High flowrate through gated tunnels may cause critical flow conditions, especially downstream of the regulating gates. Aeration is found to be the most effective and efficient way to prevent cavitation attack. Several experimental equations are presented to predict air demand in gated tunnels; however, they are restricted to particular model geometries and flow conditions and often provide differing results. In this study the current relationships are first evaluated, and then other approaches for air discharge estimation are investigated. Three machine learning techniques are compared based on the flow measurements of eight physical models, with scales ranging from 1:12–1:20, including the fuzzy inference system (FIS), the genetic fuzzy system (GFS), and the adaptive network-based fuzzy inference system (ANFIS). The Bayesian Model Average (BMA) method is then proposed as a tool to merge the simulations from all models. The BMA provides the weighted average of the predictions, by assigning weights to each model in a probabilistic approach. The application of the BMA is found to be useful for improving the design of hydraulic structures by combining different models and experimental equations.
Keywords: adaptive network-based fuzzy inference system, aeration, Bayesian model average, fuzzy logic, gated tunnel, genetic fuzzy system