CN102609820A - Buffer adjustment method for key chain based on project property - Google Patents

Abstract

The invention discloses a buffer adjustment method for a key chain based on project property, which comprises the following steps: step 1, generating an initial feasible scheduling scheme; step 2, calculating free time difference of task and identifying the key chain; step 3, calculating buffer size by two classical quantification methods, selecting the better quantification method to adjust the buffer size through an adjustment strategy; and step 4, generating a buffer scheduling scheme and outputting the result. According to the buffer adjustment method for the key chain based on project property, two important factors of network complexity and resource tightness which affect the robustness of the scheduling scheme are considered, and the network complexity and the resource tightness are introduced in the buffer quantification method, so that the scheduling scheme can respond to the uncertainty of a project in a better way. The method can obtain a reasonable buffer inserting scheme on the basis that the feasibility of the scheduling scheme is ensured. The buffer adjustment method for the key chain based on project property has practical value and vast application prospect in the technical field of optimizing and scheduling.

Description

A kind of chaining key buffering method of adjustment based on project characteristic

Technical field

The present invention relates to a kind of chaining key buffering method of adjustment, particularly relate to a kind of chaining key buffering method of adjustment under uncertain environment based on project characteristic based on project characteristic.The invention belongs to the Optimum Scheduling Technology field.

Background technology

The chaining key method has received scholars' extensive concern (seeing 1997 " chaining key " for details) since 1997 are proposed by Goldratt.Goldratt points out that the importance of all tasks of conventional project management method supposition is identical, therefore is that all tasks add safety time.Yet most safety time is not necessary, even causes the increase greatly of total duration probably.Theoretical research shows all that with practice the bottleneck that influences total duration of project is the chaining key in the project, need protect the task on the chaining key.Chaining key is meant that the group task of total duration of decision project, this group task have been formed the longest path in the project network under the situation of considering precedence constraint and dependent resource.

In order to protect the chaining key task, need insert surge time in position.Item buffer is inserted in after the chaining key, is mainly used in the task on the protection chaining key.When importing chaining key, non-key chain task need insert the buffering of confluxing, to reduce by the delay influence that brings to chaining key of non-key chain task.In the chaining key scheduling theory, the activity on the non-key chain just begins where necessary, to guarantee the chaining key task enough resources is arranged.

Can produce metastable scheduling scheme through the chaining key dispatching method, with the uncertainty in the reply project.And the size of buffering has bigger influence to scheduling scheme, need suitable buffering be set according to the degree of uncertainty of project.Excessive buffering can cause the increase of project duration, and too small buffering then can reduce the robustness of scheduling scheme.Therefore, how confirming buffer size, is that the chaining key dispatching method is applied to an emphasis problem in the actual engineering.Classical buffer setting method mainly contains shearing-mounting method (Cut and Paste Method; C&PM) and the root variance method (Root Square Error Method, RSEM), these two kinds of methods all need be estimated the duration of project; Subjective, and do not consider the characteristic of project.This often causes the final scheduling scheme buffering that produces excessive.

Aspect the improvement buffer setting, researchers at first attempt proposing new heuristic strategies buffer size are set.Yet the heuritic approach that this type is new is too high in the multinomial order problem complexity of solution, realizes that difficulty is big, is difficult in actual engineering, be applied.So mostly work on hand is the adjustment of classic method and improvement strategy research.The chaining key scheduling theory thinks that task all tends to delay.Yet, in practical implementation, be not that all tasks are all delayed, some task even can fulfil ahead of schedule.These fulfiling ahead of schedule of tasks can provide extra buffering to the task chain of its existence, improve this chain for probabilistic resistivity.Owing to ignored this extra buffering, tradition buffering quantization method can obtain bigger buffering as a rule.

The uncertainty of task is along with the carrying out of project transmitted backward, and its reason is the constraint that exists between task.These constraints have determined project for probabilistic resistivity, i.e. the robustness size.The constraint that task faces in the process of implementation has two types of temporal constraint and resource constraints, is determined by task characteristic and resource characteristics respectively.Therefore, consider that in the middle of all project characteristics, network structure and resource requirement especially can not be ignored as two key factors that influence temporal constraint and resource constraint size from project scheduling robustness angle.

