Federated Learning Based Failure Prediction With Distributed Data For Optical Networks

As the digital economy has developed,products used the technology of digital economy require a platform for optical networks transmission with large bandwidth,high speed,low latency and high reliability,as well as consideration of both economic and social values.With the surge of transmitted data in networks and the expanding scale of networks,optical devices are increasingly deployed in the optical layer,making the structure of optical networks much more complicated than before and facing severe challenges in its reliability.If a failure occurs in an optical network,it will result in the service interruption that causes heavy financial losses.If the failure in optical networks can be predicted,the severe impact of sudden faults can be effectively avoided.In previous related studies,researchers using statistical methods or alarm triggering methods to predict the failure in optical networks.However,as optical networks has been developed into a new stage,it is difficult to accurately predict the potential failure in the networks only with traditional methods.In recent years,the development of artificial intelligence technology has provided new methods for optical failure prediction.In the view of the above case,this paper adopts the federated learning architecture with the LSTM model that is more sensitive to time series data,to conduct failure prediction.It also uses the data collected from real optical networks to verify the prediction effect.The contents and innovation of this paper are as follows:First,analyzes the forming factors and effects of isolated data islands in a systematic way,and studies the resource consumption in the modeling process.The basic situation of the collected collaborative data is introduced,including the corresponding pre-processing process.Finally,based on personal engineering experience,data are labeled for subsequent use.Second,in view of the timing features of the failure data collected in optical networks,this paper explores the use of long short-term memory neural network(Long Short-Term Memory,LSTM)for failure prediction in optical networks.The experimental results show that,compared with the general machine learning model,the algorithm used in this paper can use the failure data more effectively and mine more correlated information among data that are usually neglected,so as to improve the prediction effect.Therefore,compared with general machine learning algorithms and traditional forecasting methods,the LSTM network can realize better accuracy and stability.Third,designs a system for failure prediction based on federated learning that uses collaborative data.There are some problems in conventional failure prediction methods,such as poor data security,high modeling cost,and low prediction accuracy.In this paper,the rate of accuracy and recall are taken as measurement indicators.The results show that the prediction effect of federated learning is much close to that of centralized modeling.More importantly,the failure prediction based on federated learning can effectively break the limitation of isolated data islands,realize multi-party safety modeling,and significantly reduce modeling costs,which is conducive to the practical application of fault prediction technology.In this paper,LSTM network,which is more sensitive to time series data,is used to predict optical network faults in combination with federated learning framework.The proposed model can protect the security of modeling data meanwhile it has better performance in and modeling cost than centralized modeling.