Numerous attempts have been made to address the computationally intractable open pit optimization and long-term planning problem. Heuristic methods, economic parametric analysis, operations research, and genetic algorithms have been used to formulate periodic open pit long-term schedules. In practice all these techniques have limitations in dealing with large industrial problems;​ stochastic processes governing mining technical and economic variables;​ temporal nature of exploitation;​ and mine to mill integration. In complex mining operations, small deviations from the optimal strategic plan could result in the loss of millions of dollars.

The primary research goal is to develop, analyze, and implement a 3-D intelligent open pit optimal production simulator (IOPS) based on reinforcement learning (RL) to maximize the net present value (NPV) of the venture. Also a continuous open pit simulator (COPS) based on the modified open pit geometrical model and a system of differential equations have been developed to capture the continuous-time open pit dynamics for tactical purposes. The Java Reinforcement Learning Library was chosen as the core of the IOPS application implementation. Java programming language and MATLAB were selected as the platform for programming and graphical user interface (GUI) implementation.

To verify and validate the research models, a case study on an iron ore deposit with 114,000 blocks was carried out. The final pit limits were determined using Lerch’s Grossman’s algorithm with the Whittle software. The optimized final pit limits show the total amount of 399 million tonnes of material consisting of 220 million tonnes of ore and 179 million tonnes of waste. The practical annual schedule generated by the industry standard tool — Milawa algorithm used in Whittle software — yielded an NPV of $430 million over a 21-year mine life at a discount rate of 10% per annum. The practical learned scenario of 3000 simulation iterations using IOPS yielded an NPV of $438 million over the same time span. Experiments were also performed to compare the annual stripping ratio, average grade, annual waste, and the ore and concentrate production. The outcome of the research demonstrated a strong promise towards improving the expected net present value of mining investments. The algorithms developed can be the basis of the next generation of mine design software packages.