An AMR scheduling method based on improved particle swarm optimization and rolling window

Abstract

The present invention relates to an AMR scheduling method based on an improved particle swarm algorithm and a rolling window, belonging to the technical field of autonomous mobile robot scheduling in logistics. In this method, in a logistics system consisting of an AMR intelligent scheduling center communicating with multiple AMRs via a manager, the AMR intelligent scheduling center selects and outputs the optimal scheduling solution by initializing a population, determining individual fitness values, selecting optimal group and individual values, iteratively determining and outputting the results, and then updating the population. This process effectively prevents the impact of dynamic process interference events on scheduling reliability and the redundancy of enterprise management systems, and expands the application scope and sharing capabilities of AMR scheduling systems.

Description

Technical Field

[0001] The present invention relates to an AMR scheduling method, in particular to an AMR scheduling method based on an improved particle swarm algorithm and a rolling window, and belongs to the technical field of logistics autonomous mobile robot scheduling. Background Art

[0002] With the continuous development of emerging industries such as e-commerce, express delivery, and new energy, traditional, manually-based warehousing and logistics operations are rapidly being replaced by automated and intelligent logistics systems due to their inability to meet practical operational demands. Autonomous mobile robots (AMRs) are a crucial component of intelligent logistics systems, accommodating diverse orders, small batch sizes, short cycle times, and multiple batches. Existing AMR scheduling strategies include genetic algorithms, ant colony algorithms, particle swarm algorithms, and simulated annealing algorithms. Compared to genetic algorithms, ant colony algorithms, and simulated annealing algorithms, particle swarm algorithms not only have advantages such as independence from the problem itself, continuous value solutions, and strong applicability, but also offer fast parallel computing speed, fewer variables requiring manual adjustment, and relatively simple implementation.

[0003] A search reveals that Chinese patent application number 201910799166.7, entitled "A Scheduling and Routing Method and System for AMR Clusters," discloses a scheduling method whose main steps are "discretizing the work area, having a central server plan a global route and update traffic flow, and having the AMR autonomously avoid obstacles and follow the global route to reach the destination." Furthermore, Chinese patent application number 202110590491.X, entitled "A Distributed Scheduling System and Method for Multiple AMRs," discloses another distributed scheduling method. The former implements AMR cluster scheduling by discretizing the work area and having a central server plan a global route; the latter implements AMR distributed scheduling through UDP communication units, UWB communication units, and TCP / IP communication units. Practice has demonstrated that because these existing technologies all dynamically implement AMR scheduling, they lack reliability in complex transport processes requiring speed and efficiency. Summary of the Invention

[0004] The purpose of the present invention is to provide an AMR scheduling method based on an improved particle swarm optimization algorithm and a rolling window to address the shortcomings of the existing technology, so that AMR scheduling can be performed conveniently, quickly and efficiently, and its reliability can be effectively guaranteed.

[0005] In order to achieve the above objectives, the basic technical solutions of the AMR scheduling method based on the improved particle swarm algorithm and rolling window of the present invention are as follows:

[0006] In a logistics system where an AMR intelligent dispatching center communicates with multiple AMRs via a manager, the AMR intelligent dispatching center dispatches each AMR according to the following steps:

[0007] Step 1: Initialize the population - set the population to have s particles P k (k=1,2,…s), the position code of each particle is represented as a random number sequence X consisting of the corresponding transport task number k =[x1, x2, …, x m ], speed code is represented by the random number sequence Y consisting of the corresponding transport task number k =[y1, y2, …, y m ]; where m is the number of transport tasks; the task number corresponds to at least task information including the starting point number, the end point number, the time taken to execute the task, and the required time;

[0008] Step 2: Determine the individual fitness value - the individual fitness value is determined by the Positional encoding The calculation determines at least the number of AMRs and the distance traveled by the AMRs; the position code is determined Whether the sum of the corresponding task execution end time and the time taken to continue executing the next task exceeds the required time of the next task, if not, the corresponding task and the previous task are divided into a task set M j And make an AMR execute; if yes, add another AMR and make the above judgment again until all tasks are covered; the final task set M j The number of AMRs is determined, and the travel distance of each AMR is obtained as a particle The individual fitness value of

[0009] Step 3: Select the swarm optimal value and individual optimal value - Construct a non-inferior solution set and an individual optimal particle set. The non-inferior solution set is composed of the particle set with the least number of AMRs and the shortest AMR travel distance. Randomly select a particle from the non-inferior solution set as the swarm optimal value, and randomly select a particle from the current generation particle population and the individual optimal particle set based on the minimum number of AMRs or the shortest AMR travel distance as the individual optimal value. The individual optimal particle set is composed of the individual optimal value particle set of each generation.

