Method for multi-AGV multi-task distribution

Abstract

The invention discloses a method for multi-AGV and multi-task allocation, and the method comprises the steps: obtaining real-time tasks, and sorting the tasks according to dynamic priorities; according to AGV states and task information, a charging decision is considered, an upper layer model of multi-AGV and multi-task allocation is established, scheduling optimization is carried out by using an APSO-NSGA-III hybrid algorithm, and a task queue to be executed is allocated for each AGV; according to task queue information generated by the upper-layer model, conflict resolution is considered, a lower-layer model with the lowest total path cost as a target is established, and an improved A * algorithm is used for performing optimal path planning of multiple AGVs; the optimal path planning result generated by the lower-layer model is fed back to the upper-layer model, the upper-layer model forms a new task queue scheme according to the feedback and continues to influence the lower-layer model, the upper-layer model and the lower-layer model mutually feed back, and the optimal scheme of multi-AGV multi-task allocation is generated.

Description

Technical Field

[0001] The present invention relates to a resource scheduling method, in particular to a method for multi-AGV multi-task allocation. Background Art

[0002] With the rapid development of industrial automation, Automated Guided Vehicle (AGV), as the core component of intelligent warehousing system, undertakes key tasks such as material handling and cargo sorting. The current multi-AGV task allocation method has the following problems:

[0003] Task priority and real-time issues: In practical applications, the urgency of a task, its deadline, and its dependencies on other tasks may change at any time. Existing task allocation methods are difficult to respond and adjust in real time, resulting in high-priority tasks not being completed in a timely manner, affecting overall efficiency.

[0004] AGV status and charging issues: The battery status and charging requirements of AGVs have a great impact on task execution. Existing methods often do not fully consider the battery status and charging arrangements of AGVs, resulting in the possibility of insufficient power during the AGV's task execution, increasing the risk of task delays and system downtime.

[0005] Task allocation model and optimization problem: Existing task allocation models are usually static and cannot dynamically adapt to changes in tasks and AGV states. In addition, existing optimization algorithms often have difficulty in achieving a global optimal solution when dealing with multi-objective and multi-constrained task allocation problems.

[0006] Path planning and conflict resolution issues: Path planning and conflict resolution are key issues in multi-AGV systems. Path conflicts between AGVs can lead to traffic jams and even collision accidents. Existing methods lack effective strategies for path planning and conflict resolution, resulting in inefficient path planning and affecting the overall performance of the system. Summary of the invention

[0007] Purpose of the invention: In view of the above-mentioned prior art, a method for multi-AGV multi-task allocation is proposed, which plans the optimal conflict-free path for the AGV.

[0008] Technical solution: A method for multi-AGV multi-task allocation, comprising:

[0009] S1. Obtain real-time tasks and sort them according to dynamic priorities;

[0010] S2. According to the AGV status and task information, considering the charging decision, the upper-level model of multi-AGV multi-task allocation is established, and the APSO-NSGA-Ⅲ algorithm is used for scheduling optimization to allocate a queue of tasks to be executed for each AGV;

[0011] S3. Based on the task queue information generated by the upper model, considering conflict resolution, a lower model with the goal of minimizing the total path cost is established, and the improved A* algorithm is used to perform optimal path planning for multiple AGVs;

[0012] S4. The optimal path planning result generated by the lower-level model is fed back to the upper-level model. The upper-level model forms a new task queue plan based on the feedback and continues to influence the lower-level model. The upper and lower-level models feed back each other to generate the optimal solution for multi-AGV multi-task allocation.

[0013] Beneficial effects: (1) This method can obtain task information in real time and dynamically prioritize tasks according to their urgency, deadline, and dependencies with other tasks, ensuring that high-priority tasks can be processed in a timely manner, thereby improving the response speed and flexibility of the system.

[0014] (2) The upper-level model adopts the APSO-NSGA-Ⅲ hybrid algorithm for task allocation optimization. This algorithm combines the advantages of the APSO algorithm and the NSGA-Ⅲ algorithm, and can effectively solve the problems of slow convergence speed and poor population diversity maintenance of the NSGA-Ⅲ algorithm, thereby improving the global optimality of task allocation. The lower-level model adopts the improved A* algorithm for path planning, and considers the path conflict resolution between multiple AGVs. By optimizing the path planning strategy, it reduces path conflicts and improves the efficiency and safety of path planning.

