Soft-Sensor Design for a Crude Distillation Unit Using Statistical Learning Methods
Computer Aided Chemical Engineering, Elsevier B.V., ss.2269-2274, 2018
- Yayın Türü: Kitapta Bölüm / Araştırma Kitabı
- Basım Tarihi: 2018
- Doi Numarası: 10.1016/b978-0-444-64241-7.50373-6
- Yayınevi: Elsevier B.V.
- Sayfa Sayıları: ss.2269-2274
- Anahtar Kelimeler: adaptive modeling, cross-validation, predictive modeling, predictor selection
- Boğaziçi Üniversitesi Adresli: Evet
Özet
Data-driven soft-sensors are statistical models constructed from historical data, and expected to perform well both in normal and novel process conditions. Numerous adaptive mechanisms for soft-sensors have been developed, but more work is required to develop appropriate statistical modeling tools for chemical processes, which yield highly collinear measurements in changing operating conditions. In the current study, numerous predictor subset selection and statistical learning tools are employed on 12 different data sets from a crude distillation unit simulated in SimSci PROII™ and the generalization capacities of the models are evaluated using the test points inside and outside the operating regions of the training sets. Relevance vector machine is found to be promising for adaptive modeling, with accuracies close to those of neural network models.