Event Model
EVENTModels the outages ahead — planned and unplanned — on a physical model of the grid.
Where the foundation model reads history, the Event Model reasons over the physical network — given a power-flow model of the grid, it anticipates where stress and outages are likely to emerge.
How it works
A graph neural network over the network's physical topology — buses as nodes, branches as edges — coupled to power-flow solutions rather than trained on time series alone. It ingests a power-flow model of the grid (bus and branch parameters, thermal and voltage limits, injections) and message-passes stress signals along the same connectivity power actually flows across, so it learns from historical outage and loading records where thermal, voltage, and contingency stress concentrates on the real network, not just which node happened to fail before. Planned outages are modeled as known injections about to be removed; unplanned risk is a learned hazard over conditions. A power-flow model is a required input, not optional context.
Inputs
- Power-flow model of the grid — REQUIRED (bus / branch parameters, limits)
- Network topology
- Historical outage records — planned and forced
- Loading and weather conditions
Outputs
- Spatially-resolved grid-stress likelihood — "expect stress here" — by location and time
- Outage-likelihood over the network, planned and unplanned
Three ways to consume it
product · inference · MCPIn the product
The Event Model powers Ask the Grid directly — the live map, the radar, and the grid agent read it in place, so the same model you'd call is the one already reasoning over the grid on screen.
Developer inference service
A direct, typed inference API: request forecasts or embeddings by node and horizon and get structured results back. This is a plain inference endpoint — not MCP — for pipelines, notebooks, and services that want the model output without an agent in the loop.
See the developer surfaceMCP
The same model outputs, exposed as tools your own agent can call. Point a compatible client at the Ask the Grid MCP server and the model becomes a tool in your agent's own loop.
Build with MCPIntended use
- Operators anticipating where the grid will be stressed
- Traders anticipating outage-driven price moves
- Asset teams planning around maintenance windows
Limitations
- Requires a power-flow model of the grid as input
- The earliest-stage model in the preview — the physical models follow the forecasting ones
- Outputs are likelihoods, not deterministic outage schedules