Research About Fault Location And Traceability Based On Supervised Learning

With the rapid development of telecommunication network,the scale of base station and core equipment in the network is larger and larger,and the network topology relationship is more and more complex.Once a network element fails,it often causes a series of alarms.The various alarm information brings great challenges to the fault location and traceability.In addition,a complete topology diagram is essential,but at present,due to the loss of information,complex relationship and other reasons,the topology relationship between some network elements is missing,which adds a lot of difficulty to the fault location and traceability,so the urgent task is to improve the topology relationship of the missing network elements,and the alarm data contains very important information,so this paper analyzes the current alarm data based on the analysis of the existing problems such as large quantity,information redundancy,time asynchrony,alarm attributes,etc.,this paper puts forward the {association analysis+classification} method to predict the missing topological relationship,trains and verifies the proposed algorithm for many times based on the actual telecom network alarm data.Firstly,aiming at the problem of redundancy of attribute information of alarm data,the alarm data is standardized and simplified.Then,aiming at the problem of time asynchrony,a method based on time sliding window is proposed to complete the preliminary analysis of alarm association.Then,the obtained association item set is sorted and standardized.According to the obtained association characteristics and the network element resource file,the data set of alarms with known topology relationship is obtained,called alarm correlation analysis preprocessing data set,it is used as the input set of the classification algorithm,and the trained model is verified by the test set.The experimental results show that the method is effective.In order to get better classification effect,the classification algorithm is replaced by CNN.According to the characteristics of each attribute type of the alarm association analysis preprocessing data set,the mapping operations such as quantization coding and scaling processing are carried out respectively.Then according to the time sequence characteristics of the data set,convld is constructed.Through several different training sets and verification sets,it is proved that the effect of using CNN is better.In order to prove the necessity of correlation analysis,the alarm number statistics preprocessing data set is proposed.Firstly,the network element ID is Cartesian product,and then the network element is used to extract the characteristics of the number of alarms occurred in every time window.According to the classification algorithm and CNN,the data set is transformed and quantized,and conv2d is constructed according to the formal characteristics of the data set.The limitations of the data set are illustrated by several different training sets and verification sets,the rationality and necessity of correlation analysis are proved.This research innovatively proposed the {association analysis+classification/CNN}joint algorithm,which proved its effectiveness,and proved the necessity of association analysis through comparative experiments;for the follow-up research,we can try to use other methods to extract features,in order to achieve better classification effect.