Agent-Based Optimization for Truck Dispatching in Open-Pit Mines

The mining industry has long recognized the value of dispatch systems in open pit mines as they reduce load and haul costs. Over the years, researchers have proposed many dispatch systems with various limitations and advantages. The simplest dispatch algorithms are the so called 1-truck-for-N-shovels dispatch strategy. These algorithms are limited by the fact that their objective functions do not consider all the objectives of a mine and cannot be applied to all possible truck-shovel configurations. They are also myopic in nature. However, they are simple and computationally efficient and do not require occasional updates of the upper stage problem as required in multi-stage dispatch algorithms. In this work, an agent-based truck dispatch algorithm that conceptualizes trucks as intelligent agents that make autonomous dispatching decisions to maximize their utility is proposed. The advantages of this algorithm includes utility functions that encapsulate all of management’s objectives and agent’s with broad situational awareness. They are also more suitable for autonomous trucks. We evaluate the new algorithm against a simple 1-truck-for-N-shovels dispatch strategies using discrete event simulation. The simulation results show that the new utility function has significant advantages over 1-truck-for-N-shovels inspired utility functions. Future work will incorporate adaptive behavior into the model via reinforcement learning algorithm. INTRODUCTION In an attempt to optimize truck shovel operations in open pit mines in order to reduce the cost of haulage operation, which according to (Alarie and Gamache, 2002) is estimated to be about 50-60% of the total cost of mining, several effective and efficient strategies have been developed. Optimizing truck haulage systems in an open pit mine is one of the greatest challenges of mining engineering and has been the subject of many research projects. It is important that haulage systems are designed to be efficient, in order to minimise haulage cost, improve profitability and increase the total mine value. Although optimizing the number of trucks required in the operation of open pit mines can help reduce the cost, an effective and efficient truck dispatching system add value by ensuring that the available trucks are used efficiently. With truck dispatching systems, the trucks are normally dispatched between the crusher and dumping station (e.g. waste dumps and stockpiles), and the shovels in order to maximize the total output, obtain grades targets, reduce fuel consumptions and other objectives. Truck dispatching, which can have one or more objectives, is typically subjected to certain operational constraints such as truck and shovel capacities, empty and loaded trucks speed, among others. The problems associated with truck dispatching was first mentioned by Dantzig and Ramser (1959) after they made an attempt to determine the optimum routes of a fleet of delivery trucks between bulk terminals and a service stations. The problem occurs in several industrial systems besides mining (Alarie and Gamache, 2002; Gendreau et al., 2006; Rego and Roucariol, 1995). The dispatching problem, also referred in literature as the vehicle routing problem, has seen a number of modifications and extensions with Bertsimas and Ryzin (1991) presenting a dynamic and stochastic vehicle routing problem, and Ralphs et al. (2003) also adding to the original concept, the concept of vehicle capacity routing.