Short-term wind forecasting using hybrid deep learning techniques

In parallel with the increasing energy demand, using renewable energy sources is becoming increasingly important when environmental impacts are considered. In light of the rising demand for renewable energy and the objective of attaining carbon neutrality, the wind energy sector has witnessed a period of accelerated expansion in recent times. The integration of large-scale offshore wind energy into the electricity grid presents a significant challenge to grid operation and distribution, due to the inherent variability and intermittency of wind speed. To integrate large-scale wind power plants into the electricity grid and to optimize grid reliability and efficiency, it is of great importance to have access to short-term forecasts. This thesis presents the development of innovative short-term hybrid forecasting models utilizing both onshore wind speed and offshore wind power data. In this context, the performance of primary and secondary use of decomposition methods, as well as deep learning-based forecasting approaches, is analyzed comparatively. In addition to the traditional Multilayer Perceptron (MLP), Echo State Network (ESN), Long Short-Term Memory (LSTM), and Bidirectional LSTM (BiLSTM) approaches are employed in the construction of the prediction model. The models under consideration are those employing either single or secondary decomposition. These include Empirical Mode Decomposition (EMD), Ensemble Empirical Mode Decomposition (EEMD), Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN), Swarm Decomposition (SWD), Wavelet Decomposition (WD) and Variational Mode Decomposition (VMD). All analyses are presented statistically in comparison with offshore wind power data and onshore wind speed data from three different regions. Concerning the error performance criteria, it is observed that the innovative secondary decomposition-based models proposed in this thesis yield superior results.