Deep Learning Techniques for Wind Speed Forecasting at Palembang Airport

Abstract

The Sultan Mahmud Badaruddin (SMB) II Palembang Meteorological Station is a technical implementation unit (UPT) of the Meteorology, Climatology, and Geophysics Agency (BMKG) that plays a role in disseminating actual weather information, particularly at SMB II Palembang Airport. Various weather parameters are observed, one of which is wind speed. During the take-off and landing processes, wind speed is a crucial parameter used by airport personnel, including pilots and air traffic controllers (ATC). This study focuses on analyzing and evaluating three deep learning methods using the architectures of LSTM (Long Short Term Memory), GRU (Gated Recurrent Unit), and BiLSTM (Bidirectional Long Short Term Memory). Time series data such as air pressure, rainfall, humidity, and temperature are used as predictors. The data is sourced from the AWOS (Automatic Weather Observation System) device. After processing the data using deep learning methods with the architectures above, an analysis will be conducted to determine which architecture model is the most accurate based on the lowest loss error rate in forecasting wind speed at SMB II Palembang Airport. The results show that the GRU deep learning architecture has the lowest loss value compared to the LSTM and BiLSTM architectures so that it can produce better wind speed forecasts in the next 12 hours and 24 hours, with RMSE of 1.62 and 1.77, respectively.