An evaluation model based on radial basis function neural network is proposed in order to monitor,analyze and evaluate the operation state of power grid more quickly and visually after contingency. AP clustering is used to optimize the center and the number of neurons of the hidden layer.The extended evaluation indices of power grid operation security state are established,and index weights and exponent order are used to avoid the masking effect,and PCA is used for the feature selection. The simulation on IEEE-30 bus system shows the effectiveness of the proposed model.