[0015] (3) This method adopts a two-layer model architecture, where the upper model is responsible for task allocation and scheduling optimization, and the lower model is responsible for path planning and conflict resolution. The upper and lower models feed back each other to form a closed-loop optimization, so that the task allocation plan can be dynamically adjusted according to the path planning results, forming the optimal multi-AGV multi-task allocation plan, and improving the overall efficiency of task allocation and path planning. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is an overall flow chart of a method for multi-AGV multi-task allocation of the present invention;

[0017] Figure 2 It is the flow chart of APSO-NSGA-Ⅲ hybrid algorithm in the method of the present invention;

[0018] Figure 3 is the chromosome coding map in the method of the present invention;

[0019] Figure 4is a crossover operator graph in the method of the present invention;

[0020] Figure 5 is a mutation operator graph in the method of the present invention;

[0021] Figure 6 A flowchart of improving the A* algorithm for path planning in the method of the present invention;

[0022] Figure 7 This is an example of the dynamic avoidance strategy in the method of the present invention. DETAILED DESCRIPTION

[0023] The present invention will be further explained below in conjunction with the accompanying drawings.

[0024] Reference Figure 1 , a method for multi-AGV multi-task allocation, based on the interaction between task allocation and path planning, constructs a double-layer coupled dynamic scheduling task allocation model of task scheduling and path planning, constructs an upper-layer model that allocates relatively balanced task queues to each AGV, and on this basis, constructs a lower-layer model for optimal conflict-free path planning. The upper-layer model is the final optimization result. The lower-layer model gives the optimal path conflict-free driving plan based on the optimal allocation queue of the upper model, and then feeds back to the upper-layer model, thereby further optimizing the goals of the upper-layer model. The final optimal solution for multi-AGV multi-task allocation is the result of the mutual coupling calculation of the upper and lower double-layer models. The specific steps are as follows:

[0025] S1. Get real-time tasks and sort them according to dynamic priorities. When executing tasks, give priority to tasks with high urgency and close deadlines. Also, consider the dependencies between tasks. Calculate the dynamic priorities of tasks and sort them in descending order according to their priorities:

[0026] P ij t =α·E ij t +β·L ij t +γ·D ij t

[0027] Where i is the task number; j is the AGV number; t is the time when the AGV executes the task; E ij t L is the urgency of the task; ij t is the task deadline; D ij t is the task dependency; α, β, γ are weight coefficients.

[0028] S2. According to the AGV status and task information, considering the charging decision, the upper-level model of multi-AGV multi-task allocation is established, and the APSO-NSGA-Ⅲ algorithm is used for scheduling optimization to allocate a queue of tasks to be executed for each AGV;

[0029] S3. Based on the task queue information generated by the upper model, considering conflict resolution, a lower model with the goal of minimizing the total path cost is established, and the improved A* algorithm is used to perform optimal path planning for multiple AGVs;

[0030] S4. The optimal path planning result generated by the lower-level model is fed back to the upper-level model. The upper-level model forms a new task queue plan based on the feedback and continues to influence the lower-level model. The upper and lower-level models feed back each other to generate the optimal solution for multi-AGV multi-task allocation.

[0031] Reference Figure 2 Aiming at the problems of slow convergence speed and poor population diversity maintenance of the traditional NSGA-Ⅲ algorithm, a hybrid algorithm based on APSO-NSGA-Ⅲ is proposed to optimally schedule multi-AGV multi-task allocation, including the following steps:

[0032] S201. Initialize reference points: Generate uniform distribution in M-dimensional space. The reference point set Z r ={z1,z2,...,z M}, normalized to generate reference point position Among them, H1 is the number of segments on each target; k1, k2…k M is a non-negative integer, k1+k2+...+k M =H1.

[0033] S202. Initialize the population: Generate a population of N good point sets P0 = {x1, x2, ..., x N}, good point coordinates Among them, aj and bj are the upper and lower limits of the current dimension respectively; {i·r -k} is the inverse sequence of the first (D-1) prime numbers r.

[0034] S203. Evaluate the population: Calculate the fitness of each individual in the population in the objective function space, using the multi-AGV multi-task allocation upper model as the fitness function, considering the shortest task completion time and the lowest AGV energy consumption cost under the condition of relatively balanced AGV scheduling:

[0035] F=min(f1,f2,f3)

[0036] The AGV scheduling balance index f1 considers the balance of task quantity and the utilization rate of charging stations:

[0037]

[0038] Where, m is the total number of AGVs; q j Assign the number of tasks to the jth AGV; Assign an average number of tasks to each AGV; u k is the utilization rate of the kth charging station, k is the charging station number, and p is the total number of charging stations; is the average utilization rate of each charging station.

