Research On Federal Learning Technology Of Multi-domain Optical Network

In recent years,with the rapid development of optical communication industry,great changes have taken place in the industry.The deep integration of machine learning and optical network provides a new idea for the intelligent management and control of optical networking.However,in this process,it will inevitably face many challenges.Multi domain optical network is subject to data acquisition mechanism and management constraints,and the problem of data island appears in multi domain optical network,which makes machine learning applicable In this case,how to implement machine learning applications with excellent performance is a very important research field.Based on this,this paper puts forward the research topic of joint learning technology based on multi domain optical network.Through the method of multi domain joint learning,the knowledge acquisition mode changes from single source to multi-source,solves the problem of data island,improves the performance of single domain machine learning model,and makes simulation verification in alarm prediction and optical signal-to-noise ratio prediction.The specific work and research contents of this paper are as follows:(1)In order to solve the problem of data island in multi domain optical network,this paper introduces the concept of joint learning,analyzes how joint learning and multi domain optical network are closely combined,proposes two joint learning strategies suitable for multi domain optical network scenarios,provides multi domain joint learning options for multi domain optical network,and finally proposes a multi domain joint learning strategy which can carry multi domain optical network joint learning Joint learning network architecture.(2)An alarm prediction scheme based on joint learning of multi domain optical networks is proposed.To a certain extent,the problem of uneven distribution of alarm data characteristics caused by data island is solved.The performance of the proposed strategy is verified by simulation experiments.The strategy can alleviate the data islanding problem caused by multi domain optical network,and improve the performance of the model to a certain extent.The prediction accuracy of the two sub categories of alarm prediction is improved by 3%and 5%respectively.(3)In this paper,a hybrid machine learning model based on prior knowledge and posterior knowledge is proposed,which is superior to prior knowledge and posterior knowledge.With the support of joint learning mechanism,an optical signal-to-noise ratio prediction scheme based on joint learning strategy of multi domain optical network is proposed to optimize the performance of single domain network model.The results show that the algorithm with multi domain joint learning strategy has better performance than the model algorithm with only local data set,and the accuracy is improved by about 6%.