基于RBFNN的催化再生烟气二氧化硫浓度预测研究  

Prediction of Sulfur Dioxide Concentration inCatalytic Regenerated Flue Gas Based on RBFNN

作　　者：杨文玉 Yang Wenyu(State Key Laboratory of Safety and Control for Chemicals,SINOPEC Research Institute of Safety Engineering Co.,Ltd.,Shandong,Qingdao,266104)

机构地区：[1]中石化安全工程研究院有限公司化学品安全控制国家重点实验室,山东青岛266104

出　　处：《安全、健康和环境》2023年第2期35-40,共6页Safety Health & Environment

基　　金：中国石化重大科技项目(321123),催化裂化装置绿色低碳运行智能管控关键技术研究。

摘　　要：为了提前掌握催化裂化再生烟气中二氧化硫的排放浓度,有效动态指导烟气脱硫设施运行参数调节,研究开展了RBF和BP神经网络在催化裂化再生烟气二氧化硫浓度预测中的应用。通过业务和数据分析,确定了影响再生烟气二氧化硫浓度的工艺特征变量。利用2组采用不同方法清洗的数据,对比分析了RBF和BP神经网络模型在提前15 min情况下,预测再生器出口二氧化硫排放浓度的效果,结果表明2种模型的预测精度分别为90.36%和86.43%。RBF神经网络二氧化硫浓度预测模型经过400个工业样本测试,浓度预测值的最大误差为14.01 mg/m 3,最小误差为0.05 mg/m 3,平均误差为6.08 mg/m 3,满足企业现场应用的要求。In order to obtain the sulfur dioxide(SO 2)concentration in flue gas emission at the outlet of fluid catalytic cracking(FCC)regenerator in advance,and effectively guide the adjustment of operation parameters for denitrification units from the perspective of source control,the application study of radial basis function(RBF)neural network and back propagation(BP)neural network in the prediction of SO 2 concentration in flue gas at the outlet of FCC regenerator was conducted.The process characteristic variables that affected SO 2 concentration of flue gas were determined by chemical process analysis and data analysis.Using two sets of data cleaned by different methods,the prediction effect of RBF and BP neural network models on sulfur dioxide emission concentration at the regenerator outlet was compared and analyzed 15 min in advance.The results showed that the prediction accuracy of the two models were 90.36%and 86.43%,respectively.Among the 400 industrial test samples,the maximum error of sulfur dioxide concentration prediction model of RBF neural network was 14.01 mg/m 3;the minimum error was 0.05 mg/m 3;and the average error was 6.08 mg/m 3,meeting the requirements of industrial field application.

关 键 词：催化裂化 二氧化硫 径向基神经网络 神经网络 浓度预测 

分 类 号：X742[环境科学与工程—环境工程]