Optimizing Aeration Efficiency Prediction in High-Head Sluice-Gated Conduits: Data-Driven Sensitivity and Uncertainty Analysis

Abstract

This research investigates aeration efficiency AE(20) in high-head sluice-gated conduits, focusing on conduit structure/flow characteristics. The study incorporates dimensional variables such as the aspect ratio (alpha), gate opening degree (& Oslash;), gate opening height (h(o)), conduit length (L), cross-sectional water flow area (A(w)), water velocity at the gate location (V), and air and water discharge rates (Q(air )and Q(water)), alongside nondimensional parameters including alpha, & Oslash;, L/h(o), A(w)/h(o)(2), Q(air)/h(o)(2)V, the Froude number (F), and the air-demand ratio (Q(air )/Q(water)). To forecast AE(20) , the machine learning models deep neural networks (DNNs), gradient boosting machines, random forest, Gaussian process regression, and support vector regression were applied and compared with conventional models. Our evaluation showed that the DNN model delivers superior accuracy across all types of data. However, we did not stop there; we also put its dependability to the test using rigorous uncertainty methods such as Monte Carlo simulations and the interval approach. These tests confirmed the model's reliability, proving it is ready for real-world use in optimizing conduit design. Furthermore, by using several sensitivity techniques (Morris, Sobol, correlation, Shapley, etc.), we consistently identified V as the most important factor for dimensional data and F for nondimensional data.