Abstract
This study is aimed at the issue of energy waste resulting from significant fluctuations in the energy consumption of the steam turbine system under the flexible peaking demands of coal-fired units. To accurately predict the energy consumption of these units across a wide range of load conditions, the energy consumption prediction model of eXtreme Gradient Boosting (XGBoost) steam turbine system is established. First, the model variables are chosen based on the existing measurements and an analysis of the power plant. Meanwhile, the energy consumption dataset and its distribution are calculated by the consumption rate analysis. Second, the model feature variables are screened by the maximum information coefficient (MIC) and Kendall rank correlation coefficient, and the energy consumption prediction model of the 660 MW steam turbine system based on XGBoost is established. Finally, the Bayesian optimization (BO) algorithm is employed to determine the best hyperparameters of the XGBoost model. Moreover, three energy consumption prediction models of MIC-BO-XGBoost are built for multi-objective prediction: independent modeling, chain modeling 1, and chain modeling 2. Chain modeling 2 is capable of forecasting the energy consumption of the steam turbine system in ultra-supercritical coal-fired units with greater precision under wide variations of load. It can provide the basis for the operation optimization of the steam turbine system of subsequent coal-fired units.