SCR烟气脱硝系统动态建模方法比较

Comparative Study on Dynamic Modeling Methods of SCR Denitration System

SCR系统脱硝过程普遍应用的建模方法分为机理建模和数据建模两种,但这两种方法之间的对比研究相对较少。确定两种方法各自的适用范围,从而选取适宜的建模方法是确保模型准确性的前提。文中根据E-R反应机理以及支持向量机(support vector machine,SVM)、BP神经网络(BP neural network,BPNN)、核偏最小二乘(kernel partial least squares,KPLS)等方法分别建立了SCR系统的机理和数据动态模型。采用现场实际运行数据对模型进行验证,并根据模型精度评价指标以及建模耗时对各种建模方法进行对比。结果表明,在局部工况样本条件下,机理模型计算精度更高,计算量较小。在全局工况样本条件下,数据模型的拟合和泛化能力更强,但是数据模型的计算量更大,建模耗时更长。

There are two kinds of modeling methods widely used in denitration processes of SCR system, which are mechanism modeling method and data modeling method, while the comparative study of these two methods is relatively lack. In order to ensure the accuracy of the model, it is necessary to determine the scope of application of two methods to select the appropriate modeling method. Mechanism model of SCR system was developed based on Eley-Rideal mechanism and data-driven models were construct based on support vector machine （SVM）, BP neural network （BPNN） and kernel partial least squares （KPLS） methods. The actual operation data was used to verify the models, model accuracy evaluation indexes and calculation time were employed to make the comparison between the models. The results show that when training samples are limited, the calculation of the mechanism model is more accurate and the computational complexity is less. When the training samples are sufficient, the fitting and generalization performance of the data-driven models are better. However, the calculation of the data-driven models is more complex and the modeling time is longer.