[0039] The task completion time f2 takes into account the task execution time, task waiting time, and possible charging task time:

[0040]

[0041] Decision variables:

[0042] In the formula, i is the task number, n is the total number of tasks; j is the AGV number; is the travel time to the task point; t w Waiting time for AGV; is the driving time to the charging station; t c For charging time.

[0043] The AGV energy consumption cost f3 mainly considers the driving energy consumption, including the driving energy consumption of performing load tasks, the driving energy consumption of performing no-load tasks, and the driving energy consumption of possible charging tasks:

[0044]

[0045] In the formula, is the AGV load travel distance; e load is the unit energy consumption of load driving; is the empty-load driving distance of the AGV; is the driving distance to the charging station; e unload is the unit energy consumption when driving without load.

[0046] The constraints are as follows:

[0047] Task dependency constraints: l ≥e h ,if D(l,h)=1

[0048] Battery power limit: b j ≥ b threshold

[0049] Charging station capacity constraints:

[0050] AGV speed constraints:

[0051] AGV mission constraints:

[0052] In the formula, s l is the start time of the lth task; e h is the end time of the hth task; D(l,h) = 1 means the lth task depends on the hth task; b j is the current battery charge of the jth AGV; b threshold Threshold values ​​specified for batteries; The number of AGVs charging at the kth charging station; C max is the maximum capacity of the charging station; d ij is the driving distance of the jth AGV performing the i-th task; t ij v is the travel time of the jth AGV to perform the i-th task; max is the maximum speed of the vehicle; is the number of tasks performed by the jth AGV.

[0053] S204.PMX crossover operation: Use the partial matching crossover (PMX) operator to randomly select two parent chromosomes for the task order layer. V is the length of the chromosome, and the crossover points c1, c2∈{1, 2, ..., W} are determined, c1<c2, W≤V, W is the crossover end point, and the gene fragments between the crossover points Adjust the remaining genes according to the mapping relationship to generate two offspring The AGV allocation layer is inherited from the parent.

[0054] S205.Logistic Chaotic Mapping Mutation Strategy: For the parent generation P t =(P1,P2,...,P V ) generates a uniformly distributed random number x for each variable v (v=1,2,...,V) in v ∈[0,1], use the Logistic mapping formula to generate a chaotic sequence:

[0055] x v+1 = r·x v (1-x v )

[0056] In the formula, x v is the current value of the sequence, and r is the chaos parameter.

[0057] Map the generated chaotic sequence to the index range of the task sequence, and perform mutation operations on the task sequence according to the generated mutation index to generate the child C t =(C1,C2,...,CV ). The AGV allocation layer is inherited from the parent generation.

[0058] S206. Population merging: merge the parent population P t and the offspring population C t Merge into a new candidate population R t .

[0059] S207. Fast Non-Dominated Sorting: Individual X i Dominant individual X j Conditions:

[0060]

[0061] For each target K, individual X i The target value f i K Must be less than or equal to individual X j Target value There is at least one target L, individual X i The target value f i L Must be strictly smaller than individual X j Target value

[0062] According to the dominance relationship, the candidate population R t Divided into different non-dominated fronts F1, F2, ... F K The first front F1 contains all non-dominated individuals, the second front F2 contains individuals that are dominated by the first front F1 but not dominated by other individuals, and so on.

[0063] S208.APSO Optimization: Using Fast Non-dominated Sorting to Refine R t Divided into different non-dominated levels F i , update the velocity and position of each particle:

[0064] v i (t+1)=ω·v i (t)+γ·[c1·(p i (t)-x i (t))+c2·(g(t)-x i (t))]

[0065] x i (t+1)=x i (t)+v i (t+1)

[0066]

[0067] In the formula, v irepresents speed, x i represents the position, t is the number of iterations; ω is the inertia weight; γ is a random number uniformly distributed between [0,1]; c1 and c2 are acceleration coefficients; p i (t) is the optimal position of particle i; g(t) is the global optimal position; ω max is the maximum inertia weight; ω min is the minimum inertia weight; f avg is the average value of the objective function; f worst is the worst value of the objective function; f best is the optimal value of the objective function.

[0068] Evaluate the fitness of each particle f(x i (t+1)), update the individual's best position and the global best position:

[0069]

[0070] If the optimized solution x i (t+1) dominates the original solution x i (t), then use x i (t+1) replace x i (t).

