Energy Consumption Prediction and Feature Contribution Analysis of Unmanned Mining Trucks Based on XGBoost and SHAP

Abstract

With the rapid development of mining automation, unmanned mining trucks are increasingly used in mine transportation. Accurate energy consumption prediction is crucial for optimizing energy management and control strategies. This study uses actual operational data of unmanned mining trucks and employs an XGBoost-based prediction model for energy forecasting, with SHAP used to interpret the model and quantify the contribution of each feature. Operating conditions are classified into unloaded downhill and fully loaded upslope conditions, with data analyzed by speed intervals. XGBoost models are constructed for each condition. SHAP analysis reveals that battery current and generator current significantly impact the model under unloaded downhill conditions, while battery current dominates in most speed ranges for fully loaded upslope conditions. At low speeds, generator speed has a strong influence. SHAP dependence plots show a linear relationship between battery current and energy consumption. Feature selection is performed by removing features with minimal contributions, simplifying the model and improving efficiency. The optimized model maintains predictive accuracy while reducing complexity. The results show that the XGBoost and SHAP-based model effectively predicts energy consumption, providing a basis for energy-saving optimization in smart mining operations.