基于KECA-NARX的RUL时间序列预测模型

RUL time series prediction model based on KECA-NARX

针对锂离子电池剩余使用寿命(RUL)预测中信息利用不完善的问题,提出基于核熵成分分析(KECA)的非线性自回归(NARX)神经网络的RUL时间序列预测模型算法。采用灰色关联分析方法确定预测模型的输入变量;再重构确定的输入变量,恢复系统多维非线性状态空间;最后通过混沌优化算法,使KECA的核参数达到全局最优,利用KECA对重构后的相空间时间序列进行特征提取,并将特征向量作为NARX神经网络的输入,得到预测的RUL。KECA-NARX模型的预测精度比Elman模型提高了近6%,表明提出的KECA-NARX模型具有较优的非线性动态预测能力、较高的精确度及泛化能力。

Aiming the problem of incomplete information utilization in the prediction of the remaining useful life(RUL)of Li-ion battery,the RUL time series prediction model algorithm based on the nonlinear autoregression(NARX)neural network with kernel entropy component analysis(KECA)was proposed.Grey correlation analysis method was used to determine the input variables of the prediction model.The determined input variables were reconstructed to restore the multidimensional nonlinear state space of the system.Finally,the kernel parameters of KECA were globally optimized by chaos optimization algorithm,KECA was used to extract the features of the reconstructed phase space time series,the feature vectors were used as the input of NARX neural network to obtain the predicted RUL.The prediction accuracy of KECA-NARX model was nearly 6%higher than that of Elman model.It was indicated that the proposed KECA-NARX model had better nonlinear dynamic prediction ability,higher accuracy and generalization ability.