پديد آورندگان :
اسمعيلي ، وركي نويسنده , , محمود اميدفمحمدحسين اميد، مترجم ,
كليدواژه :
stilling basin , Artificial neural network , Multilayer Perception , كشاورزي , پرش هيدروليكي واگرا , شبكه عصبي مصنوعي , Hydraulic jump
چكيده لاتين :
In this research an Artificial Neural Network (ANN), with multilayer perception structure, was adapted to model conjugate depth and gradually expanding jump length, which are especial but complex cases of hydraulic jumps. More than 3000 interpolated and experimental data on conjugate depths and jump lengths for both normal and gradually expanding jumps were used. The data was due to rectangular and trapezoidal sections, for a wide range of divergent angles and side wall slope. In developing the ANN models, seventeen configurations, each having a different number of hidden layers and/or neurons, were investigated. The optimal models were capable of predicting conjugate depth and jump length for a wide range of conditions. In each case, the configuration attained highest R^2 value was selected as the optimal model. For rectangular sections, the simplest ANN model had a 2-2-1 configuration, with one neuron in each of the two hidden layers, and R^2=0.97 (for normal x-section), and had a 4-1-1 configuration, with nine neurons in the hidden layer and R^2=0.91 (for gradually expanding x-section), respectively. The best ANN model for predicting respective jump lengths had 3-2-1 and 4-1-1 configurations with one and three neuron(s) in hidden layer(s), and R2 = 0.99 and 0.94, respectively. For trapezoidal sections, the simplest ANN model had a 4-2-1 configuration, with nine neurons in each of the hidden layers and R^2=0.99 (for normal x-section), and had a 5-2-1 configuration, with six neurons in each of the two hidden layers and R^2=0.94 (for gradually expanding x-section), respectively. The best ANN model for respective jump lengths had 4-2-1 and 5-2-1 configurations, with nine and six neurons in each of the two hidden layers, and R^2=0.90 and 0.85, respectively. The high values obtained for R^2 in all of the eight cases, suggest a close agreement between the ANNʹs output variable and the experimental data. The developed ANN models in this paper are, therefore, suitable for predicting gradually expanding hydraulic jump characteristics, which require a large amount of repetitive computations, for both rectangular and trapezoidal sections often encountered in the design of stilling basins.
