Smart meter data from 13,000 UK households is being used to train AI models that will generate entirely synthetic energy-use datasets, stripping out any personal information while preserving the statistical patterns researchers need. The problem is straightforward: the UK’s push to net zero depends on understanding how people use energy at home, but privacy laws block access to the real smart meter data that could inform better policies, reduce fuel poverty, and improve building performance. Without this data, research into time-of-use tariffs, grid planning, and clean energy investment stalls. Over six months, the team will produce “Synthetic-SERL”—the first dual-fuel synthetic dataset covering half-hourly gas and electricity use for 13,000 virtual households across full calendar years, complete with building, occupant, and weather data. The dataset, training code, and integration tools will be released under open licences. If successful, this removes a major bottleneck. Academics can run previously restricted research. Industry can test business cases for grid planning. Government can evaluate the distributional impacts of net zero policies. The project does not itself decarbonise anything—it makes the data that others need to do so freely available.
View original technical description
The Challenge The UK's journey to net zero depends on understanding how people use energy in their homes, but accessing the data needed for this research is restricted. Smart meter data from households provides crucial insights for developing better energy policies, reducing fuel poverty, and improving building performance. However, privacy regulations and data protection rules create major barriers to accessing this information, slowing down vital research that could accelerate our transition to clean energy. Our Solution We will solve this data access problem using cutting-edge artificial intelligence. Our project will train advanced AI models called Generative Pretrained Transformers (GPTs) on the UK’s world-leading SERL Observatory dataset - a unique collection of smart meter data from 13,000 representative GB households, combined with detailed information about their buildings, occupants, and energy use patterns collected over five years. The AI models will learn the complex patterns in real energy data and generate completely synthetic datasets that look like real household energy data but contain no actual personal information. This synthetic data will have all the statistical properties researchers need while being completely privacy-preserving. This project directly advances EPSRC AI for Science objectives: developing AI capabilities across research fields to accelerate scientific discovery; increasing access to well-governed, high-quality datasets for AI; building interdisciplinary collaborations between AI and energy researchers; and embedding AI as a research tool in a fair and inclusive way. What We Will Deliver Over six months, we will create "Synthetic-SERL", the first dual-fuel synthetic smart meter dataset with long temporal sequences. This will include half-hourly gas and electricity data for 13,000 virtual households across full calendar years, each with contextual information about building and occupant characteristics and weather. We will rigorously test this synthetic data to ensure it provides genuine research utility while passing strict privacy audits. The entire dataset, along with the training code and tools to integrate the data into workflows, will be published under open licences, making it freely available to researchers worldwide. Impact and Applications This research will deliver targeted benefits across three key sectors. Academia will benefit from removal of barriers to accessing high-quality data, enabling and accelerating R&D that was previously restricted by privacy regulations. Industry will gain access to data for testing business cases and grid planning that rely on high-resolution energy data, supporting investment decisions in clean energy technologies. Government will have data to evaluate distributional impacts of net zero policies like time-of-use tariffs, ensuring a fair transition for all households. By democratising access to high-quality energy data, we will unlock research that was previously restricted, accelerate innovation in the energy sector, and create new partnerships between AI researchers and energy experts. This project establishes the foundation for future research that will push the boundaries of AI for energy decarbonisation.
Plain English summaries and category classifications on this site are generated by AI and may not perfectly reflect the original research.
Is something wrong? Let us know