APPLICATION OF BAYESIAN ANALYSIS FOR WIND ENERGY SITE EVALUATIO

A common engineering situation is one in which the underlying probability distribution of a basic random variable is known with a high degree of certainty, but whose parameters are not. Bayesian analysis applied to such a situation treats the parameters as random variables with a prior distribution that is specified based on subjective opinion and previous experience. Sampling data are then combined systematically using Bayes' theorem to update the parameter distributions as well as the probabilistic assessments of the basic random variable.;The analysis of site wind characteristics is necessary for selection and assessment of sites for wind energy conversion systems and is an example of such a situation. The stochastic nature of the wind necessitates the use of statistical methods in analysis. Classical statistical methods have been employed at sites where long term data are available. Since little or no information may be available at a potential site, preliminary assessment can best be carried out using the Bayesian approach to statistical analysis.;Extensive research has indicated that the wind speed is well modelled by the single-parameter Rayleigh distribution. The site mean wind speed is given a prior distribution obtained from historical National Weather Service data of sites in the region. The prior distribution is updated to a posterior each time observed data become available. The posterior distribution is combined with the Rayleigh distribution to obtain the Bayesian distribution of wind speed. This distribution can now be used in conjunction with the turbine power characteristics of different machines to make rational decisions, based on the expected maximum utility criterion, for selecting the best turbine for a given site or the best site for a given turbine.;The Bayesian approach is also used in the analysis of wind persistence at the site. Using the single parameter (power-exponential) distribution derived for durations that the wind remains above or below a fixed speed, as well as information on the mean wind speed, a prior distribution is obtained for the probability of occurrence of a specific event of interest, such as a duration above a given level exceeding a particular length. This probability is then treated as a Bernoulli random variable which can be updated from observed run duration data at the site or from improved estimates of the mean wind speed.