Research On Related Issues Of Wind Cluster Output

With the rapid development of wind farm scalization and clustering, output fluctuations of wind cluster with stochastic volatility have negative influence to power grids’ secure and stable operation, dispatching and planning, and so on. As a consequence, researching on output fluctuation change rule of wind cluster is beneficial and valuable to regulate wind power integration effectively.Wind power fluctuation statistical characteristics of single wind farm and wind cluster will be analyzed and researched respectively based on time-amplitude domain and time-frequency domain in this paper. The main research includes the following several aspects:Firstly, overseas and domestic research status of wind power output volatility, output correlation and Hilbert-Huang transform application is illustrated. Then spacial-temporal characteristic of wind cluster output fluctuation from time-amplitude domain is studyed from three aspects, first of all, internal mechanism of wind farm output variability is deeply dug, and typical statistical indices of evaluating output variability in the time domain are presented. Secondly, the characteristics of time behavior and space disperse of single wind farm and wind cluster output are analyzed statistically. Thirdly, the probability characteristic of wind power variations is also studyed.Afterwards, Based on probability theory and mathematical statistics, central moments of wind power generation between a single wind farm and regional wind farms are established. Then two statistical indicators of skewness, kurtosis are introduced to depict the shape of wind power distribution. Relying on mean value, standard deviation, skewness and kurtosis, a distribution model of the Pearson family can be identified to represent the wind cluster generation. Then according to the given distances among wind farms and the average useable hours, an exponential relation between distances and correlations can derived empirically on the basis of analyzing typical wind cluster time series, the polynomial equations that allow the determination of these standard deviations from the wind power generation in the regarded region are also developed. At last, the practical application and analysis of wind clusters in Fujian province is given, it is testified that the Pearson modeling analysis based on four indices is accurate and practical for simulating and assessing output fluctuations of wind farm cluster.Thirdly, output correlation among wind cluster is analyzed from two different angles, one is the same time and the different space, the other is the different time and the same space. From the first point, cross correlation coefficients between two wind farms of wind cluster and autocorrelation coefficients among wind cluster are calculated, the characteristics of output cross correlation and autocorrelation are analyzed. Then from the second point, the time-sifting correlation and the probability distribution characteristic of cross correlation coefficients are researched.At last, empirical modes of wind cluster are analyzed, the Hilbert-Huang transform is first employed in analyzing the nonlinear and nonstationary time series signal of wind farm output in time-frequency domain, the basic theory of Hilbert-Huang transform is introduced. The basic principle of empirical mode decomposition, the concept of intrinsic mode functions, and the core algorithm are given. The core spectrum analysis indices of Hilbert spectrum analysis are proposed. Then through empirical mode analysis of annual output time series from single wind farm and wind cluster, their dominant frequency and sub-dominant frequency also are determined. What’s more, the distribution characteristic of output wave energy over frequency can be gotten. Finally, wind power short-term forecasting model based on empirical mode decomposition and Elman neural network is proposed, depending on the actual prediction example, through comparative analysis of calculated forcasting indices, it is found that compared with prediction model only by Elman neural network, combined forcast can gain higher prediction precision.