{"title":"Building Empirical Models of Process Plant Data by Regression or Neural Network","authors":"T. Cheung, O. Kwapong, J. Elsey","doi":"10.23919/ACC.1992.4792451","DOIUrl":null,"url":null,"abstract":"Although neural network has generally been recognized as a useful tool for empirical data modelling, its utility in modelling industrial process plant data needs to be compared against conventional statistical techniques. This paper presents such a comparison through two case studies. In each case, data from a real refinery frationator were modelled by neural network and by linear regression. The models correlate process measurements to stream properties which were measured by low frequency lab tests. Results from the two cases show that neural network is useful for modelling process data which contain nonlinearities. However, its performance cannot be better than linear regression model when nonlinearities cannot be observed in the data. Although many processes are nonlinear, weak nonlinearities may be difficult to observe in industrial process data which are often noisy. Linear regression models are more appropriate when noise in the data mask nonlinearities. Analyzing the residuals of the linear model helps determine if observable nonlinearities are present in the data.","PeriodicalId":297258,"journal":{"name":"1992 American Control Conference","volume":"19 1","pages":"0"},"PeriodicalIF":0.0000,"publicationDate":"1992-06-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"8","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"1992 American Control Conference","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.23919/ACC.1992.4792451","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 8
Abstract
Although neural network has generally been recognized as a useful tool for empirical data modelling, its utility in modelling industrial process plant data needs to be compared against conventional statistical techniques. This paper presents such a comparison through two case studies. In each case, data from a real refinery frationator were modelled by neural network and by linear regression. The models correlate process measurements to stream properties which were measured by low frequency lab tests. Results from the two cases show that neural network is useful for modelling process data which contain nonlinearities. However, its performance cannot be better than linear regression model when nonlinearities cannot be observed in the data. Although many processes are nonlinear, weak nonlinearities may be difficult to observe in industrial process data which are often noisy. Linear regression models are more appropriate when noise in the data mask nonlinearities. Analyzing the residuals of the linear model helps determine if observable nonlinearities are present in the data.