The overall operation and the voltage stability of power grid network are likely to be affected by the fluctuations of wind power. High accuracy of short-term wind power prediction can guarantee the stability and safety of power supply system. This paper proposes a KNN algorithm considering neighbors’ density on the basis of KNN algorithm, applying to short - term wind power prediction. The KNN algorithm considering neighbors’ density, firstly identifies training samples within the given domain of testing object and figures out density distribution of the training samples in each dimension; and secondly calculates the value of K, which dynamically changes at different times; and finally, the test object is classified according to the KNN algorithm. Taking a wind farm in Changzhou as an example, its historical data was analyzed and then predictions were made through the KNN algorithm considering neighbors’ density, proving accuracy and validity of the algorithm.