Prediction of stable channels geometry using soft computing

Determination of stable channel characteristics includes width, depth and slope is very important that considered for more than a century. Design of Stable channel was used in various works such as river engineering, flood control and water conveyance. The main objective of this study is evaluation of two methods of ANFIS and SVM to estimate stable channel characteristics.
325 measured data from the natural channel and laboratory investigations were used for training, testing and evaluation ANFIS and SVM methods. ANFIS system that combines neural network with fuzzy logic is the first time was introduced in 1993 by Zhang. Support Vector Machine can be applied not only to classification problems but also to the case of regression. 60% of data was used for training, 20% evaluation and the remaining 20% were used for test. To simulate the channel characteristics two input include: 1- discharge and 2- discharge and median sediment grains were used. The empirical formula Afzalimehr et al., Bray and Simmons and Albertson was used to compare with ANFIS and SVM. 1- discharge and 2- discharge and median sediment grains were used. The empirical formula Afzalimehr et al., Bray and Simmons and Albertson was used to compare with ANFIS and SVM.
ANFIS and SVM methods with input (2) to (1) estimate width 50% and 80% respectively and depth 61% and 40% respectively with a lower error. ANFIS and SVM prediction accuracy in various range of width and depth is different. Both methods could not predict the slope. Bray empirical relationship that predicted depth and width of the reasonably accurately estimates the slope with less accuracy.
The results showed that both methods with input (2) simulate changes in channel geometry with reasonable accuracy and estimate the width and depth as well. Overall, estimation capability of width more than depth and both methods with input (1) and (2) can’t estimate of slope stable channel. In depth of less than 2 meters the impact of median grain size is little on the predicted depth. Changes in slope do not depend only on the discharge and median grain size and other parameters that affect the change. The effects of the unknown parameters on slopes greater than 0.5% are high because both methods did not provide any reasonable estimates. Compared with empirical relations showed ANFIS more accurately estimate characteristic of stable channel than Simons and Albertson, Afzalimehr et al. and Bray relationship.