Research On Energy Management Optimization Of Data Centers With Distributed Energy

With the widespread application of cloud computing and big data around the world,the number and scale of data centers are growing,and the problems of high energy consumption and cost are becoming increasingly serious.More and more data center(DC)operators are trying to supply energy for their data centers with renewable energy sources such as wind or solar.However,renewable energy has the characteristics of intermittence and randomness,which brings many challenges to the power supply management of data center.Virtual power plant(VPP)can integrate a large number of distributed energy sources such as controllable loads,distributed generator units,and energy storage batteries,which provides an effective way for energy management in DCs.Based on stochastic programming,three levels of problems are discussed on the context of VPP:(1)energy management optimization of DC VPP(DCVPP)with interactive workload,(2)energy management optimization of DCVPP with batch workload,and(3)energy management optimization of DCVPP with demand response.(1)A mixed integer linear programming(MILP)model was established for energy management optimization of DCVPP with interactive workload The Monte Carlo simulation approach is utilized to describe the uncertainty of wind power output,market electricity price and workload demand.The MILP model aims to minimize the operation cost of DCVPP considering various physical and system-wide constraints such as workload balancing,distributed generation units,energy storage batteries and minimum on or off time of server clusters.Compared with the deterministic model using predicted values,all server clusters in the stochastic model have to be turned on to deal with the random affects;Although the stochastic model has low power exchange and the operating cost is slightly higher than the deterministic model by $72.508,it can deal with the influence of uncertainty more effectively.(2)A two-stage stochastic MILP model considering constraint of risk is proposed for energy management optimization of DCVPP with batch workload,which is aimed at optimizing operation of the power network dispatching of DCVPP.The model takes intoaccount the randomness of renewable energy and market electricity price as well as the constraints such as water price,workload balancing and distributed generation units and so on.Using conditional value at risk model to measure the impact of different risk levels on DC's day-ahead operations.The experimental results show that as the risk level decreases,the expected profit of DCVPP decreases.When the water prices increase sharply and during the off-peak hours,DCVPP turns on almost all server clusters to handle workloads.(3)A two-stage stochastic MILP model considering the constraints of demand response(DR)is proposed for energy management optimization of DCVPP with DR,which is aimed at optimizing operation of the power network dispatching of DCVPP.In order to maximize the profit of DCVPP,the model considers the constraints of workload balancing,distributed generator units,energy storage batteries,minimum switch time of server clusters,demand response and so on.Compared with the model without DR,the price of VPP and DR under the demand response model directly affect whether local users provide DR;Due to the existence of DR,the output of distributed generator units and the interaction of energy are relatively gentle,which effectively plays the role of "peak shaving and valley filling";The operation benefit is $1101.50 higher than the model without DR,which verifies the effectiveness of the proposed model.