Deep Learning Research

Model Performance & Research

WindAI's neural network was trained on 10M+ hourly observations from 300+ operational wind farms. Below are the headline metrics, the architecture, and the methodology paper.

Deep Neural Network

Architecture

400+ Input Features

Feature Space

300+ Training Farms

Training Data

Independently Validated

Plant-level results

View validation

Research Paper

“WindAI: A Deep Learning Approach to Global Wind Resource Assessment Using Multi-Source Reanalysis Data”

Key Findings

Trained on 10M+ hourly observations from 300+ wind farms across 8 countries
400+ features combining ERA5, MERRA2, boundary layer height, and elevation data
Independently validated on US wind farms not seen during training — see /validation
Global coverage — predict for any latitude/longitude on Earth

Ready to try it yourself?

Get AI-powered wind resource assessments for any location on Earth.