CN113779759A - Data center real-time energy management method and system - Google Patents

Data center real-time energy management method and system Download PDF

Info

Publication number
CN113779759A
CN113779759A CN202110905770.0A CN202110905770A CN113779759A CN 113779759 A CN113779759 A CN 113779759A CN 202110905770 A CN202110905770 A CN 202110905770A CN 113779759 A CN113779759 A CN 113779759A
Authority
CN
China
Prior art keywords
time
data center
module
energy
function
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202110905770.0A
Other languages
Chinese (zh)
Inventor
方家琨
吴云芸
薛熙臻
艾小猛
姚伟
文劲宇
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Huazhong University of Science and Technology
Original Assignee
Huazhong University of Science and Technology
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Huazhong University of Science and Technology filed Critical Huazhong University of Science and Technology
Priority to CN202110905770.0A priority Critical patent/CN113779759A/en
Publication of CN113779759A publication Critical patent/CN113779759A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Energy or water supply
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2111/00—Details relating to CAD techniques
    • G06F2111/04—Constraint-based CAD

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Human Resources & Organizations (AREA)
  • Theoretical Computer Science (AREA)
  • Economics (AREA)
  • Physics & Mathematics (AREA)
  • Strategic Management (AREA)
  • General Physics & Mathematics (AREA)
  • Tourism & Hospitality (AREA)
  • Health & Medical Sciences (AREA)
  • Marketing (AREA)
  • General Business, Economics & Management (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Water Supply & Treatment (AREA)
  • Game Theory and Decision Science (AREA)
  • Educational Administration (AREA)
  • Development Economics (AREA)
  • Operations Research (AREA)
  • Public Health (AREA)
  • Quality & Reliability (AREA)
  • General Health & Medical Sciences (AREA)
  • Evolutionary Computation (AREA)
  • Computer Hardware Design (AREA)
  • Primary Health Care (AREA)
  • Geometry (AREA)
  • General Engineering & Computer Science (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a real-time energy management method and a real-time energy management system for a data center, which belong to the field of electric energy management, and comprise the following steps: establishing a real-time energy management model of the data center in a current preset time period, and reconstructing the model into a Markov decision process; the optimal real-time energy management strategy of the data center is optimized in a mode of solving the Bellman equation time-interval by time-interval, value function approximation is carried out by adopting three-dimensional state variables of batch processing load quantity, electric storage quantity and cold storage quantity in the queue respectively, the problem of difficulty in solving the value function is solved, the batch processing load quantity approximate function, the electric storage quantity approximate function and the cold storage quantity approximate function in the queue obtained by offline training in advance are combined to obtain a three-dimensional approximate function, the three-dimensional approximate function is substituted into the Bellman equation in the Markov decision process, the Bellman equation is solved, and an approximate global optimal decision variable set for managing the data center is obtained. Real-time energy management can be performed on a data center operating in an uncertain environment.

