CN104698843B - A kind of data center's energy-saving control method based on Model Predictive Control - Google Patents
A kind of data center's energy-saving control method based on Model Predictive Control Download PDFInfo
- Publication number
- CN104698843B CN104698843B CN201510064079.9A CN201510064079A CN104698843B CN 104698843 B CN104698843 B CN 104698843B CN 201510064079 A CN201510064079 A CN 201510064079A CN 104698843 B CN104698843 B CN 104698843B
- Authority
- CN
- China
- Prior art keywords
- node
- air
- data center
- server
- model
- 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.)
- Expired - Fee Related
Links
Abstract
The present invention relates to a kind of data center's energy-saving control method based on Model Predictive Control, comprise the following steps：1) the inside distribution based on data center, using computational fluid dynamics emulation experiment, obtains the fast temperature estimation model of data center；2) power module based on server and air-conditioning, outlet temperature variation model and fast temperature estimation model, set up the state-space model of data center；3) state-space model is based on, the Model Predictive Control scheme suitable for data center is proposed；4) AR model prediction data central task amounts are used, the calculated load following for prediction data center；With reference to the task amount of prediction, performance model PREDICTIVE CONTROL MPC schemes optimize control to each equipment of data center.Compared with prior art, the present invention have the advantages that fast and accurately to estimate data center's key component gateway air themperature, be easy to verify and use, practicality it is good.
Description
Technical field
The present invention relates to a kind of data center's energy-saving control method, more particularly, to a kind of number based on Model Predictive Control
According to center energy-saving control method.
Background technology
Data center refers in communication and areas of information technology, the computer room for placing server.With modern computer
With developing rapidly for Internet technology, data center plays very important effect in all trades and professions.But in its computer room
The radiating and cooling of the equipment such as server consume substantial amounts of electric energy.Energy consumption problem has expedited the emergence of the concept of green data center.
Among the construction of green data center, efficient computing device and emerging cooling technology and reduction consumption of data center and carry
The basic measures of high energy source efficiency.Simultaneously for existing data center, optimize the running status of items of equipment, coordinating power point
The Optimized Measures matched somebody with somebody can also effectively improve the energy efficiency of data center.
Data center has complicated internal structure.The information system being made up of server and network equipment etc. passes through consumption
Electric energy completes calculating task and produces heat energy.The cooling system being made up of air conditioner in machine room etc. is by consuming electric energy discharge data center
Heat in computer room.The information system and cooling system of data center are the main bodys of whole consumption of data center.Information system and cold
But thermodynamic relation is complicated between system, while being dealt with relationship inside information system there is also complicated task.Data center
Energy consumption is too high mainly due to following three aspects.(1) data center's internal gas flow tissue is unreasonable, hot air reflux, cold air short circuit
Refrigeration is reduced etc. phenomenon；(2) work uncoordinated between multiple air-conditioning equipments, each parameter setting of air-conditioning is unreasonable, mistake
Degree consumes unnecessary electric power.(3) information equipment energy consumption is directly proportional to the growth of air conditioning energy consumption, reduces the energy of information equipment
Consumption, air-conditioning equipment just can be reduced simultaneously.
Although the energy ezpenditure of data center is larger, rarely has from data center's aspect and arranged using real-time Energy Saving Control
Apply.For in the research of consumption of data center problem, the control for information equipment and cooling device is relatively independent.Propose as
Task scheduling technique, fan control and air-conditioning control that Intel Virtualization Technology, dynamic voltage frequency regulation technology (DVFS), energy consumption are perceived
The control technologies such as system.These technologies are the energy efficiency that data center apparatus is lifted in terms of some.But it is whole from data center
Body considers, relation can be influenceed according to the heat transfer between rack and Air conditioner cabinet, in Intel Virtualization Technology, dynamic voltage regulation
(DVS) on the basis of the basic technology such as technology, fan control, airconditioning control, using method for real-time optimization control, reasonably optimizing letter
Cease plant capacity distribution, the working condition between coordination information equipment and air-conditioning equipment.So as to reduce the overall energy of data center
Consumption, improves energy efficiency.And lack research in this respect.
At present, the temperature treatment mode that existing data center generally uses is to set temperature sensor, is passively detected
The temperature focus that goes out in environment is simultaneously acted upon, it is impossible to which the distribution to temperature carries out overall estimation and prediction；In data center
Portion's thermodynamic behaviour is the bridge of contact details equipment energy consumption and cooling device energy consumption.In the thermodynamic study of data center,
Majority is modeled using the method for computational fluid dynamics (CFD) to whole data center, and optimizes the inside of data center
Arrangement.With CFD method can correct estimated data central temperature distribution, but calculated load is too big, and it can be used for counting
According to the design at center.But Temperature estimate is carried out with CFD among the real-time optimal control at available data center, it can not meet
The requirement of real-time.
