Missing Data Recovery for Transmission Networks Using Graph Convolutional Networks with Location Mask

Abstract

As the integration of power systems and information systems has advanced, the significance of power data has progressively increased. However, data loss stemming from information attacks and other transmission failures diminishes the value of power data. Research on missing data recovery in transmission network nodes overlooks the stochastic characteristics of missing positions and random missing scales in missing information completion. Therefore, this paper proposes a graph convolution method with a location mask. This model introduces an input‒output matrix containing all node feature information to address random missing positions. The designed location mask and the class residual network structure formed by the graph convolution layers enhance the model's information mining and generalization capabilities, effectively addressing random missing positions and random missing scales. Subsequently, through comparison with other models and through testing across 61 scenarios involving various missing patterns and scales, the model's effectiveness and superiority are validated. Finally, a series of experiments are conducted to explore the role of the location mask during the data recovery task.