能源领域合成数据的物理信息生成 AI 博士职位

Can you help unlock the data needed for the energy transition? In the NWO-funded SHARE project, you will develop AI models that generate realistic, privacy-preserving synthetic energy data for grid planning and decision-making. Working with real-world data from Alliander, you will publish at leading machine learning venues while building tools with tangible impact on the Dutch energy sector.

The Dutch energy transition depends on data that almost no one is allowed to see. Distribution system operators (DSOs), municipalities and energy communities need high-resolution grid and consumption data to plan grid reinforcements, heat networks and local flexibility, but privacy law (GDPR), commercial sensitivity and regulatory uncertainty keep this data locked away. Hence, critical infrastructure decisions are being made with incomplete information.

Synthetic data offers a way out: realistic-but-artificial datasets that preserve the statistical, temporal and physical structure of real energy data without identifying real households or companies. But energy data is not like images or text: it consists of time series living on a physical network, governed by power-flow equations. Off-the-shelf generative models produce data that looks plausible but violates physics and is therefore of limited use for grid planning.

As a PhD candidate you will develop physics-informed, domain-constrained generative models for energy-system data, the core scientific contribution of the SHARE project (Work Package 3). More concretely, your work will involve the following: • You will design and compare deep generative approaches, UAEs, GANs, diffusion models/flow matching, and Gaussian processes for realistic load, generation and voltage time series. • You will embed physical constraints into generation: power-flow consistency (Kirchhoff's laws) as soft or hard constraints, operational bounds (voltage limits, capacity, phase balance), and network topology through graph neural network architectures. • You will build validated benchmark datasets and an evaluation framework covering statistical fidelity, temporal and spatial structure, physical plausibility and downstream task performance (e.g. train-synthetic-test-real forecasting). • You will collaborate with a fellow PhD candidate and a postdoctoral researcher on integrating differential privacy into the generative pipeline, balancing privacy guarantees against data utility. • You will contribute to an open-source synthetic data toolbox that DSOs, municipalities and researchers across the Netherlands will actually use.

This is research with a direct route to impact: you will work with real operational data from Alliander, with regular on-site visits and direct access to the practitioners who will use your models for congestion forecasting, spatial energy planning and flexibility assessment. You will publish at top machine learning venues while producing open datasets and tools with tangible societal impact.

You will be expected to spend a small part of your time (up to 10%) on teaching activities, such as assisting in courses of our computing science programmes.

Would you like to learn more about what it’s like to pursue a PhD at Radboud University? Visit the page about working as a PhD candidate .