Bibliographic citations
This is an automatically generated citacion. Modify it if you see fit
Mendoza, J., (2017). Performance indicators and monitoring of multi-stage NMPC in a real-time software framework [Technische Universität Dortmund]. https://renati.sunedu.gob.pe/handle/sunedu/3335852
Mendoza, J., Performance indicators and monitoring of multi-stage NMPC in a real-time software framework []. DE: Technische Universität Dortmund; 2017. https://renati.sunedu.gob.pe/handle/sunedu/3335852
@mastersthesis{renati/4108,
title = "Performance indicators and monitoring of multi-stage NMPC in a real-time software framework",
author = "Mendoza Ramírez, Jorge Daniel",
publisher = "Technische Universität Dortmund",
year = "2017"
}
Title: Performance indicators and monitoring of multi-stage NMPC in a real-time software framework
Other Titles: Indicadores de desempeño y monitoreo de NMPC multietapa en un entorno de software en tiempo real
Authors(s): Mendoza Ramírez, Jorge Daniel
Advisor(s): Tatulea-Codrean, Alexandru; Engell, Sebastian
Keywords: Control predictivo; Control en tiempo real; Automatización
OCDE field: https://purl.org/pe-repo/ocde/ford#2.02.02
Issue Date: 15-Dec-2017
Institution: Technische Universität Dortmund
Abstract: El modelo de control predictivo (MPC) se ha utilizado ampliamente en la industria de procesos con notables beneficios económicos desde hace bastante tiempo, a pesar de que aún no se dispone de un enfoque sistemático para evaluar su rendimiento. Este inconveniente no solo dificulta el monitoreo y el mantenimiento de las implementaciones actuales, sino que también restringe las posibilidades de que las nuevas iniciativas en MPC se implementen en plantas reales.
En esta tesis se expone una metodología para la evaluación del rendimiento de estos controladores avanzados, teniendo especial consideración para aplicaciones basadas en control predictivo de modelos no lineales multietapa (NMPC multietapa), una nueva formulación propuesta para manejar sistemas no lineales con incertidumbres de manera eficiente.
En este contexto, el costo de la etapa se propone como la medida fundamental del desempeño. A partir de esta definición, se pueden calcular algunos índices de desempeño por medio de comparaciones entre el costo de etapa logrado en el proceso y el costo de etapa pronosticado por el controlador.
El costo de la etapa pronosticado como punto de referencia de rendimiento debe calcularse en general mediante simulaciones de Monte Carlo. No obstante, para casos especiales como sistemas lineales con distribuciones estadísticas normales y sistemas con no linealidades polinómicas, se pueden utilizar algunos procedimientos de cálculo eficientes.
Además, se explora el uso de cadenas de Markov para manejar casos con distribuciones complejas.
Las metodologías propuestas se ilustran con aplicaciones en estudios de casos industriales. Se muestra que la gama de posibles aplicaciones no solo se limita a la supervisión del rendimiento, sino que también cubre casos como la selección del controlador, la detección de fallas y la capacitación de los operadores.
Model predictive control (MPC) has been widely used in the process industry with reportedly remarkable economic benefits for quite a while now in spite of the fact that no systematic approach to evaluate its performance is yet available. This inconvenience not only hinders monitoring and maintenance of current implementations but also restricts the chances for new initiatives in MPC to be implemented in real plants. In this thesis a methodology for performance assessment of such advanced controllers is exposed, taking some special consideration for applications based on multi-stage nonlinear model predictive control (multi-stage NMPC), a novel proposed formulation to handle nonlinear systems with uncertainties efficiently. In this context, the stage cost is proposed as the fundamental measure of performance. From this definition, some performance indices can be computed by means of comparisons between the achieved stage cost in the process and the predicted stage cost from the controller. The predicted stage cost as a performance benchmark has to be computed in general by means of Monte Carlo simulations. Nonetheless, for special cases such as linear systems with normal statistical distributions and systems with polynomial nonlinearities some efficient computation procedures can be utilized. Additionally, the use of Markov chains is explored in order to handle cases with complex distributions. The proposed methodologies are illustrated with applications on industrial case studies. The range of possible applications is shown to be not only restricted to performance monitoring but also to cover cases such as controller selection, fault detection and training of operators.
Model predictive control (MPC) has been widely used in the process industry with reportedly remarkable economic benefits for quite a while now in spite of the fact that no systematic approach to evaluate its performance is yet available. This inconvenience not only hinders monitoring and maintenance of current implementations but also restricts the chances for new initiatives in MPC to be implemented in real plants. In this thesis a methodology for performance assessment of such advanced controllers is exposed, taking some special consideration for applications based on multi-stage nonlinear model predictive control (multi-stage NMPC), a novel proposed formulation to handle nonlinear systems with uncertainties efficiently. In this context, the stage cost is proposed as the fundamental measure of performance. From this definition, some performance indices can be computed by means of comparisons between the achieved stage cost in the process and the predicted stage cost from the controller. The predicted stage cost as a performance benchmark has to be computed in general by means of Monte Carlo simulations. Nonetheless, for special cases such as linear systems with normal statistical distributions and systems with polynomial nonlinearities some efficient computation procedures can be utilized. Additionally, the use of Markov chains is explored in order to handle cases with complex distributions. The proposed methodologies are illustrated with applications on industrial case studies. The range of possible applications is shown to be not only restricted to performance monitoring but also to cover cases such as controller selection, fault detection and training of operators.
Link to repository: https://renati.sunedu.gob.pe/handle/sunedu/3335852
Discipline: Automatización y Robótica
Grade or title grantor: Technische Universität Dortmund. Fakultät für Elektrotechnik und Informationstechnik
Grade or title: Maestro en Ciencias
Register date: 22-Dec-2022
Files in This Item:
File | Description | Size | Format | |
---|---|---|---|---|
MendozaRamirezJD.pdf | Tesis | 2.61 MB | Adobe PDF | ![]() View/Open |
Autorizacion.pdf Restricted Access | Autorización del registro | 431.1 kB | Adobe PDF | View/Open Request a copy |
This item is licensed under a Creative Commons License