Modeling And Multi-objective Optimization Of Power Plant Boiler Combustion System Based On Extreme Learning Machine

Coal-fired power plant boiler power generation plays an important role in China's energy supply.Nitrogen oxides(NOx)emitted from boiler combustion is one of the main pollutants in the atmosphere,which has long affected air quality and human health.With the construction of "Smart Power Plant",the storage capacity of data of thermal power plant operation is also expanding.How to effectively utilize the large-scale historical operation data of power plants,and on this basis to improve the thermal efficiency of boiler combustion and reduce the emission of NOx is the focus of power plants.In this paper,for a coal-fired boiler in a power plant,firstly,the large-scale historical operation data are mined,and the data are re-sampled by using the Gauss Mixture Model Clustering algorithm;Secondly,the improved kernel principal component analysis(KPCA)algorithm based on stochastic gradient descent algorithm is used to reduce the dimension of the operation parameters of the boiler combustion system,remove the correlation between the input variables,and then the extreme learning machine(ELM)is used to establish the models with the boiler combustion thermal efficiency and NOx emissions as the output respectively.Finally,in order to achieve the combustion objectives of high combustion thermal efficiency and low NOx emission,two different multiobjective optimization methods are used to optimize the combustion system of the boiler,in order to guide the combustion of the boiler under different requirements.The ideas and specific work of this paper are as follows:(1)At present,the DCS system or SIS system of most power plants store a large number of historical operation data of boilers.Although these massive data are helpful to model the boiler combustion system,the boiler operation site is complex and the information collected by different detection devices is disorderly.Therefore,it is necessary to use effective data mining methods to process and analyze these system data,and then build the boiler combustion prediction model driven by data.In this paper,the historical data of a utility boiler are preprocessed firstly,and then a resampling method based on Gauss Mixture Model(GMM resampling)is proposed,and its validity is verified by experiments.(2)Boiler combustion system is a non-linear,strong coupling and large lag system.There are many operating parameters,and there is a complex coupling relationship between them.In this paper,an improved Kernel Principal Component Analysis(SKPCA)based on stochastic gradient descent algorithm is used to reduce the dimension of boiler operation parameters,remove the correlation of input variables,and take the extracted principal component characteristic matrix as the input of ELM to establish the boiler combustion system model.(3)The search ability of existing optimization algorithms is quite different,and the effect is different in different application background.How to select an appropriate optimization algorithm to achieve multi-objective optimization of boiler combustion system is a difficult problem.In this paper,two optimization methods are used for the multi-objective optimization of boiler combustion system.One is to transform the multi-objective optimization problem into a single-objective optimization problem by using the weight coefficient method,and to optimize it by using the particle swarm optimization(PSO)algorithm.The other is to directly use the multi-objective particle swarm optimization(MOPSO)algorithm to obtain a set of Pareto for each parameter operation of boiler combustion system.According to the results of these two sets of experiments,the two optimization methods are analyzed,and reasonable suggestions for multiobjective optimization of the boiler combustion system are proposed.