Panoramic dynamic topology model of distribution network based on multimodal data fusion

The dynamic generation of distribution network topology primarily relies on various sensors and data acquisition devices to collect diverse types of data from the distribution network, including multiple parameters such as temperature, humidity, voltage, current, and power. The data exhibit diverse frequencies and structural modalities. Due to the complexity of the distribution network topology, which involves multiple operational modes and states, fine-grained topological analysis is required to derive accurate topological results. The paper proposes a panoramic dynamic topology framework for distribution networks, supporting the automatic generation of such topology. This enables the establishment of detailed topological models for different operational modes and states, thereby improving the accuracy of topological analysis. Based on a deep learning-driven multi-modal feature extraction model, an attention-driven fusion mechanism is employed to correlate multi-modal data features. This mechanism calculates attention weights between different modal features to identify an adaptive set of features. The proposed multi-modal data fusion model for dynamic topology of distribution networks can be divided into three stages: deep interactive feature extraction, hierarchical feature fusion, and fusion validation. Multi-modal features are extracted from text and audio data, and through hierarchical feature fusion, the model can perceive and capture multi-scale topological information. The comprehensively fused features, after multi-layer fusion, are input into a fully connected layer for final feature processing and integration.