Skip to main content

LUP Student Papers

LUND UNIVERSITY LIBRARIES

Wind power forecasting using random forests

Ahrling, Christoffer LU (2023) MVKM01 20232
Department of Energy Sciences
Abstract
The present thesis investigated using the random forest machine learning algorithm for
wind power forecasting. Meteorological prognoses for wind speed, wind direction, gust
winds, and humidity were used. For historical data, wind minimum and temperature was
also included. The results were evaluated using root means square error (RMSE), mean
absolute error (MAE), normalized mean absolute error (NMAE), and normalized mean
biased error (NMBE). Results from the ’Hans’ storm in 2023 were shown separately.
Historical data covered a year and three months. Meteorological prognoses covered a
month. Historical data was used to show that seasonality impacts forecasting.
The best results showed NMAE decreasing (compared to a simple polynomial... (More)
The present thesis investigated using the random forest machine learning algorithm for
wind power forecasting. Meteorological prognoses for wind speed, wind direction, gust
winds, and humidity were used. For historical data, wind minimum and temperature was
also included. The results were evaluated using root means square error (RMSE), mean
absolute error (MAE), normalized mean absolute error (NMAE), and normalized mean
biased error (NMBE). Results from the ’Hans’ storm in 2023 were shown separately.
Historical data covered a year and three months. Meteorological prognoses covered a
month. Historical data was used to show that seasonality impacts forecasting.
The best results showed NMAE decreasing (compared to a simple polynomial model)
from 8.5 % to 6.6 % using historical data, from 9.5 % to 8.6 % using meteorlogical
prognoses, and negligible improvements under storm conditions. The results indicate
that a random forest model can yield improvements in wind power forecasting. It is
simultaneously shown to be important that the models are based on good data and employ
good meteorological prognoses with high levels of agreement between them, and with
the turbine site. (Less)
Please use this url to cite or link to this publication:
author
Ahrling, Christoffer LU
supervisor
organization
course
MVKM01 20232
year
type
H2 - Master's Degree (Two Years)
subject
keywords
machine learning, wind power, forecasting, random forest
report number
LUTMDN/TMHP-23/5552-SE
ISSN
0282-1990
language
English
id
9140595
date added to LUP
2023-10-30 16:03:15
date last changed
2023-10-30 16:03:15
@misc{9140595,
  abstract     = {{The present thesis investigated using the random forest machine learning algorithm for
wind power forecasting. Meteorological prognoses for wind speed, wind direction, gust
winds, and humidity were used. For historical data, wind minimum and temperature was
also included. The results were evaluated using root means square error (RMSE), mean
absolute error (MAE), normalized mean absolute error (NMAE), and normalized mean
biased error (NMBE). Results from the ’Hans’ storm in 2023 were shown separately.
Historical data covered a year and three months. Meteorological prognoses covered a
month. Historical data was used to show that seasonality impacts forecasting.
The best results showed NMAE decreasing (compared to a simple polynomial model)
from 8.5 % to 6.6 % using historical data, from 9.5 % to 8.6 % using meteorlogical
prognoses, and negligible improvements under storm conditions. The results indicate
that a random forest model can yield improvements in wind power forecasting. It is
simultaneously shown to be important that the models are based on good data and employ
good meteorological prognoses with high levels of agreement between them, and with
the turbine site.}},
  author       = {{Ahrling, Christoffer}},
  issn         = {{0282-1990}},
  language     = {{eng}},
  note         = {{Student Paper}},
  title        = {{Wind power forecasting using random forests}},
  year         = {{2023}},
}