A deep learning model for malicious URL recognition

The proliferation of web application services in the power grid has heightened the significance of network security, particularly regarding web security issues. The detection and prevention of such malicious URLs play a critical role in safeguarding the information network security of the power grid. Traditional approaches to identifying malicious URLs mainly rely on blacklists. However, these methods have limitations when it comes to defending against unknown or non-blacklisted malicious attacks. Additionally, maintaining a comprehensive malicious URL library requires substantial resources. To address these challenges, this paper introduces a model that combines convolutional neural networks (CNN) and attention models to enhance the recognition of malicious URLs. The CNN component of the model captures local semantic features, allowing it to identify patterns and characteristics associated with malicious URLs. The integration of attention mechanisms further enhances the model's performance by enabling it to focus on important features within the URL data. Additionally, the XGBoost algorithm is employed to improve the overall accuracy of the model. Through experimental evaluations on multiple datasets, the proposed model demonstrates superior accuracy compared to various existing models. The results validate its effectiveness in accurately identifying malicious URLs, thereby contributing to the advancement of web security in the power grid domain.