Sum up, work on hand has been ignored the influence of project characteristic mostly on the buffer setting problem, thereby has too estimated the uncertainty of project, and excessive buffering is set when scheduling scheme generates.In engineering reality, this scheduling scheme may cause a large amount of resources idles, in the scheduling of multinomial order even cause the significantly extension of project, causes the increase of total cost.Therefore, the buffer setting method of consideration project characteristic more is of practical significance.The present invention just is being based on such consideration.

Summary of the invention

A kind of chaining key buffering method of adjustment of the present invention based on project characteristic; Its objective is: to the uncertainty in the project practical implementation; Two types of important project characteristics are incorporated in the chaining key dispatch buffer quantification problem, and then propose a kind of new buffering adjustment strategy, overcome the excessive deficiency of classical way buffer setting; Utilize ratio thereby in practical application, increase buffering, reduce the delivery time of resources idle and project.

A kind of chaining key buffering method of adjustment of the present invention based on project characteristic, its design philosophy is: at first adopt the heuritic approach calculation task right of priority based on priority rule, adopt the serial scheduling method to generate initial feasible schedule scheme; Confirm chaining key through calculating free float then, and calculate buffer size according to the chaining key of identification; Adopt adjustment strategy adjustment buffer size at last and generate the buffer scheduling scheme.

Based on top thought, following mask body is introduced technical scheme of the present invention, and concrete design procedure is following:

The first step generates initial feasible schedule scheme

The present invention adopts Late Start, and (Last Start Time LST) as the right of priority of priority rule calculation task, adopts the serial scheduling method to generate scheduling scheme.If the start time of task i is ST _i, the concluding time is FT _i, tight back set of tasks is S (i), number of tasks is that n algorithm concrete steps are:

(1) do not consider resource constraint between task, begin successively scheduler task forward from last task, the concluding time of each task equals the start time of the tight back task of its early start.The start time of logger task, each task earliest start time EST of initialization was 0 as its Late Start LST;

(2) from small to large task is sorted (being that the little task of Late Start is dispatched earlier) according to the LST value, generate priority scheduling set E (S);

(3) from E (S), choose first task i and dispatch, make its start time ST _i=EST _i, judge task i execution time section [ST _i, FT _i] in whether produce resource contention; If the resource use amount surpasses available quantity, then make ST _i=ST _i+ 1, FT _i=FT _i+ 1 judges again;

(4) if time period [ST _i, FT _i] interior No Assets conflict, task i dispatches completion, upgrades the earliest start time of its all tight back tasks:

if FT _i＞EST _j

then?EST _j＝FT _i，j∈S(i)

(5) deletion task i from priority scheduling set E (S); If E (S) non-NULL changes (three); Otherwise, finishing scheduling.

The free float of the second step calculation task, the identification chaining key

Move the free float that operation comes calculation task about use task of the present invention, and then identification item purpose chaining key.For task i, if change its start time ST _i, not producing under any conflict situation of (comprising temporal constraint and resource constraint), the start time of other tasks does not all change in the project, and this operation just can be defined as task and moves so.Based on such definition, fixterm purpose execution time section is [ST ₀, FT _N+1] (task 0 is empty task with task n+1); Promptly, ST is arranged for task i ₀≤ST _i≤ST _N+1And FT ₀≤FT _i≤FT _N+1, the concrete steps of identification chaining key are:

(1) scheduling scheme that generates for the first step, from the outset between the latest task begin, select task i from back to front successively;

(2) the single step task that moves to right: the start time ST that makes task i _i=ST _i+ 1, FT _i=FT _i+ 1, judgement time section [ST _i, FT _i] in whether produce conflict (comprising timing conflict and resource contention); If produce conflict, cancel this right-shift operation, i.e. ST _i=ST _i-1, FT _i=FT _i-1, task i mobile end is selected next task; Do not produce if there is conflict, task i is continued to carry out the single step right-shift operation;

(3) when the end that moves to right of all tasks, the start time of record task this moment is Late Start LST _i=ST _i, i=1,2 ..., n;

