Research On The Model Of The Output Power Of Wind Farms Based On Cluster Analysis

As the problem of energy and environmental more and more seriously day bay day, renewable energy has become a major issue researched by the world. Among them, the wind power technology is one of which are growing fastest and the most mature. However, it has the shortcomings of intermittent and volatility and so on. With the installed capacity increasing rapidly and connected to the gird, the proportion fo wind power in the gird continues to improve, wind power connected to the grid will definitely bring range of issues such as system stability, power quality and power system scheduling.The key problem to address the difficulties brought by wind power connected to the grid is that we can accuratly forecasting the output power of the wind farm. Although the issue to research wind power technology continued to deepen, but, the reaearch concerned wind power prediction has not yet reach a very satisfied level. Under this context, choose the short-term forecasting as the study of this paper, proposed a model to forecasting the output power basede on cluster analysis.When established the forecasting model, firstly describe the principles and specific steps of dynamic clustering algorithm in this paper, then introduced the principles and applications of SOM network and the principles and processess of radial basis function (RBF) network and genetic algorithm, finally, according to the impacts which effect output power, presented the model based on cluster analysis.In the numerical example, firstly, using the classical method called time series as a forecast of the output power, then according to chose the K-means algorithm or dterative self-organizing datat analysis algorithm when making categories, the different dimensions of input parameters of RBF network, whether to adopt the actual power as the previous time in the one-dimensional feature vector parameters respectively making study using examples, finally, using combinationed model doing research.In this paper, doing comparative analysis from the aspects of average error, mean square error, the means of relative error and convergence time first, and then proposed two methods to evaluation the forecasting models—posterior difference test, comprehensive evaluation method.