基于EEMD和卷积神经网络的电站NO_(x)排放预测

Prediction of NO_(x) Emissions from Power Plants based on EEMD and Convolutional Neural Network

精准的NO_(x)排放预测模型能够提高SCR系统的脱硝效率,为此本文分析了一维卷积神经网络在NO_(x)预测领域的应用,并提出了一种结合集成经验模态分解和卷积神经网络的NO_(x)排放预测方法。首先,对原始数据进行预处理,并采用互信息法确定输入变量。然后,采用集成经验模态分解算法对NO_(x)数据进行分解处理,降低NO_(x)数据的预测难度。最后,基于一维卷积神经网络构建各分量的预测模型并进行重构,得到最终的NO_(x)预测结果。基于某电厂的实际运行数据进行实验,实验结果表明,所提出模型预测结果的平均绝对百分比误差为3.34%。一维卷积神经网络的超参数实验说明了Adam优化方法和合适的输入步长有利于模型的训练,但是dropout正则化不利于模型的性能提升。

Accurate NO_(x) emission prediction model can improve the denitration efficiency of SCR system.For this,this paper analyzes the application of one-dimensional convolutional neural network in the field of NO_(x) prediction,and proposes a combination of ensemble empirical mode decomposition(EEMD)and convolutional neural network for NO_(x) emission prediction methods.Firstly,the original data is preprocessed,and the mutual information method is used to determine the input variables.Secondly,the ensemble empirical mode decomposition algorithm is applied to decompose the NO_(x) data to reduce the difficulty of NO_(x) data prediction.Finally,the prediction model of each classification is constructed based on the one-dimensional convolutional neural network and reconstructed to obtain the final NO_(x) prediction result.Based on the actual operating data of a power plant,the experiment results show that the average absolute percentage error of the proposed model′s prediction results is 3.34%.The hyperparameter experiment of the one-dimensional convolutional neural network shows that the Adam optimization method and the appropriate input step size are conducive to the training of the model,but the dropout regularization is not conducive to the performance improvement of the model.