基于改进 LightGBM 的电动汽车电池剩余使用寿命在线预测

In order to achieve the remaining useful life (RUL) on-line prediction and reduce the impact of outlier value on prediction accuracy, this paper proposes an on-line prediction method based on the improved light gradient boosting machine (LightGBM). Firstly, in order to accomplish RUL online prediction, the health indicator is selected according to the relationship between isobaric time series and capacity. Then, in order to reduce the impact of outliers on the prediction accuracy, the prediction model based on LightGBM is built, and Bagging learning method is adopted, which ignores the weights of outliers. The improved LigthGBM based on adaptive robust loss function is established to reduce the impact further. Parameter α is utilized to limit the saturation value for first-order derivative of loss function, so that the influence of residual error on the gradient is reduced. Finally, the effectiveness of the established health indicator and the proposed RUL prediction method is verified by experimental data, and the RUL prediction performance based on different loss functions are compared. The results demonstrate that the proposed method has higher prediction accuracy and better robustness.