A heterogeneous AMR scheduling method and device based on a bidirectional link, and a medium

Abstract

This invention relates to the field of resource scheduling technology, and discloses a heterogeneous AMR scheduling method, apparatus, and medium based on bidirectional links. The method includes: acquiring historical environmental information and calculating the maintenance estimate of historical AMRs; acquiring target partition AMR information corresponding to emergency task information based on local scheduling links; marking the target partition AMR information in a preset global scheduling link to obtain a first marked AMR; obtaining a candidate AMR dataset based on the maintenance estimate; inputting the candidate AMR dataset into a preset comprehensive scheduling function to calculate AMR fitness; determining the target AMR scheduling instruction based on the AMR fitness, marking the target AMR in the global scheduling link to obtain a second marked AMR, and then determining the scheduling scheme; by obtaining the target AMR scheduling instruction, tasks can be directly assigned to the second marked AMR in the first marked AMR where the execution scheme has been determined, improving the rationality of the entire scheduling scheme.

Description

Technical Field

[0001] This invention relates to the field of AMR scheduling technology, and in particular to a heterogeneous AMR scheduling method, apparatus, and medium based on bidirectional links. Background Technology

[0002] AMR is an autonomous mobile robot used in industrial logistics; a "lights-out factory" is an unmanned intelligent workshop that relies on AMR autonomous mobile robots, machine vision, intelligent scheduling systems, and automated warehousing equipment to automate the entire process of raw material transfer, processing and assembly, quality inspection, and warehousing; no on-site operators are required during production periods, and production can continue with the lights off.

[0003] Existing centralized scheduling methods mostly employ a single scheduling link, processing both regular and emergency tasks within the same link. Once an emergency task is triggered, the system must globally rearrange the task sequences of all AMRs across the entire plant, resulting in high recalculation time. During this period, the entire plant's logistics scheduling is at a standstill, severely disrupting continuous production. More critically, the global rearrangement disrupts the established task sequences of many AMRs; even if an emergency task only affects a single partition, regular tasks across the entire plant will be unnecessarily affected. Furthermore, existing scheduling methods typically use only a single metric as the decision-making basis when assigning emergency tasks, such as selecting the nearest AMR or the currently idle AMR. This one-sided decision-making easily leads to secondary problems such as channel congestion and AMR deadlock collisions, further exacerbating the actual execution delay of emergency tasks. Existing AMR scheduling models all iteratively search from a completely random initial state after a task is triggered, resulting in long computation times and difficulty in meeting task requirements.

[0004] In heterogeneous AMR (Autonomous Mobile Robot) hybrid operation scenarios, heavy-duty AMRs are typically responsible for long-distance bulk material transfer, while lightweight AMRs are typically responsible for high-frequency, flexible workstation replenishment. The two types of robots differ significantly in dynamic characteristics, energy consumption levels, and execution stability. When emergency tasks are added, resource mismatch can easily occur, with larger robots performing lighter tasks and smaller robots undertaking heavier tasks, leading to decreased equipment utilization, frequent localized congestion, and overall efficiency loss.

[0005] Therefore, in order to address the problem of unreasonable AMR scheduling schemes generated due to emergency order insertions, and the problem of excessive recalculation time caused by emergency order insertions, it is necessary to propose a new method to solve the problems of unreasonable AMR scheduling schemes and excessive calculation time after emergency order insertions, improve the rationality of AMR scheduling schemes generated after emergency order insertions, and reduce the calculation time required to generate AMR scheduling schemes. Summary of the Invention

[0006] This invention provides a heterogeneous AMR scheduling method, device, and medium based on bidirectional links, which improves the rationality of the AMR scheduling scheme generated after emergency order insertion and reduces the computation time required to generate the AMR scheduling scheme.

[0007] The first aspect of this invention discloses a heterogeneous AMR scheduling method based on bidirectional links, the method comprising: Acquire historical environmental information; the historical environmental information includes historical AMR body state information and channel occupancy state information in the environment; based on the historical environmental information, obtain the maintenance estimate of the historical AMR; the maintenance estimate is used to describe the overall performance of the AMR in performing tasks; When it is determined that an emergency task information has been received, a preset local scheduling link is initiated; based on the local scheduling link, the target partition AMR information corresponding to the emergency task information is obtained; the target partition AMR information in the preset global scheduling link is marked to obtain a first marked AMR; Based on the first labeled AMR, the emergency task information, and the maintenance estimate, a candidate AMR dataset is obtained; The candidate AMR dataset is input into a preset comprehensive scheduling function to obtain the AMR fitness; the AMR fitness is used to reflect the rationality score of the generated scheduling scheme; based on the AMR fitness, the target AMR scheduling instruction is determined; based on the target AMR scheduling instruction, the target AMR in the global scheduling link is marked to obtain the second marked AMR; A scheduling scheme is determined based on the second marked AMR and the target AMR scheduling instruction.

[0008] As an optional implementation, in the first aspect of the present invention, obtaining the maintenance estimate of the historical AMR based on the historical environmental information includes: Based on the historical AMR ontology state information, the estimated resource utilization rate and the estimated task execution latency are obtained; the estimated task execution latency is used to describe the response rate of the AMR. Based on the channel occupancy status information, a conflict risk estimate is obtained; the conflict risk estimate is used to reflect the probability of channel conflicts occurring during task execution. The maintenance estimate is obtained based on the resource utilization rate estimate, the task execution delay estimate, and the conflict risk estimate.

[0009] As an optional implementation, in a first aspect of the invention, obtaining a candidate AMR dataset based on the first labeled AMR, the emergency task information, and the maintenance estimate includes: When it is determined that the task load in the emergency task information is less than or equal to the load of the first marked AMR, and the latest execution time of the AMR in the emergency task information is greater than or equal to the earliest execution time of the first marked AMR, the filtered AMR dataset is obtained. Based on the filtered AMR dataset and the maintenance estimate, the filtered AMR maintenance estimate is obtained; When it is determined that the estimated value of the screened AMR is greater than the preset screening threshold, a candidate AMR dataset is obtained; Furthermore, the comprehensive scheduling function is constructed in the following manner:

[0010] In the above formula, For the comprehensive scheduling function, Let the resource utilization rate function be... This is the task execution delay function. Let be the interval conflict risk function. The resource utilization coefficient, This represents the task execution delay coefficient. This is the interval conflict coefficient.

[0011] As an optional implementation, in the first aspect of the present invention, the step of inputting the candidate AMR dataset into a preset comprehensive scheduling function to obtain the AMR fitness includes: The population size is determined based on the candidate AMR dataset; the population size includes chromosomes corresponding to different candidate AMRs; the chromosomes corresponding to the candidate AMRs are used to represent a scheduling scheme. Based on the population size, determine the first seed size, the second seed size, and the third seed size; When it is determined that the estimated value of the candidate AMR corresponding to the candidate AMR dataset is greater than the preset elite AMR threshold, an elite AMR is obtained; based on the elite AMR and the first seed size, the first AMR chromosome is obtained. Weighted random sampling is performed on the candidate AMR estimates corresponding to the candidate AMR dataset to obtain ordinary AMRs; based on the ordinary AMRs and the second seed size, the second AMR chromosome is obtained; Random sampling is performed on the candidate AMR dataset to obtain random AMRs; based on the random AMRs and the third seed size, the third AMR chromosome is obtained; The total AMR chromosome is obtained based on the first AMR chromosome, the second AMR chromosome, and the third AMR chromosome; The total AMR chromosomes are calculated using the comprehensive scheduling function to obtain a comprehensive scheduling score; the total AMR fitness is obtained based on the comprehensive scheduling score; and the total AMR fitness is calculated using genetic iteration to obtain the AMR fitness.

[0012] As an optional implementation, in the first aspect of the present invention, the step of performing weighted random sampling on the candidate AMR estimates corresponding to the candidate AMR dataset to obtain ordinary AMRs includes: Based on the candidate AMR prediction values ​​corresponding to the candidate AMR dataset, the extraction weights corresponding to the candidate AMR dataset are obtained; all the extraction weights are normalized to obtain the candidate AMR probability corresponding to each candidate AMR in the candidate AMR dataset; the candidate AMR probabilities are sampled using roulette wheel sampling to obtain ordinary AMRs; And, obtaining the second AMR chromosome based on the ordinary AMR and the second seed size includes: Based on the second seed size, a preset time delay is randomly assigned to each of the ordinary AMRs to obtain the second sub-AMR chromosome; based on all the second sub-AMR chromosomes, the second AMR chromosome is obtained.

[0013] As an optional implementation, in the first aspect of the present invention, the step of performing genetic iteration calculation on the total AMR fitness to obtain the AMR fitness includes: When the total AMR fitness is determined to be greater than or equal to a preset fitness threshold, the retained AMR fitness is obtained; based on the retained AMR fitness, the first offspring AMR chromosome corresponding to the retained AMR fitness is obtained; Based on the total fitness of all AMRs, a variety of AMR chromosomes are obtained by roulette wheel sampling; and a mixed AMR chromosome is obtained based on the variety of AMR chromosomes and the first progeny AMR chromosome. The hybrid AMR chromosomes are cross-matched to obtain cross-matched AMR chromosome pairs; the cross-matched AMR chromosomes are used for progeny calculation, and a preset encoding correction algorithm is used to constrain them during the calculation process to obtain the second progeny AMR chromosome; the second progeny AMR chromosomes are adjusted using a preset mutation algorithm to obtain the third progeny AMR chromosome. All the third-generation AMR chromosomes are calculated using the comprehensive scheduling function to obtain an iterative comprehensive scheduling score; the iterative AMR fitness is obtained based on the iterative comprehensive scheduling score. When it is determined that the fitness of the iterative AMR meets the preset iteration requirements, the fitness of the AMR is obtained; the fitness of the AMR is the maximum fitness obtained after sorting all the fitness of the iterative AMR.

