Wind site turbulence de‐trending using statistical moments: Evaluating existing methods and introducing a Gaussian process regression approach

Abstract This paper considers the problem of retrospectively de‐trending wind site data when only statistical moments, in form 10‐min means and standard deviations speed, are available. Low‐frequency trends present speed known to bias fatigue damage estimates, and, hence, removal their influence is important for accurate life estimation. When raw available, this procedure straightforward; however, many sites, significant quantities which contain moments. Additional value therefore unlocked if can also take place context. Existing methods, Models 1 2, introduced, performance viability appraised, respectively. A Gaussian process (GP) regression implementation developed, seeks incorporate characteristics real extracted from into fitting via an appropriately chosen lengthscale hyperparameter. Results indicate that Model recommended method previous work, suffers fundamental issues, with implication it should no longer be used. GP results shown very similar at turbulence distribution level. finding interpreted as a validation indication may already performing well hoped for, given information available current formulation. Theoretical overheads associated GPs, addition similarities mentioned above, lead being best approach moment‐based time.