[0010] Step 4: Iteration judgment output - judge whether the number of iterations reaches the set value. If so, output the individual optimal value as the optimal scheduling result of AMR; if not, proceed to the next step;

[0011] Step 5. Update the population - update the current velocity code of the particles in the current population to the new velocity code of the random exchange sequence RES, individual exchange sequence IES and group exchange sequence GES, and exchange the internal order of the position code of the current position code according to the random exchange sequence RES, individual exchange sequence IES and group exchange sequence GES of the new velocity code to update it to the new position code; return to step 2.

[0012] The above process of the present invention cleverly utilizes and improves the particle swarm algorithm to complete the static and reasonable scheduling of AMRs. First, the number of particles in the population and the initial position code and speed code of each particle are initialized and determined; then the number of AMRs used by each particle and the AMR travel distance (i.e., the total travel distance) are calculated; then a non-inferior solution set and an individual optimal particle set are constructed, and the individual optimal value and the group optimal value are selected; then the position code and speed code of the particles in the population are continuously updated and iterated a predetermined number of times; finally, the optimal AMR scheduling solution is output, which lays the foundation for the organic combination of the static process and the rolling window. As a result, the impact of interference events in the dynamic process on the scheduling reliability and the redundancy of the enterprise management system can be effectively avoided, and the application scope and sharing of the AMR scheduling system are expanded.

[0013] A further improvement of the present invention is to decompose the current dynamic transport task into m static transport tasks according to the task demand time and priority, forming a rolling window with corresponding scheduling time; when a certain scheduling time is reached, the optimal scheduling result solution for the corresponding static transport task is output for AMR scheduling; and then the rolling window is rotated until all transport tasks are completed. In this way, the AMR dynamic transport task is divided into multiple static transport tasks, and after solving the optimal scheduling solution for each static transport task, it is reasonably assigned to the AMR for execution in chronological order, and the rolling window is rotated until all AMR dynamic transport tasks are completed. This organically combines dynamic scheduling with static scheduling, effectively ensuring reliability and completing complex AMR scheduling quickly and efficiently.

[0014] A further improvement of the present invention is that in step five, the current position codes of the particles in the current population are exchanged in internal order according to the random exchange sequence RES, individual exchange sequence IES and group exchange sequence GES of the new speed code, and are retained and updated to the new position code according to the set probability.

[0015] A further improvement of the present invention is that: in step 1, X k The random number x m No repetition, Y k The random number y m Duplicates are allowed. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 The figure is a schematic diagram of the hardware structure of an embodiment of the present invention.

[0017] Figure 2 The figure is a schematic diagram of a scheduling process according to an embodiment of the present invention.

[0018] Figure 3 for Figure 2 Schematic diagram of the route for an AMR to perform a transportation task.

[0019] Figure 4 for Figure 2 Schematic diagram of one of the routes for an AMR to perform a transportation task in the embodiment.

[0020] Figure 5 for Figure 2 Schematic diagram of the second transport task route performed by an AMR in the embodiment.

[0021] Figure 6 for Figure 2 Schematic diagram of the exchange sequence generation principle for population update in an embodiment. DETAILED DESCRIPTION

[0022] The present invention will be explained in detail below with reference to the accompanying drawings and specific embodiments.

[0023] Example 1

[0024] The hardware environment of this embodiment is based on the improved particle swarm algorithm and the rolling window AMR scheduling method. Figure 1 As shown, similar to the prior art, the AMR intelligent dispatching center is connected to multiple AMRs through a manager to form a logistics system. The AMR intelligent dispatching center of this embodiment is composed of Alibaba Cloud servers and AMR intelligent systems. Figure 2 As shown, the following steps are used to implement the scheduling of each AMR through static scheduling:

[0025] Step 1: Initialize the population—set the population to s particles Pk (k = 1, 2, ... s). Considering that a larger number of particles s in the population leads to better algorithm convergence but slower computational speed, after experimenting with various values, we settled on a commonly used value of 50-1000 for particle swarm algorithms based on the performance.

