Research And Implementation Of Mobile Network Performance Evaluation Method For Specific Scenario

With the explosion of mobile network users and performance demand,mobile network performance evaluation,as an evaluation reference to measure the performance of mobile network,has attracted more and more attention of the maj or operators.At present,the traditional mobile network performance evaluation methods investigate the performance of mobile network from multiple dimensions by establishing a set of key performance indicators.However,the traditional methods rely heavily on expert opinions,and cannot give comprehensive evaluation results in all dimensions.Moreover,they lack flexibility in application scenario transformation.Therefore,it is imperative to establish a more intelligent,comprehensive and flexible mobile network performance evaluation system in specific scenarios.According to the background above,this thesis establishes a machine learning based mobile network performance evaluation model,which is validated on the data set of mobile network performance indicators provided by a Chinese operator.The experimental process is mainly divided into three parts:data mining,modeling and result verification.The data mining part aims at exploring the trend of each performance indicator and the correlation between them.The core of the modeling part is based on multivariate time series clustering and autoencoder clustering algorithm,which is subdivided into three steps:the first step is to calculate the similarity between the time series data based on FastDTW method and construct a multi-dimensional similarity matrix.In the second step,the auto-encoder is used to reduce the dimension of the similarity matrix,and the fusion multi-dimensional similarity matrix is obtained.The third step is to cluster the corresponding data in the similarity matrix based on K-medoids algorithm.The validation part aims verifying the validity of the proposed method by establishing a mobile network performance evaluation model on real data in the scenario of universities.Firstly,this thesis introduces the background and significance of the topic,summarizes and analyses the current research on mobile network performance evaluation methods at home and abroad,and on this basis,puts forward the detailed research content and objectives.Then,the background technologies used in this thesis are introduced,including mobile network performance evaluation,FastDTW algorithm,auto-encoder,K-medoids algorithm and so on.Then,the feasibility of different machine learning algorithms in mobile network performance evaluation is analyzed by full-scene data analysis.Based on this analysis,appropriate algorithms are selected and a set of experimental process is designed.The validity is verified by comparing with KPI evaluation system.Finally,this thesis establishes a mobile network performance evaluation model for the scenario of universities using this evaluation method,and analyses the distribution and change rules of mobile network performance in detail according to the experimental results.