Makine Öğrenmesi İle Aksu Deresi’nde Akış Tahmin Modeli Geliştirilmesi

Turkish

In this study, future runoff modeling has been made for the Aksu Stream sub-basin in the Eastern Black Sea Basin, where a major flood disaster occurred in 2020. Runoff data obtained from the State Hydraulic Works, precipitation and vapor pressure data obtained from the General Directorate of Meteorology were used for modeling. Artificial Neural Networks (ANN) and Classical Regression (CR) methods were preferred for the runoff model to be created. 14 different input models were built by using vapor pressure, precipitation and historical discharge data. These input models created were tested with the runoff prediction model created by the Multi Layered ANN (ML-ANN). The estimation performances of the runoff prediction models were determined using the Root Mean Square Error (RMSE), Correlation Coefficient (r), Relative Error (RE), Nash-Sutcliffe Coefficient (E) and Mean Absolute Error (MAE) criteria and their performance in the ML-ANN model was determined. The highest input model was tested with the Classic Multiple Regression (CMR) method. The model with the highest runoff estimation performance in the region was the model operated with the M10 input set created with ML-ANN. In the study, the estimated future runoff rates were evaluated according to the exceedance probabilities used to determine the project runoff rate in the design phase of structures such as flood protection facilities, hydroelectric power plant facilities, treatment facilities. As a result, it has been determined that the use of day-delayed input set increases the performance in the CMR method, as in the models that perform machine learning, the ML-ANN method is more successful than the CMR method for the runoff estimation modeling in the region, and it is suitable for determining the project runoff rate.