[0071] S209. Selection strategy based on reference point selection: for the i-th objective function f i Perform adaptive normalization and get:

[0072]

[0073] In the formula, a i is the intercept of the plane formed by the extreme points and the coordinate axis; f i min is the minimum value of the i-th objective function.

[0074] For each individual x i , the normalized objective function is worth the direction vector:

[0075] Calculate individual x i With reference point z M Distance between:

[0076] Assign individuals to the nearest reference point. The number of niches at each reference point is the number of individuals assigned to the reference point. Starting from the frontier of non-dominated sorting, give priority to individuals at reference points with smaller numbers of niches to construct a new population S with a population size of N. t .

[0077] S210. Iteration: Iterate until the stop condition is met, and generate the best task queue to be executed for each AGV.

[0078] Reference Figure 3 According to the complexity of multi-AGV multi-task assignment, in order to effectively coordinate the matching of tasks and AGVs and improve the scheduling efficiency and the overall performance of the system, a double-layer real number encoding mechanism based on task sequence and AGV sequence is adopted to optimize the process of m AGVs executing n scheduling tasks. The chromosome consists of two layers. The first layer uses the task number i to encode, indicating the execution sequence of each task; the second layer uses the AGV number j to encode, indicating the order in which each task is assigned to a specific AGV.

[0079] Reference Figure 4 In order to retain the good genes of the parent individuals and avoid illegal combinations to ensure the legitimacy of the offspring individuals, the partial matching crossover (PMX) operator is considered to cross the task sequences sorted by priority. The AGV sequence is inherited from the parent and no crossover is performed. Take 10 tasks as an example:

[0080] Randomly select a pair of chromosomes and determine the crossover point positions c1=4, c2=7:

[0081]

[0082] Swap the positions of the two sets of genes:

[0083]

[0084] Perform conflict detection and establish a mapping relationship based on the exchanged genes. It can be found that there are two genes 3 in Proto-child1. At this time, it is transformed into gene 4 through the mapping relationship, and so on, to ensure that there is no conflict in the offspring genes:

[0085]

[0086] The final result is:

[0087]

[0088] Reference Figure 5 , taking 10 tasks as an example:

[0089] Parent:4 10 9 3 8 1 5 7 2 6

[0090] Initialize x0 = 0.7, select r = 4, and use the Logistic mapping formula x n+1 = r·x n (1-x n)Generate chaotic sequence: x = [0.84, 0.5376, 0.99549184, 0.017625608, 0.069204283].

[0091] Since the length of the task sequence is 10, the generated chaotic sequence is mapped to the index range of the task sequence, and its integer part is taken as the index: x = [8, 5, 9, 0, 0].

[0092] According to the index, select mutation positions 9 and 6, 10 and 1, and exchange them to get the final result after mutation:

[0093] Child:6 10 9 3 8 2 5 7 1 4

[0094] Reference Figure 6 ,The path cost minimum target in the lower layer model of multi-AGV multi-task assignment is planned using the improved A* algorithm, which includes the following steps:

[0095] S301. Generate a grid map based on storage environment information;

[0096] S302. Determine the starting point of the path planning according to the task information (x s ,y s ) and the end point (x e ,y e );

[0097] S303. Initialize the OpenList and CloseList lists, add the obstacle point to the CloseList, and the starting point to the OpenList;

[0098] S304. The five-neighborhood search strategy selects feasible nodes around the parent node and adds the searched nodes to the OpenList;

[0099] S305. Introduce dynamic weighting factors and obstacle rates into the cost function to avoid obliquely crossing static obstacle vertices and improve the speed and accuracy of the search:

[0100] f(n)=g(n)+μ·(1-log(θ))h(n)

[0101]

[0102] Where f(n) is the distance from the starting point to the target point through the current point (x n ,y n ) is the total path cost; g(n) is the actual distance cost from the starting point to the current point; h(n) is the distance cost from the current point to the end point; μ is the dynamic weighting factor; θ is the obstacle rate between the starting point and the end point; l is the length of the grid map; w is the width of the grid map.

[0103] S306. Determine the node with the minimum cost function. If the node is the end point, connect the parent nodes in sequence to obtain the optimal path for the current task. Use the SmoothFloyd algorithm to optimize the generated optimal path, delete redundant nodes, reduce turns, and increase the smoothness of the path.

[0104] S307. Check whether there is a driving conflict between the planned optimal paths, determine the AGV to be avoided according to the task obstacle avoidance strategy, and perform a two-by-two cyclic detection until there is no driving conflict, and then output the conflict-free optimal path planning result.

