Adaptive just-in-time and relevant vector machine based soft-sensors with adaptive differential evolution algorithms for parameter optimization. (23rd November 2017)

• Record Type: Journal Article
• Authors: Liu, Yiqi
• Is Part Of: Chemical engineering science. Volume 172(2017)
• Page Start: 571
• Page End: 584
• Publication Date: 2017-11-23
• Subjects: Soft-sensors -- Just-in-time -- Relevant vector machine -- Differential evolution -- Parameters; Chemical engineering -- Periodicals; Génie chimique -- Périodiques; Chemical engineering; Periodicals; Electronic journals; 660
• DOI: 10.1016/j.ces.2017.07.006
• Languages: English
• ISSNs: 0009-2509

Abstract

Highlights: A Just-in-time (JIT) and Relevant-vector-machine (RVM) to perform perditions. Nature-inspired optimized algorithm that ensures optimal parameters selection for JIT and RVM. A JADE evolution algorithm to optimize parameters for JIT and RVM without hyper-parameter setting. A moving window methodology for improvement of the JIT and RVM model. Proposed method is efficiently attractive in a wastewater plant monitoring. Abstract: Just-in-time (JIT) and Relevant vector machine (RVM) are two of commonly used models for soft-sensors modeling, the efficiency of which is governed by few critical parameters and hyper-parameters significantly. These parameters are routinely selected by trial and error or experience, thus leading to over- or under-fitting for the prediction. Adaptive differential evolution with optional external archive (JADE) has been used to optimize the parameters of JIT and RVM in this paper. The resulted JADE-JIT and JADE-RVM based soft-sensors are further enhanced into an adaptive format by the moving window (WM) technique. The proposed methodologies are applied to prediction of hard-to-measured variables in the wastewater treatment plants (WWTPs) and successful results are obtained.