车用锂离子动力电池剩余寿命非线性组合预测研究

针对单一预测模型难以准确预测锂电池的剩余寿命（remaining useful life, RUL）难题，提出了非线性组合预测方法；利用相空间重构，对实验采集到的数据进行重构，将重构后的数据对改进Elman神经网络和非线性自回归（nonlinear autoregressive with exogenous input，NARX）神经网络这2个单项预测模型进行训练和预测；采用RBF神经网络对2个单项模型的预测值进行非线性组合，获得最终的RUL预测值．结果表明:非线性组合预测方法的均方根误差比PCA-NARX方法提高了近1%，比NARX方法提高了近2%，比改进Elman方法提高了近3%；非线性组合预测方法具有较高的精度及泛化能力，采用相空间重构技术有利于提高非线性组合方法的预测精度． 

Nonlinear combination prediction of remaining useful life of automotive Lithium-ion batteries

Due to difficulty of accurately predicting residual life of Lithium batteries with a single prediction model, a nonlinear combination prediction method is proposed in this work．Phase space reconstruction was used to reconstruct data collected from experiments．Reconstructed data were trained and predicted on two single prediction models, improved Elman neural network and nonlinear autoregression neural network．RBF neural network was used to combine predicted values of the two single prediction models, final RUL predicted value was then obtained．The proposed nonlinear combination forecast method of mean square error was found to be nearly 1% higher than PCA-NARX, nearly 2% higher than NARX, nearly 3% higher than improved Elman．Nonlinear combination forecasting method had higher precision and generalization ability．It is conluded that phase space reconstruction technology is helpful to improve prediction precision of nonlinear combination method．