The demands for mass individualization and networked collaborative manufacturing are increasing, bringing significant challenges to effectively organizing idle distributed manufacturing resources. To improve production efficiency and applicability in the distributed manufacturing environment, this paper proposes a multi-agent and cloud-edge orchestration framework for production control. A multi-agent system is established both at the cloud and the edge to achieve the operation mechanism of cloud-edge orchestration. By leveraging Digital Twin (DT) technology and Industrial Internet of Things (IIoT), real-time status data of the distributed manufacturing resources are collected and processed to perform the decision-making and manufacturing execution by the corresponding agent with permission. Based on the generated data of distributed shop floors and factories, the cloud production line model is established to support the optimal configuration of the distributed idle manufacturing resources by applying a systematic evaluation method and digital twin technology, which reflects the actual manufacturing scenario of the whole production process. In addition, a rescheduling decision prediction model for distributed control adjustment on the cloud is developed, which is driven by Convolutional Neural Network (CNN) combined with Bi-directional Long Short-Term Memory (BiLSTM) and attention mechanism. A self-adaptive strategy that makes the real-time exceptions results available on the cloud production line for holistic rescheduling decisions is brought to make the distributed manufacturing resources intelligent enough to address the influences of different degrees of exceptions at the edge. The applicability and efficiency of the proposed framework are verified through a design case.

大规模个性化和网络化协同制造的需求不断增加，对有效组织闲置的分布式制造资源带来了重大挑战。为了提高分布式制造环境下的生产效率和适用性，本文提出了一种用于生产控制的多智能体和云边编排框架。在云端和边缘建立多代理系统，实现云边编排的运行机制。利用数字孪生（DT）技术和工业物联网（IIoT），收集和处理分布式制造资源的实时状态数据，由相应的代理人在获得许可后进行决策和制造执行。基于分布式车间和工厂生成的数据，建立云产线模型，应用系统的评估方法和数字孪生技术，支持分布式闲置制造资源的优化配置，反映整体实际制造场景生产过程。此外，开发了一种基于卷积神经网络（CNN）结合双向长短期记忆（BiLSTM）和注意力机制驱动的云上分布式控制调整的重新调度决策预测模型。引入了一种自适应策略，使实时异常结果可在云端生产线上进行整体重新调度决策，使分布式制造资源足够智能，以应对边缘不同程度异常的影响。通过设计案例验证了所提出框架的适用性和效率。