Multi-AGV path planning method based on path scheme and passing priority coevolution

By splitting the population into path scheme subpopulations and pass priority subpopulations iterates in parallel, the problem of unconsidered impact of path schemes and pass priority decisions in the prior art is solved, and multi-AGV path planning in complex large-scale equipment assembly lines is realized The efficient collaboration of the multi-AGV collaborative handling tasks are suitable for multi-AGV collaborative handling tasks.

CN118917767APending Publication Date: 2024-11-08NANJING UNIV OF AERONAUTICS & ASTRONAUTICS +1
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202410126615.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-01-30
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

The existing multi-AGV path planning method fails to effectively consider the decision-making impact between path scheme and traffic priority. Especially in complex large-scale equipment assembly lines, it is difficult to end the aggregation process of multiple AGVs at the same time, and it is not suitable for multi-AGV collaborative handling tasks. .

Method used

Using a method based on path scheme and pass priority co-evolution, the population is split into path scheme subpopulations and pass priority subpopulations, and iteratively evolves in parallel. The initial population is generated through task hierarchical strategies and random solution generation rules, and the initial population is generated and utilized. The intelligent optimization algorithm is updated iteratively until the termination condition is met, and the optimal path planning scheme is output.

Benefits of technology

The decision-making impact between the path scheme and the pass priority in multi-AGV path planning is fully considered, and the path planning efficiency and quality of multi-AGV collaborative handling tasks are improved, and a better multi-AGV path planning scheme is obtained.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118917767A_ABST
    Figure CN118917767A_ABST
Patent Text Reader

Abstract

The invention relates to a multi-AGV path planning method based on path scheme and passing priority coevolution, and belongs to the field of intelligent manufacturing. When task allocation operation is executed, firstly, a task allocation scheme and a task execution sequence are obtained; then, an initial population is formed, the initial population is composed of an initial passage priority sub-population and an initial path scheme sub-population, the initial passage priority sub-population is formed according to a passage priority random solution generation rule based on a task allocation strategy, and the initial path scheme sub-population is formed according to a path scheme random solution generation rule; and finally, performing iterative computation by using an intelligent optimization algorithm to obtain a multi-AGV path planning scheme. According to the method, an effective multi-AGV path planning method is provided for a complex large-scale equipment assembly line with a multi-AGV cooperative carrying task; compared with the existing method in research, the method disclosed by the invention has the advantages that the path scheme and the passing priority are iteratively changed, and the optimal multi-AGV path planning scheme is more opportunity to obtain.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of intelligent manufacturing, and in particular to a multi-AGV path planning method based on the co-evolution of path scheme and passage priority. Background Art

[0002] An automatic guided vehicle (AGV) is an unmanned vehicle that can be used for material handling. An AGV system consisting of multiple AGVs has the advantages of efficient transportation, automation, flexibility, and rapid response. It is widely used in manufacturing workshops, warehousing, and docks. For AGV systems, multi-AGV path planning has always been a hot research topic. Its core lies in two points: one is to plan a high-quality path for each task that each AGV needs to perform, and the other is to avoid path conflicts between multiple paths. Specifically, it is the decision-making problem of the path plan for each task and the decision-making problem of the passage priority of the AGV performing each task on the conflicting path. The path plan determines the driving distance of each task and the possible path conflict; the passage priority determines how the AGV performing a low passage priority task adjusts the occupation time window of each path segment in its path plan to avoid path conflicts. Therefore, the path plan and passage priority jointly determine the performance of the multi-AGV path planning solution. It should be noted that the adjustment of the occupied time window to resolve the existing path conflict may lead to new path conflicts, that is, the path conflicts that need to be resolved under different traffic priority schemes for the same path plan are different. In addition, under the same traffic priority scheme, the path conflicts that need to be resolved by different path plans are also different. Therefore, in order to obtain a high-quality multi-AGV path planning solution, it is worthwhile to consider the decision-making impact between the path plan and the traffic priority. However, in the current multi-AGV path planning method, the strategy usually adopted is independent decision-making, that is, planning a high-quality path for each task based on a certain goal, determining the traffic priority of the AGV performing the task based on a certain rule, and then resolving the path conflicts based on time window adjustment or local path update. For example, the Chinese invention patent with the publication number CN116679694A and the patent name of a method for optimizing multi-AGV scheduling first obtains the optimal path of the delivery task of each AGV based on a genetic algorithm, then determines whether there is a conflict between the optimal paths of different delivery tasks based on the time window method, and avoids the conflict by setting a waiting time for the optimal path with a lower task priority or replacing the optimal path, and finally obtains the conflict-free and shortest delivery path for multiple AGVs. In this patent, the priority of the task, that is, the priority of passage, is fixed, and the disclosed method does not consider the decision-making influence between the path plan and the priority of passage.

[0003] In addition, in complex large-scale equipment assembly lines, the weight and size of materials vary greatly, and the AGV system needs to simultaneously handle single AGV material handling tasks for conventional-sized components and multi-AGV collaborative handling tasks for large-sized components. For multi-AGV collaborative handling tasks, it is performed collaboratively by multiple AGVs, involving the assembly process of each AGV and the delivery process performed by all AGVs. Since only after all AGVs have completed the assembly process can the AGV queue formation be performed, and then the loading is completed and the delivery process is entered, there is a unique problem when planning the path for multi-AGV collaborative handling tasks, that is, the assembly process of multiple AGVs should be ended as simultaneously as possible. However, in the existing multi-AGV path planning research, the research object is concentrated on single AGV material handling tasks, such as the Chinese invention patent application with the publication number CN116224923A and the patent name of a multi-AGV path planning method considering conflict avoidance. In the above research, there is no need to consider the unique problems of multi-AGV collaborative handling tasks, so it is not suitable for complex large-scale equipment assembly line scenarios.

[0004] Therefore, there is an urgent need for a multi-AGV path planning method that is suitable for complex large-scale equipment assembly lines and considers the decision-making impact between path plans and traffic priorities. Summary of the invention

[0005] In order to solve the technical defects existing in the prior art, the present invention proposes a multi-AGV path planning method based on the co-evolution of path scheme and traffic priority. By splitting the population into two sub-populations - the path scheme sub-population and the traffic priority solution sub-population, the two sub-populations are iteratively evolved in parallel, and the quality of each solution in the other sub-population is verified by each other during the iterative evolution, thereby fully considering the decision-making influence between the path scheme and the traffic priority in the multi-AGV path planning problem; by splitting the task into a pickup sub-task and a delivery sub-task, the problem is converted into a path scheme decision and traffic priority decision problem for each sub-task, and the unique problems of multi-AGV collaborative handling tasks can be taken into account in the iterative evolution.

[0006] The present invention is achieved through the following technical solutions:

[0007] The multi-AGV path planning method based on the co-evolution of path scheme and traffic priority includes the following steps:

[0008] Step 1: Determine whether the task allocation operation has been performed. If so, obtain the task allocation plan and task execution order and proceed to step 2. If not, proceed to step 1.

[0009] The task allocation scheme is the allocation relationship between each task and each AGV.

[0010] Step 2: According to the task stratification strategy, a task stratification table consisting of multiple task layers is formed.

[0011] The task hierarchy table stores all tasks involved in this task allocation operation.

[0012] The task consists of a single AGV handling task and a multi-AGV collaborative handling task.

[0013] The single AGV handling task consists of a pickup subtask and a delivery subtask.

