Onderzoek naar optimaal datagedreven voorspellingsmodellen voor de activatie van energieflexibiliteit in woningen
Year: 2024
Type: dissertation
Abstract: The integration of Renewable Energy Sources (RES) into the energy mix offers promise for reducing greenhouse gas emissions by 2050. The building sector plays a crucial role to achieve this objective. However, the intermittent nature of RES poses challenges in maintaining the demand-supply balance of energy. Demand Response (DR) programs have been proposed to address this imbalance, focusing on shifting electricity use within buildings. To successfully participate in DR programs, buildings are required to provide energy flexibility to the grid. However, to exploit the inherent energy flexibility potential of buildings building energy management systems should be upgraded to integrate advanced control systems like Model Predictive Control (MPC). The key to harness energy flexibility through MPC lies in developing a dynamic predictive model of the building, a process that serves as an important challenge in implementing MPC. Typically, these models have a data-driven nature, and the collection, logging, and pre-processing of such datasets constitute a time-consuming and costly step in the model development pipeline. This thesis investigates some of the challenges and solutions in deploying MPC for energy flexibility activation in buildings. It develops a framework to evaluate the impact of different dataset features and modelling techniques on performance of predictive models within MPC. The study finds that considering multi-step ahead prediction error is crucial to achieve high flexibility levels through MPC. Thereafter, the impact of sampling rate on the performance of predictive models for harnessing flexibility potential is investigated. It is shown that choosing an inappropriate sampling rate causes MPC to fail to tap into 50% of potential flexibility of a building. Next, the impact of dataset size and excitation method on the predictive model used in flexibility activation context is investigated. It is shown that the suitable choice of excitation technique depends on the modelling technique and even with normal operational data of a building, MPC can successfully harness potential flexibility of a building. Regarding dataset size, it is demonstrated that certain modelling techniques exhibit greater sensitivity to dataset size, concerning the final flexibility achieved through predictive models constructed using those modelling techniques. This study also investigates the importance of dataset's level of detail, showing that information regarding some specific variables is essential to have a high quality MPC. On the other hand, it demonstrates that a higher level of detail in all variables does not necessarily improve MPC's performance. Overall, the findings underscore the importance of appropriate modelling techniques and carefully choosing dataset features which are to be used to generate predictive models of an MPC with the aim of maximizing energy flexibility in buildings. These findings contribute to advancements in sustainability of building sector.
Source: Lirias
Author: Erfani Beyzaee, Arash
Institution: KU Leuven
Language: English
Subfield: Building and Construction
Field: Engineering
Domain: Physical Sciences
Keywords: building energy flexibility, model predictive control, data-driven predictive models, demand response, building energy management systems, renewable energy integration, multi-step forecasting, sampling frequency, excitation methods, dataset size, residential buildings, building energy simulation
Open Access status: green
OpenAlex ID: W7051676184
