一种基于卷积特征的NO_(x)排放浓度深度预测模型

A Deep Prediction Model of NO_(x)Emission Concentration Based on Convolution Features

为提高燃煤锅炉出口NO_(x)排放浓度的预测精度,该文提出一种基于卷积特征提取的NO_(x)排放浓度深度学习预测模型。首先,分析锅炉燃烧产生NO_(x)的机理,确定模型初选变量;然后,计算不同初选变量与NO_(x)排放浓度之间的最大相关系数,确定变量延迟时间;其次,为挖掘输入变量深层交互信息,设计二维卷积网络进行特征提取,获得高维预测模型输入候选集合;同时,通过偏最小二乘法计算候选集变量与NO_(x)排放浓度之间相关性,降低输入变量维数,确定最终模型的输入变量;最后,设计深度神经网络建立NO_(x)排放预测模型,预测NO_(x)排放浓度。基于1000MW锅炉实际运行数据的实验结果表明,所提出卷积深度神经网络预测算法的平均相对百分比误差小于4%,预测精度能够满足实际生产的需求。

In order to improve the prediction accuracy of NO_(x) emission concentration at the outlet of coal-fired boilers,a deep learning model based on convolution feature extraction is proposed.First,the generation mechanism of NO_(x) is analyzed to determine the primary selection variables of the model.Then,maximal information coefficient(MIC)between the different primary variables and the NO_(x) emission concentration are calculated to determine the delay time.Next,to dig out the depth interactive information of primary variable in a two-dimensional convolution network,feature extraction is developed to obtain a high-dimensional predictive model input candidate set.The partial least square method is employed to calculate the importance of candidate set variables and NO_(x) emission concentration to reduce the dimensionality of input variables and determine the input variables of the final model.Finally,a deep neural network is designed to establish a NO_(x) emission prediction model to predict the NO_(x) emission concentration.Experimental results based on actual operation data of 1000MW boilers illustrated that the MAPE of the proposed convolution deep neural network is less than 4%.And the prediction accuracy can meet actual production requirements.