Self-supervised CNN for user behavior analysis on smart meter data

Smart meters generate extensive data on individual consumer electricity usage, providing valuable insights that can aid in identifying demographic information and advancing the development of smart grids. Current research has primarily focused on traditional machine learning approaches for this task, with relatively few studies exploring deep learning methods, despite their potential for more accurate and efficient analysis. To address this gap, this paper proposes a self-supervised deep learning approach based on Convolutional Neural Network (CNN) to identify demographic information from smart meter data. The model leverages the Fast Fourier Transform (FFT) to detect frequency cycles within the dataset, which are then used to optimize the sizes of convolutional kernels. This design enhances periodic stability during shallow feature extraction, improving the model’s ability to capture meaningful patterns in the data. Furthermore, the model incorporates a self-supervised pre-training strategy to predict temporal and spatial interactions in load signals, effectively enhancing representation learning without relying on extensive labeled data. This approach ensures the model’s robustness and adaptability to different datasets. Comprehensive experiments were conducted on a publicly available Irish dataset to evaluate the model’s performance. Results demonstrate that the proposed model surpasses a series of state-of-the-art (SOTA) methods, achieving superior performance in demographic information identification. These findings highlight the effectiveness of integrating FFT-based kernel design and self-supervised learning in improving feature extraction and representation learning for smart meter data.