Flow forecasting in multiple sections of a river system

• Published in: KSCE journal of civil engineering Vol. 21; no. 2; pp. 512 - 522
• Main Authors: Tripura, Joseph, Roy, Parthajit
• Format: Journal Article
• Language: English
• Published: Seoul Korean Society of Civil Engineers 01.02.2017; Springer Nature B.V; 대한토목학회
• Subjects: Artificial neural networks; Civil Engineering; Design Optimization and Applications in Civil Engineering; Economic forecasting; Engineering; Floods; Flow; Forecasting; Freshwater; Geotechnical Engineering & Applied Earth Sciences; Industrial Pollution Prevention; Lead time; Mathematical models; Neural networks; Nonlinearity; Recurrent neural networks; Rivers; Tributaries; 토목공학; India, Assam; India; river systems; short range forecasting; NARX; PRNN; MIMO model
• ISSN: 1226-7988; 1976-3808
• DOI: 10.1007/s12205-017-1514-9

A river system includes the combination of flows occurring simultaneously in the main river and its contributing tributaries. Any change in the flow condition of the river system is caused due to changes in flow of the main river and/or contributing tributaries. An accurate flow forecasting at multiple sections of a river system is worthy for issuing early warning to the imminent floods and in regulating the reservoir outflows. The application of multiple-inputs and multiple-outputs (MIMO) model is an effective way for simultaneous flow forecasting as it provides interrelation among multiple input and multiple output variables simultaneously. In the present study an Artificial Neural Networks (ANN) based MIMO model has been developed for Barak river system in Assam, India using Partially Recurrent Neural Network (PRNN) and Nonlinear Autoregressive with Exogenous Inputs (NARX) approaches. Performance of the model using both NARX and PRNN provide an efficacy with coefficient of efficiency (CE) > 0.87 and Mean Absolute Percentage Error (MAPE) < 7.66% at 12 hour lead time forecasting. This indicates satisfactory model performances for simultaneous flow forecasting at multiple sections of a river system however the results obtained by MIMO using NARX (MIMONARX) perform better than MIMO using PRNN (MIMO-PRNN) in terms of statistical performance criterion.