文献类型：期刊文献

英文题名：Intelligent Collaborative Resource Allocation for Mechanical Manufacturing Edge Networks: A Deep Reinforcement Learning Approach

作者：Xiao, Kaile[1]; Wang, Xiajing[2]; Wang, Jing[1]

第一作者：Xiao, Kaile

机构：[1] School of Applied Science and Technology, Beijing Union University, Beijing, China; [2] The Open University of China, Beijing, China

第一机构：北京联合大学应用科技学院

年份：2025

卷号：12

期号：12

外文期刊名：EAI Endorsed Transactions on Scalable Information Systems

收录：EI(收录号：20262821064659)

语种：英文

外文关键词：Concurrency control - Decision making - Deep learning - Deep reinforcement learning - Distributed computer systems - Intelligent agents - Multi agent systems - Reinforcement learning - Smart manufacturing - Topology

摘要：Industry 4.0 is transforming mechanical manufacturing systems into edge-enabled, networked, and intelligent environments, where concurrent task execution, heterogeneous resource coordination, and dependency-aware scheduling have become critical requirements. In such scenarios, resource allocation must jointly consider computational demand, storage demand, business priority, deadline urgency, and inter-task dependencies, while enabling coordinated decisions among distributed edge agents. However, existing single-agent reinforcement learning methods have limited capability to model complex dependency relationships and heterogeneous resource collaboration under concurrent workloads, whereas conventional multi-agent systems often rely on coarse-grained task modeling and simplified cooperation mechanisms. To address these limitations, this paper proposes MIRA, a multi-agent deep reinforcement learning-based method for resource allocation in mechanical manufacturing edge networks. MIRA first decomposes tasks into fine-grained dependent subtasks, constructs a deadline-aware multi-metric priority function, and introduces a dynamic weight adjustment mechanism to balance computational demand, storage demand, normalized business priority, and deadline urgency. It then employs an adjacency matrix to characterize topology-aware agent interactions, enabling coordinated decision-making between computing agents and storage agents. Furthermore, MIRA incorporates an event-triggered state-exchange mechanism that updates subtask priorities and agent policies under changing workload, deadline, resource, and topology conditions. Experimental results on a PCB-derived simulated scheduling workload show that MIRA outperforms the selected baselines across multiple scheduling metrics. Copyright ? 2026 Kaile Xiao et al., licensed to EAI. This is an open access article distributed under the terms of the CC BY-NC-SA 4.0, which permits copying, redistributing, remixing, transformation, and building upon the material in any medium so long as the original work is properly cited.