【摘要】 当前，如何充分利用异构化、碎片化、失衡化的算力资源满足差异化的智能工业任务要求，是工业互联网的智能化发展过程中急需解决的难题。本文聚焦工业互联网的云边端协同调度，突破传统基于小模型的调度策略，深入探讨了网络大模型在该领域的应用潜能。首先，从云边端的数据异构、算力异构、算法异构三个维度深入剖析了现存挑战；然后，结合NetGPT、NetLLM和LAMBO等典型网络大模型的实际应用案例，探讨了它们在降低算力、优化网络任务和提高效率等方面的显著优势；最后，进一步提出了多模态模型融合、大模型云边端协同部署、行业定制模型开发等未来发展方向，论述了实施难点及潜在解决方案，旨在为工业互联网的智能化演进提供新的思路与策略。更多还原

【Abstract】 Currently, how to make full use of heterogeneous, fragmented, and imbalanced computing resources to satisfy the requirements of differentiated intelligent industrial tasks is an urgent problem to be solved in the process of intelligent development of industrial Internet. This paper focuses on the cloud-edge-end collaborative scheduling of industrial Internet, breaks through the traditional scheduling strategy based on small models, and deeply explores the application potential of network large models in this field. First, the existing challenges are analyzed from the three dimensions of data heterogeneity, computing power heterogeneity, and algorithm heterogeneity at the cloud-side end. Then, combined with the practical application cases of typical network big models such as Net GPT, Net LLM and LAMBO, their significant advantages in reducing computing power consumption, optimizing network tasks and improving efficiency are discussed. Finally, future development directions such as multimodal model fusion, cloud-edge-end collaborative deployment of large models, and industry-customized model development are further proposed, and implementation difficulties and potential solutions are discussed, aiming to provide new ideas and strategies for the intelligent evolution of industrial Internet.更多还原