Abstract
New digital technologies have enabled mining complexes to acquire new information on the performance of its different components and the flow of materials from mines to products. At the same time, existing complementary technologies cannot fully integrate such emerging digital information to adapt short-term production planning. A new self-learning artificial intelligence framework for a mining complex that learns, from its own experience, to adapt short-term production scheduling decisions by responding to incoming new information and quantifying related risk is presented. After providing context, several approaches are overviewed starting with early experiments based on the AlphaGo program from Google DeepMind, which is rewritten to continuously learn and adapt short-term production schedules. This is followed by feeding uncertainty models to a neural network that is trained using a policy gradient reinforcement learning algorithm to adapt the short-term flow of materials from mines to processing facilities in a mining complex. Lastly, an Actor-Critic reinforcement learning approach that jointly adapts short-term production scheduling, equipment allocation and material destination policies based on new incoming real-time information is presented. Aspects of the self-learning framework are demonstrated using examples from a copper-gold mining complex composed of two pits and multiple processing stream options.