The Research on a Lightweight Intelligent Distributed Feeder Automation Model Based on Knowledge Distillation

With the rapid advancement of smart grids, Distributed Feeder Automation Systems (DFAS) have become crucial in enhancing the reliability of distribution networks. However, the integration of existing deep learning methods into DFAS faces challenges due to computational resource constraints. To address these challenges, this paper proposes a lightweight intelligent distributed feeder automation model based on knowledge distillation. First, a deep learning-based teacher model is designed, capable of effectively processing time-series data from distribution networks and capturing the complex interrelationships between electrical quantities. Through knowledge distillation, a lightweight student model is developed, employing techniques such as depthwise separable convolution, the Squeeze-and-Excitation module, and a lightweight temporal attention mechanism to achieve efficient model compression. Extensive experimental results demonstrate that the proposed model maintains over 90% fault classification accuracy, despite a significant reduction in parameters. This research provides a promising solution for improving the operational efficiency and reliability of smart distribution networks.