[0014] As an optional implementation, in the first aspect of the invention, the interval conflict risk function is constructed in the following manner: Based on the historical environment information, the total time window corresponding to resolving historical emergency tasks is obtained; the total time window is divided into several time slice windows according to preset time slices; based on the historical environment information, overcapacity judgment is performed on the channels in each time slice window to obtain the overcapacity AMR number; Based on the time slice window and the overcapacity AMR number, an interval conflict risk function is constructed. The specific formula for the interval conflict risk function is as follows:

[0015] In the above formula, The number of time slice windows where conflicts occur. The number of overcapacity AMRs within the time slice window where the conflict occurs. This represents the total number of time slice windows. This represents the number of overcapacity AMRs in the total time slice window.

[0016] A second aspect of this invention discloses a heterogeneous AMR scheduling device based on bidirectional links, the device comprising: The valuation acquisition module is used to acquire historical environmental information, including historical AMR body status information and channel occupancy status information in the environment; based on the historical environmental information, the maintenance estimate of the historical AMR is obtained; the maintenance estimate is used to describe the overall performance of the AMR in performing tasks. The data marking module is used to activate a preset local scheduling link when it is determined that an emergency task information has been received; based on the local scheduling link, obtain the target partition AMR information corresponding to the emergency task information; and mark the target partition AMR information in the preset global scheduling link to obtain a first marked AMR. The scheme scoring module is used to obtain a candidate AMR dataset based on the first labeled AMR, the emergency task information, and the maintenance estimate; input the candidate AMR dataset into a preset comprehensive scheduling function to obtain the AMR fitness; the AMR fitness is used to reflect the rationality score of the generated scheduling scheme. The scheme determination module is used to determine the target AMR scheduling instruction based on the AMR fitness; mark the target AMR in the global scheduling link based on the target AMR scheduling instruction to obtain the second marked AMR; and determine the scheduling scheme based on the second marked AMR and the target AMR scheduling instruction.

[0017] As an optional implementation, in a second aspect of the present invention, the specific operation method by which the valuation acquisition module obtains the maintenance estimate of the historical AMR based on the historical environmental information includes: Based on the historical AMR ontology state information, the estimated resource utilization rate and the estimated task execution latency are obtained; the estimated task execution latency is used to describe the response rate of the AMR. Based on the channel occupancy status information, a conflict risk estimate is obtained; the conflict risk estimate is used to reflect the probability of channel conflicts occurring during task execution. The maintenance estimate is obtained based on the resource utilization rate estimate, the task execution delay estimate, and the conflict risk estimate.

[0018] As an optional implementation, in a second aspect of the present invention, the specific operation method by which the scheme scoring module obtains the candidate AMR dataset based on the first labeled AMR, the emergency task information, and the maintenance estimate includes: When it is determined that the task load in the emergency task information is less than or equal to the load of the first marked AMR, and the latest execution time of the AMR in the emergency task information is greater than or equal to the earliest execution time of the first marked AMR, the filtered AMR dataset is obtained. Based on the filtered AMR dataset and the maintenance estimate, the filtered AMR maintenance estimate is obtained; When it is determined that the estimated value of the screened AMR is greater than the preset screening threshold, a candidate AMR dataset is obtained; Furthermore, the comprehensive scheduling function is constructed in the following manner:

[0019] In the above formula, For the comprehensive scheduling function, Let the resource utilization rate function be... This is the task execution delay function. Let be the interval conflict risk function. The resource utilization coefficient, This represents the task execution delay coefficient. This is the interval conflict coefficient.

[0020] As an optional implementation, in a second aspect of the present invention, the specific operation method by which the scheme scoring module inputs the candidate AMR dataset into a preset comprehensive scheduling function to obtain the AMR fitness includes: The population size is determined based on the candidate AMR dataset; the population size includes chromosomes corresponding to different candidate AMRs; the chromosomes corresponding to the candidate AMRs are used to represent a scheduling scheme. Based on the population size, determine the first seed size, the second seed size, and the third seed size; When it is determined that the estimated value of the candidate AMR corresponding to the candidate AMR dataset is greater than the preset elite AMR threshold, an elite AMR is obtained; based on the elite AMR and the first seed size, the first AMR chromosome is obtained. Weighted random sampling is performed on the candidate AMR estimates corresponding to the candidate AMR dataset to obtain ordinary AMRs; based on the ordinary AMRs and the second seed size, the second AMR chromosome is obtained; Random sampling is performed on the candidate AMR dataset to obtain random AMRs; based on the random AMRs and the third seed size, the third AMR chromosome is obtained; The total AMR chromosome is obtained based on the first AMR chromosome, the second AMR chromosome, and the third AMR chromosome; The total AMR chromosomes are calculated using the comprehensive scheduling function to obtain a comprehensive scheduling score; the total AMR fitness is obtained based on the comprehensive scheduling score; and the total AMR fitness is calculated using genetic iteration to obtain the AMR fitness.

[0021] As an optional implementation, in the second aspect of the present invention, the specific operation method of the scheme scoring module to perform weighted random sampling on the candidate AMR prediction values ​​corresponding to the candidate AMR dataset to obtain ordinary AMRs includes: Based on the candidate AMR prediction values ​​corresponding to the candidate AMR dataset, the extraction weights corresponding to the candidate AMR dataset are obtained; all the extraction weights are normalized to obtain the candidate AMR probability corresponding to each candidate AMR in the candidate AMR dataset; the candidate AMR probabilities are sampled using roulette wheel sampling to obtain ordinary AMRs; And, obtaining the second AMR chromosome based on the ordinary AMR and the second seed size includes: Based on the second seed size, a preset time delay is randomly assigned to each of the ordinary AMRs to obtain the second sub-AMR chromosome; based on all the second sub-AMR chromosomes, the second AMR chromosome is obtained.

[0022] As an optional implementation, in a second aspect of the present invention, the specific operation method by which the scheme scoring module performs genetic iteration calculation on the total AMR fitness to obtain the AMR fitness includes: When the total AMR fitness is determined to be greater than or equal to a preset fitness threshold, the retained AMR fitness is obtained; based on the retained AMR fitness, the first offspring AMR chromosome corresponding to the retained AMR fitness is obtained; Based on the total fitness of all AMRs, a variety of AMR chromosomes are obtained by roulette wheel sampling; and a mixed AMR chromosome is obtained based on the variety of AMR chromosomes and the first progeny AMR chromosome. The hybrid AMR chromosomes are cross-matched to obtain cross-matched AMR chromosome pairs; the cross-matched AMR chromosomes are used for progeny calculation, and a preset encoding correction algorithm is used to constrain them during the calculation process to obtain the second progeny AMR chromosome; the second progeny AMR chromosomes are fine-tuned using a preset mutation algorithm to obtain the third progeny AMR chromosome. All the third-generation AMR chromosomes are calculated using the comprehensive scheduling function to obtain an iterative comprehensive scheduling score; the iterative AMR fitness is obtained based on the iterative comprehensive scheduling score. When it is determined that the fitness of the iterative AMR meets the preset iteration requirements, the fitness of the AMR is obtained; the fitness of the AMR is the maximum fitness obtained after sorting all the fitness of the iterative AMR.

[0023] As an optional implementation, in a second aspect of the invention, the interval conflict risk function is constructed in the following manner: Based on the historical environment information, the total time window corresponding to resolving historical emergency tasks is obtained; the total time window is divided into several time slice windows according to preset time slices; based on the historical environment information, overcapacity judgment is performed on the channels in each time slice window to obtain the overcapacity AMR number; Based on the time slice window and the overcapacity AMR number, an interval conflict risk function is constructed. The specific formula for the interval conflict risk function is as follows:

[0024] In the above formula, The number of time slice windows where conflicts occur. The number of overcapacity AMRs within the time slice window where the conflict occurs. This represents the total number of time slice windows. This represents the number of overcapacity AMRs in the total time slice window.

[0025] A third aspect of the present invention discloses an apparatus comprising a memory and a processor, the apparatus comprising: Memory containing executable program code; A processor coupled to the memory; The processor calls the executable program code stored in the memory to execute some or all of the steps in the heterogeneous AMR scheduling method based on bidirectional links according to any of the first aspects of the present invention.

[0026] The fourth aspect of the present invention discloses a computer storage medium storing computer instructions, which, when invoked, are used to execute some or all of the steps in the heterogeneous AMR scheduling method based on bidirectional links as described in any of the first aspects of the present invention.

[0027] Compared with the prior art, the present invention has the following beneficial effects: By acquiring historical environmental information and calculating the estimated maintenance value of historical AMRs, and by collecting, calculating, and storing existing data in advance before activating local scheduling links, the pre-set local scheduling link can be activated upon receiving an emergency task message. This allows for direct access to the already calculated data, improving the computation speed of subsequent algorithms.

[0028] Based on the local scheduling link, the target partition AMR information corresponding to the emergency task information is obtained; the target partition AMR information in the preset global scheduling link is marked to obtain the first marked AMR; the AMR is marked in advance so that the global scheduling link will not assign new tasks to the AMR in the target partition. After the local scheduling link generates a new scheduling scheme, the global scheduling link will unlock the remaining AMRs; by marking in advance, the scheduling process of AMR can be made more reasonable and conflicts between different AMRs can be avoided; at the same time, locking the first marked AMR in advance can also make the subsequent calculations start directly from the first marked AMR, improving the calculation speed.

[0029] Based on the first labeled AMR, emergency task information, and maintenance estimate, a candidate AMR dataset is obtained; the candidate AMR dataset is input into a preset integrated scheduling function to calculate the AMR fitness; based on the AMR fitness, the target AMR scheduling instruction is determined; based on the target AMR scheduling instruction, the target AMR in the global scheduling link is labeled to obtain the second labeled AMR; based on the second labeled AMR and the target AMR scheduling instruction, the scheduling scheme is determined. By obtaining the target AMR scheduling instruction, the second marked AMR to be executed is then determined. When the local scheduling link is activated, the first marked AMR is marked, but at this time the first marked AMR has not been assigned a task by the global scheduling link. Therefore, after the local scheduling link determines the scheme, the task can be directly assigned to the second marked AMR in the first marked AMR whose execution scheme is determined, thereby improving the rationality of the entire scheduling scheme. Attached Figure Description

[0030] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0031] Figure 1 This is a flowchart illustrating a heterogeneous AMR scheduling method based on bidirectional links disclosed in an embodiment of the present invention. Figure 2 This is a schematic diagram of the structure of a heterogeneous AMR scheduling device based on a bidirectional link disclosed in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of a device including a memory and a processor disclosed in an embodiment of the present invention; Figure 4 This is a flowchart illustrating a heterogeneous AMR distributed collaborative scheduling method for "lights-out" factories disclosed in an embodiment of the present invention. Detailed Implementation

[0032] To enable those skilled in the art to better understand the present invention, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0033] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this invention are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, apparatus, product, or end that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or ends.