Description

Data center real-time energy management method and system
Technical Field
The invention belongs to the field of electric energy management, and particularly relates to a real-time energy management method and system for a data center.
Background
With the development of internet + and cloud computing, the energy consumption of data centers is rapidly increased, the problems of high energy consumption and high electricity consumption cost are increasingly prominent, and energy management and optimization of the data centers are important means for operators to improve market competitiveness. Existing data center energy management methods are mostly directed to data centers operating in a determined environment. In fact, when the data center operates, uncertain situations of different actual values and predicted values of data load, new energy output and power grid electricity price occur, and the uncertain situations can cause that the optimal decision obtained in the determined environment is no longer the optimal decision in the actual operation. Therefore, the existing method cannot perform real-time energy management on the data center operating in the uncertain environment.
When the data center runs in a day, the data center needs to make corresponding decisions according to the real-time information of the random factors, and the real-time change of the random factors can seriously affect the safety and the economy of the real-time running of the data center. How to ensure the operation economy of a data center in the real-time operation considering the uncertainty of data load, power grid electricity price and new energy output is an urgent problem to be solved.
Disclosure of Invention
Aiming at the defects and the improvement requirements of the prior art, the invention provides a data center real-time energy management method and a data center real-time energy management system, and aims to solve the problem that the existing energy management method cannot perform real-time energy management on a data center operating in an uncertain environment.
To achieve the above object, according to an aspect of the present invention, there is provided a data center real-time energy management method, including: s1, establishing a real-time energy management model based on the accumulation of the product of the difference between the energy consumption and the energy generation of the data center at each moment and the corresponding coefficient in the current preset time period, wherein the data center comprises a power supply module, an energy consumption module, an electricity storage module, a cold accumulation module and an electric energy exchange module; s2, reconstructing the real-time energy management model into a Markov decision process by taking batch processing load quantity executed by the energy consumption module, charging and discharging power of the electricity storage module and cold storage and discharge power of the cold storage module as a decision variable set; s3, respectively carrying out value function approximation on the batch processing load quantity, the electricity storage quantity and the cold storage quantity in the state variable queue in the Markov decision process, combining a batch processing load quantity approximate function, an electricity storage quantity approximate function and a cold storage quantity approximate function in the queue obtained by off-line training in advance to obtain a three-dimensional approximate function, and substituting the three-dimensional approximate function into a Bellman equation in the Markov decision process; and S4, solving the Bellman equation to obtain a global optimal decision variable set, and managing the energy consumption module, the electricity storage module and the cold accumulation module based on the global optimal decision variable set.
Further, S1 is preceded by: s0', generating a plurality of training scenes based on the day-ahead predicted values of the random factors and the error distribution information; s0', taking the first training scene as an initial scene, solving the globally optimal decision variable set of the data center under the current training scene, and taking the value function of the current training scene in the solving process as the initial value of the value function of the next training scene; and S0 ', repeatedly executing the S0' until the decision variable set of the global optimum of the data center under the last training scene is solved, wherein the value function of the last training scene in the solving process is a value function obtained by pre-discrete training.
Further, the real-time energy management model is:
Figure BDA0003203121120000021
wherein F is the total electric energy cost of the data center within the current preset time period T, and alphatAnd betatThe input energy cost coefficient and the output energy cost coefficient of the data center at the time t are respectively, delta t is the time length between two adjacent times,
Figure BDA0003203121120000022
and
Figure BDA0003203121120000023
respectively the consumed energy and the generated energy of the data center at the time t.
Still further, the energy consumption module includes an IT device and a refrigeration device, and the real-time energy management model satisfies an IT device constraint, a refrigeration device constraint, an electricity storage module constraint, and a cold storage module constraint, wherein: the IT device constraints include: service level agreement constraint, data center capacity constraint, IT equipment quantity constraint, CPU utilization rate margin constraint and IT equipment total energy consumption constraint; the refrigeration equipment restraint comprises: the energy consumption of the refrigeration equipment is restricted by total energy consumption, the energy consumption of a water chilling unit, the energy consumption of a CRAH unit and the energy consumption of a water pump; the power storage module restraint includes: the method comprises the following steps of (1) power storage charging and discharging state constraint, power storage charging and discharging power upper and lower limit constraint and power storage module capacity and lowest electric quantity level constraint; the cold storage module restraint includes: the cold storage and release state constraint, the cold storage and release power upper and lower limit constraint, the cold storage module capacity and the lowest energy level constraint and the cold water machine and cold storage module interactive constraint.
Further, the markov decision process is:
Figure BDA0003203121120000031
Figure BDA0003203121120000039
Figure BDA0003203121120000032
wherein M istIs a stand forMarkov decision process, StIs a set of state variables, xtFor the set of decision variables, WtIn order to be a set of external information,
Figure BDA0003203121120000033
is a state transition equation, CtFor the energy management objective function, Δ t is the duration between two adjacent moments, dt-ΔtThe batch load, Q, being performed for time t- Δ ttFor the batch load in the queue at time t, BatttAnd TStThe time t is the electric storage quantity of the electric storage module and the cold storage quantity, a of the cold storage module respectivelytAnd btRespectively the interactive load arriving at time t and the batch load arriving,
Figure BDA0003203121120000034
for the energy generated by the power supply module at time t, αtFor a cost parameter of the electrical energy input to the data center,
Figure BDA0003203121120000035
are respectively at、bt、
Figure BDA0003203121120000036
αtThe prediction error of (2).
Further, for solving the set S of state variables at time t + Δ tt+ΔtEquation of state transition of
Figure BDA0003203121120000037
Comprises the following steps:
St+Δt(1)＝xt (1)
Figure BDA0003203121120000038
Figure BDA0003203121120000041
Figure BDA0003203121120000042
Figure BDA0003203121120000043
where k is the index, St+Δt(k)、St(k)、xt(k)、
Figure BDA0003203121120000044
Wt+Δt(k) Are respectively St+Δt、St、xt、
Figure BDA0003203121120000045
Wt+ΔtThe k-th element of (a) is,
Figure BDA0003203121120000046
a set of ante-dated prediction values for the random factor at time t + at,
Figure BDA0003203121120000047
for the charging and discharging efficiency of the electricity storage module,
Figure BDA0003203121120000048
for the cold storage and discharge efficiency of the cold storage module,
Figure BDA0003203121120000049
are respectively at、bt、
Figure BDA00032031211200000410
αtThe predicted value of (c).
Further, the three-dimensional approximation function is:
Figure BDA00032031211200000411
Figure BDA00032031211200000412
wherein the content of the first and second substances,
Figure BDA00032031211200000413
for the set of state variables after the decision at time t,
Figure BDA00032031211200000414
for the purpose of said three-dimensional approximation function,
Figure BDA00032031211200000415
respectively as a batch processing load quantity approximate value function, an electricity storage quantity approximate value function and a cold accumulation quantity approximate value function,
Figure BDA00032031211200000416
