Abstract
In this study, an artificial neural network (ANN) model is developed to predict the stability number of breakwater armor stones based on the experimental data reported by Van der Meer in 1988. The harmony search (HS) algorithm is used to determine the near-global optimal initial weights in the training of the model. The stratified sampling is used to sample the training data. A total of 25 HS-ANN hybrid models are tested with different combinations of HS algorithm parameters. The HS-ANN models are compared with the conventional ANN model, which uses a Monte Carlo simulation to determine the initial weights. Each model is run 50 times and the statistical analyses are conducted for the model results. The present models using stratified sampling are shown to be more accurate than those of previous studies. The statistical analyses for the model results show that the HS-ANN model with proper values of HS algorithm parameters can give much better and more stable prediction than the conventional ANN model.
Highlights
Artificial neural network (ANN) models have been widely used for prediction and forecast in various areas including finance, medicine, power generation, water resources and environmental sciences
Both models were run 50 times, and the statistical analyses were conducted for the model results
Lee [31] compared the computational time between the conventional ANN model and models with various combinations of HMCRofand
Summary
Artificial neural network (ANN) models have been widely used for prediction and forecast in various areas including finance, medicine, power generation, water resources and environmental sciences. The Pollack [15] demonstrated generally uses a gradient descent algorithm [11] It can give a local minimum of that the BP training algorithm has large dependency on the initial weights by performing value a Monte the error function as shown in Figure and it is sensitive to thebased initialon weights and biases. ANNs and tothe improve model accuracy methods have other hand, as expected in most cases, if they are selected to be far from the optimal values as shown been implemented to enhance the accuracy of the model They are shown to overcome the by ‘Start’ in Figure the final destination would be the weights local minimum markedalgorithms by ‘End’ inand the dependency of the 1, ANN model on the initial but alsothat on is training figure. Data structure [20,21,22,23]
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