[0034] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0035] This invention discloses a heterogeneous AMR scheduling method, apparatus, and medium based on bidirectional links, which reduces the computation time required to generate AMR scheduling schemes and improves the rationality of AMR scheduling schemes. These are described in detail below.

[0036] Example 1 Please see Figure 1 , Figure 1 This is a flowchart illustrating a heterogeneous AMR scheduling method based on bidirectional links disclosed in an embodiment of the present invention. Wherein, Figure 1 The described bidirectional link-based heterogeneous AMR scheduling method can be applied to a bidirectional link-based heterogeneous AMR scheduling device, which can be integrated into a cloud server or a local server; this embodiment of the invention is not limited thereto. Figure 1 As shown, the heterogeneous AMR scheduling method based on bidirectional links can include the following operations: Step 101: Obtain historical environmental information; based on the historical environmental information, obtain the estimated maintenance value of the historical AMR.

[0037] In this embodiment of the invention, historical environmental information can be used to describe historical AMR body status information and channel occupancy status information in the environment; maintenance estimate can be used to describe the scheduleability of the AMR. Specifically, historical AMR body status information may include historical AMR rated load, current load, task queue, operating status, time points during operation, remaining power, current speed, and the current real-time location of the AMR, as well as the time required for historical AMRs to resolve historical emergency order insertion tasks; environmental channel occupancy status information may include each channel topology, channel capacity, current number of occupied channels, channel reservation information, channel length, maximum speed limit, and the historical transport routes of each AMR.

[0038] In this embodiment of the invention, the maintenance estimate of historical AMRs can be calculated from the relevant data of historical AMRs in the "lights-out" factory when the local scheduling link is in a silent state. This estimate is used to provide advance knowledge of the data, status, and performance of each AMR when an emergency task occurs. This allows for quick location of the AMRs with the best performance data in the partition where the emergency task is located when it starts, and then the most suitable AMR and its scheduling scheme can be selected through calculation. The maintenance estimate can also be used to describe a continuously updated capability profile for each AMR when there are no emergency tasks. It is not specific to any particular task but is always ready to immediately identify the most suitable AMR and the AMR with the best overall performance when an emergency task occurs. Overall performance describes whether the AMR is cost-effective to use and can reduce empty running waste; whether the AMR responds quickly and can react promptly; and whether the AMR is safe and prone to causing channel congestion. The maintenance estimate will be continuously updated as factors such as the AMR's real-time location, channel occupancy, and historical statistics are updated, ensuring that when an emergency task is actually triggered, it can provide meaningful guidance to the algorithm based on the latest factory situation.

[0039] By pre-compiling data with a small portion of resources when the local scheduling link is not activated, the AMR that can execute the task can be quickly located when an emergency task is inserted. This way, subsequent calculations do not need to calculate all AMRs, which can significantly save computing resources and improve the running speed of the calculation.

[0040] This invention proposes a bidirectional scheduling link control method. The bidirectional scheduling link can include a global scheduling link and a local scheduling link. The global scheduling link can be an existing mature scheduling system in a lights-out factory, which is always running and responsible for all routine handling tasks. Its scheduling scope covers all logical partitions and all AMRs in the entire factory. The local scheduling link controls the operation of AMRs in the logical partition to which the emergency task belongs. During operation, it is prohibited to trigger any task rescheduling of the global scheduling link, and a temporary locking mechanism is used to mark the AMRs in the logical partition to prevent instruction conflicts between the two scheduling links for the same AMR.

[0041] Step 102: When an emergency task information is received, a preset local scheduling link is started; based on the local scheduling link, the target partition AMR information corresponding to the emergency task information is obtained; the target partition AMR information in the preset global scheduling link is marked to obtain the first marked AMR.

[0042] In this embodiment of the invention, the emergency task information may include the location of the pickup point, the location of the delivery point, the weight of the materials, and the latest start time limit.

[0043] In this embodiment of the invention, the determination of emergency task information can be as follows: when an emergency order insertion task is received, the material load in the emergency order insertion task is first determined to see if the material load can be executed by the AMR of the corresponding target partition; if it can be executed, the next local scheduling link is directly started to calculate and generate a scheduling scheme.

[0044] In this embodiment of the invention, the acquisition of the first marked AMR can be achieved by, after determining the target partition AMR information corresponding to the emergency task information, marking the AMRs in the target partition that meet the conditions in the global scheduling link, so that the global scheduling link will not assign new tasks to these marked AMRs in the following time. Meeting these conditions can be achieved by first screening AMRs that meet the material load and latest start time limits, and then determining candidate AMRs by calculating the maintenance estimate of historical AMRs. By using the maintenance estimate, AMRs with significantly poor estimated performance are eliminated in advance, narrowing the search space of subsequent algorithms and making the search more focused.

[0045] Step 103: Based on the first labeled AMR, emergency task information, and maintenance estimate, obtain the candidate AMR dataset; input the candidate AMR dataset into the preset comprehensive scheduling function to obtain the AMR fitness.

[0046] In this embodiment of the invention, the candidate AMR dataset can be obtained by first using the material load and latest start time information in the emergency task information to filter the first marked AMRs; then judging the filtered AMRs by the calculated maintenance estimate, and removing the AMRs with poor scores to obtain the candidate AMR dataset; in this way, when performing subsequent algorithm calculations, it is not necessary to calculate these poorly scored AMRs, making the search range more accurate and improving the calculation speed of the algorithm; the AMR fitness can be used to reflect the score of the generated scheduling scheme.

[0047] In this embodiment of the invention, the comprehensive scheduling function can be constructed as follows: the comprehensive scheduling function can be composed of a resource utilization function, a task execution delay function, and an interval conflict risk function; wherein, the resource utilization function can be the proportion of effective delivery time to the total time spent, the task execution delay function can be the total time spent from task triggering to delivery, and the interval conflict risk function can be the risk of channel conflicts during task execution. Then, the candidate AMR dataset is transformed into an initial chromosome population in a genetic algorithm, with each chromosome representing a scheduling scheme. All initial chromosome populations are input into the comprehensive scheduling function for calculation to obtain the initial AMR fitness. The initial AMR fitness is then sorted, and the chromosomes corresponding to the higher AMR fitness are retained. The chromosomes corresponding to the remaining AMR fitness continue to generate different chromosomes according to the method of generating excellent offspring in the genetic algorithm. The offspring chromosomes and the retained chromosomes are then input into the comprehensive scheduling function for calculation. When the calculation is performed after a preset number of rounds or when the calculated AMR fitness does not change significantly, the final AMR fitness is obtained. AMR fitness can be used to reflect the rationality of the generated scheduling scheme; AMR fitness can also be the scheduling scheme itself. In order to facilitate the visualization and observation of the scheduling scheme, AMR fitness is used to reflect the rationality of the scheduling scheme.

[0048] Step 104: Determine the target AMR scheduling instruction based on the AMR fitness; mark the target AMR in the global scheduling link based on the target AMR scheduling instruction to obtain the second marked AMR; determine the scheduling scheme based on the second marked AMR and the target AMR scheduling instruction.

[0049] In this embodiment of the invention, the target AMR scheduling instruction may be as follows: after determining the fitness of an AMR, the scheme corresponding to the determined fitness of the AMR (i.e., the chromosome generated subsequently) is decoded to obtain the scheduling scheme data; the scheduling scheme data includes the number of the AMR selected for the task execution and the time delay amount most suitable for scheduling in that AMR; and a scheduling instruction for the AMR is generated based on the scheduling scheme data.

[0050] In this embodiment of the invention, the local scheduling link is activated to handle emergency task insertion. After the local scheduling link calculates the solution for handling the emergency task insertion, it marks the target AMR in the global scheduling link. This is because although an emergency task insertion is received, the AMR in the corresponding partition is still allocated according to the plan made by the global scheduling link before the emergency task insertion. However, after receiving the emergency task, the global scheduling link marks the AMR but no longer assigns tasks to the first marked AMR. This continues until the local scheduling link calculates the solution for handling the emergency task and marks the AMR corresponding to the emergency task solution in the global scheduling link, obtaining the second marked AMR. At this point, after determining the AMR to execute the emergency task, the remaining AMRs continue to be calculated and allocated tasks by the global scheduling link using the traditional method. Once the plan allocated by the global scheduling link before the emergency task insertion is completed, the local scheduling link uses the target AMR scheduling instruction to schedule the second marked AMR to execute the emergency task. This method allows for better coordination of scheduling different AMRs, improving the rationality of the scheduling scheme.

[0051] In this embodiment of the invention, the second marked AMR can be obtained by marking the target AMR in the dark factory that needs to perform an emergency order insertion task. The second marked AMR is then scheduled to perform the emergency order insertion task using target scheduling instructions. Based on the channel occupancy status information in the environment, the historical task transportation route is known, thus obtaining the transportation route for the emergency order insertion task. The time delay of the second marked AMR is also obtained through AMR fitness, thus confirming the final scheduling scheme. After confirming the final scheduling scheme, the marking of the remaining AMRs in the target partition is removed, allowing the global scheduling link to continue performing unified scheduling on the AMRs in that partition again. Through this invention, emergency order insertion tasks can be handled with minimal cost.

[0052] As can be seen, in this embodiment of the invention, historical environmental information is acquired and the estimated maintenance value of historical AMR is calculated; existing data is collected, calculated, and saved in advance before the local scheduling link is activated. When an emergency task information is received, the preset local scheduling link is activated, and the already calculated data information can be directly called, improving the calculation speed of subsequent algorithms.

