The amount of sand moving parallel to a coastline forms a prerequisite formany harbor design projects. Such information is currently obtained throughvarious empirical formulae. Despite so many works in the past an accurateand reliable estimation of the rate of sand drift has still remained as aproblem. The current study addresses this issue through the use ofartificial neural networks (ANN). Feed forward networks were developed topredict the sand drift from a variety of causative variables. The bestnetwork was selected after trying out many alternatives. In order to improvethe accuracy further its outcome was used to develop another network. Suchsimple two-stage training yielded most satisfactory results. An equationcombining the network and a non-linear regression is presented for quickfield usage. An attempt was made to see how both ANN and statisticalregression differ in processing the input information. The network wasvalidated by confirming its consistency with underlying physical process.
展开▼
