Long Short-Term Memory Algorithm for Rainfall Prediction Based on El-Nino and IOD Data

作     者：Dina Zatusiva Haq Dian Candra Rini Novitasari Abdulloh Hamid Nurissaidah Ulinnuha Arnita Yuniar Farida RR. Diah Nugraheni Rinda Nariswari Ilham Hetty Rohayani Rahmat Pramulya Ari Widjayanto 

作者机构：Department of Mathematics Universitas Negeri Medan Medan Indonesia 20221 Department of Environmental Engineering UIN Sunan Ampel Surabaya Indonesia 60237 Statistics Department School of Computer Science Bina Nusantara University Jakarta Indonesia 11480 Department of Information System UIN Sunan Ampel Surabaya Indonesia 60237 Department of Information Technology Adiwangsa Jambi University Jambi Indonesia 36125 Faculty of Agriculture University Teuku Umar Aceh Indonesia Meteorological Climatological and Geophysics Agency Surabaya Indonesia 60165 

出 版 物：《Procedia Computer Science》 (计算机科学会议集)

年 卷 期：2021年第179卷

页      面：829-837页

学科分类：08[工学] 0812[工学-计算机科学与技术（可授工学、理学学位）] 

主　　题：Deep Learning Long Short-Term Memory LSTM Rainfall Forecasting 

摘      要：Rainfall has the highest correlation with adverse natural disasters. One of them, rainfall can cause damage to the hot mud embankments in Sidoarjo, East Java, Indonesia. Therefore, in this study, rainfall prediction is carried out to anticipate the damage to the embankments. The rainfall prediction was carried out using Long Short-Term Memory (LSTM) based on rainfall parameters: El-Nino and Indian Ocean Dipole (IOD). Experiments were carried out with two schemes: the first scheme used the El-Nino and IOD parameters, while the second scheme used rainfall time series pattern. Each scheme used varied number of hidden layers, batch size, and learn drop period. The prediction results using El-Nino and IOD parameters obtained MAAPE values ​​of 0.9644 with hidden layer, batch size and learn rate drop period values ​​of 100, 64, and 50. The prediction results using rainfall parameters resulted in a more accurate prediction with a MAAPE value of 0.5810. The best prediction results were obtained with the number of hidden layers, batch size and learn rate drop period of 100, 32, and 150 respectively.