An agent-based framework for open pit mine planning

Long-term production scheduling optimization has been a challenging issue for the mining industry because of the size and complexity of the problem. The current planning algorithms have limitations in terms of addressing the stochastic variables underlying the mine planning problem. In this paper, an intelligent agent-based mine planning framework based on reinforcement learning is introduced. The long-term mine planning is modelled as a dynamic decision network. The intelligent agent interacts with the block model by means of a stochastic simulation and employs a Q-learning algorithm to learn the sequence of pushbacks that maximizes the net present value of the mining operation. The intelligent open pit simulator (IOPS) was implemented with an object-oriented design in Java®. A comparative application case study was carried out to verify and validate the models. The proposed method was used in planning an iron ore deposit and the results were compared to the Milawa scheduler used in Whittle® software. The outcome of the study demonstrates that the intelligent agent framework provides a powerful basis for addressing real-size open pit mine planning problems.