Time series prediction of wind speed
Date Issued
September 2004
Author(s)
Abstract
In this paper a time series prediction of wind speed with artificial neural networks is presented.
For this purpose the mean hourly wind speed records for the area of Kourris dam, located at the
south of Cyprus, are used. Wind data for ten consecutive years (1991-2000) are available for
this area. The network was trained to predict the mean monthly hourly wind speed of a year
(e.g. 1994) by using the values of wind speed for the same month, same hour for the three
previous years (e.g. 1991-1993), consecutively. The data for the wind speed up to the year 1999
have been used for the training of the network whereas those for the years 1997-1999 (input)
and 2000 (output) were used for the validation of the network. It should be noted that the data
for the year 2000 were completely unknown to the network. The wind speed for the validation
data set was predicted with a correlation coefficient of 0.82 which is satisfactory for wind speed
which is very unstable. Therefore the method proved to be very promising both for predicting
missing values and for forecasting.
For this purpose the mean hourly wind speed records for the area of Kourris dam, located at the
south of Cyprus, are used. Wind data for ten consecutive years (1991-2000) are available for
this area. The network was trained to predict the mean monthly hourly wind speed of a year
(e.g. 1994) by using the values of wind speed for the same month, same hour for the three
previous years (e.g. 1991-1993), consecutively. The data for the wind speed up to the year 1999
have been used for the training of the network whereas those for the years 1997-1999 (input)
and 2000 (output) were used for the validation of the network. It should be noted that the data
for the year 2000 were completely unknown to the network. The wind speed for the validation
data set was predicted with a correlation coefficient of 0.82 which is satisfactory for wind speed
which is very unstable. Therefore the method proved to be very promising both for predicting
missing values and for forecasting.
File(s)![Thumbnail Image]()
Name
C65-Met.pdf
Size
151.08 KB
Format
Adobe PDF
Checksum (MD5)
b6a0314b793b790814b6d9163c6eeca4