[0014] The multi-AGV collaborative handling task consists of multiple picking-up subtasks and one delivery subtask.

[0015] The picking-up subtask is a task for executing the AGV to travel from its current location to the material loading point of the task.

[0016] The delivery subtask is the task of all AGVs executing the task to travel from the material loading point to the material unloading point.

[0017] Step 3: Based on the task hierarchy table and the random solution generation rule for traffic priority, multiple traffic priority solutions are randomly generated to form an initial traffic priority sub-population.

[0018] The passing priority solution is composed of the codes of the passing priorities of the subtasks of all tasks.

[0019] The passage priority of the subtask is the priority of each AGV in obtaining the path use right of the conflicting path segment when there is a path conflict between multiple AGVs executing the subtask.

[0020] Step 4: According to the random solution generation rules of the path plan, randomly generate multiple path plan solutions to form an initial path plan subpopulation.

[0021] The path solution consists of the encoding of the path solutions of the subtasks of all tasks.

[0022] Step 5: Form an initial population consisting of an initial traffic priority subpopulation and an initial path solution subpopulation, calculate the fitness of each solution in the initial population according to the fitness solution method, and determine the pairing solution of each solution.

[0023] The traffic priority sub-population and the path solution sub-population are complementary sub-populations to each other.

[0024] The fitness solution is specifically as follows:

[0025] When solving the fitness of a solution in a subpopulation, first pair the solution with each solution in the complementary subpopulation one by one to form a complete individual, then decode the complete individual to obtain the path plan and passage priority of each subtask, detect the existing path conflicts, and adjust the occupation time windows of multiple conflicting subtasks on the conflicting paths based on the passage priority of the subtasks, so as to avoid path conflicts, then calculate the fitness of all individuals, and finally take the fitness of the best individual among all complete individuals as the fitness value of the solution.

[0026] The pairing solution is a solution of a complementary subpopulation that forms an optimal individual with a solution.

[0027] Step 6: Use the intelligent optimization algorithm to iteratively update the population until the termination condition is met, output the optimal solution and its paired solution, and decode to obtain a multi-AGV path planning solution consisting of the path plan of each subtask and the passage priority of each subtask.

[0028] Furthermore, the task layering strategy includes the following steps:

[0029] Step 2.1: Let x be 1, let s be 1, put the sth task in the task execution order into the xth layer of the task hierarchy table, and put all the execution AGVs of the sth task into the task layer AGV set of the xth layer of the task hierarchy table, let s be s+1, and let a be x.

[0030] The task layer AGV set is a set of AGVs that execute all tasks placed in the task layer.

[0031] Step 2.2: Determine whether the task execution AGV set of the sth task in the task execution order has an intersection with the task layer AGV set of the ath layer of the task hierarchy table. If not, let a be a-1 and go to step 2.3. If yes, let a be a+1 and go to step 2.4.

[0032] The task execution AGV set is the set of all AGVs assigned to execute the task.

[0033] Step 2.3: Determine whether a=0 is true. If so, set a to a+1, put the sth task in the task execution order into the ath layer of the task hierarchy table, and put all the executing AGVs of the sth task into the task layer AGV set of the ath layer of the task hierarchy table, and go to step 2.5. If not, go to step 2.2.

[0034] Step 2.4: Place the sth task in the task execution sequence into the ath layer of the task hierarchy table, and place all execution AGVs of the sth task into the task layer AGV set of the ath layer of the task hierarchy table, and determine whether a>x holds. If so, let x be a and proceed to step 2.5. If not, proceed to step 2.5.

[0035] Step 2.5: Let s be s+1, and determine whether there is an sth task in the task execution order. If so, let a be x and go to step 2.2. If not, end the call of this strategy.

[0036] Furthermore, the random solution generation rule of the right of way includes the following steps:

[0037] Step 3.1: Let b be 1, copy the task hierarchy table, and obtain a temporary task hierarchy table.

[0038] Step 3.2: Set the bth layer of the temporary task hierarchy table as the coding layer, package all tasks of the bth layer of the temporary task hierarchy table to form a coding layer task set, package all variable layer tasks in the coding layer task set to form a coding layer variable layer task set, randomly generate access priorities for the subtasks of all tasks in the coding layer task set, and obtain the access priority of the bth layer.

[0039] The random generation of the right of way must meet the following conditions:

[0040] The access priority of the subtasks of all tasks in the b-th layer must be lower than the access priority of any subtask of any task in the b-1-th layer.

[0041] The variable layer task is defined as follows:

[0042] Assume that a task is located at the dth layer of the task hierarchy table, and the task hierarchy table has the d+1th layer. If all the executing AGVs of the task have no identical AGVs with the executing AGVs of all the tasks located at the d+1th layer of the task hierarchy table, then the task is a variable-layer task.

[0043] Step 3.3: Determine whether the coding layer variable layer task set is empty. If so, set b to b+1 and proceed to step 3.7. If not, set c to 1 and proceed to step 3.4.

[0044] Step 3.4: Determine whether there is a cth task in the coding layer variable layer task set. If so, set the cth task in the coding layer variable layer task set as a pending task and go to step 3.5. If not, set b to b+1 and go to step 3.7.

[0045] Step 3.5: Determine whether the b+1th layer is the task variable layer of the cth task. If so, proceed to step 3.6. If not, set c to c+1 and proceed to step 3.4.

[0046] The task variable layer is defined as follows:

[0047] Assume that a variable-layer task is located at the dth layer of the task hierarchy table. If all the execution AGVs of this task have no identical AGVs with the execution AGVs of all tasks located at the d+1th layer to the d+eth layer of the task hierarchy table, then the dth layer to the d+eth layer of the task hierarchy table are the task variable layers of this variable-layer task.

[0048] Step 3.6: Determine whether the access priorities of all subtasks of the pending task are lower than the access priorities of all subtasks of non-variable layer tasks in the coding layer task set. If so, cancel the access priorities generated for all subtasks of the pending task, and move the pending task from the bth layer to the b+1th layer of the temporary task hierarchy table, set c to c+1, and enter step 3.4. If not, set c to c+1, and enter step 3.4.

[0049] Step 3.7: Check whether the temporary task hierarchy table contains the bth layer. If so, proceed to step 3.2. If not, merge the access priority codes of all layers to obtain a randomly generated priority solution and terminate the call of this rule.

[0050] Furthermore, the path plan random solution generation rule comprises the following steps:

[0051] Step 4.1: Get the subtasks of all tasks to form a subtask set, and set f to 1.

[0052] Step 4.2: Get the task starting point and task target point of the fth subtask in the subtask set, determine the path library that matches the subtask, randomly select a path plan from the matching path library and assign it to the subtask, get the path plan code of the subtask, and let f be f+1.

[0053] The path library is composed of a plurality of path solutions with the shortest path between a specific starting point and a specific target point.

[0054] Step 4.3: Determine whether there is an fth subtask in the subtask set. If so, proceed to step 4.2. If not, merge the path plan codes of all subtasks to obtain a randomly generated path plan solution, and end the call of this rule.

[0055] Furthermore, the intelligent optimization algorithm is an improved honey badger algorithm, and its improvement includes at least one of an improved density factor update formula, a neighborhood search strategy and a cluster elimination strategy.

[0056] Furthermore, the improved density factor update formula is as follows:

[0057]

[0058] In the formula, α iis the density factor of the i-th honey badger in the population, C is a constant, the default value is 2, W is the current iteration number, W max is the maximum number of iterations, f i is the fitness value of the i-th honey badger; f max is the fitness value of the worst individual in the previous generation; f min is the fitness value of the best individual in the previous generation.

