Naar het karakteriseren van energieflexibiliteit in residentiële districten

The transition toward decarbonized and flexible energy systems requires buildings, as major energy consumers, to become active participants in demand response (DR). This dissertation investigates how building stock can be characterized, controlled, and clustered to provide energy flexibility at district scale under realistic operational constraints. Its central aim is to bridge the gap between high-fidelity building modeling and practical demand response applications by introducing scalable frameworks for emulator generation, systematic flexibility characterization, and flexibility-driven clustering. The work contributes to both the methodological foundations of control-oriented urban building energy modeling (UBEM) and to the operational integration of building flexibility in energy systems. The thesis begins with the development of UCEGe (Urban Control-oriented Emulator Generator), a scalable framework for generating modular, interoperable, and computationally efficient urban building emulators. Implemented in Modelica/Dymola using the IDEAS library and linked through the Functional Mock-up Interface (FMI), UCEGe integrates building thermal models, heating systems, and controllers. It supports high temporal resolution and real-time co-simulation, enabling large-scale district emulation that is suitable for predictive control, flexibility analysis, and digital twin applications. UCEGe thus provides the foundational modeling infrastructure for subsequent analyses. Building upon this foundation, the thesis examines how price signal characteristics (duration, amplitude, and shape) and building features (Heat Loss Coefficient, Thermal Capacity, Time Constant, and heating system type) determine the performance of flexibility under different DR regimes. A district of 480 residential buildings, each equipped with price-aware Model Predictive Controllers (MPCs), was subjected to seven price signals: two sets of impulse signals (2 h and 6 h, each with three amplitudes) and a 24 h step signal. A comprehensive framework of twelve Key Performance Indicators (KPIs) was introduced to quantify flexibility in terms of load shifting (SEI, SED, SER), peak shaving (PP, DP, PSE), ramping behavior (EIFF, EDFF, MRU, MRD), and timing (PPIT, DPIT). Results showed that signal characteristics decisively shape flexibility outcomes, with long-duration signals enhancing shifting potential but reducing efficiency under saturation, while shorter, moderate-amplitude signals provide efficient and predictable responses. Feature-driven analysis demonstrated that building properties strongly condition flexibility, with distinct trade-offs across time scales and control regimes. Recognizing the limitations of quantile-based classification, the thesis introduced an unsupervised clustering framework that groups buildings directly by their flexibility response profiles rather than by predefined property categories. Using normalized flexible power (NFP), Dynamic Time Warping (DTW) distance, and Ward's linkage, the study revealed distinct flexibility archetypes emerging from flexibility-driven grouping. However, replicating the framework with different signal types showed that cluster memberships can vary significantly across different DR signals, highlighting the signal dependence of clustering outcomes. To overcome this, the thesis developed a consensus clustering methodology that integrates multiple DR regimes to derive stable, signal-agnostic clusters. By aggregating clustering results across signals and evaluating robustness through CDF analysis, silhouette scores, and variance of flexible power (FP), the method identified building groups that consistently exhibit similar flexibility behavior regardless of the signal. These consensus clusters are interpretable in terms of building properties, and they offer a robust tool for DR program design, and enhanced forecasting models, enabling aggregators and system operators to characterize and reliably mobilize building flexibility. The thesis concludes by outlining its contributions: (i) the creation of UCEGe as a scalable control-oriented UBEM framework; (ii) the development of a comprehensive KPI-based methodology for flexibility characterization; (iii) systematic feature-driven analysis of district flexibility under short-, mid-, and long-term DR regimes; (iv) the introduction of unsupervised clustering for emergent flexibility archetypes; and (v) the advancement of consensus clustering as a robust and transferable methodology for flexibility-oriented aggregation. Limitations include the reliance on white-box modeling, the use of baselines for flexibility characterization, the exclusive assumption of MPC-equipped buildings, the computational intensity of high-fidelity simulations, and presuming ideal conditions of having a market mechanism for the buildings flexibility exploitation. Future work should address these by validating consensus clustering in field trials, extending the framework to heterogeneous control environments, improving data acquisition infrastructure, integrating uncertainty, and scaling to diverse building stocks and regulatory contexts. In sum, this dissertation advances the methodological and operational understanding of urban building flexibility, offering a pipeline that links modeling, control, characterization, and clustering. By doing so, it provides scientific and practical tools to enable the integration of buildings as active, reliable, and scalable participants in future smart energy systems.