TY - JOUR
T1 - Artificial neural networks in the estimation of monthly capacity factors of WECS in Taiwan
AU - Tu, Yi Long
AU - Chang, Tsang Jung
AU - Hsieh, Cheng I.
AU - Shih, Jen Ying
PY - 2010/12
Y1 - 2010/12
N2 - Measured wind speed and power output time series with 10-min resolution at Jhongtun wind power station and 30-min resolution at Mailiao station have been used to estimate monthly energy output by artificial neural networks (ANNs) over a period spanned between 2002 and 2006. The widely used back propagation algorithm is used in the network. The available database of these two stations is divided into two parts - data from Year 2002 to Year 2005 is for training and Year 2006 data is used for the validation of training results. For the purpose of investigating the adequate training length of ANN simulations, four training periods (Year 2002-2005, 2003-2005, 2004-2005 and 2005) together with four training intervals (yearly, half-yearly, seasonal, and monthly) are input into the ANN model to estimate monthly capacity factors of the two stations at Year 2006 and compared with the measured data. The results show that ANN is an efficient tool for estimating wind power production. A training length with 12 months can provide satisfactory estimation of monthly capacity factors in both of the stations. Moreover, the half-yearly training interval, which is derived from the real wind characteristics in Taiwan, gives the best estimation of monthly capacity factors compared to other training intervals.
AB - Measured wind speed and power output time series with 10-min resolution at Jhongtun wind power station and 30-min resolution at Mailiao station have been used to estimate monthly energy output by artificial neural networks (ANNs) over a period spanned between 2002 and 2006. The widely used back propagation algorithm is used in the network. The available database of these two stations is divided into two parts - data from Year 2002 to Year 2005 is for training and Year 2006 data is used for the validation of training results. For the purpose of investigating the adequate training length of ANN simulations, four training periods (Year 2002-2005, 2003-2005, 2004-2005 and 2005) together with four training intervals (yearly, half-yearly, seasonal, and monthly) are input into the ANN model to estimate monthly capacity factors of the two stations at Year 2006 and compared with the measured data. The results show that ANN is an efficient tool for estimating wind power production. A training length with 12 months can provide satisfactory estimation of monthly capacity factors in both of the stations. Moreover, the half-yearly training interval, which is derived from the real wind characteristics in Taiwan, gives the best estimation of monthly capacity factors compared to other training intervals.
KW - Artificial neural network
KW - Monthly capacity factor
KW - Training length
KW - Wind power
UR - http://www.scopus.com/inward/record.url?scp=77955472387&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=77955472387&partnerID=8YFLogxK
U2 - 10.1016/j.enconman.2010.06.035
DO - 10.1016/j.enconman.2010.06.035
M3 - Article
AN - SCOPUS:77955472387
SN - 0196-8904
VL - 51
SP - 2938
EP - 2946
JO - Energy Conversion and Management
JF - Energy Conversion and Management
IS - 12
ER -