Assessment of GEP and ANN for Predicting Suspended Sediment Load: A Case Study of Ghatoor and Aland Rivers, West Azerbaijan, Iran

Document Type : Original Article

Authors

1 Department of Civil Engineering, Faculty of Civil Engineering, University of Tehran, Iran

2 Water Resource Engineering, West Azerbaijan Regional Water Authority, Urmia, Iran

3 Department of Soil Science, Faculty of Agriculture, Isfahan University of Technology

4 Master of Science in Power Engineering, Khazar Holding, Mashin Sazi Khorram Abad, Tehran, Iran

10.22044/jhwe.2023.13552.1028

Abstract

Estimation of the volume of suspended sediment load of rivers, especially when dam constructed on it, is one of the tremendous challenges that civil engineers faced. It is crucial to accurately predict the suspended sediment load to effectively mitigate the negative consequences of this phenomenon. To estimate the total suspended sediment accumulated behind the Aland and Ghatoor dams, two models of artificial intelligence, Gene Expression Programming (GEP) and Artificial Neural Network (ANN), were employed in this study. The performance of these two AI models compared with the traditional method, Sediment Rating Curve (SRC), for estimating the suspended sediment volume using hydrometric stations from 1969 to 2017. Unfortunately, the appropriate data from 2017 to the present is not available from authorities of the West Azerbaijan province, so inevitably, we used the hydrologic records till the end of the year 2017 in this article. Two statistical indices were used to evaluate the models: the coefficient of determination (R-squared) and the Mean Absolute Error (MAE). Based on these indices, the intelligent models performed better than the SRC in estimating the suspended sediment volume. In comparing the GEP and ANN models, the performance criteria show that the ANN model produces better results. For the Ghatoor River, the performance indicators of the ANN model were MAE=993.1 ton/day and R^2=0.910, which is 45% and 43% higher than the GEP model and SRC method, respectively. For Aland River, the performance indicators of the ANN model were MAE=519.2 ton/day and R^2=0.961, which is 12% and 57% higher than the GEP model and SRC method, respectively. In conclusion, for predicting the suspended sediment load in Ghatoor and Aland Rivers, the ANN model can be the best choice for this purpose.

Keywords