In the evolving landscape of energy-efficient building technologies, digital twins are emerging as powerful tools to simulate, control, and optimize complex systems. One of the recent initiatives taken forward by Work Package 3: Heat Sub Digital Twin, focused on building a digital twin of a heat pump system integrated with Thermal Energy Storage (TES) — a project that blends physics-based modeling with advanced machine learning for smarter energy management.
Building the Digital Twin
The work package created a physics-based model of the heat pump and TES system. This model represents the real-world thermal and dynamic behaviours of the system with high fidelity. The goal was to capture critical dynamics like heat transfer, storage efficiency, and heat pump performance, enabling accurate virtual testing and scenario analysis.
Exploring Model Adjustment with Limited Data
Since we didn’t have access to a real system, we were not able to fully adjust the model using real-time measurements. To move forward, we used a combination of historical and simulated data to explore how AI could support model adjustment. While this stage didn’t solve the data challenge entirely, it helped us test different approaches and prepare for the optimization phase.
Machine Learning Meets Control Strategy
With a well-adjusted digital twin in place, we shifted focus to optimisation. Leveraging advanced ML techniques, we explored optimal control strategies that minimize energy consumption and improve system responsiveness.
From Simulation to Real-Time Control Capability
The final step focused on enabling real-time integration by developing a control block designed to receive live data and generate actionable outputs for physical system interaction.
