GridForge
Projects
Student contributions organised by cohort year. Each year's work builds on the last.
Four groups running in parallel. Two are implementing and extending the platform; one is a masters research group exploring new foundational methods for topology comparison and evaluation; a fourth joined in second semester to work on optimal battery placement.
Digital Twin Infrastructure
Implementing established research methods for topology estimation and phase grouping within the GridForge platform, and refining those published methods where they don't hold up on realistic LV data. Alongside the research integration, this group is upgrading the platform's infrastructure: a containerised (Docker) locally-hosted deployment.
AI Topology Evaluation
Developing a machine learning approach to evaluate the quality of topology estimates produced by different estimation methods, using AMI data as the signal. When multiple topology estimation methods are in use, this work provides a principled way to assess which estimated topology most likely reflects the true network structure, without needing ground-truth labels for training.
Topology Comparison & Metrics
New research into how topology estimates should be compared and evaluated. The central question: what makes an estimated topology "close to" or "far from" the ground truth? The group is developing formal similarity metrics, and using them as an objective rather than just a score: a candidate tree is refined by proposing small mutations, scoring them, and keeping the ones that improve it. That points towards simpler, more computationally tractable ways of estimating topology.
Optimal Battery Placement
Determining where community-scale batteries should sit on an LV feeder to get the most value out of them. The problem is well studied when the network model is known; the interesting case here is when it isn't. Without a ground-truth topological model, placement has to be decided from estimated or uncertain network structure, which raises the question of how sensitive an "optimal" placement really is to the topology it was derived from.
The first GridForge cohort. One student focused on pilot-building the platform from scratch: establishing the Streamlit-based GUI, connecting Python analysis components, and writing backend code for synthetic data generation and simulation.
GridForge Prototyping
Built the initial GridForge platform from the ground up: a Streamlit-based frontend wiring together a set of Python analysis components, a backend for generating synthetic LV network data, and simulation infrastructure using OpenDSS. Established the project's architecture and development patterns for future cohorts to build on.