Sử dụng mạng nơ-ron nhân tạo dự báo mực nước sông chịu ảnh hưởng của thủy triều

Abstract

Recurrent Neural Network (RNN) is widely used in many different fields, including irrigation. RNN models have been applied to forecast river water levels, reservoir’s inflow... In this paper, the authors developed a Long Short-Term Memory Network model (LSTM), a special type of RNN, to predict water levels downstream of Cau Cat Culvert in the Bac Hung Hai irrigation system. The input data of the model are just the water levels downstream of Cau Cat Culvert in the past, the predicted result is the water level there for 6 hours, 12 hours, 18 hours and 24 hours of lead-time in the future. The model proposed by the authors provides results with high accuracy and stability, with Nash coefficients ranging from 95.3% to 91.6% corresponding to the predicted cases. Therefore, it is possible to use this model to forecast the water level in the tidal sluice, helping to operate the culvert safely and effectively.

Key words: Recurrent Neural Network (RNN), LSTM,  forecast river water levels, Cau Cat Culvert.