فاطمه انوار

Cloud manufacturing, as a novel approach in managing manufacturing processes due to its high scalability and flexibility, has become a key method in optimizing production workflows. Among the core challenges in cloud manufacturing is the integration of services and optimal resource allocation to combine multiple services and deliver a final product with efficient use of resources and maximum added value. This research addresses the issue of dynamic service composition in multi-product manufacturing environments with diverse processes, yet shared service requirements and production resources, taking into account logistics parameters and distribution to demand centers. The study's approach focuses on improving the manufacturing process by utilizing maximum available resources with the goal of enhancing service quality in the delivered product. Most recent studies in this area have employed metaheuristic methods, which, despite requiring complex structural designs, lack necessary adaptability in dynamic environments. Deep Reinforcement Learning (DRL), due to its capabilities in managing complex decision-making under uncertainty, has gained attention as an effective method in recent years, particularly in cloud manufacturing environments. These algorithms provide intelligent and adaptive solutions that contribute to improving service quality and reducing production time. To address the problem, a service composition approach based on Deep Deterministic Policy Gradient (DDPG) with prioritized experience replay is proposed. Through iterative training, it learns optimal service composition solutions. The performance of this model is eva‎luated and compared with Q-learning and DDQN algorithms. The results indicate that the proposed model achieves faster convergence and optimal service quality compared to the other two models. Although the response time of the DDPG model is higher due to its structural complexity, its other advantages compensate for this delay. Furthermore, in a practical setting, the model demonstrates at least a 200% improvement in production time and reduced waiting time, while also showing robust tolerance to up to 20% changes in resources.

يادگيري تقويتي , گراديان سياست قطعي عميق , تخصيص منابع , تركيب خدمات , كيفيت خدمت

Reinforcement learning , Deep Deterministic Policy Gradient , resource matching , service composition , Quality of service

Fatemeh Anvar

Dr Mohammad Reza Rasouli