Improving Renewables and Demand forecasting with ML-driven weather models
Abstract
GraphCast and GenCast are state-of-the-art AI models able to make medium-range weather
forecasts with unprecedented accuracy. They can be customized to predict arbitrary labels or
objective functions and (in the case of GenCast) can produce well-calibrated probabilistic
forecasts (forecasts ensembles of trajectories for hundreds of weather variables, up to 15 days
at 1 degree resolution globally). Google Research and Google DeepMind are currently
exploring their use for applications in the energy grid, such as wind and solar supply, and energy
demand. We are particularly interested in predicting distributions of outcomes, such as the
probability of an energy production shortfall or unexpected rise in demand.
Carl Elkin will discuss early-stage Research into how these models can be used to improve
reliability of the electric grid, including progress and current challenges.
forecasts with unprecedented accuracy. They can be customized to predict arbitrary labels or
objective functions and (in the case of GenCast) can produce well-calibrated probabilistic
forecasts (forecasts ensembles of trajectories for hundreds of weather variables, up to 15 days
at 1 degree resolution globally). Google Research and Google DeepMind are currently
exploring their use for applications in the energy grid, such as wind and solar supply, and energy
demand. We are particularly interested in predicting distributions of outcomes, such as the
probability of an energy production shortfall or unexpected rise in demand.
Carl Elkin will discuss early-stage Research into how these models can be used to improve
reliability of the electric grid, including progress and current challenges.
