Abstract: ToIn order to efficiently allocate limited network resources in cloud computing whileto ensuringe service quality, while and improving resource utilization and management efficiency, an encoder-decoder-based deep reinforcement learning algorithm (ED-DRL) was proposed for service function chain (SFC) deployment. this paper proposes a deep reinforcement learning algorithm based on an encoder-decoder structure (ED-DRL) for the deployment of Service Function Chains (SFCs). The SFC placement was formulated as a Markov decision process, and a reinforcement learning approach with a graph attention network (GAT) encoder and a gated recurrent unit (GRU) decoder was employed to effectively extract network topology features and inter-node dependencies.The algorithm first treats SFC placement as a Markov decision process, then employs a reinforcement learning method with a Graph Attention Network (GAT) encoder and a Gated Recurrent Unit (GRU) decoder structure to efficiently extract network topology features and inter-node dependencies. By integratingCombined with the Asynchronous Advantage Actor-Critic (A3C) method, the algorithm was capable of generates secure and reliable SFC placement strategies in dynamic environments.SFC placement strategies that are reliable and secure in dynamic environments. Simulation results demonstrateshow that the encoder-decoder structure reinforcement learning approachmethod, which considerings security and reliability, achieved an acceptance rate of 70.1% and an average reward of 0.0635, outperforming existing algorithmsoutperforms existing algorithms in terms of acceptance rate, average reward, and average running time.

Key words: Service Function Chain Deployment, Reinforcement Learning, Markov Decision Process, Asynchronous Advantage Actor-Critic, Graph Attention Network, Gated Recurrent Unit

摘要： 为了在云计算中高效分配有限网络资源以确保服务质量，同时提高资源利用率和管理效率，提出了一种基于编解码结构的深度强化学习算法(ED-DRL)用于服务功能链(SFC)部署。该算法首先将SFC放置看作一个马尔科夫决策过程，采用图注意力网络(GAT)编码器和门控循环单元(GRU)解码器结构的强化学习方法高效提取网络拓扑特征和节点间的依赖关系，结合异步优势Actor-Critic(A3C)方法，算法能在动态环境中生成安全可靠的SFC放置策略。仿真结果表明 ，考虑安全与可靠性的编解码结构强化学习方法能获得了70.7%的够在接受率与、0.0635的平均收益，与平均运行时间上优于现有算法。

关键词: 服务功能链部署, 强化学习, 马尔科夫决策过程, 异步优势Actor-Critic, 图注意力网络, 门控循环单元