after the decision is made at the moment t, the electric energy storage quantity of the electric energy storage module, the cold accumulation quantity of the cold accumulation module, the batch processing load quantity in the queue, va，k，tFor the slope of the k-th piecewise linear function at time t, kaIs the number of segments of the a-th piecewise linear function, alpha is 1,2,3, ra，kIs the length of the k-th segment mapped to the horizontal axis,
Figure BDA00032031211200000417
is a set of state variables that are approximated by a value function.
Further, after obtaining the three-dimensional approximation function in S3, the method further includes: solving the slope of the three-dimensional approximation function in a differential mode, and updating the slope in an iterative mode:
Figure BDA00032031211200000418
Figure BDA00032031211200000419
wherein the content of the first and second substances,
Figure BDA0003203121120000051
for the slope sample value of the kth segment of the a-th piecewise linear function at the time of the nth iteration t,
Figure BDA0003203121120000052
for the nth iteration at time t the a-th state variable,
Figure BDA0003203121120000053
for the change of the a-th state variable at time t of the nth iteration,
Figure BDA0003203121120000054
as state variables
Figure BDA0003203121120000055
And amount of change of state variable
Figure BDA0003203121120000056
An approximation function of the sum of the values,
Figure BDA0003203121120000057
as state variables
Figure BDA0003203121120000058
Is used to approximate the function of (a) to (b),
Figure BDA0003203121120000059
for the a-th state variable at the time of the nth iteration t-delta t after decision making,
Figure BDA00032031211200000510
the slope value theta of the kth segment of the a-th piecewise linear function at the time of the nth iteration t-delta tn-1The step size is updated for the slope,
Figure BDA00032031211200000511
for the slope value of the kth segment of the a-th piecewise linear function at the time of the (n-1) th iteration t-deltat,
Figure BDA00032031211200000512
the slope sampling value of the kth segment of the a-th piecewise linear function at the time t of the nth iteration is obtained.
Further, the energy management objective function is:
Ct(St，xt，Wt)＝Gb，t(St，xt，Wt)-Gs，t(St，xt，Wt)
the Bellman equation is:
Figure BDA00032031211200000513
wherein, Ct(St，xt，Wt) As said energy management objective function, Gb，t(St，xt，Wt) An income value, G, of energy output for said data centers，t(St，xt，Wt) Cost of inputting energy for said data center, Vt(St) And
Figure BDA00032031211200000514
respectively as a function of the pre-decision and post-decision values, gamma is an attenuation factor,
Figure BDA00032031211200000515
is a state variable after decision.
According to another aspect of the present invention, there is provided a data center real-time energy management system, including: the model establishing module is used for establishing a real-time energy management model based on the accumulation of the product of the difference between the energy consumption and the energy generation of the data center at each moment and the corresponding coefficient in the current preset time period, and the data center comprises a power supply module, an energy consumption module, an electricity storage module, a cold storage module and an electric energy exchange module; the reconstruction module is used for reconstructing the real-time energy management model into a Markov decision process by taking batch processing load quantity executed by the energy consumption module, charging and discharging power of the electricity storage module and cold storage and discharge power of the cold storage module as a decision variable set; the processing module is used for respectively carrying out value function approximation on the batch processing load quantity, the electric storage quantity and the cold storage quantity in the state variable queue in the Markov decision process, combining a batch processing load quantity approximate value function, an electric storage quantity approximate value function and a cold storage quantity approximate value function in the queue obtained by off-line training in advance to obtain a three-dimensional approximate value function, and substituting the three-dimensional approximate value function into a Bellman equation in the Markov decision process; and the solving and managing module is used for solving the Bellman equation to obtain a global optimal decision variable set, and managing the energy consumption module, the electricity storage module and the cold storage module based on the global optimal decision variable set.
Generally, by the above technical solution conceived by the present invention, the following beneficial effects can be obtained: reconstructing an original real-time energy management model by using Markov decision process modeling, performing value function approximation on state variables related to time-interval coupling constraints, converting a multi-time-interval optimization problem into a series of single-time-interval optimization problems, and reducing the difficulty of a data center in solving the real-time energy management problem; in addition, real-time changes of new energy processing, power grid electric energy cost parameters, interactive loads and batch processing loads in the power supply module under an uncertain environment are fully considered, and an approximate value function with excellent performance is obtained through offline training, so that an approximately globally optimal real-time energy management strategy can be obtained when the data center operates on line, and the economical efficiency of the operation of the data center is ensured; the energy storage device can be matched with multiple energy storage coordination operations to use electric energy more economically, and the energy consumption cost of the data center is reduced.
Drawings
Fig. 1 is a flowchart of a method for real-time energy management of a data center according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of a data center information flow and energy flow system provided by an embodiment of the present invention;
fig. 3A to 3E are training scene diagrams of an interactive charge day-ahead predicted value, a batch load day-ahead predicted value, a wind power day-ahead predicted value, a photovoltaic day-ahead predicted value, and a grid power price day-ahead predicted value, respectively, according to an embodiment of the present invention;
fig. 4 is an optimal power supply plan obtained by the data center real-time energy management method according to the embodiment of the present invention for the scenario shown in fig. 3;
fig. 5 is an optimal power consumption obtained by the data center real-time energy management method according to the embodiment of the present invention for the scenario shown in fig. 3;
fig. 6 is a relationship between data load processing and power grid electricity price obtained by the data center real-time energy management method according to the embodiment of the present invention for the scenario shown in fig. 3;
fig. 7 is a block diagram of a data center real-time energy management system according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. In addition, the technical features involved in the embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.
In the present application, the terms "first," "second," and the like (if any) in the description and the drawings are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order.
Fig. 1 is a flowchart of a data center real-time energy management method according to an embodiment of the present invention. Referring to fig. 1, a detailed description is given to a method for real-time energy management of a data center in this embodiment with reference to fig. 2 to 6. Referring to FIG. 1, the method includes operation S1-operation S4.
Operation S1 is to establish a real-time energy management model based on the accumulation of the product of the difference between the energy consumed and the energy generated by the data center at each moment and the corresponding coefficient in the current preset time period, where the data center includes a power module, an energy consumption module, an electricity storage module, a cold storage module, and an electric energy exchange module.
Referring to fig. 2, in the embodiment, the data center includes a power module, an energy consumption module, an electricity storage module, a cold storage module, and an electric energy exchange module. The power supply module comprises, for example, a wind farm, a photovoltaic panel, or a combination of both; the energy consumption module comprises IT equipment (such as a server cluster in FIG. 2), refrigeration equipment (such as a refrigeration system in FIG. 2) and stray power consumption items (such as lighting equipment in FIG. 2); the electricity storage module is, for example, a battery; the cold accumulation module is, for example, a cold accumulation tank; the electric energy interaction module is used for realizing electric energy interaction between the data center and an external power grid.
In this embodiment, before the data center real-time energy management method is executed, technical parameters of each element in the data center need to be collected. For the data center shown in fig. 2, the technical parameters of each element specifically include:
relevant technical parameters of the IT equipment: number of servers
