Predicting Haul Truck Travel Times in Underground Mines

Abstract

Accurately predicting haul truck (HT) travel times (TT) in underground mines is essential for enhancing operational planning, as it allows planners to forecast extraction rates at each work face, minimize queue-related downtime, and ultimately increase productivity. However, in underground environments where GPS signals are unavailable, beacon-based locating systems have not yet been utilized for this predictive purpose. This study addresses that gap by introducing a machine learning approach for HT TT prediction that relies exclusively on beacon detection data, thus eliminating the need for traditional telemetry. The proposed method combines three route-segmentation strategies—full-route, short-segment, and major-segment predictions—with Gaussian mixture models, long short-term memory networks, and a stacking ensemble. Validated on two underground mines, it outperformed industry benchmarks, reducing prediction error by up to 34% on ascending routes and 18% on descending routes while achieving even greater precision for autonomous HTs. It showcases the untapped potential of beacon-based location systems for predictive applications, supporting mine planners.