PRESTO: Predictive REcommendation of Surrogate models To approximate and Optimize. (15th February 2022)

• Record Type: Journal Article
• Authors: Williams, Bianca; Otashu, Joannah; Leyland, Simon; Eden, Mario R.; Cremaschi, Selen
• Is Part Of: Chemical engineering science. Volume 249(2022)
• Publication Date: 2022-02-15
• Subjects: Model selection -- Surface approximation -- Surrogate-based optimization -- Meta-learning; Chemical engineering -- Periodicals; Génie chimique -- Périodiques; Chemical engineering; Periodicals; Electronic journals; 660
• DOI: 10.1016/j.ces.2021.117360
• Languages: English
• ISSNs: 0009-2509

Abstract

Highlights: Selecting a surrogate modeling technique depends on the characteristics of the data being modeled. We identified attributes of data that are appropriate for use in selecting models. We developed PRESTO, Predictive REcommendation of Surrogate Models to Approximate and Optimize. PRESTO selects surrogate models for data based on its attributes. PRESTO provides accurate model selections for a case study of cumene process simulation data. Abstract: Surrogate models are used to map input data to output data when the actual relationship between the two is unknown or computationally expensive to evaluate. Many techniques exist for surrogate modeling; however, selecting suitable techniques for a given application remains an open challenge. This work describes PRESTO, a Random Forest classifier-based tool, to recommend appropriate surrogate modeling techniques for a given dataset for surface approximation and surrogate-based optimization, using attributes calculated only using the input and output data. The tool identifies the techniques for surface approximation with an accuracy of 91% and a precision of 90% and for surrogate-based optimization with an accuracy of 98% and a precision of 99%. PRESTO was tested on data generated from a high fidelity process model of the cumene production process. Its performance on this case study was comparable to the training data. PRESTO enables computational time savings for selecting surrogate model forms by avoiding expensive trial-and-error methods.