SBIR Phase I: Reducing Numerical Weather Forecasting Computational Expense Using Machine Learning

The broader impact/commercial potential of this Small Business Innovation Research (SBIR) project is the development of artificial intelligence (AI) weather forecasting tools to improve forecast accuracy and strengthen power grid resiliency. The United States is currently transitioning to an increasingly renewable energy-based economy. The availability of the leading renewable sources, solar and wind, vary with weather, making forecasts critical to plan for daily plant and energy grid operations.

When weather forecasts are wrong, the grid suffers inefficiencies, causing power prices to spike and hurt consumers. Worldwide, national weather agencies continually release weather data to the public. This project demonstrates the feasibility of improving weather forecasting by collecting, organizing, and leveraging large public weather datasets in real-time.

AI will sort through these data to deliver advanced weather insights to the energy industry as well as other commercial consumers requiring high fidelity forecasts. These insights will allow energy market participants to anticipate market movements, correct inefficiencies, and lead to a more resilient energy grid and power distribution.

This SBIR Phase I project proposes to advance the state of weather forecasting by augmenting physics-based atmospheric models with artificial intelligence. The intellectual merit of this project is the development of a toolkit of algorithms that in real-time applies AI to detect errors in weather forecasts and uses machine learning accessing historical forecasts to provide error correction.

The project’s objectives are 1) development of a database of historical wind forecast data corresponding to major Texas wind farm sites; 2) training and validation of an ensemble of AI algorithms, e.g. artificial neural networks, random forests, and probabilistic analogs, that detect errors in historical forecast data; and 3) development of software that ingests publicly available weather forecasts from physics-based models, applies AI algorithms, and delivers advanced weather forecast insights in real-time. This research will provide feasibility that AI algorithms can easily integrate with high-fidelity models to complement the existing weather forecasting infrastructure.

The technology developed in this project is a general environmental forecasting framework that demonstrates the use of AI/ML applications to a broad set of weather-dependent scientific and engineering challenges.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.