Estudio de simulación sobre métodos de identificación de parámetros multilineales para un sistema de calefacción de un edificio industrial

Abstract

[ES] La identificación de parámetros está en el núcleo de todas las técnicas modernas de control basadas en modelos. La tesis respalda el objetivo del proyecto SIOSTA: control óptimo basado en modelos de los sistemas de calefacción para un edificio industrial en Hamburgo, Alemania. El objetivo de esta tesis es una comparación de diferentes métodos de identificación para modelos de espacio de estado utilizando datos de entrada-salida. Los enfoques lineales estándar construyen la línea base y se investigan métodos multilineales modernos y los resultados de ambos se validan, comparan y se realizan investigaciones sobre los efectos del ruido. Desafortunadamente, debido al brote de COVID-19, no es posible acceder a mediciones reales desde el edificio. Por ello, los datos artificiales son generados por MATLAB Simulink a partir de modelos multilineales con propiedades de ruido adicionales.

[EN] Heating systems are responsible for a large part of the total energy consumption in Germany. Currently, almost 40% of the energy consumed has as its final destination the building sector, in which non-residential buildings play an important role. This consumed energy is mostly intended for regulating the room temperatures of the building. The solution to this high consumption ratio includes high efficiency standards for new buildings but also optimization of energy consumption systems in the already existing buildings. The utilisation of noneffective control systems causes a misuse of energy. However, a 30% of the energy can be saved by implementing effective control methods. The thesis supports the goal of the SIOSTA project which is the implementation of optimal model-based control of the heating systems for an industrial building in Hamburg. As parameter identification is in the core of all modern model-based control techniques, the focus of this thesis is a comparison of different identification methods for state space models using input-output data. Standard linear approaches build the baseline and modern multilinear methods are investigated, and the results of both are validated and compared. Unfortunately, due to the outbreak of COVID-19, it is not possible to access to real measurements from the building. Instead, artificial data is generated by MATLAB Simulink from multilinear models.