(4) for the scheduling scheme that moves to right of this moment, from the outset between the earliest task begin, select task j from front to back successively;

(5) the single step task that moves to left: the start time ST that makes task j _j=ST _j-1, FT _j=FT _j-1, judgement time section [ST _j, FT _j] in whether produce conflict (comprising timing conflict and resource contention); If produce conflict, cancel this shift left operation, i.e. ST _j=ST _j+ 1, FT _j=FT _j+ 1, task j mobile end is selected next task; Do not produce if there is conflict, task j is continued to carry out the single step shift left operation;

(6) when the end that moves to left of all tasks, the start time of record task this moment is earliest start time EST _i=ST _i, i=1,2 ..., n;

(7) the free float TS of calculation task _i=LST _i-EST _i, i=1,2 ..., n; If TS _i=0, then task i is the chaining key task; Otherwise be non-key chain task.

The 3rd step buffering quantizes and adjustment

The quantification and the adjustment of buffering are divided into following a few step:

(1) computational grid complexity and resource elasticity

Network complexity may be defined as the ratio (all not considering empty task) of the precedence relationship number that exists in the network and the theoretical maximum precedence relationship number that exists, and is used for representing the temporal constraint intensity of network.Wherein, the precedence relationship number that exists in the network is all forerunner's nodes (comprising indirect forerunner's node) sum of each task, and maximum precedence relationship number is n (n-1)/2, and wherein n representes non-empty task number.Therefore network complexity can be expressed as:

The resource elasticity may be defined as the average use amount of resource and the ratio between the total amount, expression resource constraint intensity:

${\mathrm{RC}}_k=\frac{{\stackrel{\‾}{r}}_k}{a_k}$

Wherein, a _kThe total amount of expression renewable resources k,

The average use amount of then representing resource k, promptly

${\stackrel{\‾}{r}}_k=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}{r}_{\mathrm{ik}}/\underset{i=1}{\overset{n}{\mathrm{\Σ}}}m,$ $m=\left\{\begin{array}{cc}1,& r_{\mathrm{ik}}>0\\ 0& r_{\mathrm{ik}}=0\end{array}\right.$

(2) buffering quantizes

The present invention adopts classical buffering quantization method to calculate buffer size, comprises shearing-mounting method and root variance method.If the i safety time of task is estimated as S _i, expression task 90% probability is accomplished the corresponding duration of institute; Be estimated as A averaging time _i, expression task 50% probability is accomplished the pairing duration, and then the task i duration is D _i=A _iSuppose that task k～k+m belongs to same task chain, then the buffer size B of this chain is set to:

Shearing-mounting method:

The root variance method: $B_{\mathrm{RSEM}}=\mathrm{Sqrt}\left(\underset{i=k}{\overset{k+m}{\mathrm{\Σ}}}{(S_i-A_i)}^2\right);$

(3) buffering adjustment

The item buffer and the buffering of confluxing of two kinds of method computational items above adopting, the option buffering smaller adjust.The robustness of disparity items is different, and is different with the robustness notion of broad sense, and the robustness is here refered in particular in the project implementation, when some factor changes, and the probability that the project duration changes.When representing that like the task robustness task takes place to delay, the probability of extension takes place in the project duration.When between task about beam intensity the time, the extension of a task can cause a chain of effect of inter-related task, causes the extension of total duration.When the task constraint was weak, the extension of some task maybe not can have any impact to project.The present invention uses network complexity OS and resource elasticity RC description entry purpose task robustness R respectively _TaskWith resource robustness R _ResourceBecause OS and RC have represented the size of temporal constraint and resource constraint respectively, therefore adopting these two kinds of indexs to describe project is rational for the resistibility of task change and resource changing.

Item buffer (PB) is determined that by number of tasks on the chaining key and task length the buffering of confluxing (FB) is then by non-key chain task decision.In general, chaining key is longer than non-key chain, and task is also more.So item buffer often is better than the buffering of confluxing for probabilistic resistivity, can carry out bigger reduction to item buffer, buffering reduces by a small margin to confluxing.Therefore, the present invention is according to the task robustness R of project _TaskAnd resource robustness R _ResourceThe adjustment weights omega is set _FAnd ω _P, adjust conflux buffering and resource pooling respectively.Thereby on the basis that the assurance project completes on schedule, reduce date of payment, increase the buffering proportion of utilization.

