Abstract
Due to the unpredictability of loads and the complexity of fault propagation, modern power distribution networks require advanced technologies for fault identification and localization. In traditional systems, there is a high rate of false alarms, slow response times, and limited precision. For fault identification, the Multimodal Deep Feature Hybrid Deep Learning Model (MDF-HDL) utilizes LiDAR, optical images, and sensor data. This model was developed to help overcome these obstacles. The model utilizes deep learning layers to provide elaborate, multimodal feature representations. Additionally, Kalman filtering is used to enhance feature fusion. Classification results can be refined using decision trees, which are optimized using the Adam algorithm. This helps to reduce mistake rates. Through the use of GIS mapping, faults are precisely identified, facilitating efficient maintenance planning. While maintaining a low computational complexity, the MDF-HDL model, implemented in Python, achieves an accuracy of 98.91%, a precision of 98.7%, a recall of 98.3%, an F1-score of 98.5%, and an inference time of 12.5 milliseconds. Through the incorporation of multimodal data and sophisticated algorithms, the system is able to transcend standard constraints, guaranteeing fault management that is both dependable and effective in complicated grid contexts.