Research on Network Security Risk Monitoring Method Based on Big Data in Power Grid Heterogeneous Network Fusion Scenarios

With the rapid development of information technology, critical infrastructures such as power grids face increasingly complex cybersecurity threats. To improve the performance of traditional Convolutional Neural Networks (CNNs) in intrusion detection for power grid heterogeneous networks, this paper proposes an improved CNN model designed to enhance feature extraction capabilities, accelerate the training process, and improve generalization. The model adapts to the characteristics of power grid traffic data by optimizing the convolutional layer structure, selecting appropriate activation functions, and incorporating batch normalization (BN) layers and Dropout regularization. The multi-level convolutional network structure extracts potential features from power grid traffic at different scales, while the BN layer addresses the issues of gradient vanishing and explosion, improving training stability and accelerating model convergence. The ReLU activation function effectively avoids the gradient vanishing problem caused by traditional activation functions, while also improving computational efficiency. To prevent overfitting, the Dropout layer enhances the model's generalization ability. During optimization, the Adam optimizer is used, which combines momentum and adaptive learning rate strategies, dynamically adjusting the learning rate to improve training efficiency and stability. Experimental results show that the proposed improved CNN model achieves significant results in intrusion detection for power grid heterogeneous networks, with an accuracy rate of 95.3%, outperforming Autoencoder-based Intrusion Detection System (Autoencoder-based IDS) and Variational Autoencoder (VAE). Comparative experiments verify the superiority of this model in handling complex network traffic data, demonstrating strong robustness and good performance