Short-Term Wind Forecasting with Weather Data using Deep Learning - Case Study in Baron Techno Park

Abstract

In a microgrid with small-scale renewable sources, the unpredictable and highly variable nature of wind necessitates the adoption of reliable wind forecasting technologies. This study employs artificial neural networks (ANNs), specifically the Long Short-Term Memory (LSTM) and Multi-Layer Perceptron (MLP), which are classified as Deep Learning (DL) networks. These models integrate diverse weather data, such as wind speed, temperature, humidity, and atmospheric pressure, derived from actual measurements collected in Baron Techno Park, an isolated microgrid situated in the coastal region of Yogyakarta, Indonesia. For various scenarios, the root-mean-square error (RMSE) and mean absolute error (MAE) performances of the proposed ANN-based multivariable model are given and contrasted. Furthermore, it examines the impact of incorporating multiple local variables in contrast to solely relying on wind power, comparing against the persistence method. The findings reveal that the model incorporating a comprehensive set of weather data as inputs attains the lowest RMSE and MAE values. It is also can be concluded that additional weather data, even though they show almost no correlation to wind power in Baron Techno Park can improve short-term wind power prediction, with an improvement of 2.3% for every addition of weather parameter.