CN105225016A - A kind of in the cloud computing system of renewable energy supply based on the energy distributing method of cooperative game - Google Patents

Abstract

The invention discloses a kind of in the cloud computing system of renewable energy supply the energy distributing method based on cooperative game, mainly comprise the following steps: the time interval calculating this energy distribution; Predict the utilisable energy in this time interval; Judge that whether energy is sufficient; Theory based on cooperative game carries out modeling to energy distribution problem, and changes into the optimization problem of a belt restraining; This optimization problem is changed into its dual problem; By this dual problem of gradient projection method solution.In the present invention, role is mainly divided into user and cloud service provider, with conventional cloud calculate unlike, service provider will consider the singularity of energy supply while providing service.It is unstable for transforming by collection outside energy the electric energy obtained, and how energy reasonable distribution is given each user when the electric energy transformed is not enough by the present invention's consideration.In this case, cloud service provider needs the demand taking into account each user, and also will consider the cost of oneself, the present invention's cooperative game carries out modeling to this simultaneously, and being converted into the optimization problem of a belt restraining, namely the solution of this optimization problem is corresponding energy distribution scheme.

Description

A kind of in the cloud computing system of renewable energy supply based on the energy distributing method of cooperative game

Technical field

The present invention relates to cloud computing, game theory relevant knowledge, particularly relate to a kind of energy distribution scheme considering fairness and harmony; Specifically a kind of in the cloud computing system of renewable energy supply based on the energy distributing method of cooperative game.

Background technology

Regenerative resource can be used as the power supply source of computer system, but can series of problems be produced like this, whether comprise the abundance of electricity, whether stablizing of electric energy, whether the operation of computer system is influenced etc., and domestic and international expert has carried out deep research with regard to these problems.JingChen devises one and considers the characteristic of task and processor by the real-time system of regenerative resource energy supply, and author proposes a Static and dynamic algorithm respectively to miss rate closing time of the energy consumption and task that reduce system.LongboHuang proposes a networking by regenerative resource energy supply, the stored energy that the external world collects is in the battery of a finite capacity, and author proposes algorithm on a line and manages the energy that collects and be reasonably allocated to by energy on each node in networking.ShaoboLiu proposes one is selected (DVFS) energy management algorithm based on dynamic voltage frequency, is intended to the service quality that the extraneous energy collected of Appropriate application improves system simultaneously.JingYang studies a communication system by renewable energy energy supply, and owing to needing when communicating the speed considering information transmission, therefore, author designed algorithm, minimizes call duration time under the prerequisite of finite energy.

Cloud computing is as a novel business prototype of computer realm, and become the important means changing computing machine use-pattern, each large IT giant is building the Cloud Server of oneself.Cloud computing model comprises the service of following level: namely infrastructure serve (IaaS), and namely platform serves (PaaS) and namely software serve (SaaS).IaaS (Infrastructure-as-a-Service): namely infrastructure serve, consumer can obtain service from perfect computer based Infrastructure by Internet; In fact PaaS (Platform-as-a-Service): namely platform serves, PaaS refer to using the platform of research and development of software as one service, submit to user with the pattern of SaaS, and therefore, PaaS is also the one application of SaaS pattern; SaaS (Software-as-a-Service): namely software serve, it is a kind of pattern being provided software by Internet, and user without the need to buying software, but rents the software of sing on web to provider, carry out management enterprise business activities.

Summary of the invention

The object of the invention is propose consider the demand of user and the cost of service provider when carrying out energy distribution for the energy distribution problem in the cloud computing system by regenerative resource energy supply, make every effort to not only a justice but also efficient allocative decision, namely a kind of in the cloud computing system of renewable energy supply based on the energy distribution scheme of cooperative game.In the process exploring the program, carry out modeling, and be converted into the optimization problem of a belt restraining with cooperative game to energy distribution, namely the solution of this optimization problem is corresponding energy distribution scheme.

The object of the present invention is achieved like this:

Based on an energy distributing method for cooperative game in the cloud computing system of renewable energy supply, feature is that the method comprises the following steps:

Step one: the time interval determining this energy distribution;

Step 2: predict the utilisable energy in this time interval, this utilisable energy is as the energy source of system;

Step 3: judge that whether energy is sufficient, go to step four time inadequate, turns distribution according to need energy time sufficient, and goes to step six;

Step 4: modeling is carried out to energy distribution with game theory, and change into the optimization problem of belt restraining, then convert it into dual problem;

Step 5: solve dual problem by gradient projection method, and the Global Optimality separated is asked in checking;

Step 6: distribute and terminate.

