A novel deep learning intelligent clustered hybrid models for wind speed and power forecasting. (15th December 2020)

• Record Type: Journal Article
• Authors: Aly, Hamed H.H.
• Is Part Of: Energy. Volume 213(2020)
• Publication Date: 2020-12-15
• Subjects: Wind speed forecasting -- ANN -- FS -- WNN -- RKF -- Clustering techniques; Power resources -- Periodicals; Power (Mechanics) -- Periodicals; Energy consumption -- Periodicals; 333.7905
• DOI: 10.1016/j.energy.2020.118773
• Languages: English
• ISSNs: 0360-5442

Abstract

Abstract: Wind energy is playing a compromising role in the new generation of sustainable energy and promising to increase more. Forecasting of the fluctuated wind speed and its output power is playing an essential role in the smart power system grid. The wind power integration is based on the accuracy of the wind speed and power forecasting models. This paper is proposing highly accurate hybrid deep learning clustered models for wind speed and power forecasting using different artificial intelligent systems for optimal performance. Various combinations of Recurrent Kalman Filter (RKF), Fourier Series (FS), Wavelet (WNN) and Artificial Neural Network (ANN) are used in this work. Twelve different hybrid models are proposed and tested. The novelty of this work is the applied clustered segments with the deep learning hybrid models to improve the aggregated system performance. This work is validated by using different unseen data set with the proposed models as well as using K-fold cross validation method. All the proposed models are performing well with high accurate results, but the hybrid clustered model of WNN and RKF outperforms all other models. Highlights: Twelve different hybrid models are proposed for wind speed and power forecasting. The hybrid approaches are combinations of Wavelet, ANN, FS and RKF. The most accurate hybrid model is the model of WNN and RKF with nRMSE of 0.04446. The models are tested and validated using K-fold cross validation and another data set. The proposed models are tested using data collected from Halifax in 2008.