Over the past few months, the tool behind EOSC-SIESTA’s third use case, the Energy Domain, has taken a significant step forward. Led by Predictia Intelligent Data Solutions, the use case set out to build a user-friendly service for forecasting energy generation and consumption in solar installations, which can in turn be used to optimize battery usage. That service is now live on the project dashboard, and it has grown beyond its original audience: what began as a tool for data rights holders now also serves the researchers who want to build on top of this data. The result is a single environment where sensitive energy data can be put to work and, at the same time, safely shared for science.
The use case is grounded in a real scenario, with models trained on data from solar installations in Spain. For any household or installation with solar generation and a battery, the same question comes up every day: when should you sell the energy you produce, when should you buy from the grid, and when should you store the surplus in your battery to keep your bill as low as possible? The tool answers that question in three moves. First, it predicts how much energy the installation will produce over the next 24 hours, combining weather data such as solar radiation with the installation’s own historical generation. Second, it predicts how much energy the household will consume, based on historical consumption data, which is the most sensitive piece of the puzzle because consumption patterns directly reflect the daily behaviour of the people living in the home.
Finally, with both forecasts in hand and by monitoring the market electricity price, these predictions provide the basis for deciding moment by moment whether to store, sell or draw energy.
Two roles in one dashboard
The most important change is who can use the tool, and how. On one side are the data rights holders, the people and organisations who actually hold generation and consumption records. They can now feed their own information directly into the tool, providing historical solar production data, historical consumption data and installation metadata such as latitude, longitude and other characteristics of the installation. In return, the tool sends back a solar production forecast and a consumption forecast for the next 24 hours. Beyond the immediate benefit to the owner, every upload enriches the pool of data the service works with, making the use case more representative and more valuable over time.
On the other side, and this is the newly opened part, are the researchers. From their own dedicated interface, distinct from the data owner’s, they can access and work with the energy data to carry out their scientific studies. Crucially, they never see the raw, sensitive records. What they receive, served through a dedicated anonymised-database API, is an anonymised dataset that is realistic enough to experiment with and train models on, but stripped of the information that could expose the individuals behind it. Each role corresponds to a different application, and both are available on the SIESTA dashboard.
A step-by-step tutorial walks new users through each interface. In short, data owners feed the system with real data and get practical value back, while researchers get usable, privacy-preserving data to advance their work.
Bridging those two roles safely is only possible because of the anonymisation layer SIESTA brings to the table. To generate the shared dataset, the use case relies on TrasgoDP, one of the project’s anonymisation tools, developed within SIESTA. TrasgoDP implements mechanisms for local differential privacy, and here it is applied to the location of the installations, which can itself be identifying: it adds carefully calibrated noise while offering formal privacy guarantees. This is what allows researchers to work with realistic energy data without ever compromising the households behind it, and is particularly useful when preparing data for publication or for training anonymised models, ensuring that no single installation, or the people connected to it, can be singled out.
Like the rest of EOSC-SIESTA, the Energy Domain components are developed openly and hosted on IFCA’s GitLab, and the use case is put together from four building blocks. A [data rights holder frontend] is the interface where data owners upload their production, consumption and installation data and receive their 24-hour forecasts, while a separate [researcher frontend] is where researchers explore and download the anonymised data for their studies. Behind them sit two services: a [solar forecasting API] that runs the deep-learning models and returns the production and consumption forecasts, and an [anonymised database API] that exposes the anonymised dataset to the researcher frontend.
Taken together, the Energy Domain use case is a compact illustration of what EOSC-SIESTA is about: making genuinely sensitive data reusable for research without sacrificing privacy. By combining accurate deep-learning forecasts, a trusted execution environment and differential-privacy tools such as TrasgoDP, the tool turns private solar and consumption data into two things at once, a concrete cost-saving service for the people who own the data and an open, anonymised resource for the scientific community that wants to build on it. With both sides now available on the dashboard, that promise moves from concept to something people can actually use.