Wave transmission at various types of low-crested structures using neural networks

Abstract

Wave transmission is often an important criterion in the design of a low-crested breakwater structure and influences early decisions on the type of structure and the applied construction material. In design stages classical formulae are used to predict the wave transmission coefficient Kt [-], but these formulae still show a considerable scatter, probably due to a limited number of parameters included. Neural networks are mathematical devices that have proven to be powerful tools in many fields of technology and can be used as prediction method for complex cases. During this study, neural networks have been used to find relations for parameters influencing wave transmission. Finally, a prediction model is made with an ensemble of neural networks, based on a database consisting of 3,934 transmission tests from test facilities from all over the world for various types of low-crested structures. An extensive database is composed, describing all specific hydraulic and structural parameters for each single test in a homogeneous way. In total 9 governing parameters have found to be important for obtaining a reliable prediction of the wave transmission coefficient. The present prediction model shows significantly improved predictions of the wave transmission coefficient compared to the existing empirical formulae. The prediction model has proven to be capable of handling both mound (rough) and smooth structures within one model, although these structures are found to behave completely different on wave trans-mission.