Data-Driven Grinding Control Using Reinforcement Learning

Abstract

In the mineral industry, the grinding circuit (GC) is the most critical unit for mineral processing operations.
The goal for GC control optimisation is to ensure the outputs of the controlled processes best follow the control actions and to ensure that the grinding product quality and efficiency are well controlled within the optimal ranges. However, it is hard to achieve these goals at the level of basic feedback control where global operational indices are not considered. Therefore, the higher-level advanced control mechanism is required for grinding operations. In this paper, we present our work using a big data driven and reinforcement learningbased approach for optimising GC processes. With our approach, it is not necessary to manually construct a system process model as it can be learnt from the historical GC log data automatically. To evaluate our method, a series of experiments have been conducted, and the experiment results show evident enhancement with regards to both product quality and grinding process efficiency.