随着工业4.0与智能制造的快速发展，工业网络对高灵活性、低时延及强可靠性的需求日益迫切。软件定义网络（Software-Defined Networking，SDN）通过集中控制与数据平面分离的特性，为构建多业务协同的工业网络提供了新的技术框架。而网络切片作为SDN的核心功能之一，能够为不同工业应用场景提供定制化的网络服务。本文基于SDN架构对工业网络切片资源分配实现更加智能化的分配策略，以提高服务质量（Quality of Service，QoS）目标及综合网络性能，主要研究内容如下：

（1）针对现有工业网络切片资源分配方案多以公平性和吞吐量作为优化目标，且在平衡资源公平分配与网络收益最大化方面存在不足，提出基于分层联合优化的资源分配策略，综合考虑切片间和切片内资源调度问题。在切片间资源分配中，根据业务特点定义切片优先级，并在比例公平（Proportional Fairness，PF）算法基础上引入速率、时延和丢包率因子，权衡公平性与业务需求，提出PF-Q算法。在切片内资源调度中，引入斯塔克尔伯格（Stackelberg）博弈模型，将基站和用户的资源分配决策建模为动态博弈过程，基站作为领导者优化资源分配和迁移策略以最大化网络收益，用户作为跟随者动态调整需求。仿真结果表明，联合优化策略相较于传统博弈策略在网络收益约提升了11%，同时对比基准算法资源分配的公平性达到0.96以上。

（2）针对工业网络中业务需求多样性和QoS要求差异性导致的网络资源利用低问题，提出一种基于深度强化学习的网络切片动态资源分配策略。该策略运用深度Q网络（Deep Q-Network，DQN）算法优化网络切片资源分配的准入控制，通过智能体在特定时间窗口内处理资源请求，并根据不同网络切片的QoS要求及请求准入结果进行资源的动态分配。实验结果表明，所提出的策略相比基准算法在提高网络收益、资源利用率和接收率方面分别提升了8.33%、9.84%和8.57%，在保障服务质量的同时，也提高了资源利用效率与系统收益。

With the rapid development of Industry 4.0 and smart manufacturing, industrial networks increasingly demand high flexibility, low latency, and robust reliability. Software-Defined Networking (SDN), featuring centralized control and separated data planes, provides a novel technical framework for constructing multi-service collaborative industrial networks. As a core functionality of SDN, network slicing enables customized network services for diverse industrial application scenarios. This paper proposes an intelligent resource allocation strategy for industrial network slicing under the SDN architecture to enhance Quality of Service (QoS) objectives and overall network performance. Key research contributions include:

(1) In view of the existing industrial network slicing resource allocation schemes, most of which take fairness and throughput as optimization objectives, and have shortcomings in balancing fair resource allocation and network benefit maximization, a resource allocation strategy based on joint optimization is proposed to comprehensively consider inter-slice and intra-slice resource scheduling problems. In the resource allocation between slices, slice priority is defined according to the service characteristics, rate, delay and packet loss rate are introduced based on Proportional Fairness (PF) algorithm, and PF-Q algorithm is proposed to balance fairness and service requirements. In intra-slice resource scheduling, Stackelberg game model is introduced to model resource allocation decisions between base stations and users as a dynamic game process. Base stations act as leaders to optimize resource allocation and migration strategies to maximize network benefits, and users act as followers to dynamically adjust demand. The simulation results show that compared with the traditional game strategy, the network income of the joint optimization strategy is increased by about 11%, and the fairness of resource allocation of the benchmark algorithm is more than 0.96.

(2) To resolve low network resource utilization caused by diverse service demands and varying QoS requirements in industrial networks, we propose a deep reinforcement learning-based dynamic resource allocation strategy. Utilizing the Deep Q-Network (DQN) algorithm, this approach optimizes admission control for slice resource allocation. Intelligent agents process resource requests within specific time windows, dynamically allocating resources based on QoS specifications and admission outcomes. Experimental results show the proposed strategy improves network revenue, resource utilization, and request acceptance rates by 8.33%, 9.84%, and 8.57% respectively compared to benchmark algorithms, while ensuring service quality, it also improves resource utilization efficiency and system revenue.