Intelligent detection and big data analysis methods for power grid fault handling based on multimodal image fusion

Abstract

In order to solve the problems of distributed power network inspection images, the shortage of single mode fault characterization, and the difficulty of directly supporting the response decision, a new approach is presented. In this method, a uniform sample representation of visible, infrared, partial discharge (PD), PD, and video frame sequence (VFS) is built, which includes the property of the device, the space position, the fault category, the risk level, and the response record. Based on this, the multi-source image is normalized, the area is cut, and the cross mode is registered, and the cross-modal feature fusion network is built to realize the cooperative representation of vision damage, temperature anomaly, discharge trace and time state. Then, the fusion feature is fed into the fault recognition and localization module, which outputs the fault category, the destination position, and the confidence score. These are then combined with the history of the defect database, the importance of the device, and the response records, and the results show that the proposed approach is stable in the fields of power grid fault identification, target location, and response priority generation, which can reduce the vulnerability of single-mode recognition to environment disturbance and feature loss. The research shows that the combination of multi-modality image fusion and large data correlation analysis provides a new technology path for power system failure.