An Industrial Load Identification Method for Distribution Grid Smart Terminal

Abstract

This paper proposes a non‐intrusive industrial load identification method for distribution grid smart terminals using low‐frequency power data. Existing high‐frequency feature‐based methods face challenges such as high sampling costs and processing complexity in industrial scenarios. To address this, a hybrid CNNSE2‐LSTM model is developed: a 1D convolutional network with Squeeze‐Excitation blocks (CNNSE2) extracts spatial features by enhancing local correlations and recalibrating channel weights, while a parallel LSTM captures long‐term temporal dependencies. The model is trained on 15‐min interval smart terminal data from a manufacturing enterprise, covering three environmental protection devices and four production devices. Experimental results show average accuracy, precision, recall, F1 score, and G‐means exceeding 90%, outperforming LSTM, CNN‐LSTM, and attention‐based FCN‐LSTM models. The method reduces reliance on costly high‐frequency sampling while maintaining high identification accuracy, offering a cost‐effective solution for grid security and energy management in industrial applications. Future work will focus on improving generalization for diverse equipment via transfer learning and unsupervised techniques. © 2026 Institute of Electrical Engineers of Japan and Wiley Periodicals LLC.