Managing propagation of site-wide disturbances using model predictive control with forecasting
File(s)
Author(s)
Borghesan, Francesco
Type
Thesis
Abstract
Disturbances are undesired deviations of process variables from their set-points. Many disturbances are characterised by the repetitions of fast and complex patterns, and by the fact that such disturbances can propagate. If they propagate within a plant they are called plant-wide disturbances. If they propagate also to utilities of a process site, they are called site-wide disturbances.
Control systems play a role in the propagation of disturbances. A common requirement of a control system is to maintain a controlled variable at its set point by rejecting a disturbance. However, rejecting a disturbance can lead to further propagation of the disturbance elsewhere in the system. Hence there is a need for a trade-off. Plant operators can prefer to manually control a section of a production site to manage this trade-off, when a disturbance occurs.
To automatically manage the response to plant-wide and site-wide disturbances, this thesis proposes Model Predictive Control (MPC) with disturbance forecasting. Since there have been very few publications focusing on the causes and mechanisms of propagation of site-wide disturbances, this thesis contains a review of the possible disturbances occurring between process plants, steam utility and electrical utility. This analysis identifies the important scenarios where the proposed MPC framework might be useful, and also provides the research community a structured discussion about the topic of site-wide disturbances.
The thesis presents a k-nearest neighbours method for the prediction of disturbances. The algorithm uses a novel weighting method that makes the prediction more robust. The method is compared with other k-nearest neighbours methods and autoregressive models on industrial datasets.
The proposed MPC framework with disturbance forecasting is compared with other industrial frameworks on three simulated case studies. The case studies are a laboratory CSTH tank, an industrial steam boiler and an industrial distillation column. Incorporation of a prediction of the future evolution of a disturbance in model predictive control reduces the propagation of disturbances in most cases by a value between 65% and 95%, and improves the tracking of set-points up to 10%. This allows management of the trade-off between disturbance propagation and rejection of the disturbance from the controlled variables.
The approaches and novel results of the research are examined critically, with suggestions for future work.
Control systems play a role in the propagation of disturbances. A common requirement of a control system is to maintain a controlled variable at its set point by rejecting a disturbance. However, rejecting a disturbance can lead to further propagation of the disturbance elsewhere in the system. Hence there is a need for a trade-off. Plant operators can prefer to manually control a section of a production site to manage this trade-off, when a disturbance occurs.
To automatically manage the response to plant-wide and site-wide disturbances, this thesis proposes Model Predictive Control (MPC) with disturbance forecasting. Since there have been very few publications focusing on the causes and mechanisms of propagation of site-wide disturbances, this thesis contains a review of the possible disturbances occurring between process plants, steam utility and electrical utility. This analysis identifies the important scenarios where the proposed MPC framework might be useful, and also provides the research community a structured discussion about the topic of site-wide disturbances.
The thesis presents a k-nearest neighbours method for the prediction of disturbances. The algorithm uses a novel weighting method that makes the prediction more robust. The method is compared with other k-nearest neighbours methods and autoregressive models on industrial datasets.
The proposed MPC framework with disturbance forecasting is compared with other industrial frameworks on three simulated case studies. The case studies are a laboratory CSTH tank, an industrial steam boiler and an industrial distillation column. Incorporation of a prediction of the future evolution of a disturbance in model predictive control reduces the propagation of disturbances in most cases by a value between 65% and 95%, and improves the tracking of set-points up to 10%. This allows management of the trade-off between disturbance propagation and rejection of the disturbance from the controlled variables.
The approaches and novel results of the research are examined critically, with suggestions for future work.
Version
Open Access
Date Issued
2021-05
Date Awarded
2021-10
Copyright Statement
Creative Commons Attribution NonCommercial Licence
License URL
Advisor
Thornhill, Nina
Sponsor
Horizon 2020 European Commission
Grant Number
Grant Agreement no 675215
Publisher Department
Department of Chemical Engineering
Publisher Institution
Imperial College London
Qualification Level
Doctoral
Qualification Name
Doctor of Philosophy (PhD)