Short-term Prediction Of NO_x Emission In SCR Denitrification System Based On Jorda N Circulating Neural Network

In China,combined air pollution is becoming more and more serious,NOx is the main pollutant leading to acid rain and other problems.In order to reduce environmental pollution,in recent years,the state has made stricter regulations on NOx emission from power plants,requiring the concentration of NOx emission from coal-fired power plants to be lower than 50mg/Nm3.Among numerous denitrification technologies,Selective Catalytic Reduction(SCR)has become the most widely used and technologically mature flue gas denitrification technology throughout the world due to its advantages of efficient denitrification,simple structure and easy maintenance.The amount of SCR emission is an important basis for the evaluation of its function,which can reflect the denitrification efficiency,unit load and a series of operating conditions,so the prediction of NOx emission has an important significance for NOx emission control and SCR denitrification system operating conditions controlThere are massive data accumulation and different types of influencing factors in the denitrification system.The diversity of power plant data provides massive data for the digital construction of intelligent power plant.But the accumulation of a large number of data has also brought high data jumbled,computational complexity and other problems.To solve this problem,a multi-constraint framework based on deep learning was proposed in the field of computer vision.This framework optimized the network by minimizing luminosity errors and added constraints between discontinuous images to improve the performance of the model.The results show that the prediction based on deep learning is more accurate.Using this method for reference will provide a new method for data value extraction and emission prediction of power plantsIn this paper,SCR denitrification unit of a coal-fired power plant in Baotou was taken as the research object.In view of the characteristics of multi-variable strong coupling and timeliness of denitrification unit,principal component analysis(PCA)was used to preprocess SCR denitrification system data before modeling.Then,the Jordan circulating neural network model,which is different from traditional modeling,is used to combine the operation data of SCR denitration unit with time series to predict NOX emission.The main work content is as follows:(1)The current situation of NOx emission in China's thermal power industry is analyzed,and the NOx emission standards and control technologies in China's thermal power industry are systematically elaborated,so as to provide relevant data and theoretical support for NOx emission prediction in the thermal power industry(2)The principal component analysis method in data mining technology was used for feature extraction.More than 40 groups of relevant data were screened to select the variables that mainly affect NOx emission,eliminate the interference of redundant characteristic variables on the prediction results,and lay a foundation for the establishment of NOx prediction model(3)As DCS stores a dynamic time series data,this paper constructs a Jordan circular neural network prediction model that can dynamically remember historical information,and determines the optimal parameters of the model through comparison experiments.According to the RMS error and accuracy of the prediction results,the Jordan cyclic neural network prediction model has a good effect.(4)to establish the traditional feedforward neural network and least squares support vector machines(LS-SVM)forecasting model with Jordan cycle neural network forecasting model to forecast the NOx emissions of the export of SCR system and comparison,the results confirmed that Jordan cycle root mean square error of neural network model is the LS-SVM decreased 0.68,at the same time data correlation a feed-forward neural network model is improved about 16%.It can be concluded that the Jordan circular neural network modeling method in deep learning can greatly improve the prediction accuracy and reduce the prediction error compared with the traditional modeling method,and the prediction accuracy is obviously better than the feedforward neural network method and LS-SVM methodWith the continuous improvement of China's industrial intelligence,the high-dimensional trend of massive industrial control system data is becoming more and more serious.How to use this data for classification and prediction is a major problem in the future This paper will provide some references for the application of deep learning in the industrial field.