Research On Modeling And Optimize Method Of NO_x Emission Based On Deep Learning

National Energy Administration statistics show that in 2020,the national thermal power generation capacity reached 52798.7 billion k Wh,about 71.2% of the total national power generation,accounting for an absolutely dominant position.However,the NO_x produced by the combustion of coal-fired boilers has a significant impact on the ecological environment.With the increasing national attention to environmental protection,the prediction and control of NO_x emission have been widely concerned.Aiming at the problems of low prediction model accuracy and difficulty in optimal control in the process of NO_x emission control in coal-fired power plants,a 1000 MW boiler was taken as the research object,and the research was carried out from four aspects of data processing,high-precision modeling,multi-objective optimization and system design.The specific contents are as follows:(1)In view of the problem that the field data collected contains a large number of outliers,the local outlier factor algorithm is adopted to identify the outliers in the data.The missing values and outliers were filled by moving average interpolation method.Min-max algorithm is used to eliminate the dimensionality effect of the data and make the target data in the same order of magnitude for normalization processing.(2)Aiming at the problem of low accuracy of NO_x emission prediction model of power plant,a prediction modeling method(EMD-DNN)based on deep learning was proposed.MI algorithm was used to analyze the correlation of input variables;In order to verify the influence of time delay on subsequent modeling accuracy,the time delay between the input data and the NO_x output was analyzed;The empirical mode decomposition algorithm is used to decompose the variables.Four data sets were established according to different data collection times.According to different data sets,the deep learning algorithm is adopted and a deep neural network is selected to forecast NO_x emissions.(3)Aiming at the problem of multi-objective optimal control of NO_x emissions,the secondary air volume is selected as the decision variables.Establish an optimization model with the goal of minimizing NO_x emission and improving boiler efficiency.Choose NSGAII algorithm model,obtaining the decision variables of pareto optimal solution set.(4)The deep learning-based NO_x emission optimization control system is developed by using C# language.The system includes five functional modules: user login,data loading,prediction modeling,multi-objective optimization and result display.Users can log in to the system according to their account and password;In the data loading interface,the variable information to be queried can be independently selected and graphed for display;In the prediction modeling interface,the appropriate modeling algorithm can be ticked for NO_x emission modeling;The multi-objective optimization operation based on NSGA-II algorithm comprehensively optimized NO_x emissions and boiler efficiency,and provided a variety of optimization suggestions for energy saving and efficient production of thermal power plants.The system is used for reference by the operation personnel of the power plant.The experimental results show that the proposed deep learning prediction method not only accurately predicts the boiler emissions,but also has a prediction error of less than 2%,The proposed multi-objective optimization control method can effectively reduce the emission of nitrogen oxides and improve the combustion efficiency of the boiler.