Long short term memory (LSTM) recurrent neural network (RNN) for discharge level prediction and forecast in Cimandiri river, Indonesia

• Published in: IOP conference series. Earth and environmental science 2019-07, Vol.299 (1), p.12037
• Main Authors: Sudriani, Y, Ridwansyah, I, A Rustini, H
• Format: Article
• Language: English
• Subjects: Algorithms; Decision making; Drinking water; Dry season; Fisheries; Fishery industry; Irrigation; Irrigation water; Long short-term memory; Machine learning; Neural networks; Power plants; Recurrent neural networks; Rice fields; River discharge; River flow; Rivers; Short term; Tourism; Water discharge; Water management
• ISSN: 1755-1307; 1755-1315; 1755-1315
• DOI: 10.1088/1755-1315/299/1/012037

Cimandiri watershed in Sukabumi prefecture of West Java, Indonesia, has been used for profitable activities such as power plant, rafting tourism, drinking water, and municipal, industries, agriculture, and fishery, and irrigation. More than 60% water source of PDAM, which supplies irrigation water of 1,217 Ha of rice fields and hundreds of industries, is obtained from the river. The irrigation in Cimandiri is even designed to be the model of irrigation in West Java. Low river discharge during the dry season can generate disadvantages to irrigation. This paper presents a method to predict and forecast the discharge of five days ahead to help the decision maker to control the operation of irrigation. We use a Deep Learning algorithm which involves Recurrent Long Short Term Memory (LSTM) Neural Network. Daily discharge data at two river gauge stations were analyzed. These stations are Leuwilisung (17 years data), and Tegaldatar (13 years data). The result shows that the relative errors are below 10% which is acceptable. In this study, dynamic changes of discharge level are evaluated to give a contribution to irrigation and water management control in Cimandiri River, Indonesia.