In document and patent related to the present invention, document Parolini L, Sinopoli B, Krogh B H, et
al.A cyber–physical systems approach to data center modeling and control for
energy efficiency[J].Proceedings of the IEEE,2012,100(1):254-268. Primary Studies exist
Application model forecast Control Algorithm in data center, discusses potential efficiency and improves.But it only considered linearisation in document
The task processing model such as model and server power of model simplification, due to the simplicity of model, it does not consider server shape
The control of state switching at runtime and service quality.A kind of method and device (CN of computer room temperature Based Intelligent Control of Patents
102818338A) this patent proposes a kind of device of computer room temperature Based Intelligent Control, utilizes snmp management agreement, periodic polling computer room
Real-time air intake vent temperature and key chip temperature on the interior worst network equipment of environment, are adjusted by air intake vent temperature
The basal temperature of air-conditioning, further according to by the way of prediction, weighting adjusts the temperature of air-conditioning to key chip temperature；According to final
Temperature control of the air-conditioner temperature realization to air-conditioning equipment in computer room of acquisition is calculated, so that the regulation to temperature in computer room is realized,
Simultaneously there is provided a kind of method of computer room temperature Based Intelligent Control.This patent does not account for the Temperature Distribution of whole data center, only
The point inquiry temperature of high temperature is likely to occur in computer room；The control method provided only adjusts air-conditioner temperature, rather than to whole
The task distribution of data center and air-conditioner temperature carry out global control.
The content of the invention
The purpose of the present invention is exactly the defect in order to overcome above-mentioned prior art presence and provides a kind of based on model prediction
Data center's energy-saving control method of control.
The purpose of the present invention can be achieved through the following technical solutions：
A kind of data center's energy-saving control method based on Model Predictive Control, it is characterised in that comprise the following steps：
1) the inside distribution based on data center, using computational fluid dynamics emulation experiment, obtains the fast of data center
Fast temperature estimation model；
2) power module based on server and air-conditioning, outlet temperature variation model and fast temperature estimation model, set up
The state-space model of data center；
3) state-space model is based on, the Model Predictive Control scheme suitable for data center is proposed；
4) AR model prediction data central task amounts are used, the calculated load following for prediction data center；With reference to pre-
The task amount of survey, performance model PREDICTIVE CONTROL MPC schemes optimize control to each equipment of data center.
Control section includes the distribution of real-time task, server state switching and the regulation etc. in real time of air conditioner in machine room temperature；Control
During system, the data correction fast temperature gathered in real time using temperature sensor estimates the deviation of model.
The described fast temperature estimation specific acquisition process of model is as follows：
11) according to the concrete structure of data center, CFD modelings are carried out to data center；Modeling is main to consider data center
Server cabinet and air-conditioning equipment, ignore illuminating lamp heating etc. to the less part of ambient temperature effect.During modeling with reference to
The air mass flow of real data center's rack ventilation and air-conditioning, and air-conditioning refrigerating capacity and the operation work(of each rack
Rate.It may be assumed that the heat that server is produced is that the air-flow promoted by server fan is discharged herein, do not consider metal cabinet
The heat transfer of itself.When the Main physical part in data center is fixed, internal gas flow situation when it runs should have solid
Fixed pattern.The CFD of this step has modeled i.e. by Computer Simulation these patterns.
12) node division, by one or more adjacent server cabinets as a server node, data center is every
Individual cabinet type refrigeration air-conditioner is used as an air-conditioning node；
13) according to typical data center's operation conditions, operation CFD is calculated, and is obtained data center under stable state and is respectively saved
Point out inlet temperature；
System has M server node, N number of air-conditioning node, and the heating power for setting i-th of server node is
Pref,i, the air mass flow by the node is fi, wherein 1≤i≤M；The cold wind outlet temperature for setting j-th of air-conditioning node is
Tout,j, the air mass flow by the node is fj, wherein 1≤j≤N, then when CFD computings are completed, obtain at steady state,
Server node i air intake and air exit temp is respectively Tin,iAnd Tout,i, air-conditioning node j discharges room air
Temperature T at air intakein,j；
14) change each node power or temperature setting, re-start multigroup CFD experiments；Specific power or temperature are set
The mode of putting is, only changes the power or temperature setting of node in experiment every time, keeps other nodes to set with the 3rd step not
Become, i.e., on the basis of the experiment of the 3rd step, only change the power of server node at one every time, M CFD is carried out altogether and is calculated, often
The cold wind outlet temperature of air-conditioning node, carries out n times calculating at secondary change one；M+N calculating, record experiment every time are carried out altogether
As a result；
15) correlation matrix A is calculated, computing formula is as follows：
A=I- (Pnew-Pref)(Tout,new-Tout,ref)-1·K-1
In formula, A is correlation matrix, is the square formation of one (M+N) (M+N), and I is the unit matrix that (M+N) is tieed up, Pnew
Each each node power arranges value being classified as in the 4th step in once testing, PrefEach row be it is each in the 3rd step
Individual node power is set, Tout,newEach temperature for being classified as each node air outlet slit once tested in the 4th step, Tout,ref
Each row are each node air exit temp in the 3rd step, and each described node air outlet slit includes server node
Discharge hot air outlet and air-conditioning node cold-air vent；
The power of air-conditioning is calculated by the air mass flow f and Air Conditioning gateway temperature difference of air-conditioning：
Pj=ρ Cp·fj·(Tout,j-Tin,j)
ρ is atmospheric density, C in formulapFor air specific heat, Tout,jFor air-conditioning node j air temperature, Tin,jFor intake
Indoor hot air temperature；
16) data center's fast temperature estimation model is obtained, obtained temperature model is：
Tin=φ Tout
Wherein φ=K-1A ' K, K are that (M+N) ties up diagonal matrix, and each element of its leading diagonal is by KjComposition, Kj=ρ Cp·
fj, the model according to each node exit Temperature Distribution, quickly estimate each node inlet temperature be distributed, so as to for estimating
Count the influence produced during outlet temperature change to inlet temperature.