[0053] Based on the local scheduling link, the target partition AMR information corresponding to the emergency task information is obtained; the target partition AMR information in the preset global scheduling link is marked to obtain the first marked AMR; the AMR is marked in advance so that the global scheduling link will not assign new tasks to the AMR in the target partition. After the local scheduling link generates a new scheduling scheme, the global scheduling link will unlock the remaining AMRs; by marking in advance, the scheduling process of AMR can be made more reasonable and conflicts between different AMRs can be avoided; at the same time, locking the first marked AMR in advance can also make the subsequent calculations start directly from the first marked AMR, improving the calculation speed.

[0054] Based on the first labeled AMR, emergency task information, and maintenance estimate, a candidate AMR dataset is obtained; the candidate AMR dataset is input into a preset integrated scheduling function to calculate the AMR fitness; based on the AMR fitness, the target AMR scheduling instruction is determined; based on the target AMR scheduling instruction, the target AMR in the global scheduling link is labeled to obtain the second labeled AMR; based on the second labeled AMR and the target AMR scheduling instruction, the scheduling scheme is determined. By obtaining the target AMR scheduling instruction, the second marked AMR to be executed is then determined. When the local scheduling link is activated, the first marked AMR is marked, but at this time the first marked AMR has not been assigned a task by the global scheduling link. Therefore, after the local scheduling link determines the scheme, the task can be directly assigned to the second marked AMR in the first marked AMR whose execution scheme is determined, thereby improving the rationality of the entire scheduling scheme.

[0055] In an optional embodiment, the maintenance estimate of the historical AMR is obtained based on historical environmental information, including: Based on historical AMR ontology status information, we obtain the estimated resource utilization rate and the estimated task execution delay. Based on the channel occupancy status information, a conflict risk estimate is obtained; Based on the estimated resource utilization rate, the estimated task execution delay, and the estimated conflict risk, the estimated maintenance rate is obtained.

[0056] In this optional embodiment, the maintenance estimate is calculated first to obtain the parameter information of each AMR. When an emergency task is interrupted, the local scheduling link can quickly mark the available AMRs. At this time, the global scheduling link continues to arrange scheduling schemes for all AMRs. However, the AMRs marked by the local scheduling link are also marked in the global scheduling link. For a short period of time afterward, the global scheduling link will not assign tasks to the marked AMRs (i.e., the first marked AMRs). After the local scheduling link completes its calculation and selects the most suitable AMR, the global scheduling link will then uniformly allocate tasks to the remaining AMRs in the target partition and all AMRs in all global partitions. Using the scheme in this optional embodiment, emergency task interruptions can be reasonably scheduled without affecting global scheduling.

[0057] In this optional embodiment, the resource utilization rate estimate is used to describe the utilization efficiency of the AMR; the task execution latency estimate is used to describe the response rate of the AMR; the conflict risk estimate is used to reflect the probability of channel conflict during task execution; and historical environmental information and historical task information can be used to calculate and save previous data in advance.

[0058] In this optional embodiment, the formula for calculating the resource utilization rate function that yields the estimated resource utilization rate can be as follows:

[0059]

[0060] In the above formula, For the first Estimated resource utilization rate of Taiwan AMR A non-zero positive integer represents one of the AMRs. This refers to the travel time when carrying cargo. This refers to the travel time when the vehicle is unloaded. For loading time, For unloading time, For queuing time, This represents the total time.

[0061] The travel time when unloaded can be the time required for the AMR to travel from its current location to the virtual pickup point, or it can be calculated using a path planning algorithm based on the path length and unloaded speed from the AMR's current coordinates to the pickup point coordinates; the travel time when loaded is calculated in the same way.

[0062] The formula for calculating the estimated task execution latency function can be as follows:

[0063] In the above formula, For the first Estimated mission execution latency of Taiwan AMR This is the estimated queuing time based on the current channel occupancy. The estimated task execution latency represents the total time it takes for an AMR to execute a typical emergency task from triggering to delivery; a lower value indicates a faster response.

[0064] The formula for calculating the interval conflict risk function of the conflict risk estimate can be as follows:

[0065] In the above formula, For the first Conflict risk estimate for Taiwan AMR The number of time slice windows where conflicts occur. The number of overcapacity AMRs within the time slice window where the conflict occurs. For the total time slice window, This represents the number of overcapacity AMRs within the total time slice window. The maintenance estimate is the predicted probability of a channel conflict occurring when this AMR performs an emergency task; a lower value indicates greater safety.

[0066] The formula for calculating the estimated value can be as follows:

[0067] In the above formula, For the first Maintenance estimate for Taiwan AMR The resource utilization coefficient, This represents the task execution delay coefficient. is the interval conflict coefficient. Wherein, ; 0.4 can be taken. 0.3 can be taken. 0.3 can be taken.

[0068] As can be seen, in this optional embodiment, the maintenance estimate can be obtained by summing the resource utilization estimate, task execution delay estimate, and conflict risk estimate calculated from historical AMR operation data and historical task information. By understanding the status information of each AMR in advance before the local scheduling link is started, this calculation does not occupy the computing resources of the global scheduling link. However, in subsequent calculations, the calculated maintenance estimate can be directly used to directly select the AMRs that meet the requirements, thereby reducing the computing burden of the local scheduling link and improving the computing speed of the local scheduling link.

[0069] In another optional embodiment, a candidate AMR dataset is obtained based on the first labeled AMR, emergency mission information, and maintenance estimate, including: When it is determined that the workload of the emergency task information is less than or equal to the workload of the first marked AMR, and the latest execution time of the AMR in the emergency task information is greater than or equal to the earliest execution time of the first marked AMR, the filtered AMR dataset is obtained. Based on the selected AMR dataset and maintenance estimate, the maintenance estimate of the selected AMR is obtained; When it is determined that the estimated value of the selected AMR is greater than the preset selection threshold, a candidate AMR dataset is obtained. Furthermore, the comprehensive scheduling function is constructed in the following way:

[0070] In the above formula, For the comprehensive scheduling function, Let the resource utilization rate function be... This is the task execution delay function. Let be the interval conflict risk function. The resource utilization coefficient, This represents the task execution delay coefficient. This is the interval conflict coefficient.

[0071] In this optional embodiment, since it is an emergency insertion task, the insertion task is not particularly heavy. Therefore, in this optional embodiment, the first labeled AMR in the target partition is filtered when local scheduling is started. First, AMRs in the target partition that meet the requirements of the emergency insertion task are filtered out. Then, based on the maintenance estimate calculated before starting local scheduling, the maintenance estimate of these filtered AMR datasets is found. Since the maintenance estimate reflects the scheduling performance of the AMR, selecting the AMRs with better performance for calculation can improve the rationality of the generated scheduling scheme, thereby improving the efficiency of completing the emergency insertion task. Since the scheduling scheme is better, the completion efficiency will be greatly improved, thus minimizing the impact on global scheduling.

[0072] In this optional embodiment, the fact that the task load in the emergency task information is less than or equal to the load of the first marked AMR, and the latest execution time of the AMR in the emergency task information is greater than or equal to the earliest execution time of the first marked AMR, can be as follows: In order to ensure that the selected AMR can complete the task content of this emergency order insertion, the hard task indicators of the AMR itself must be completed. Therefore, while the task load in the emergency task information is less than or equal to the load of the first marked AMR, the latest execution time of the AMR in the emergency task information is greater than or equal to the earliest execution time of the first marked AMR. For example, if the latest execution time of the AMR in the emergency task information is 9:00, then the earliest execution time of the first marked AMR cannot be later than 9:00. If the latest execution time of the AMR in the emergency task information is 8:50, and the earliest execution time of the first marked AMR to execute the emergency task after completing the tasks previously assigned by the global scheduling link is 9:00, then if the AMR with the earliest execution time of 9:00 is selected, the emergency order insertion task requirement cannot be completed.

[0073] In this optional embodiment, when the AMR maintenance estimate is greater than a preset screening threshold, the AMR maintenance estimate can be obtained by calculating the maintenance estimate before starting the local scheduling link; and the preset screening threshold can be to sort the AMR maintenance estimates from smallest to largest and retain only the top ones. Taiwan was selected as the final candidate AMR set. The value can be a positive integer greater than 0, such as 5, 6, 7, 8, etc. It can also be a selected value that meets certain conditions. , This refers to the specific number of AMR datasets to be filtered. This involves retaining specific candidate AMR datasets. By utilizing historical statistical estimates, AMRs with significantly poor predicted performance are eliminated in advance, narrowing the search space for subsequent genetic algorithms and making the search more focused.

[0074] In this optional embodiment, the comprehensive scheduling function may be,

[0075] In the above formula, For the comprehensive scheduling function, Let the resource utilization rate function be... This is the task execution delay function. Let be the interval conflict risk function. The resource utilization coefficient, This represents the task execution delay coefficient. This is the interval conflict coefficient; after chromosome allocation for the candidate AMR dataset, the chromosomes are input into this scheduling function to calculate the AMR fitness.

[0076] As can be seen, in this optional embodiment, a candidate AMR dataset is obtained by filtering using the first labeled AMR, emergency task information, and maintenance estimate data. This filtering ensures that the selected AMRs can complete the task content of this emergency order insertion. The reduced number of filtered AMRs narrows the search space for subsequent algorithms, making the search more focused, thereby reducing the resources required for algorithm computation and improving computational efficiency. The constructed comprehensive scheduling function is used to process the candidate AMR dataset and input it into the function for calculation, obtaining a visual comprehensive index. Through continuous iterative calculation, the rationality of the scheduling scheme is improved.

[0077] In another optional embodiment, inputting the candidate AMR dataset into a preset comprehensive scheduling function to obtain the AMR fitness may include: Determine the population size based on the candidate AMR dataset; Based on the population size, determine the first seed size, the second seed size, and the third seed size; When the estimated value of the candidate AMR corresponding to the candidate AMR dataset is greater than the preset elite AMR threshold, the elite AMR is obtained; based on the elite AMR and the first seed size, the first AMR chromosome is obtained. Weighted random sampling is performed on the candidate AMR prediction values ​​corresponding to the candidate AMR dataset to obtain ordinary AMRs; based on the ordinary AMRs and the second seed size, the second AMR chromosome is obtained. Random sampling is performed on the candidate AMR dataset to obtain random AMRs; based on the random AMRs and the third seed size, the third AMR chromosome is obtained; The total AMR chromosome is obtained based on the first AMR chromosome, the second AMR chromosome, and the third AMR chromosome; The total AMR chromosomes are calculated using a comprehensive scheduling function to obtain a comprehensive scheduling score; the total AMR fitness is obtained based on the comprehensive scheduling score; and the total AMR fitness is calculated using genetic iteration to obtain the AMR fitness.

