Octopus’s Faraday and OpenSync push synthetic smart‑meter data for research

Gus Chadney of Octopus Energy’s Centre for Net Zero presented Faraday, a generative‑AI pipeline (autoencoder + Gaussian mixture on latent space) that produces synthetic household smart‑meter time series at 30‑minute granularity and conditions generation on household features such as technology ownership and tariff type. Chadney said Faraday was trained on 1.8 billion rows of Octopus customer data and that the synthetic outputs are designed to be non‑identifiable while preserving statistical fidelity for use in forecasting, distribution modeling and product development.

Octopus has contributed synthetic datasets and algorithms to an open project, OpenSync, intended as a shared repository where utilities and researchers can generate or share synthetic datasets to support benchmarking and model development. Chadney emphasized evaluation across three axes — fidelity (statistical match to real data), utility (usefulness for forecasting/classification), and privacy (resistance to re‑identification) — and described validation work with universities and system operators in the UK and The Netherlands.

Why it matters: Access to high‑granularity demand data is restricted by privacy rules and commercial concerns. Synthetic datasets can let researchers and planners run bottom‑up simulations, test tariffs and design network studies without exposing customer identities. Octopus argues that a community approach (OpenSync) will broaden the set of regional datasets and models available for research and planning.

Next steps: Octopus plans geographic expansions (including U.S. markets where they operate) and enhancements to the Faraday model (location awareness, scenario conditioning); OpenSync aims to curate multiple synthetic datasets and tools on a common portal.