Faraday & OpenSynth: Open access, AI-generated data for the grid

Charlotte Avery – Centre for Net Zero

The transition to a decarbonised, decentralised, and flexible power system requires modelling that accurately captures the complexity of consumer demand across diverse geographies with varied technology adoption rates. Progress is currently limited by data access: strict privacy regulations inhibit sharing consumer data, and anonymised, aggregated datasets prevent the development of truly robust models that account for the relevant diversity in consumer demand.

This talk will demonstrate how high-resolution, open-access AI-generated synthetic data can be a solution – democratising data whilst ensuring privacy – accelerating the breakthroughs in intelligent power system applications necessary for the smart systems of the future.

Key Highlights:
– Faraday V5: We present a custom deep generative model designed to synthesise open-access, half-hourly electricity demand profiles, trained on real data from approximately one million households. We detail how Faraday V5 moves beyond national averages to provide locational awareness, outputting demand profiles unique to specific regions.
– Modelling the “Hard-to-Sample”: We demonstrate how Faraday V5 is conditioned on specific household characteristics and low-carbon technology adoption allowing the generation of high-fidelity profiles that are representative of future electrification scenarios rather than historical norms.
– OpenSynth: We present recent updates in our global community initiative, co-founded with LF Energy, to standardise and scale AI-generated and AI-ready grid data. OpenSynth is a data repository; space for algorithm development; and peer-led community ensuring that synthetic data is validated, reproducible, and ready for use in training AI models.