Figure BDA0003203121120000081
Capacity F of each server, capacity margin σ of the server, and energy consumption p when each server is idleidlePeak power consumption p for each serverpeakMaximum queue delay time T of batch processing loadmax。
Refrigeration equipment related technical parameters: empirical model parameter gamma of energy consumption of water chiller of refrigeration system1、γ2、γ3Performance coefficient EER of water chiller and minimum power P for running of water chillerchiller min。
Relevant technical parameters of the power storage module: initial capacity of battery Batt0Allowable minimum charge level of batteryBattBattery capacity
Figure BDA0003203121120000082
Maximum charge-discharge power of battery
Figure BDA0003203121120000083
And
Figure BDA0003203121120000084
charge and discharge efficiency of battery
Figure BDA0003203121120000085
The related technical parameters of the cold accumulation module are as follows: initial cold volume level TS of cold storage tank0Allowable minimum energy level of cold storage tankTSCapacity of cold storage tank
Figure BDA0003203121120000086
Maximum storage and discharge cooling power of cold storage tank
Figure BDA0003203121120000087
And
Figure BDA0003203121120000088
cold storage and discharge efficiency of cold storage tank
Figure BDA0003203121120000089
In this embodiment, the parameters of the system server and the energy storage are shown in table 1:
TABLE 1
Figure BDA00032031211200000810
Considering factors such as data load scheduling, server dormancy, multi-type energy storage coordinated operation, interaction with a power grid and the like, the established real-time energy management model is as follows:
Figure BDA00032031211200000811
wherein F is the total electric energy cost of the data center in the current preset time period T,
Figure BDA00032031211200000812
a revenue value for the data center to output energy at time t,
Figure BDA00032031211200000813
cost of energy input to the data center at time t, αtAnd betatRespectively at time t for the data centerThe input energy cost coefficient and the output energy cost coefficient of the moment, delta t is the time length between two adjacent moments,
Figure BDA0003203121120000091
and
Figure BDA0003203121120000092
respectively the consumed energy and the generated energy of the data center at the time t. []+When the internal value is positive, the value is set to be]+And if not, taking the value of 0. Input energy cost coefficient alphatFor example, the price of purchasing electricity from the outside of the data center, the output energy cost coefficient betatFor example, the price of the output power for the data center, it should be noted that the existing α is directly utilized in the present embodimenttAnd betatIs not aligned with alphatAnd betatAny administrative controls are made.
With respect to the data center shown in figure 2,
Figure BDA0003203121120000093
and
Figure BDA0003203121120000094
respectively as follows:
Figure BDA0003203121120000095
Figure BDA0003203121120000096
Figure BDA0003203121120000097
wherein the content of the first and second substances,
Figure BDA0003203121120000098
for the total energy consumption of the IT devices in the energy consumption module,
Figure BDA0003203121120000099
is the total energy consumption of the refrigeration equipment in the energy consumption module,
Figure BDA00032031211200000910
to charge the power storage module with power,
Figure BDA00032031211200000911
for cooling power of cold storage module, PMisThe power consumption of the stray power consumption items in the power consumption module,
Figure BDA00032031211200000912
is the condition of the output of the power supply module,
Figure BDA00032031211200000913
for the situation of the wind power output,
Figure BDA00032031211200000914
in order to be in the photovoltaic output situation,
Figure BDA00032031211200000915
in order to store the amount of electricity released by the electricity storage module,
Figure BDA00032031211200000916
the cold energy released by the cold accumulation module.
The real-time energy management model meets IT equipment constraint, refrigeration equipment constraint, electricity storage module constraint and cold storage module constraint. IT equipment constraints include: service level agreement constraints, data center capacity constraints, IT equipment quantity constraints, CPU utilization margin constraints, and IT equipment total energy consumption constraints. The refrigeration equipment constraints include: the energy consumption of the refrigeration equipment is restricted by total energy consumption, the energy consumption of a water chilling unit, the energy consumption of a CRAH unit and the energy consumption of a water pump. The power storage module constraints include: the system comprises an electricity storage charging and discharging state constraint, an electricity storage charging and discharging power upper and lower limit constraint and an electricity storage module capacity and lowest electricity level constraint. The cold storage module constraint includes: the cold storage and release state constraint, the cold storage and release power upper and lower limit constraint, the cold storage module capacity and the lowest energy level constraint and the cold water machine and cold storage module interactive constraint.
Specifically, for the IT equipment, the batch processing load Q in the initial queue of the t periodtComprises the following steps:
Figure BDA00032031211200000917
wherein, BtCumulative amount of batch load arriving at the data center and processed for time t and all previous times, Dt-1The cumulative amount of batch load that arrived at the data center and was processed for time t-1 and all previous times, bτFor batch load of the server arriving at time τ, dτThe batch processing load of the server process is time τ.
The service level agreement constraints are:
Figure BDA0003203121120000101
wherein, TmaxThe maximum queue delay time for the batch load.
The data center capacity constraints are:
μt＝at+dt
0≤μt≤N
wherein, mutIs the IT demand in time period t, atAnd N is the total capacity of the data center.
The number of IT devices (servers) constraints are:
Figure BDA0003203121120000102
wherein m istNumber of servers that are active;Mthe minimum number of open servers;
Figure BDA0003203121120000103
the total number of servers in the data center.
The CPU utilization margin constraint is:
Figure BDA0003203121120000104
Figure BDA0003203121120000105
Figure BDA0003203121120000106
Figure BDA0003203121120000107
wherein the content of the first and second substances,
Figure BDA0003203121120000108
for server i CPU utilization, F for each server capacity, stσ is the capacity margin of the server for the CPU utilization of the data center.
The total energy consumption constraint of the IT equipment is as follows:
Figure BDA0003203121120000109
Figure BDA0003203121120000111
wherein the content of the first and second substances,
Figure BDA0003203121120000112
energy consumption for active servers i, pidleFor energy consumption when the server is idle, ppeakIn order for the peak power consumption of the server,
Figure BDA0003203121120000113
the total energy consumption of the IT equipment system.
The total energy consumption constraint of the refrigeration equipment is as follows:
Figure BDA0003203121120000114
wherein the content of the first and second substances,
Figure BDA0003203121120000115
in order to provide the overall energy consumption of the refrigeration system,
Figure BDA0003203121120000116
the energy consumption of the water chilling unit is reduced,
Figure BDA0003203121120000117
in order to consume the energy of the CRAH unit,
Figure BDA0003203121120000118
the energy consumption of the water pump is reduced.
The energy consumption constraint of the water chilling unit is as follows:
Figure BDA0003203121120000119
wherein, PIT maxFor the peak power consumption of IT equipment system, parameter gamma1＝0.32，γ2＝0.11，γ3When the EER is 0.63, the performance coefficient of the chiller is.
The CRAH unit energy consumption constraints are:
Figure BDA00032031211200001110
Figure BDA00032031211200001111
wherein the content of the first and second substances,
Figure BDA00032031211200001112
for the fundamental energy consumption of the CRAH unit,
Figure BDA00032031211200001113
for the heat dissipation energy consumption of the CRAH unit, f is the air flow, and the air flow f ═ f needed by the machine roommax·st，ηheatIs the coefficient of performance.
The energy consumption constraint of the water pump is as follows:
Figure BDA00032031211200001114
the charge and discharge state constraint of the electricity storage (battery) is as follows:
Figure BDA00032031211200001115
wherein the content of the first and second substances,
Figure BDA00032031211200001116
and
Figure BDA00032031211200001117
the charge and discharge states of the battery are respectively.
The upper and lower limits of the power of the stored electricity are restricted as follows:
Figure BDA00032031211200001118
Figure BDA0003203121120000121
wherein the content of the first and second substances,P chandP disrespectively the minimum charge and discharge power of the battery;
Figure BDA0003203121120000122
and
Figure BDA0003203121120000123
the charging and discharging power of the battery is respectively measured in the t period,
