Research On Secondary Air Prediction Of CFB Boiler Based On Stacking Fusion Model

Although thermal power generation has brought many benefits to economic development,it has also brought some negative effects,such as excessive exploitation of coal resources and excessive discharge of pollutants.Circulating fluidized bed boiler is a boiler using clean combustion technology,which promotes the transformation of thermal power generation from quantity supply to quality supply.Circulating fluidization technology can improve the combustion efficiency and enhance the adaptability of coal.It is one of the current development directions of energy technology and has been widely used in thermal power generation.However,the operation control of the circulating fluidized bed has not been fully automated,and some operations are still carried out through human experience at present.This control method has problems such as poor stability and poor economy.This article first describes the characteristics of circulating fluidized bed boilers and the current status of modeling theory.In view of the current status of industrial big data,the article introduces commonly used data preprocessing and mathematical modeling methods for power plant operation data.From the principle point of view,the combustion mechanism of the circulating fluidized bed boiler is analyzed.Combined with the historical operation data of the power plant,a stacking multi-model fusion algorithm based on Ridge Regression,XGBoost,Gate Recurrent Unit Neural Network(GRU)is proposed to predict and model the secondary air of the circulating fluidized bed boiler.The secondary wind prediction model that can adapt to various variable working conditions is obtained.Finally,the historical operating data of 214 days of non-heating period of a 350 MW supercritical circulating fluidized bed boiler of a power plant is used as training set and test set.After preprocessing steps such as data cleaning and data standardization,and the feature selection and integration,the model is trained.And compared the prediction ability of single algorithm and fusion model algorithm.The experimental results show that the MAPE value of the ridge regression model on the test set is 13.59%,the MAPE value of the GRU model on the test set is 13.55%,the MAPE value of XGBoost model on the test set is 8.96%.However,the MAPE value on the test set of the fusion model based on Stacking is only 5.19%.Therefore,the fusion algorithm based on Stacking has better adaptability and better prediction effect under variable working conditions.