电池SOC估计模型

Genetic Algorithm Optimized Neural Network SOC Estimation Model for Battery Packs

针对Back Propagation(BP)神经网络建立的动力电池荷电状态(State of Charge,简称SOC)估计模型存在的精度、稳定性等问题,使用遗传算法(Genetic Algorithm,简称GA),对SOC估计网络模型进行优化。使用以轮盘赌算法配合最佳个体保存法做为遗传算法核心选择算法,避免传统选择算法中最佳个体可能丢失的情况。实验对象选用新能源氢电混动重卡4X2/6X2T FCV的辅助动力电池包。模型考虑了环境温度、放电电压、放电倍率以及直流内阻的影响,对不同隐含层数量对模型估计精度的影响进行了对比分析。测试结果表示,使用遗传算法优化后的神经网络SOC估计模型的精度和稳定性得到大幅提升。

For the accuracy and stability of thestate of charge(SOC)estimation model established by the Back Propagation(BP)neural network,a Genetic Algorithm(GA)is used to estimate the network model for the SOC optimize.Use the roulette algorithm combined with the best individual preservation method as the core selection algorithm of the genetic algorithm to avoid the situation in which the best individual in the traditional selection algorithm may be lost.The test object selected the auxiliary power battery pack of 4X2/6X2T FCV,a new energy hydrogen-electric hybrid heavy truck.The model considers the effects of ambient temperature,discharge voltage,discharge rate and DC internal resistance,and compares and analyzes the effects of the number of different hidden layers on the model estimation accuracy.The test results show that the accuracy and stability of the neural network SOC estimation model optimized by genetic algorithm have been greatly improved.