Figure BDA0003203121120000124
and
Figure BDA0003203121120000125
the maximum charge and discharge power of the battery is respectively.
The power storage module capacity and minimum power level constraints are:
Figure BDA0003203121120000126
Figure BDA0003203121120000127
wherein, BatttIs the energy level of the battery at the beginning of the t period,
Figure BDA0003203121120000128
in order to achieve the charge-discharge efficiency of the battery,Battis the lowest energy level of the battery,
Figure BDA0003203121120000129
is the capacity of the battery.
The cold storage and discharge state constraint is as follows:
Figure BDA00032031211200001210
wherein the content of the first and second substances,
Figure BDA00032031211200001211
and
Figure BDA00032031211200001212
respectively the cold storage state and the cold discharge state of the cold storage system.
The upper and lower limits of the cold storage and discharge power are restricted as follows:
Figure BDA00032031211200001213
Figure BDA00032031211200001214
wherein the content of the first and second substances,P standP rerespectively the minimum cold storage and discharge power of the cold storage tank,
Figure BDA00032031211200001215
and
Figure BDA00032031211200001216
the cold storage power of the cold storage tank is respectively t time period,
Figure BDA00032031211200001217
and
Figure BDA00032031211200001218
respectively the maximum cold storage and discharge power of the cold storage tank.
The capacity and minimum energy level constraints of the cold accumulation module are as follows:
Figure BDA00032031211200001219
Figure BDA00032031211200001220
wherein, TStIs the energy level at the beginning of the t period,
Figure BDA00032031211200001221
in order to improve the cold storage and discharge efficiency of the cold storage tank,TSis the lowest energy level of the cold accumulation tank,
Figure BDA00032031211200001222
is the capacity of the cold storage tank.
The interactive constraint of the water cooler and the cold accumulation module is as follows:
Figure BDA00032031211200001223
wherein, Pchiller minThe minimum power is operated for the water chiller of the refrigeration system.
In operation S2, the real-time energy management model is reconstructed into a markov decision process by taking the batch processing load executed by the energy consumption module, the charging and discharging powers of the electricity storage module, and the cold storage and discharge powers of the cold storage module as a decision variable set.
Reconstructing a real-time energy management model of the data center into a Markov decision process:
Figure BDA0003203121120000131
wherein M istFor Markov decision process, StSet of state variables, x, that need to be known to solve the data center energy management problem at time ttSet of decision variables, W, obtained when solving the data center energy management problem for time ttFor an external information set with uncertainty at time t,
Figure BDA0003203121120000132
solving the State transition equation of the State variable at the next moment for time t, CtAnd performing an energy management objective function for the data center at the time t.
Specifically, solving the data center energy management problem at time t requires knowing the batch processing load d executed at time t- Δ t of the last periodt-ΔtAnd the batch processing load Q in the queue at the time ttAnd the electric storage quantity Batt of the electric storage module at the moment ttAnd the cold accumulation amount TS of the cold accumulation module at the time ttInteractive load a arriving at time ttBatch load b arriving at time ttEnergy generated by power module at time t
Figure BDA00032031211200001310
And a cost parameter alpha of the electric energy input to the data center at the moment ttSet of state variables S at time ttComprises the following steps:
Figure BDA00032031211200001311
the batch processing load d executed by the energy consumption module in the current time period needs to be decided for solving the energy management problem of the data center at the time ttCharging power of electricity storage module
Figure BDA0003203121120000133
Discharge power of the energy storage module
Figure BDA0003203121120000134
Cold storage power of cold storage module
Figure BDA0003203121120000135
And cold discharge power of cold storage module
Figure BDA0003203121120000136
Set of decision variables x at time ttComprises the following steps:
Figure BDA0003203121120000137
considering the uncertainty of data load arrival, new energy output and power grid electricity price, and the external information set W at the moment ttAnd a set of day-ahead predictive values for random factors
Figure BDA0003203121120000138
Comprises the following steps:
Figure BDA0003203121120000139
Figure BDA0003203121120000141
wherein the content of the first and second substances,
Figure BDA0003203121120000142
in order to be able to predict the error of the interactive load,
Figure BDA0003203121120000143
in order to predict the error for a batch load,
Figure BDA0003203121120000144
the prediction error for the energy produced by the power module,
Figure BDA0003203121120000145
for the prediction error of the cost parameter of the electric energy input to the data center,
Figure BDA0003203121120000146
for a day-ahead prediction of the interactive load,
Figure BDA0003203121120000147
for a day-ahead prediction of batch load,
Figure BDA0003203121120000148
a predicted value of the energy produced by the power module in the day ahead,
Figure BDA0003203121120000149
is a predicted value of a cost parameter of the electrical energy input to the data center.
The transfer function can solve the state variable at the next moment by using the time interval coupling relation, and is specifically used for solving the state variable set S at the t + delta t momentt+ΔtEquation of state transition of
Figure BDA00032031211200001410
Comprises the following steps:
St+Δt(1)＝xt (1)
Figure BDA00032031211200001411
Figure BDA00032031211200001412
Figure BDA00032031211200001413
Figure BDA00032031211200001414
where k is the index, St+Δt(k)、St(k)、xt(k)、
Figure BDA00032031211200001415
Wt+Δt(k) Are respectively St+Δt、St、xt、
Figure BDA00032031211200001416
Wt+ΔtThe k-th element of (a) is,
Figure BDA00032031211200001417
a set of ante-dated prediction values for the random factor at time t + at,
Figure BDA00032031211200001418
for the charging and discharging efficiency of the electricity storage module,
Figure BDA00032031211200001419
the cold storage efficiency of the cold storage module is improved.
the energy management objective function of the data center energy management problem at the time t is as follows:
Ct(St，xt，Wt)＝Gb，t(St，xt，Wt)-Gs，t(St，xt，Wt)
wherein, Ct(St，xt，Wt) As said energy management objective function, Gb，t(St，xt，Wt) An income value, G, of energy output for said data centers，t(St，xt，Wt) A cost of inputting energy for the data center.
Operation S3 is to perform value function approximation on the batch processing load, the electric storage capacity, and the cold storage capacity in the state variable queue in the markov decision process, respectively, combine the batch processing load approximate function, the electric storage capacity approximate function, and the cold storage capacity approximate function in the queue obtained by offline training in advance to obtain a three-dimensional approximate function, and substitute the three-dimensional approximate function into the bellman equation in the markov decision process.
According to the embodiment of the invention, before the operation S1, an offline training is required in advance to obtain the batch processing load quantity approximation function, the stored electricity quantity approximation function and the cold storage quantity approximation function in the queue, specifically, the operation S0 '-operation S0' ".
In operation S0', a plurality of training scenarios are generated based on the day-ahead predicted values of the random factors and the error distribution information. In particular, a plurality of training scenarios are generated, for example using the monte carlo method.
In operation S0 ″, the first training scenario is used as an initial scenario, a decision variable set of the data center global optimum in the current training scenario is solved, and a value function of the current training scenario in the solving process is used as an initial value of a value function of the next training scenario.
In operation S0' ″, operation S0 ″ is repeatedly performed until a decision variable set of the data center global optimum in the last training scenario is solved, and the value function of the last training scenario in the solving process is a value function obtained by pre-discrete training.
Respectively processing batch load quantity Q in state variable queue with time interval coupling constraint by adopting monotonically increasing piecewise linear functiontBattery capacity BatttAnd cold storage tank cold energy TStPerforming value function approximation, and combining the value function into a three-dimensional approximation function, wherein the obtained three-dimensional approximation function is as follows:
Figure BDA0003203121120000151
Figure BDA0003203121120000152
wherein the content of the first and second substances,
