Velocity Field Prediction in Open-Channels Junction using Data Driven Models

The measurement of velocity in rivers confluences and open-channels junction is important in terms of hydraulic and environmental aspects. In this research, the performance of data driven models including ANN, ANFIS and GEP in estimating horizontal component of flow velocity in the open-channels junction was investigated using Khosravinia (2012) laboratory data. In the mentioned study, effects of a 45o side slope in the main channel on hydraulic characteristics of flow were investigated and compared with those at a 90o side slope. In this regard, the velocity field was measured for side slop angles of 45o and 90o using an ADV. For estimating the horizontal component of the flow velocity in the junction region, the discharge ratio and the dimensionless coordinates of measured points in three dimensional space of flow field were used. The performance of the models and comparison of their results were evaluated by coefficient of determination (R2) and root mean square error (RMSE). In addition to statistical indicators, for objective accuracy checking and performance of data driven models, scatterplot, box-plot, and Taylor diagram were used. Comparison of the results of different models using the best pattern indicates that the GEP model with the highest determination coefficient (R2 = 0.967), the lowest root mean square error (RMSE = 0.142) and the lowest mean absolute error (MAE = 0.094) in validation step has shown better performance than other models in (U/U0) estimation for the side slop angle of 45o. Similarly, the mentioned values were achieved 0.956, 0.184 and 0.128 using the GEP model for the side slop angle of 90o. This is while the ANFIS and ANN models ranked second and third. According to the scatterplots of the GEP model, nearly all the points are concentrated around the one-to-one line, which indicates the high level of predictive capability of this model in (U/U0) estimation for the both side slop angles. Also, according to the box plots, the statistical distribution of the GEP model in the lower and upper quartiles and median 50 percentile had a better performance than the other models, for the both side slop angles. The under and over estimation conditions of the ANFIS and ANN models were evident in these ranges. Moreover, according to the Taylor diagrams, the GEP model was closer to the observations and its superiority to the other models was tangible, with the lowest RMSE and the highest correlation coefficient for the both side slop angles. According to the results, GEP model as a powerful model can be used to replace the direct methods of velocity measurement in the junction region. In other words, using the mathematical equations derived from the GEP model in the present study, the longitudinal velocity field in the open-channels junction for side slopes of 45° and 90° can be predicted accurately.