PhD position ‘Adaptive Privacy and Differential Privacy Energy Data’

Can you help make energy data usable without giving away sensitive information?

The energy transition in the Netherlands urgently requires upgrading of the electricity grid, for instance to accommodate solar panels, wind turbines, and electric vehicles. To efficiently plan and manage these upgrades, stakeholders such as grid operators, municipalities and energy communities need to share detailed data about electricity consumption, generation and grid operation. This data sharing however is prevented by concerns on the competitively sensitive nature and privacy of data, as it reveals detailed information about households, buildings and operational assets.

Synthetic data generated by generative AI-models offers a promising alternative, but preserving statistical, temporal, and physical characteristics of real energy data while protecting sensitive information are major challenges. We are offering a PhD position in the SHARE project that addresses these challenges.

As a PhD candidate, you will develop methods for adaptive privacy protection of energy data
, combining differential privacy, empirical privacy attacks and privacy-utility optimisation. Your research is part of the Adaptive Privacy and Legal Compliance work package in the SHARE project. More specifically, your challenges are to:

• Investigate how temporal patterns, spatial information, rare events and other distinctive features contribute to re-identification and inference risk. You will work with metrics such as uniqueness, entropy, attack success rates and time-series-specific indicators.
• Implement and study attacks such as membership inference, reconstruction, re-identification and attribute inference to understand where information leakage occurs in both original and synthetic datasets.
• Investigate methods that allocate stronger protection to sensitive components and lighter perturbation to lower-risk components, thereby preserving as much analytical utility as possible.
• Formulate privacy protection as an optimisation problem balancing formal privacy guarantees against statistical, temporal and physical utility requirements for energy-system applications.
• Collaborate closely with researchers at Radboud University, who develop physics-informed generative models for synthetic energy data, and investigate approaches such as differentially private training and post-generation privacy calibration.
• Work closely with distribution system operator Alliander and other consortium partners to evaluate whether the developed methods provide meaningful protection while retaining the information required for practical energy-system applications.
• Ultimately contribute to an open-source privacy layer for the SHARE toolbox.

Requirements

You hold, or will soon obtain, an MSc degree in Computer Science, Cybersecurity, Artificial Intelligence, Data Science, Applied Mathematics, or a closely related discipline. You should have:

• a strong background in privacy-enhancing technologies, artificial intelligence, cybersecurity or statistical modelling;
• strong programming skills, particularly in Python;
• preferably, some familiarity with deep generative models (VAEs, GANs, diffusion models) and probabilistic modelling;
• good written and spoken English skills;
• knowledge of Dutch language preferred not mandatory.

Fixed-term contract : for 4 years.

The PhD candidate will be appointed for a period of 15 months. The appointment will be extended to 4 years when progress and performance are good. A PhD training program is part of the agreement.

The salary is determined in accordance with salary scale P of Appendix A of the Collective Labour Agreement of Dutch Universities…