ω _F＝max{OS，RC}

ω _P＝min{OS，RC}

FB′＝ω _F×FB

PB′＝ω _P×PB

The 4th step generated the buffer scheduling scheme;

Adjustment result according to the 3rd step moves to left non-key chain task, upgrades its start time, obtains the buffer scheduling scheme.

The present invention is with the advantage that existing method is compared: the present invention has considered to influence two key factors, network complexity and the resource elasticity of scheduling scheme robustness.Be introduced in the buffering quantization method, make the uncertainty in the scheduling scheme reply project better.The present invention can reasonably be cushioned interleaved plan on the basis of the feasibility that guarantees scheduling scheme.

Description of drawings

Fig. 1 heuritic approach process flow diagram of the present invention;

Fig. 2 chaining key recognizer of the present invention process flow diagram;

The temporal constraint figure of Fig. 3 embodiment of the invention;

The initial feasible schedule scheme Gantt chart of Fig. 4 embodiment of the invention;

The scheduling Gantt chart that moves to right of Fig. 5 embodiment of the invention;

The scheduling Gantt chart that moves to left of Fig. 6 embodiment of the invention;

The buffering of Fig. 7 embodiment of the invention is inserted synoptic diagram as a result;

The result contrasts synoptic diagram (changing OS) before and after Fig. 8 item buffer proportion of utilization adjustment of the present invention;

Fig. 9 the present invention is confluxed, and the result contrasts synoptic diagram (changing OS) before and after the buffering proportion of utilization adjustment;

The result contrasts synoptic diagram (changing RC) before and after Figure 10 item buffer proportion of utilization adjustment of the present invention;

Figure 11 the present invention is confluxed, and the result contrasts synoptic diagram (changing RC) before and after the buffering proportion of utilization adjustment;

Figure 12 FB(flow block) of the present invention.

Symbol and label declaration among the figure are following:

Circle is represented test assignment among Fig. 3, and arrow is represented temporal constraint, the duration of the other digitized representation task of circle, i.e. execution time;

Horizontal ordinate is the time in Fig. 4～6, and ordinate is the numbering of task, and square is represented task, and the length of square is represented task execution time, and the left margin of each square is its zero hour, and right margin is its finish time;

Horizontal ordinate is the time among Fig. 7, and ordinate is a task chain, and square is represented task, and the length of square is represented task execution time, and the left margin of each square is its zero hour, and right margin is its finish time; The FB buffering of representing to conflux, PB representes item buffer;

Horizontal ordinate is represented the different experiments use-case in Fig. 8～11, and ordinate representes to cushion proportion of utilization;

Embodiment

The present invention explains its embodiment through the simplified example of 4 kinds of renewable resources of 10 tasks.Wherein, task 0 is empty task with task 11, i.e. duration and resource requirement is 0; The available quantity of each resource is 10, and the task network chart is as shown in Figure 3, retrains between task like following table:

Table 1

According to above data, see Figure 12, practical implementation step of the present invention is following:

The first step generates initial feasible schedule scheme

Adopt the heuritic approach that the first step proposes in the summary of the invention, according to the flow process among Fig. 1 the scheduling scheme of project is found the solution, concrete steps are:

(1) calculates priority scheduling set E (S)

Do not consider resource constraint, begin from big to small to carry out back scheduling according to task symbol, get ST from task 11 beginnings ₁₁=FT ₁₁=0.The tight back of the early start of task 10 task is a task 11, so FT ₁₀=ST ₁₁=0, ST ₁₀=FT ₁₀+ D ₁₀=3; The tight back of the early start of task 9 task is a task 11, so FT ₉=ST ₁₁=0, ST ₉=FT ₉+ D ₉=7; In like manner can obtain the start time set J (ST of scheduling scheme _i)={ 26,19,12,5,10,10,1,9,7,3}, owing to be back scheduling, so need deduct the Late Start LST that the current duration 26 just obtains task, so the set of the Late Start of task is J (LST _i)={ 0,7,14,21,16,16,25,17,19,23}.Therefore, priority scheduling set for E (S)=1,2,3,5,6,8,9,4,10,7}

