Forecast Studies On Improving Short-term Wind Power Of Neural Nets Based On Regional Cluster Analysis

With the rapid development of science and technology, our country attached more importance to environmental protection which also brings about an increasing demand for clean energy. New energy industry grows up fast. Being a new way of power generation, the wind power’s accessing to the electrical power grid is apt to be the norm. The volatility and intermittency of wind power leads to the output fluctuation of wind farm. The operation security, economical dispatching and quality of the power are being challenged by this way.If the coming output of the wind farm could be accurately forecast in advance, related personnel could adjust the dispatching plan timely, to reduce the detrimental effects of the wind power integration to the stability of grid-connected wind power system. Consequently,the precise prediction of the output power of grid-connected wind power system could eliminate the effect of inherent nature of wind energy.In view of the above-mentioned facts, I choose the short-term wind power prediction with complex topographic as my object of study. In earlier work, we found the significant influence of space distribution, wind farm feature data, and the relativity of collection sample on the results. This article sifts through the specific data and extracts data which has great influence. It divides the wind farm into sections by method of clustering based on the distribution of wind turbine generator. This main aims of this paper are as follows:Firstly, divide the entire wind farm into sub domain: clustering environmental factors,operating conditions, reducing iterations and improving the efficiency of division;Secondly, use regional wind turbines power predict the wind power of all, regional wind turbine output is determined by the wind generator and sub- regional of greater relevance, the standard unit is stetted by the greater one;Thirdly, find the closest data of the output by studying one example of Guangxi wind farm, selecting standard unit to build MATLAB simulation model, collecting weather parameters from numerical weather prediction, obtaining the prediction of regional and the whole area power by building BP neural network;What’s more, chose the scientific forecast cycle by contrasting the predictive value of 8hours, 4 hours, 2 hours, and 1 hour with the measured values of traditional methods. And then obtaining the whole wind power forecast, carrying error index evaluation to verify reliability forecasting method in this paperLast but not least, analyze the shortage of the prediction method used in this paper,summarizes the existing problems and discusses the key points to improve the method.