Estimation of daily suspended sediment load using a new hybrid artificial neural network model combined with observer-teacher-learner- based- optimization method

siyamak doroudi ¹ ( Department of Civil Engineering, Science and Research Branch, Islamic Azad University, Tehran, Iran )
ahmad sharafati ² ( Department of Civil Engineering, Science and Research Branch, Islamic Azad University, Tehran, Iran )

Submited date : 2023-06-03 Accepted date : 2024-02-14

Keywords: Artificial neural network, suspended sediment load, optimization algorithm,

Abstract :

Introduction: Suspended sediment load (SSL) is one of the complex hydrological phenomena, and its prediction is difficult. This study uses the artificial neural network method to predict suspended sediment load. Since the accuracy of artificial neural networks depends on their parameters, the benefit of meta-heuristic algorithms can be effective in increasing their performance. The case study is the catchment area of the Kosar Dam located in the southwest of Iran.

Methods: River discharge and rainfall were considered as inputs, features for predicting models. Five input compounds were selected. OTLBO and PSO meta-heuristic algorithms were used to find the optimal ANN values, and ANN-OTLBO and ANN-PSO prediction models were developed. Predicting models were evaluated using different numerical and visual indicators.

Findings: The results show that the ANN-OTLBO model provides higher prediction performance than other models used in this study. Specifically, the ANN-OTLBO-M5 model shows superior values (R=0.96358, RMSE=258.14, PBIAS=2.6752, and NSE=0.92674). Also, based on the scatter plot, heat map, and box plot, the closest predicted data to the observed data belongs to the ANN-OTLBO-M5 model.

Conclusion: Among all ANN, ANN-OTLBO, and ANN-PSO models, the ANN-OTLBO model has the best performance. The mentioned model can accurately predict high, medium, and low amounts of sediment. Among all ANN, ANN-OTLBO, and ANN-PSO models, the ANN-OTLBO model has the best performance. The mentioned model can accurately predict high, medium, and low amounts of sediment. In addition, the combination of meta-heuristic algorithms with artificial neural networks increases the accuracy of sediment prediction.