Estimating Longitudinal Dispersion Coefficient of Pollutants in Open Channel Flows Using Artificial Neural Networks

The longitudinal dispersion of pollutants is one of the most effiective phases of the pollutants dilution process, which having insight about it is of importance. The complexity of measuring longitudinal dispersion coefficient in rivers increases the necessity of using appropriate methods of modeling to predict it. One of the most efficient methods for modeling is the artificial neural network which is one of the artificial intelligence techniques. In this model, without applying the complex nonlinear equations, the dynamics of the system can be extracted and, by this way the output of the model can be predicted. In this study, the longitudinal dispersion coefficient was predicted by artificial neural network (ANN), using hydraulic and geometric parameters of the streams as input parameters. Results indicated that the feed forward perceptron network had a suitable precision in estimating the longitudinal dispersion coefficient. Sensitivity analysis indicated that in the model, for which the ratio of velocity to the shear velocity was considered as an input variable, the determination coefficient and error function were equal to 0.84 and 0.87%, respectively. However, inthe modelwith an input variable of width to flow depth ratio, the determination coefficient and error function were obtained 0.7 and 1.01%, respectively. Therefore, the ratio of the velocity to the shear velocity or roughness coefficient had a greater impact on longitudinal dispersion coefficient, as compared with the last one. The proposed methodology is an efficient approach to estimate dispersion coefficient in streams and can be implemented into mathematical models of pollutant transfer.