Steam Temperature Optimization of Coal-Fired Units Coupled with LSTM-XGBoost-GA Intelligent Algorithm

Data driven modeling techniques have been extensively applied to optimize power plant operations and enable intelligent control. Techniques including support vector machines, neural networks, and long short term memory (LSTM) architectures have significantly enhanced energy efficiency as well as operational safety. In this study, LSTM and eXtreme Gradient Boosting (XGBoost) models are used to predict main steam temperature, reheat steam temperature and heat storage difference, achieving coefficients of determination ($\mathrm{R}^{2}$) above 0.97. A genetic algorithm (GA) is then employed to optimize steam temperatures, keeping their variations within the unit's stable operating range. The results indicate that simultaneously optimizing main and reheat steam temperatures leads to a maximum temperature deviation$0.56-0.90^{\circ} \mathrm{C}$greater than optimizing main steam alone; this deviation is reduced by$0.50-1.18^{\circ} \mathrm{C}$through considering the heat storage difference in the objective. This research realizes the accurate prediction of heat storage deviation and provides some detailed guides for reducing the steam temperature changes.