Comparison and analysis of long short-term memory and gated recurrent unit models in wind power prediction

With its clean and environmentally friendly profile, wind energy is a prominent renewable energy source on a global scale. However, meteorological factors and wind speed fluctuations make wind energy systems unstable, and necessitate the prediction of wind power. Wind power prediction plays a critical role in maintaining grid balance, efficient usage of renewable energy resources and optimizing energy trading, maintenance-repair operations and energy generation planning. For this purpose, in this thesis, long short-term memory and gated recurrent unit deep learning architectures were structured as single, two and three layers, and trained with adaptive moment estimation, root mean square propagation and stochastic gradient descent with momentum algorithms. Additionally, the optimal hyper-parameter values for each deep learning model were determined by utilizing the Bayesian optimization algorithm. The performance of the created prediction models were analyzed in terms of mean absolute percentage error, mean absolute error and root mean square error metrics. As a result of the analyses performed, the single-layer long short-term memory model based on the adaptive moment estimation training algorithm became the deep learning model that provided the highest prediction accuracy.