Project

Exploring Graph foundation model for power utility usecases

Objective

Graph Foundation Models (GFMs) are designed to work with graph data (networks of interconnected entities). They are pre-trained on massive graph datasets to learn the inherent structure and patterns within networks. It is excel at network analysis, ideal for the interconnected electric grid. They can predict outages, optimize power flow, integrate renewables, and manage assets better. But the transferability is the key issue. A GFM trained on one grid might not work well on another due to grid differences (size, structure) and data variations. To improve GFM transferability across different electricity grids, a key area of development is creating GFM tokenizers for graph data normalization. This would enable GFMs to learn transferable knowledge and be more widely applicable.

Outcome

The internship will involve researching existing GFM applications and then developing a tokenizer that adapts GFMs for a chosen downstream task in the electric power grid industry. papers to read - https://arxiv.org/pdf/2402.02216 https://arxiv.org/pdf/2403.01121 https://arxiv.org/pdf/2101.10025 https://openreview.net/pdf?id=ntSP0bzr8Y https://www.osti.gov/servlets/purl/1508032