Described step 2) be specially：Assuming that data center possesses M server node, N number of air-conditioning node,
21) server node power module is set up：
Wherein Hi(t) it is the server number of units of activation in server node, S is the total number of units of node server, PidleFor clothes
The power that business device is consumed when idle, PcoFor positive coefficient, Ui(t) in moment t node server average utilization, P0To treat
The server consumption power of machine, N is server node number；
22) air-conditioning node power model is set up：
Wherein KiFor with air mass density, pass through the relevant amount of the air velocity and specific heat capacity of air-conditioning, Tin,i(t) be into
Mouth temperature, Tout,i(t) it is outlet temperature, COP is the refrigerating efficiency of air-conditioning node, and it is relevant with air conditioner outlet temperature, and C is air-conditioning
Node number；
23) server node outlet temperature variation model is set up：
Wherein kiFor the inverse of above-mentioned differential equation of first order time constant, CniFor power to the conversion coefficient of heat, Tin,i(t)
For inlet temperature, Tout,i(t) it is outlet temperature, Tin,iConstraints need to be met, inlet temperature should be less than threshold value
24) air-conditioning node exit temperature change model is set up：
1/kiFor time constant, C is conversion coefficient of the power to heat, Tin,i(t) it is inlet temperature, Tout,i(t) it is outlet
Temperature, TrefConstraints need to be met, in setting minimum thresholdAnd max-thresholdsBetween；
25) task processing model is set up：
Exemplified by handling the server of internet task.By each server node be considered a M M C queue systems,
Task of the node is arrived at when available free server is probabilistically assigned to idle server, when no server free, is also had
When task is unallocated, task is waited in queue, for i-th of server node, during wait of the task in the queuing model
Between can be simulated with equation below：
Wherein siFor the quantity for the server currently opened in i-th of server node；μiFor the task of server in node
Average treatment speed, λiThe task amount arrived at for the node, Br is the blocking rate of queue system, represents that a task adds queue
Probability, Br theoretical estimation formulas is as follows：
Therefore, the server handled for the internet task simulated by queuing model, in order to meet service quality, its
Blocking rate Br or task average latency should keep within limits；
26) state-space model：
Y (t)=Cx (t)
X (t)=[Tout,1(t),...,Tout,N+M(t)]T
Y (t)=[Tin,1(t),...,Tin,N+M(t)]T
U (t)=[P1(t),...,PN(t),Tref,1(t),...,Tref,M(t)]T
X (t) is quantity of state, and u (t) is controlled quentity controlled variable, and y (t) is output quantity, and matrix A is state matrix, and matrix B is input square
Battle array, Matrix C is output matrix.
Described utilization AR model prediction data central task amounts, the calculated load following for prediction data center is specific
For：
41) fitting historic task amount data obtain AR model parameters, based on the following MPC controller control step of AR model predictions
The change of data center's task amount in long Tp；
42) state of state-space model in controller is updated based on data center's status data.
In current time k, performance model budget control (MPC) method calculates the optimum control amount of current control period.
Performance index functionFor the power sum of server node and air-conditioning, J is defeated on state space
Enter, the nonlinear function of quantity of state and output quantity.Wherein y, u, Δ u are the predicted value of each controlling cycle in prediction step,
Wherein y is the system prediction output quantity with being obtained according to the state space difference equation iteration after discrete, and u is controlled quentity controlled variable, and Δ u is
Controlling increment, because y and u is the function that is determined by Δ u, therefore performance index function J is changed into a function J only related to Δ u
=J (Δ u).
Constraints C includes ymin≤y≤ymax, umin≤u≤umax, Δ umin≤Δu≤Δumax, xmin≤x≤xMax,About
Beam condition is determined by the requirement such as server running environment, equipment physical limit, service quality.Running environment is mainly clothes
Device inlet temperature of being engaged in limitation.Equipment physical limit includes number of devices, each side such as performance.Data center services quality standard is each
Different, for different service types, the constraint of service quality is different.Such as step 25) described in internet task processing clothes
Business device, quality of server is constrained to the average latency less than certain scope.Different restriction range settings can be produced to result
Raw influence.Server node power sum during wherein u includes server node power distribution, each controlling cycle should be big
The power sum needed for the task amount of processing prediction.
