TY - JOUR
T1 - A New Combinatory Approach for Wind Power Forecasting
AU - Abedinia, Oveis
AU - Bagheri, Mehdi
AU - Naderi, Mohammad Salay
AU - Ghadimi, Noradin
N1 - Funding Information:
Manuscript received June 29, 2019; revised October 23, 2019; accepted December 6, 2019. Date of publication January 20, 2020; date of current version September 2, 2020. This work was supported in part by the Program-Targeted Funding of the Ministry of Education and Science of the Republic of Kazakhstan through the Innovative Materials and Systems for Energy Conversion and Storage for 2018–2020 under Grant BR05236524 and in part by the Faculty Development Research Grant of Nazarbayev University, SoE 2018018. (Corresponding author: Oveis Abedinia.) O. Abedinia and M. Bagheri are with the Electrical and Computer Engineering Department, Nazarbayev University, Nursultan 010000, Kazakhstan and also with the National Laboratory Astana, Nursultan 010000, Kazakhstan (e-mail: oveis.abedinia@un.edu.kz; mehdi.bagheri@nu.edu.kz).
PY - 2020/9
Y1 - 2020/9
N2 - Wind power generation is considerably dependent on the weather condition and it is correlated closely with the air density, wind speed, and its direction through considering the transformer tap change cost. Therefore, an accurate prediction model is required to forecast and adopt with the complicated signals. In this article, an accurate prediction method trained with improved wavelet transform (IWT) to decompose the original signal to subsignals, new feature selection based on the maximum dependence, maximum relevancy, and minimum redundancy to filter the signal and select the best candidate inputs and a synthetic forecasting engine with minimum forecasting error is proposed. The proposed method of forecasting engine composed two-dimensional convolution neural network (TDCNN) and trained by improved optimization algorithm based on particle swarm optimization. The proposed improved optimization algorithm will fine-tune the weights of TDCNN to increase the prediction accuracy of the forecast engine. The training optimization processes and effectual fast classification are the main duty of the proposed intelligent algorithm. The efficiency of the suggested prediction approach is widely evaluated and compared to other prediction approaches using the practical power market data. A simulation study is conducted over the trained model and the results are discussed in detail. Obtained numerical results and also analysis in short-term and long-term forecasting horizons validate the high performance and advantages of the introduced approach.
AB - Wind power generation is considerably dependent on the weather condition and it is correlated closely with the air density, wind speed, and its direction through considering the transformer tap change cost. Therefore, an accurate prediction model is required to forecast and adopt with the complicated signals. In this article, an accurate prediction method trained with improved wavelet transform (IWT) to decompose the original signal to subsignals, new feature selection based on the maximum dependence, maximum relevancy, and minimum redundancy to filter the signal and select the best candidate inputs and a synthetic forecasting engine with minimum forecasting error is proposed. The proposed method of forecasting engine composed two-dimensional convolution neural network (TDCNN) and trained by improved optimization algorithm based on particle swarm optimization. The proposed improved optimization algorithm will fine-tune the weights of TDCNN to increase the prediction accuracy of the forecast engine. The training optimization processes and effectual fast classification are the main duty of the proposed intelligent algorithm. The efficiency of the suggested prediction approach is widely evaluated and compared to other prediction approaches using the practical power market data. A simulation study is conducted over the trained model and the results are discussed in detail. Obtained numerical results and also analysis in short-term and long-term forecasting horizons validate the high performance and advantages of the introduced approach.
KW - Accidental floater particle swarm optimization (AFPSO)
KW - and minimum redundancy (MDMRMR)
KW - convolution neural network (CNN)
KW - feature selection (FS)
KW - improved wavelet transform (IWT)
KW - maximum dependence
KW - maximum relevancy
KW - wind signal
UR - http://www.scopus.com/inward/record.url?scp=85083956744&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85083956744&partnerID=8YFLogxK
U2 - 10.1109/JSYST.2019.2961172
DO - 10.1109/JSYST.2019.2961172
M3 - Article
AN - SCOPUS:85083956744
VL - 14
SP - 4614
EP - 4625
JO - IEEE Systems Journal
JF - IEEE Systems Journal
SN - 1932-8184
IS - 3
M1 - 8963864
ER -