tailieunhanh - Ứng dụng mạng sai số lan truyền ngược dự báo dòng chảy kiệt trên sông Tả Trạch tỉnh Thừa Thiên Huế

Bài viết đã thiết lập mạng EBN gồm 1 lớp neuron đầu vào, 1 lớp neuron đầu ra và 2 lớp ẩn. Lớp đầu vào là chuỗi giá trị mực nước, lưu lượng trung bình thời đoạn trước và lượng mưa thời đoạn sau tương ứng với giá trị lượng mưa dự báo theo thời đoạn. Lớp đầu ra là lưu lượng trung bình thời đoạn dự báo. | Transport and Communications Science Journal Vol 71 Issue 8 10 2020 1000-1015 Transport and Communications Science Journal APPLICATION OF EBN TO PREDICT THE DRY DISCHARGE IN TA TRACH RIVER THUA THIEN HUE PROVINCE VIETNAM Hoang Nam Binh Le Thi Viet Ha University of Transport and Communications No 3 Cau Giay Street Hanoi Vietnam ARTICLE INFO TYPE Research Article Received 28 5 2020 Revised 17 8 2020 Accepted 9 9 2020 Published online 28 10 2020 https Corresponding author Email binhhn@ Abstract. Predicting dry season flows play an important role in distributing and managing water resources. Hydrological models can predict the flow with good quality results for the flow in large basins mainly affected by rainfall and buffer surface properties. The discharge is usually very small flow in the dry season and influenced by many factors. But none of them has a strong weight so it is difficult for experts to predict the flow. The article presents the results of the application of artificial neural network ANN with error backpropagation networks EBN for predicting the dry season discharge for Thuong Nhat station on Ta Trach river Thua Thien Hue province Vietnam. The structure of ANN is similar to the human brain so it is possible to find the relationship between inputs and outputs data by quot learning quot from existing data. EBNs have been established with an input and output neuron layer and two hidden layers. The input layer includes the water level discharge in the previous period and rainfall in the later period corresponding to the predicted rainfall. The output layer is predicting discharge. The results of sort and mid-term prediction have good quality with more than 80 output neuron satisfy the target error. The 24-hour prediction has the best accuracy with good pattern rate reaches in July. The 30-day prediction showed the lowest quality with good pattern rate of . Keywords Artificial Neural Network Error .