提出了一种动态安全分析神经网络输入特征的优选方法。该方法将决策表最优特征子集理论与粗糙集理论的元素离散化方法相结合,在保证较高精度的运行点分类和稳定裕度计算的基础上,从大维数的注入元中优选特征元作为神经网络输入特征,以降低其输入特征的个数,较好地解决了神经网络动态安全域方法用于大系统所面临的“维灾难”问题。通过对电力系统分析综合程序(PSASP)中EPRI-36测试系统的仿真计算,验证了该方法的有效性。 A method of dynamic security analysis of neural network input features is proposed. This method combines the theory of decision table subset of features and element discretization of rough set theory. Based on the classification of operating points and the calculation of stability margins, As a feature input of neural network, feature elements are reduced to reduce the number of input features and solve the problem of “dimension disaster” facing large systems. Through the simulation of EPRI-36 test system in PSASP, the validity of this method is verified.