In this study we develop and apply new methods of data analysis for high resolution wind power and system load time series, to improve our understanding of how to characterize highly variable wind power output and the correlations between wind power and load. These methods are applied to wind and load data from the ERCOT region, and wind power output from the PJM and NYISO areas. We use a wavelet transform to apply mathematically well-defined operations of smoothing and differencing to the time series data. This approach produces a set of time series of the changes in wind power and load (or "deltas"), over a range of times scales from a few seconds to approximately one hour. A number of statistical measures of these time series are calculated. We present sample distributions, and devise a method for fitting the empirical distribution shape in the tails. We also evaluate the degree of serial correlation, and linear correlation between wind and load. Our examination of the data shows clearly that the deltas do not follow a Gaussian shape; the distribution is exponential near the center and appears to follow a power law for larger fluctuations. Gaussian distributions are frequently used in modeling studies. These are likely to over-estimate the probability of small to moderate deviations. This in turn may lead to an over-estimation of the additional reserve requirement (hence the cost) for high penetration of wind. The Gaussian assumption provides no meaningful information about the real likelihood of large fluctuations. The possibility of a power law distribution is interesting because it suggests that the distribution shape for of wind power fluctuations may become independent of system size for large enough systems.

%I LBNL %C Berkeley %8 12/2010 %2 LBNL-4147E