Research On The Simulating Model Of Wind Farm Based On Measured Operating Data

An accurate simulating model of wind farm is demanded for researches about the impacts of large scale wind generation on the power grid. Influenced by the park effect and wake effect, the operational points of all wind turbines in the wind farm are not the same and the one machine equivalent model is not applicable. In order to aggregate wind turbines in complex terrain or irregular layout, the clustering approaches for wind farm is proposed based on the measured operating data. The main content of this thesis is as follows:(1) Combined with the operation principle of wind farm, its mathematical model is established, which including wind speed model, aerodynamics model, shaft system model, pitch angle control model, generator model and control system model.(2) In order to aggregate wind turbines in complex terrain or irregular layout, the clustering approach for wind farm based on the K-means clustering algorithm for wind farm based on immune-outlier data and immune-sensitive initial center is proposed. Firstly, considering the fact that the outliers are existed in the measured operating data of wind turbines, the data processing is performed to immune the outliers. Secondly, the traditional K-means algorithm is sensitive to the initial center and is easily getting to the trap of a local solution. To immune the sensitiveness, the improved max-min distance means is adopted to benefit the optimization of initial cluster centers. Finally, an actual wind farm is employed to the clustering and the dynamic simulation for the wind farm is performed. Results show that the equivalent model proposed in this paper show a high degree of correspondence with the detail model, both for gust disturbance and for short circuit. The proposed clustering method for wind farm is effective for the study of wind power system.(3) The wind turbine grouping method for wind farm dynamic equivalence based on semi-supervised split hierarchy spectral clustering algorithm is proposed. K-means clustering algorithm is restricted to normally distributed clusters and the clusters must be convex datasets. If the distribution of the dataset isn’t convex, the clustering result may fall into local optimum. Aiming at this issue, firstly, the renormalized eigenvector matrix is formed. The matrix can reflect the initial sample space of the measured wind speed data. Even more, it provides more useful information for the clustering. With the available prior information of some samples, the sample groups in the matrix are divided into several clusters based on semi-supervised split hierarchy spectral clustering algorithm. Then the capacity weighting method is applied to calculate the parameters of the equivalent WTG for each cluster. Finally, to test and verify the proposed method, an actual wind farm is used for the clustering. Results show that the dynamic equivalence method for wind farm proposed in this paper is effective for the study of wind power system.(4) An agglomerative hierarchical aggregating method for WTG clusters based on information entropy is explored. If the WTG clusters are too many to hinder the simulation of wind farm, it is necessary to aggregate the clusters. In the process of clusters aggregation, the definition and property of information entropy is taken advantage for the database mining. The information entropy is adopted to evaluate clustering quality, which is fed to the upper database to improve the clusters aggregation.