基于随机森林的火电机组SCR脱硝反应器建模

Modeling of the SCR Denitration Reactor in a Thermal Power Unit Based on Random Forest Algorithm

火电机组选择性催化还原技术(SCR)脱硝反应器过程复杂多变,采用机理建模的SCR脱硝反应器出口NO_x质量浓度预测难以取得良好的效果。基于火电厂的历史运行数据,将主成分分析(PCA)和随机森林(RF)相结合建立了SCR脱硝反应器出口NO_x质量浓度预测模型。在建模过程中,采用主成分分析方法计算各个变量的贡献率来筛选变量,进而对随机森林模型进行试验验证,并与支持向量机(SVM)模型和BP神经网络模型的预测性能进行对比。结果表明:采用PCA变量选择方法确定SCR系统模型的输入变量是可行和有效的;与SVM和BP神经网络模型相比,RF算法得到的SCR系统模型具有更好的预测效果。

In view of the complex and variable process of the SCR denitration reactor in a thermal power unit,the NO_x concentration at the reactor outlet is hard to be accurately predicted by mechanism modeling.Based on the historical operation data of thermal power plants,by combining the principal component analysis(PCA)with random forest(RF)algorithm,a dynamic model was established for the SCR denitration reactor.In the modeling process,the PCA method was used to calculate the contribution rate of each variable to filter the variables,then the random forest model was experimentally verified,and its prediction performance was compared with that of the support vector machine(SVM)and back propagation neural network(BPNN).Results show that it is feasible and effective to use the PCA variable selection method to determine the input variables of the SCR system model.Compared with SVM and BPNN,the SCR system model obtained by RF algorithm has better prediction effect.