Abstract: Aiming at the problems of highly non-stationary cloud platform resource load data and low prediction accuracy due to random noise,a cloud platform resource load prediction method is proposed by combining signal decomposition and deep learning technologies.Firstly,the original data is decomposed using empirical mode decomposition(EMD) method to obtain multiple IMF components;then a prediction model based on temporal convolutional network(TCN) is constructed to predict the IMF components separately;finally,the prediction results are combined to obtain the final prediction value.The proposed method is compared with traditional prediction methods and deep learning prediction methods,and a comparative experiment is carried out on Alibaba’sopen source data center resource monitoring log data set.Experimental data results show that the prediction errors of the proposed method reduces by 36.75%,23.5%,24.44%,and 24.53% compared with ARIMA,Bi-LSTM,GRU and TCN,respectively,and the prediction results have the best accuracy.

Key words: Cloud computing, Load prediction, Temporal convolutional network, Empirical mode decomposition