ABSTRACT In response to the problem that the energy consumption of pure electric vehicles buses is difficult to accurately predict due to the influences of multiple factors, this paper proposes an energy consumption prediction model based on the extreme gradient boosting (XGBoost) fusion algorithm, which considers driver behavior patterns, vehicle operating environment, and vehicle performance. Firstly, a multiple linear regression model is introduced to preliminarily predict energy consumption. The influence of driver behavior patterns on vehicle energy consumption is analyzed using a Gaussian mixture model. The results of the above two models are input into a fusion algorithm based on extreme gradient boosting for energy consumption prediction. Bayesian optimization methods are used to adaptively optimize the hyperparameters of the XGBoost algorithm. This article takes gradient boosting decision tree (GBDT) and random forest algorithms as control groups to compare and analyze the performance of prediction models. The experimental results show that the prediction model using XGBoost algorithm yields the lowest mean absolute percentage error (MAPE). It also proves that the use of driving time, mileage, and cumulative output energy of the moto (COEM) features, as well as adaptive optimization of hyperparameters, can further improve the accuracy of the prediction model.
Read full abstract