[0078] In this optional embodiment, the AMR fitness can be calculated using an improved genetic algorithm. The genetic algorithm generates multiple schemes through selection, crossover, and mutation, and calculates visualized data based on the comprehensive scheduling function constructed in this scheme. Its core purpose is to find the optimal / approximate optimal solution among a large number of schemes, which is particularly suitable for solving the problems in this scheme.

[0079] In this optional embodiment, the population size may include chromosomes corresponding to different candidate AMRs; and the chromosomes corresponding to candidate AMRs may represent a scheduling scheme; wherein, the population size can be calculated by constructing several chromosomes for each AMR, allocating a preset time delay to an AMR to construct one chromosome; and the population size is determined in advance based on the candidate AMR dataset to determine the number of chromosomes needed for better resource allocation; this population size can be considered as all time delays, because allocating one time delay to an AMR results in one chromosome; therefore, all time delays can be considered as the population size. The population size can be determined by formula... The calculation yielded, where In the above formula Population size based on chromosome number The number of AMRs in the candidate AMR dataset; by using the formula The population size is calculated to ensure a lower limit for the population, so that even if there is only one candidate, a minimum of 8 populations are retained; and the candidate AMR data cannot be 0, so an upper limit is also controlled to prevent the calculation time from being too long. The upper limit of 20 can also be set and adjusted according to the computing resources.

[0080] The preset time delay can be set to 0.1s, 3s, or the average of historical time delays. Setting it to 0.1s means the AMR can start instantly without waiting. Setting it to 3s represents a general delay for the AMR; compared to the 0.1s option, it only adds a negligible 3-second delay, but may avoid a serious path conflict. Compared to the average traffic delay option, its delay is smaller, making it suitable for areas where the overall traffic flow is not congested, but there is a risk of instantaneous conflicts. Setting it to the average of historical time delays can provide an optimal guess based on the current real-time traffic conditions in the physical world, providing an efficient search starting point for the algorithm. The subsequent crossover and mutation tasks of the genetic algorithm are precisely to perform a fine search around this reasonable starting point of average traffic delay, to find the optimal delay value that perfectly balances risk and efficiency. In other words, the preset time delay is randomly allocated from (0.1, 3, average time delay), and the population size is the amount of time delay allocated from (0.1, 3, average time delay).

[0081] In this optional embodiment, the first seed size can be allocated to the elite AMR generation chromosome obtained after screening the candidate AMR dataset, the second seed size can be the AMR generation chromosome obtained by weighted random sampling of the candidate AMR dataset, and the third seed size can be the AMR generation chromosome obtained by purely random selection of the candidate AMR dataset.

[0082] Once the total population size for seed allocation is obtained, the seed size is divided into three parts according to a fixed ratio. The first seed size can be divided into 1 / 4 of the total population size. If the proportion is too high, it will crowd out the space for guiding randomness, resulting in insufficient population diversity. Once the prediction is inaccurate, the algorithm will have almost no ability to correct it. If the proportion is too low, the guidance will be insufficient, and the acceleration effect will not be obvious. The second seed size is selected based on the probability of the prediction. The better the prediction, the higher the probability of the AMR being selected, and all candidate AMRs have a chance. Therefore, this part accounts for 1 / 2 of the population size. The main proportion ensures that the population fully explores areas with high probability. At the same time, because the delay is completely random, it covers multiple time-series strategies, providing rich genetic material for subsequent crossover and mutation. The third seed size is a safety net mechanism to prevent the overall prediction from being biased in the wrong direction, leading to inaccurate selections in the previous stages. Through random selection, purely random individuals may accidentally select the truly optimal AMR, which will be retained and amplified in subsequent evolution. This part can be set to 1 / 4, which can play a safety net role without dragging down the overall search efficiency.

[0083] In this optional embodiment, the elite AMR threshold can be the value of the top 3 predicted values ​​of all candidate AMRs. The first AMR chromosome can be obtained by sorting the elite AMRs, and then assigning a first seed to each of the sorted elite AMRs in sequence according to the first seed size. If there are any remaining seeds after the first round of allocation, the allocation continues in sequence until the first seed size is exhausted. For example, if 3 AMRs are selected, they are encoded to obtain an encoding sequence, and the positional relationship can be [AMR1, AMR2, AMR3]. Then, seeds are assigned to the AMRs. For example, [average time delay, 0, 0] indicates that the first candidate AMR is selected, and its original task queue is delayed by the average time delay to make room for the emergency task execution window. [0, 3, 0] indicates that the second candidate AMR is selected, and its original task queue is delayed by 3 seconds to make room for the emergency task execution window. This is a first AMR chromosome. [0, 0, 0] is an invalid chromosome (no non-zero genes) and is discarded directly.

[0084] In this optional embodiment, the weighted random sampling process can be the following:

[0085]

[0086] In the above formula, The estimated value for candidate AMRs, It is a very small positive number. and The weights of candidate AMRs, The probability of being selected. The total number of candidate AMRs.

[0087] Then according to probability distribution Perform roulette wheel sampling, and each time a candidate AMR is selected, a normal AMR is obtained. A time delay (second seed) is randomly assigned to the selected normal AMR. The time delay (second seed) is generated uniformly and randomly within a preset reasonable range, and finally the second AMR chromosome is obtained. Repeat roulette wheel sampling and seed assignment until the second seed size is exhausted, and finally the second AMR chromosome is obtained.

[0088] In this optional embodiment, the acquisition of the third AMR chromosome may be achieved by randomly sampling the candidate AMR dataset to obtain random AMRs; allocating a time delay (third seed) to the selected random AMRs to obtain the third AMR chromosome; repeating random sampling and seed allocation until the second seed size is fully allocated; the time delay (third seed) is also uniformly and randomly generated within a preset reasonable range.

[0089] In this optional embodiment, the final calculation process of inputting the total AMR chromosome into the integrated scheduling function can be as follows: first, decode the total AMR chromosome, extract the selected AMR and time delay, and by obtaining historical environmental information, know the real-time location of the selected AMR and the current task queue.

[0090] By analyzing the actual pickup and delivery points, material weight, and real-time channel occupancy status and road network topology of the emergency task, the earliest available time can be calculated (based on the current task queue, the time it takes to complete all assigned tasks). Adding the earliest available time to the time delay allows us to calculate the actual start time. Using a path planning algorithm, we can plan the specific paths from the current location to the pickup point and from the pickup point to the delivery point. Knowing the specific paths allows us to determine the waiting time required in the queues. Based on the map and road conditions, we can find a route from the current AMR location to the pickup point. By planning the fastest or shortest route to the pickup point, the AMR will clearly specify which lanes it will pass through in sequence. Based on the route length and real-time traffic conditions, the empty-run time can be estimated. Similarly, by planning the route from the pickup point to the delivery point after loading, the loading time can be estimated. Emergency task insertion information reveals the loading and unloading times for that task. Knowing which lanes the AMR will pass through and when it will pass through allows for assessment of potential conflicts with other AMRs on those lanes, enabling precise calculation of conflict risks. All this data allows for the calculation of all the data in the integrated scheduling function, ultimately yielding the integrated scheduling score.

[0091] In this optional embodiment, the total AMR fitness can be calculated as follows:

[0092] In the above formula, (This is a comprehensive scheduling score.) For total AMR fitness, The selected AMR.

[0093] In this optional embodiment, the total AMR fitness is calculated using a genetic algorithm through genetic iteration to obtain the final AMR fitness. As can be seen, in this optional embodiment, the population size is first determined using the candidate AMR dataset; then, the candidate AMR dataset is screened, weighted random sampling is performed, and random selection is performed to select different AMRs; then, the first AMR chromosome, the second AMR chromosome, and the third AMR chromosome are generated; finally, the comprehensive scheduling score is calculated using the comprehensive scheduling function, and thus the AMR fitness is obtained; by using the chromosomes generated from the screened AMRs to calculate the final AMR fitness using a genetic algorithm, not only can the computational efficiency of the entire algorithm process be improved, but the rationality of the scheme can also be improved through genetic iteration.

[0094] In yet another optional embodiment, weighted random sampling is performed on the candidate AMR estimates corresponding to the candidate AMR dataset to obtain ordinary AMRs, including: Based on the candidate AMR prediction values ​​corresponding to the candidate AMR dataset, the extraction weights corresponding to the candidate AMR dataset are obtained; all extraction weights are normalized to obtain the candidate AMR probability corresponding to each candidate AMR in the candidate AMR dataset; the candidate AMR probabilities are sampled using roulette wheel sampling to obtain ordinary AMRs. And, based on the ordinary AMR and the second seed scale, the second AMR chromosome is obtained, including: Based on the second seed size, a preset time delay is randomly assigned to each ordinary AMR to obtain the second sub-AMR chromosome; based on all the second sub-AMR chromosomes, the second AMR chromosome is obtained.

[0095] In this optional embodiment, the calculation process for the extraction weights is as follows:

[0096] In the above formula, The estimated value for candidate AMRs, It is a very small positive number. The weights of the candidate AMRs.

[0097] Normalization can be performed as follows:

[0098] In the above formula, The weights of candidate AMRs, The probability of being selected. The total number of candidate AMRs.

[0099] In this optional embodiment, according to probability distribution Perform roulette wheel sampling, and each time a candidate AMR is selected, a normal AMR is obtained. A time delay (second seed) is randomly assigned to the selected normal AMR. The time delay (second seed) is generated uniformly and randomly within a preset reasonable range, and finally the second AMR chromosome is obtained. Repeat roulette wheel sampling and seed assignment until the second seed size is exhausted, and finally the second AMR chromosome is obtained.

