Abstract: To enhance the adaptability to changes in ore properties, improve concentrate production and grade, a precision control algorithm for fineness-concentration in the grinding-flotation collaborative optimization system under layered reinforcement learning is studied. Modeling the grinding-flotation collaborative optimization system, establishing mechanism models for equipment such as cyclones, ball mills, and spiral classifiers to determine key control variables that affect changes in fineness and concentration. Construct a fineness-concentration control framework for a grinding-flotation collaborative optimization system based on hierarchical reinforcement learning. Train the hierarchical controller using the strategy gradient method, generate fineness-concentration control objectives that meet the optimal comprehensive benefits through the upper optimization layer, and guide the lower control layer to adjust various key control variables to achieve fineness-concentration control in the grinding-flotation collaborative optimization system. The experimental results show that the algorithm can achieve fineness-concentration control in the grinding-flotation collaborative optimization system, with control errors of 1.20% and 0.75%, control time of 1.5 seconds, and concentrate production, grade, and drug consumption of 87.5 t/h, 59.2%, and 11.8 kg/t, respectively.