TR-SVR: Trends and Residuals Support Vector Regression in Thermal Power Units

With the rapid development of clean energy sources, the role of thermal power units in the power grid has gradually shifted toward co-regulation. To address the volatility of renewable energy, thermal power units are required to undergo frequent load fluctuations. Constructing a coal consumption prediction model based on the operating status of these units enables more accurate forecasting, thereby supporting the transformation of China's power generation structure. To improve prediction accuracy, this study proposes a Trends and Residuals Support Vector Regression(TR-SVR)model that combines global trend fitting with local residual capturing. A self-learning K-means algorithm is employed to identify a minimal input subset, thereby enhancing model training efficiency. The model is validated using real-world operational data from a power company in North China. Results show that the model maintains high accuracy, and the training outcomes based on the minimum subset are nearly identical to those using the full dataset.