(2) generate initial feasible schedule scheme

The earliest start time EST of all tasks of initialization _i=0, from E (S), choosing first task is task 1, makes ST ₁=EST ₁=0, FT ₁=ST ₁+ D ₁=7, owing to be the task of first scheduling, obviously No Assets conflict this moment; Deletion task 1 from E (S) is upgraded its tightly earliest start time EST of back task ₂=EST ₃=EST ₄=EST ₉=EST ₁₀=7; Select the task 2 among the E (S), make ST ₂=EST ₂=7, FT ₂=ST ₂+ D ₂=17, No Assets conflict this moment; Deletion task 2 from E (S) is upgraded its tightly earliest start time EST of back task ₇=EST ₈=17; Select the task 3 among the E (S), make ST ₃=EST ₃=7, FT ₃=ST ₃+ D ₃=11, because at this moment in the moment 7, the demand of resource 3 and resource 4 has surpassed available quantity, has promptly produced resource contention, so task 3 is shifted to the right to up to No Assets conflict generation, ST ₃=FT ₂=17, FT ₃=ST ₃+ D ₃=21; In like manner can get the start time and the concluding time of other tasks.The initial feasible schedule set of task is: J (S)=0,7,17,7,21,30,55,39,48, and 30}, as shown in Figure 4;

The free float of the second step calculation task, the identification chaining key

According to the feasible schedule scheme that the first step obtains, adopt summary of the invention second to go on foot the free float TS of the algorithm computation task of mentioning _i, the recognizer flow process is as shown in Figure 2; Fixedly the project duration is identical with task right-shift operation in the first step, is J with move to right the one by one scheduling scheme that obtains moving to right of task _R(S)=0,7,17,13,21,30,55,39,48,36}; Be J with move to left the one by one scheduling scheme that obtains moving to left of task again _L(S)=0,7,17,7,21,30,55,39,48,30} deducts the start time that moves to left through the start time that moves to right thus, and the time difference of task be J (TS)=0,0,0,6,0,0,0,0,0,6} (shown in Fig. 5～6).Therefore the chaining key task is 1,2,3,5,6,7,8,9, and non-key chain task is 4,10; Specifically introduce in the face of right-shift operation down:

(1) earlier each task (not comprising empty task) is sorted by the start time from big to small,, then number big priority of task: J={7 if the start time is identical, 9,8,10,6,5,3,4,2,1};

(2) task 7 is because FT ₇=ST ₁₁=56, promptly equal the project duration, so can not move to right;

(3) task 9 can not move to right.Task 9 unit that moves to right if this is then reaches 20 in 55 tasks 9 constantly with the resource 1 that task 7 consumes, and surpasses the available quantity of 10 unit;

(4) task 8 can not move to right.Task 9 unit that moves to right if this is then reaches 20 in 48 tasks 8 constantly with the resource 1 that task 9 consumes, and surpasses the available quantity of 10 unit;

(5) task 10 6 units that can move to right, its start time becomes 36;

(6) task 6 is tight preceding tasks of task 7, and task 7 can not move to right, so task 6 also can not move to right;

(7) task 5 can not move to right.Task 5 unit that moves to right if this is then reaches 13 in 30 tasks 5 constantly with the resource 1 that task 6 consumes, and surpasses the available quantity of 10 unit;

(8) task 3 can not move to right.Task 3 unit that moves to right if this is then reaches 14 in 21 tasks 3 constantly with the resource 1 that task 5 consumes, and surpasses the available quantity of 10 unit;

(9) task 46 units that can move to right, its start time becomes 13;

(10) task 2 is tight preceding tasks of task 3, and task 3 can not move to right, so task 2 also can not move to right;

(11) task 1 is the tight preceding task of task 2, and task 2 can not move to right, so task 1 also can not move to right;

The 3rd step buffering quantizes and adjustment

According to the 3rd step of summary of the invention, get the safety time of all tasks and estimate S _i=2D _i, estimate S averaging time _i=D _iBy the chaining key that second step was confirmed, the item buffer that adopts shearing-mounting method and root variance method to try to achieve respectively is respectively:

