Neural network for design and reliability analysis of rubble mound breakwaters

Determination of Optimal Initial Weights of an Artificial Neural Network by Using the Harmony Search Algorithm: Application to Breakwater Armor Stones

“…They employed six input variables: P, N w , S, ξ m , h{H s , and the spectral shape parameter SS, where h is the water depth in front of the structure. Kim and Park [6] followed the approach of Mase et al [3] except that they used 641 data including low-crested structures. Since, in general, the predictive ability of an ANN is improved as the dimension of input variables increases, they split the surf similarity parameter into structure slope and wave steepness, and the wave steepness further into wave height and period.…”

“…Table 1 shows the correlation coefficients of previous studies, which will be compared with those of the present study later. Kim and Park [6] 0.902 to 0.952 641 Including data of low-crested structures Balas et al [9] 0.906 to 0.936 554 ANN-PCA hybrid models…”

“…Therefore, in general, the size of the training data is taken to be larger than that of the test data. In the previous studies of Mase et al [3] and Kim and Park [6], however, only 100 randomly sampled data were used for training the models, which is much smaller than the total number of data, 579 or 641. This might be one of the reasons why the ANN models do not show superior performance compared with the empirical formula (see Table 1).…”

“…Table 2 shows the input and output variables in the ANN model. The surf similarity parameter was split into cotα, H s , and T p as done by Kim and Park [6]. The peak period, T p , was used instead of T m because it contains the information about spectral peak as well as mean wave period.…”

“…Although the basic concept of artificial neurons was first proposed in 1943 [1], applications of ANNs have blossomed after the introduction of the back-propagation (BP) training algorithm for feedforward ANNs in 1986 [2], and the explosion in the capabilities of computers accelerated the employment of ANNs. The ANN models have also been used in various coastal and nearshore research [3][4][5][6][7][8][9][10].…”