Traditionally, the operation and maintenance of optical networks rely on the experience of engineers to configure network parameters, involving command-line interface, middle-ware scripting, and troubleshooting. However, with the emerging of newly B5G applications, the traditional configuration cannot meet the requirement of real-time automatic configuration. Operators need a new configuration way without manual intervention at an underlying optical transport network. To cope with this issue, we propose an intent defined optical network (IDON) architecture toward artificial intelligence-based optical network automated operation and maintenance against service objective, by introducing a self-adapted generation and optimization (SAGO) policy in a customized manner. The IDON platform has three key innovations including intent-orient configuration translation, self-adapted generation and optimization policy, and close-loop intent guarantee operation. Focusing specifically on communication requirements, the IDON uses natural language processing to construct semantic graphs to understand, interact, and create the required network configuration. Then, deep reinforcement learning (DRL) is utilized to find the composition policy that satisfies the requirement of intent through the dynamic integration of fine-grained policies. Finally, the deep neural evolutionary network (DNEN) is introduced to achieve the intent guarantee at the milliseconds level. The feasibility and efficiency are verified on enhanced SDN testbed. Finally, we discuss several related challenges and opportunities for unveiling a promising upcoming future of intent defined optical network.

传统上，光网络的运营和维护依靠工程师的经验来配置网络参数，包括命令行界面，中间件脚本和故障排除。但是，随着新的B5G应用的兴起，传统的配置已不能满足实时自动配置的要求。运营商需要一种新的配置方式，而无需人工干预基础的光传输网络。为解决此问题，我们通过以定制方式引入自适应的生成和优化（SAGO）策略，针对基于服务目标的基于人工智能的光网络自动化操作和维护提出了意图定义的光网络（IDON）体系结构。IDON平台具有三项关键创新，包括面向意图的配置转换，自适应的生成和优化策略以及闭环意图保证操作。IDON专门针对通信需求，使用自然语言处理来构造语义图，以理解，交互和创建所需的网络配置。然后，利用深度强化学习（DRL）通过动态集成细粒度策略来找到满足意图要求的合成策略。最后，引入了深度神经进化网络（DNEN）以在毫秒级别实现意图保证。在增强的SDN测试平台上验证了可行性和效率。最后，