Predicting scour around offshore wind turbines using soft computing techniques - Comparing Genetic Programming with Existing Scour Prediction Methods.

摘要

In this report a prediction method is developed for scour around monopiles. A soft computing technique called genetic programming (GP) is used to create a scour prediction formula that can compute scour in all offshore conditions, meaning current-induced, wave-induced and combined current- and wave-induced scour. The GP was trained with an extensive database of laboratory scour measurements from multiple sources, to ensure that a wide range of conditions was represented. Furthermore, only dimensionless parameters were used to create a formula that is also applicable for field tests. The formulas where analyzed both on their mathematical and physical behavior and it was concluded that they could accurately predict scour in all conditions. The new scour prediction method was compared to various existing scour prediction methods and it was seen that the formula created in this study predicted more accurate scour depths, especially for test with larger scour depths. The study was finalized with a comparison to a second soft computing method: the neural network. It was found that the GP is less successful in predicting the scour depth compared to the NN. However, the high accuracy of the NN could not have been achieved without the knowledge of the parameter behavior obtained by the GP.