[0059] Furthermore, the neighborhood search strategy specifically searches for all neighborhood solutions of a solution. If a neighborhood solution is better than a current solution, the neighborhood solution is used to replace the original solution in the subpopulation, and the fitness of the solution and the paired solution are updated.

[0060] Furthermore, the neighborhood solution is obtained as follows:

[0061] For a path solution, a certain bit in the solution code is changed to the code of the next path solution in the matching path library, and the other bits remain unchanged, so that a neighborhood solution of the solution can be obtained;

[0062] For the passage priority solution, the codes of any two adjacent positions in the encoding of the solution are exchanged, and the other positions remain unchanged, so as to obtain a neighborhood solution of the solution.

[0063] Furthermore, the cluster elimination strategy comprises the following steps:

[0064] Step 6.1: After a subpopulation of each generation is updated, set the subpopulation as the target subpopulation, arrange the solutions in the target subpopulation from best to worst to form a sequence subIP, set the first solution in subIP as the cluster center of the first cluster, and obtain the encoding length N of the solution in the target subpopulation. RP , set y to 2, set z to 2.

[0065] Step 6.2: Determine whether there is a zth solution in subIP. If so, proceed to step 6.3; if not, proceed to step 6.4.

[0066] Step 6.3: Calculate the Hamming distance between the zth solution in subIP and the centers of all clusters, and determine whether the Hamming distances are all greater than Ceil (0.1×N RP ), if yes, set the solution as the cluster center of the y-th cluster and let y+1, if no, assign the solution to the cluster closest to it and mark it as an eliminated solution.

[0067] Step 6.4: Let z be z+1 and go to step 6.2.

[0068] Step 6.5: For each eliminated solution, generate a random solution to replace the solution in the target subpopulation, and update the fitness and pairing solution of the solution.

[0069] Compared with the prior art, the present invention has at least the following beneficial effects or advantages:

[0070] The present invention proposes a multi-AGV path planning method based on the co-evolution of path schemes and traffic priority. The method divides the population into two sub-populations, namely, the path scheme sub-population and the traffic priority solution sub-population. The two sub-populations evolve iteratively in parallel and verify the quality of each solution in the other sub-population during the iterative evolution. This method fully considers the decision-making influence between the path scheme and the traffic priority in the multi-AGV path planning problem. The method divides the task into a pickup sub-task and a delivery sub-task, and converts the problem into the path scheme decision and traffic priority decision problem of each sub-task. The unique problems of the multi-AGV collaborative handling task can be considered during the iterative evolution. In summary, the present invention provides an effective multi-AGV path planning method for complex large-scale equipment assembly lines with multi-AGV collaborative handling tasks. Compared with the methods in existing research, the path scheme and traffic priority in the method disclosed by the present invention are both iteratively changed, and there is a greater chance of obtaining the optimal multi-AGV path planning solution. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] The present invention will be further described in detail below with reference to the accompanying drawings:

[0072] Figure 1 It is a flow chart of the multi-AGV path planning method based on the co-evolution of path scheme and passage priority of the present invention;

[0073] Figure 2 An example of a task allocation scheme and the order in which tasks are to be performed;

[0074] Figure 3 for Figure 2 An example of a randomized pass priority solution for an example of . DETAILED DESCRIPTION

[0077] The present invention will be described in detail below with reference to the accompanying drawings, but the present invention is not limited thereto. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0078] Figure 1 This is a flow chart of the multi-AGV path planning method based on the co-evolution of path scheme and traffic priority according to the present invention, which specifically includes the following steps:

[0079] Step 1: Determine whether the task allocation operation has been performed. If so, obtain the task allocation plan and task execution order and proceed to step 2. If not, proceed to step 1.

[0080] The task allocation scheme is the allocation relationship between each task and each AGV.

[0081] The task allocation operation is to allocate the required number of AGVs to each task and determine the execution order of each task. The path planning of a task requires two key information: the starting point of the task and the target point of the task. The former is the target point of the previous task of the AGV executing the task, which is determined by the result of the task allocation operation. Therefore, before executing multi-AGV path planning, it is necessary to obtain the task allocation plan and the task execution order.

[0082] Step 2: According to the task stratification strategy, a task stratification table consisting of multiple task layers is formed.

[0083] The task hierarchy table stores all tasks involved in this task allocation operation.

[0084] The task consists of a single AGV handling task and a multi-AGV collaborative handling task.

[0085] The single AGV handling task consists of a pickup subtask and a delivery subtask.

[0086] The multi-AGV collaborative handling task consists of multiple picking-up subtasks and one delivery subtask.

[0087] The picking-up subtask is a task for executing the AGV to travel from its current location to the material loading point of the task.

[0088] The delivery subtask is the task of all AGVs executing the task to travel from the material loading point to the material unloading point.

[0089] The execution process of a task is divided into a picking process and a delivery process. The former is the process of the AGV driving from the current position to the material loading point of the task, and the latter is the process of the AGV driving from the material loading point of the task to the material unloading point. The former is also called a picking subtask, and the latter is also called a delivery subtask. The number of picking subtasks involved in each task is determined by the number of AGVs executing it. For example, a 4-AGV collaborative handling task needs to be assigned 4 AGVs to execute it. Therefore, the task involves 4 picking subtasks, and each picking subtask corresponds to the process of 1 AGV executing the task from its current position to the material loading point. In addition, each task only involves one delivery task, because whether it is a single AGV handling task or a multi-AGV collaborative handling task, the driving path of the AGV is the same. In the existing multi-AGV path planning method, the path for executing each task is usually planned one by one from the perspective of the AGV. In the problem solved by the present invention, if the path is planned from the perspective of the AGV, there will be a problem that the path of the delivery task of the multi-AGV collaborative handling task is repeatedly planned, which increases the workload and may have the problem of inconsistent paths. Therefore, the method proposed in the present invention is to plan the path from the perspective of subtasks to solve the above-mentioned problems.

[0090] The task stratification strategy is to divide all tasks involved in this task allocation operation into different task layers, thereby forming a task stratification table consisting of multiple task layers. The task stratification table will be used for the random generation of the passage priority solution in the subsequent steps.

[0091] The following steps are involved:

[0092] Step 2.1: Let x be 1, let s be 1, put the sth task in the task execution order into the xth layer of the task hierarchy table, and put all the execution AGVs of the sth task into the task layer AGV set of the xth layer of the task hierarchy table, let s be s+1, let a be x;

[0093] The task layer AGV set is a set of execution AGVs of all tasks placed in the task layer;

[0094] Step 2.2: Determine whether the task execution AGV set of the sth task in the task execution order has an intersection with the task layer AGV set of the ath layer in the task hierarchy table. If not, let a be a-1 and go to step 2.3. If yes, let a be a+1 and go to step 2.4.

[0095] The task execution AGV set is the set of all AGVs assigned to execute the task;

[0096] Step 2.3: Determine whether a=0 holds. If so, set a to a+1, put the sth task in the task execution order into the ath layer of the task hierarchy table, and put all the executing AGVs of the sth task into the task layer AGV set of the ath layer of the task hierarchy table, and go to step 2.5. If not, go to step 2.2;

[0097] Step 2.4: put the sth task in the task execution sequence into the ath layer of the task hierarchy table, and put all the execution AGVs of the sth task into the task layer AGV set of the ath layer of the task hierarchy table, and judge whether a>x is true. If so, let x be a and go to step 2.5. If not, go to step 2.5;

[0098] Step 2.5: Let s be s+1, and determine whether there is an sth task in the task execution order. If so, let a be x and go to step 2.2. If not, end the call of this strategy.

[0099] Figure 2This is an example of a task allocation plan and task execution sequence. This example is used to illustrate the task hierarchical strategy. Go to step 2.1, let x and s both be 1, put the first task T1 in the task execution sequence into the first layer of the task hierarchy table, update the task layer AGV set of the first layer to {G1, G2}, let s be 2, let a be 1, and go to step 2.2; since the task execution AGV set {G2, G3} of the second task T2 has an intersection with the task layer AGV set {G1, G2} of the first layer, let a be 2, and go to step 2.4; put T2 into the second layer of the task hierarchy table, and update the task layer AGV set of the second layer to {G2, G3}. Since a is 2 and x is 1, a>x holds. Let x be 2 and go to step 2.5; let s be 3. There is a third task T3 in the order, so let a be 2 and go to step 2.2; since the task execution AGV set {G2} of the third task T3 intersects with the task layer AGV set {G2, G3} of the second layer, let a be 3 and go to step 2.4; put T3 into the third layer of the task hierarchy table, and update the task layer AGV set of the third layer to {G2}. Since a is 3 and x is 2, a>x holds. Let x be 3 and go to step 2.5; let s be 4. Since there is a fourth task T4 in the task execution order, let a be 3 and go to step 2.2; since the task execution AGV set {G1} of the fourth task T4 intersects with the task layer AGV set {G2, G3} of the third layer, 2} have no intersection, so let a be 2 and go to step 2.3; since a is 2, a=0 does not hold, go to step 2.2; since the task execution AGV set {G1} of the 4th task T4 has no intersection with the task layer AGV set {G2,G3} of the 2nd layer, let a be 1 and go to step 2.3; since a is 1, a=0 does not hold, go to step 2.2; since the task execution AGV set {G1} of the 4th task T4 has an intersection with the task layer AGV set {G1,G2} of the 1st layer, let a be 2 and go to step 2.4; put T4 into the 2nd layer of the task hierarchy table, and update the task layer AGV set of the 2nd layer to {G1,G2,G3}, Since a is 2 and x is 3, a>x does not hold, and we proceed to step 2.5; let s be 5. Since there is a fifth task T5 in the task execution sequence, let a be 3 and proceed to step 2.2; since the task execution AGV set {G1, G2} of the fifth task T5 intersects with the task layer AGV set {G2} of the third layer, let a be 4 and proceed to step 2.4; put T5 into the fourth layer of the task hierarchy table, and update the task layer AGV set of the fourth layer to {G1, G2}. Since a is 4 and x is 3, a>x holds, let x be 4 and proceed to step 2.5; let s be 6. Since there is no sixth task in the task execution sequence, the call of this strategy ends. In summary, Figure 2The task hierarchy table obtained according to the task hierarchy strategy consists of 4 task layers. The first layer includes T1, the second layer includes T2 and T4, the third layer includes T3, and the fourth layer includes T5. The task layer AGV set of the first layer is {G1, G2}, the task layer AGV set of the second layer is {G1, G2, G3}, the task layer AGV set of the third layer is {G2}, and the task layer AGV set of the fourth layer is {G1, G2}.

[0100] Step 3: Based on the task hierarchical table and the random solution generation rule of the traffic priority, a plurality of traffic priority solutions are randomly generated to form an initial traffic priority sub-population;

[0101] The passage priority solution is composed of the codes of the passage priorities of the subtasks of all tasks;

[0102] The passage priority of the subtask is the priority of each AGV to obtain the path use right of the conflicting path segment when there is a path conflict between multiple AGVs executing subtasks;

[0103] The multi-AGV path planning method based on the co-evolution of path scheme and traffic priority proposed in the present invention adopts the strategy of co-evolution of traffic priority sub-population and path scheme sub-population. Therefore, before iterative calculation, multiple initial traffic priority solutions and path scheme solutions need to be generated respectively to form the traffic initial priority sub-population and the initial path scheme sub-population. For the solutions that constitute the population, it is necessary to satisfy that the solution is feasible.

[0104] by Figure 2 Taking the example of the task allocation scheme and task execution order in as an example, the subtasks of each task are as follows: T1 includes 2 pickup subtasks ST 1,1 、ST 1,2 , 1 delivery subtask T 1,D ; T2 includes 2 pickup subtasks ST 2,1 、ST 2,2 , 1 delivery subtask T 2,D ; T3 includes 1 pickup subtask ST 3,1 , 1 delivery subtask T 3,D ; T4 includes 1 pickup subtask ST 4,1 , 1 delivery subtask T 4,D ; T5 includes 2 pickup subtasks ST 5,1 、ST 5,2 , 1 delivery subtask T 5,D .by Figure 2 Taking the example of the task allocation scheme and task execution order in as an example, the subtask execution order of each AGV is: the subtask execution order of G1 is ST 1,1 -ST 1,D -ST4,1 -ST 4,D -ST 5,1 -ST 5,D ; The execution order of G2's subtasks is ST 1,2 -ST 1,D -ST 2,1 -ST 2,D -ST 3,1 -ST 3,D -ST 5,2 -ST 5,D ; The execution order of G3's subtasks is ST 2,2 -ST 2,D .

[0105] Figure 3 for Figure 2 An example of a random priority solution for an example, the code of a randomly generated priority solution is [0.51 0.53 0.52 0.81 0.01 0.05 0.54 0.55 0.56 0.82 0.06 0.07 0.080.09 0.57 0.58]. The decoding rule is that the smallest 6 numbers in the code correspond to the 6 subtasks that G1 needs to perform, the next smallest 8 numbers correspond to the 8 subtasks that G2 needs to perform, and the remaining 2 numbers correspond to the 2 subtasks that G3 needs to perform. The decoded priority is ST 1,2 >ST 1,D >ST 2,1 >ST 2,2 >ST 1,1 >ST 1,D >ST 2,D >ST 3,1 >ST 3,D >ST 2,D >ST 4,1 >ST 4,D >ST 5,1 >ST 5,D >ST 5,2 >ST 5,D According to the decoding, ST 1,2 The priority is the highest among all subtasks. For task T1, its delivery subtask ST 1,D Appeared 2 times, only in the T1 pickup subtask ST 1,1 and ST 1,2 The first ST that appears after 1,D Valid, that is, located in ST 1,2 After ST 1,D is invalid, located in ST 1,1 After ST 1,D It is valid. From the decoding results, it can be found that ST 2,1 The priority is higher than ST1,D , since G2 needs to execute ST 1,D ST can be executed after 2,1 , so the random pass priority solution is infeasible.

[0106] The initial traffic priority subpopulation is composed of a plurality of randomly generated traffic priority solutions, which are generated based on the task hierarchy table and according to the traffic priority random solution generation rules.

[0107] The random solution generation rule of right of way includes the following steps:

[0108] Step 3.1: Let b be 1, copy the task hierarchy table, and obtain a temporary task hierarchy table.

[0109] Step 3.2: Set the bth layer of the temporary task hierarchy table as the coding layer, package all tasks of the bth layer of the temporary task hierarchy table to form a coding layer task set, package all variable layer tasks in the coding layer task set to form a coding layer variable layer task set, randomly generate access priorities for the subtasks of all tasks in the coding layer task set, and obtain the access priority of the bth layer.

[0110] The random generation of the right of way must meet the following conditions:

[0111] The access priority of the subtasks of all tasks in the b-th layer must be lower than the access priority of any subtask of any task in the b-1-th layer.

[0112] The variable layer task is defined as follows:

[0113] Assume that a task is located at the dth layer of the task hierarchy table, and the task hierarchy table has the d+1th layer. If all the executing AGVs of the task have no identical AGVs with the executing AGVs of all the tasks located at the d+1th layer of the task hierarchy table, then the task is a variable-layer task.

[0114] from Figure 3 From the example, we can see that there are situations where the randomly generated traffic priority solution is infeasible. From the specific steps of the task stratification strategy, we can see that the task stratification strategy follows the following rules: first, tasks are divided into the earlier task layers as much as possible; second, two tasks with an intersection in the task execution AGV set need to be divided into different task layers; then, for two tasks with an intersection in the task execution AGV set, the task with an earlier task execution order should be divided into an earlier task layer. Therefore, based on the task stratification table, a traffic priority lower than the subtasks of each task in the previous layer is assigned to the subtasks of each task in the current layer layer by layer, so as to avoid the situation where the generated random traffic priority is infeasible.

[0115] by Figure 2Taking the task hierarchy table obtained according to the task hierarchy strategy in the example of task allocation plan and task execution order as an example, T4 is located at the 2nd level of the task hierarchy table, and the task hierarchy table has a 3rd level. All the execution AGVs of T4 are G1, and the execution AGVs of all tasks located at the 3rd level of the task hierarchy table are G2. There is no identical AGV, so T4 is a variable-level task.

[0116] Step 3.3: Determine whether the coding layer variable layer task set is empty. If so, set b to b+1 and proceed to step 3.7. If not, set c to 1 and proceed to step 3.4.

[0117] Step 3.4: Determine whether there is a cth task in the coding layer variable layer task set. If so, set the cth task in the coding layer variable layer task set as a pending task and go to step 3.5. If not, set b to b+1 and go to step 3.7.

[0118] Step 3.5: Determine whether the b+1th layer is the task variable layer of the cth task. If so, proceed to step 3.6. If not, set c to c+1 and proceed to step 3.4.

[0119] The task variable layer is defined as follows:

[0120] Assume that a variable-layer task is located at the dth layer of the task hierarchy table. If all the execution AGVs of this task have no identical AGVs with the execution AGVs of all tasks located at the d+1th layer to the d+eth layer of the task hierarchy table, then the dth layer to the d+eth layer of the task hierarchy table are the task variable layers of this variable-layer task.

[0121] by Figure 2 Taking the task hierarchy table obtained according to the task hierarchy strategy as an example of the task allocation plan and task execution order, T4 is a variable-layer task located at the 2nd layer of the task hierarchy table. The execution AGVs of all tasks from the 3rd layer to the 4th layer of the task hierarchy table are G1 and G2, and there is no identical AGV. Therefore, the variable layer of T4's task is from the 3rd layer to the 4th layer of the task hierarchy table.

[0122] Step 3.6: Determine whether the access priorities of all subtasks of the pending task are lower than the access priorities of all subtasks of non-variable layer tasks in the coding layer task set. If so, cancel the access priorities generated for all subtasks of the pending task, and move the pending task from the bth layer to the b+1th layer of the temporary task hierarchy table, set c to c+1, and enter step 3.4. If not, set c to c+1, and enter step 3.4.

[0123] If the next layer is still a variable layer of a variable layer task, when the access priority of all subtasks of the variable layer task is lower than that of the subtasks of the non-variable layer task in the encoding of this layer, the encoding of the task in this layer can be cancelled and the task can be put into the next layer for encoding. This can improve the diversity of solutions.

[0124] Step 3.7: Check whether the temporary task hierarchy table contains the bth layer. If so, proceed to step 3.2. If not, merge the access priority codes of all layers to obtain a randomly generated priority solution and terminate the call of this rule.

[0125] by Figure 2 Taking the example of the task allocation scheme and task execution order in as an example, the random solution generation rule for the right of way is explained. First, let b be 1, copy the task hierarchy table, get the temporary task hierarchy table, and go to step 3.2; set the first layer of the temporary task hierarchy table as the coding layer, pack all the tasks of the first layer of the temporary task hierarchy table to form the coding layer task set {T1}. Since there is no variable layer task in the coding layer task set, the coding layer variable layer task set is empty, which is the three subtasks ST of T1. 1,1 ,ST 1,2 ,ST 1,D Randomly generate the priority of passage. It should be noted that although T1 has only 3 subtasks, the execution order of its subtasks of AGV G1 and G2 involves ST 1,D To facilitate subsequent decoding operations, the encoding length will be increased by 1 bit to 4 bits, and a priority order of ST will be randomly generated. 1,1 >ST 1,2 >ST 1,D >ST 1,D , the corresponding execution AGV sequence is G1-G2-G1-G2; go to step 3.3, since the coding layer variable layer task set is empty, set b to 2, and go to step 3.7; since the temporary task hierarchy table has the second layer, go to step 3.2; set the second layer of the temporary task hierarchy table as the coding layer, pack all the tasks of the second layer of the temporary task hierarchy table to form the coding layer task set {T2, T4}, form the coding layer variable layer task set {T4}, randomly generate a pass priority sequence for the second layer, and randomly generate a priority sequence of ST 2,1 >ST 2,D >ST 2,2 >ST 2,D >ST 4,1 >ST 4,D, the corresponding execution order of AGV is G2-G2-G3-G3-G1-G1. To meet the conditions for random generation of traffic priority, we only need to put the priority of the second layer after the priority of the first layer when merging the traffic priorities of each layer, and then go to step 3.3; since the coding layer variable layer task set is not empty, let c be 1, and go to step 3.4; the first task in the coding layer variable layer task set exists, which is T4, and T4 is set as the pending task, and then go to step 3.5; since the third layer is the task variable layer of T4, we go to step 3.6; since the subtask ST of T4 4,1 and ST 4,D The priority of the subtasks of the coding layer is lower than that of the subtasks of other tasks. Therefore, the priority order of the second layer is modified to ST 2,1 >ST 2,D >ST 2,2 >ST 2,D , the corresponding execution AGV sequence is G2-G2-G3-G3, let c be 2, go to step 3.4; there is no second task in the coding layer variable layer task set, so let b be 3, go to step 3.7; since there is a third layer in the temporary task hierarchy table, go to step 3.2; set the third layer of the temporary task hierarchy table as the coding layer, pack all tasks of the third layer of the temporary task hierarchy table to form the coding layer task set {T3, T4}, form the coding layer variable layer task set {T4}, randomly generate a pass priority sequence for the third layer, and randomly generate a priority sequence of ST 3,1 >ST 4,1 >ST 3,D >ST 4,D , the corresponding execution AGV sequence is G2-G1-G2-G1, go to step 3.3; since the coding layer variable layer task set is not empty, set c to 1, go to step 3.4; the first task in the coding layer variable layer task set exists, which is T4, set T4 as the pending task, go to step 3.5; since the third layer is the task variable layer of T4, go to step 3.6; since the subtask ST of T4 4,1 and ST 4,D The passage priority of is not lower than the passage priority of the subtask of T3, so let c be 2 and go to step 3.4; there is no second task in the coding layer variable layer task set, so let b be 4 and go to step 3.7; since there is a 4th layer in the temporary task hierarchy table, go to step 3.2; set the 4th layer of the temporary task hierarchy table as the coding layer, pack all the tasks of the 4th layer of the temporary task hierarchy table to form the coding layer task set {T5}, the coding layer variable layer task set is empty, randomly generate a passage priority sequence for the 4th layer, and randomly generate a priority sequence of ST 5,1 >ST 5,2 >ST 5,D >ST5,D , the corresponding execution AGV sequence is G1-G2-G2-G1, go to step 3.3; since the encoding layer variable layer task set is empty, let b be 5, go to step 3.7; since the temporary task hierarchy table does not have a 5th layer, the combined access limit of all layers is ST 1,1 >ST 1,2 >ST 1,D >ST 1,D >ST 2,1 >ST 2,D >ST 2,2 >ST 2,D >ST 3,1 >ST 4,1 >

[0126] ST 3,D >ST 4,D >ST 5,1 >ST 5,2 >ST 5,D >ST 5,D , the corresponding execution AGV order is G1-G2-G1-G2-G2-G2-G3-G3-G2-G1-G2-G1-G1-G2-G2-G1. Randomly generate 16 numbers between (0,1), randomly place the 6 smallest numbers at position G1 in the execution AGV order, the next smallest 8 numbers at position G2 in the execution AGV order, and the remaining numbers at position G3 in the execution AGV order. Get a randomly generated code for the passage priority solution and end the call of this rule.

[0127] Step 4: According to the random solution generation rule of the path plan, randomly generate multiple path plan solutions to form an initial path plan subpopulation;

[0128] The path solution consists of the encoding of the path solutions of the subtasks of all tasks;

[0129] The random solution generation rule of the path plan includes the following steps:

[0130] Step 4.1: Get the subtasks of all tasks to form a subtask set, and set f to 1;

[0131] Step 4.2: Get the task starting point and task target point of the fth subtask in the subtask set, determine the path library matching the subtask, randomly select a path plan from the matching path library and assign it to the subtask, get the path plan code of the subtask, and let f be f+1;

[0132] The path library is composed of a plurality of path solutions with the shortest path between a specific starting point and a specific target point;

[0133] In an actual manufacturing environment, material storage points, material demand points, and AGV parking points are countable, and the starting point and target point of any subtask belong to one of these three. Therefore, multiple path libraries can be built to cover the path solutions between the starting point and the target point of all subtasks. Each path library stores multiple path solutions with the shortest path between a specific starting point and a specific target point. In this implementation case, each path library stores 100 paths. Assume that the subtask ST 1,D Assign the 21st path solution in the matching path library, then ST 1,D The path scheme is coded as 0.21.

[0134] Step 4.3: Determine whether there is an fth subtask in the subtask set. If so, proceed to step 4.2. If not, merge the path plan codes of all subtasks to obtain a randomly generated path plan solution, and end the call of this rule.

[0135] Step 5: Form an initial population consisting of an initial traffic priority subpopulation and an initial path solution subpopulation, calculate the fitness of each solution in the initial population according to the fitness solution method, and determine the pairing solution of each solution;

[0136] The passage priority sub-population and the path solution sub-population are complementary sub-populations to each other;

[0137] The fitness solution is specifically as follows:

[0138] When solving the fitness of a solution in a subpopulation, first pair the solution with each solution in the complementary subpopulation one by one to form a complete individual, then decode the complete individual to obtain the path plan and passage priority of each subtask, detect the existing path conflicts, and adjust the occupation time windows of multiple conflicting subtasks on the conflicting paths based on the passage priority of the subtasks, so as to avoid path conflicts, then calculate the fitness of all individuals, and finally take the fitness of the best individual among all complete individuals as the fitness value of the solution;

[0139] The pairing solution is a solution of a complementary subpopulation that forms an optimal individual with a solution;

[0140] The fitness of the solution is related to the objective function. In this implementation case, the objective function is set as follows:

[0141] f=min(w1×C u +w2×C l +w3×C T ) (1)

[0142] In the formula, w1, w2 and w3 are weights, C u is the average idle time of the task, C l is the average load duration of the task, CT is the average delay time of the task. w1 is 0.1, w2 is 0.2, and w3 is 0.7.

[0143] The fitness solution method is to solve and calculate the fitness value of the solution based on the objective function. The solution is based on the occupation time window of each task on each path segment, so as to obtain the indicators concerned in the objective function, such as the average idle time of the task, the average load time of the task and the average delay time of the task in this implementation case. It should be noted that the absence of conflict between the paths of multiple AGVs is a necessary requirement for solving the multi-AGV path planning problem. Therefore, the occupation time window based on the fitness solution method must be satisfied, and each task will not conflict when traveling according to this time window, that is, a conflict-free occupation time window. In the present invention, the solution is divided into two categories, one is the path plan solution, and the other is the traffic priority solution. Decoding the path plan solution can obtain the path plan of each subtask, and then obtain the theoretical occupation time window of each path segment when the AGV executes each subtask. There may be path conflicts, so a traffic priority solution is required to adjust the occupation time windows of the conflicting paths for multiple conflicting subtasks. The priority of passage is the priority of each AGV to obtain the right to use the path of the conflicting path segment when there is a path conflict between multiple AGVs executing subtasks. When adjusting the occupied time windows of the conflicting paths of multiple conflicting subtasks, the occupied time windows of the conflicting paths of the low-pass priority subtasks should be adjusted to avoid path conflicts with the high-priority subtasks on the conflicting paths. Similarly, decoding the passage priority solution can obtain the passage priority of each subtask, that is, the priority of the path use rights of each subtask to the conflicting path when the path conflict occurs. It also requires a path plan solution to obtain the conflict-free occupied time window of each subtask for each path segment. Therefore, when solving the fitness of a solution, it is necessary to pair the solution with the solutions in the complementary subpopulation one by one to form a complete individual that can be decoded as a conflict-free occupied time window. When pairing, multiple complete individuals can be obtained, and the fitness of the best individual is selected as the fitness of the solution, and the solution in the complementary subpopulation that is paired with the solution to form the best individual is marked as the paired solution of the solution.

[0144] Step 6: Use the intelligent optimization algorithm to iteratively update the population until the termination condition is met, output the optimal solution and its paired solution, and decode to obtain a multi-AGV path planning solution consisting of the path plan of each subtask and the passage priority of each subtask.

[0145] In this implementation case, the termination condition is set to reach the maximum number of iterations.

[0146] Using intelligent optimization algorithms, in each iteration, each solution in the path plan subpopulation is updated, and each solution in the traffic priority subpopulation is updated. When updating, the fitness and paired solutions corresponding to each solution need to be updated simultaneously. When the termination condition is met, a solution with the best fitness is selected from the two subpopulations of the current generation, and the optimal solution and its paired solution are output and decoded to obtain the path plan and traffic priority of each subtask, which is the final multi-AGV path planning solution.

[0147] In this implementation case, the intelligent optimization algorithm used is an improved honey badger algorithm, and the improvements of the improved honey badger algorithm include at least one of an improved density factor update formula, a neighborhood search strategy, and a cluster elimination strategy.

[0148] In the standard honey badger algorithm, the honey badger position update formula is as follows:

[0149]

[0150] Where P prey is the prey position, i.e., the current optimal position; F is a flag used to change the search direction, with a value of 1 or -1; β≥1 represents the ability of the honey badger to capture food; I i is the intensity of the prey odor, which is related to the distance between the prey and the honey badger and the distance between the honey badgers; r, r1, r2 and r3 are random numbers between [0,1]. When r<0.5, the mining mode is executed, otherwise the honey collection mode is executed; α is the density factor, which is used to balance the global development and local exploration capabilities of the algorithm; d i is the distance between the prey and the i-th honey badger.

[0151] In the standard honey badger algorithm, the density factor update formula is as follows:

[0152]

[0153] Where C is a constant, the default value is 2, W is the current iteration number, and W max is the maximum number of iterations.

[0154] The improved density factor update formula is as follows:

[0155]

[0156] In the formula, α i is the density factor of the i-th honey badger in the population, f i is the fitness value of the i-th honey badger; f max is the fitness value of the worst individual in the previous generation; f min is the fitness value of the best individual in the previous generation.

[0157] From the honey badger position update formula of the standard honey badger algorithm, it can be seen that the density factor value affects the search range of the honey badger. From the density factor update formula of the standard honey badger algorithm, it can be seen that it is only related to the current number of iterations and ignores the quality differences between honey badgers. Since the better the quality of the honey badger individual, the greater the hope that there is a better solution nearby, so using formula (4) to update the density factor may cause the skipping of neighboring better solutions, thereby affecting the convergence speed of the algorithm.

[0158] The improved density factor update formula fully considers the quality differences of different honey badger individuals. The better the quality of the honey badger individual, the more it tends to search in the neighboring area, thereby increasing the probability of finding a better solution in the neighborhood and improving the convergence speed of the algorithm.

[0159] The neighborhood search strategy specifically searches for all neighborhood solutions of the solution. If the neighborhood solution is better than the current solution, the neighborhood solution is used to replace the original solution in the subpopulation, and the fitness and pairing solution of the solution are updated.

[0160] The neighborhood solution is obtained as follows:

[0161] For a path solution, a certain bit in the solution code is changed to the code of the next path solution in the matching path library, and the other bits remain unchanged, so that a neighborhood solution of the solution can be obtained;

[0162] For the passage priority solution, the codes of any two adjacent positions in the encoding of the solution are exchanged, and the other positions remain unchanged, so as to obtain a neighborhood solution of the solution.

[0163] Introducing a neighborhood search strategy into the standard honey badger algorithm can improve the local search capability of the algorithm, thereby accelerating the convergence of the algorithm.

[0164] The cluster elimination strategy includes the following steps:

[0165] Step 6.1: After a subpopulation of each generation is updated, set the subpopulation as the target subpopulation, arrange the solutions in the target subpopulation from best to worst to form a sequence subIP, set the first solution in subIP as the cluster center of the first cluster, and obtain the encoding length N of the solution in the target subpopulation. RP , set y to 2, set z to 2;

[0166] Step 6.2: Determine whether there is a z-th solution in subIP. If so, proceed to step 6.3. If not, proceed to step 6.4.

[0167] Step 6.3: Calculate the Hamming distance between the zth solution in subIP and the centers of all clusters, and determine whether the Hamming distances are all greater than Ceil (0.1×N RP), if yes, set the solution as the cluster center of the y-th cluster, let y+1, if no, assign the solution to the cluster closest to it and mark it as an eliminated solution;

[0168] Step 6.4: Let z be z+1 and go to step 6.2;

[0169] Step 6.5: For each eliminated solution, generate a random solution to replace the solution in the target subpopulation, and update the fitness and pairing solution of the solution.

[0170] From the honey badger position update formula of the standard honey badger algorithm, it can be seen that it focuses on the optimal solution position, so the algorithm is prone to fall into local optimality. The task completion time of multi-AGV collaborative handling tasks is related to the assembly completion time of each executing AGV, that is, the time it takes for each executing AGV to complete the pickup path of the corresponding pickup subtask. Since the assembly completion time is only related to the completion time of the pickup path of the corresponding pickup subtask of the AGV that arrives at the assembly point last, iteratively updating the pickup paths of other pickup subtasks will not improve the assembly completion time. Therefore, for the multi-AGV path planning problem involving multi-AGV collaborative handling tasks, it has the problem of many local optimal traps, which will further aggravate the premature maturity of the algorithm.

[0171] Introducing a cluster elimination strategy into the standard honey badger algorithm can ensure the population diversity of each sub-population during the algorithm iteration process, thereby alleviating the problem of premature maturity when the algorithm solves problems.

[0172] The above description is only a preferred embodiment of the present invention. It should be pointed out that a person skilled in the art can make several improvements without departing from the principle of the present invention, and these improvements should also be regarded as within the protection scope of the present invention.

Claims

1. Multi-AGV path planning method based on co-evolution of path scheme and traffic priority, It is characterized in that The following steps are involved: Step 1: Determine whether the AGV has executed the task allocation operation. If so, obtain the task allocation plan and task execution order and proceed to step 2. If not, proceed to step 1. The task allocation scheme is the allocation relationship between each task and each AGV; Step 2: According to the task stratification strategy, a task stratification table consisting of multiple task layers is formed; The task hierarchy table stores all tasks involved in this task allocation operation; The tasks include single AGV handling tasks and multi-AGV collaborative handling tasks; The single AGV handling task consists of a pickup subtask and a delivery subtask; The multi-AGV collaborative handling task consists of multiple pickup subtasks and one delivery subtask; The picking-up subtask is a task for executing the AGV to travel from its location to the material loading point of the task; The delivery subtask is the task of all AGVs executing the task to travel from the material loading point to the material unloading point; Step 3: Based on the task hierarchical table and according to the random solution generation rule of the traffic priority, a plurality of traffic priority solutions are randomly generated to form an initial traffic priority sub-population; The passage priority solution is composed of the codes of the passage priorities of the subtasks of all tasks; The passage priority of the subtask is the priority of each AGV to obtain the path use right of the conflicting path segment when there is a path conflict between multiple AGVs executing subtasks; Step 4: According to the random solution generation rule of the path plan, randomly generate multiple path plan solutions to form an initial path plan subpopulation; The path solution consists of the encoding of the path solutions of the subtasks of all tasks; Step 5: Form an initial population consisting of an initial traffic priority subpopulation and an initial path solution subpopulation, calculate the fitness of each solution in the initial population according to the fitness solution method, and determine the pairing solution of each solution; The passage priority sub-population and the path solution sub-population are complementary sub-populations to each other; The fitness solution is specifically as follows: When solving the fitness of a solution in a subpopulation, first pair the solution with each solution in the complementary subpopulation one by one to form a complete individual, then decode the complete individual to obtain the path plan and passage priority of each subtask, detect the existing path conflicts, and adjust the occupation time windows of multiple conflicting subtasks on the conflicting paths based on the passage priority of the subtasks, so as to avoid path conflicts, then calculate the fitness of all individuals, and finally take the fitness of the best individual among all complete individuals as the fitness value of the solution; The pairing solution is a solution of a complementary subpopulation that forms an optimal individual with a solution; Step 6: Use the improved honey badger algorithm to iteratively update the population until the termination condition is met, output the optimal solution and its paired solution, and decode to obtain a multi-AGV path planning solution consisting of the path plan of each subtask and the passage priority of each subtask.

2. The multi-AGV path planning method based on the co-evolution of path scheme and traffic priority according to claim 1, It is characterized in that The forming of a task hierarchy table consisting of multiple task layers according to the task hierarchy strategy comprises the following steps: Step 2.1: Let x be 1, let s be 1, put the sth task in the task execution order into the xth layer of the task hierarchy table, and put all the execution AGVs of the sth task into the task layer AGV set of the xth layer of the task hierarchy table, let s be s+1, let a be x; The task layer AGV set is a set of execution AGVs of all tasks placed in the task layer; Step 2.2: Determine whether the task execution AGV set of the sth task in the task execution order has an intersection with the task layer AGV set of the ath layer in the task hierarchy table. If not, let a be a-1 and go to step 2.

3. If yes, let a be a+1 and go to step 2.

4. The task execution AGV set is the set of all AGVs assigned to execute the task; Step 2.3: Determine whether a=0 holds. If so, set a to a+1, put the sth task in the task execution order into the ath layer of the task hierarchy table, and put all the executing AGVs of the sth task into the task layer AGV set of the ath layer of the task hierarchy table, and go to step 2.

5. If not, go to step 2.2; Step 2.4: put the sth task in the task execution sequence into the ath layer of the task hierarchy table, and put all the execution AGVs of the sth task into the task layer AGV set of the ath layer of the task hierarchy table, and judge whether a>x is true. If so, let x be a and go to step 2.

5. If not, go to step 2.5; Step 2.5: Let s be s+1, and determine whether there is an sth task in the task execution order. If so, let a be x and go to step 2.

2. If not, end the call of this strategy.

3. The multi-AGV path planning method based on the co-evolution of path scheme and traffic priority according to claim 1, It is characterized in that The step 3 comprises the following steps: Step 3.1: Let b be 1, copy the task hierarchy table, and obtain a temporary task hierarchy table; Step 3.2: Set the bth layer of the temporary task hierarchy table as the coding layer, pack all tasks of the bth layer of the temporary task hierarchy table to form a coding layer task set, pack all variable layer tasks in the coding layer task set to form a coding layer variable layer task set, randomly generate access priorities for the subtasks of all tasks in the coding layer task set, and obtain the access priority of the bth layer; The randomly generated traffic priority must meet the following conditions: The priority of all subtasks of tasks in layer b must be lower than the priority of any subtask of any task in layer b-1; The variable layer task is defined as follows: Assume that a task is located at the dth layer of the task hierarchy table, and the task hierarchy table has the d+1th layer. If all the executing AGVs of the task have no identical AGVs with the executing AGVs of all the tasks located at the d+1th layer of the task hierarchy table, then the task is a variable-layer task. Step 3.3: Determine whether the coding layer variable layer task set is empty. If so, set b to b+1 and proceed to step 3.

7. If not, set c to 1 and proceed to step 3.

4. Step 3.4: Determine whether there is a c-th task in the coding layer variable layer task set. If so, set the c-th task in the coding layer variable layer task set as a pending task and proceed to step 3.

5. If not, set b to b+1 and proceed to step 3.

7. Step 3.5: Determine whether the b+1th layer is the task variable layer of the cth task. If so, proceed to step 3.

6. If not, set c to c+1 and proceed to step 3.

4. The task variable layer is defined as follows: Assume that a variable-layer task is located at the dth layer of the task hierarchy table. If all the executing AGVs of this task have no identical AGVs with the executing AGVs of all the tasks located at the d+1th layer to the d+eth layer of the task hierarchy table, then the dth layer to the d+eth layer of the task hierarchy table are the task variable layers of this variable-layer task. Step 3.6: Determine whether the access priorities of all subtasks of the pending task are lower than the access priorities of all subtasks of the non-variable layer tasks in the coding layer task set. If so, cancel the access priorities generated for all subtasks of the pending task, and move the pending task from the bth layer to the b+1th layer of the temporary task hierarchy table, set c to c+1, and go to step 3.

4. If not, set c to c+1, and go to step 3.

4. Step 3.7: Check whether the temporary task hierarchy table contains the bth layer. If so, proceed to step 3.

2. If not, merge the access priority codes of all layers to obtain a randomly generated priority solution.

4. The multi-AGV path planning method based on path scheme and traffic priority co-evolution according to claim 1, It is characterized in that The S4 comprises the following steps: Step 4.1: Get the subtasks of all tasks to form a subtask set, and set f to 1; Step 4.2: Get the task starting point and task target point of the fth subtask in the subtask set, determine the path library matching the subtask, randomly select a path plan from the matching path library and assign it to the subtask, get the path plan code of the subtask, and let f be f+1; The path library is composed of a plurality of path solutions with the shortest path between a specific starting point and a specific target point; Step 4.3: Determine whether there is an fth subtask in the subtask set. If so, proceed to step 4.

2. If not, merge the path plan codes of all subtasks to obtain a randomly generated path plan solution.

5. The multi-AGV path planning method based on the co-evolution of path scheme and traffic priority according to claim 1, It is characterized in that The improved honey badger algorithm comprises one of an improved density factor update formula, a neighborhood search strategy and a cluster elimination strategy.

6. The multi-AGV path planning method based on the co-evolution of path scheme and traffic priority according to claim 5, It is characterized in that The improved density factor update formula is: In the formula, α i is the density factor of the i-th honey badger in the population, C is a constant, the default value is 2, W is the current iteration number, W max is the maximum number of iterations, f i is the fitness value of the i-th honey badger; f max is the fitness value of the worst individual in the previous generation; f min is the fitness value of the best individual in the previous generation.

7. The multi-AGV path planning method based on path scheme and traffic priority co-evolution according to claim 5, It is characterized in that The neighborhood search strategy specifically searches for all neighborhood solutions of a solution. If a neighborhood solution is better than a current solution, the neighborhood solution is used to replace the original solution in the subpopulation, and the fitness of the solution and the paired solution are updated.

8. The multi-AGV path planning method based on path plan and traffic priority co-evolution according to claim 5, It is characterized in that The cluster elimination strategy comprises the following steps: Step 6.1: After a subpopulation of each generation is updated, set the subpopulation as the target subpopulation, arrange the solutions in the target subpopulation from best to worst to form a sequence subIP, set the first solution in subIP as the cluster center of the first cluster, and obtain the encoding length N of the solution in the target subpopulation. RP , set y to 2, set z to 2; Step 6.2: Determine whether there is a z-th solution in subIP. If so, proceed to step 6.

3. If not, proceed to step 6.

4. Step 6.3: Calculate the Hamming distance between the zth solution in subIP and the centers of all clusters, and determine whether the Hamming distances are all greater than Ceil (0.1×N RP ), if yes, set the solution as the cluster center of the y-th cluster, let y+1, if no, assign the solution to the cluster closest to it and mark it as an eliminated solution; Step 6.4: Let z be z+1 and go to step 6.2; Step 6.5: For each eliminated solution, generate a random solution to replace the solution in the target subpopulation, and update the fitness and pairing solution of the solution.

9. The multi-AGV path planning method based on path scheme and traffic priority co-evolution according to claim 7, It is characterized in that The neighborhood solution is obtained as follows: For a path solution, a certain bit in the solution code is changed to the code of the next path solution in the matching path library, and the other bits remain unchanged, thereby obtaining a neighborhood solution of the solution; For a passage priority solution, the codes of any two adjacent positions in the encoding of the solution are swapped, and the other positions remain unchanged, to obtain a neighborhood solution of the solution.

Citation Information

Cited By

  • AGV dynamic path planning method based on multi-modal navigation and Internet of Things data fusion

    CN121613900A