Automated Probabilistic Analysis and Parametric Modelling of the Seasonal-Diurnal Wind Vector

Abstract

A refined and extended version of the Offset Elliptical Normal mixture model has been developed to parameterise the seasonal diurnal wind vector automatically. Automated using R scripts, the method eliminates any potential risk of confirmation bias posed by the manual supervision in the original method. Refinements to the method include the latest algorithms for clustering of Gaussian mixtures, with Bayesian regularisation to set the number of components and to limit the predisposition to overfit. A new extension uses fuzzy logic to evaluate the probability distributions, autocovariances and spectra of the random perturbations around the mean seasonal-diurnal variations for each component of the mixture. These additional parameters allow the predictions of the OEN model to be validated and its automated application demonstrated using the hourly METAR reports of mean wind speeds at Adelaide, South Australia, showing significant improvements over the previously published analysis. The OEN mixture model is directly applicable to a wide range of wind engineering applications where seasonal and diurnal variation is of importance.