Research On Network Equipment Indicators Prediction Based On Time Series Analysis

With the continuous development of information technology,the Internet has become an indispensable part of daily life.With the continuous deepening research on network theories and technologies,the network has become larger and larger.Such huge network is essentially composed of local area networks,so the management and analysis of the local area network has become an important research object.At the same time,with the rapid development of artificial intelligence technology in recent years,the combination of network analysis system and artificial intelligence technology has become an unstoppable trend.In the intelligent network analyzer,indicator prediction is an important component.However,in the existing systems,due to the lack of similar research,the current indicator prediction algorithm is weak and has insufficient prediction accuracy.In view of the above problems,five indicators under the representative S-CSCF registration module are selected from the network equipment CSC3300,according to the predictability of indicators and the principle of whether or not under the same module(including the number of initial registration requests,the number of re-registration times,the number of third-party registrations,the number of third-party registration successes and registered user number).The main work is as follows:(1)The widely used LSTM model based on unary time series for the experimental data is first exploited in order to verify the prediction accuracy in the test sets.This experiment is served as the benchmark.(2)A method of time series partitioning,based on the characteristics of date type,is proposed.A combined model experiment is carried out with the divided data sets.This experiment distinguishes between the holiday data and non-holiday data according to the different Gaussian distribution of holiday data and non-holiday data shown by the network equipment indicators.The corresponding training models are then designed.The experimental results show that the combined model has a significant improvement in the forecasting ability of holiday data compared to the benchmark experiment.(3)The data format conversion method based on unary time series is improved,which is suitable for multivariate time series.A multivariate time series model experiment is then carried out.This experiment is based on the characteristics of a certain synergy relationship under the same module.With the help of the LSTM model,it can handle the characteristics of multivariate time series and have a powerful learning ability of the LSTM model on data laws.A multivariate time series experiment is carried out.The experimental results show that after adding the inter-indicator synergy information to the LSTM model,the prediction ability of the multi-time series model is significantly improved when compared with the benchmark experiment.Through the above experiments,this thesis deeply explores the indicator prediction part in the scene of the intelligent network analysis system,greatly improves the accuracy of indicator prediction,and has a positive effect on the research and development of intelligent network analyzer.At the same time,the experimental ideas and experimental procedures of this thesis can be referenced for the prediction of indicators in the similar intelligent analysis systems.