Channel junctions are one of the most important hydraulic structures that are constructed for diverting or consolidating the flow in open channel networks. However, these structures can complicate the analysis of flow characteristics. In the present study, the performance of an analytical model along with machine learning (ML) models in estimation of flow characteristics (discharge and water depth) in an open channel junction was investigated. For this purpose, ANFIS (adaptive neuro fuzzy inference system), and a hybrid model (ANFIS-GWO: ANFIS coupled with the gray wolf optimisation algorithm) were employed. Geometrical and hydraulic parameters were used as inputs of ML models. In terms of estimating outlet discharge from the main channel, both ML models demonstrated significantly better performance compared to the analytical model. In this regard, the hybrid ANFIS-GWO model achieved RMSE values of 0.8 and 0.78 for the training and testing sets, respectively, resulting in a reduction of estimation error by 58% and 49% compared to the analytical model. ANFIS-GWO was also more efficient at estimating outflow depth in the main channel, reducing the error rate by 45% and 21% compared to analytical model and ANFIS, respectively. The analytical model (RMSE = 0.79) did slightly better than ANFIS (RMSE = 0.9) in estimating outflow depth of tributary channel, but once again was outperformed by the hybrid model (RMSE = 0.73). The novelty of the present research lies in the determination of the flow diversion angle from the main inlet to the tributary outlet and vice versa, as well as in the pioneering application of both ANFIS and ANFIS-GWO for analyzing flow characteristics in a four-branch junction.
Hosseini et al. (Tue,) studied this question.