An intelligent rule-based decision-making system for preliminary truck dispatching within open-pit mines

Abstract

Preliminary truck dispatching involves directing trucks to appropriate destinations before addressing specific optimization objectives such as maximizing ore production or minimizing waiting times. While rule-based systems are commonly used for preliminary dispatching, they lack adaptability to unforeseen scenarios. This study presents an intelligent rule-based system that integrates reinforcement learning to generate labeled data and supervised learning to train a deep neural network on the collected data. A modified Q-learning algorithm, RapidQ, was introduced to expedite the data collection process. The system was implemented in a simulated open-pit mine case study, which incorporated a broader range of dispatching features that are underexplored but essential compared to previous studies. The simulation was designed to handle uncertainties such as weather conditions, blasting needs, truck and shovel failures, and maintenance schedules. When evaluated against a conventional rule-based system, the proposed intelligent dispatching system achieved 10 times fewer incorrect dispatches, 4% fuel savings, a 10% reduction in queuing time, and a 14% increase in ore production. The developed system can be positioned as a potential upper-stage solution in future multi-stage intelligent dispatching systems, complemented by specific dispatching algorithms at the lower stage.