基于PCA-LMBP神经网络模型的SCR脱硝催化剂工艺特性预测  

Process characteristics forecasting for SCR denitration catalyst based on PCA-LMBP neural network model

作　　者：林正根 姚杰 庄柯 金定强 吴碧君 LIN Zhenggen;YAO Jie;ZHUANG Ke;JIN Dingqiang;WU Bijun(Guodian Environmental Protection Research Institute Co.,Ltd.,Nanjing 210031,China)

机构地区：[1]国电环境保护研究院有限公司

出　　处：《热力发电》2019年第11期108-114,共7页Thermal Power Generation

基　　金：国家重点研发计划项目(2016YFC0208102)~~

摘　　要：为了利用烟气脱硝催化剂几何及理化特性实现对催化剂固定工况下脱硝效率和活性的预测,本文分析了选择性催化还原(SCR)脱硝催化剂工艺特性与几何及理化特性之间的关联特征,利用相关系数矩阵分析、主成分分析(PCA)简化输入参数,提出改进的神经网络预测模型。通过大量样本数据的训练,建立了催化剂脱硝效率和活性的PCA-LMBP神经网络预测模型,利用该模型对实际测量数据进行模拟预测。模型预测结果与实际值对比表明,所建立的PCA-LMBP神经网络预测模型具有较高的准确性,对于烟气脱硝催化剂性能检测、质量监控及相关的技术服务有指导意义。To realize prediction of denitration efficiency and activity of the catalyst under stationary test condition by using geometric and physicochemical characteristics of the flue gas denitration catalyst,the correlation between the process characteristics and geometric and physical/chemical properties of selective catalytic reduction(SCR)denitrification catalyst was analyzed.The correlation coefficient matrix analysis and principal component analysis(PCA)were adopted to simplify the input parameters,and the improved neural network predictive model was proposed.Through training of a large amount of samples,the prediction model of denitration efficiency and activity for SCR denitration catalyst was established and then used to carry out prediction on the basis of the actual measurement data.The results indicate that,the PCA-LMBP has a high accuracy,which has great guide meaning and application value on the performance test and quality monitoring works for SCR denitration catalysts.

关 键 词：SCR烟气脱硝 催化剂 主成分分析 LMBP神经网络 脱硝效率 工艺特性 活性 

分 类 号：X511[环境科学与工程—环境工程]