Analysis Of Energy Management For Green Data Center Based On Model Predictive Control Technologies

Driven by a growing level of environmental awareness of the general public over recent years,the energy efficiency and supply,as well as structural design of modern green data centers have become the focus of relevant scientific communities around the world.Utilising renewable energy locally and efficiently while reducing operational costs are some of the critical research areas of the energy management of future data center.To achieve the stable operation of green data centers,coordination of multi-energy systems and scheduling of tasks submitted by each user,we put forward a durable,reliable,economical,efficient and universal energy management model for green data centers.A series of scientific explorations have been conducted around the design and verification of data center system model and scheduling strategy.The main contents of our work are summarised below:First,computational tasks at green data centers are classified according to their characteristics and different types of task models are built.The energy model of the on-site photovoltaic power plants,the energy exchange model with the external power grid and the energy storage model of the green data center are also established.Together these models form the overall energy management model for green data centers.Second,an improvement is achieved for the photovoltaic power energy forecasting.By considering the influence of the intermittent and uncontrollable nature of photovoltaic power energy on data center power consumption and incorporating the data from the numerical weather prediction(NWP)model,a new photovoltaic power generation model based on the Boosting algorithm is proposed.Through the improvement of the renewable energy forecasting method,the accuracy of the prediction of the renewable energy generation is enhanced,which acts as a reliable input data for the green data center energy management model.Finally,a closed-loop energy management process based on model predictive control(MPC)is designed.By adding MPC to the energy management process and scheduling tasks accordingly,the improper task scheduling caused by the inaccurate prediction data is minimised.Our approach also takes measures to control the charging,discharging and the exchange with the external power grid of the energy storage system.Therefore the management process is constantly readjusted during the piratical execution to reduce the impact of errors in the renewable energy forecast.