Research On Fault Localization Methods In Communication Networks Based On Logical Rules And Graph Neural Networks

Fault localization,as a crucial process in network fault management,is the process of deducing the exact source of a failure from a sequence of observed symptoms.The existing methods to solve fault localization are either based on logical rules or machine learning.However,as the communication networks become more complex,the knowledge-driven methods based on logical rules are facing the problem of low efficiency and lack of flexibility.In addition,data-driven machine learning algorithms have not been widely accepted by the industry due to their dependence on large-scale training sets and lack of explainability.This paper attempts to combine knowledge-driven methods and data-driven methods in fault localization-at the same time,use the advantages of the two methods to complete the root cause localization task,and try to use data to automatically model the propagation dynamics of fault signals in communication networks,to provide new ideas for the research of root cause localization in telecom networks and promote the development of this field.Based on the above background,the main research contents of this paper are as follows:(1)Inspired by the dual process theory in cognitive psychology,this paper proposes a dual system method,named Dual Sys for fault localization.The work is divided into two parts.Firstly,this paper implements the perception system(System 1)and logic system(System 2)for fault localization respectively.In System 1,the fault localization task is modeled as a node classification task,the Graph Neural Network is used to model and represent the communication network topology,and the classifier is used to classify the nodes on the network;In System 2,this paper models the fault localization task as a search optimization task realizes a simulator that can simulate the generation of alarms and a verifier that can calculate the similarity of fault scenarios,and finds the location where the fault is most likely to occur by searching.Second,when the two systems are combined in turn,system 2 depends on the output of system 1.When the result of system 1 is wrong,we regard it as a conflict between the two systems.To ensure the accuracy of the overall results,this paper proposes two conflict mitigation mechanisms.(2)In this paper,an automatic modeling method of network dynamics based on a residual graph neural network is proposed to automatically learn the propagation dynamics process of fault signals on communication network topology.This method provides a data-driven solution for the simulator in System 2.The proposed automatic modeling method of network dynamics mainly expresses the factors affecting the propagation of fault signals into vector form and then uses the message transmission process of Graph Neural Networks to simulate the propagation of fault signals on the communication network.At the same time,the residual structure is used to continuously change the fault signals on the communication networks,so as to more accurately simulate the propagation process of fault signals on the communication network.Through the experimental analysis realized on the real network topology,this paper verifies that Dual Sys has the same accuracy and interpretability as the knowledge-driven root cause location method on the data set from the real family communication network,and has a shorter running time.It has the advantages of both knowledge-driven methods and data-driven methods.Therefore,the method in this paper provides network operators with a promising choice for efficient fault localization.