Comparison the Performance of M5 Tree Model with the Artificial Neural Network and Support Vector Machine Models in Derivation of Flow Duration Curve, Case Study: Khazangah Station of Aras River

Abstract

Flow duration curve is one of the most important and applicable signals of hydrologic response of a basin. This curve was used for analyzing the frequency of low and flood flows of a river in many hydrologic uses. Also, the flow duration curve (FDC) was used to display the complete domain of river discharge from minimum up to maximum flood. Therefore, accurate derivation of this curves with the least error is necessary. In this study, applicability of M5 Tree Model in derivation of flow duration curve in Khazangah station located on Aras River, East Azerbaijan province was investigated and compared with the results of Artificial Neural Network (ANN) and Support Vector Machine (SVM) models. The results of M5Tree Model showed competition of 80 percent of data for training and the remaining for the testing has the best performance in presenting the flow duration curve with values of R2=0.992, RMSE=5.47 m3/s and MAE=4.38 m3/s. The results of different structures of Neural Network showed the best model (2 neurons for hidden layer) was obtained with values of R2=0.997, RMSE=3.91 m3/s and MAE=3.30 m3/s. Also the performance of RBF kernel of Support Vector Machine Showed this model has the best ability in simulation of flow duration curve, so that this model has lowest error values of RMSE=2.98 m3/s, MAE=2.66 m3/s and highest value of R2=0.998. Comparison the results between the intelligence models showed that each three models have proper performance in determining the discharge values of flow duration curve. From the practical view, M5Tree Model has more applicability in derivation of flow duration curve because of the simplicity of the proposed equations and calculations.

Keywords

• duration
• discharge flow
• aras river
• error value
• intelligent models