Resource Usage Prediction Based on BILSTM-GRU Combination Model

Abstract

With the rapid development of cloud computing, accurate resource usage prediction has become a key technology for the efficient utilization of cloud data center resources. Aiming at the problems of low prediction accuracy and long prediction time of the current load prediction model, a combined prediction model BILSTM-GRU based on bidirectional long short-term memory network (BILSTM) and gated recurrent unit (GRU) is proposed, which effectively combines BILSTM network with high prediction accuracy and short prediction time of the GRU network. It is compared and verified with various classical time series prediction algorithms on the Google cloud computing data set. Experimental results show that the mean square error (MSE) of BILSTM-GRU combined prediction model is reduced by about 5, and the prediction time is shortened by about 5% compared with the existing combined prediction model. The experimental results verify that BILSTM-GRU combined model has higher prediction accuracy and shorter prediction time, which provides an important scientific basis for automatic expansion and shrinkage of cloud computing containers using the prediction results of resource usage.