Described step one specifically comprises:

Steps A 1: the minimum energy required for computing system:

E _min(Δt)＝M*min{δ _is _i ²c*N _i|,i＝1,2,…,M}

Wherein: M is the number of server, s _ifor the frequency of server, N _ifor the number of task on server, c is the execution frequency of task, δ _ifor circuit efficiency factor;

Steps A 2: the maximum energy required for computing system:

E _max(Δt)＝M*max{δ _is _i ²c*N _i|,i＝1,2,…,M}

Wherein: M is the number of server, s _ifor the frequency of server, N _ifor the number of task on server, c is the execution frequency of task, δ _ifor circuit efficiency factor;

Steps A 3: the energy distribution time interval, Δ t was defined as E _harv(Δ t) is in E _min(Δ t) and E _maxtime between (Δ t):

Δt＝min{Δt|E _harv(t+Δt)∈[E _min(Δt),E _max(Δt)]}。

Described step 2 specifically comprises:

Step B1: the utilisable energy in this time interval predicted is:

$E_{harv}\left(\mathrm{\Δ}t\right)={\∫}_t^{t+\mathrm{\Δ}t}{P}_{harv}\left(t\right)dt,P_{harv}\left(\mathrm{\Δ}t\right)$

Wherein, P _harvt () is energy acquisition power.

Described step 3 specifically comprises:

Step C1: the energy that computing system needs altogether:

$E_{demand}\left(\mathrm{\Δ}t\right)=\underset{i=1}{\overset{M}{\Σ}}{E}_{cons,i}$

E _cons，i＝N _i*E _cons(τ，S _i)

$E_{cons}(\mathrm{\τ},s_i)={\mathrm{\δ}}_i*{s_i}^3*\frac{c}{s_i}={\mathrm{\δ}}_i*{s_i}^2*c$

Wherein, E _demand(Δ t) gross energy required for M server, E _{cons, i}be the energy that i-th server needs, E _cons(τ, s _i) energy that consumes for each task, when task is s in frequency _iserver on when running; N _iit is the number of task on i-th server;

Step C2: judge that whether energy is sufficient

Work as E _harv(Δ t) > E _demandtime (Δ t), system capacity is sufficient, otherwise system capacity is inadequate.

Described step 4 specifically comprises:

Step D1: modeling is carried out to energy distribution problem with game theory, and be converted into the optimization problem of a belt restraining, that is:

$max:\underset{i=1,2,\mathrm{...},M}{\Π}(\frac{E_{aloc,i}}{{\mathrm{\δ}}_i{s_i}^{2}c}-{{\mathrm{\μ}}_i}^0)$

s.t:∑ _{i＝1,2,…,M}E _aloc,i＝E _avl(Δt)

$\frac{E_{aloc,i}}{{\mathrm{\&delta;}}_i{s_i}^{2}c}<N_i,i=1,2,...,M$

${{\mathrm{\&mu;}}_i}^0\&le;\frac{E_{aloc,i}}{{\mathrm{\&delta;}}_i{s_i}^{2}c},i=1,2,...,M$

Wherein, E _{aloc, i}be the energy that i-th server distributes, μ _i ⁰be the handling capacity that i-th server subsistence level meets, M is the number of server, s _ifor the frequency of server, N _ifor the number of task on server, c is the execution frequency of task, δ _ifor circuit efficiency factor;

Step D2: Lagrangian function corresponding to this optimization problem is:

$\begin{array}{c}L(E,\mathrm{\&alpha;},\mathrm{\&beta;},\mathrm{\&gamma;})=-\underset{i=1}{\overset{M}{\&Sigma;}}\ln (\frac{E_{aloc,i}}{{\mathrm{\&delta;}}_i*{s_i}^2*c}-{{\mathrm{\&mu;}}_i}^0)+\mathrm{\&alpha;}(\underset{i=1}{\overset{M}{\&Sigma;}}{E}_{aloc,i}-E_{aloc,i}\left(\mathrm{\&Delta;}t\right))\\ +{\&Sigma;}_{i=1}^M{\mathrm{\&beta;}}_i(\frac{E_{aloc,i}}{{\mathrm{\&delta;}}_i*{s_i}^2*c}-N_i)+\underset{i=1}{\overset{M}{\&Sigma;}}{\mathrm{\&gamma;}}_i({{\mathrm{\&mu;}}_i}^0-\frac{E_{aloc,i}}{{\mathrm{\&delta;}}_i*{s_i}^2*c})\end{array}$