The task amount of described combination prediction, performance model PREDICTIVE CONTROL MPC schemes are carried out to each equipment of data center
Optimal control is specially：
51) by J=J, (Δ u) represents that constraints is represented by C to object function, and wherein Δ u is controlling increment, will be set
It is the Solution of Nonlinear Optimal Problem that J, constraints are C that standby optimal control, which is converted into object function,；
52) solved using SQP methods and obtain optimal solution, i.e. optimum control increment Delta u, then optimal control needed for current time k
Amount processed is u (k)=u (k-1)+Δ u (k), i.e., desired server power distributionAnd desired reference temperature is set
53) it is distributed according to desired server powerCalculate the minimum service that each server node needs
Device number of units and task apportionment ratio, it is assumed that each server executable unit task amount needs to consume power p, every server maximum work
Rate is Pmax, for server node i=1 ..., N., when expectation server power is assigned asWhen, in current control period
Need activation number of servers beIt can distribute to the maximum task amount of the server node simultaneously and be
Above-mentioned data center's energy-saving control method based on Model Predictive Control, method is mainly imitated by computational fluid dynamics
Very, temperature model parameters are calculated, model predictive control method application, and Solution of Nonlinear Optimal Problem is solved and data center is real-time
The link such as information measurement and feedback is constituted.The forecast Control Algorithm that methods described this patent is provided is data center's aspect control
Method, effective control decision is provided for lower floor's control.Be not related in method in data center control command implementation procedure and
Actuator etc..This method is not related to premised on lower floor's information collecting device can effectively provide each equipment running status information
The gatherer process and device of the every terms of information of each equipment working state.This method devises each node temperature prison of a data center
Examining system, detects that each node of data center actual temperature in controlled process is distributed, in practical application model by sensor
Each node temperature state of data center can be modified in PREDICTIVE CONTROL, improve precise control.Methods described is applied to number
Control to also need to each Floor layer Technology support in real time according to center, now this control method can be compared relative in traditional data by emulation
Advantage of the heart apparatus control method in terms of energy consumption.
Compared with prior art, the present invention has advantages below：
1) data center's key component gateway air themperature can be fast and accurately estimated, without being answered
Miscellaneous fluid mechanical emulation is calculated, and calculated load is small during operation, can meet the demand of real-time；Fluid thermodynamic based on CFD
Learn calculating process complicated, it is time-consuming longer, it is not easily applicable to predict different control orders to the change of data central temperature in control in real time
Influence.
2) it is easy to verify and use, the Temperature Distribution method of estimation provided using this patent only needs to observe data center itself
Physical characteristic, be that can obtain the estimation of Temperature Distribution by simple computation；Simultaneously in control process, temperature sensor system
Each node temperature actual value can be gathered in real time, available for temperature data in real-time amendment "current" model.
3) practicality is good, and this method uses the model predictive control method of great application prospect.Model Predictive Control is one
Roll stablized loop method is planted, is conducive to introducing optimum control among data center's efficiency control.Meeting data center
In the case that task handles quality and operating ambient temperature constraint, the overall energy consumption of data center is effectively reduced；
4) this method is managed for information equipment and cooling device simultaneously, compared to the control for relatively considering single equipment in the past
Method processed, this method optimizes each equipment running status according to the relation between information system and cooling system.From information equipment energy
Consumption and the aspect of cooling system energy consumption two are started with, and effectively reduce the overall energy consumption of data center.
5) application cost is relatively low.In expected application, in addition to it may need sensor to be mounted and top level control device, this
Method simultaneously additionally need not add and install in the data center other equipment.It can be completed with relatively low cost in data
The energy consumption control of the heart.
Brief description of the drawings
Fig. 1 is the overall flow figure of inventive algorithm；
Fig. 2 is embodiment data center inside top schematic diagram；
Fig. 3 is 3D schematic diagrames inside embodiment data center；
Fig. 4 is control structure schematic diagram；
Fig. 5 is fast temperature estimation unknown distribution of model parameters schematic diagram in embodiment.
Embodiment
The present invention is described in detail with specific embodiment below in conjunction with the accompanying drawings.
Embodiment
Fig. 1 is the idiographic flow of data center's energy-saving control method based on Model Predictive Control.For the data of application
In center, first according to data center's internal unit distribution and equipment volume, data center's computational fluid dynamics (CFD) is set up
Following steps are described in detail by model according to Fig. 1：
In step 1, data center's CFD model is set up, correlation matrix is influenceed by CFD experimental calculations temperature.CFD
Model is as shown in Figures 2 and 3.Correlation matrix is as shown in Figure 5 in embodiment.
