Application of Markov decision processes to mine optimisation : a real option approach

This thesis describes preliminary research on the application of Markov Decision Processes (MDPs) to the optimisation of mine scheduling in an uncertain environment. The MDP framework is a novel approach to scheduling in a mining operation and option valuation. The task of scheduling in mining operations is dependent on the availability of models that permit the representation of some of the key stochastic properties of the environment, such as grade and price uncertainty. The tools used to model these processes are respectively sequential Gaussian simulation and Geometric Brownian motion. Three cases of increasing size are used to illustrate the results of the model and demonstrate its suitability to mine scheduling and option valuation. The computational efficiencies of solving an MDP formulation by Policy Iteration and Value Iteration are compared. The impact of the discount rate on the optimal policy is assessed. To determine the value of one or several options, an optimal policy without options is generated and valued. Then, the exercise is repeated with the relevant options to value (e.g., production rate, cut-off grade and time of mine closure). By comparing the values obtained in both cases, the financial benefit of having operational flexibility is determined, thus yielding the option value. A full size case study is conducted to validate the applicability of MDPs to real mining projects