A Hybrid Deep Reinforcement and Swarm Optimization Strategy for Intelligent Cloud Service Composition

Cloud service composition has emerged as a complex optimization problem, particularly in dynamic, large-scale, and heterogeneous environments where Quality of Service requirements vary continuously. To address the limitations of conventional approaches, this study proposes a novel hybrid optimization framework that integrates Deep Reinforcement Learning with Particle Swarm Optimization for intelligent and adaptive cloud service composition. The DRL component models the service environment as a Markov Decision Process, enabling the system to learn optimal service selection policies through interaction and feedback. Simultaneously, PSO enhances global search efficiency, ensuring faster convergence and optimality in the composition process. The proposed method is evaluated against three state-of-the-art meta-heuristic algorithms—Grey Wolf Optimizer, Ant Colony Optimization, and Improved Bat Algorithm—using standard benchmark cloud service datasets. Experimental results confirm that the proposed hybrid DRL-PSO framework significantly outperforms existing algorithms across all major performance metrics. Specifically, it achieves a 12.5% improvement in overall QoS utility compared to the best-performing baseline (GA-PSO at 0.78 vs. DRL-PSO at 0.91), an 18.3% reduction in execution time (from 12.9s with DRL-KF to 10.1s with DRL-PSO), and a 9.7% increase in service success rate (from 87.2% with GA-PSO to 95.3% with DRL-PSO). These gains validate the effectiveness of combining deep learning and swarm intelligence for cloud service composition. Additionally, the framework demonstrates superior scalability by maintaining low convergence times (15 episodes) and consistent performance across dynamic QoS fluctuations, proving its robustness in high-dimensional, real-time environments. This research advances the state of cloud computing by introducing a self-adaptive, learning-based optimization model that enables efficient, resilient, and QoS-aware service composition for complex and evolving cloud ecosystems.