[0100] As can be seen, in this optional embodiment, a large-scale probability search of candidate AMR data is performed by weighted random sampling, which increases the number of chromosome types, provides multiple scheduling schemes, and improves the rationality of the generation scheme.

[0101] In another optional embodiment, the total AMR fitness is calculated using genetic iteration to obtain the AMR fitness, including: When the total AMR fitness is determined to be greater than or equal to the preset fitness threshold, the retained AMR fitness is obtained; based on the retained AMR fitness, the first offspring AMR chromosome corresponding to the retained AMR fitness is obtained; Based on the total fitness of all AMRs, a variety of AMR chromosomes are obtained by roulette wheel sampling; based on the variety of AMR chromosomes and the first progeny AMR chromosomes, a mixed AMR chromosome is obtained. The hybrid AMR chromosomes are cross-matched to obtain cross-matched AMR chromosome pairs; the cross-matched AMR chromosomes are used for progeny calculation, and a preset encoding correction algorithm is used to constrain them during the calculation process to obtain the second progeny AMR chromosome; the second progeny AMR chromosomes are adjusted using a preset mutation algorithm to obtain the third progeny AMR chromosome. All third-generation AMR chromosomes are calculated using a comprehensive scheduling function to obtain an iterative comprehensive scheduling score; the iterative AMR fitness is then obtained based on the iterative comprehensive scheduling score. When it is determined that the fitness of the iterative AMR meets the preset iteration requirements, the fitness of the AMR is obtained; the fitness of the AMR is the maximum fitness obtained after sorting all the fitness of the iterative AMR.

[0102] In this optional embodiment, the fitness threshold can be determined by arranging all total AMR fitness values ​​in order and setting the fitness values ​​of the top three positions as the fitness threshold. The first progeny AMR chromosome can be selected only if the total AMR fitness value is greater than this fitness threshold, thus retaining the selected total AMR chromosomes that meet the criteria. This selected total AMR chromosome is named the first progeny AMR chromosome and used as part of subsequent crossover and mutation operations. By retaining the optimal AMR, the optimal solution is prevented from being destroyed by subsequent selection, crossover, and mutation operations. In this optional embodiment, the diverse AMR chromosomes can be selected by sampling all the total AMR fitnesss using roulette wheel sampling to select the second part of the AMR chromosomes, which is the diverse AMR chromosomes; the roulette wheel sampling can be used to randomly select AMR chromosomes for crossover and mutation from the current population according to the probability ratio of fitness; individuals with higher fitness have a greater probability of being selected, but individuals with low fitness still have a chance.

[0103] In this optional embodiment, the mixed AMR chromosome can be obtained by mixing the calculated diverse AMR chromosomes with the first progeny AMR chromosome; the mixed AMR chromosome serves as the source AMR chromosome data for subsequent crossover mutation operations.

[0104] Crossover matching of diverse AMR chromosomes can be achieved by pairing mixed AMR chromosomes together to obtain crossover AMR chromosome pairs. For crossover AMR chromosome pairs, linear crossover can be used to generate offspring genes. For example, a crossover AMR chromosome pair could be crossover chromosome 1. Cross chromosome 2 Then the offspring genes generated can be in, In the above formula It is a random number within [0,1], generated digit by digit independently or uniformly.

[0105] In this optional embodiment, after linear crossover, the offspring may have multiple non-zero genes or all-zero genes, violating the coding rules. Therefore, a coding correction algorithm is needed to correct this. The coding correction algorithm can be as follows: when there are more than one non-zero gene, randomly retain one non-zero gene and reset the remaining non-zero genes to 0; for example, in crossover chromosome 1... This indicates that the second AMR has been selected, with a 3.0-second delay and crossover chromosome 2. This indicates that the first AMR has been selected, with a 1.5-second delay. If 0.4 is randomly selected, then the offspring genes... This would result in the selection of two AMRs. However, this solution addresses the issue of choosing the most reasonable AMR scheduling scheme in emergency task insertion, which means selecting the most suitable AMR to execute the task. This involves using the shortest possible computation time to best complete the emergency task insertion, minimizing its impact on the overall scheduling. Therefore, the encoding correction algorithm could randomly retain one non-zero gene and set the other to zero. For example, randomly selecting and retaining the second gene would result in the corrected offspring being... The result was that the second AMR was selected, with a delay of 1.2 seconds.

[0106] In this optional embodiment, the mutation algorithm can fine-tune each offspring chromosome generated by crossover with a preset probability, such as 2%, 3%, 4%, 5%, etc. The mutation algorithm can be calculated in two ways: first, keeping the non-zero gene position unchanged and randomly increasing or decreasing its value by a small amount (e.g., the change range is within ±1.0s, but within a reasonable range); second, randomly moving the non-zero gene position to the corresponding position of another candidate AMR, and regenerating the delay amount randomly. Randomly moving the non-zero gene position to the corresponding position of another candidate AMR and regenerating the delay amount randomly is because if only "delay fine-tuning" is done during mutation, it's equivalent to perpetually patching up the current AMR; no matter how it's adjusted, it's still the same AMR. However, if the AMR itself is not optimal, no amount of fine-tuning will find the globally optimal solution. We first force it out of the existing AMR selection and try a completely different AMR. When we switch to a different AMR, the original latency becomes meaningless because each AMR has a different position, task queue, and travel path. A latency suitable for AMR-1 might be completely unreasonable for AMR-3. Therefore, it's better to randomly generate a new value, giving this new AMR an independent and entirely new starting point. In the early stages of the search, we usually favor the AMRs with good predictions. If the prediction is wrong, the truly optimal AMR might have a very low percentage in the population. At this point, switching the mutation operation between different AMRs has the opportunity to forcibly pull a chromosome that originally selected another AMR to this buried optimal AMR during the evolutionary process, and preserve it through the subsequent fitness-optimal retention mechanism. Selecting and retaining AMRs with high fitness makes it easier for superior genes to spread, crossover generates new latency combinations, and mutation introduces random perturbations to maintain population diversity. These three factors work together to drive the population towards the optimal solution.

[0107] In this optional embodiment, the third-generation AMR chromosome obtained through the mutation algorithm is further calculated using a comprehensive scheduling function to obtain the iterative AMR fitness. When the iterative AMR fitness meets a preset iteration requirement, the AMR fitness required by this scheme can be obtained. The preset iteration requirement can be that the fitness value of the maximum value among the iterative AMR fitness values ​​after five iterations is taken as the AMR fitness; or, if the optimal fitness value no longer increases after two consecutive iterations, the iteration terminates, and the fitness value at this point is taken as the AMR fitness.

[0108] As can be seen, in this optional embodiment, by selecting and retaining AMRs with high fitness, superior genes are more easily propagated; by generating new combinations of delay amounts through crossover, different time delay amounts are generated; and by introducing random perturbations through mutation algorithms, population diversity is maintained. Through the synergistic driving of these three factors, the generated AMR population evolves towards the optimal solution, improving the accuracy of the generated time offset and thus enhancing the rationality of the generation scheduling scheme.

[0109] In yet another alternative embodiment, the interval conflict risk function is constructed as follows: Based on historical environmental information, the total time window corresponding to resolving historical emergency tasks is obtained; the total time window is divided into several time-slice windows according to preset time slices; based on historical environmental information, overcapacity judgment is performed on the channels in each time-slice window to obtain the overcapacity AMR number; Based on the time slice window and the overcapacity AMR number, the interval conflict risk function is constructed. The specific formula for the interval conflict risk function is as follows:

[0110] In the above formula, The number of time slice windows where conflicts occur. The number of overcapacity AMRs within the time slice window where the conflict occurs. This represents the total number of time slice windows. This represents the number of overcapacity AMRs in the total time slice window.

[0111] In this optional embodiment, based on historical environmental information, the operating trajectory of the AMR can be obtained from historical environmental information, including which channels each AMR is planned to pass through in the future, at what time, the length of each channel, the maximum number of AMRs allowed to be accommodated simultaneously, and speed limits. The total time window can be the total time required to resolve all emergency insertion tasks in history. The preset time slice can be obtained by cutting continuous time into equal-length segments, each segment being called a time slice, and the time slice can be set to a fixed 0.5 seconds, 0.6 seconds, 0.7 seconds, etc., thus dividing the total time window into multiple time slice windows; for example, if the total time window is 40 seconds and the preset time slice is 0.5 seconds; then the first time slice window: Seconds, second time slice window: Seconds, ..., 40th time slice window: Second.

[0112] In this optional embodiment, the overcapacity AMR count can be determined by: in each time slot window, determining the location of each AMR; predicting, based on path planning and speed, which lane (or segment of a lane) each AMR will be in within the partition during that time slot; counting the number of AMRs on each lane; for each lane within the partition, counting how many AMRs are simultaneously on that lane during that time slot; if the number of AMRs on a lane exceeds the design capacity of that lane (e.g., a single lane has a capacity of 1, but there are 2 AMRs simultaneously), it is recorded as a conflict event; for the lane where a conflict occurs, calculating the difference between the actual number of AMRs and the lane capacity, which is the overcapacity AMR count for that lane during that time slot. If multiple lanes are overcapacity simultaneously, these overcapacity counts are accumulated.

[0113] In this optional embodiment, the number of time slices where a conflict occurs can be determined by how many time slice windows in the total time slice window experience at least one channel overcapacity event. For example, if a conflict is detected only in the 3rd and 7th time slice windows, then the number of time slices where a conflict occurs is 2.

[0114] It is evident that calculating the conflict risk estimate by using the interval conflict risk function can guide the initial population size of the algorithm away from AMRs that are predicted to cause high conflict risk, thereby accelerating the search process and improving the computation speed. Furthermore, by constructing and maintaining the conflict risk estimate, the selection of AMRs can be truncated, and AMRs with extremely high predicted conflict risk can be eliminated in advance, thus improving the rationality of the generated solution.

