Abstract

Buildings are evolving from passive consumers into grid-interactive agents capable of providing demand-side flexibility, a cornerstone for achieving renewable integration and system reliability. Delivering such building-to-grid services requires supervisory control frameworks that combine predictive algorithms with interoperable, machine-readable infrastructures. Yet, current automation systems remain fragmented across proprietary platforms, while most digital twin applications focus on monitoring and visualization rather than predictive control. In parallel, model-based predictive control (MPC) has proven effective in coordinating distributed energy resources (DERs) but is rarely integrated with semantic digital twins (SDTs), leaving a critical gap between semantic interoperability and optimal operational planning.
This thesis introduces a methodological framework for developing operational SDTs that embed ontological reasoning directly into supervisory MPC. The framework progresses from data acquisition and entity structuring to layered semantic modeling, knowledge graph generation and validation, and integration with predictive control. A central contribution is the Resistance-Capacitance Ontology (RCOnt), which semantically formalizes grey-box thermal RC networks for control-oriented applications. The SDT framework integrates Brick Schema, ASHRAE Standard 223P, Energy Flexibility Ontology, and RCOnt, binding semantic entities to live telemetry streams and enabling query-driven discovery, model retrieval, and automated supervisory and local equipment control configuration.
The methodology is applied and validated in Hydro-Québec’s Experimental Houses for Building Energetics, two identical, instrumented residential testbeds relying solely on electricity as the energy carrier, equipped with rooftop photovoltaics, battery storage, and smart thermostats. Economic MPC for tariff-responsive load shifting achieves 35–75% reductions in electricity costs while maintaining indoor air temperature predominantly within ±1 °C of the reference setpoint. Heuristic MPC for post-outage cold load pickup mitigation reduces rebound demand by approximately 50%, with only a modest extension of thermal recovery time. Replication across houses demonstrates the transferability of the architecture, indicating the potential for scalability, as semantic abstraction enables reproducible, vendor-agnostic supervisory control for residential buildings with DERs.
This thesis establishes SDTs as a practical foundation for scalable grid service orchestration by integrating semantic modeling, physics-based representations, and predictive control within a single operational workflow. The results demonstrate that embedding semantics into supervisory MPC not only enhances interoperability and automation but also provides a transferable pathway for enabling real-time flexibility services in future grid-interactive buildings.