| Thermal power has two obvious drawbacks: one is the huge consumption of fossil energy,the other is the emission of flue gas contains a variety of pollutants.With the constant improvement of the human environmental protection consciousness and the "double carbon" policy,under the influence of thermal power once mighty status has been shaken,so you need to optimize the traditional thermal power transformation,to ensure the thermal power plants to meet under the premise of high efficiency,environmental protection,can for our country electric power structure caused by thermal power to provide cushion of the transformation of new energy power generation.In this paper,taking the 330 MW boiler of a power plant as the research object,the multi-objective optimization of combustion state based on neural network and intelligent calculation is carried out to reduce NOx emissions without affecting the thermal efficiency of the boiler.First of all,the NOx formation mechanism of coal-fired power plant boiler and boiler thermal efficiency characteristic has made the detailed analysis,using the method of single control variables of the boiler was carried out including oxygen,secondary air,burning wind,the combustion adjustment test and found a set of NOx emissions and boiler thermal efficiency with the parameters change rule,used to guide subsequent combustion tuning and optimization results correctness verification.Secondly,the deep learning algorithm is applied to the boiler combustion system modeling,and the DNN prediction model of boiler NOx emission and thermal efficiency is established.The data used for modeling were from the boiler’s historical operation data in DCS.The data were cleaned before modeling,and PCA was used to eliminate redundant variables and simplify the input of the model.Then,Preason correlation coefficient method was used to analyze the time-delay correlation between the inputs and outputs of the model.The sampling interval was1 min,and the input at the moment of the most significant correlation was selected as the final modeling input.Finally confirmed 16 input,4 hidden layer,double output network model of the structure,before and after treatment of time delay and in the history of boiler operation data of the model test,the test results show that the model of NOx emissions within DNN better prediction effect and the thermal efficiency of the boiler,to delay the data after it possesses much higher prediction accuracy of the model.Finally,the multi-objective evolutionary algorithm based on decomposition of MOEA/D join population migration mechanism,using the improved I-MOEA/D algorithm for boiler operation parameters optimization,will be built within DNN prediction model as the algorithm of the fitness evaluation function,set the adjustable parameter range,and contrasted the performance optimization of before and after the improvement algorithm,the optimization results show that The Pareto frontier searched by the improved I-MOEA/D algorithm is more ideal,the objective function of NOx emission and boiler heat loss are reduced,and the optimized operating parameters are in line with the low nitrogen combustion theory and the conclusion of combustion adjustment test. |
