Abstract
The typical approach for estimating the average annual power produced by a WEC requires the use of a parametric spectra. The parametric spectra is fitted to measured wave spectra, e.g. from oceanographic buoys, and the resulting parameter space is divided into a number of sub-regions either through a power matrix, i.e. equally spaced bins, or through the use of clustering algorithms. A representative set of parameters (and therefor representative wave spectrum) is selected for each bin or cluster, and the WEC power is estimated by modeling the WEC dynamics. Finally the average annual power is estimated by summing all these representative mean power from the representative sea states, weighted by the probability of each bin or cluster, i.e. the proportion of measured sea states that fall within the bin or cluster. The use of significant wave height and either energy period or peak period as the two parameters for representing sea states is ubiquitous in practice for modeling WECs in realistic sea states. This is the case even as several authors have shown the inability of two parameters to represent the entire range of wave spectra present at WEC locations of interest. The variability seen by real wave spectra that have the same values for these two parameters is significantly beyond what can be attributed to random sampling, and therefor the base assumption of the underlying sea state being an ergodic Gaussian process is broken. This issue is particularly problematic when estimating the mean annual power, which requires evaluating the WEC's performance at all possible sea states. In this study we demonstrate the effect of spectra parametrization on the mean annual power estimate using a typical 2-parameter spectra, a 4-parameter semi-empirical spectra from literature, and a 4-parameter machine learning-based spectra developed here. The results show the failure of the 2-parameter representation and the ability of both 4-parameter spectra to be used for estimating mean annual power. The pros and cons of the different parametrization, as well as remaining open questions and areas of research are discussed.