${\mathrm{PB}}_{\mathrm{RSEM}}=\mathrm{sqrt}\left(\underset{i=k}{\overset{k+m}{\mathrm{\Σ}}}{(S_i-A_i)}^2\right)=\sqrt{7^2+{10}^2+4^2+9^2+9^2+9^2+7^2+1^2}=21.4$

Because the item buffer that adopts the root variance method to obtain is less, therefore select the root variance method to adjust.The embodiment of the invention has two non-key chains, comprises a task respectively.Therefore, conflux the buffering be respectively:

${\mathrm{FB}}_{\mathrm{RSEM}}\left(1\right)=\sqrt{3^2}=3$

${\mathrm{FB}}_{\mathrm{RSEM}}\left(2\right)=\sqrt{4^2}=4$

For buffering is adjusted, need the network complexity and the resource elasticity of computational item.

(1) computational grid complexity then need be considered the precedence relationship (not considering empty task) in the network; Therefore, the present invention weighs precedence relationship through the successor node sum of calculation task.With embodiment is example, uses Suc (i) to describe the follow-up set of tasks of task, calculates from big to small according to mission number:

(1) task 10～7 does not have the tight back task of removing empty task, thus Suc (i)=Φ, i=7,8,9,10;

(2) task 6 has only a tight back task 7, so Suc (6)={ 7} ∪ Suc (7)={ 7};

(3) task 5 has only a tight back task 7, so Suc (5)={ 7} ∪ Suc (7)={ 7};

(4) task 4 has only a tight back task 7, so Suc (4)={ 7} ∪ Suc (7)={ 7};

(5) task 3 has only a tight back task 8, so Suc (3)={ 8} ∪ Suc (8)={ 8};

(6) task 2 has two tight back tasks, is respectively task 7 and task 8, so have:

Suc(2)＝{7，8}∪Suc(7)∪Suc(8)＝{7，8}；

(7) the tight back task of task 1 has task 2, task 3, and task 4, task 9 and task 10, so have:

Suc(1)＝{2，3，4，9，10}∪Suc(2)∪Suc(3)∪Suc(4)∪Suc(9)∪Suc(10)

＝{2，3，4，9，10}∪{7，8}∪{8}∪{7}∪Φ∪Φ＝{2，3，4，7，8，9，10}

Therefore, the precedence relationship that comprises of network adds up to 7+2+1+1+1+1+0+0+0+0=13.The network complexity of task is:

(2) the resource elasticity is then weighed the ratio of the demand of resource according to resource available quantity and task, in the middle of embodiment, and concerning resource 1,

$\stackrel{\&OverBar;}{{\mathrm{<figure-callout\; id="1"\; label="r"\; filenames="HDA0000136315200000022.png,HDA0000136315200000031.png"\; state="\{\{state\}\}">r</figure-callout>}}_1}=\underset{i=1}{\overset{\mathrm{<figure-callout\; id="1"\; label="n\; r\; i"\; filenames="HDA0000136315200000022.png,HDA0000136315200000031.png"\; state="\{\{state\}\}">n</figure-callout>}}{\mathrm{\&Sigma;}}}{r}_i{1}_{}/\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}m=\frac{7+10+4+4+9+9+1+9+7+3}{10}=7$

${\mathrm{<figure-callout\; id="1"\; label="RC"\; filenames="HDA0000136315200000022.png,HDA0000136315200000031.png"\; state="\{\{state\}\}">RC</figure-callout>}}_1=\frac{\stackrel{\&OverBar;}{{\mathrm{<figure-callout\; id="1"\; label="r"\; filenames="HDA0000136315200000022.png,HDA0000136315200000031.png"\; state="\{\{state\}\}">r</figure-callout>}}_1}}{a_1}=\frac{7}{10}=0.7$

In like manner can get RC ₂=RC ₃=RC ₄=0.7, averaging to get RC=0.7;

According to the adjustment strategy that summary of the invention was mentioned in the 3rd step, the item buffer that obtains is adjusted:

ω _F＝max{OS，RC}＝0.7

ω _P＝min{OS，RC}＝0.29

FB′(1)＝ω _F×FB(1)＝0.7×3＝2.1

FB′(2)＝ω _F×FB(2)＝0.7×4＝2.8

PB′＝ω _P×PB＝0.29×21.4＝6.206

The 4th step buffer scheduling scheme generates

Adjustment result according to the 3rd step moves to left non-key chain task, upgrades its start time, obtains the buffer scheduling scheme, and is as shown in Figure 7.

