Research And Implementation Of Prediction System For Daily Consumption Of Cloud Resource

Accurate prediction of the cloud resources daily consumption has significant sense to improve the service quality of cloud service providers.Generally,the time series of cloud resources daily usage can be classified,by its characteristics,into stationary non-random,non-stationary with minor fluctuations and non-stationary with large fluctuations.This thesis mainly predicts the time series of cloud resources daily usage with non-stationary and minor fluctuations,and then predicts the cloud resources daily consumption.The ARIMA model is commonly adopted to predict the time series with non-stationary and minor fluctuations,but using it alone always leads to low accuracy.Focusing on the fact that the cloud resources are susceptible to many factors,and taking into consideration the actual needs of daily operation and maintenance together with its sales work of JD Cloud,this thesis develops the cloud resources daily consumption prediction system based on the operation and maintenance system platform of JD Cloud.The specific research contents and results are as follows:(1)If ARIMA model is used alone for prediction,an increase in the number of period will lead to a significant increase in variance.To solve this problem,this thesis uses the typical moving average method to correct the predicted value of ARIMA model in order to reduce the error.In addition,considering the fact that there always exists singularity in the time sequence of cloud resources daily usage,the preprocessing operation is added to smooth the singularity and reduce its effect on prediction results before the ARIMA model of the daily usage time sequence is made.Based on the above-mentioned thoughts,the ARIMA-DCCR algorithm is proposed on the basis of ARIMA model,aiming at predicting the cloud resources daily consumption.(2)This thesis analyzes the experiment of ARIMA-DCCR algorithm by using the daily usage time series of JD Cloud BGP,and obtains preliminary predicted values(ARIMA model predicted values)and modified predicted values(ARIMA-DCCR algorithm predicted values).Then,this thesis make a quantitative error analysis of the two prediction series,showing that the MAE,MAPE and RMSE of ARIMA-DCCR algorithm decrease by 54.41%,52.21%and 43.67%respectively,compared with ARIMA model,which indicates the ARIMA-DCCR algorithm reduces the error of predicting the cloud resources daily usage by using ARIMA model,thereby improving the prediction accuracy of cloud resources daily consumption.(3)Adopting Object-Oriented technology,UML technology,Python and Java,this thesis provides analysis,design and implementation of the cloud resources daily consumption prediction system based on ARIMA-DCCR algorithm,and provides the software system model of requirement analysis,system overall process,system overall framework and part design of database,as well as the implementation scheme of main modules.When the cloud resources daily consumption prediction system is applied to the JD Cloud operation and maintenance and sales,the ARIMA-DCCR algorithm improves the prediction accuracy of cloud resources daily consumption,which proves that the prediction system has important practical significance in improving the efficiency of cloud resources operation and maintenance and formulating the sales strategy of cloud resources.