[0105] Reference Figure 7 In the dynamic avoidance principle, different types of dynamic conflicts are identified and conflict avoidance decisions are made based on the dynamic priority of the task and the status of the AGV. When multiple AGVs are driving simultaneously in a single-lane two-way channel, there will be same-direction conflicts, reverse conflicts, and turning conflicts. For vehicles with different load states, empty vehicles are given priority for avoidance; when the load states are the same, vehicles with sufficient power and lower task dynamic priority are considered for avoidance. Empty vehicles have the ability to drive through the bottom of the shelf and can flexibly avoid; loaded vehicles can only drive along a single-lane two-way channel and have a higher priority.

[0106] When a same-direction conflict occurs, for AGVs in the negative-empty state, the empty vehicle will avoid the loaded vehicle by driving under the shelf; for AGVs in the empty-empty state, vehicles with sufficient power and low task priority will be selected to drive under the shelf to avoid it. If the power and task priority are equal, vehicles closer to the task to be executed will be considered for avoidance; for AGVs in the negative-negative state, if the avoidance cannot be completed by waiting at a feasible point in advance, choose to replan the route and find an alternative path.

[0107] When a reverse conflict occurs, for AGVs in the negative-empty state, the unloaded vehicle will pass under the shelf to avoid the loaded vehicle; for AGVs in the empty-empty state, a vehicle with sufficient power and low task priority will be selected to pass under the shelf to avoid it. If the power and task priority are equal, the vehicle closer to the task to be executed will be considered for avoidance; for AGVs in the negative-negative state, according to the priority of the task to be executed, the vehicle with low priority will be selected to slow down and avoid it, or according to the real-time traffic conditions, the front vehicle will be considered to accelerate through the conflict area.

[0108] When a steering conflict occurs, for AGVs in the negative-empty state, the unloaded vehicle will pass under the shelf to avoid the loaded vehicle; for AGVs in the empty-empty state, vehicles with sufficient power and low task priority will be selected to pass under the shelf to avoid the load. If the power and task priority are equal, vehicles closer to the task to be executed will be considered for avoidance; for AGVs in the negative-negative state, vehicles with low task priority will be selected to wait at the conflict-free point in advance, and the waiting time will be dynamically adjusted to reduce the impact on the overall efficiency of the system.

[0109] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for multi-AGV multi-task allocation, characterized in that: include: S1. Obtain real-time tasks and sort them according to dynamic priorities; S2. According to the AGV status and task information, considering the charging decision, the upper-level model of multi-AGV multi-task allocation is established, and the APSO-NSGA-Ⅲ algorithm is used for scheduling optimization to allocate a queue of tasks to be executed for each AGV; S3. Based on the task queue information generated by the upper model, considering conflict resolution, a lower model with the goal of minimizing the total path cost is established, and the improved A* algorithm is used to perform optimal path planning for multiple AGVs; S4. The optimal path planning result generated by the lower-level model is fed back to the upper-level model. The upper-level model forms a new task queue plan based on the feedback and continues to influence the lower-level model. The upper and lower-level models feed back each other to generate the optimal solution for multi-AGV multi-task allocation.

2. The method for multi-AGV multi-task allocation according to claim 1, characterized in that: In S1, when executing tasks, priority is given to tasks with high urgency and close deadlines, and the dependencies between tasks also need to be considered; the dynamic priorities of the tasks are calculated, and then they are sorted in descending order according to the priority: P ij t =α·E ij t +β·L ij t +γ·D ij t Where i is the task number; j is the AGV number; t is the time when the AGV executes the task; E ij t The urgency of the task; L ij t The deadline for the task; D ij t for task dependencies; α, β, and γ are weight coefficients.

3. The method for multi-AGV multi-task allocation according to claim 1, characterized in that: The APSO-NSGA-Ⅲ algorithm is used for scheduling optimization, including the following steps: S201. Generate H uniformly distributed reference point sets Z in M-dimensional space r ; S202. Generate a good point set P0 with a population size of N as an initial population in the M-dimensional space using the reference point set; S203. The multi-AGV multi-task allocation upper model is used as the fitness function to calculate the fitness of each individual in the population in the objective function space; S204. Using the partial matching crossover operator, for the task order layer, randomly select two parent chromosomes P1 and P2, determine the crossover points c1 and c2, adjust the chromosome genes according to the mapping relationship, and generate two daughter chromosomes C1 and C2; the AGV allocation layer is inherited from the parent generation; S205. For each variable v in the parent generation P, generate a uniformly distributed random number x v ∈[0,1], use the Logistic mapping formula to generate a chaotic sequence, map the generated chaotic sequence to the index range of the task sequence, and perform a mutation operation on the task sequence according to the generated mutation index to generate a child C t ; S206. The parent population P t and the offspring population C t Merge into a new candidate population R t ; S207. According to the dominance relationship, the candidate population R t Divided into different non-dominated frontiers F i ; The first front contains all non-dominated individuals, the second front contains individuals dominated by the first front but not dominated by other individuals, and so on; S208. Initialize the velocity v of each particle i (0) and position x i (0), initialize the individual best position p i (0) and the global optimal position g(0), evaluate the fitness of each particle f(x i (t+1)), update the best position p of the individual i (t+1) and the global optimal position g(t+1), update the velocity v of each particle i (t+1) and position x i (t+1), if the optimized solution x i (t+1) dominates the original solution x i (t), then use x i (t+1) replace x i (t); S209. Adaptively normalize the i-th objective function f i ′, calculate the normalized direction vector of the objective function Calculate individual x i With reference point z M The distance between i , assign individuals to the nearest reference point. The number of niches at each reference point is the number of individuals assigned to the reference point. Starting from the frontier of non-dominated sorting, individuals at reference points with smaller numbers of niches are given priority to construct a new population S with a population size of N. t ; S210. Iterate and execute until the stop condition is met, and generate the best task queue to be executed for each AGV.

4. The method for multi-AGV multi-task allocation according to claim 1, characterized in that: The improved A* algorithm is used to perform optimal path planning for multiple AGVs, including the following steps: S301. Generate a grid map based on storage environment information; S302. Determine the starting point of the path planning according to the task information (x s ,y s ) and the end point (x e ,y e ); S303. Initialize the OpenList and CloseList lists, add the obstacle point to the CloseList, and the starting point to the OpenList; S304. The five-neighborhood search strategy selects feasible nodes around the parent node and adds the searched nodes to the OpenList; S305. Introducing dynamic weighting factors and obstacle rates into the cost function to avoid obliquely crossing static obstacle vertices and improve search speed and accuracy; S306. Determine the node with the minimum cost function. If the node is the end point, connect the parent nodes in sequence to obtain the optimal path for the current task. Use the Smooth Floyd algorithm to optimize the generated optimal path, delete redundant nodes, reduce turns, and increase the smoothness of the path. S307. Check whether there is a driving conflict between the planned optimal paths, determine the AGV to be avoided according to the task obstacle avoidance strategy, and perform a two-by-two cyclic detection until there is no driving conflict, and then output the conflict-free optimal path planning result.

5. The method for multi-AGV multi-task allocation according to claim 4, characterized in that: In order to meet the characteristics of multi-AGV multi-task allocation, the chromosome gene encoding method of the population in NSGA-Ⅲ adopts a double-layer real number encoding mechanism based on task sequence and AGV sequence to optimize the process of m AGVs performing n scheduling tasks; the chromosome consists of two layers. The first layer is encoded with task number i, indicating the execution sequence of each task; the second layer is encoded with AGV number j, indicating the order in which each task is assigned to a specific AGV.

6. The method for multi-AGV multi-task allocation according to claim 4, characterized in that: In S203, the upper-level model of multi-AGV multi-task allocation considers that when the AGV scheduling is relatively balanced, the task completion time is the shortest and the AGV energy consumption cost is the lowest; specifically, the AGV scheduling balance index considers the balance of task quantity and the balance of charging station utilization; the task completion time considers the task execution time, task waiting time and possible charging task time; the AGV energy consumption cost mainly considers driving energy consumption, including the driving energy consumption of executing load tasks, the driving energy consumption of executing no-load tasks and the driving energy consumption of possible charging tasks.

7. The method for multi-AGV multi-task allocation according to claim 4, characterized in that: In the avoidance principle of S307, conflict avoidance decisions are made by identifying different types of dynamic conflicts and combining the dynamic priority of the task with the status of the AGV. When multiple AGVs are traveling simultaneously in a single-lane two-way channel, same-direction conflicts, reverse conflicts, and turning conflicts may occur. For vehicles with different load states, unloaded vehicles are given priority for avoidance. When the load states are the same, vehicles with sufficient power and lower dynamic priority of the task are considered for avoidance. Unloaded vehicles have the ability to travel through the bottom of the shelf and can flexibly avoid. Loaded vehicles can only travel along the single-lane two-way channel and have a higher priority.

Images