Realtime Weather Prediction System Using GRU with Daily Surface Observation Data from IoT Sensors
Hendri Darmawan, Mike Yuliana, Moch. Zen Samsono Hadi
Abstract
Weather is an essential part of everyone's life, but its dynamic nature makes weather forecasting difficult. In this research, we propose a system that makes daily weather predictions in Sedati, an administrative district in the Sidoarjo Regency, Indonesia. The system uses Gated Recurrent Unit (GRU) algorithm that is fed with historical weather data from the BMKG Juanda Meteorological Station. On the test data, the GRU regression model resulted in the best average RMSE for four days' worth of prediction with humidity, wind speed, temperature, sea level pressure, maximum wind speed, minimum temperature, maximum temperature, and dew point at 3.7325; 1.31; 0.64; 0.85; 2,065; 0.7125; 0.6825; and 0.705. However, the average RMSE score of those variables after we integrated the model with local IoT devices were 5.6; 2.3; 1.56; 0.57; 2.47; 3.04; 1.78; and 0.55, respectively. The sequence GRU model for rain classification achieved an accuracy of 0.88, and the accuracy after the model was integrated with the IoT devices was 0.75.