Prediction of Circulating Load Ratio for Semi-Autogenous Grinding Process Using Fuzzy <i>C</i> -Means and Bayesian-Optimized Random Forest

Grinding in mineral processing critically depends on real-time control of the circulating load ratio (CLR) to optimize efficiency and reduce energy consumption. While investigating the impact of the CLR on the grinding process from a mechanistic perspective can optimize production, it fails to achieve real-time perception of its dynamic variations during operation. This limitation hinders timely adjustments to operational parameters. Focusing on an actual semi-autogenous grinding (SAG) process, this paper presents a hybrid fuzzy C -means (FCM) and Bayesian-optimized random forest (BO-RF) framework that explicitly addresses nonlinearity and operational variability in the SAG process. Key parameters influencing CLR are first identified through mechanistic analysis. Operating conditions are clustered via FCM, followed by BO-RF submodel construction for each cluster. A nearest-neighbor criterion dynamically activates submodels for real-time prediction. Validated with industrial data from an iron concentrate plant, the method achieves more than 90% prediction accuracy. This approach establishes a generalizable framework for complex industrial processes with multivariate dynamics.