New formulation for predicting total dissolved gas supersaturation in dam reservoir: application of hybrid artificial intelligence models based on multiple signal decomposition

• DOI: 10.1007/s10462-024-10707-4
• Publication Date: 2024-03-09T08:01:16Z
• Authors: Salim Heddam; Ahmed M. Al-Areeq; Mou Leong Tan; Iman Ahmadianfar; Bijay Halder; Vahdettin Demir; Huseyin Cagan Kilinc; Sani I. Abba; Atheer Y. Oudah; Zaher Mundher Yaseen
• Keywords: Perceptron; Multilayer perceptron

Abstract Total dissolved gas (TDG) concentration plays an important role in the control of aquatic life. Elevated TDG can cause gas-bubble trauma fish (GBT). Therefore, controlling fluctuation has become great importance for different disciplines surface water environmental engineering.. Nowadays, direct estimation is expensive and time-consuming. Hence, this work proposes a new modelling framework predicting based on integration machine learning (ML) models multiresolution signal decomposition. The proposed ML were trained validated using hourly data obtained from four stations at United States Geological Survey. dataset are composed from: ( i ) temperature T w ), ii barometric pressure BP iii discharge Q which used as input variables prediction. strategy conducted two steps. First, six singles model namely: multilayer perceptron neural network, Gaussian process regression, random forest iv vector functional link, v adaptive boosting, vi Bootstrap aggregating (Bagging), developed , their performances compared. Second, was introduced combination empirical mode decomposition (EMD), variational (VMD), wavelet transform (EWT) preprocessing algorithms with building hybrid models. signals decomposed to extract intrinsic functions (IMFs) by EMD VMD methods analysis (MRA) components EWT method. Then after, IMFs MRA selected regraded integral part thereof. single prediction compared several statistical metrics namely, root mean square error, absolute coefficient determination R 2 Nash–Sutcliffe efficiency (NSE). times high number repetitions, depending kind modeling process. results gave good agreement between predicted situ measured dataset. Overall, Bagging performed better than other five NSE values 0.906 0.902, respectively. However, extracted EMD, have contributed improvement models’ performances, significantly increased reaching 0.996 0.995. Experimental showed superiority more importantly improving predictive accuracy TDG. Graphical abstract