Available Projects

Artificial intelligence (AI)-based technologies are being adopted across various sectors at an unprecedented scale. However, the computing resources required to run these algorithms, i.e., the AI data centers, are extremely power hungry, thus significantly increasing the burden on the electrical grid. The proposed PhD project will develop analytical methods to identify the threats from AI data centers to power grids and develop solutions to mitigate the risks.

Primary supervisor: Dr Subhash Lakshminarayana - Subhash.Lakshminarayana@warwick.ac.uk

Project detail:
Artificial intelligence (AI)-based technologies are being adopted across various sectors at an unprecedented scale. However, the computing resources required to run these algorithms, i.e., the AI data centres, are extremely power hungry, thus significantly increasing the burden on the electrical grid. More importantly, AI data centres have unique load patterns which can threaten the power grid stability, such as quick ramp-up/ramp down in power consumption. Cyclic demand consumption patterns (such as powering up and down repeatedly due to AI training cycles) can lead to unforeseen load patterns, potentially threatening the grid stability.

The proposed PhD project will develop analytical methods to identify the threats from AI data centers to power grids and develop solutions to mitigate the risks. To this end, we will adopt a rare-event sampling approach to uncover network-threatening load attack patterns. The rationale is that the most impactful behaviours, such as those that threaten system stability or safety, often occur with extremely low probability and thus are poorly captured by standard random sampling. For instance, power grids are designed to be highly reliable and operate across a wide range of scenarios. Yet, grid operators today are increasingly witnessing rare but disruptive events that threaten network stability. Most existing studies in this area are often based on engineering intuitions or heuristic methods such as Monte Carlo sampling of the load patterns. These approaches are not scalable to large, realistic system settings representative of real-world power grids [1]. There remains a lack of analysis into how rare-event sampling schemes should be systematically designed and tuned for domain-specific objectives in power systems. Addressing this gap requires moving beyond heuristic application towards a more theoretically grounded understanding of sampling and performance in complex physical systems. Working with our industrial partner, National Grid, the project will ensure the real-world applicability of the proposed analysis and mitigation solutions. This will be achieved using the following tasks.

How to apply for admission:www.warwick.ac.uk/pgrengineering