Ensuring the security and resilience of power systems is crucial in mitigating risks associated with component failures. Traditional contingency screening methods, such as the N-1 criterion, focus on single-component failures, limiting their effectiveness in handling simultaneous disruptions. This paper presents a novel Deep Convolutional Neural Network (Deep CNN)-based contingency screening framework capable of efficiently managing both N-1 and N-2 contingencies in large-scale power networks. Unlike conventional approaches, the presented model leverages advanced architectural optimization and hyperparameter tuning in order to enhance accuracy while significantly reducing computational complexity. The proposed method has been rigorously validated on IEEE bus systems and applied to Indonesia's Java-Madura-Bali (JAMALI) 500 kV grid, the country's largest interconnected power system. Experimental results demonstrate that the presented Deep CNN model outperforms traditional accuracy and computational efficiency techniques, particularly in identifying and ranking high-risk N-2 contingencies. This research provides a scalable and real-world applicable AI-driven solution for power system security, paving the way for more intelligent, reliable, and resilient grid management strategies worldwide. © 2025 Praise Worthy Prize S.r.l. - All rights reserved.