In step 2, based on data center information equipment task processing model, the power module of items of equipment and quick temperature
Degree estimation model obtains data center's state-space model and the mathematical expression of constraints.Then step 3 is performed.
In step 3, acquisition server and running state of air conditioner information, by temperature sensor acquisition node entrance and go out
State space quantity of state in mouth temperature distribution information, correction model predictive controller.Then step 4 is performed；
In step 4, based on historic task amount data, using the following a period of time task in AR model prediction datas center
Amount, into step 5；
In steps of 5, using Model Predictive Control Algorithm, asked with nonlinear optimization under SQP Algorithm for Solving constraintss
Topic.Obtain the optimal server node power distribution of current control period and air-conditioning design temperature.Subsequently into step 6；
In step 6, based on the desired power distribution of server node.Consider number of servers and distribution task amount
Influence to service quality.Under the premise of quality of service requirement is met, using power needed for server executable unit task amount as according to
According to the activation server order of magnitude task sendout that each server node of calculating needs.Then step 7 is performed；
In step 7, judge whether whether evaluation algorithm meets end condition, end condition is data center's task herein
It is finished, closes data center.Such as no, execution step 8；
In step 8, application control makes：Each node activation server number of units, task apportionment ratio and air-conditioning design temperature.Protect
Controlled quentity controlled variable is held to next sampling period.Then step 3 is performed.
As shown in Figures 2 and 3, data center includes 28 server cabinets and 4 air conditioner in machine room units in embodiment.Wherein
Each rack is a server node, and each server node includes 40 same model servers, and each air-conditioning unit is one
Individual air-conditioning node.Rack and conditioned space position are arranged as shown in the figure inside data center.
As shown in figure 4, model of this method control structure comprising a task predicting unit, a data center's layer face is pre-
Survey controller and information acquisition system.The corresponding controller of lower floor is included in each server node and air-conditioning unit.Task is pre-
The task forecast that unit is responsible for providing following a period of time is surveyed, embodiment is 2 hours length.Model predictive controller is adopted
It is 6 steps with prediction step, it is 3 steps to control step-length.Often walk duration 20 minutes.Information acquisition system is gathered comprising equipment running status
System and node gateway temperature acquisition sensor (embodiment is emulation, this part equipment for needed for being implemented mode).Server
Node lower floor controller is responsible for controlling server state, controls activation server quantity, and distribution and scheduler task are for oral administration to node
Business device.Air-conditioning Node Controller is responsible for adjustment air-conditioning design temperature etc. and (assumes that lower floor controls accordingly in embodiment simulation process
Device perfect can perform the corresponding command).
Fig. 5 show the temperature relative influence system in the fast temperature of the data center of corresponding diagram 2 in embodiment estimation model
Matrix number.
Claims (4)
1. a kind of data center's energy-saving control method based on Model Predictive Control, it is characterised in that comprise the following steps：
1) the inside distribution based on data center, using computational fluid dynamics emulation experiment, obtains the quick temperature of data center
Degree estimation model；
2) power module based on server and air-conditioning, outlet temperature variation model and fast temperature estimation model, set up data
The state-space model at center；
3) state-space model is based on, the Model Predictive Control scheme suitable for data center is proposed；
4) AR model prediction data central task amounts are used, the calculated load following for prediction data center；With reference to prediction
Task amount, performance model PREDICTIVE CONTROL MPC schemes optimize control to each equipment of data center；
The described fast temperature estimation specific acquisition process of model is as follows：
11) according to the concrete structure of data center, CFD modelings are carried out to data center；
12) node division, by one or more adjacent server cabinets as a server node, each cabinet of data center
Formula refrigeration air-conditioner is used as an air-conditioning node；
13) according to typical data center's operation conditions, operation CFD is calculated, and is obtained each node of data center under stable state and is gone out
Inlet temperature；
System has M server node, and N number of air-conditioning node, the heating power for setting i-th of server node is Pref,i, lead to
The air mass flow for crossing the node is fi, wherein 1≤i≤M；The cold wind outlet temperature for setting j-th of air-conditioning node is Tout,j, pass through
The air mass flow of the node is fj, wherein 1≤j≤N, then when CFD computings are completed, obtain at steady state, server section
Point i air intake and air exit temp is respectively Tin,iAnd Tout,i, the air intake that air-conditioning node j discharges room air
Locate temperature Tin,j；
14) change each node power or temperature setting, re-start multigroup CFD experiments；Specific power or temperature setting side
Formula is, only changes the power or temperature setting of node in experiment every time, keeps other nodes to set and step 13) it is constant,
I.e. in step 13) on the basis of experiment, only change the power of server node at one every time, M CFD calculating is carried out altogether, every time
Only change the cold wind outlet temperature of air-conditioning node at one, carry out n times calculating；M+N calculating is carried out altogether, and record tests knot every time
Really；
15) correlation matrix A is calculated, computing formula is as follows：
A=I- (Pnew-Pref)(Tout,new-Tout,ref)-1·K-1
In formula, A is correlation matrix, is one (M+N) × (M+N) square formation, and I is the unit matrix that (M+N) is tieed up, Pnew's
It is each to be classified as step 14) in once test in each node power arranges value, PrefEach row be step 13) in it is each
Individual node power is set, Tout,newEach be classified as step 14) in the temperature of each node air outlet slit once tested,
Tout,refEach row are step 13) in each node air exit temp, described each node air outlet slit includes clothes
The discharge hot air outlet of business device node and the cold-air vent of air-conditioning node；
The power of air-conditioning is calculated by the air mass flow f and Air Conditioning gateway temperature difference of air-conditioning：
Pj=ρ Cp·fj·(Tout,j-Tin,j)
ρ is atmospheric density, C in formulapFor air specific heat, Tout,jFor air-conditioning node j air temperature, Tin,jFor the interior of intake
Hot air temperature；
16) data center's fast temperature estimation model is obtained, obtained temperature model is：
Tin=φ Tout
Wherein φ=K-1A ' K, K are that (M+N) ties up diagonal matrix, and each element of its leading diagonal is by KjComposition, Kj=ρ Cp·fj, should
Model quickly estimates that each node inlet temperature is distributed, so as to for estimating according to each node exit Temperature Distribution
The influence produced during mouth temperature change to inlet temperature.
