Development of a debutanizer column soft sensor by means of evolutionary machine learning approaches

Soft Sensors have been widely employed to predict variables from a process that cannot be measured. In literature, there are many techniques that can be used to develop a soft sensor. The robustness of the model is directly related to the complexity of the techniques employed in the model. However, it does not mean that simple models are not able to achieve good results. Support Vector Machines (SVM) and Decision Trees (DT) has been shown as good candidates to build soft sensors. Two experiments are proposed in this work. First, three benchmark regression data sets are studied. Finally, a debutanizer column is provided as study case in order to build a soft sensor. Using Genetic Algorithm (GA) as a tool to improve the developed models, it is possible to perceive that there are no significant differences between the results achieved with SVM and DT in terms of RMSE.