Abstract
Being able to predict the output of a wind farm online for a day or two in advance has significant advantages for utilities, such as better possibility to schedule fossil fuelled power plants and a better position on electricity spot markets. In this paper prediction methods based on Numerical Weather Prediction (NWP) models are considered. The spatial resolution used in NWP models implies that these predictions are not valid locally at a specific wind farm. Furthermore, due to the non-stationary nature and complexity of the processes in the atmosphere, and occasional changes of NWP models, the deviation between the predicted and the measured wind will be time dependent. If observational data is available, and if the deviation between the predictions and the observations exhibits systematic behavior, this should be corrected for; if statistical methods are used, this approaches is usually referred to as MOS (Model Output Statistics). The influence of atmospheric turbulence intensity, topography, prediction horizon length and auto-correlation of wind speed and power is considered, and to take the time-variations into account, adaptive estimation methods are applied. Three estimation techniques are considered and compared, Extended Kalman Filtering, recursive least squares and a new modified recursive least squares algorithm. (au)
More>>
Joensen, A;
Giebel, G;
Landberg, L;
[1]
Madsen, H;
Nielsen, H A
[2]
- Risoe National Lab., Roskilde (Denmark)
- The Technical Univ. of Denmark, Dept. of Mathematical Modelling, Lyngby (Denmark)
Citation Formats
Joensen, A, Giebel, G, Landberg, L, Madsen, H, and Nielsen, H A.
Model output statistics applied to wind power prediction.
Denmark: N. p.,
1999.
Web.
Joensen, A, Giebel, G, Landberg, L, Madsen, H, & Nielsen, H A.
Model output statistics applied to wind power prediction.
Denmark.
Joensen, A, Giebel, G, Landberg, L, Madsen, H, and Nielsen, H A.
1999.
"Model output statistics applied to wind power prediction."
Denmark.
@misc{etde_679636,
title = {Model output statistics applied to wind power prediction}
author = {Joensen, A, Giebel, G, Landberg, L, Madsen, H, and Nielsen, H A}
abstractNote = {Being able to predict the output of a wind farm online for a day or two in advance has significant advantages for utilities, such as better possibility to schedule fossil fuelled power plants and a better position on electricity spot markets. In this paper prediction methods based on Numerical Weather Prediction (NWP) models are considered. The spatial resolution used in NWP models implies that these predictions are not valid locally at a specific wind farm. Furthermore, due to the non-stationary nature and complexity of the processes in the atmosphere, and occasional changes of NWP models, the deviation between the predicted and the measured wind will be time dependent. If observational data is available, and if the deviation between the predictions and the observations exhibits systematic behavior, this should be corrected for; if statistical methods are used, this approaches is usually referred to as MOS (Model Output Statistics). The influence of atmospheric turbulence intensity, topography, prediction horizon length and auto-correlation of wind speed and power is considered, and to take the time-variations into account, adaptive estimation methods are applied. Three estimation techniques are considered and compared, Extended Kalman Filtering, recursive least squares and a new modified recursive least squares algorithm. (au) EU-JOULE-3. 11 refs.}
place = {Denmark}
year = {1999}
month = {Mar}
}
title = {Model output statistics applied to wind power prediction}
author = {Joensen, A, Giebel, G, Landberg, L, Madsen, H, and Nielsen, H A}
abstractNote = {Being able to predict the output of a wind farm online for a day or two in advance has significant advantages for utilities, such as better possibility to schedule fossil fuelled power plants and a better position on electricity spot markets. In this paper prediction methods based on Numerical Weather Prediction (NWP) models are considered. The spatial resolution used in NWP models implies that these predictions are not valid locally at a specific wind farm. Furthermore, due to the non-stationary nature and complexity of the processes in the atmosphere, and occasional changes of NWP models, the deviation between the predicted and the measured wind will be time dependent. If observational data is available, and if the deviation between the predictions and the observations exhibits systematic behavior, this should be corrected for; if statistical methods are used, this approaches is usually referred to as MOS (Model Output Statistics). The influence of atmospheric turbulence intensity, topography, prediction horizon length and auto-correlation of wind speed and power is considered, and to take the time-variations into account, adaptive estimation methods are applied. Three estimation techniques are considered and compared, Extended Kalman Filtering, recursive least squares and a new modified recursive least squares algorithm. (au) EU-JOULE-3. 11 refs.}
place = {Denmark}
year = {1999}
month = {Mar}
}