Multi-dimensional deep learning-based anomaly detection and adaptive security strategy for sustainable power grid operation
Qingqing Ren, Wanqing Kang, Xuehui Yang, Qingpeng Wang, Qiang Huang
Abstract
Traditional grid anomalous behaviour identification methods usually rely on fixed rules and single-dimension data analysis, which are difficult to meet the requirements for anomaly detection in complex and changing grid operation (PGO) environments, and cannot effectively ensure grid security. In this paper, we propose an intelligent strategy for detecting abnormal behaviours in power grid operation by combining multi-dimensional digital portraits with deep neural networks (DNNs). Traditional methods rely on fixed rules and unidimensional analyses, which are insufficient to cope with complex grid environments. This study can process grid operation data across time series, spatial and frequency dimensions to create a comprehensive digital portrait. By combining Convolutional Neural Networks (CNNs) for spatial and frequency feature extraction with Recurrent Neural Networks (RNNs) for time-series analysis, the hybrid model achieves high accuracy in anomaly detection, and excels in the anomaly category D (accuracy: 0.965, F1 score: 0.827). In addition, an adaptive security protection strategy is proposed that enhances the stability of the grid over time, as evidenced by the significant decrease in the frequency of anomalous behaviours from 0.133 per day to 0.034 per day in one year. These innovations demonstrate the practical value of the present model in ensuring secure and sustainable operation of the grid.