[0026] The position code of each particle is represented as a random number sequence X consisting of the corresponding transport task number. k =[x1, x2, …, x m ], speed code is represented by the random number sequence Y consisting of the corresponding transport task number k =[y1, y2, ..., y m], where m is the number of transport tasks obtained by combining the rolling window to decompose the current dynamic transport tasks according to the task demand time and priority, X k The random number x in m No repetition, Y k The random number y in m Repeatable; the task number at least corresponds to the task information including the starting point number, end point number, execution time, and required time.

[0027] To facilitate understanding of the present invention, a simplified example is given: assuming that the number of transport tasks is 3 and there are 3 particles in the population, the initial position code and initial velocity code of each particle are random number sequences composed of the numbers of the three tasks in Table 1: X1=[1,2,3], X2=[3,2,1], X3=[1,3,2] and Y1=[1,2,2], Y2=[2,3,1], Y3=[2,3,3].

[0028] Table 1

[0029] Task Number Starting point number End point number Time taken to execute the task Demand time Load capacity 1 2 5 3 minutes 10:00:00 100 2 3 7 3 minutes 09:55:00 200 3 1 4 2 minutes 10:05:00 150

[0030] Step 2: Determine the individual fitness value - based on each particle Positional encoding Calculate and determine the individual fitness value consisting of at least the number of AMRs and the distance traveled by AMRs. Whether the sum of the corresponding task execution end time and the time taken to continue executing the next task exceeds the required time of the next task, if not, the corresponding task and the previous task are divided into a task set M j And make an AMR execute; if yes, add another AMR and make the above judgment again, and so on until all tasks are covered. j The number of AMRs is determined, and the travel distance of each AMR (the sum of their travel distances) is obtained as the particle The individual fitness value of .

[0031] As in the previous example: particles The initial position code of 1 is X1=[1,2,3], which is combined with Table 1. Figure 3As can be seen, Task 1, Task 2, and Task 3 are executed in order. Assuming all AMRs begin executing tasks at 09:50:00, the first AMR completes Task 1—that is, transporting the object from starting point 2 to ending point 5 (and similarly for the following tasks)—at 09:53:00. Continuing with Task 2, since it takes 2 minutes to move from the ending point 5 of Task 1 to the starting point 3 of Task 2, and Task 2 takes 3 minutes, the time to complete Task 2 is 09:58:00, which exceeds the required time of 09:55:00 for Task 2. Therefore, Task 1 is divided into a separate task set. Next, the second AMR is assigned to perform Task 2. The time to complete Task 2 is 09:53:00. Continuing to perform Task 3, it takes 4 minutes to move from the corresponding end point to the starting point of Task 3, Position 1. It takes 2 minutes to perform Task 3, so the time to complete Task 3 is 09:59:00. This does not exceed the required time of Task 3, 10:05:00, and there are no subsequent tasks. Therefore, Task 2 and Task 3 are divided into one task set. The time taken to complete each task and reach the next task is shown in Table 2. The route of the first AMR to perform Task 1 is as follows: Figure 4 As mentioned above, the travel distance is 9; the route of the second AMR to perform tasks 2 and 3 is as follows Figure 5 As shown, the travel distance is 23. In summary, it can be determined that the particle The individual fitness value of 1 is determined by the number of AMRs being 2 and the distance traveled by AMRs being 32.

[0032] Table 2

[0033] Task transfer location Time taken After completing Task 1, move to the starting point of Task 2 2 minutes After completing Task 1, move to the starting point of Task 3 3 minutes After completing Task 2, move to the starting point of Task 1 4 minutes After completing Task 2, move to the starting point of Task 3 4 minutes After completing Task 3, move to the starting point of Task 1 2 minutes After completing Task 3, move to the starting point of Task 2 1 minute

[0034] Step 3. Select the group optimal value and individual optimal value - construct a non-inferior solution set and an individual optimal particle set. The non-inferior solution set is composed of the particle set with the least number of AMRs and the shortest AMR travel distance. The individual optimal particle set is composed of the particle set with the individual optimal value of each generation; randomly select a particle from the non-inferior solution set as the group optimal value, and randomly select a particle from the current generation particle population and the individual optimal particle set based on the least number of AMRs or the shortest AMR travel distance as the individual optimal value.

