针对SCR入口NO_(x)浓度的EMD-Informer长序列预测综合模型

Comprehensive model for SCR inlet NO_(x) concentration prediction under long sequence prediction

准确预测选择性催化还原系统(SCR)入口NO_(x)浓度并量化喷氨是提高SCR效率和降低NO_(x)排放的关键。然而,用于测量发电厂NO_(x)浓度的连续排放监测系统(CEMS)存在严重的延迟问题,因此需要进行长序列预测来抵消这种延迟。本文提出了一种综合预测模型,结合了特征选择、数据预处理和深度学习,用于预测300 MW亚临界自然循环汽包锅炉的SCR入口NO_(x)浓度。首先,通过主成分分析法和基于知识的方法筛选特征变量,然后利用经验模态分解(EMD)将原始历史数据分解为一系列分量序列。随后,采用Informer模型对每个分量进行预测,最后将这些预测的分量重构得到NO_(x)浓度的预测。与其他深度学习预测方法相比,该模型在长序列预测任务中表现出色,为精确控制SCR系统提供了一种有前景的方法。

Accurate prediction of NO_(x)concentration and quantified injection of ammonia are crucial for increasing the efficiency of Selective Catalytic Reduction(SCR)systems and reducing NO_(x)emissions.However,the Continuous Emission Monitoring System(CEMS)used to measure NO_(x)concentration in power plants has significant delay issues.Therefore,long sequence prediction tasks of NO_(x)concentration are necessary.This paper introduces a novel data-driven hybrid approach,based on Empirical Mode Decomposition(EMD)and the Informer model,for predicting the SCR inlet NO_(x)concentration of a 300 MW subcritical natural circulation drum boiler as the research subject.This model employs Empirical Mode Decomposition(EMD)to decompose the original historical data into a series of component sequences,effectively separating trend signals from noise signals.Subsequently,the Informer model is used to predict each component,and these predictions are then reconstructed to form the predicted NO_(x)concentration.Compared to other deep learning prediction methods,this model exhibits outstanding performance in long sequence forecasting tasks,offering a promising approach for precise control of Selective Catalytic Reduction(SCR)systems for denitrification.