Figure BDA0003203121120000153
for the set of state variables after the decision at time t,
Figure BDA0003203121120000154
in order to be a three-dimensional approximation function,
Figure BDA0003203121120000155
respectively as a batch processing load quantity approximate value function, an electricity storage quantity approximate value function and a cold accumulation quantity approximate value function,
Figure BDA0003203121120000156
respectively the electric quantity stored by the electric storage module after the decision at the moment t, the cold accumulation quantity of the cold accumulation module, and the batch processing load quantity in the queue, va，k，tFor the slope of the k-th piecewise linear function at time t, kaIs the number of segments of the a-th piecewise linear function, alpha is 1,2,3, ra，kIs the length of the k-th segment mapped to the horizontal axis,
Figure BDA0003203121120000157
is a set of state variables that are approximated by a value function.
After obtaining the three-dimensional approximation function in operation S3, the method further includes: solving the slope of the three-dimensional approximation function in a differential mode, and updating the slope in an iterative mode:
Figure BDA0003203121120000161
Figure BDA0003203121120000162
wherein the content of the first and second substances,
Figure BDA0003203121120000163
for the slope sample value of the kth segment of the a-th piecewise linear function at the time of the nth iteration t,
Figure BDA0003203121120000164
for the nth iteration at time t the a-th state variable,
Figure BDA0003203121120000165
for the change of the a-th state variable at time t of the nth iteration,
Figure BDA0003203121120000166
as state variables
Figure BDA0003203121120000167
And amount of change of state variable
Figure BDA0003203121120000168
An approximation function of the sum of the values,
Figure BDA0003203121120000169
as state variables
Figure BDA00032031211200001610
Is used to approximate the function of (a) to (b),
Figure BDA00032031211200001611
for the a-th state variable at the time of the nth iteration t-delta t after decision making,
Figure BDA00032031211200001612
the slope value theta of the kth segment of the a-th piecewise linear function at the time of the nth iteration t-delta tn-1The step size is updated for the slope,
Figure BDA00032031211200001613
for the n-1 th iteration t-delta tThe slope value of the kth segment of the a-th piecewise linear function,
Figure BDA00032031211200001614
the slope sampling value of the kth segment of the a-th piecewise linear function at the time t of the nth iteration is obtained.
Specifically, in this embodiment, a leveling algorithm is used to ensure monotonic increase of the slope of the piecewise linear function, that is:
Figure BDA00032031211200001615
further, substituting the three-dimensional approximation function into a Bellman equation in the Markov decision process:
Figure BDA00032031211200001616
wherein, Vt(St) And
Figure BDA00032031211200001617
respectively as a function of the pre-decision and post-decision values, gamma is an attenuation factor,
Figure BDA00032031211200001618
is a state variable after decision.
Operation S4, solving the Bellman equation to obtain a global optimal decision variable set, and managing the energy consumption module, the electricity storage module and the cold accumulation module based on the global optimal decision variable set.
In the embodiment, the data center can make a current operation decision through a value function obtained by off-line training and external real-time information obtained at the current moment, and the decision has the characteristic of approximate global optimum.
To further illustrate the real-time energy management method for the data center provided by this embodiment, 500 training scenes are generated by a monte carlo method using the day-ahead prediction information and error distribution, a three-dimensional approximation function is trained, and an approximation obtained by the training is obtainedThe value function will be used for online optimization. During online optimization, a Monte Carlo method is used to regenerate test scenes to simulate real-time information in the day. The interactive training scenario for the predictive value of charge day ahead is shown in fig. 3A, the batch processing training scenario for the predictive value of load day ahead is shown in fig. 3B, the training scenario for the predictive value of wind day ahead is shown in fig. 3C, the training scenario for the predictive value of photovoltaic day ahead is shown in fig. 3D, and the training scenario for the predictive value of electricity price day ahead is shown in fig. 3E. Specifically, the prediction errors considering the arrival of two types of data loads, the wind power output, the photovoltaic output and the power grid price are subjected to normal distribution and satisfy Wt～N(0，0.12)。
The simulation results are as follows: the total time of the off-line training is 4209.2s, which is less than two hours, and the approximate function required by the in-day on-line optimization can be solved in the day ahead. During online optimization, the average calculation time of a single test scene is 2.20s, and the method is suitable for real-time application of actual engineering. The optimal power supply plan of the data center is shown in fig. 4, the optimal power consumption is shown in fig. 5, and the relationship between the data load processing and the power grid price is shown in fig. 6. As can be seen in conjunction with fig. 4-6: when the electricity price is high, by delaying the batch processing load, the power consumed by the data center is reduced, the electric energy purchased from the outside is reduced, the discharge of the battery and the cooling of the cold storage tank are possible, and the electric energy cost is reduced; when the electricity price is low, the data center chooses to perform the batch processing load as much as possible, the consumed power increases, the electric power purchased from the outside increases, and the battery and the cold storage tank perform the charging and cold storage operations. Therefore, the real-time energy management method for the data center can enable batch processing load scheduling to actively adapt to electricity price change and simultaneously cooperate with multi-energy storage coordinated operation to more economically utilize electric energy, and reduce the energy consumption cost of the data center.
Fig. 7 is a block diagram of a data center real-time energy management system according to an embodiment of the present invention. Referring to fig. 7, the data center real-time energy management system 700 includes a model building module 710, a reconstruction module 720, a processing module 730, and a solving and management module 740.
The model building module 710 performs, for example, operation S1, and is configured to build a real-time energy management model based on an accumulation of products of differences between energy consumption and energy generation at various times of the data center in a current preset time period and corresponding coefficients, where the data center includes a power module, an energy consumption module, an electricity storage module, a cold storage module, and an electric energy exchange module.
The reconstruction module 720 performs operation S2, for example, to reconstruct the real-time energy management model into a markov decision process with the batch processing load performed by the energy consumption module, the charging and discharging powers of the electricity storage module, and the cold storage and discharge powers of the cold storage module as the decision variable sets.
The processing module 730, for example, performs operation S3, and is configured to perform value function approximation on the batch processing load, the electric storage capacity, and the cold storage capacity in the state variable queue in the markov decision process, respectively, combine the batch processing load approximation function, the electric storage capacity approximation function, and the cold storage capacity approximation function in the queue obtained through offline training in advance to obtain a three-dimensional approximation function, and substitute the three-dimensional approximation function into the bellman equation in the markov decision process.
The solving and managing module 740, for example, performs operation S4 to solve the bellman equation to obtain a globally optimal decision variable set, and manages the energy consumption module, the electricity storage module, and the cold storage module based on the globally optimal decision variable set.
The data center real-time energy management system 700 is used to perform the data center real-time energy management method in the embodiment shown in fig. 1-6. For details that are not described in the present embodiment, please refer to the data center real-time energy management method in the embodiments shown in fig. 1 to fig. 6, which is not described herein again.
It will be understood by those skilled in the art that the foregoing is only a preferred embodiment of the present invention, and is not intended to limit the invention, and that any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the scope of the present invention.

