Abstract
Offshore wind resource assessment requires in-situ wind observation of at least one year. On the other hand, at initial stages of an offshore wind project, it would be useful if annual wind conditions could be estimated based on shorter-term data and used for initial feasibility studies as soon as observation data are collected at the site. This study tried to estimate annual wind conditions from several months of wind observations by using the MCP (Measure-Correlate-Predict) method. The MCP methods tested in this study include those based on linear regression and machine learning. As a result, it was found that the linear regression (No wind direction classification) and the Random Forest difference model have the highest estimation accuracy. For short observation periods of less than six months, it was found that the season of the observation greatly influence estimation accuracy and the accuracy can be worse than the estimation without MCP. In the estimation using the Random Forest, it was found that the estimation accuracy can be greatly improved by constructing a model that uses wind speed differences as the training data, instead of directly estimating wind speeds.
