A framework development for predicting the longitudinal dispersion coefficient in natural streams using an artificial neural network

Authors

  • R. Noori,

    Corresponding author
    1. Department of Water Resources Research, Water Research Institute, Ministry of Energy, Tehran, Iran
    2. Department of Environmental Engineering, Graduate Faculty of Environment, University of Tehran, Tehran, Iran
    • Department of Water Resources Research, Water Research Institute, Ministry of Energy, Tehran, Iran
    Search for more papers by this author
  • A.R. Karbassi,

    1. Department of Environmental Engineering, Graduate Faculty of Environment, University of Tehran, Tehran, Iran
    Search for more papers by this author
  • H. Mehdizadeh,

    1. Research Institute of Petroleum Industry (RIPI), Tehran, Iran
    Search for more papers by this author
  • M. Vesali-Naseh,

    1. Department of Water Resources Research, Water Research Institute, Ministry of Energy, Tehran, Iran
    2. Department of Environmental Engineering, Graduate Faculty of Environment, University of Tehran, Tehran, Iran
    Search for more papers by this author
  • M.S. Sabahi

    1. Department of Environmental Engineering, Graduate Faculty of Environment, University of Tehran, Tehran, Iran
    Search for more papers by this author

Abstract

The main objective of the present investigation is to predict longitudinal dispersion coefficient (Kx) in natural streams using artificial neural network (ANN) technique based on most famous training functions such as Trainlm, Trainrp, Trainscg, Trainoss, and so on. To achieve the goal, hydraulic and geometric data (shear velocity, channel width, local flow depth, and mean longitudinal velocity) that are easily obtained in natural streams are used. First, we have tried to review the most well-known of published work in the field due to find out deficiencies of them. Second, new approach of ANN model based on the famous training functions is applied for predicting Kx in natural streams and then the best architectures for each training functions is selected by trial and error. Finally, Levenberg-Marquardt training function (Trainlm) is selected as the best choice for training the network parameters. Determination coefficient (R2) and mean absolute error for ANN (Trainrp) model were equal to 0.94 and 33 in the training and 0.95 and 30 in the testing steps, respectively. It is hoped that the presented methodology in the research, can be useful in river water quality management studies. © 2010 American Institute of Chemical Engineers Environ Prog, 2011.

Ancillary