<p>Alumina concentration is a key parameter to enhance electrolytic efficiency and minimize energy consumption in aluminum electrolysis. To address the deficiencies in conventional alumina concentration control, specifically the absence of intelligent setpoint optimization and the slow response in control systems, a novel control strategy based on a two-layer model predictive control (TL-MPC) algorithm is proposed. Firstly, a prediction model of alumina concentration is established based on process mechanisms and industrial data. Then, in the steady-state target computation layer (SSTC), the index of DC power consumption per tonne of aluminum is designed by combining the threshold-driven strategy to realize the online optimization of the alumina concentration setpoint. In the dynamic control layer, a predictive control-based alumina feeding strategy is proposed to ensure the precise regulation of alumina concentration. Finally, based on production data of an aluminum plant, the accuracy of the proposed alumina concentration model is validated. The results show that the DC power consumption of the TL-MPC strategy is reduced by 15.07 kWh per tonne of aluminum compared with interval control. Compared with PID and LQG algorithms, the average root mean square error (RMSE) of the TL-MPC is reduced by 0.0124 and 0.003, which shows stronger robustness and accuracy.</p> Graphic Abstract <p></p>