LightGBM-Based Prediction of Remaining Useful Life for Electric Vehicle Battery under Driving Conditions

Abstract

Remaining useful life (RUL) prediction for electric vehicle battery (EV battery) is of great significance for the early replacement and regular maintenance of batteries with potential safety hazards in driving conditions. This paper proposes a novel method based on light gradient boosting machine (LightGBM) to predict the RUL of the battery under driving conditions. LightGBM uses the histogram optimization strategy to reduce the number of traversal of the data sample set and improve the robustness of the method; the depth-first splitting (leaf-wise) strategy reduces the risk of overfitting; the gradient-based one-sided sampling strategy (GOSS), reduce the data dimension; use the exclusive feature bundling strategy (EFB) to reduce the feature dimension. However, the LightGBM method has the difficulty of parameter setting. Therefore, this paper uses Hyperopt based on distributed asynchronous algorithm configuration/hyperparameter optimization to optimize its complicated hyperparameters. Subsequently, the method was applied to the prediction of battery RUL under simulated driving conditions. Based on the comparative cases, the results show that this method can guarantee the rapidity, accuracy and robustness of RUL prediction under the condition of low memory usage.