| In recent years,the number and size of data centers have become larger,resulting in a significant increase in energy consumption in data centers,which can cause a series of economic and environmental problems.It is one of the main energy consumption of data centers when IT equipment consumes energy,of which server energy consumption accounts for the highest proportion,so reducing server energy consumption helps data centers save energy.Predicting the energy consumption of the server can assist the data center to implement the scheduling strategy based on the energy consumption and reduce the energy consumption of the data center.From the perspective of time series analysis,this paper conducts an in-depth study on the data center energy consumption prediction problem and improves the accuracy of the forecasting model.This paper turned the prediction of server’s energy consumption into a prediction problem in time series analysis.For the case of high dimension of time series,a method of using the selfsupervised learning paradigm for time series representation learning is proposed,and this method is applied to the commonly used time series analysis task.The effectiveness of the proposed method is verified by experiments on multiple publicly available datasets.In addition,this paper collected the server’s system metrics data in a simulated data center environment and modeled the time series energy consumption prediction task.This paper studies the use of popular Transformer models in deep learning and refines the disadvantages of Transformer models to make them suitable for time series forecasting.Combining the representation learning of time series with the Transformer model,a CPC-Transformer model is proposed,which uses contrast prediction coding for pre-training.In two datasets of different tasks collected by ourselves,the CPCTransformer model has shown excellent performance.Finally,this paper develops an algorithm library module for data center energy consumption assessment and prediction,which can provide a reference for data center management in a timely and accurate manner. |
