Production scheduling in industrial mining complexes with incoming new information using tree search and deep reinforcement learning

Abstract

Industrial mining complexes have implemented digital technologies and advanced sensors to monitor and gather real-time data about their different operational aspects, starting from the supply of materials from the mineral deposits involved to the products provided to customers. However, technologies are not available to respond in real-time to the incoming new information to adapt the short-term production schedule of a mining complex. A short-term production schedule determines the daily/weekly/monthly sequence of extraction, the destination of materials and utilization of processing streams. This paper presents a novel self-learning artificial intelligence algorithm for mining complexes that learns, from its own experience, to adapt the short-term production scheduling decisions by responding to incoming new information. The algorithm plays the game of short-term production scheduling on its own using a Monte Carlo tree search to train a deep neural network agent that adapts the short-term production schedule with incoming new information. The deep neural network agent evaluates the short-term production scheduling decisions and, in parallel, performs searches using the Monte Carlo tree search to generate experiences. The experiences are then used to train the agent. The agent improves the strength of the tree search, which results in an even stronger self-play to generate better experiences. An application of the proposed algorithm at a real-world copper mining complex shows its exceptional performance to adapt the 13-week short-term production schedule almost in real-time. The adapted production schedule successfully meets the different production requirements and makes better use of the processing capabilities, while also increasing copper concentrate production by 7% and cash flows by 12% compared to the initial production schedule. A video of the proposed algorithm can be found at https://youtu.be/_gSbzxMc_W8.