Attention-Based Multiscale Network for Efficient Load Signal Classification

Load signal classification (LSC) is essential for improving power system reliability, operational efficiency, and intelligent decision-making. Recent advancements in artificial intelligence, particularly deep learning, have revolutionized the ability to extract and interpret complex signal patterns. This paper presents a novel framework for load signal recognition based on deep metric learning. At the core of the framework is an attention-based multiscale atrous convolutional network (AMACN), designed to capture distinctive features across multiple temporal resolutions. The network dynamically emphasizes the most salient features, effectively extracting meaningful patterns from load signals. Its efficiency is further enhanced through group convolutions and channel shuffle techniques without compromising performance. To ensure accurate classification, an aggregation loss function is introduced, minimizing intra-class variability while maximizing inter-class separability. Extensive experiments on real-world datasets validate the effectiveness of the proposed framework, demonstrating state-of-the-art classification accuracy. These results underscore its potential to advance load signal classification and contribute to intelligent power system management.