Caculation of the hydraulic jump length in horizontal rectangular channels using artificial neural network models

Tóm tắt

This study aims to develop artificial neural network (ANN) models that are effective and simple in forecasting the length of a hydraulic jump in a horizontal rectangular canal. This study also recommends applying the Buckingham Pi theorem to ascertain four inputs and one output of the ANN models, corresponding to five dimensionless Pi terms. The effect of the channel roughness, fluid viscosity, and gravity has been considered to improve the model's performance. The ten models, each containing an input layer with four neurons, an output layer with one neuron, and one or two hidden layers, were meticulously developed and evaluated using the four statistical indicators MAE, RMSE, NSE, and MAPE. According to the test results and comparison with empirical equations, all models perform better than empirical equations in forecasting the hydraulic jump length. The study's findings show that the number of neurons per hidden layer should be between 16 and 32. The best model comprises two hidden layers, each containing sixteen neurons. This model's NSE value reaches 0.9905 during testing. Two-hidden-layer models require fewer epochs but are more efficient than one-hidden-layer models. Consequently, when the inflow Froude number ranges from 1.73 to 19.67, the proposed ANN models might be utilized instead of empirical formulae for determining the hydraulic jump length. Keywords: ANN, hydraulic jump, machine learning, model