Research On Optical Network Traffic Analysis And Intelligent Optical Transceiver Technology Based On Machine Learning

In this era of rapid Internet development,optical network technology is facing more challenges.On the one hand,it is necessary to conduct a comprehensive analysis of optical network traffic data,summarize the laws,guide the rational allocation of optical network resources,optimize optical network transmission performance.On the other hand,more intelligent and refined resource management methods are needed to adapt to the current complicated optical network environment with rapid traffic growth and strong services suddenness.At first,this thesis conducts an in-depth study on the synthesis of optical network traffic data,and proposes adaptive traffic data augmentation scheme in different scenarios.Next,this thesis proposes a prediction algorithm that combines the traffic augmentation scheme with the traditional optical network traffic prediction technology,which can improve the traffic prediction effect.Finally,a prediction-based intelligent optical transceiver solution is proposed,which can significantly reduce the network blocking rate and improve the performance of the optical network.The main work and innovations of the thesis are as follows:Firstly,research and analyze the characteristics of optical network traffic.In view of the poor dynamic performance of traditional optical network traffic synthesis methods and the need to rely on expert experience to set traffic model parameters,an adaptive optical network traffic augmentation scheme based on Generative Adversarial Networks(GAN)is proposed.With the help of machine learning algorithms to automatically learn the distribution of real traffic data,the adaptive optical network traffic augmentation scheme can realize adaptive expansion of traffic data in different scenarios,and we use three typical dimensions of optical network traffic such as mean,variance and Hurst exponent to evaluate the effectiveness of the augmented optical network traffic data.The experimental results show that the augmented optical network traffic data and the real traffic are similar in statistical characteristics such as mean,variance and Hurst index.Therefore,the proposed scheme can efficiently and adaptively generate augmented traffic in different optical network traffic scenarios.Secondly,the traditional optical network traffic prediction method has a weak generalization ability and poor robustness due to insufficient training traffic data,which results in poor prediction effect on the test set.To solve this problem,an optical network traffic prediction algorithm combining GAN and Long Short Term Memory(LSTM)is proposed.The experimental results show that this scheme can provide better performance than other models when there are few data sets.Lastly,the traditional optical transceiver has a certain delay in the process of switching the optical transceiver configuration,and cannot quickly switch to the corresponding configuration when a higher transmission capacity is required,resulting in an increase in the network blocking rate.Similarly,when a low transmission bandwidth is required,it cannot be quickly switched to the corresponding configuration,resulting in a decrease in spectrum resource utilization.To solve this problem,a smart optical transceiver scheme based on traffic prediction is proposed.Simulation results show that this scheme can significantly reduce the network blocking rate and improve the performance of the optical network.