Research On Intelligent Fault Diagnosis And Prediction Method For LTE-R Network

With the rapid development of national economy and the continuous progress of railway technology,“Heavy-haul freight” has become an important development direction of freight railway technology in China.In order to solve the communication demand of 20,000-ton heavy-haul train,Shuohuang Railway Company applied LTE-R(Long Term Evaluation-Railway)to heavy-haul Railway for the first time in the world.The LTE-R wireless communication network carries the core communication services of heavy-haul train.Therefore,the reliability and stability of LTE-R network directly affect the operation safety of Shuohuang railway.The intelligent fault diagnosis and prediction method for LTE-R network can effectively find the problems existing in LTE-R network,or early warning the possible problems according to the operation status of the network,which has become an important technical means to ensure the reliable and stable operation of LTE-R network system.For this purpose,this paper takes Shuohuang Railway LTE-R network system as the research object and summarizes the basic theory,related methods and evaluation metrics.Then,based on the theory of imbalanced data classification and time series prediction,this paper utilizes big data,deep learning and intelligent optimization algorithm as technical means to propose corresponding intelligent fault diagnosis and prediction methods for the problems encountered in the fault diagnosis and prediction of Shuohuang Railway LTE-R network.The research of this paper provides the necessary theoretical basis and technical means for the intelligent operation and maintenance of LTE-R network,which has a certain value of theory and application.The main research results and conclusions are shown as follows:(1)To solve the problem of inefficient storage,reading and processing of massive and high-dimensional LTE-R network operation and maintenance data,this paper propose a LTE-R handover and coverage problem detection method based on big data and Geographic Information System(GIS)technology.In order to realize the structured storage of massive and high-dimensional LTE-R network operation and maintenance data,this paper utilizes HDFS,Hive and Presto to store and read these data,and uses Spark to improve the time-consuming steps in the process of handover and coverage problem detection.Moreover,all the detected problems are displayed on the GIS system,which enables the network operation and maintenance personnel to intuitively and quickly obtain the information of the problem area.Practical application shows that this method can accurately and efficiently identify common LTE-R network handover and coverage problems from massive operation and maintenance data,and can intuitively display these problems,which has high application value.(2)In the detection of LTE-R performance degradation cell,the number of performance degradation cells is far less than that of normal cells,which can be regarded as an imbalanced binary classification problem.To solve this problem,this paper first utilizes the K-means method to transform the key performance index data of each cell,so as to build a data set of LTE-R communication performance degradation cell detection.Then,to solve the problem of imbalanced binary classification at the level of potential features of data,this paper constructs a dual encoder denosing auto-encoder neural network,and introduces the Generative Adversarial Networks(GAN)to conduct a layer-wise training.In addition,to further improve the classification performance,this paper constructs multiple optimization objectives based on Fisher criterion and AUC,and utilizes Non-Dominated Sorting Genetic Algorithm-Ⅲ(NSGA-Ⅲ)to optimize the network parameters according to these optimization objectives.Finally,we use Technique for Order Preference by Similarity to Ideal Solution(TOPSIS)to select the best one from multiple non dominated solutions as the final parameter set of the network,so as to improve the separability between the majority class and the minority class.Experiments on both public imbalanced binary data set and the degraded cell detection data set of LTE-R show that the proposed method has better classification performance than contrast methods.(3)In order to further improve the accuracy of communication performance degradation cell detection,this paper proposes a LTE-R communication performance degradation cell detection method based on the feature extraction of class imbalance sequence.Aiming at the problem of how to capture the local features of the key performance index sequence more effectively,this paper constructs a neural network structure based on convolution neural network(CNN),and through the custom convolution kernel and distance calculation layer,the network can calculate the minimum distance between each shaplets and each segment of the original sequence in the way of sliding window.Thus,the local features of sequence data can be extracted by using shaplets transformation.Then,to solve the class-imbalanced problem of sequence data,we construct a optimization objective based on Fisher criterion and apply the differential evolution algorithm to train the whole network.Experimental results show that the proposed method can detect LTE-R communication performance degradation cells very accurately,and has high application value(4)Aiming at the prediction of LTE-R cell communication performance,this paper uses the Evolved Radio Access Bearer(E-RAB)abnormal release ratio of core service carried by LTE-R network as the evaluation index of cell communication performance,and proposes a time series prediction method named PA-LSTM to predict the communication performance of LTE-R cell.Firstly,PA-LSTM utilizes the Complete Ensemble Empirical Mode Decomposition with Adaptive Noise(CEEMDAN)to preprocess the original data,so as to eliminate the high-frequency components in the original data,which can make the data smooth and the changing trend of data more obvious.Then,PA-LSTM introduces the Long Short-Term Memory(LSTM)and attention mechanism to build the prediction model.Moreover,PA-LSTM introduces Particle Swarm Optimization(PSO)algorithm to adjust the attention weight,which can improve the prediction accuracy.Experiments on several LTE-R cell communication performance data sets show that the proposed prediction method can accurately predict the communication performance of LTE-R cell,so as to provide suggestions for the operation and maintenance of LTE-R network.(5)Considering that LTE-R network can carry multiple services at the same time,and the communication signals of current cell and adjacent cell will affect each other,in order to predict the communication performance of LTE-R cell more accurately,this paper proposes a multidimensional time series prediction method for LTE-R cell communication performance prediction.This method selects the historical E-RAB abnormal release ratio sequence of the current cell and the adjacent cells as the original data,and utilizes discrete Binary Particle Swarm Optimization(BPSO)algorithm to select a set of time series with the max-relevance and min-redundancy from the multiple time series.Then,in order to fully extract the depth features contained in the selected time series,this paper designs a deep neural network structure based on LSTM and CNN,and introduces the attention mechanism to weight the extracted features.Experimental results show that the proposed method can predict the communication performance of LTE-R cells more accurately than contrast methods.