[0115] Example 2 Please see Figure 2 , Figure 2 This is a schematic diagram of a heterogeneous AMR scheduling device based on a bidirectional link disclosed in an embodiment of the present invention, as shown below. Figure 2As shown, the device may include: The valuation acquisition module 201 is used to acquire historical environmental information, including historical AMR body status information and channel occupancy status information in the environment. Based on the historical environmental information, the maintenance estimate of the historical AMR is obtained. The maintenance estimate is used to describe the overall performance of the AMR in performing tasks. The data marking module 202 is used to start a preset local scheduling link when it is determined that an emergency task information has been received; based on the local scheduling link, obtain the target partition AMR information corresponding to the emergency task information; and mark the target partition AMR information in the preset global scheduling link to obtain the first marked AMR. The scheme scoring module 203 is used to obtain a candidate AMR dataset based on the first labeled AMR, emergency task information, and maintenance estimate; input the candidate AMR dataset into a preset comprehensive scheduling function to obtain the AMR fitness; the AMR fitness is used to reflect the rationality score of the generated scheduling scheme. The scheme determination module 204 is used to determine the target AMR scheduling instruction based on the AMR fitness; based on the target AMR scheduling instruction, the target AMR in the global scheduling link is marked to obtain the second marked AMR; and the scheduling scheme is determined based on the second marked AMR and the target AMR scheduling instruction.

[0116] As can be seen, in this embodiment of the invention, historical environmental information is acquired and the estimated maintenance value of historical AMR is calculated; existing data is collected, calculated, and saved in advance before the local scheduling link is activated. When an emergency task information is received, the preset local scheduling link is activated, and the already calculated data information can be directly called, improving the calculation speed of subsequent algorithms.

[0117] Based on the local scheduling link, the target partition AMR information corresponding to the emergency task information is obtained; the target partition AMR information in the preset global scheduling link is marked to obtain the first marked AMR; the AMR is marked in advance so that the global scheduling link will not assign new tasks to the AMR in the target partition. After the local scheduling link generates a new scheduling scheme, the global scheduling link will unlock the remaining AMRs; by marking in advance, the scheduling process of AMR can be made more reasonable and conflicts between different AMRs can be avoided; at the same time, locking the first marked AMR in advance can also make the subsequent calculations start directly from the first marked AMR, improving the calculation speed.

[0118] Based on the first labeled AMR, emergency task information, and maintenance estimate, a candidate AMR dataset is obtained; the candidate AMR dataset is input into a preset integrated scheduling function to calculate the AMR fitness; based on the AMR fitness, the target AMR scheduling instruction is determined; based on the target AMR scheduling instruction, the target AMR in the global scheduling link is labeled to obtain the second labeled AMR; based on the second labeled AMR and the target AMR scheduling instruction, the scheduling scheme is determined. By obtaining the target AMR scheduling instruction, the second marked AMR to be executed is then determined. When the local scheduling link is activated, the first marked AMR is marked, but at this time the first marked AMR has not been assigned a task by the global scheduling link. Therefore, after the local scheduling link determines the scheme, the task can be directly assigned to the second marked AMR in the first marked AMR whose execution scheme is determined, thereby improving the rationality of the entire scheduling scheme.

[0119] Example 3 Please see Figure 3 , Figure 3 This is a schematic diagram of a device including a memory and a processor, as disclosed in an embodiment of the present invention. Figure 3 As shown, the device including memory and processor may include: Memory 301 storing executable program code; Processor 302 coupled to memory 301; The processor 302 calls the executable program code stored in the memory 301 to execute some or all of the steps in any of the heterogeneous AMR scheduling methods based on bidirectional links in Embodiment 1 of the present invention.

[0120] Example 4 This invention discloses a computer storage medium storing computer instructions. When these computer instructions are invoked, they are used to execute some or all of the steps in any of the heterogeneous AMR scheduling methods based on bidirectional links disclosed in Embodiment 1 of this invention.

[0121] Example 5 Please see Figure 4 , Figure 4 This is a flowchart illustrating a heterogeneous AMR distributed collaborative scheduling method for "lights-out" factories, as disclosed in an embodiment of the present invention. The method revolves around a three-pronged technical approach of "global optimization—interval prediction—local fast response," and is implemented according to the following steps: Step 1: Construct a heterogeneous AMR industrial handling scenario model, including three-stage task modeling, zoning scenario modeling, AMR heterogeneous attribute modeling, and time-of-use electricity price parameter organization; Step 2: Construct a joint scheduling code, representing the candidate scheduling scheme as three parts: task execution order, task partition allocation, and AMR assignment; Step 3: Establish a comprehensive objective function that integrates total completion time, comprehensive energy consumption cost, and interval conflict risk penalty term, and achieve multi-objective scheduling strategy switching through weight coefficients; Step 4: Establish a conflict prediction model driven by interval logic, extend the nominal execution time to an execution time interval, and evaluate the potential conflict risk by the degree of interval overlap; Step 5: Employ a hybrid optimization algorithm based on the improved chameleon concept, using global evolutionary optimization to generate an initial, relatively optimal schedule, and utilizing a local fast response mechanism to handle urgent task insertions. Step 6: Perform local rescheduling under dynamic perturbation, identify the set of affected tasks, and only locally update the relevant task order, AMR assignment, and time schedule; Step 7: Output the scheduling scheme, system evaluation indicators and visualization results. The visualization results include convergence curves, scheduling Gantt charts, emergency task response charts, multi-objective strategy comparison charts and comparison experiment charts.

[0122] Step 1 is implemented in the following steps: Step 1.1, Task Modeling. To make the scheduling process closely resemble the actual industrial material handling operation chain, this solution abstracts a single material handling task into three consecutive stages: the first stage is traveling empty to the pickup point, the second stage is loading or picking up the goods upon arrival, and the third stage is traveling loaded to the target workstation or delivery point.

[0123] Compared to treating tasks as a single "point-to-point transport" action, the three-stage modeling approach can more accurately reflect the differences in resource consumption, execution time, and energy consumption at different stages, while also facilitating the generation of engineering-interpretive time schedules in the subsequent scheduling results.

[0124] Step 1.2, Zonal Scene Modeling. Considering that actual factories are usually not a single open plane, but consist of multiple work areas, sections or partitions, this solution abstracts the factory into several logical partitions.

[0125] The main functions of partition modeling are twofold: It enables the system to distinguish between the hierarchical relationship between "local area execution" and "global unified coordination"; It reserves a natural interface for subsequent expansion to edge collaborative deployment or multi-workshop parallel operation.

[0126] Step 1.3, Heterogeneous AMR Modeling. The AMRs in the system are not homogeneous resources, but rather have different dynamic and operational attributes. Let the attribute vector of the i-th AMR be:

[0127] in, Indicates AMR type; Indicates the execution time fluctuation coefficient; Indicates no-load operating power; This indicates the operating power of the load.

[0128] Among them, large heavy-duty AMRs typically have greater... This means that its execution time is more significantly affected by environmental disturbances; while lightweight AMRs typically have smaller [performance / performance / etc.]. This indicates that its execution process is relatively stable. Through this heterogeneous attribute modeling, the system can reflect the differentiated division of labor logic during scheduling, where "larger vehicles are suitable for high-load tasks and smaller vehicles are suitable for flexible, high-frequency tasks."

[0129] Step 1.4: Time-of-use electricity price data modeling. To ensure that the scheduling results conform to the constraints of industrial operating costs, the system sets time-of-use electricity price parameters to calculate the operating costs in different time periods.

[0130] Step 2 is implemented in the following steps: Step 2.1, Decision Variable Design. To simultaneously describe task order, region allocation, and AMR assignment, this invention adopts a joint encoding approach, representing a candidate scheduling scheme as three parts:

[0131] in, Indicates the task sequence encoding. Indicates task partition allocation. This represents the AMR assignment result. This representation directly describes the three core decisions in the scheduling problem: "what to do first", "where to do it", and "who to do it".

[0132] Step 2.2, Encoding Significance. Compared to single-layer models that only optimize task ordering or path allocation, this ternary joint encoding method is more suitable for comprehensive scheduling solutions in heterogeneous AMR scenarios. The task execution order encoding determines the order in which tasks enter the scheduling queue; the partition allocation encoding limits the work area where the task is located; and the AMR assignment encoding determines which specific robot will execute the task.

[0133] Step 2.3, Decoding Mechanism. After candidate solutions are encoded, they must be decoded into an executable scheduling timeline. The system decoding module allocates resources sequentially according to task order and calculates the actual start and finish times for each task by combining partition constraints, AMR availability time, and task phase duration. The decoding results are used for both fitness evaluation and subsequent Gantt chart and runtime state diagram output.

[0134] Step 3 is implemented in the following steps: Step 3.1, Objective Function Design. This invention does not solely focus on the shortest completion time, but rather comprehensively considers production efficiency, operating costs, and safety risks, constructing the following comprehensive objective function:

[0135] in, Total completion time To consider the overall energy cost, This is a penalty item for the risk of inter-regional conflict. , , Let the weighting coefficients satisfy: α+β+γ=1, α,γ≥0 The significance of this objective function lies in: if only minimizing... The system may frequently activate high-power AMRs and exacerbate shared path contention; if only minimization is achieved... This could lead to an overly conservative system and delayed task completion; if ignored... If the scheduling scheme may hold true under an ideal model, it carries a high risk of conflict in actual implementation. Therefore, a comprehensive objective function can more realistically characterize the multi-objective decision-making needs in industrial settings.

[0136] Step 3.2, Completion Time Model. Assume the task... The completion time is The total completion time is then defined as:

[0137] This metric describes the total time required for the entire system to complete all tasks and is a core indicator for measuring scheduling efficiency and throughput. Since this invention is geared towards industrial scheduling scenarios, it adopts... Using average completion time for a single task as the main efficiency indicator is more in line with the overall perspective of production organization.

[0138] Step 3.3, Energy Cost Model. This invention estimates the energy consumption of the AMR during task execution based on the operational status at different stages.

[0139] Its AMR energy consumption can be expressed as:

[0140] in, Indicates the equivalent power during the loading phase. 、 、 These represent the idle driving time, loading and processing time, and loaded driving time, respectively. i represents the equipment, and j represents the task.

[0141] The total energy consumption of the system is:

[0142] in, This represents the number of AMRs.

[0143] By further combining time-of-use pricing, the comprehensive energy cost can be obtained:

[0144] in, Indicates the first Total energy consumption generated within a given electricity price period This represents the electricity price for time period q, where Q is the total number of time periods. Once this module is implemented, the system can directly incorporate energy consumption factors into the scheduling evaluation, instead of being limited to simply optimizing for the shortest time.

