Image: Cortland Johnson, PNNL
ChatGrid has been developed as a generative AI solution to simplify and speed up decision making for power grid operators in the US.
ChatGrid, the brainchild of Pacific Northwest National Laboratory (PNNL) optimisation and grid modelling researcher Shrirang Abhyankar, was conceived to exploit the question-and-answer approach of generative AI tools to support grid operators in their decision making.
With this, the user can then ask a question such as: “What is the generation capacity of the top five wind power generators in the Western Interconnection?” with the response of a visualisation showing the desired information.
Users can ask questions about generation capacity, voltage, power flow and more, while customising the visualization to show different information layers.
Have you read?
Tech talk | On the grid edge
Europe’s grid action plan: a step in the right direction, but is it enough?
Abhyankar says the aim is to simplify the experience for grid operators who have to make many decisions as they monitor the grid in real-time.
“We’re envisioning a new way to look at data through questions. ChatGrid allows someone to query the data – in a literal sense – and get an instantaneous answer.”
ChatGrid runs on a publicly available large language model, with the input data the synthesised data from the Exascale Grid Optimisation (ExaGO) model, which was developed to take advantage of the power of supercomputing to simulate the nation’s power grid in real time.
Moreover, to further protect grid security, the model was not trained on the data itself.
Instead, the data was compiled in an internal database in columns with headings such as ‘capacity’ and ‘location’ of power plants and the model used to produce a ‘structured query language’ (SQL) that allows ChatGrid to search this database for its answers.
Work in progress
For Abhyankar, ChatGrid remains very much a work in progress.
He hopes that once grid operators start using ChatGrid and providing feedback, a better version can be built that they can then safely use in their control rooms with real-life data.
For that to work, ExaGO’s developers need the data to be useful on regular computers as well, however.
Further, while ChatGrid is available for download on GitHub, the process takes a few steps and once the feedback starts rolling in, the development of a one-step download process is anticipated.
Users also are invited to play around with the phrasing prompts and questions to assess how to produce the best answers.
