Optimum Radial Basis Neural Network for open channel junction velocity prediction

Open channel junctions are commonly used in the practical situations. Junction’s downstream flow is comprises from two zones, the separation zone and the contraction zone. Therefore, the downstream zone has a complex 3D structure. In order to prevent from the erosion in the contraction zone and deposition in the separation zone, simulating the accurate flow field is one of the most important topics in designing the junctions. In this study, the longitudinal velocity of open channel junction is modeled using the Radial Basis Neural Network (RBNN) and the results are compared with the nonlinear regression method (REG). Modeling is done by using the 3D coordinates of the flow and the ratio of upstream to downstream discharge as input variables. Results showed that the RBNN model with Mean Squared Error (MSE) of 0.067, performs significantly better compare with the REG method with MSE of 0.118.

کلیدواژه ها:

Open channel junction ، Velocity ، Radial Basis Neural Network ، Regression method