Neuro-Fuzzy GMDH-Based Evolutionary Algorithms to Predict Flow Discharge in Straight Compound Channels

In this study, neuro-fuzzy-based group method of data handling (NF-GMDH) as an adaptive learning network is used to predict the flow discharge in straight compound channels. The NF-GMDH network is developed by using the particle swarm optimization (PSO) and gravitational search algorithm (GSA). The depth ratio (ratio of water depth in floodplain to that in main channel), coherence parameter, and the discharge ratio [ratio of flow discharge calculated from vertical divided channel method (VDCM) to the bank full discharge] are considered as input parameters to represent a functional relationship between input and output parameters. The performances of training and testing stages for NF-GMDH models were quantified in terms of statistical error parameters. Also, the results of performances were compared with those obtained by using linear genetic programming, nonlinear regression methods, and VDCM. Evaluation of the proposed model demonstrated that NF-GMDH-GSA network provides a more accurate prediction than the NF-GMDH-PSO network. Finally, statistical error parameters indicated that the NF-GMDH networks as a new soft-computing tool produced better prediction of flow discharge in comparison with linear genetic programming, nonlinear regression methods, and VDCM.