Ask local derviation to Lagrangian function, and make local derviation equal 0, obtaining allocative decision is: known allocative decision is Lagrange multiplier (α, β _i, γ _i) function, ask these variablees by its dual problem;

Step D3: will substitute into Lagrangian function, obtaining its dual problem is:

$\begin{array}{c}\min -\underset{i=1}{\overset{M}{\&Sigma;}}\ln ({\mathrm{\&delta;}}_i*\mathrm{\&alpha;}*{s_i}^2*c+{\mathrm{\&beta;}}_i)+{\mathrm{\&alpha;E}}_{avl}\left(\mathrm{\&Delta;}t\right)+\underset{i=1}{\overset{M}{\&Sigma;}}{\mathrm{\&beta;}}_i*N_i-\underset{i=1}{\overset{M}{\&Sigma;}}\mathrm{\&alpha;}*{E_{aloc,i}}^0\\ -\underset{i=1}{\overset{M}{\&Sigma;}}\frac{{\mathrm{\&beta;}}_i*{E_{aloc,i}}^0}{{\mathrm{\&delta;}}_i*{s_i}^2*c}+M\end{array}$

s.t.β _i≥0,i＝1,2,…,M

$\frac{{\mathrm{\&delta;}}_i{s_i}^{2}c}{{\mathrm{\&alpha;\&delta;}}_i{s_i}^{2}c+{\mathrm{\&beta;}}_i-{\mathrm{\&gamma;}}_i}+{E_{aloc,i}}^0\&le;{s_i}^2*c*N_i.$

Described step 5 specifically comprises:

Step e 1: solve dual problem by gradient projection method.First provide the initial value of algorithm, the solution of dual problem can be obtained after algorithm convergence, α ^*, β _i ^*, γ _i ^*, substituted into $E_{aloc,i}=\frac{{\mathrm{\&delta;}}_i{s_i}^{2}c}{{\mathrm{\&alpha;\&delta;}}_i{s_i}^{2}c+{\mathrm{\&beta;}}_i-{\mathrm{\&gamma;}}_i}+{E_{aloc,i}}^0,$ Obtain final energy distribution scheme;

Step e 2: the Global Optimality separated is asked in checking

First, the gloomy matrix in sea of dual problem objective function is obtained, i.e. function $-{\mathrm{\&Sigma;}}_{i=1}^{M}ln({\mathrm{\&delta;}}_i*\mathrm{\&alpha;}*{s_i}^2*c+{\mathrm{\&beta;}}_i)+{\mathrm{\&alpha;E}}_{avl}\left(\mathrm{\&Delta;}t\right)+$ ${\mathrm{\&Sigma;}}_{i=1}^M{\mathrm{\&beta;}}_i*N_i-{\mathrm{\&Sigma;}}_{i=1}^M\mathrm{\&alpha;}*{E_{aloc,i}}^0-{\mathrm{\&Sigma;}}_{i=1}^M\frac{{\mathrm{\&beta;}}_i*{E_{aloc,i}}^0}{{\mathrm{\&delta;}}_i*{s_i}^2*c}+M$ The gloomy matrix in sea; In the gloomy matrix in sea, each is calculated as follows:

$\frac{\&part;d^{*2}}{\&part;{\mathrm{\&alpha;}}^2}=\underset{i=1}{\overset{M}{\&Sigma;}}\frac{{{\mathrm{\&delta;}}_i}^2{c}^2{s_i}^4}{{({\mathrm{c\&delta;}}_i{{\mathrm{\&alpha;s}}_i}^2+{\mathrm{\&beta;}}_i)}^2}\&GreaterEqual;0,i=1,2,...,M$

$\frac{\&part;d^{*2}}{\&part;\mathrm{\&alpha;}\&part;{\mathrm{\&beta;}}_i}=\frac{{{\mathrm{cs}}_i}^2}{{({\mathrm{c\&alpha;\&delta;}}_i{s_i}^2+{\mathrm{\&beta;}}_i)}^2}>0,i=1,2,...,M$

$\frac{\&part;d^{*2}}{\&part;{\mathrm{\&beta;}}_i\&part;\mathrm{\&alpha;}}=\frac{{{\mathrm{cs}}_i}^2}{{({\mathrm{c\&alpha;\&delta;}}_i{s_i}^2+{\mathrm{\&beta;}}_i)}^2}>0,i=1,2,...,M$

$\frac{\&part;d^{*2}}{\&part;{{\mathrm{\&beta;}}_i}^2}=\frac{1}{{({\mathrm{c\&alpha;\&delta;}}_i{s_i}^2+{\mathrm{\&beta;}}_i)}^2}>0,i=1,2,...,M$

