Conditional Diffusion Model For Probabilistic Disaggregation of Pv Systems and Heat Pumps

Behind-the-meter (BTM) distributed energy resources and flexible electrified loads, such as photovoltaic (PV) systems and heat pumps (HPs), are expanding rapidly, thereby increasing the complexity and uncertainty of the distribution grid. This paper proposes a probabilistic methodology to disaggregate BTM energy components based on a multivariate conditional diffusion model. Leveraging low-frequency (LF) smart meter data and conditioning signals such as irradiance and temperature, the model jointly reconstructs PV generation and HP consumption, distinguishing between domestic hot water and space heating. Evaluated on real residential data from the Netherlands, the methodology demonstrates strong deterministic and probabilistic performance, providing reliable uncertainty estimates across multiple time scales. The model exhibits robustness to reduced training sets, and the impact of seasonality on disaggregation performance is analyzed. By jointly disaggregating BTM PV generation and HP loads while quantifying uncertainty, the proposed approach provides distribution system operators with a practical tool to improve observability, planning, and flexibility management under LF metering constraints.