Intelligent Resource Scheduling in Cloud-Fog Computing: An Integrated Approach Based on Deep Learning, Game Theory, and Heuristic Algorithms

The rapid development of global information technology and the digital economy has triggered a surge in data. It has also caused a sharp rise in the number of IoT devices. This poses unprecedented challenges to traditional computing architectures and scheduling methods. Cloud computing and fog computing have significantly enhanced data processing capabilities and real-time responsiveness. They have also introduced complex issues such as uneven resource distribution, dynamic load variations, fluctuating network latency, task diversity, and uncertain arrival rates. Traditional scheduling methods based on rules or single heuristic algorithms cannot meet the strict requirements of real-time performance, global optimality, and system fairness in practical applications. To address this situation, this paper proposes an integrated scheduling framework. It incorporates advanced techniques such as Convolutional Neural Networks (CNN), game theory, deep reinforcement learning, and genetic algorithms. First, CNN is used for deep feature extraction of high-dimensional spatiotemporal data and historical load image analysis of computing nodes and tasks. This enables accurate prediction of future load and resource demands. Next, a dynamic resource balancing mechanism for multiple nodes is constructed by combining a non-cooperative game model with deep reinforcement learning. This ensures a balance between local optimization and global fairness. Finally, global iterative optimization is performed using heuristic strategies such as genetic algorithms. The model parameters and scheduling strategies are continuously updated through closed-loop feedback. Experimental results show that the proposed framework significantly outperforms traditional methods in key performance indicators such as average response time, resource utilization, system stability, scheduling fairness, and energy consumption. This study provides a new theoretical basis and technical pathway for intelligent resource management and task scheduling. It promotes further optimization of large-scale distributed systems in practical applications.