Claims (10)

1. A real-time energy management method for a data center is characterized by comprising the following steps:
s1, establishing a real-time energy management model based on the accumulation of the product of the difference between the energy consumption and the energy generation of the data center at each moment and the corresponding coefficient in the current preset time period, wherein the data center comprises a power supply module, an energy consumption module, an electricity storage module, a cold accumulation module and an electric energy exchange module;
s2, reconstructing the real-time energy management model into a Markov decision process by taking batch processing load quantity executed by the energy consumption module, charging and discharging power of the electricity storage module and cold storage and discharge power of the cold storage module as a decision variable set;
s3, respectively carrying out value function approximation on the batch processing load quantity, the electricity storage quantity and the cold storage quantity in the state variable queue in the Markov decision process, combining a batch processing load quantity approximate function, an electricity storage quantity approximate function and a cold storage quantity approximate function in the queue obtained by off-line training in advance to obtain a three-dimensional approximate function, and substituting the three-dimensional approximate function into a Bellman equation in the Markov decision process;
and S4, solving the Bellman equation to obtain a global optimal decision variable set, and managing the energy consumption module, the electricity storage module and the cold accumulation module based on the global optimal decision variable set.
2. The data center real-time energy management method of claim 1, wherein the S1 is preceded by:
s0', generating a plurality of training scenes based on the day-ahead predicted values of the random factors and the error distribution information;
s0', taking the first training scene as an initial scene, solving the globally optimal decision variable set of the data center under the current training scene, and taking the value function of the current training scene in the solving process as the initial value of the value function of the next training scene;
and S0 ', repeatedly executing the S0' until the decision variable set of the global optimum of the data center under the last training scene is solved, wherein the value function of the last training scene in the solving process is a value function obtained by pre-discrete training.
3. The data center real-time energy management method of claim 1, wherein the real-time energy management model is:
Figure FDA0003203121110000021
wherein F is the total electric energy cost of the data center within the current preset time period T, and alphatAnd betatThe input energy cost coefficient and the output energy cost coefficient of the data center at the time t are respectively, delta t is the time length between two adjacent times,
Figure FDA0003203121110000022
and
Figure FDA0003203121110000023
respectively the consumed energy and the generated energy of the data center at the time t.
4. The data center real-time energy management method of any one of claims 1-3, wherein the energy consumption modules comprise IT equipment and refrigeration equipment, and the real-time energy management model satisfies IT equipment constraints, refrigeration equipment constraints, electricity storage module constraints, and cold storage module constraints, wherein:
the IT device constraints include: service level agreement constraint, data center capacity constraint, IT equipment quantity constraint, CPU utilization rate margin constraint and IT equipment total energy consumption constraint;
the refrigeration equipment restraint comprises: the energy consumption of the refrigeration equipment is restricted by total energy consumption, the energy consumption of a water chilling unit, the energy consumption of a CRAH unit and the energy consumption of a water pump;
the power storage module restraint includes: the method comprises the following steps of (1) power storage charging and discharging state constraint, power storage charging and discharging power upper and lower limit constraint and power storage module capacity and lowest electric quantity level constraint;
the cold storage module restraint includes: the cold storage and release state constraint, the cold storage and release power upper and lower limit constraint, the cold storage module capacity and the lowest energy level constraint and the cold water machine and cold storage module interactive constraint.
5. The data center real-time energy management method of claim 1, wherein the markov decision process is:
Mt＝{St,xt,Wt,ft trans,Ct}
Figure FDA0003203121110000025
Figure FDA0003203121110000024
wherein M istFor the Markov decision process, StIs a set of state variables, xtFor the set of decision variables, WtAs a set of external information, ft transIs a state transition equation, CtFor the energy management objective function, Δ t is the duration between two adjacent moments, dt-ΔtThe batch load, Q, being performed for time t- Δ ttFor the batch load in the queue at time t, BatttAnd TStThe time t is the electric storage quantity of the electric storage module and the cold storage quantity, a of the cold storage module respectivelytAnd btRespectively the interactive load arriving at time t and the batch load arriving,
Figure FDA0003203121110000031
for the energy generated by the power supply module at time t, αtFor a cost parameter of the electrical energy input to the data center,
Figure FDA0003203121110000032
are respectively at、bt、
Figure FDA0003203121110000033
αtThe prediction error of (2).
6. The data center real-time energy management method of claim 5, used for solving a set S of state variables at time t + Δ tt+ΔtEquation of state transition of (f)t trans(St,xt,Wt+Δt) Comprises the following steps:
St+Δt(1)＝xt(1)
Figure FDA00032031211100000319
Figure FDA0003203121110000034
Figure FDA0003203121110000035
Figure FDA0003203121110000036
where k is the index, St+Δt(k)、St(k)、xt(k)、
Figure FDA0003203121110000037
Wt+Δt(k) Are respectively St+Δt、St、xt、
Figure FDA0003203121110000038
Wt+ΔtThe k-th element of (a) is,
Figure FDA0003203121110000039
a set of ante-dated prediction values for the random factor at time t + at,
Figure FDA00032031211100000310
for the charging and discharging efficiency of the electricity storage module,
Figure FDA00032031211100000311
for the cold storage and discharge efficiency of the cold storage module,
Figure FDA00032031211100000312
are respectively at、bt、
Figure FDA00032031211100000313
αtThe predicted value of (c).
7. The data center real-time energy management method of claim 5, wherein the three-dimensional approximation function is:
Figure FDA00032031211100000314
Figure FDA00032031211100000315
wherein the content of the first and second substances,
Figure FDA00032031211100000316
for the set of state variables after the decision at time t,
Figure FDA00032031211100000317
for the purpose of said three-dimensional approximation function,
Figure FDA00032031211100000318
respectively as a batch processing load quantity approximate value function, an electricity storage quantity approximate value function and a cold accumulation quantity approximate value function,
Figure FDA0003203121110000041
after the decision is made at the moment t, the electric energy storage quantity of the electric energy storage module, the cold accumulation quantity of the cold accumulation module, the batch processing load quantity in the queue, va,k,tFor the slope of the k-th piecewise linear function at time t, kaIs the number of segments of the a-th piecewise linear function, a is 1,2,3, ra,kIs the length of the k-th segment mapped to the horizontal axis,
Figure FDA0003203121110000042
is a set of state variables that are approximated by a value function.
8. The method for real-time energy management of a data center according to claim 5, wherein after obtaining the three-dimensional approximation function in S3, the method further includes:
solving the slope of the three-dimensional approximation function in a differential mode, and updating the slope in an iterative mode:
Figure FDA0003203121110000043
Figure FDA0003203121110000044
wherein the content of the first and second substances,
Figure FDA0003203121110000045
for the slope sample value of the kth segment of the a-th piecewise linear function at the time of the nth iteration t,
Figure FDA0003203121110000046
for the nth iteration at time t the a-th state variable,
Figure FDA0003203121110000047
for the change of the a-th state variable at time t of the nth iteration,
Figure FDA0003203121110000048
as state variables
Figure FDA0003203121110000049
And amount of change of state variable
Figure FDA00032031211100000410
An approximation function of the sum of the values,
Figure FDA00032031211100000411
as state variables
Figure FDA00032031211100000412
Is used to approximate the function of (a) to (b),
Figure FDA00032031211100000413
for the a-th state variable at the time of the nth iteration t-delta t after decision making,
Figure FDA00032031211100000414
the slope value theta of the kth segment of the a-th piecewise linear function at the time of the nth iteration t-delta tn-1The step size is updated for the slope,
Figure FDA00032031211100000415
for the slope value of the kth segment of the a-th piecewise linear function at the time of the (n-1) th iteration t-deltat,
Figure FDA00032031211100000416
the slope sampling value of the kth segment of the a-th piecewise linear function at the time t of the nth iteration is obtained.
9. The data center real-time energy management method of any one of claims 5-8, wherein the energy management objective function is:
Ct(St,xt,Wt)＝Gb,t(St,xt,Wt)-Gs,t(St,xt,Wt)
the Bellman equation is:
Figure FDA00032031211100000417
wherein, Ct(St,xt,Wt) As said energy management objective function, Gb,t(St,xt,Wt) An income value, G, of energy output for said data centers,t(St,xt,Wt) Cost of inputting energy for said data center, Vt(St) And
Figure FDA0003203121110000051
respectively as a function of the pre-decision and post-decision values, gamma is an attenuation factor,
Figure FDA0003203121110000052
is a state variable after decision.
10. A data center real-time energy management system, comprising:
the model establishing module is used for establishing a real-time energy management model based on the accumulation of the product of the difference between the energy consumption and the energy generation of the data center at each moment and the corresponding coefficient in the current preset time period, and the data center comprises a power supply module, an energy consumption module, an electricity storage module, a cold storage module and an electric energy exchange module;
the reconstruction module is used for reconstructing the real-time energy management model into a Markov decision process by taking batch processing load quantity executed by the energy consumption module, charging and discharging power of the electricity storage module and cold storage and discharge power of the cold storage module as a decision variable set;
the processing module is used for respectively carrying out value function approximation on the batch processing load quantity, the electric storage quantity and the cold storage quantity in the state variable queue in the Markov decision process, combining a batch processing load quantity approximate value function, an electric storage quantity approximate value function and a cold storage quantity approximate value function in the queue obtained by off-line training in advance to obtain a three-dimensional approximate value function, and substituting the three-dimensional approximate value function into a Bellman equation in the Markov decision process;
and the solving and managing module is used for solving the Bellman equation to obtain a global optimal decision variable set, and managing the energy consumption module, the electricity storage module and the cold storage module based on the global optimal decision variable set.
CN202110905770.0A 2021-08-09 2021-08-09 Data center real-time energy management method and system Pending CN113779759A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110905770.0A CN113779759A (en) 2021-08-09 2021-08-09 Data center real-time energy management method and system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110905770.0A CN113779759A (en) 2021-08-09 2021-08-09 Data center real-time energy management method and system

