Intelligent recognition and sustainable security protection strategies for abnormal behavior of power grid operation data based on multidimensional digital portrait and deep neural networks

Traditional methods for identifying abnormal behavior in the power grid typically rely on fixed rules and single-dimensional data analysis, making it difficult to meet the anomaly detection requirements in complex and changing power grid operation (PGO) environments, and unable to effectively ensure the safety of the power grid, limiting their effectiveness in complex environments. The article presents an intelligent strategy combining multidimensional digital portraits with deep neural networks (DNN). Power grid operation (PGO) data is cleaned, normalized, and analyzed across time series, spatial, and frequency dimensions to create a multidimensional digital portrait. CNN extracts spatial and frequency features, while RNN processes time series data, enabling accurate anomaly detection. The model performs well, especially for anomaly category D, achieving an accuracy of 0.965 and an F1 score of 0.827. Trend analysis of one year’s grid data shows a decrease in abnormal behavior frequency from 0.133 times/day on day 90 to 0.034 times/day on day 365, indicating improved system stability over time. These results confirm the model's practical value for ensuring the safe operation of power grids.