基于互信息–图卷积神经网络的燃煤电站NOx排放预测

Prediction of NO;Emissions of Coal-fired Power Plants Based on Mutual Information-graph Convolutional Neural Network

燃煤电站NO_(x)排放预测模型可提高脱硝经济性。NO_(x)排放机理复杂,相关性变量众多,有效的融合相关变量之间的信息,能提高NO_(x)排放预测精度。提出了一种基于互信息-图卷积神经网络的NO_(x)排放预测模型。基于某660MW燃煤电站的运行参数,计算影响NO_(x)排放的特征变量之间的互信息,设计特征变量间的邻接关系,获取特征邻接矩阵,构建了基于图卷积神经网络的NO_(x)排放预测模型。将所提出的NO_(x)预测模型与基于LSTM、BPNN和LS-SVM的典型NO_(x)预测模型进行对比,实验结果表明,MI-GCN预测模型具有较好的泛化能力和较高的预测精度。

NO_(x) emission prediction model of coal-fired power plant can improve denitrification economy.The NOx emission mechanism is complex,and there are many variables that effect the NO_(x) emissions.The effective fusion of the information between the correlation variables can improve the NO_(x) emission prediction accuracy.This paper presented a NO_(x) emission prediction model through mutual information-graph convolution neural network(MI-GCN).Based on the operation parameters of the 660MW coal-fired power plant,the mutual information between characteristic variables affecting NO_(x) emission was calculated,the adjacency relationship between characteristic variables was designed,the characteristic adjacency matrix was obtained,and the NO_(x) emission prediction model based on graph convolution neural network was constructed.The proposed NO_(x) prediction model was compared with the typical NO_(x) prediction models based on long short time memory(LSTM),BPNN and least squares support vector machine(LS-SVM).The experimental results show that the MI-GCN prediction model has better generalization ability and higher prediction accuracy.