$\frac{\&part;d^{*2}}{\&part;{\mathrm{\&beta;}}_i\&part;{\mathrm{\&beta;}}_j}=0,i,j=1,2,...,M$

Visible, in extra large gloomy matrix, each is all more than or equal to 0, extra large gloomy matrix positive semidefinite, and in dual problem, objective function is convex function, therefore, and the solution α asked by gradient projection ^*, β _i ^*, γ _i ^*global optimum, therefore final energy distribution scheme $E_{aloc,i}=\frac{{\mathrm{\&delta;}}_i{s_i}^{2}c}{{\mathrm{\&alpha;\&delta;}}_i{s_i}^{2}c+{\mathrm{\&beta;}}_i-{\mathrm{\&gamma;}}_i}+{E_{aloc,i}}^0$ Ye Shi global optimum.

The present invention considers energy abundance or supplements the situation of foot.When energy is sufficient, distribute corresponding energy according to the actual needs of each server; When energy is inadequate, adopt cooperative game to carry out energy distribution, final energy distribution scheme has not only considered user's health check-up but also has looked after the cost of cloud service provider, meets the demand of user and service provider both sides to a certain extent simultaneously.

Accompanying drawing explanation

Fig. 1 is process flow diagram of the present invention;

Energy distribution schematic diagram when Fig. 2 is 3 servers;

Energy distribution schematic diagram when Fig. 3 is 5 servers;

Energy distribution schematic diagram when Fig. 4 is 8 servers;

When Fig. 5 is 3 servers and changes utilisable energy, the present invention and the naive method contrast schematic diagram in fairness and harmony;

When Fig. 6 is 5 servers and changes utilisable energy, the present invention and the naive method contrast schematic diagram in fairness and harmony;

When Fig. 7 is 8 servers and changes utilisable energy, the present invention and the naive method contrast schematic diagram in fairness and harmony;

When Fig. 8 is 3 servers and changes load, the present invention and the naive method contrast schematic diagram in fairness and harmony;

When Fig. 9 is 5 servers and changes load, the present invention and the naive method contrast schematic diagram in fairness and harmony;

When Figure 10 is 8 servers and changes load, the present invention and the naive method contrast schematic diagram in fairness and harmony;

Figure 11 is the present invention and the contrast schematic diagram of MT scheme in handling capacity;

Figure 12 is the present invention and the contrast schematic diagram of MT scheme in fairness.

Embodiment

Below in conjunction with drawings and the specific embodiments, the present invention is described in further detail.

The present invention adopts sun power as energy source, and following formula presents the time dependent characteristic of solar panels capacitation:

$P_h\left(t\right)=|10*N\left(t\right)*cos\left(\frac{t}{70\mathrm{\&pi;}}\right)*cos\left(\frac{t}{100\mathrm{\&pi;}}\right)|$

Wherein, N (t) is a stochastic variable of obeying 0-1 Gaussian distribution.

Task in the present invention arrives has randomness, unpredictability, therefore the load on each server is random generation, in test, also the load constantly changed on each server carrys out the validity of proof scheme, experiment shows, how to change regardless of the load on server, the allocative decision carried can take into account cloud service provider and user preferably.In addition, task in the present invention has the identical average performance period, do not change in test the validity of tasks carrying cycle verification algorithm, because the performance period of all tasks is the same, it is not very necessary for changing this parameter, however, the performance period of task can be set to realistic any value.

In the present invention, server is also random, here be embodied in two aspects at random.The first, the number of server is random, and cloud service provider decides the quantity of the virtual server that can externally provide according to one's own hardware resource situation.The second, the frequency of server is random, and in actual cloud service system, the expense that the frequency of virtual server is paid by user determines, therefore the frequency of server can process as a random value.Constantly change the validity that the quantity of server and frequency verify allocative decision in test, result shows, when number of servers and frequency shift, the allocative decision proposed in article has met expected effect preferably.

The distribution mechanism proposed in the present invention is intended to consider the handling capacity on the handling capacity of high in the clouds entirety and the server of each user oneself simultaneously, namely a good allocative decision takes into account the scheme of performance and fairness, how to represent that a scheme has " efficiency ", how to represent a scheme be " justice ", how to weigh the degree that a scheme takes into account performance and fairness.

