Research On Datacenter Power Forecast Based On Tensor

With the rise of the development of Internet of Things applications and cloud computing,the scale of data centers is gradually expanding.As a result,the power consumption of the data center increases greatly,which not only has a great impact on the management of the data center but also causes a series of environmental problems.This paper studies the power forecast problem to implement the power-aware scheduling strategy,which is favorable to reducing the power consumption of the data center.Based on building a small data center and collecting data,we adopted a cost-sensitive strategy to overcome the cost imbalance problem of the data center power forecast.The cost-sensitive strategy can adjust the deviation of the power forecast to be suitable for the high-power forecast.The application of cost-sensitive strategies makes the forecast performance of the model better and improves the overall service level.Moreover,this paper also selects key features for different types of tasks in the data center and proposes a tensor-based dual-stage attention long-term short-term memory network model based on the cost-sensitive loss function.Compared with the popular time series forecasting model,the proposed model can effectively model the temporal pattern and feature correlatio,better capture the fluctuation of power,and improve the accuracy of forecasting.Last,we develop a data center power forecast system.The data center power forecast system is based on Spring Cloud microservices architecture,which integrates the power forecast algorithms.The system can provide stable and reliable power forecast services to improve the efficiency of data center scheduling and management.