An expert system for predicting longitudinal dispersion coefficient in natural streams by using ANFIS

Longitudinal dispersion coefficient in rivers and natural streams usually is estimated by simple inaccurate empirical relations, because of the complexity of the phenomena. So, in this study using adaptive neuro-fuzzy inference system (ANFIS), which have the ability of perception and realization of phenomenon without need for mathematical governing equations, a new flexible tool is developed to predict the longitudinal dispersion coefficient. The process of training and testing of this new model is done using a set of available published filed data. Several statistical and graphical criterions are used to check the accuracy of the model. The dispersion coefficient values predicted by the ANFIS model satisfactorily compared with the measured data. The predicted values were also compared with those predicted using available empirical equations that have been suggested in the literature and it was found that the ANFIS model with R2 = 0.99 and RMSE = 15.18 in training stage and R2 = 0.91 and RMSE = 187.8 in testing stage is superior in predicting the dispersion coefficient than the best accurate empirical equation with R2 = 0.48 and RMSE = 295.7. The presented methodology in this paper is a new approach in estimating dispersion coefficient in streams and can be combined with mathematical models of pollutant transfer or real-time updating of these models.