Neural networks for wind power generation forecasting in Uruguay

This article explores the application of Artificial Neural Networks in predicting wind power generation. The precise prediction of power output is crucial for establishing a reliable wind power generation framework. Accurate forecasts play a key role in enabling efficient planning, management, and distribution strategies for the generated power, leading to improved performance of electrical systems. A specific case study is addressed: the nation-wide wind power generation in Uruguay. The study considers real wind power generation data from Uruguay, spanning the period from 2018 to 2022. The forecasting is carried out using different artificial neural network architectures, including Convolutional Neural Network, Long Short-Term Memory artificial neural network, Encoder-Decoder architectures, and hybrid models. Univariate and multivariate models are studied. The main results indicate that the best proposed hybrid approach demonstrates high prediction accuracy, with a median root mean square error of 0.12, median absolute error of 0.09, and mean absolute percentage error of 14.9%, significantly improving over other models. The computed forecasts can be effectively utilized for planning and scheduling to enhance the quality of services provided by the smart electric grid.