Neuro-Fuzzy GMDH Approach to Predict Longitudinal Dispersion in Water Networks

• Published in: Water resources management Vol. 29; no. 7; pp. 2205 - 2219
• Main Authors: Najafzadeh, Mohammad, Sattar, Ahmed M. A.
• Format: Journal Article
• Language: English
• Published: Dordrecht Springer Netherlands 01.05.2015; Springer Nature B.V
• Subjects: Algorithms; Analysis; Atmospheric Sciences; Average velocity; Civil Engineering; Coefficients; Contaminants; Contamination; Dispersion; Dispersions; Earth and Environmental Science; Earth Sciences; Environment; Flow velocity; Friction; Geotechnical Engineering & Applied Earth Sciences; Hydrogeology; Hydrology/Water Resources; Machine learning; Mathematical models; Networks; Optimization; Pipe; Pipe flow; Pipelines; Reynolds number; Sensitivity analysis; Studies; Turbulence; Turbulent flow; Variables; Water quality; Water utilities; Longitudinal dispersion coefficient; NF-GMDH; Pipe network; PSO algorithm
• ISSN: 0920-4741; 1573-1650
• DOI: 10.1007/s11269-015-0936-8

Longitudinal dispersion in pipelines leads to changes in the characteristics of contaminants. It is critical to quantify these changes because the contaminants travel through water networks or through chemical reactors. The essential characteristics of longitudinal dispersion in pipes can be described by the longitudinal dispersion coefficient. This paper presents the application of the adaptive Neuro fuzzy group method of data handling to develop new empirical formulae for the prediction of longitudinal dispersion coefficients in pipe flow using 233 experimental case studies of dispersion coefficient with a R e range of 900 to 500,000 spanning laminar, transitional and turbulent pipe flow. The NF-GMDH network was improved using particle swarm optimization based evolutionary algorithm. The group method data handling is used to develop empirical relations between the longitudinal dispersion coefficient and various control variables, including the Reynolds number, the average velocity, the pipe friction coefficient and the pipe diameter. GMDH holds advantage in the case of small data samples due to the optimal choice of the model complexity with automatic adaptation to an unknown level of the data uncertainties. Sensitivity analysis is performed on the developed model and the weight and importance of each control variable is presented. The results indicate that the proposed relations are simpler than previous numerical solutions and can effectively evaluate the longitudinal dispersion coefficients in pipe flow.