State of charge prediction for mine-used power battery

It was difficult to optimize parameters of regularization and kernel function when least squares support vector machine(LSSVM) was used to predict state of charge(SOC) of mine-used power battery, and grey wolf optimization(GWO) algorithm was prone to early maturity, poor stability and local optimization when solving constraint optimization problem alone. In view of above problems, on the basis of differential evolution GWO(DE-GWO) algorithm, non-linear convergence factor in the form of exponential function is used to improve the DE-GWO algorithm. The non-linear convergence factor has low attenuation rate in the early stage of iterative process and the global optimal solution can better be found, while it has high attenuation rate at the end of iterative process and the local optimal solution can be found more accurately, which effectively balances global search ability and local search ability. The experimental results show that the maximum absolute error and the maximum relative error of SOC prediction model for mine-used power battery are 3.7% and 5.3% respectively after LSSVM parameters are optimized by the improved DE-GWO algorithm.