Intelligent Modelling And Optimization Of NO_x Emission Concentration Based On Deep Learning

With a national thermal power generation capacity of 5,888.79 billion k Wh in2022,accounting for 66.55% of total power generation,thermal power genera tion remains the most significant source of electricity for the grid.Coal is the main energy source in thermal power generation,and its combustion produ ces harmful gases such as NO_x,which causes damage to the ecological environment.Optimization of the boiler combustion process is an effective way to reduce NO_x emissions.In order to be able to reduce the NO_x emission concentration while still ensuring that the boiler main steam temperature is not reduced,the intelligent modelling and optimization model for this study was designed as follows:(1)To address the issue of pre-processing the data of the primary variables,power plant boilers in actual opera tion are affected by the operating environment,resulting in the existence of outliers in the data col lected by the power plant DCS system,so the 3δ criterion is used to identify the outliers and replace the identified outliers with local averages to ensure the continuity of the data;the Min-Max method is used to normalize the magnitude The influence of the difference in magnitude on the modelling accuracy was eliminated.(2)Considering the large time delay,redundancy and coupli ng characteristics of the boiler combustion process.The time delay correlation analysis is carried out on the primary variables,and the time delay time with the highest correlation is selected to improve the correlation of the primary variables.To addre ss the characteristics of redundancy and coupling,the design uses convolution operations to extract features from the state va riables and partial least squares to carry out importance analysis to reduce the dimensionality of the model input variables,while reducing the impact of redundant variables on the prediction model and improving the model accuracy.(3)For boiler combustion process prediction model accuracy problem,for boiler combustion process prediction model accuracy problem,power plant boiler combustion process to generate NO_x is very complex,where the unit load,air distribution method and tail flue gas temperature and other factors are closely related to this process.As deep learning has good learning performance,it can effectively solve the problem of non-linear boiler combustion process,so deep learning neural network algorithm is used to establish the predict ion model of NO_x emission concentration and boiler main steam temperature.(4)For the NO_x emission optimization problem,in the actual production operation of the power plant,reducing the NO_x emission concentration will cause changes in the boiler main s team temperature at the same time.For the sake of the economy and environmental protection of the power plant boiler,the NSGA-II algorithm is used for bi-objective optimization,and two fitness functions are constructed.The Pareto optimal solution set c an be obtained to obtain the optimal decision variables.(5)For the use of operators,this study designs and develops an intelli gent optimization system for the boiler combustion process.The system contains five functional modules: user login interface,data management interface,intelligent algorithm interface,optimization management interface and user management interface.The system can meet the operational needs of the actual staff.The application value of the system is demonstrated by the intuitive query of the actual results,predicted results and optimization results of NO_x emission concentrations.Based on the operational data of actual production,the method proposed in this study is used to conduct experiments,and the prediction error of the o btained experimental results is within 4%,which can accurately predict the outlet NO_x emission concentration,and the constructe d dual-objective optimization algorithm can obtain the optimization requirements of NO_x emission concentration to satisfy the field operators.