A Comparative Study of LSTM and GRU Models for Wind Forecasting

Abstract

The use of deep learning in the current technological era is increasingly widespread, including in the field of meteorology to support aviation safety. As an archipelagic country, Indonesia faces significant challenges in ensuring flight safety due to unpredictable weather conditions, particularly wind direction and speed, which greatly influence takeoff and landing operations. To address these challenges, the Automatic Weather Observing System (AWOS) plays a crucial role in providing real-time weather data. This study aims to compare the performance of two popular deep learning models for time series data, namely Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU), in forecasting wind direction and speed based on AWOS data from Sultan Hasanuddin International Airport for the period of January 2020–December 2022, obtained from the National Oceanic and Atmospheric Administration (NOAA) website. After preprocessing, five out of eight attributes were used for modeling. The evaluation results show that the LSTM model consistently outperformed GRU in all forecasting scenarios (30 minutes, 1 hour, and 1.5 hours). For wind direction, LSTM achieved MAE values of 10.92°–11.01°, MSE 242.45–247.89, and RMSE 15.57°–15.74°, all lower than those of GRU. For wind speed, LSTM recorded MAE values of 30.32–31.72 knots, MSE 1868.53–2013.92, and RMSE 43.23–44.88 knots, also outperforming GRU. This research is expected to contribute to the development of risk mitigation systems and the advancement of weather forecasting technology in the future.