2. a kind of data center's energy-saving control method based on Model Predictive Control according to claim 1, its feature exists
In described step 2) be specially：
Assuming that data center possesses M server node, N number of air-conditioning node,
21) server node power module is set up：
Pi(t)=Hi(t)[Pidle+PcoUi(t)]+[S-Hi(t)]Po, i=1 ..., N
Wherein Hi(t) it is the server number of units of activation in server node, S is the total number of units of node server, PidleFor server
The power consumed when idle, PcoFor positive coefficient, Ui(t) in moment t node server average utilization, P0To be standby
Server consumes power, and N is server node number；
22) air-conditioning node power model is set up：
Wherein KiFor with air mass density, pass through the relevant amount of the air velocity and specific heat capacity of air-conditioning, Tin,i(t) it is entrance temperature
Degree, Tout,i(t) it is outlet temperature, COP is the refrigerating efficiency of air-conditioning node, and it is relevant with air conditioner outlet temperature, C is air-conditioning node
Number；
23) server node outlet temperature variation model is set up：
Wherein, kiIt is the inverse of above-mentioned differential equation of first order time constant, CniFor power to the conversion coefficient of heat, Tin,i(t) it is
Inlet temperature, Tout,i(t) it is outlet temperature, Tin,iConstraints need to be met, inlet temperature should be less than threshold value24) set up empty
Point of adjustment outlet temperature variation model：
1/kiFor time constant, CniFor power to the conversion coefficient of heat, Tin,i(t) it is inlet temperature, Tout,i(t) it is outlet temperature
Degree, TrefConstraints need to be met, in setting minimum thresholdTre fAnd max-thresholdsBetween；
25) task processing model is set up：
By each server node be considered a M M C queue systems, the task of the node is arrived at when available free server
It is probabilistically assigned to idle server, when no server free, when also task is unallocated, task is waited in queue, right
In i-th of server node, stand-by period of the task in the queuing model can be simulated with equation below：
Wherein siFor the quantity for the server currently opened in i-th of server node；μiIt is averaged for the task of server in node
Processing speed, λiThe task amount arrived at for the node, Br is the blocking rate of queue system, represents that a task adds the general of queue
Rate, Br theoretical estimation formulas is as follows：
Therefore, the server handled for the internet task simulated by queuing model, in order to meet service quality, it blocks
Rate Br or task average latency should keep within limits；
26) state-space model：
Y (t)=Dx (t)
X (t)=[TOut, 1..., T (t)Out, N+M(t)]T
Y (t)=[TIn, 1..., T (t)In, N+M(t)]T
U (t)=[P1..., P (t)N(t), TRef, 1..., T (t)Ref, M(t)]T
X (t) is quantity of state, and u (t) is controlled quentity controlled variable, and y (t) is output quantity, and matrix A is state matrix, and matrix B is input matrix, square
Battle array D is output matrix.
3. a kind of data center's energy-saving control method based on Model Predictive Control according to claim 2, its feature exists
In, described utilization AR model prediction data central task amounts, the calculated load following for prediction data center is specially：
41) fitting historic task amount data obtain AR model parameters, based on the following MPC controller control step-length Tp of AR model predictions
The change of interior data center's task amount；
42) state of state-space model in controller is updated based on data center's status data.
