Özet:

In the detection of levels of pollution in the streams, the supply of use and drinking water, it is very important that the water structures such as the bar and the connectivity are properly detected in the project phase of the sediment load. This study tested methods such as an adaptable network-based fool output system (ANFIS), artificial nerve networks (YSA) and multi-linear regression (MLR) for the three current observation stations (GI) located on the Euphrates Bay. When selecting the group numbers of the ANFIS models created, the K-means aggregation algorithm was used. The study used sediment (Qs), rain (P), debi (Q) and temperature (P) data for each station. With these data, the sediment prediction model for each station has been developed. In the model created the input variable is the daily value (P) where the rain occurs, the value one day before the day when the rain occurs (P-1), the debi and temperature values, and the output variable is used as the sediment concentration. This model is compared for all stations in both training and test stages, respectively, in terms of regression rate (R2) and average percentage error (OYH). The analysis found that the K-means aggregation algorithm and the number of subgroups determined the ANFIS model had both the number of subgroups found to be more successful than the randomly created ANFIS models and the YSA and MLR models. It has also been shown that the YSA and ANFIS methods have achieved results closer to the observed values compared to the MLR method model.

Anahtar Kelimeler: