Research And Implementation Of Resource Early Warning Model Combining Critical Point Prediction And Real-time Monitoring

To solve the limitations of the time series method in the forecast of critical time point,the step size is fixed,the indicators are single,and the forecast results may not include critical points,etc.,and to provide more sufficient preparation time for the flexible strategy,this thesis considers the impact of network traffic and special dates on critical time points(critical points).Proposing a resource early warning model combining critical point prediction and real-time monitoring.This model uses historical matching database as the core,uses long-period historical data to train the critical point prediction model,and builds historical matching database.The data center is regarded as a black box,and real-time monitoring of black box monitoring indicators is used to provide input data for the predictive model and complete the update of the historical matching database.Combining the two to realize a resource early warning model combining critical point prediction and real-time monitoring.The main research contents of this thesis are as follows.First,aiming at the limitations of traditional time series forecasting elastic time points,a critical point forecasting model is proposed.Dividing the multivariate time series into a collection of sub-sequences by analysing the impact of network traffic,network traffic trends,and special dates on critical time points.The KPM algorithm that combines clustering,dimensionality reduction,and matching uses the critical time difference of historically similar sub-sequences as a strong reference for the prediction of the critical point of the future sequence in coarse-grained time.Simultaneously build historical matching database.This kind of forecasting method has a wider forecasting range and more time for operation and maintenance and flexibility.Second,for the similarity measurement of time series,this thesis designs a time series distance(TSD)algorithm.A method for calculating the trend value of time series based on the change of sine angle is proposed.Based on the dynamic time warping algorithm,the time series measurement algorithm(DTWK)that combines the value series and the trend series is introduced to realize the similarity measurement of time series.Applying the TSD algorithm to the KPM algorithm to realize the prediction of critical time points.Third,to provide the input data of the critical point prediction model and to realize the update of the historical matching database,a resource early warning model combining the critical point prediction model and real-time monitoring is proposed.Providing services for predictive models and historical matching databases through real-time monitoring of black box monitoring indicators,and providing specific combinations of the two.Finally,in view of the problem that the monitoring sampling frequency is not synchronized with the time granularity of the data required by the prediction model,the time granularity synchronization processing function is used in real-time monitoring to perform statistical processing on the collected fine time granularity monitoring indicators,and the data is converted into the data required by the prediction model.Coarse time-granularity data structure to realize the structure of the input data of the predictive model.