风电场输出功率的预测对降低风电场运行成本及合理安排电力系统调度具有重要意义。提高风能功率的预测精度，可以有效地减轻风电对电网的影响，同时提高风电场在电力市场中的竞争能力。在建立风能短期功率预测模型时，由于样本的选取对预测的精度有较大的影响，因此研究样本的选取方法具有重要意义。本文提出了一种利用K-均值算法对历史功率数据样本进行聚类，根据得到的聚类结果训练学习向量量化(LVQ)神经网络，利用训练好的神经网络对待预测数据进行自动分类，最后使用最小二乘法建立风能功率预测模型的方法。通过实验验证了该方法的有效性和可行性，对于风电调度具有一定的参考意义。 Forecasting the output power of wind farm play a vital role in the reducing of the running cost of wind power plants and the reasonable arrangements for the dispatch of power systems. Improving the prediction accuracy of wind power can contribute to lower the detrimental impact of wind power plants on power grid as well as improve the competitiveness of wind power plants against others in electricity markets. As in the establishment of short-term wind power forecasting model, the sample selection has a greater impact on prediction accuracy, so the study of sample selection method is very important. In this paper, a new method is proposed in which the history wind power data is clusterred by K-means algorithm, data is classified through the LVQ net and the prediction model of wind power is established with the least-squares method. The practical application shows that the method can be utilized to predict the wind power effectively and precisely, and it is quite significant for the regulation of wind power.