Towards trustworthy residential load profile synthesis using a diversity driven diffusion model

Driven by the penetration of Low Carbon Technologies, power and energy systems are undergoing a digital transformation. This shift demands the building of robust distribution networks supported by intelligent systems to observe, estimate, control and manage network loadings to manage new dynamics. In this evolving landscape, customers’ load profiles are critical resources to support data analytics applications in modern power systems. However, there are usually insufficient historical load profiles for power system analysis, due to collection cost and data privacy issues. This presents an opportunity for the new range of high performant generative models which, while capable of learning the distribution of real data through traditional statistical methods, may fall short in maintaining electrical realism. Consequently, to establish trust in these deeply opaque AI systems, human-centered design as linked to domain-specific insights is key for robust and realistic modelling. To build trust in these models, we explore the impact of loss functions on a Diffusion model’s performance with the aim of producing diversity faithful domestic load profiles for a group of customers.