Choose many group experiment use-cases respectively and adopt Monte Carlo method that scheduling scheme is carried out 1000 emulation, write down the actual construction time of its 90% probability deadline as project to the scheduling scheme before and after the adjustment.Calculate the proportion of utilization (shown in Fig. 8～11) of buffering, can find out, the method for adjustment that the present invention proposes has effectively increased the proportion of utilization of buffering and item buffer that confluxes, thereby has reduced the project delivery time.

Claims (1)

1. the chaining key based on project characteristic cushions method of adjustment, and it is characterized in that: it may further comprise the steps:

Step 1: generate initial feasible schedule scheme

Adopting Late Start is Last Start Time, and LST adopts the serial scheduling method to generate scheduling scheme as the right of priority of priority rule calculation task; If the start time of task i is ST _i, the concluding time is FT _i, tight back set of tasks is S (i), number of tasks is that n algorithm concrete steps are:

(1) do not consider resource constraint between task, begin successively scheduler task forward from last task, the concluding time of each task equals the start time of the tight back task of its early start; The start time of logger task, each task earliest start time EST of initialization was 0 as its Late Start LST;

(2) from small to large task being sorted according to the LST value is that the little task of Late Start is dispatched earlier, generates priority scheduling set E (S);

(3) from E (S), choose first task i and dispatch, make its start time ST _i=EST _i, judge task i execution time section [ST _i, FT _i] in whether produce resource contention; If the resource use amount surpasses available quantity, then make ST _i=ST _i+ 1, FT _i=FT _i+ 1 judges again;

(4) if time period [ST _i, FT _i] interior No Assets conflict, task i dispatches completion, upgrades the earliest start time of its all tight back tasks:

if FT _i＞EST _j

then?EST _j＝FT _i，j∈S(i)

(5) deletion task i from priority scheduling set E (S); If E (S) non-NULL changes (three); Otherwise, finishing scheduling;

Step 2: the free float of calculation task, identification chaining key

Move the free float that operation comes calculation task about the use task, and then identification item purpose chaining key; For task i, if change its start time ST _i, not producing under the situation that any conflict promptly comprises temporal constraint and resource constraint, the start time of other tasks does not all change in the project, and this operation just can be defined as task and moves so; Based on such definition, fixterm purpose execution time section is [ST ₀, FT _N+1], task 0 is empty task with task n+1; Promptly, ST is arranged for task i ₀≤ST _i≤ST _N+1And FT ₀≤FT _i≤FT _N+1, the concrete steps of identification chaining key are:

(1) scheduling scheme that generates for the first step, from the outset between the latest task begin, select task i from back to front successively;

(2) the single step task that moves to right: the start time ST that makes task i _i=ST _i+ 1, FT _i=FT _i+ 1, judgement time section [ST _i, FT _i] in whether produce timing conflict and resource contention; If produce conflict, cancel this right-shift operation, i.e. ST _i=ST _i-1, FT _i=FT _i-1, task i mobile end is selected next task; Do not produce if there is conflict, task i is continued to carry out the single step right-shift operation;

(3) when the end that moves to right of all tasks, the start time of record task this moment is Late Start LST _i=ST _i, i=1,2 ..., n;

(4) for the scheduling scheme that moves to right of this moment, from the outset between the earliest task begin, select task j from front to back successively;

(5) the single step task that moves to left: the start time ST that makes task j _j=ST _j-1, FT _j=FT _j-1, judgement time section [ST _j, FT _j] in whether produce timing conflict and resource contention; If produce conflict, cancel this shift left operation, i.e. ST _j=ST _j+ 1, FT _j=FT _j+ 1, task j mobile end is selected next task; Do not produce if there is conflict, task j is continued to carry out the single step shift left operation;

