User-Friendly Reliable ANN Model for Stability Number of Rock Armors and Tetrapods

Abstract

The stability number of rubble mound breakwaters determines the appropriate weight of armor units of concrete or rock required to resist the wave condition. Therefore, the prediction of suitable stability number is necessary for the stability of the breakwaters. Many empirical formulas have been developed for the stability number since Hudson (1959). To improve the empirical formulas which had significant differences between observed data and prediction data, the machine learning, ANN in particular, has been used during the last two decades. However, most of ANN models did not deal with reliability assessment such as confidence interval. In addition, they are seldom used by practicing engineers probably because most of them did not provide them with an explicit calculation method. In this study, to solve these problems, bootstrap resampling technique was used to make the information or assessment of the reliability in prediction. Also, Excel files made with the by-products of the ANN model such as weights and biases are provided, so that practicing engineers can easily use ANN model.