A scheme has the throughput ratio of the expression high in the clouds whole system of " efficiency " higher, namely the summation of Servers-all handling capacity is larger, and under the prerequisite of limited utilisable energy, entire system handling capacity is higher, scheme selected by expression is more efficient, also more meets the wish of service provider.

A scheme " " should be showed by the handling capacity of all users, if the handling capacity difference between user is comparatively large, then can cause that user's is discontented, this also illustrates that the fairness of the program is poor to fairness.Therefore, " fairness " can represent by the variance of user throughput, and the less explanation fairness of variance is better, and variance larger explanation fairness is poorer.

Represent that a scheme takes into account degree to fairness and harmony by fairness/efficiency in the present invention, this value is less, then take into account better is described, the larger explanation scheme of this value does not well take into account service provider and user.

In order to the advantage of the allocative decision proposed in the present invention is described, experimental section will carry out according to following steps:

1. checking is at different utilisable energies, and when different load, the method for proposition can converge to a globally optimal solution, namely can obtain a unique solution based on the allocative decision of cooperative game.

2. the load in fixed server, changes utilisable energy, and in different processor number, when different task scale, checking proposes the validity of distribution method.

3. fixing utilisable energy, changes server load, and in different processor number, when different task scale, the validity of checking put forward the methods scheme.

4. the method for proposition and the scheme with maximum throughput are contrasted.

Embodiment

Step one: determine this energy distribution time interval.In the invention process process, adopt 8 servers, 50-150 task carrys out presentation process.

Frequency on server is:

Server1	Server2	Server3	Server4	Server5	Server6	Server7	Server8
								2.2	2.3	2.1	2.5	2.4	2.5	2.6	2.8

Circuit factor delta on server _ifor:

δ 1

δ 2

δ 3

δ 4

δ 5

δ 6

δ 7

δ 8

13*10 -9

12*10 -9

11*10 -9

11*10 -9

14*10 -9

15*10 -9

11*10 -9

13*10 -9

Load in service, namely task number is:

Load1	Load2	Load3	Load4	Load5	Load6	Load7	Load8
								70	80	100	120	130	110	90	80

By E _min(Δ t)=M*min{ δ _is _i ²c*N _i|, i=1,2 ..., the minimum energy required for M} calculates, by E _max(Δ t)=M*max{ δ _is _i ²c*N _i|, i=1,2 ..., the maximum energy required for M} calculates, and by Δ t=min{ Δ t|E _harv(t+ Δ t) ∈ [E _min(Δ t), E _max(Δ t)] } to calculate this energy distribution time interval be 130 minutes.

Step 2: use formula predict that this utilisable energy is 14580J.

Step 3: use formula the energy calculating actual needs is 18500J, and therefore, utilisable energy is inadequate, adopts and distributes based on game theoretic allocative decision.

Step 4: solve dual problem and basis by gradient projection method calculating final allocative decision is: 1250J, 2500J, 2100J, 2500J, 1000J, 2500J, 1750J, 990J.

Step 5: the uniqueness of checking allocative decision

First, the number of stochastic generation server, the load on each server and utilisable energy are also given at random, and constantly change these parameters, and constantly change the initial value of algorithm, whether verification algorithm converges to an optimum solution.

Respectively different when: three servers, task scale is 0-20; Five servers, task scale is 20-50; Eight servers, task scale is 50-150, observes the relation of different initial values and gradient projection iterations.

Table 1

As shown in table 1, when server is 3, change the initial value of gradient projection, iterations there occurs change, but final energy distribution scheme is consistent, as shown in Figure 2.

When server number is 5,8, change the initial value of gradient projection, final energy distribution scheme is also unique, respectively as shown in Figure 3 and Figure 4.

Visible, the allocative decision proposed in the present invention is the scheme of a global optimum.

Step 6: change energy estimate methods load, the validity of the allocative decision proposed in checking the present invention

First, the load on stochastic generation server number and server, and the load in fixed server, change extraneous utilisable energy, and compare with demand assigned naive method, verifies the validity proposed a plan.When contrasting each time, respectively the number of server is fixed as 3,5,8, and keeps the load on each server not change, change extraneous utilisable energy simultaneously.

Fig. 5 is three servers, and time task scale is 0-20, extraneous utilisable energy is respectively 100J, distributes when 110J, 120J, 130J, 140J, 150J to energy, and ordinate represents the ratio of fairness/efficiency.After being assigned, calculate the ratio of justice/efficiency, in figure, light bar post is naive method, and dark bars post is allocative decision of the present invention, can find out, the scheme proposed strictly is better than naive method.