4. a kind of data center's energy-saving control method based on Model Predictive Control according to claim 3, its feature exists
In the task amount that described combination is predicted, performance model PREDICTIVE CONTROL MPC schemes optimize control to each equipment of data center
System is specially：
51) by J=J, (Δ u) represents that constraints is represented by E to object function, and wherein Δ u is controlling increment, and equipment is excellent
Change that to control to be converted into object function be the Solution of Nonlinear Optimal Problem that J, constraints are E；
52) solved using SQP methods and obtain optimal solution, i.e. optimum control increment Delta u, then optimum control amount needed for current time k
For u (k)=u (k-1)+Δ u (k), i.e., desired server power distributionAnd desired reference temperature is set
53) it is distributed according to desired server powerCalculate the minimum service device number of units that each server node needs
And task apportionment ratio, it is assumed that each server executable unit task amount needs to consume power p, and every server peak power is
Pmax, for server node i=1 ..., N., when expectation server power is assigned asWhen, needed in current control period
The number of servers of activation isIt can distribute to the maximum task amount of the server node simultaneously and be
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510064079.9A CN104698843B (en) | 2015-02-06 | 2015-02-06 | A kind of data center's energy-saving control method based on Model Predictive Control |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510064079.9A CN104698843B (en) | 2015-02-06 | 2015-02-06 | A kind of data center's energy-saving control method based on Model Predictive Control |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104698843A CN104698843A (en) | 2015-06-10 |
| CN104698843B true CN104698843B (en) | 2017-07-11 |
Family
ID=53346087
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510064079.9A Expired - Fee Related CN104698843B (en) | 2015-02-06 | 2015-02-06 | A kind of data center's energy-saving control method based on Model Predictive Control |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104698843B (en) |
Families Citing this family (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106507640B (en) * | 2016-10-13 | 2018-12-04 | 内蒙古工业大学 | A kind of server management method of green data center temperature sensing |
| CN106681453A (en) * | 2016-11-24 | 2017-05-17 | 电子科技大学 | Dynamic heat treatment method of high-performance multi-core microprocessor |
| CN106659080B (en) * | 2016-12-27 | 2018-11-13 | 贵州电网有限责任公司信息中心 | A kind of analysis method for data center's automation refrigeration |
| CN106949598B (en) * | 2017-03-15 | 2019-07-30 | 华北电力大学 | Network center's machine room energy-saving optimization method when network traffic load changes |
| CN107681656B (en) * | 2017-09-27 | 2019-09-13 | 华中科技大学 | A kind of congestion cost bi-level programming method considering real time execution risk |
| CN108089440A (en) * | 2017-12-06 | 2018-05-29 | 北京百度网讯科技有限公司 | Energy-saving control method and device |
| CN108317683A (en) * | 2018-01-19 | 2018-07-24 | 四川斐讯信息技术有限公司 | A kind of prediction technique and system of indoor temperature and humidity |
| CN110188932A (en) * | 2019-05-20 | 2019-08-30 | 国核电力规划设计研究院有限公司 | Consumption of data center prediction technique based on evaluation optimization |
| CN110244797B (en) * | 2019-05-22 | 2022-04-05 | 平安科技（深圳）有限公司 | Computer room temperature control method and device, computer equipment and storage medium |
| WO2021042339A1 (en) * | 2019-09-05 | 2021-03-11 | 阿里巴巴集团控股有限公司 | Heat dissipation control and model training method, device and system, and storage medium |
| JP7258172B2 (en) * | 2019-10-07 | 2023-04-14 | 三菱電機株式会社 | Air conditioner controller, air conditioner and program |
| CN111174375B (en) * | 2019-12-11 | 2021-02-02 | 西安交通大学 | Data center energy consumption minimization-oriented job scheduling and machine room air conditioner regulation and control method |
| US11314212B2 (en) | 2020-01-27 | 2022-04-26 | Kyndryl, Inc. | HTM-based predictions for system behavior management |
| CN114580688A (en) * | 2020-11-30 | 2022-06-03 | 中兴通讯股份有限公司 | Control model optimization method of water cooling system, electronic equipment and storage medium |
| CN113091262B (en) * | 2021-04-12 | 2022-08-09 | 国家计算机网络信息与安全管理中心 | Data center temperature and humidity set value determination method based on model predictive control |
| CN115875809A (en) * | 2021-09-26 | 2023-03-31 | 中国移动通信集团浙江有限公司 | Energy-saving method and device for heat exchange equipment of machine room and computer readable storage medium |
| CN113961410B (en) * | 2021-10-29 | 2023-11-14 | 苏州浪潮智能科技有限公司 | Digital twinning-based debugging method and system for immersed liquid cooling server |
| CN114459134B (en) * | 2022-01-14 | 2023-11-28 | 科华数据股份有限公司 | Air conditioner control method, control terminal and computer readable storage medium |
| CN117330205A (en) * | 2023-10-23 | 2024-01-02 | 广州市资拓科技有限公司 | IDC environment monitoring and early warning method and system and storage medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPS60129845A (en) * | 1983-12-16 | 1985-07-11 | Fujitsu Ltd | Control system of information processing device |
