Research On Capacity Analysis And Planning Methods For Telecom Operation And Maintenance

With the rapid development of 5G and Internet of Things technologies,the rapid growth of the number of network equipment has promoted the continuous increase of telecom business volume and business types.At the same time,telecom operation and maintenance will inevitably face the problem of optimization and upgrade,and there is an urgent need for AIOps.Capacity can be understood as the upper limit of resources preallocated to a specific application system.The capability framework of the AIOps white paper mainly describes three common application scenarios of efficiency,quality,and cost,with their own research points.Quality aspects include abnormal detection,root cause analysis,etc.;cost aspects include performance prediction,capacity management,etc.;efficiency aspects include intelligent decision-making,intelligent question and answer,etc.Existing research on AIOps mainly focuses on quality anomaly detection and root cause analysis,and there is very little research on cost capacity management.Therefore,this article combines the telecommunications industry to study the capacity management part of AIOps.The research of this thesis is mainly carried out from the aspects of capacity data classification,indicator trend prediction,capacity planning and algorithm engineering.First of all,in view of the wide variety of capacity performance indicators in telecom operation and maintenance,and the different data characteristics,a DTW-based method for dividing the data types of telecom capacity is proposed,which divides the performance indicator data series into periodic,trend and irregular types.Secondly,in view of the uneven prediction effect obtained by using the same method to predict all capacity indicator data,an indicator trend prediction method based on the characteristics of capacity data is proposed.According to the periodic capacity data,this method proposes a busy-idle distribution analysis method,and designs a periodic capacity indicator prediction model based on a bidirectional recurrent neural network that makes full use of the busy-idle distribution information.This model performs well on the data set provided by telecom operators and the Azure data set.Third,a method for capacity planning of business systems based on genetic algorithms is designed.This method uses the GBDT method to build a model,analyzes the relationship between the system's hardware performance indicators and business data indicators,and provides a basis for capacity planning;for new systems,the target call volume and business call volume prediction models are substituted into the fitness function,and the genetic algorithm is used to calculate the capacity planning plan;for the running system,propose a system capacity planning method based on indicator trend prediction,use the performance prediction method based on capacity data characteristics proposed in this thesis to predict the next moment business call volume,and use it as the target call volume.Substitute the fitness function together with the multi-dimensional indicator model to obtain the capacity planning scheme.Finally,combined with the telecommunications industry,the application of the indicator trend prediction method based on the characteristics of capacity data and the system capacity planning method in the actual AIOps capacity analysis and management system is studied.