Investigation of neural network and fuzzy inference neural network and their optimization using meta-algorithms in river flood routing

Abstract

Flood routing is one of the methods of flood forecasting in rivers to manage and control the flood. Today, the new technique of using the intelligent models is widely reported in various fields of science and engineering, particularly water resources. In this research, flood routing was studied using artificial neural network (ANN) and adaptive neuro-fuzzy inference system (ANFIS) models. By using the bat algorithm and imperialist competitive algorithm (ICA), the structure of ANN models was optimized. This process was repeated for combining genetic algorithm and particle swarm optimization algorithm with the ANFIS model. Four input patterns were used for network training, which It−7, It−6, Qt−1, Qt−2 pattern was the best pattern for network input according to the evaluation test. Results of routing of 8 flood hydrographs (6 hydrographs for network training and 2 hydrographs for network testing) indicated that the ANN–ICA predicted the hydrograph volume, peak flow and flood time more accurately. The statistical analyses at the training stage were: RMSE = 0.33, MARE = 0.32, SI = 0.05, BIAS = 0.18 and at the testing stage were: RMSE = 0.3, MARE = 0.32, SI = 0.04, BIAS = 0.08. Also, according to the sensitivity analysis, It−6 has the highest impact on flood discharge. Finally, the flood hydrograph was predicted for a return period of 10,000 years.