| CN102393628A (en) * | 2011-09-14 | 2012-03-28 | 哈尔滨工业大学 | Dynamic matrix control (DMC) engineering method based on model simplification and prediction error correction |
| CN102818338A (en) * | 2012-08-07 | 2012-12-12 | 杭州华三通信技术有限公司 | Method and device for intelligently controlling temperature of machine room |
| CN102841540A (en) * | 2012-09-10 | 2012-12-26 | 广东电网公司电力科学研究院 | MMPC-based supercritical unit coordination and control method |
| US8560657B2 (en) * | 2003-05-16 | 2013-10-15 | Time Warner Cable Enterprises Llc | Data transfer application monitor and controller |
-
2015
- 2015-02-06 CN CN201510064079.9A patent/CN104698843B/en not_active Expired - Fee Related
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPS60129845A (en) * | 1983-12-16 | 1985-07-11 | Fujitsu Ltd | Control system of information processing device |
| US8560657B2 (en) * | 2003-05-16 | 2013-10-15 | Time Warner Cable Enterprises Llc | Data transfer application monitor and controller |
| CN102393628A (en) * | 2011-09-14 | 2012-03-28 | 哈尔滨工业大学 | Dynamic matrix control (DMC) engineering method based on model simplification and prediction error correction |
| CN102818338A (en) * | 2012-08-07 | 2012-12-12 | 杭州华三通信技术有限公司 | Method and device for intelligently controlling temperature of machine room |
| CN102841540A (en) * | 2012-09-10 | 2012-12-26 | 广东电网公司电力科学研究院 | MMPC-based supercritical unit coordination and control method |
Non-Patent Citations (5)
| Title |
|---|
| A Cyber–Physical Systems Approach to Data Center Modeling and Control for Energy Efficiency;Luca Parolini等;《Proceedings of the IEEE》;20120131;第100卷(第1期);第254-268页 * |
| Model Predictive Control of Data Centers in the Smart Grid Scenario;Luca Parolini等;《Preprints of the 18th IFAC World Congress》;20110902;第10505-10510页 * |
| Predictive Control for Dynamic Resource Allocation in Enterprise Data Centers;Wei Xu等;《10th IEEE/IFIP Network Operations and Management Symposium》;20060407;第1-11页 * |
| 基于AR模型时延预测的改进GPC网络控制算法;时维国等;《控制与决策》;20120331;第27卷(第3期);第477-480页 * |
| 基于CFD非稳态模型的温室温度预测控制;周伟等;《农业机械学报》;20141231;第45卷(第12期);第335-340页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104698843A (en) | 2015-06-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104698843B (en) | A kind of data center's energy-saving control method based on Model Predictive Control | |
| CN109800066B (en) | Energy-saving scheduling method and system for data center | |
| Macarulla et al. | Implementation of predictive control in a commercial building energy management system using neural networks | |
| CN101782258B (en) | Energy-saving method for air conditioner | |
| Beghi et al. | A PSO-based algorithm for optimal multiple chiller systems operation | |
| US8155793B2 (en) | System and method for controlling air conditioning facilities, and system and method for power management of computer room | |
| Zhong et al. | Energy-aware modeling and scheduling for dynamic voltage scaling with statistical real-time guarantee | |
| CN106949598B (en) | Network center's machine room energy-saving optimization method when network traffic load changes | |
| CN108321793B (en) | Active power distribution network modeling and optimal scheduling method integrating flexible loads of intelligent building | |
| WO2016011937A1 (en) | Temperature and humidity control method and apparatus for air-conditioner | |
| Chang | An innovative approach for demand side management—optimal chiller loading by simulated annealing | |
| CN104049716B (en) | Computer energy-saving method and system combined with temperature sensing | |
| CN104158754B (en) | Based on heat load data center's adaptivity Power Management method in a balanced way | |
| WO2023160110A1 (en) | System frequency modulation method and system for thermostatically controlled load cluster, and electronic device and storage medium | |
| CN101655272A (en) | Energy-saving control management system of network central air conditioner and method thereof | |
| Afram et al. | Effects of dead-band and set-point settings of on/off controllers on the energy consumption and equipment switching frequency of a residential HVAC system | |
| CN114970358A (en) | Data center energy efficiency optimization method and system based on reinforcement learning | |
| TW201027014A (en) | Method for managing air conditioning power consumption | |
| Hou et al. | Real-time optimal control of HVAC systems: Model accuracy and optimization reward | |
| Chen et al. | Power and thermal-aware virtual machine scheduling optimization in cloud data center | |
| Beghi et al. | Load forecasting for the efficient energy management of HVAC systems | |
| Li et al. | Reinforcement learning-based demand response strategy for thermal energy storage air-conditioning system considering room temperature and humidity setpoints | |
| Shen et al. | Advanced control framework of regenerative electric heating with renewable energy based on multi-agent cooperation | |
| Kaushik et al. | A dynamic scheduling technique to optimize energy consumption by ductless-split acs | |
| CN108224692B (en) | Consider the air-conditioning flexible control responding ability prediction technique of outside air temperature prediction error |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20170711 Termination date: 20200206 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |