CN116090742A - Unmanned cluster distributed task allocation algorithm considering time window constraint - Google Patents

Abstract

The invention discloses an unmanned cluster distributed task allocation algorithm considering time window constraint, comprehensively considering task time window constraint and unmanned aerial vehicle capability constraint, and providing an unmanned cluster distributed cooperative task allocation algorithm by taking the minimum average time cost of tasks as an optimization target. The gain and marginal gain of the task for the unmanned aerial vehicle are defined to describe the change condition of the time cost of the unmanned aerial vehicle to execute the task. A distributed allocation algorithm is then established that includes three phases of task addition, conflict resolution, and task reassignment. The invention can quickly respond to task change, improve calculation speed, reduce the bandwidth requirement of the communication link, and simultaneously avoid the problem of distribution interruption caused by single-point failure in the distribution process. This provides the basis for solving task allocation problems in complex and diverse environments.

Description

Unmanned cluster distributed task allocation algorithm considering time window constraint

Technical Field

The invention belongs to the software and information technology, and particularly relates to an unmanned cluster distributed task allocation algorithm considering time window constraint.

Background

Because the unmanned aerial vehicle has advantages such as strong flexibility, assembly are convenient, cost consumption is low, unmanned aerial vehicle has used in fields such as search and rescue, traffic inspection, express delivery transportation, remote sensing survey and drawing widely. A single unmanned aerial vehicle often cannot guarantee efficient execution of tasks, and therefore multiple unmanned aerial vehicles are required to combine or cooperatively execute tasks in an unmanned cluster. The unmanned cluster cooperative task allocation refers to comprehensively considering task characteristics, platform performance and task environments, taking an optimization target as traction, and allocating tasks to a plurality of unmanned aerial vehicles, so that the cost of executing the tasks of the whole system is the lowest or the efficiency is the highest. In recent years, the problem of unmanned cluster task allocation has become a key problem to be solved in the unmanned cluster collaborative planning field. The unmanned cluster task allocation method is divided into two ideas, namely a centralized one and a distributed one. The centralized method has simple structure, poor system scale expandability and single-point faults can lead to the shutdown of the whole system, and the distributed collaborative task scheduling method becomes a research hot spot for task allocation problems in recent years.

In the distributed task allocation method, each unmanned aerial vehicle is used as an independent individual with certain intelligence, and the self information and the task information are shared, negotiated and decided among the unmanned aerial vehicles to further obtain a task allocation scheme, so that the distributed method has strong flexibility, has certain robustness to single-point faults, and can reduce safety risks and cost. The existing distributed method comprises an auction algorithm, a contract net algorithm, a corresponding improved algorithm and the like. Existing distributed task allocation algorithms rarely consider some elements in the real environment. Such as unmanned aerial vehicle flight capability, task execution time window constraints, constraints on the length of task execution time, and the like.

Disclosure of Invention

Aiming at the problem of unmanned cluster cooperative task allocation with time window constraint, the invention provides a distributed task allocation algorithm, and the method is used for the problem of task allocation with time window constraint in a complex environment, and the specific steps can be summarized as follows:

step1: and the unmanned aerial vehicle performs new task bidding according to the information such as tasks to be allocated, time window constraints of the tasks and the like, and executes a task adding stage.

Step2: and each unmanned aerial vehicle circularly executes a conflict resolution stage by utilizing a negotiation mechanism, and removes tasks which are repeatedly added with tasks and do not meet the time window constraint until all task allocation results are unchanged.

Step3: step1 and Step2 are executed in a loop until no tasks are added or removed by all unmanned aerial vehicles, or the number of iterations exceeds 20. Calculating the unassigned task set Ca if

The algorithm ends.

Step4: each unmanned aerial vehicle performs new task bidding on tasks in Ca, and performs task adding in a task reassignment stage.

Step5: and each unmanned aerial vehicle utilizes a negotiation mechanism to execute task reassignment stage conflict resolution, and the reassignment negotiation stage is circularly executed to remove tasks which are repeatedly added with tasks and do not meet the time window constraint until the gain list of all unmanned aerial vehicles is not changed.

Step6: and (4) cycling Step4 and Step5 until all unmanned aerial vehicles do not add or remove tasks, and ending the algorithm.

The distributed task allocation method is adopted, so that task change can be responded quickly, calculation speed is improved, the bandwidth requirement on a communication link is reduced, and the problem of allocation interruption caused by single-point failure in the allocation process is avoided. This provides the basis for solving task allocation problems in complex and diverse environments.

The invention provides a distributed task allocation algorithm with minimum task average completion time as an optimization target based on a distributed PI algorithm framework and by considering the problem of unmanned cluster cooperative task allocation with time window constraint. The time window constraint of the present invention pertains to a "hard constraint," i.e., tasks must be performed within the time window, otherwise not assigned. The whole algorithm runs on all unmanned aerial vehicles in parallel, and task distribution results meeting time window constraint are obtained by repeatedly executing three stages of task adding, conflict resolution and task redistribution. The invention has the main advantages that:

(1) Embedding the time window constraint of the task into a PI distributed algorithm architecture, so that the distributed tasks all meet the time window constraint;

(2) Compared with the traditional PI algorithm, a new third stage is provided for the task which cannot be allocated because the time window constraint is not met: task reassignment strategy with unmanned aerial vehicle local time cost minimum as optimization target. The task reassignment stage greatly improves the task assignment success rate;

(3) The task allocation success rate of the algorithm is high, the average completion time of the task is small, the algorithm is not influenced by the communication topology change of the unmanned aerial vehicle, and full communication can be carried out under various topologies;

(4) Constraint conditions such as unmanned aerial vehicle flight capacity, task execution time window constraint, task execution time length constraint and the like are considered.

Drawings

FIG. 1 is a flow chart of a task allocation algorithm.

Detailed Description

1. Optimization objective

The invention takes the average time cost of the tasks as an optimization target, researches the problem of unmanned cluster cooperative task allocation, and the corresponding mathematical model is as follows:

wherein (1) is an objective function representing the average time of the taskThe minimum, m is the number of tasks, n is the number of unmanned aerial vehicles, P _i Is unmanned plane u _i Task execution path of F _j (P _i ) Indicating completion of task t _j The time cost required; (2) Specifying that each task can only be assigned to one drone and cannot be repeatedly allocated; (3) Provision for unmanned aerial vehicle u _i Cannot be performed at a time beyond L _i Tasks, N _u Is an unmanned aerial vehicle set; (4) Specifying that the start execution time of a task satisfies a time window constraint, i.e. task t _j Must not be earlier than time T _{j_start} Executing no later than time T _{j_end} And executing.

2. Unmanned cluster distributed task allocation algorithm

The invention provides a distributed task allocation algorithm considering time window constraint. The proposed algorithm runs in parallel on all unmanned aerial vehicles. Information is exchanged among unmanned aerial vehicles through local communication topology, and information such as distribution results of all tasks and task gains is recorded. The algorithm mainly comprises three stages: task addition, conflict resolution, and task reassignment. The algorithm is firstly cycled and iterated to execute the first two stages, and when the task allocation object lists recorded by all unmanned aerial vehicles are consistent and unchanged for a period of time, the third stage is started to be executed, and some unallocated tasks are continuously allocated to the unmanned aerial vehicles. And after multiple iterations, finally obtaining the conflict-free task allocation result with minimum task average time cost and meeting the time window constraint.

First, each unmanned plane u is defined _i A list of recorded information.

1) Task allocation object list

Z _i ＝[Z _i1 ,…,Z _im ] ^T Representation unmanned plane u _i The distribution object, component Z, of all tasks recorded _ij K represents u _i Consider task t _j Is assigned to u _k . Note that if u _i Let t be _j Unassigned, then Z _ij Approaching infinity, i.e. Z _ij →∞。

2) Task gain list

Q _i ＝[Q _i1 ,…,Q _im ] ^T Representation unmanned plane u _i Gain for all tasks recorded. The concept of task gain will be defined in section 2.1. When Z is _ij When=k, Q _ij Equal to task t _j From P _k After deletion of u _k Is a reduction in the time cost of (a) the system. Note that if Z _ij → infinity, let Q _ij →∞。

3) Task start execution time list

T _i ＝[T _i1 ,…,T _im ] ^T Representation unmanned plane u _i The recorded start execution times of all tasks. When Z is _ij When=k, T _ij Reflecting u _k Begin to execute task t _j Is a time of (a) to be used. If Z _ij To → infinity, let T _ij →∞。

4) Timestamp list

s _i ＝[s _i1 ,…,s _in ] ^T Unmanned recording machine u _i From unmanned aerial vehicle u _h The moment of obtaining the latest information, the component s _ih The following calculation was performed:

wherein τ _r Represents u _i Receiving u _h The moment of the information.

Task gain list Q _i And task allocation object list Z _i Indicating the current time u _i Task allocation status of the whole system under view angle, u _i Recorded task start time list T _i For verifying whether the task satisfies the time window constraint. Initially, let the

Z _ij →∞，Q _ij →∞，/>

Unmanned plane u _i The four kinds of information are exchanged in communication with the neighboring unmanned aerial vehicle.

2.1 task addition stage

In the task adding stage, unmanned plane u _i With the aim of minimizing the local time cost of the task, selecting some tasks to be inserted into the task path P of the task on the premise of meeting the constraint of a time window _i . This section first gives the concept of "gain" and "marginal gain" of a task for a drone, describing when inserting a task into P, respectively _i And from P _i Unmanned plane u after middle deletion _i Own time cost phi (P _i ) Is a variation of (2).

1) Task gain. Assume task t _j Located at P _i In order to make

Representing t _j From P _i Task sequence remaining after removal of (a) gain->

Representing t _j From P _i Unmanned plane u after deletion _i The amount of change in time cost is calculated as follows:

gain of

T is reacted with _j For u _i "contribution value" of the current time cost. Furthermore, when->

Time, order

2) Task margin gain. Assume that

If t is taken _j Insertion P _i After affirmation can cause unmanned plane u _i The increase of the time cost makes the marginal gain +.>

Representing the minimum of such an increment, is calculated by: />

Wherein, the liquid crystal display device comprises a liquid crystal display device,

representation handle t _j Inserted into P _i A task sequence obtained after the kth position of (c). When t _j ∈P _i When in use, let->

Giving the unmanned plane u in the task adding stage _i When assigning tasks, only when task t _j At the same time, the following two conditions are satisfied, and the insertion of the two conditions into P is allowed _i Is a kind of medium.

Condition 1: z is Z _ij Not equal to i and

condition 2: order the

Let t _j Insertion P _i After position p of (2), the resulting sequence->

The time window constraint is satisfied by all tasks in (a).

In condition 1, Z _ij Not equal to i represents t _j Not assigned to u _i Let s=z _ij . If s is not equal to infinity, Q _ij Record will t _j From path P _s U after upper deletion _s Is a reduction in the time cost of (a) the number of time slots,

representing t _j Insertion P _i Rear u _i By a minimum increase in the time cost of (a)

Representing the requirement to be t _j Insertion P _i The cost of post global time is reduced; if s → infinity, u is represented by _i Let t be _j Not yet allocated, at this time Q _ij To increase the task allocation success rate, t is preferentially set to _j Assigned to u _i . Condition 2 indicates that t is _j Insertion P _i The time window constraint is still satisfied after the appropriate position.

Thus, to reduce global time costs as much as possible or to prioritize tasks that have not yet been allocated, we have been assembled from the set according to equation (12)

Conditions 1 and 2 for t _j Hold select task insert P _i Is the position of (2)

Let t _k Assigned to u _i Then, let Z _ik ＝i，

And recalculate P according to (1) _i Start execution time of all tasks in the list T is updated _i 。

2.2 Conflict resolution phase

The task adding process is executed in parallel on all unmanned aerial vehicles, so that the same task can be allocated to a plurality of unmanned aerial vehicles at the same time, and allocation conflicts are generated. Therefore, in order to obtain a collision-free task allocation result, a collision resolution process is also required to be performed. At this stage, the unmanned aerial vehicle repeatedly and iteratively performs two steps of negotiation and task deletion until all conflicts are eliminated and the task allocation information of all unmanned aerial vehicles remains consistent, so as to achieve global consensus, namely Q _i ＝Q _h ，

u _h ∈N _u And u is _i ≠u _h 。

2.2.1 agree step

Table 1 unmanned aerial vehicle communication rules

Suppose G _ih =1, unmanned plane u _i And adjacent unmanned plane u _h Information exchange via local communication topology, sharing task allocation object list Z _i And Z _h Task gain list Q _i And Q _h Task start execution time list T _i And T _h Time stamp s _i Sum s _h And updating corresponding information according to a certain communication rule. In fact, when unmanned plane u _i Receiving the neighboring u _h After the information is sent, the information is according to Z _i Sum s _i To determine whether the received information is task t _j With the aim of reducing the own time cost, it is determined whether to update the own stored information Z according to the communication rules shown in Table 1 _i ，Q _i And T _i 。

Table 1 contains the following three operations altogether:

update: z is Z _ij ＝Z _hj ，Q _ij ＝Q _hj ，T _ij ＝T _hj ；

Retention: no operation is performed;

reset: z is Z _ij →∞，Q _ij →∞，T _ij →∞。

Wherein the first two columns of Table 1 are sender u _h And receiver u _i Task t under view _j Is a distributed object of (a); the third column gives u _i For task t _j The actions taken, default actions are reserved. It should be noted that the storage is carried out in the unmanned plane u _i Timestamp s on _i Represents the latest time to obtain information, therefore u _i S is updated according to the formula (9) after each time of receiving information _i 。

2.2.2 task deletion step

After the step of negotiating by using the local communication network, unmanned plane u _i Checking the current task path P _i Deleting the set a= { t therefrom _j ∈P _i |Z _ij Not equal to i and set b= { t } _j ∈P _i |T _j (P _i )<T _{j_start} Or T _j (P _i )>T _{j_end} Task in }. Wherein A represents Z _i And P _i The unmatched task set, B, represents a task set that does not meet the time window constraints.

First, unmanned plane u _i First, according to equation (13), task t is calculated that minimizes execution time costs _k From P _i And a simultaneous removal in a. This process is repeated until all tasks in a no longer satisfy (13) or a is empty.

Wherein the method comprises the steps of

Representing the currentTime task t _j For unmanned plane u _i Is a gain in time cost of (a).

If the process is finished

Resetting the executor of the task in A to u _i I.e. for all remaining tasks t _j E A, let Z _ij ＝i。

Then, unmanned plane u _i Checking the self sequence P _i Once task t exists _j ∈P _i If the time window constraint is not satisfied, then t _j From sequence P _i Removed and the information list Z is reset _ij →∞，Q _ij →∞，T _ij And → infinity. Unmanned plane u _i Every time a task t is removed _j All that is required is to recalculate P _i And judging whether a new task which does not meet the time window constraint appears or not according to the starting execution time of each task. The process is repeatedly and iteratively executed until all P _i All tasks on the table meet the time window constraint. When the task deletion operation is finished, P is recalculated _i Each task t of (a) _j Gain Q of (2) _ij And a task start time T _ij 。

Unmanned plane u _i And repeating the two steps of negotiation and task deletion, wherein when the task allocation result is not changed after multiple iterations, the conflict-free task allocation is realized and the time window constraint is met, and the conflict resolution stage is completed.

2.3 task reassignment phase

In general, the task allocation result obtained after the first two phases are executed can cover all tasks, but in the case that the time window constraint is severe, some tasks may not be allocated. Let Ca denote the unassigned set of tasks after the first two phases. In order to have more tasks allocated under the condition that the time window constraint is satisfied, a third phase, task reassignment, is also performed on the tasks in Ca. The reassignment stage comprises two steps of task adding and conflict resolving, wherein only the tasks in the Ca are removed or added from the unmanned aerial vehicle, and the tasks which are assigned outside the Ca are not operated.

2.3.1 task adding step

To avoid repeated adding and deleting of tasks by the same unmanned plane, unmanned plane u _i Defining a vector M _i . Initial ream M _i [t _j ]＝0，

If t _j Quilt u _i Added to P _i But from P due to violation of the time window constraint _i After removal, let M _i [t _j ]＝1。

Task t _j E Ca is allowed to be inserted into unmanned plane u only if the following conditions 3 and 4 are satisfied _i Path P of (2) _i Is a suitable position in the (c):

condition 3: z is Z _ij ≠i，M _i [t _j ]＝0；

Condition 4: order the

Let t _j Insertion P _i After position p of (2), the resulting sequence->

The time window constraint is satisfied by all tasks in (a).

Order the

Condition 3 and condition 4 for t _j All true }. We aim at global time cost minimization from +.>

Selecting a task t _k Insertion P _i Position->

Different from the formula (12) in the section 2.1, the task t in the formula (14) is added by taking only marginal gain of the task to the unmanned aerial vehicle into consideration _k Added to P _i After that, unmanned plane u _i The local time cost increase is minimal. Repeating the task adding process until (14) is no longer satisfied or u _i The number of tasks performed has reached the maximum load.

2.3.2 Conflict resolution step

Similar to section 2.2, task reassignment conflict resolution is also divided into two parts, negotiation and task deletion.

1) Agreement with each other

Because it is a reassignment process for tasks in Ca, during this communication process, the drones only exchange information related to tasks in Ca. The agreed rules are as in table 1, and the operations are specifically performed as the update, hold and reset operations in 2.2.1.

2) Task deletion

Unmanned plane u _i Need to delete a' = { t _j ∈P _i |Z _ij Not equal to i } and B' = { t _j ∈P _i |T _j (P _i )<T _{j_start} Or T _j (P _i )>T _{j_end} Task in }. Wherein u is _i The method of removing the task in A' is the same as in section 2.2.2. And for task t _j E, B', task t _j From sequence P _i When the medium is removed, the operation is performed: z is Z _ij →∞，Q _ij →∞，T _ij →∞，M _i [t _j ]=1. At this time M _i [t _j ]Set to 1 to ensure that in the current task reassignment phase, t _j Will not be added to P _i Therefore, invalid task adding and removing are avoided, and task distribution efficiency is improved. Unmanned plane u _i Every time a task t is removed _j All that is required is to recalculate P _i And judging whether a new task which does not meet the time window constraint appears or not according to the starting execution time of the task.

Unmanned plane u _i Iteratively removing tasks in B' repeatedly until

Namely P _i All tasks in (a) satisfy the time window constraint. After the task deletion process is finished, calculate Q _ij And T _ij ，/>

/>

Claims (9)

1. An unmanned cluster distributed task allocation algorithm taking time window constraints into consideration, comprising the steps of:

step1: the unmanned aerial vehicle performs new task bidding according to the information such as tasks to be allocated, time window constraints of the tasks and the like, and performs a task adding stage;

step2: each unmanned aerial vehicle circularly executes a conflict resolution stage by utilizing a negotiation mechanism, and removes tasks which are repeatedly added with tasks and do not accord with time window constraint until all task allocation results are not changed;

step3: step1 and Step2 are circularly executed until all unmanned aerial vehicles do not add or remove tasks or the iteration number exceeds 20, the unallocated task set Ca is calculated, if

Ending the algorithm;

step4: each unmanned aerial vehicle performs new task bidding on tasks in Ca, and performs task adding in a task reassignment stage;

step5: each unmanned aerial vehicle utilizes a negotiation mechanism to execute task reassignment stage conflict resolution, and the reassignment negotiation stage is executed circularly to remove tasks which are repeatedly added with tasks and do not accord with time window constraint until gain lists of all unmanned aerial vehicles are not changed;

step6: and (4) cycling Step4 and Step5 until all unmanned aerial vehicles do not add or remove tasks, and ending the algorithm.

2. An unmanned cluster distributed task allocation algorithm taking into account time window constraints according to claim 1,the method is characterized in that the unmanned aerial vehicle and task information in the step1 comprise the following steps: position information, speed information, task allocation information, task gain information, task execution time window information, assume task t _j Is located at unmanned plane u _i To-be-tasklist P of (a) _i Wherein the task gain information is defined as t _j From P _i And after deleting the variable quantity of the time cost of the unmanned aerial vehicle, each unmanned aerial vehicle selects tasks which can reduce the time cost of the system as much as possible to be included in a task list to be executed, and the time cost of the system cannot be reduced until the number of the tasks in the task list to be executed reaches the maximum capacity or any tasks are added.

3. The unmanned cluster distributed task allocation algorithm according to claim 1, wherein the negotiation procedure in step2 is: each unmanned aerial vehicle exchanges information of each task allocation situation, task gain information and task starting execution time information with adjacent unmanned aerial vehicles through communication topology, removes tasks repeatedly bidding according to the obtained information and the principle of minimum overall time cost, removes tasks which do not meet time window constraint, repeats information exchange and task deletion processes of each unmanned aerial vehicle, and completes a conflict resolution stage when task allocation results are not changed after multiple iterations.

4. The unmanned cluster distributed task allocation algorithm according to claim 1, wherein the Ca in step3 is a set of tasks that are not currently included in the list of tasks to be performed by any unmanned aerial vehicle.

5. The unmanned cluster distributed task allocation algorithm according to claim 1, wherein task allocation in the task allocation stage in step4 is only performed for tasks in Ca, and each unmanned aerial vehicle selects tasks that can reduce the cost of system time as much as possible to be included in the task list to be executed until the number of tasks in the task list to be executed reaches the maximum capacity or no additional tasks can reduce the cost of system time.

6. The unmanned cluster distributed task allocation algorithm according to claim 1, wherein the task reallocation phase conflict resolution in step5 only exchanges information and deletes tasks in Ca, and does not perform any operation on tasks allocated before step 3.

7. The unmanned cluster distributed task allocation algorithm according to claim 1, wherein the unmanned aerial vehicle in step6 does not add or remove tasks until the current system time cost is minimized, and cannot be reduced.

8. An unmanned cluster distributed task allocation algorithm that takes into account the time window constraints according to claim 1, wherein the task addition stage, the system time cost W, is calculated as follows:

/>

wherein m is the number of tasks, n is the number of unmanned aerial vehicles, and P _i Is unmanned plane u _i Task execution path of F _j (P _i ) Indicating completion of task t _j The time cost required.

9. An unmanned cluster distributed task allocation algorithm according to claim 2, wherein the task gain information, task t, is assumed _j Located at P _i In order to make

Representing t _j From P _i Task sequence remaining after removal of the middle, task gain->

Representing t _j From P _i Unmanned plane u after deletion _i The amount of change in time cost is calculated as follows:

q _ij (P _i ！t _j )＝F(P _i )-F(P _i ！t _j )

wherein Φ (P) _i ) Representation unmanned plane u _i Own time cost, gain

T is reacted with _j For u _i "contribution value" of the current time cost, furthermore, when +.>

When in use, let->

/>

Images