Development of Neural Network Model Based on Conjugate Gradient and Resilient Back-Propagation Training Function for Estimation of Longitudinal Dispersion Coefficient in Rivers

Abstract

Determining the longitudinal dispersion coefficient (LDC) for Advection-Diffusion equation is the first step in water quality modeling for one-dimensional water bodies such as rivers. In this research, an artificial neural network (ANN) model has been developed based on the standard numerical optimization algorithms and heuristic techniques to determine the LDC. In this regard, conjugate gradient (CG) training functions including Fletcher-Reeves, Polak-Ribiére, Powell-Beale and scaled conjugate gradient functions from the standard numerical optimization algorithms category and resilient back-propagation (Trainrp) training function from the heuristic algorithms, have been applied to optimizing ANN parameters. Then, the best model has been selected for each of the training functions according to indices that are used to evaluate results. Among the selected models, the ANN model with the Trainrp training function has been selected as the best model to predict the LDC due to DDR statistic. Finally, a comparison has been undertaken between the selected model and other suggested artificial intelligent methods by the researchers. According to the implemented comparisons, the Trainrp function acquired the best performance.

Keywords

• longitudinal dispersion coefficient
• water pollutant
• intelligence models
• training algorithm