Combining Learning-Based Cop Proxies and Formal Optimization for Residential Heat Pump System Operation

Residential heat pump systems are key in achieving climate policy objectives in the building sector. Due to the variability in technical installation criteria, manufacturer requirements, local regulations and the spatial or geographical restrictions associated with buildings, these heating systems are highly heterogeneous. The operating conditions measured under standardized settings do not always align with those observed in real-world systems. This poses challenges for operational planning tools that aim to coordinate the systems' power consumption and increasingly exploit their flexibility when interacting with markets or system operators' congestion management mechanisms. We propose a predict-then-optimize approach that combines learning-based COP proxies with modelbased optimization methods to use real-world operational data when making prescriptive scheduling decisions.