Thesis ESTIMACIÓN DE PERTURBACIONES PARA SU INCORPORACIÓN EN LA OPTIMIZACIÓN EN TIEMPO REAL EN SISTEMAS DE SUPERVISIÓN DE PROCESOS
Date
2018-08
Authors
GONZÁLEZ PASMIÑO, FÉLIX ALEJANDRO
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
La optimización en tiempo real (RTO) es una metodología sistemática e iterativa que busca las condiciones óptimas de un proceso, a pesar de la presencia de incertidumbres en el modelo que lo describe. Las perturbaciones pueden afectar el desempeño de esta metodología debido al cambio que estas generan en las condiciones del sistema, por lo que es necesario incluir información de éstas en la RTO. El presente trabajo tiene como objetivo general estudiar e implementar metodologías de estimación de perturbaciones para su incorporación en optimización en tiempo real con adaptación de modificadores. Para esto se utilizará el modelo del Reactor de Otto Williams como sistema experimental, al cual se le realizarán cambios en el flujo de alimentación de uno de sus componentes, asumiendo que este cambio no puede ser medido. La optimización en tiempo real con adaptación de modificadores permite llegar al óptimo del proceso aún en presencia de incertidumbre estructural, ya que ocupa un sistema de aprendizaje automático basado en el cálculo de los gradientes del proceso, utilizando información pasada del sistema. Debido a que esta metodología está basada en información pasada, la presencia de perturbaciones afecta directamente a la estimación de los gradientes del proceso, puesto que, al generar un cambio en las condiciones del sistema, es necesario evaluar qué parte del cambio observado en las variables dependientes es producto de las perturbaciones y qué parte se explica por modificaciones en las variables de decisión. Puesto que las perturbaciones no siempre son medidas, esto podría comprometer la efectividad de los modificadores a la hora de alcanzar el óptimo del proceso. Para el estudio de estimadores de perturbación y cómo afectan éstos al rendimiento de la RTO al Reactor de Otto Williams, se añadieron perturbaciones de dos formas: Una en escalón con una tendencia definida y tasa de cambio constante; y otra generada con una función ARIMA, la cual permite tener un acercamiento más real a las perturbaciones al agregar autocorrelación y aleatoriedad a estas. Las estimaciones realizadas son utilizadas para mejorar el cálculo de los gradientes del proceso mediante la definición de derivadas direccionales y el valor obtenido de la perturbación es utilizado en el modelo.
The Real-Time Optimization (RTO) is a systematic and iterative methodology that seeks the optimal conditions of a process, despite the presence of uncertainties in the model that describes it. Since the disturbances can affect the performance of this methodology, due to the change that these generate in the system conditions, it is necessary to include information in the RTO.The main objective of this work is to study and implement perturbation estimation methodologies to incorporate them in the Real-Time Optimization with Modifier Adaptation (MA). For this purpose, the Otto Williams Reactor model will be used as an experimental system and, as a disturbance, the feed flow of one of its components will be changed, assuming that this change can not be measured. The RTO with MA allows reaching the optimum of the process, even in the presence of structural uncertainty, since it occupies an automatic learning system based on the calculation of the process gradients, using past measurement from the system. Because of this methodology is based on past information, the presence of disturbances directly affects the estimation of the process gradients, since, when generating a change in the conditions of the system, it is necessary to evaluate which part of the change observed in the dependent variables is a result of the disturbances, and which one is explained by changes in the decision variables. Since disturbances are not always measured, this could compromise the effectiveness of the modifiers in reaching the optimum of the process. For the study of perturbation estimators, and how they affect the performance of the RTO to the Otto Williams Reactor, disturbances were added in two ways: One in step with a definite trend and constant rate of change; and another generated with an ARIMA function, which allows a more realistic approach to disturbances, by adding autocorrelation and randomness. The estimations made are used to improve the calculation of process gradients by defining directional derivatives and the value obtained from the disturbance is used in the model.
The Real-Time Optimization (RTO) is a systematic and iterative methodology that seeks the optimal conditions of a process, despite the presence of uncertainties in the model that describes it. Since the disturbances can affect the performance of this methodology, due to the change that these generate in the system conditions, it is necessary to include information in the RTO.The main objective of this work is to study and implement perturbation estimation methodologies to incorporate them in the Real-Time Optimization with Modifier Adaptation (MA). For this purpose, the Otto Williams Reactor model will be used as an experimental system and, as a disturbance, the feed flow of one of its components will be changed, assuming that this change can not be measured. The RTO with MA allows reaching the optimum of the process, even in the presence of structural uncertainty, since it occupies an automatic learning system based on the calculation of the process gradients, using past measurement from the system. Because of this methodology is based on past information, the presence of disturbances directly affects the estimation of the process gradients, since, when generating a change in the conditions of the system, it is necessary to evaluate which part of the change observed in the dependent variables is a result of the disturbances, and which one is explained by changes in the decision variables. Since disturbances are not always measured, this could compromise the effectiveness of the modifiers in reaching the optimum of the process. For the study of perturbation estimators, and how they affect the performance of the RTO to the Otto Williams Reactor, disturbances were added in two ways: One in step with a definite trend and constant rate of change; and another generated with an ARIMA function, which allows a more realistic approach to disturbances, by adding autocorrelation and randomness. The estimations made are used to improve the calculation of process gradients by defining directional derivatives and the value obtained from the disturbance is used in the model.
Description
Catalogado desde la version PDF de la tesis.
Keywords
INCERTIDUMBRE , OPTIMIZACION DE PROCESOS , OPTIMIZACION EN TIEMPO REAL
