Research on network workload resource prediction based on hybrid model

In view of the most of the research focuses on using a single model to forecast the network workload, ignoring other factors effect on the internal network resources, lead to large amount of data implied information loss, often difficult to obtain accurate results. This paper proposes two hybrid model prediction methods. The hybrid model takes advantage of ARIMA model, Kalman filter model and BP neural network model, and combines the ARIMA model with Kalman filter and BP neural network. Experimental results show that with a single time series prediction method of integral (autoregressive moving average model, BP neural network and kalman filtering), compared two methods of hybrid model has higher prediction accuracy, effectively improve the utilization rate of resources, effectively improve the efficiency of the on-demand scheduling of virtual machine resources.