Abstract

In recent years, the offshore wind turbine has played an important role in the development and utilization of offshore wind energy and brought considerable economic benefits. However, the marine environment is extremely harsh, and the design and construction cost of offshore wind power structures is high. Besides, the service economy of offshore wind power structure is often negatively correlated with the service time. If the operation and maintenance (O & M) cost of offshore wind power structure can not be effectively controlled, the lifecycle operation income of offshore wind power is difficult to guarantee. To reduce the O & M cost of the offshore wind turbine, a data-driven condition evaluation method is proposed here. The condition evaluation method starts with the modal identification of monitoring data, which is validated by cross-checking between different methods; then it uses the identified modal frequencies and modal damping ratio as dominant features to perform supervised classification. Three supervised learning algorithms (i.e. Decision Trees, k Nearest Neighbors, and Support Vector Machines) are used here. The results show that though the three methods all have the nonlinear capacity, they seem to need some improvement concerning their training and test performance. Further investigations should be conducted on whether the undesirable performance is due to a lack of training samples or the selected features.

Introduction

Wind power development is very active in recent years due to technological progress. However, the cost of wind project development and construction is still very high, so it is necessary to monitor and evaluate the condition of wind turbines to reduce downtime and optimize operation and maintenance as much as possible.

Most of the wind turbines are equipped with condition monitoring sensors after installation, which conducts intensive sampling for different signals, 24 hours a day and 365 days a year. The large amount of data generated brings challenges to data analysis.

This content is only available via PDF.
You can access this article if you purchase or spend a download.