18650型锂电池荷电状态的估计

The SOC estimation of 18650 lithium battery

摘要:

锂电池状态的准确估计,能够延长电池的使用寿命和减少安全事故的发生。为提高BP神经网络估计锂电池荷电状态的精度,提出一种使遗传粒子群算法有目的性的优化BP神经网络初始权值的改进方法。该算法引入K均值算法优化遗传粒子群算法初始粒子分布的随机性带来的误差问题,寻找BP神经网络算法初始权值的权重分配与输出误差的关系,在遗传粒子群算法随机产生的粒子群中进行最优粒子群选优,以降低误差。通过对采集到的18650型锂电池的充放电数据和未改进遗传粒子群算法优化的BP神经网络训练产生的200组BP神经网络的初始权值数据的研究分析,得到具有锂电池特性的BP神经网络的初始权值特征公式。并用MATLAB和FPGA联合仿真验证了改进BP神经网络方法的可行性。该方法也优化了遗传粒子群算法,减小了初值不确定带来的误差。

Abstract:

The accurate estimation of the lithium battery’s state can prolong the service life of the battery and reduce the occurrence of safety accidents. In order to improve the accuracy of back propagation (BP) neural network to estimate the state of charged (SOC), an improved method is proposed to optimize the initial weights of BP neural networks by using genetic particle swarm optimization algorithm (GA-PSO). The K-mean algorithm is introduced to optimize the error caused by the randomness of the initial particle distribution in the genetic particle swarm algorithm, and seeks the relationship between the initial weights and the output error of the BP neural network algorithm, the optimal particle swarm optimization is carried out in the particle swarm generated by the genetic particle swarm optimization algorithm, which can reduce the error. According to the analysis of the charging and discharging data of the 18650 lithium battery and the 200 sets of data is produced by the BP neural network training that is optimized by the original genetic particle swarm optimization algorithm. Then the initial weight characteristic formula of BP neural network with lithium battery characteristics is obtained. And the feasibility of the improved BP neural network method is verified by the cosimulation of MATLAB and FPGA .The method also optimizes the GA-PSO, and reduces the error caused by the uncertainty of initial values.