Adaptive heating curves using data-driven methods to improve the energy efficiency of buildings

Heat pumps are a key technology for electrifying and decarbonising heating in the building sector. Their efficiency is strongly coupled to the heat supply temperature: the lower the supply temperature, the higher the coefficient of performance. In practice, the supply temperature is governed by the heating curve, which relates the supply temperature setpoint to the outdoor air temperature. This heating curve is often poorly configured, as proper parametrisation requires testing and detailed knowledge of the specific building. This paper presents a method to optimise the heating curve automatically using data-driven techniques, which can be implemented as a cloud-based service, thereby eliminating the need for manual parametrisation by a technician. The method first generates diverse operational data through an automated system-identification procedure. A gaussian process regressor is then trained to predict the room temperature deviation. The regressor is finally used to derive a self-optimising heating curve. The approach is evaluated in a simulation study using a high-order, multi-zone model of a single-family home coupled to an air-source heat pump, across three climate locations and two heat transfer systems (radiators and underfloor heating). The results show that the method generates suitable heating curves for all cases, achieving thermal comfort and heat pump efficiency comparable or better to reference heating curves that were parametrised optimally with full system knowledge. A model predictive controller outperforms the heating curves in almost all cases, but requires higher development effort and online optimization.