Real Time Fault Prediction Of LTE Base Station E-RAB On Road Test Data Sets

LTE-R system is a new mobile bandwidth access standard proposed by 3GPP to meet the operational and business requirements of high-speed railway.The continuous transmission of signal plays an important role in the train operation.The use of Dingli background software can be used to conduct statistics on KPI of train signal quality data,like RSRP(Reference Signal Receiving Power,SINR(Signal to Interference plus Noise Ratio),Attach,Ping,cell handover,up-throughput and down-throughput speed,etc.Analysis of problem points requires manual playback of each LOG one by one to find the problem points.By analyzing the signaling data of base station side and UE side,the adjustment scheme is proposed.In LTE-R,,E-RAB(Radio Access satellite)is the carrier of the user plane,used to transmit voice,data and multimedia services between the user's device and the core network.After E-RAB is successfully established,a basic business is established and the user's equipment enters the business usage process.When E-RAB is successfully established,the release of the exception will cause the business line to drop,affecting the operation of the entire system.This paper mainly analyzes the road measurement data and E-RAB dropping rate,and proposes a computer solution to assist the manual to analyze faults,because this line carries the wireless reconnection and dispatching command on the railway.If the failure occurs,it will cause a large loss.The project mainly uses the test software of Dingli,combining with the CPE(Customer Premises Equipment)of Huawei to collect the data of network road test.Base station E-RAB drop rate is used as the label of the road test data according to the base station coverage.The method of time series classification is used to classify the road test data of each interval.It will get the level of dropouts between the regions.Then a detection model is created to assist the manual detection and analysis.This paper is based on the clustering method of Shapelets and SVM(Support Vector Machines)+Bagging+ CCS(Cluster Centroid SVM)classification method.It will organize each time series into multiple sub-sequences of equal length according to a certain rule to form a set of sub-sequences.Clustering algorithm is used to divide the sub-sequence into different clusters.A SVM classifier model is trained in each cluster.After splitting the new time series with the same rule,the classification of eachsub-sequence will be judged,and the classification result will be obtained by using SVM.After combining the result with the weight,the final result will be obtained.The research content of this paper mainly includes the following aspects:Research point 1: the determination of the length and interval of Shapelets in clustering and the determination of the number of clustering K.Research Point 2: the determination of the weight coefficients corresponding to each sub-classifier in the CCS method.