Recurrent Neural Network Predictions for Water Levels at Drainage Pumping Stations in an Agricultural Lowland

Drainage management in a complicated system an agricultural lowland must operate pumps flexibly and quickly, based on the water level at pumping station. A data-driven model without any physical-based information was implemented drainage to predict of lagoon near main We employed long shortterm memory (LSTM) as advanced neural network utilize field datasets obtained from water-related facilities sensors over about eight years input data. performed sensitivity tests for accuracy with different types data locations using cross-validation error quantity between observed predicted levels The results showed that LSTM all available better than models several parts or it roughly equivalent those entire period 3-h 6-h lead times. In addition, only inputs rainfall by stations subperiod, including severest flood event.