[0035] As in the previous example: According to the above three tasks, the individual fitness values are calculated as follows: 1. The number of AMRs is 2, the distance traveled by AMRs is 32, and the particle 2AMR quantity is 1, AMR travel distance is 35, particle 3AMR quantity is 2, and AMR travel distance is 30; due to the particle 2 has the least number of AMRs, and particles The AMR of 3 has the shortest travel distance, so particle 3 is selected. 2 and Construct a non-inferior solution set; randomly select a particle in the non-inferior solution set, and then randomly select As the optimal value of the group; randomly select individuals based on the minimum number of AMRs and the shortest AMR travel distance, and then randomly select particles based on the minimum number of AMRs As the individual optimal value, it is added to the individual optimal particle set.

[0036] Step 4: Iteration judgment output - judge whether the number of iterations reaches the set value. If so, output the individual optimal value as the optimal scheduling result solution of AMR; if not, proceed to the next step.

[0037] As in the previous example: After 20 pre-set iterations of transport tasks 1, 2, and 3, the optimal AMR scheduling scheme for the three transport tasks is [2, 1, 3]. Only one AMR is needed to complete the three transport tasks, and the AMR has a short travel distance.

[0038] Step 5. Population update - update the current velocity code of the particles in the current population to the new velocity code of the random exchange sequence RES, individual exchange sequence IES and group exchange sequence GES, and exchange the internal order of the position code of the current position code of the particles in the current population according to the random exchange sequence RES, individual exchange sequence IES and group exchange sequence GES of the new velocity code, and retain and update it to the new position code according to the set probability (see the following formula).

[0039] According to the particle swarm optimization algorithm, the above terms are as follows: the exchange sequence refers to the process in which the current sequence of particles is converted into the target sequence by exchanging the position order of their own random numbers; the random exchange sequence RES is a randomly generated sequence; the individual exchange sequence IES is a sequence sequence obtained by comparing the individual optimal value position code and the current position code; the group exchange sequence GES is a sequence sequence obtained by comparing the group optimal value position code and the current position code.

[0040] RES

[0041] Y k = IES

[0042] GES

[0043] X1 1 = (X k ⊕RES)×θ1

[0044] X k= X1 2 = (X k 1 ⊕IES)×θ2

[0045] X1 3 = (X k 2 ⊕GES)×θ3

[0046] For example: The basic principle of generating exchange sequence is as follows: Figure 6 As shown, through its basic principle, it can be seen that particles RES of 1 can be randomly generated as [3,1,2]; the obtained individual optimal value position is encoded as X 个优 =[2,1,3], the group optimal value position is encoded as X 群优 =[1,3,2], since the particle The position code of particle is X1=[1,2,3], then The IES is [2,2,3] and the GES is [1,3,3].

[0047] The current position code is updated by RES first, and the new position code is X1 1 =[2,1,3], and retain it according to the set probability θ1; then update it according to IES to get the new position code X1 2 =[1,2,3], and retain it according to the set probability θ2; finally, update it according to GES to get the new position code X1 3 =[1,3,2], and are retained with the set probability θ3.

[0048] The values of θ1, θ2, and θ3 are all in the range of 0-1.0. Probabilistic retention means: suppose the initial number of particles in the original population 1 is 50, then first update the 50 particles according to RES to obtain a new population 2 (there are 50 particles in population 2, and the position code of each particle is the new position code obtained by updating the position code of the corresponding particle in population 1 according to RES), and then randomly select a particles from population 2 (the value of a is 50×θ1) and b particles from population 1 (the value of b is 50×(1-θ1)) according to the value of θ1 to form population 3 (the total number of particles in population 3 is still 50); population 3 is then randomly selected according to I ES is updated to obtain population 4, and according to the value of θ2, c particles from population 4 (the value of c is 50×θ2) and d particles from population 3 (the value of b is 50×(1-θ2)) are randomly selected to form population 5 (the total number of particles in population 5 is still 50); population 5 is then updated according to GES to obtain population 6, and according to the value of θ3, e particles from population 6 (the value of e is 50×θ3) and f particles from population 5 (the value of f is 50×(1-θ3)) are randomly selected to form population 7 (the total number of particles in population 7 is still 50).

[0049] In practice, the above static scheduling can be performed within the required rolling window of the AMR dynamic scheduling. This means that the dynamic scheduling process is divided into multiple scheduling moments. When a scheduling moment arrives, the current dynamic transport task is decomposed into several static transport tasks based on task demand time and priority. Static scheduling is then performed to obtain and output the optimal scheduling plan for these static transport tasks for AMR scheduling. This process is then repeated, rolling over and over again, until all transport tasks are completed.

[0050] Experiments have shown that compared with the existing technology, the significant advantages of this embodiment are: AMR scheduling based on the improved particle swarm algorithm and rolling window, by organically combining static scheduling with dynamic scheduling, is easy to implement, and can quickly and efficiently provide a reasonable and reliable AMR scheduling solution, thus laying an ideal foundation for efficient and safe AMR operation management.

Claims

1. An AMR scheduling method based on an improved particle swarm optimization algorithm and a rolling window, in a logistics system where an AMR intelligent scheduling center communicates with multiple AMRs via a manager, is characterized by: The AMR intelligent dispatching center implements the dispatching of each AMR according to the following steps: Step 1: Initialize the population - set the population to have s particles P k [k=1,2,…s], the position code of each particle is represented as a random number sequence X consisting of the corresponding transport task number. k =[x1, x2, ..., x m ], speed code is represented by the random number sequence Y consisting of the corresponding transport task number k =[y1, y2, ..., y m ]; where m is the number of transport tasks; the task number corresponds to at least task information including the starting point number, the end point number, the time taken to execute the task, and the required time; Step 2: Determine the individual fitness value - the individual fitness value is determined by the individual fitness value of each particle P k Positional encoding X k =[x1, x2, ..., x m ] The calculation is determined to be at least composed of the number of AMRs and the distance traveled by the AMRs; the position code X is determined k Whether the sum of the corresponding task execution end time and the time taken to continue executing the next task exceeds the required time of the next task, if not, the corresponding task and the previous task are divided into a task set M j And make an AMR execute; if yes, add another AMR and make the above judgment again until all tasks are covered; the final task set M j The number of AMRs is determined, and the travel distance of each AMR is obtained as the particle P k The individual fitness value of Step 3: Select the swarm optimal value and individual optimal value - Construct a non-inferior solution set and an individual optimal particle set. The non-inferior solution set is composed of the particle set with the least number of AMRs and the shortest AMR travel distance. Randomly select a particle from the non-inferior solution set as the swarm optimal value, and randomly select a particle from the current generation particle population and the individual optimal particle set based on the minimum number of AMRs or the shortest AMR travel distance as the individual optimal value. The individual optimal particle set is composed of the individual optimal value particle set of each generation. Step 4: Iteration judgment output - judge whether the number of iterations reaches the set value. If so, output the individual optimal value as the optimal scheduling result of AMR; if not, proceed to the next step; Step 5. Update the population - update the current velocity code of the particles in the current population to the new velocity code of the random exchange sequence RES, individual exchange sequence IES and group exchange sequence GES, and exchange the current position code to the new position code by performing internal position code exchange according to the random exchange sequence RES, individual exchange sequence IES and group exchange sequence GES of the new velocity code; Return to step 2; The current dynamic transport task is decomposed into m static transport tasks according to the task demand time and priority, forming a rolling window with corresponding scheduling time. When a certain scheduling time is reached, the optimal scheduling result plan for the corresponding static transport task is output for AMR scheduling. Then the cycle is repeated until all transport tasks are completed.

2. The AMR scheduling method based on the improved particle swarm optimization algorithm and rolling window according to claim 1 is characterized in that: In step 1, X k The random number x m No repetition, Y k The random number y m Duplicates are allowed.

3. The AMR scheduling method based on the improved particle swarm optimization algorithm and rolling window according to claim 2 is characterized in that: In step five, the current position codes of the particles in the current population are exchanged in order within the position codes according to the random exchange sequence RES, individual exchange sequence IES and group exchange sequence GES of the new velocity code, and are retained and updated to the new position code according to the set probability.

4. The AMR scheduling method based on the improved particle swarm optimization algorithm and rolling window according to claim 1 is characterized in that: The AMR intelligent dispatch center consists of Alibaba Cloud servers and AMR intelligent systems.

Images