Collaborative Prediction of Baseline Functions of Wind, Coal and Water Based on Stacked Generalized Integrated Learning for a Coal-fired Plant

Under the background of China's construction of a new type of power system, thermal power is faced with the new situation of frequent peak and frequency adjustments, which puts forward higher requirements for the flexibility. Wind, coal and water are three key variables in the control of coal-fired units. With time going on, they mismatch with each other for the shift of combustion characteristics of the boiler and the deviation between the coal type in use and the designed coal type; and then economy and flexibilities decrease. To solve the problem, a method based on machine learning to predict the wind-coal-water coordination is proposed. Firstly, the R-Value algorithm and local outlier algorithm are used to eliminate the anomalies; secondly, the Least Squares algorithm, the Support Vector Machine algorithm and the Polynomial Fitting algorithm are used as the basis function to fit the wind-coal-water function under different load conditions, respectively; finally, the stacking generalization ensemble learning model is used as the second-level integration algorithm to make the final prediction. The application of the modified wind-coal-water matching function in a 1000MW ultra-supercritical unit shows that the wind-coal-water matching function is more coordinated, critical parameters’ tracking performance is improved, and the unit operates more smoothly and economically.