Fig. 6 is five servers, task scale is 20-50, outside energy is respectively 650J, 700J, 750J, 800J, carry out distribute energy by naive method and the present invention respectively when 850J, 900J, ordinate represents the ratio of fairness/efficiency, can significantly find out, the scheme that the present invention proposes will strictly be better than naive method to the degree of taking into account of fairness and harmony.

Fig. 7 is eight servers, task scale is 50-150, and extraneous utilisable energy is respectively 3500J, 4000J, 4500J, when 5000J, 5500J, 6000J, energy distribution is carried out respectively by the scheme that naive method and the present invention propose, ordinate represents the ratio of fairness/efficiency, is not difficult to find out, energy distribution scheme of the present invention has good effect.

As can be seen from Fig. 5, Fig. 6, Fig. 7, when number of servers and task scale are fixed, when outside energy changes, the allocation algorithm based on cooperative game well can take into account performance and fairness.

Step 7: change load fixed energies, the validity of the allocative decision proposed in checking the present invention

First, stochastic generation server number and utilisable energy, constantly change the load on server, and when system has the server of varying number, " validity " and " fairness " of proof scheme.

Fig. 8 is 3 servers, and respectively in different loads situation (load as shown in FIG.), the contrast situation of two schemes, from figure, in the present invention, allocative decision is strictly better than naive method.

Fig. 9 is 5 servers, the contrast under different loads situation, and because load data is more, can not show in figure, therefore load data is enumerated in table 2.From in figure, the allocative decision in the present invention is better than naive method.

Table 25 server, different loads

Load1	Load2	Load3	Load4	Load5	Load6
						20，25，30，33，40	20，28，32，25，45	23，30，35，38，50	20，32，40，45，50	22，34，42，41，48	24，33，39，40，45

Figure 10 is 8 servers, and the contrast under different loads situation, load data is enumerated in table 3.From in figure, be better than naive method based on game theoretic allocative decision.

Table 38 server, different loads

Step 8: contrast with the scheme (MT) with maximum throughput, the validity of allocative decision in checking the present invention

The scheme (maximumthroughputstrategy, MT) that scheme in the present invention and one have maximum throughput is contrasted; Due to square being directly proportional of the energy consumption of individual task and the length of task and frequency, and all task length is identical in the present invention, therefore, if energy priority is assigned on the server of low frequency, then system can obtain higher handling capacity, because same energy will support more task run.Two schemes will contrast respectively in validity and fairness.

Figure 11 is the contrast diagram of handling capacity (contrast of validity) corresponding to two schemes, and horizontal ordinate is utilisable energy, and ordinate represents handling capacity, E _needfor completing the minimum energy that all required by task are wanted.As can be seen from Figure, when utilisable energy is less than E _needtime, handling capacity outline of the present invention is lower than MT scheme, and its reason provides at the preceding paragraph.When utilisable energy increases gradually, throughput of system also can increase, and when utilisable energy is more than or equal to E _needtime, handling capacity corresponding to two kinds of energy distribution schemes is equal.

Figure 12 is the contrast diagrams of two kinds of allocative decisions in fairness, and represent fairness when contrasting by standard deviation, because the numerical value of variance is excessive, ordinate represents standard deviation.Compared with MT scheme, scheme in the present invention has good fairness, this is because when carrying out energy distribution at every turn, all more fair by energy distribution on each server, and energy priority is only just assigned to above the server of low frequency by MT scheme, therefore, MT scheme can not guarantee fairness.When utilisable energy is more than or equal to E _needtime, both have identical fairness, because now both allocative decisions are identical.

Energy, under the prerequisite of finite energy, is reasonably allocated to each processor by the present invention, makes the handling capacity of single process higher, and simultaneously overall handling capacity is also higher.

Claims (6)

1. in the cloud computing system of renewable energy supply based on an energy distributing method for cooperative game, it is characterized in that the method comprises the following steps:

Step one: the time interval determining this energy distribution;

Step 2: predict the utilisable energy in this time interval, this utilisable energy is as the energy source of system;

Step 3: judge that whether energy is sufficient, go to step four time inadequate, turns distribution according to need energy time sufficient, and goes to step six;

Step 4: modeling is carried out to energy distribution with game theory, and change into the optimization problem of belt restraining, then convert it into dual problem;

Step 5: solve dual problem by gradient projection method, and the Global Optimality separated is asked in checking;

Step 6: distribute and terminate.

