| As an important part of the energy supply system,coal-fired boilers play an important role in the fields of petroleum,chemical industry,metallurgy,and civil heating.At the same time,the nitrogen oxides(NOx)generated by the combustion of raw coal in the furnace pose certain hazards to the natural environment and human health.How to effectively monitor pollutant emissions in real-time,reduce emissions while improving boiler combustion thermal efficiency is currently an important task faced by the coal-fired boiler industry.Aiming at the problems of low accuracy,poor real-time performance and complex optimization process of nitrogen oxide measurement in the combustion process,this paper takes a 300 MW coal-fired boiler as the research object,and conducts research from four aspects: data preprocessing,correlation analysis and feature selection,Stacking model fusion and multi-objective optimization.The main research work includes the following aspects:(1)Screen data collected in DCS data system and eliminate outliers through Laida criteria;The linear interpolation method is used to fill the vacant values,so that the time labels match and the data integrity is improved.(2)Through the Pearson correlation coefficient method,the maximum mutual information coefficient method(MIC),random forests embedding method combination of feature selection for operation variables.Empirical mode decomposition algorithm was used to decompose variables for secondary feature selection,and the linear and nonlinear relationship between variable data was fully mined.Through feature selection,the problem that there are many influencing factors of NOx emission and they are coupled with each other could be overcome,so as to prepare for subsequent modeling experiments.(3)The k-means clustering algorithm is adopted to sample,to distinguish the data belongs to different conditions.A hierarchical model based on the Stacking integrated method is established,the model parameters are optimized,and the high-precision prediction models for boiler NOx emission and combustion efficiency under different stacking conditions are established.(4)The basis of the above to establish forecasting model,the structure of combustion optimization objective function,using two different optimization methods were analyzed,one is a single objective optimization based on genetic algorithm(GA).Another is to use non dominated sorting genetic algorithm(NSGA-Ⅱ)combustion system of decision variables is obtained Pareto optimal solution set.The results show that the constructed Stacking fusion model can accurately predict NOx emission and boiler thermal efficiency with errors less than 5%.Multi-objective optimization strategy can effectively at the same time the implementation of the boiler combustion efficient and low emission,optimization of reference for power plant energy saving production. |
