Cloud Task Prediction Model Based on Singular Spectrum Analysis, Empirical Mode Decomposition and Temporal Convolutional Network

As we all know, the nonlinearity and non-stationarity of the cloud task series bring great challenges for accurate prediction. To solve this problem, this paper proposes a hybrid method which consists of empirical model decomposition (EMD) and time convolutional networks (TCN) to predict the numbers of the future tasks arriving the datacenters. In this approach, the singular spectrum analysis is used to eliminate the noises and extracts the trend information of the original cloud task series; the cloud task series was decomposed into several subsequences by EMD, which attenuates the interaction between subsequences in the original sequence; and then the TCN is used to predict the sub-sequences and get the prediction result of the trend of each sub-sequences. The prediction results of all the sub-sequences are reconstructed to obtain the final prediction results. Experimental results show that the proposed EMD-TCN algorithm has higher prediction accuracy.