2. energy distributing method as claimed in claim 1, is characterized in that described step one specifically comprises:

Steps A 1: the minimum energy required for computing system:

E _min(Δt)＝M*min{δ _is _i ²c*N _i|,i＝1,2,…,M}

Wherein: M is the number of server, s _ifor the frequency of server, N _ifor the number of task on server, c is the execution frequency of task, δ _ifor circuit efficiency factor;

Steps A 2: the maximum energy required for computing system:

E _max(Δt)＝M*max{δ _is _i ²c*N _i|，i＝1，2，...，M}

Wherein: M is the number of server, s _ifor the frequency of server, N _ifor the number of task on server, c is the execution frequency of task, δ _ifor circuit efficiency factor;

Steps A 3: the energy distribution time interval, Δ t was defined as E _harv(Δ t) is in E _min(Δ t) and E _maxtime between (Δ t):

Δt＝min{Δt|E _harv(t+Δt)∈[E _min(Δt),E _max(Δt)]}。

3. energy distributing method as claimed in claim 1, is characterized in that described step 2 specifically comprises:

Step B1: the utilisable energy in this time interval predicted is:

$E_{harv}\left(\mathrm{\&Delta;}t\right)={\&Integral;}_t^{t+\mathrm{\&Delta;}t}{P}_{harv}\left(t\right)dt,P_{harv}\left(\mathrm{\&Delta;}t\right)$

Wherein, P _harvt () is energy acquisition power.

4. energy distributing method as claimed in claim 1, is characterized in that described step 3 specifically comprises:

Step C1: the energy that computing system needs altogether:

$E_{demand}\left(\mathrm{\&Delta;}t\right)=\underset{i=1}{\overset{M}{\&Sigma;}}{E}_{cons,i}$

E _cons,i＝N _i*E _cons(τ,s _i)

$E_{cons}(\mathrm{\&tau;},s_i)={\mathrm{\&delta;}}_i*{s_i}^3*\frac{c}{s_i}={\mathrm{\&delta;}}_i*{s_i}^2*c$

Wherein, E _demand(Δ t) gross energy required for M server, E _{cons, i}be the energy that i-th server needs, E _cons(τ, s _i) energy that consumes for each task, when task is s in frequency _iserver on when running; N _iit is the number of task on i-th server;

Step C2: judge that whether energy is sufficient

Work as E _harv(Δ t) > E _demandtime (Δ t), system capacity is sufficient, otherwise system capacity is inadequate.

5. energy distributing method as claimed in claim 1, is characterized in that described step 4 specifically comprises:

Step D1: modeling is carried out to energy distribution problem with game theory, and be converted into the optimization problem of a belt restraining, that is:

$max:\underset{i=1,2,...,M}{\&Pi;}\left(\frac{E_{aloc,i}}{{\mathrm{\&delta;}}_i{s_i}^{2}c}-{{\mathrm{\&mu;}}_i}^0\right)$

s.t:∑ _{i＝1，2，…,M}E _aloc,i＝E _avl(Δt)

$\frac{E_{aloc,i}}{{\mathrm{\&delta;}}_i{s_i}^{2}c}<N_i,i=1,2,...,M$

${{\mathrm{\&mu;}}_i}^0\&le;\frac{E_{aloc,i}}{{\mathrm{\&delta;}}_i{s_i}^{2}c},i=1,2,...,M$

Wherein, E _{aloc, i}be the energy that i-th server distributes, μ _i ⁰be the handling capacity that i-th server subsistence level meets, M is the number of server, s _ifor the frequency of server, N _ifor the number of task on server, c is the execution frequency of task, δ _ifor circuit efficiency factor;

Step D2: Lagrangian function corresponding to this optimization problem is:

$\begin{array}{c}L\left(E,\mathrm{\&alpha;},\mathrm{\&beta;},\mathrm{\&gamma;}\right)=-\underset{i=1}{\overset{M}{\&Sigma;}}\ln \left(\frac{E_{aloc,i}}{{\mathrm{\&delta;}}_i*{s_i}^2*c}-{{\mathrm{\&mu;}}_i}^0\right)+\mathrm{\&alpha;}\left(\underset{i=1}{\overset{M}{\&Sigma;}}{E}_{aloc,i}-E_{aloc,i}\left(\mathrm{\&Delta;}t\right)\right)\\ +{\&Sigma;}_{i=1}^M{\mathrm{\&beta;}}_i\left(\frac{E_{aloc,i}}{{\mathrm{\&delta;}}_i*{s_i}^2*c}-N_i\right)+\underset{i=1}{\overset{M}{\&Sigma;}}{\mathrm{\&gamma;}}_i\left({{\mathrm{\&mu;}}_i}^0-\frac{E_{aloc,i}}{{\mathrm{\&delta;}}_i*{s_i}^2*c}\right)\end{array}$

