Statistical Evaluation of Computational Intelligence Techniques for the Prediction of Hydrodynamic Performance of Quarter Circle Breakwater

Abstract

The development of a country generally depends on its per capita power consumption and hence its trade. Coastal areas are vital for developing Ports and harbours, which are to be planned and prepared properly for safely handling ships and boats for efficient trade by constructing proper protective structures. Breakwater is one such structure constructed in ocean water to attenuate the wave energy to create a tranquil region on its lee side. With the increasing demand for coastal development and the increase in the cost of construction of the conventional type of breakwater, innovative breakwaters have been developed in the last two decades. The various coastal developmental proposal, such as the size and length of the breakwater, its orientation, and sediment studies, were used to be studied in a laboratory before their implementation in arriving at an optimal solution. These laboratory studies are not only involved huge costs but also time-consuming. The advancement both in computing methods and enhancement of knowledge in using artificial intelligence technology is being applied in various fields of application and has encouraged even to tackle complex non-linear problems in coastal engineering. The thesis aims to evaluate various computational intelligence techniques to predict the hydrodynamic performance of quarter circle breakwater subjected to wave–structure interaction with perforation on its seaside. Computational methods such as ANN, ANFIS, SVM, and Deep learning are trained and tested for non-dimensional input parameters obtained from earlier Physical model studies for four different perforation conditions viz., S/D=2. 3, 4, and 5. The data is segregated into 70% and 30% for training and testing the network. In the case of the ANN model Levenberg Marquardt and Conjugate gradient algorithms were considered. For ANFIS, Gambel’s and triangular memberships were considered. For the SVM model, Gaussian and Sigmoidal kernels were considered, and in deep learning, Decision tree and Deep forests were studied. Model performance Index from the statistical study is adopted for evaluating the performance of each model and suggesting the best model for the prediction. The experimental study is very robust and efficient in handling issues involving structures having arbitrary configurations in finite water depths. The physical problems are analyzed in the context of linear water wave theory. The role of several physical and numerical parameters associated with wave scattering and trapping are investigated by non-perforated and perforated ii caisson structures. In addition, an attempt is made to study the effect of slotted barriers on the hydrodynamic performance of the caisson breakwater by introducing the horizontal and vertical slotted barriers in front of the caisson breakwater. Both the cases of horizontal and vertical structures are considered in different cases for handling problems of varied configurations. The numerical convergence of the solution is also analyzed using finite volume code Fluent. The numerical results are validated against the existing literature and experimental results for most of the physical problems considered in this thesis.