Answers are grounded in this paper and cite their sources.
This study develops an integrated framework to model and explain fuel eco-efficiency [L/(t·km)] of underground haul trucks using truck–day–shift operational data from a mine in Pasco, Peru. After consistency screening and 99th-percentile trimming, a fixed-effects panel model was estimated as an econometric benchmark, accounting for truck-specific heterogeneity and key operational variables. In parallel, global time-series models (LSTM, DeepAR, and Temporal Fusion Transformer) were trained using rolling windows of log-transformed fuel intensity with operational and calendar covariates. The Transformer achieved the best typical forecasting performance (MAE = 0.544; MedAE = 0.357; MAPE = 49%), outperforming the panel benchmark in relative-error terms (MAPE = 92.6%). To capture nonlinear relationships and improve interpretability, an XGBoost regressor combined with SHAP achieved MAE = 0.152, RMSE = 0.242, MAPE = 12.4%, and R 2 = 0.905. SHAP results indicated that haul distance and lagged fuel intensity contributed most to higher predicted fuel intensity, whereas higher tonnage and productivity were associated with lower predicted values. Overall, the framework integrates econometric inference, global forecasting, and explainable machine learning to support fuel-efficiency assessment, fleet monitoring, dispatch decisions, energy management, and emissions mitigation in underground haulage under variable operating conditions and across multiple haul trucks.
No takes yet. Share an insight, caveat, or question.
Mamani-Quispe et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: