Machine learning-facilitated algorithm for operations optimization in thermal power plants

Coal remains a dominant energy source currently, necessitating that coal-fired power plants strictly control pollutant emissions while maintaining a stable electricity supply. To address this challenge, this study proposes a mixed-integer linear programming (MILP) model designed to optimize coal allocation and the combustion rates of generating units, aiming to minimize sulfur emissions. To efficiently solve this complex optimization problem, we develop a machine learning- facilitated column generation (ML-CG) algorithm. Building upon the CG, this algorithm integrates a deep neural network to optimize the column selection process. Furthermore, the deep neural network is refined by incorporating synthetic minority oversampling technique and a class-weighted loss function to enhance predictive accuracy in imbalanced datasets. The results demonstrate that the ML-CG algorithm achieves an average computational time reduction of 22.4% compared to traditional CG. Using numerical experiments and sensitivity analyses based on real operational data from a coal-fired power plant in China, this study provides managerial insights for optimizing coal resource management and combustion rate control in coal-fired power plants.