Research On Intelligent Failure Prediction And Analysis Technology Of Optical Network System

With the large-scale application of the 5th generation mobile communication technology(5G),the optical network carries more and more data traffic,and its failure prediction and analysis is very important to ensure the stable and reliable operation of the optical network.Traditional methods of failure prediction and analysis rely on expert experience and threshold system,which have limitations such as poor fault tolerance and long construction period of knowledge base,and it is difficult to achieve active defense.Moreover,in the aspect of active defense,it is difficult to identify the cause of the failure before it occurs,which brings difficulties to the actual deployment of active defense in optical networks.Therefore,intelligent failure prediction and analysis of optical networks is still a valuable research topic.In this paper,the intelligent failure prediction and analysis of optical networks are studied,and the core innovations are as follows:(1)Aiming at the traditional method of failure prediction relying on the threshold preset by expert experience,a temporal data-driven intelligent failure prediction scheme is proposed.Based on the historical operation data of optical network equipment,an improved recurrent neural network model which is good at dealing with time-series data is used for failure prediction modeling without relying on human experience.The prediction accuracy rate is higher than 99%when the equipment operation state is predicted one day in advance,and the false negative rate and false positive rate are both lower than 1%.(2)Aiming at the opacity of decision-making results and decision-making process of intelligent failure prediction and analysis,a cause-aware intelligent failure analysis scheme is proposed.Based on understanding the decision results,a failure analysis scheme based on interpretable eXtreme gradient boosting(XGBoost)is proposed.Based on the feature importance score obtained by the proposed scheme,the input features with high correlation with the faults of two types of optical network equipment are found,which are the average value of environmental temperature and the maximum value of environmental temperature.Combined with industry experience,it is inferred that the failure cause of the two types of optical equipment may be fan failure.Moreover,the influence of data imbalance on failure cause analysis is taken into account.The experimental results show that the input features with the highest correlation with the two types of optical equipment based on the proposed scheme are not affected by the data imbalance characteristics,that is,the causes most related to equipment failure will not be affected.(3)Aiming at the problem that it is difficult to identify the cause before the fault occurs,a potential failure cause identification scheme driven by attention mechanism is proposed.Firstly,a potential failure cause identification scheme based on dot-product attention mechanism is proposed,which can predict the faults and identify the potential faults of optical network equipment based on attention weight.Then,the identification of potential failure cause of the deep learning model based on additive attention mechanism and the deep learning model based on dot-product attention mechanism are studied.The experimental results show that the attention weights obtained by the two types of attention mechanisms can identify the potential failure causes of optical network equipment.Moreover,compared with the deep learning model without attention mechanism,the deep learning model based on attention mechanism also improves the performance of failure prediction.Its average accuracy and F1 score are 98.73%and 97.19%respectively,and its false negative rate and false positive rate are 2.6%and 0.91%,respectively.