Ask local derviation to Lagrangian function, and make local derviation equal 0, obtaining allocative decision is: known allocative decision is Lagrange multiplier (α, β _i, γ _i) function, ask these variablees by its dual problem;

Step D3: will substitute into Lagrangian function, obtaining its dual problem is:

$\begin{array}{c}\min -\underset{i=1}{\overset{M}{\&Sigma;}}\ln \left({\mathrm{\&delta;}}_i*\mathrm{\&alpha;}*{s_i}^2*c+{\mathrm{\&beta;}}_i\right)+{\mathrm{\&alpha;E}}_{avl}\left(\mathrm{\&Delta;}t\right)+\underset{i=1}{\overset{M}{\&Sigma;}}{\mathrm{\&beta;}}_i*N_i-\underset{i=1}{\overset{M}{\&Sigma;}}\mathrm{\&alpha;}*{E_{aloc,i}}^0\\ -\underset{i=1}{\overset{M}{\&Sigma;}}\frac{{\mathrm{\&beta;}}_i*{E_{aloc,i}}^0}{{\mathrm{\&delta;}}_i*{s_i}^2*c}+M\end{array}$

s.t.β _i≥0,i＝1,2,…,M

$\frac{{\mathrm{\&delta;}}_i{s_i}^{2}c}{{\mathrm{\&alpha;\&delta;}}_i{s_i}^{2}c+{\mathrm{\&beta;}}_i-{\mathrm{\&gamma;}}_i}+{E_{aloc,i}}^0\&le;{s_i}^2*c*N_i.$

6. energy distributing method as claimed in claim 1, is characterized in that described step 5 specifically comprises:

Step e 1: solve dual problem by gradient projection method.First provide the initial value of algorithm, the solution of dual problem can be obtained after algorithm convergence, α ^*, β _i ^*, γ _i ^*, substituted into obtain final energy distribution scheme;

Step e 2: the Global Optimality separated is asked in checking

First, the gloomy matrix in sea of dual problem objective function is obtained, i.e. function the gloomy matrix in sea; In the gloomy matrix in sea, each is calculated as follows:

$\frac{\&part;d{*}^2}{\&part;{\mathrm{\&alpha;}}^2}=\underset{i=1}{\overset{M}{\&Sigma;}}\frac{{{\mathrm{\&delta;}}_i}^2{c}^2{s_i}^4}{{({\mathrm{c\&delta;}}_i{{\mathrm{\&alpha;s}}_i}^2+{\mathrm{\&beta;}}_i)}^2}\&GreaterEqual;0,i=1,2,...,M$

$\frac{\&part;d{*}^2}{\&part;\mathrm{\&alpha;}\&part;{\mathrm{\&beta;}}_i}=\frac{{{\mathrm{cs}}_i}^2}{{({\mathrm{c\&alpha;\&delta;}}_i{s_i}^2+{\mathrm{\&beta;}}_i)}^2}>0,i=1,2,...,M$

$\frac{\&part;d{*}^2}{\&part;{\mathrm{\&beta;}}_i\&part;\mathrm{\&alpha;}}=\frac{{{\mathrm{cs}}_i}^2}{{({\mathrm{c\&alpha;\&delta;}}_i{s_i}^2+{\mathrm{\&beta;}}_i)}^2}>0,i=1,2,...,M$

$\frac{\&part;d{*}^2}{\&part;{{\mathrm{\&beta;}}_i}^2}=\frac{1}{{({\mathrm{c\&alpha;\&delta;}}_i{s_i}^2+{\mathrm{\&beta;}}_i)}^2}>0,i=1,2,...,M$

$\frac{\&part;d{*}^2}{\&part;{\mathrm{\&beta;}}_i\&part;{\mathrm{\&beta;}}_j}=0,i,j=1,2,...,M$

Visible, in extra large gloomy matrix, each is all more than or equal to 0, extra large gloomy matrix positive semidefinite, and in dual problem, objective function is convex function, therefore, and the solution α asked by gradient projection ^*, β _i ^*, γ _i ^*global optimum, therefore final energy distribution scheme $E_{aloc,i}=\frac{{\mathrm{\&delta;}}_i{s_i}^{2}c}{{\mathrm{\&alpha;\&delta;}}_i{s_i}^{2}c+{\mathrm{\&beta;}}_i-{\mathrm{\&gamma;}}_i}+{E_{aloc,i}}^0$ Ye Shi global optimum.