Thermal Process Data Modeling Based On Mutual Information Variable Selection

In recent years,with the continuous improvement of the information level of power plants,thermal process data modeling based on historical data driven has gradually become one of the hot topics of research.The method effectively overcomes the defects of the insufficient precision of the mechanism modeling and the large workload of the experimental modeling.The NOx generation of the boiler combustion system is a typical thermal process.There are many factors affecting the generation of NOx,and the coupling of variables is serious.According to the combustion system of power plant boiler,the correlation variables that affect the generation of NOx are analyzed by mutual information method,and accurate prediction of NOx generation volume by using appropriate data modeling method is an important part of effectively reducing NOx emissions.This paper first discusses the research background of data modeling for thermal processes,and then to the boiler combustion system NOx production process as the research object,detailed introduces the significance and status of the variable selection and modeling of NOx data;the second chapter is the basic framework of the variable selection methods of different key generation and evaluation criteria for the subset of the advantages and disadvantages are introduced the commonly used variable selection methods and each method,selection algorithm for the next chapter of the variable theory;innovation the third chapter puts forward a selection method of mutual information variables based on dynamic data,the theory of k nearest neighbor mutual information and order mutual information based on the correlation metric and use conditional mutual information and forward search strategy to ensure that the selected subset of variables of total amount,the effectiveness of the proposed method is confirmed by simulation experiment.Finally use the combustion system of a power plant boiler,to identify the impact of relevant variables on the NOx generation;the fourth chapter will dominate factor screening in Chapter third as input variables,and then use genetic algorithm to optimize the LSSVM parameters,establishes the dynamic NOx content prediction model.Through the actual operation data,the dynamic prediction model can effectively improve the prediction accuracy of the NOx generation.