West Virginia University
Triggered Online System Re-Identification Applied to Model Predictive Control Using Gaussian Processes
Abstract
dc:description.abstractSafety, product quality, enhanced performance, and increased profit all depend on the control of chemical and energy processes. However, operational issues can lead to control challenges, especially when processes are subject to disturbances during their operation. Process control methods work best when processes operate close to their designed operat- ing conditions, but lack of performance or other issues may occur when the process is far from such conditions. To overcome these challenges, in this dissertation, online model re- identification is proposed for Model Predictive Control (MPC). This involves reassessing the predictive model of an advanced controller, namely MPC, when re-identification conditions are met based on activating a pre-defined trigger condition that considers the mismatch be- tween model prediction and actual process values. To re-identify the model, in this case a Gaussian process (GP), the size of the dynamic dataset and its sampling time need to be established, besides the regressors of the model. A pre-assessment of the GP model is used to select the appropriate dataset size and time step to satisfy dynamic transitions. The se- lection of the regressors, past inputs, and outputs required for Gaussian process nonlinear autoregressive with exogenous inputs (GP-NARX) regression model is also performed. Addi- tionally, the steady-state detection (SSD) analysis for data classification and dataset selection is carried out to ensure the minimum amount of dynamic and steady-state data is part of the dataset. To guarantee improved performance, the new model is verified with a regressed dataset and validated with new data before replacing the current MPC model. The proposed framework is of significant importance, and it is demonstrated in the context of a continuous stirred tank reactor (CSTR), where a first-order reaction occurs and a large-scale fuel-fired power plant. The processes input, state, and output variables are stored using a simulated Industry 4.0 infrastructure. The infrastructure simulates an interconnected environment, where various components can communicate and share data for use in both academic and industrial contexts. The existing infrastructure consists of a distributed control system, data analytics components, online load demand, and a power plant model. The examples’ findings show how the developed infrastructure can regulate the process and adjust the predictions.
Degree
thesis:*- Name thesis:degree_name
- PhD
- Level thesis:degree_level
- Dissertation
- Discipline thesis:degree_discipline
- Chemical and Biomedical Engineering
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Kestering, Daniel Augusto
- Contributors dc:contributor
-
- Fernando V. Lima
- Debangsu Bhattacharyya
- Stephen Zitney
- Heleno Bispo
- Paolo Pezzini
- Mario Perhinschi
Subjects
dc:subject × 6Identifiers
dc:identifier.*- Identifier
-
<p><a href="https://orcid.org/0009-0000-0790-8861">https://orcid.org/0009-0000-0790-8861</a></p>
https://researchrepository.wvu.edu/etd/12643