Abstract: Due to the complexity of the flotation process mechanism, which makes it difficult to meet the requirements of advanced process control, a system identification modeling method was adopted. To address the low fitting accuracy of traditional identification methods, a Whale Optimization Algorithm (WOA)-based Gated Recurrent Unit (GRU) system identification model (WOA-GRU) was proposed. This model leveraged the GRU's capability to effectively handle the time-delay characteristics inherent in the flotation process, while the WOA was used to optimize network parameters, further improving the identification accuracy. Considering that most existing coal preparation plants use single-input single-output PID controllers, which struggle to manage multi-input multi-output systems, Model Predictive Control (MPC) was introduced to better handle the multivariable coupling in the flotation process. Using production data from the Daichi Dam coal preparation plant, MPC simulations were conducted using both the WOA-GRU and NARX identification models. The results showed that the WOA-GRU model achieved 51.84% higher fitting accuracy than the NARX model. After integrating MPC, the WOA-GRU model could maintain ash content fluctuations within ±4% of the setpoint, outperforming the NARX model. Field trial results indicated that the proportion of data with ash fluctuation between 5% and 10% decreased by 10.8%, and the proportion exceeding 10% decreased by 3.9% compared to before MPC implementation. These results demonstrate that the WOA-GRU model not only offers higher accuracy and stability but also reduces ash content fluctuations, providing a reference for intelligent control and practical application in coal slime flotation.