A Multi-Functional Genetic Algorithm-Neural Network Model for Predicting Suspended Sediment Loads

• Published in: Water resources management Vol. 39; no. 5; pp. 2033 - 2048
• Main Authors: Afan, Haitham Abdulmohsin, Wan Mohtar, Wan Hanna Melini, Aksoy, Muammer, Ahmed, Ali Najah, Khaleel, Faidhalrahman, Khan, Md Munir Hayet, Kamel, Ammar Hatem, Sherif, Mohsen, El-Shafie, Ahmed
• Format: Journal Article
• Language: English
• Published: Dordrecht Springer Netherlands 01.03.2025; Springer Nature B.V
• Subjects: Accuracy; Algorithms; Atmospheric Sciences; Civil Engineering; Drought; Earth and Environmental Science; Earth Sciences; Environment; Genetic algorithms; Geotechnical Engineering & Applied Earth Sciences; Hydrogeology; Hydrology/Water Resources; Neural networks; Optimization; prediction; River discharge; River flow; Rivers; Root-mean-square errors; Sediment; sediment contamination; Sediment load; Sediments; stakeholders; Statistical analysis; Suspended load; suspended sediment; Suspended sediments; system optimization; water; Water discharge; Sediment Load Prediction; Neural Network; Johor River; Artificial Intelligence; Genetic Algorithm
• ISSN: 0920-4741; 1573-1650
• DOI: 10.1007/s11269-024-04054-w

The ability to accurately predict suspended sediment load (SSL) in a river is vital for various stakeholders. Predicting SSL can help inform efforts to reduce the negative impacts of floods and droughts and help inform mitigation efforts for extreme environmental events that have a significant impact on the availability of clean water. In this regard, this study proposes a multifunctional Genetic Algorithm-Neural Network (GA-NN) model to predict the SSL using flow discharge and SSL data at Johor River. Furthermore, a comparison study was conducted between the results obtained with the proposed model and with traditional input selection, as well as another optimization method (GHS algorithm). The findings of this study indicate that the GA-NN model is a proficient instrument for forecasting Suspended Sediment Load (SSL) utilizing river discharge and sediment load data from the Johor River. Furthermore, a superior improvement in prediction accuracy was achieved using the GA algorithm, compared to the traditional input selection and GHS algorithm. Based on several statistical matrices and graphical appraisals, the optimum results were achieved within five inputs by providing low margins of errors in terms of Mean Absolute Error (MAE) of 14.366 and Root Mean Square Error (RMSE) of 24.560 and higher correlation accuracy in terms of coefficient of determination (R 2 ) of 0.911. Thus, the Genetic Algorithm (GA) proved its ability to select input patterns, which is considered a critical step in modeling, as it helps to simplify the process of finding the optimal solution to obtain more accurate predictions.