Subway Energy Consumption Prediction based on XGBoost Model

Abstract

In the process of urban rail transit operation and management, accurate prediction of subway energy consumption is beneficial for establishing a reasonable operational organization mode and evaluating energy efficiency. Due to the multitude of factors affecting train energy consumption, traditional mathematical regression methods struggle to guarantee predictive accuracy. Thus, a energy consumption prediction method based on XGBoost is proposed. To enhance model training efficiency and accuracy, the Lasso model is utilized for feature selection of subway energy consumption influencing factors. Additionally, the K-means++ algorithm is employed for clustering subway energy consumption. Using the operational energy consumption data of Qingdao Subway Line 3 as an example for validation, XGBoost algorithm is employed to predict subway energy consumption. The results are then compared with those of the SVR and LSTM algorithms using three evaluation metrics. It is found that the XGBoost algorithm provides predictions of subway energy consumption that are closer to the experimental values.