A Composite Power Prediction Model for Semi-au-togenous Grinding Mill Based on Mechanistic Approach and XGBoost

Semi-autogenous grinding (SAG) mill has been widely used in mineral processing plants. It is large heavy equipment with high energy consumption. Therefore, modeling of its power is of significance for optimization of the operating. How-ever, due to the complex mechanism of the mill, the power is composite of several parts and difficult to be calculated accurately. This paper presents a hybrid power prediction model for SAG mills that integrates a mechanistic model with an XGBoost-based data-driven approach. Initially, the model applies physical equations to characterize the dynamic grinding process within the mill, capturing the system's essential physical behavior. In this formulation, the mill load power is decomposed into kinetic and frictional power, with kinetic power calculated and its proportion estimated to develop a comprehensive power equation. Subsequently, an XGBoost-based data-driven compensation model is introduced. This model incorporates three key internal state parameters predicted by the mechanistic model, enhancing insights into the mill's internal conditions. The residual between the mechanistic model's predicted power and the actual power is used as the target variable, to address the inherent limitations of the mechanistic model. The predictive performance of the proposed hybrid model is validated by using real industrial data from a mining plant, with comparisons made to other prediction methods. The results demonstrate that the hybrid model delivers superior accuracy in power prediction and can effectively capture fluctuations in power trends.