热虹吸式再沸器换热量软测量模型  

Soft-sensor for duty of thermosiphon reboiler

作　　者：孙蒋子豪 李红阳 祝铃钰[1] 

机构地区：[1]浙江工业大学化学工程学院,浙江杭州310014 [2]浙江鸿盛化工有限公司,浙江上虞201203

出　　处：《计算机与应用化学》2016年第12期1267-1271,共5页Computers and Applied Chemistry

基　　金：NSFC-浙江两化融合联合基金(U1509209)

摘　　要：在文献基础上分析了热虹吸式再沸器的两段机理模型,分析了影响再沸器运行的各变量关系。选择了液位、导热油入口温度与流量、精馏塔釜温度作为数据模型预测再沸器热量的输入变量。采用浅层神经网络方法建立数据模型,当隐含层只有两层时表现出了较好的运算速度与拟合精度的平衡。采用工业现场数据建立再沸器热量的软测量模型。考虑到再沸器的性能在不断变化,比较了几种数据采集时间策略。当训练样本不包含检测样本时,可以看到再沸器运行过程中,拟合精度将会随着设备参数、控制参数的变化而逐渐下降,均方根误差依然保持在2.5%以内。但当现场设备参数发生较大变化后,数据模型的适用性变差,需要重新训练以达到原来的精度。First principle model of the thermosiphon reboiler is introduced to analyze the relationship between variables. Input variables including liquid level, inlet temperature and the flowrate of the thermo oil, temperature at bottoms of the column are used to set up a neural network data-based model for heat duty. The neural network data model which has two hidden-layer is used in the project to trade off the accuracy and calculation expense. Industrial data is used to setup the soft-sensor model of reboiler heat duty. Considering the gradual changing of the reboiler performance, data selection strategy are discussed. It is proved that, as long as the thermosiphon re- boiler working, the performance of this equipment keep changing, then the accuracy of data model decreasing. The prediction accuracy is still higher than 97.5%. However, the model need to be re-trained in case of a big event like revamp of the equipment.

关 键 词：热虹吸式再沸器 神经网络 软测量 

分 类 号：TQ051.5[化学工程]