基于CEEMD-LSTM-TCN的云资源预测OA

Cloud resource loads are highly dynamic and non-stationary,and traditional prediction methods often suffer from lack of accuracy and modeling limitations when facing such complex sequences.In order to improve the accuracy and robustness of cloud resource utilization prediction,a combined prediction model based on the fusion of complete ensemble empirical modal decomposition(CEEMD),long-short-term memory network(LSTM)and temporal convolutional network(TCN)is proposed.In this model,the CEEMD is used to decompose the original CPU utilization time series into a number of modal components with different frequency features,and classify them into two categories of high frequency and low frequency by means of the dominant frequency thresholds.In allusion to different frequency components,TCN and LSTM sub-models are constructed respectively for modeling:the high-frequency component can capture the short-term disturbance characteristics by means of TCN,and the low-frequency component can conduct the long-term trend modeling by means of LSTM.The prediction results of each sub-model are fused to obtain the final load prediction value.The experiments are conducted on the AliCloud dataset,and the results show that the proposed model are better than the comparison model in a number of evaluation metrics such as MSE,RMSE,MAE,and the coefficient of determination(R²),and has higher prediction accuracy and fitting ability,especially maintaining good stability and consistency in the mutation interval.This method can provide a strong technical support for the dynamic scheduling and elastic allocation of resources in cloud computing platforms.