Machine learning and geographic information systems for large-scale wind energy potential estimation in rural areas

Dan Assouline,Nahid Mohajeri,Jean-Louis Scartezzini,Dasaraden Mauree

doi:10.1088/1742-6596/1343/1/012036

Dan Assouline, Nahid Mohajeri + Show 2 more

Open Access

https://doi.org/10.1088/1742-6596/1343/1/012036

Copy DOI

Abstract

Clean, safe, affordable and available in the long-term, wind is one of the most promising sources of renewable energy. Its optimized and profitable use, however, requires an estimation of the potential in locations of interest, given its very volatile behavior in various settings. In the present study, we propose a methodology using a combination of Machine Learning (Random Forests), Geographic Information Systems and wind parametric models to estimate the large-scale theoretical wind speed potential in rural areas over the entire Switzerland. The monthly wind speed over rural areas is estimated based on wind speed measurements and several meteorological, topographic, and wind-specific features available accross the country. Wind speed values and their associated uncertainty are computed at the scale of 200 x 200 [m2] pixels covering the territory, at a typical height for rural commercial wind turbine installation, that is, z=100m. The developed methodology, is, however, applicable to any large region, given the availability of data of interest. The results show that in the case of Switzerland, wind turbines could approximately represent an non-negligible installed power capacity of, for each pixel and for each turbine installation, on average 80 kW in Swiss rural areas, and up to 1600 kW in most suitable pixels.

Highlights

Given its clean nature and the efficiency of modern wind turbines, wind energy has become very popular over the last decade and have seen its installed capacity growing all around the world [1]
We propose a methodology using a combination of Machine Learning (Random Forests), Geographic Information Systems and wind parametric models to estimate the largescale theoretical wind speed potential in rural areas over the entire Switzerland
The steps of the methodology are as follows: (i) collect data and extract significant features impacting the behavior of wind speed in Switzerland, (ii) train monthly Random Forests (RFs) with wind monitored data at 10m as output and the features previously extracted as inputs, (iii) use trained RFs to estimate monthly wind speed maps at z=10m in rural areas, (iv) vertically extrapolate the wind speed to a height of 100m, using a taditional log wind profile, (v) compute the potential wind power generated by turbines over the territory

Summary

Introduction

Given its clean nature and the efficiency of modern wind turbines, wind energy has become very popular over the last decade and have seen its installed capacity growing all around the world [1]. The main source of available wind maps Switzerland is offered by the Swiss Wind Atlas, of the Swiss Federal Office of Energy (SFOE) (www.uvek-gis.admin.ch/BFE/storymaps/EE Windatlas) These maps are computed using Weibull distributions fitted with parameters estimated based on wind measurements. To perform an accurate mapping of monthly wind speed, we here propose a hybrid methodology that combines Machine Learning (Random Forests) and GIS It is applied for the theoretical potential estimation of wind energy in Swiss rural areas. The steps of the methodology are as follows: (i) collect data and extract significant features impacting the behavior of wind speed in Switzerland (meteorological, topographic, and other wind-related variables), (ii) train monthly Random Forests (RFs) with wind monitored data at 10m as output and the features previously extracted as inputs, (iii) use trained RFs to estimate monthly wind speed maps at z=10m in rural areas, (iv) vertically extrapolate the wind speed to a height of 100m, using a taditional log wind profile, (v) compute the potential wind power generated by turbines over the territory

Data and methods

50 Kilometers

Findings

Conclusion