Forecasting of volumetric flow rate of Ergene river using machine learning

Abstract

Today, the significance of the estimation of physical parameters has considerably increased; for example, the prediction of water flow rate (WFR) is one of the types that will gain a substantial importance among the others. The predictions of WFR of rivers play a prominent role in the plans and constructions of new water dams, or to operate the ones that were formerly constructed. In this study, a variety of machine learning algorithms have been suggested in the estimations of one-ahead instantaneous measurement of river WFR. In this regard, four different prediction algorithms including long short-term memory (LSTM) neural network, adaptive neuro-fuzzy inference system (ANFIS) with fuzzy c-means (FCM), ANFIS with subtractive clustering (SC), and the ANFIS with grid partition (GP) were initially trained and secondly tested. The study region was selected as Inanli measurement station, which is located on the Ergene River. A cumulative of 134 different models were formed using these algorithms. Instantaneous WFR predictions in terms of WFR data have shown that ANFIS-FCM and ANFIS-SC had given the best statistical error outcomes. Accordingly, the values of 0.34 m3/s, 0.72 m3/s, and 0.9728 have been found as a result of machine learning establishment corresponding to statistical accuracy results of MAE, RMSE, and R values, respectively. Namely, it has been concluded that error values of the ANFIS-FCM and ANFIS-SC computations have outcome to be sufficiently good. Also, it was concluded and shown in the current study that both tools of the ANFIS can be two efficacious methods to WFR forecasting.