Publications (1)

Publication Number Publication Date
CN113779759A true CN113779759A (en) 2021-12-10

Family

ID=78837153

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110905770.0A Pending CN113779759A (en) 2021-08-09 2021-08-09 Data center real-time energy management method and system

Country Status (1)

Country Link
CN (1) CN113779759A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114116183A (en) * 2022-01-28 2022-03-01 华北电力大学 Data center service load scheduling method and system based on deep reinforcement learning

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109599856A (en) * 2018-11-12 2019-04-09 国网天津市电力公司电力科学研究院 Electric car management of charging and discharging optimization method and device in a kind of more building of microgrid
CN110458443A (en) * 2019-08-07 2019-11-15 南京邮电大学 A kind of wisdom home energy management method and system based on deeply study

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109599856A (en) * 2018-11-12 2019-04-09 国网天津市电力公司电力科学研究院 Electric car management of charging and discharging optimization method and device in a kind of more building of microgrid
CN110458443A (en) * 2019-08-07 2019-11-15 南京邮电大学 A kind of wisdom home energy management method and system based on deeply study

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
ZHI CHEN 等: "Effective Load Carrying Capability Evaluation of Renewable Energy via Stochastic Long-Term Hourly Based SCU", 《IEEE TRANSACTIONS ON SUSTAINABLE ENERGY》, vol. 6, no. 1, 31 January 2015 (2015-01-31), pages 188 - 196, XP011567727, DOI: 10.1109/TSTE.2014.2362291 *
张树本: "云计算数据中心的能耗成本建模与优化研究", 《中国博士学位论文全文库》, 15 September 2015 (2015-09-15), pages 17 - 25 *
邢静宇;张立臣;: "信息物理融合系统数据中心能量消耗模型研究", 《计算机与现代化》, no. 06, 15 June 2013 (2013-06-15) *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114116183A (en) * 2022-01-28 2022-03-01 华北电力大学 Data center service load scheduling method and system based on deep reinforcement learning
CN114116183B (en) * 2022-01-28 2022-04-29 华北电力大学 Data center service load scheduling method and system based on deep reinforcement learning

Similar Documents

Publication Publication Date Title
Wang et al. Two-stage mechanism for massive electric vehicle charging involving renewable energy
CN106532751B (en) A kind of distributed generation resource efficiency optimization method and system
CN110350518B (en) Power grid energy storage capacity demand assessment method and system for peak shaving
CN113708365A (en) Virtual power plant energy management and control optimization method and system based on end edge cloud architecture
CN113824111A (en) Energy storage capacity configuration and scheduling method in optical energy storage scene
CN116151486B (en) Multi-time-scale random optimization method and device for photovoltaic charging station with energy storage system
CN116451883A (en) Multi-energy collaborative optimization method, device, equipment and medium for park
CN114611772B (en) Multi-agent reinforcement learning-based multi-microgrid system collaborative optimization method
CN116470543A (en) Operation control method, device, equipment and medium of virtual power plant
CN113779759A (en) Data center real-time energy management method and system
Li et al. Optimal dispatch for PV-assisted charging station of electric vehicles
Liu et al. Two-stage scheduling strategy for integrated energy systems considering renewable energy consumption
CN112952876A (en) Distributed data center type substation cooperative operation and optimized scheduling method
CN112883630A (en) Day-ahead optimized economic dispatching method for multi-microgrid system for wind power consumption
CN115099725B (en) Energy management and distribution method and system for comprehensive energy system
CN115622100B (en) Enterprise power consumption distribution monitoring system and method using photovoltaic energy storage battery
CN116822719A (en) Multi-target planning method and device for power distribution network
Fang et al. Energy scheduling and decision learning of combined cooling, heating and power microgrid based on deep deterministic policy gradient
CN112561299A (en) Accurate figure system is stored up in energy source lotus of garden
CN111932401A (en) Multi-party trust service interaction method based on block chain
Benaissa et al. Standalone data-center sizing combating the over-provisioning of the it and electrical parts
Liu et al. Wind-ES Coordinated Scheduling Optimization Model Considering Electric Vehicles as Flexible Load
de Groot et al. Development of a charge path optimization controller block for a battery energy storage system
Rodríguez Gallegos et al. On the Dispatch Strategy Optimization for PV Hybrid Systems in Real Time
Li et al. Flexibility model of integrated energy system for peak load regulation demand of power grid

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination