عنوان مجله
|
Journal of Hydraulic Structures
|
چکیده
|
In the present study, an artificial neural network and its combination with wavelet theory are
used as the computational tool to predict the depth of local scouring from the bridge pier. The
five variables measured are the pier diameter of the bridge, the critical and the average
velocities, the average diameter of the bed aggregates, and the flow depth. In this study, the
neural wavelet method is used as a preprocessor. The data was passed through the wavelet filter
and then passed to the artificial neural network. Among the various wavelet functions used for
preprocessing, the dmey function results in the highest correlation coefficient and the lowest
RMSE and is more efficient than other functions. In the wavelet-neural network compilation
method, the neural network activator function was replaced by different wavelet functions. The
results show that the neural network method with the Polywog4 wavelet activator function with a
correlation coefficient of 87% is an improvement of 8.75% compared to the normal neural
network model. By performing data filtering by wavelet and using the resulting coefficients in
the neural network, the resulting correlation coefficient is 82%, only a 2.5% improvement
compared to the normal neural network. By analyzing the results obtained from neural network
methods, the wavelet-neural network predicted errors compared to experimental observations
were 8.26, 1.56, and 1.24%, respectively. According to the evaluation criteria, combination of
the best effective hydraulic parameters, the combination of wavelet function and neural network
and the number of neural network neurons achieved the best results.
|