(6) when the end that moves to left of all tasks, the start time of record task this moment is earliest start time EST _i=ST _i, i=1,2 ..., n;

(7) the free float TS of calculation task _i=LST _i-EST _i, i=1,2 ..., n; If TS _i=0, then task i is the chaining key task; Otherwise be non-key chain task;

Step 3: buffering quantizes and adjustment

The quantification and the adjustment of buffering are divided into following a few step:

(1) computational grid complexity and resource elasticity

Network complexity is defined as the ratio of the precedence relationship number that exists in the network and the theoretical maximum precedence relationship number that exists, and is used for representing the temporal constraint intensity of network; Wherein, the precedence relationship number that exists in the network comprises indirect forerunner's node sum for all forerunner's nodes of each task, and maximum precedence relationship number is n (n-1)/2, and wherein n representes non-empty task number, so network complexity is expressed as:

The resource elasticity is defined as the average use amount of resource and the ratio between the total amount, expression resource constraint intensity:

${\mathrm{RC}}_k=\frac{{\stackrel{\&OverBar;}{r}}_k}{a_k}$

Wherein, a _kThe total amount of expression renewable resources k, The average use amount of then representing resource k, promptly

${\stackrel{\&OverBar;}{r}}_k=\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}{r}_{\mathrm{ik}}/\underset{i=1}{\overset{n}{\mathrm{\&Sigma;}}}m,$ $m=\left\{\begin{array}{cc}1,& r_{\mathrm{ik}}>0\\ 0& r_{\mathrm{ik}}=0\end{array}\right.$

(2) buffering quantizes

Adopt classical buffering quantization method to calculate buffer size, comprise shearing-mounting method and root variance method; If the i safety time of task is estimated as S _i, expression task 90% probability is accomplished the corresponding duration of institute; Be estimated as A averaging time _i, expression task 50% probability is accomplished the pairing duration, and then the task i duration is D _i=A _iSuppose that task k～k+m belongs to same task chain, then the buffer size B of this chain is set to:

Shearing-mounting method:

The root variance method: $B_{\mathrm{RSEM}}=\mathrm{Sqrt}\left(\underset{i=k}{\overset{k+m}{\mathrm{\&Sigma;}}}{(S_i-A_i)}^2\right);$

(3) buffering adjustment

The item buffer and the buffering of confluxing of two kinds of method computational items above adopting, the option buffering smaller adjust; The robustness of disparity items is different, and is different with the robustness notion of broad sense, and the robustness is here refered in particular in the project implementation, when some factor changes, and the probability that the project duration changes; When representing that like the task robustness task takes place to delay, the probability of extension takes place in the project duration; When between task about beam intensity the time, the extension of a task can cause a chain of effect of inter-related task, causes the extension of total duration; When the task constraint was weak, the extension of some task maybe not can have any impact to project; Use network complexity OS and resource elasticity RC description entry purpose task robustness R respectively _TaskWith resource robustness R _ResourceBecause OS and RC have represented the size of temporal constraint and resource constraint respectively, therefore adopting these two kinds of indexs to describe project is rational for the resistibility of task change and resource changing;

Item buffer PB is determined that by number of tasks on the chaining key and task length the buffering of confluxing FB is then by non-key chain task decision; In general, chaining key is longer than non-key chain, and task is also more; So item buffer often is better than the buffering of confluxing for probabilistic resistivity, can carry out bigger reduction to item buffer, buffering reduces by a small margin to confluxing; Therefore, according to the task robustness R of project _TaskAnd resource robustness R _ResourceThe adjustment weights omega is set _FAnd ω _P, adjust conflux buffering and resource pooling respectively, thereby on the basis that the assurance project completes on schedule, reduce date of payment, increase the buffering proportion of utilization;

ω _F＝max{OS，RC}

ω _P＝min{OS，RC}

FB′＝ω _F×FB

PB′＝ω _P×PB

Step 4: generate the buffer scheduling scheme;

According to the adjustment result of step 3, non-key chain task is moved to left, upgrade its start time, obtain the buffer scheduling scheme.

Images