Estimation on River Water Flow Rate Using Long Short-term Memory (LSTM) Neural Network and Adaptive Neuro-fuzzy Inference Systems (ANFIS)

Authors : Akın İlhan

Pages : 20-42

View : 39 | Download : 25

Publication Date : 2025-04-30

Article Type : Research Paper

Abstract :The forecasting of river water flow rate (RWFR) plays a prominent role in planning and constructing of new hydraulic dams, or running the ones that were formerly built. This study suggests algorithms of machine learning to predict future water flow rate values for river flow. Namely, estimation models were advanced according to the past time-series RWFR to obtain future RWRF values. Accordingly, long short-term memory (LSTM) neural network, adaptive neuro-fuzzy inference system (ANFIS) with fuzzy c-means (FCM), were advanced for the aim of RWFR predictions. A measurement station (MS), settled at the border of Türkiye and Bulgaria, named as Svilengrad MS was selected on the Maritsa River, as the study region. Accordingly, it was concluded that FCM algorithm of ANFIS have generated better results compared with respect to the LSTM algorithm. The comparisons of the data estimations according to the real observed water flow values were accomplished depending on the statistical error values including mean absolute error (MAE), root mean square error (RMSE), else the correlation coefficient (R). Eventually, it was concluded and shown that the superior model of FCM have generated those statistical accuracy values, respectively to correspond 3.13 m3/s MAE, 4.90 m3/s RMSE, and 0.9978 R, among the total of 49 tested models using FCM and LSTM.
Keywords : Uyarlanabilir Nöro Bulanık Çıkarım Sistemi (ANFIS), Bulanık C-ortalamaları (FCM), Uzun Kısa Süreli Bellek (LSTM)