Informing broad-scale wind energy planning using machine learning to model spatially variable power density for the United States

Authors: Dylan Harrison-Atlas*, National Renewable Energy Laboratory, Galen Maclaurin , National Renewable Energy Laboratory, Eric Lantz, National Renewable Energy Laboratory
Topics: Sustainability Science, Spatial Analysis & Modeling, Quantitative Methods
Keywords: wind energy, machine learning, predictive modeling, renewable energy
Session Type: Paper
Presentation File: No File Uploaded

Emerging technologies are reshaping the energy landscape with an emphasis on securing sustainable, resilient and cost-effective solutions. Expansion of wind energy poses challenges from an energy planning perspective whose needs include a robust accounting of resource potential across time and space. Using a machine learning methodology we analyzed spatial variation in wind power density, specifically investigating the factors that influence power density and attempting to capture these relationships through predictive modeling at the national scale. We found a mean modeled national power density of 2.82  0.75 MW/km2 and substantial regional variation with power densities ranging between 1.6 and 5.7 MW/km2. Employing model interrogation techniques we discovered that the most important explanatory features were related to wind resource characteristics, accessibility to urban centers, and forest cover. Our efforts to model geospatial variation in power density are an important step towards providing a robust, spatially-explicit foundation from which to assess wind energy potential. Derived insights are likely to illuminate land use implications of an expanding renewable energy portfolio and can further inform the ways in which growing energy needs may be met by wind deployment in diverse settings.

Abstract Information

This abstract is already part of a session. View the session here.

To access contact information login