Research On The Deployment And Application Of Artificial Intelligence In Optical Networks

Nowadays,as an important infrastructure that provides low-latency and large-capacity transmission capabilities,optical networks,combined with the computing and storage capabilities provided by data centers,carry massive amounts of heterogeneous data.Artificial Intelligence(AI)is regarded as one of the most promising methods to realize the automatic control and fault location of optical networks by virtue of its powerful data analysis capabilities.In terms of the deployment of artificial intelligence technology in optical networks,the existing research has the following shortcomings:Under the traditional cloud computing paradigm,the original data needs to be aggregated to the central cloud node for processing,which is difficult to meet the needs of some low-latency services.At the same time,it increases the load on network bandwidth.In the application of artificial intelligence technology in optical networks,although many studies have proved the advantages of AI in optical performance detection and decision management,few studies have proposed using AI technology to improve the physical layer Security.The main content and results of this article include the following two points:(1)Propose an optical network model deployment and resource allocation mechanism for artificial intelligence inference services.Aiming at the deployment of artificial intelligence in optical networks to meet the problem of low-latency AI reasoning services,this paper designs a Deep Neural Network(DNN)model based on edge computing to provide strategies on demand.First,analyze the computing resources,network bandwidth resources,and DNN model characteristics required by services in the edge and cloud nodes in the network,and establish a cost model and a delay model.Then,comprehensively considering the computing and bandwidth resources in the network,a heuristic algorithm is proposed to flexibly split the DNN between the edge and cloud nodes to achieve load balancing under the demand for delay constraints.Finally,the simulation results show that the algorithm proposed in this paper improves the success rate of low-latency DNN inference services,while effectively reducing the bandwidth load of the entire network and the computing load of cloud nodes.(2)Propose an artificial intelligence-based fiber optic wiretap detection scheme.Aiming at the problem of fiber eavesdropping in the physical layer security of optical networks,this paper proposes a fiber eavesdropping detection scheme based on deep learning.First,the eavesdropping is simulated by light splitting,and the eye diagrams of the optical fiber signal under different light splitting ratios are collected,and then preprocessed into a training set.Secondly,establish a Convolutional Neural Network(CNN)model,and then train it to recognize eye patterns under different light splitting ratios.Finally,experiments verify the feasibility and accuracy of the scheme.The experimental results show that the scheme can still guarantee a high eavesdropping behavior recognition rate even with a low eavesdropping splitting ratio.