Research on power data security full-link monitoring technology based on alternative evolutionary graph neural architecture search and multimodal data fusion

Abstract This paper proposes a new framework for power data security monitoring, which aims to cope with complex and changeable network attacks. The current evolutionary graph neural architecture search (GNAS) method mainly focuses on the topological connection and feature fusion between network layers, but it often requires a lot of computing resources, and the real-time performance of the GNN (graph neural network) model is difficult to meet the requirements when facing dynamic attacks and changing data. To solve this problem, this paper proposes a hybrid method that combines multimodal data-aware attacks with Light Gradient Boosting Machine (LightGBM) and Support Vector Regression (SVR) agent models. By using Particle Swarm Optimization-Genetic Algorithm (PSO-GA) for optimal architecture search and combining the dynamic adaptability of Deep Q-Network (DQN) algorithm, this method can automatically identify the most suitable GNN architecture for power data monitoring, thereby improving the adaptive detection and defense efficiency of the system. Experiments were conducted on the full-link monitoring platform of four power companies in Shanghai using a public network traffic dataset. The results showed that the detection efficiency of the GNAS-LightGBM + SVR (PSO-GA) model reached 96.34%, 2.79% higher than that of GNAS (PPO). At the same time, the defense efficiency of DQN reached 97.45%. The experimental results show that the proposed method significantly improves the detection and defense efficiency of power data security.