[0145] Step 4 is implemented in the following steps: Step 4.1, Interval Modeling Concept. Traditional scheduling algorithms typically assume that task execution time is a fixed value, meaning that once a nominal time is given, the task is assumed to start and end at a fixed time. However, in actual factories, robot operation is affected by factors such as ground conditions, local congestion, communication latency, and loading fluctuations, resulting in fluctuations in execution time. If the scheduling system ignores these fluctuations, situations may arise where there are theoretically no conflicts, but contention or collisions occur in actual operation.

[0146] To address this, this invention introduces interval logic, extending the execution time of a certain task phase of the AMR from a fixed value to an interval value. If the nominal execution time of a certain phase is T, the corresponding AMR fluctuation coefficient is... Then the actual execution interval of this stage can be represented as:

[0147] Among them, T - and T + These represent the lower and upper bounds of the execution time for this stage, respectively. In this way, the difference in uncertainty between large and small vehicles is directly encoded into the scheduling model: AMRs with higher volatility coefficients have wider execution intervals and a larger potential scope for occupying shared resources; AMRs with lower volatility coefficients have relatively narrower execution intervals and a smaller scope for risk propagation.

[0148] Step 4.2, Interval Overlap Risk Measurement. Let the potential time intervals of two AMRs occupying shared resources be:

[0149] in, This is the estimated time when the first AMR will begin to occupy shared resources. This is the estimated time when the second AMR will begin to occupy shared resources. The estimated time for the first AMR to release / end its occupation of shared resources. This is the estimated time when the second AMR will release / end its occupation of shared resources.

[0150] The system calculates the overlap length between the two:

[0151] And the total envelope length:

[0152] Based on this definition, the risk of interval overlap is:

[0153] The proportion of risk of interval overlap; The system's overall conflict risk penalty item is written as:

[0154] in, This represents a set of task pairs that have a shared resource and are in competition with each other. Let u be the time at which the shared resource begins to be occupied in the u-th time interval. This is the moment when the shared resource begins to be occupied in the u-th time interval.

[0155] Step 5 is implemented in the following steps: Step 5.1, Hybrid Optimization Algorithm Design Idea. This invention adopts a hybrid optimization algorithm that integrates evolutionary search and the chameleon swarm optimization concept. Its core objective is not to reproduce the original chameleon swarm optimization algorithm in the standard sense, but to absorb its biomimetic inspiration of "environment scanning - target locking - rapid predation" in the global scheduling optimization framework, in order to enhance the system's ability to quickly handle local disturbances, especially urgent single-task insertion.

[0156] The global phase primarily searches for a better scheduling solution through population evolution; the local phase utilizes a "tongue-tapping" fast solver to quickly select suitable candidate AMRs and complete task takeover when new tasks arrive. This design addresses two requirements: firstly, it needs to find a better overall scheduling scheme in a large combinatorial solution space; secondly, it needs to have a sufficiently fast local response speed in actual execution, rather than relying on large-scale global rearrangement for every perturbation.

[0157] Step 5.2, Initial Population and Fitness Evaluation. Let the initial population size be... Then the initial solution set is:

[0158] Among them, each individual A candidate scheduling scheme consists of three parts: task ordering, region allocation, and AMR assignment. For each generation of individuals in the population, a comprehensive objective function is used. Perform fitness calculations, i.e.:

[0159] in, ) is the comprehensive objective function. For fitness.

[0160] This fitness expression can transform the original objective function of "the smaller the better" into a fitness value of "the larger the better", which facilitates subsequent selection operations.

[0161] Step 5.3, Global Evolution Update. During the global search phase, the algorithm continuously generates a new generation of the population through operations such as selection, crossover, mutation, and elite retention. Let the... generation population Then its update process can be abstracted as:

[0162] in, This indicates that the elite retains a set of individuals. This indicates that a set of individuals is generated through cross-generation. This represents the set of individuals generated by mutation. This stage belongs to the classical evolutionary optimization idea. The focus of this invention is to combine this type of global evolutionary framework with the heterogeneous AMR joint encoding method, and to incorporate the interval conflict risk and time-of-use electricity price energy consumption into the fitness, while superimposing a chameleon-heuristic local fast-response module.

[0163] Step 5.4: Local Fast Response Based on the "Tongue-Striking" Concept. In industrial settings, restarting the global optimization process when an emergency task arrives, while theoretically potentially yielding a better solution, is often too costly. Therefore, this invention designs a local fast response operator to simulate the chameleon's rapid locking and capture mechanism after detecting a target.

[0164] Let the candidate AMR set be Then for each candidate AMR Its order cost function is defined as follows:

[0165] in, This represents the distance cost from the current state of AMR to the starting point of the new task; This indicates the waiting cost incurred if the order is received by this AMR; This indicates the cost of increased energy consumption; This indicates a new risk of interval conflict; This represents the weight of the local response cost.

[0166] The rapid takeover (AMR) for emergency missions can then be expressed as:

[0167] In the above formula, It is the conflict risk penalty value corresponding to the candidate solution.

[0168] The essence of this mechanism is not to replace global optimization, but to add an efficient local decision-making layer to the original scheduling framework, enabling the system to respond quickly to new tasks while maintaining the basic stability of the global main structure.

[0169] Step 6 is implemented in the following steps: Step 6.1, Local Rescheduling under Dynamic Disturbances. After completing global scheduling, the system enters the execution phase. Since industrial environments may experience order insertions, path congestion, or local execution deviations, the updability of scheduling results during operation must be considered. To avoid large-scale system reconfiguration after each disturbance, this invention adopts a "local impact, local repair" strategy.

[0170] Let the original scheduling scheme be When a new task is added or a local state change occurs, the system first identifies the set of tasks affected:

[0171] Then, only for The relevant task sequence, AMR assignments, and schedules are partially updated, while unaffected parts remain unchanged. Let the solution after this partial fix be... The system optimization objective can then be expressed as:

[0172] in, This indicates the degree of structural difference between the old and new scheduling schemes. This is a penalty coefficient used to limit excessive rearrangement.

[0173] The significance of this strategy lies in the fact that the system not only strives for a better new solution after local disturbances, but also hopes to maintain a high degree of continuity with the original plan in order to reduce on-site execution switching costs and system oscillations.

[0174] Step 7 is implemented in the following steps: Step 7.1, System Overall Architecture. This system is built around heterogeneous AMR collaborative scheduling tasks, adopting a modular implementation approach to organize data input, scheduling solution, risk assessment, dynamic response, and result output into a complete chain. With scheduling algorithms at its core and interpretable result output as support, the system can complete task data reading, heterogeneous AMR attribute modeling, initial schedule generation, interval conflict risk calculation, rapid response to emergency tasks, and graphical result output.

[0175] The system is divided into four parts: data layer, solution layer, evaluation layer, and presentation layer. The data layer is responsible for reading and organizing task parameters, partition parameters, AMR attribute parameters, and time-of-use electricity price parameters; the solution layer is responsible for global scheduling optimization and dynamic task response; the evaluation layer is responsible for calculating the total completion time, comprehensive energy consumption cost, and interval conflict risk corresponding to the scheduling scheme; and the presentation layer is responsible for outputting the final results in the form of convergence curves, Gantt charts, and emergency task response diagrams.

[0176] Step 7.2, Task Data Implementation. The system models each handling task as three consecutive stages. The program uses the task number as an index to record the nominal execution time of each of the three stages.

[0177] Step 7.3, AMR Attribute Data Implementation. The program uses a unified structure to record AMR attributes, enabling different types of robots to participate in calculations within the same scheduling framework.

[0178] Step 7.4: Implementing zoning and electricity price data.

[0179] Step 7.5, Global Scheduling Optimization Implementation. The global scheduling optimization module is used to generate an initial, relatively optimal scheduling scheme. During implementation, the system first constructs candidate solutions based on task order, partition allocation, and AMR assignment; then, it decodes each candidate solution to obtain the start and completion times of each task on different AMRs; next, it calculates the total completion time, comprehensive energy consumption, and conflict risk of the scheme, and performs fitness evaluation accordingly; finally, it continuously optimizes the scheduling results through population updates.

[0180] Step 7.6: Decoding and Time Scheduling. After the candidate solutions are generated through encoding, the decoding module allocates resources one by one according to the task order, and calculates the actual start and finish times of each task by combining partition constraints, AMR availability time, and task phase duration.

[0181] The decoding module not only serves the fitness calculation, but more importantly, it provides direct data support for the subsequent Gantt chart drawing. Therefore, this module is the key link between the algorithm output and the visualization results.

[0182] Step 7.7, Energy Consumption Calculation Implementation. After the scheduling scheme is determined, the system calculates the energy consumption during task execution according to the power parameters of different stages of AMR.

[0183] Once this module is implemented, the system can directly incorporate energy consumption factors into the scheduling evaluation, instead of being limited to simply optimizing for the shortest time.

[0184] Step 7.8: Implementation of the Interval Conflict Assessment Module. After the scheduling solution is obtained, the system calls the interval conflict assessment module to calculate the risk of potential conflicts on shared resources. Specifically, based on the fluctuation coefficient of the AMR, the nominal execution time is extended into an interval representation. Step 7.9, Emergency Task Insertion and Real-time Response Mechanism. The system implements a rapid emergency task response module based on global scheduling. When a new task arrives, the system does not re-initialize globally. Instead, it first identifies the set of candidate AMRs that can accept the task, then calculates the local cost of different candidate AMRs accepting the task, and finally selects the optimal AMR to complete the task takeover and updates the local scheduling results.

[0185] Step 7.10: Result Output and Visualization. The system outputs the solution results as a structured result object, and generates a convergence curve, a scheduling Gantt chart, and an emergency task response diagram based on this. The convergence curve is used to show the change of the objective function value with the number of iterations; the scheduling Gantt chart is used to show the task execution arrangement of different AMRs on the time axis; and the emergency task response diagram is used to show how the system selects an AMR and completes local takeover after an emergency task is triggered.

[0186] Beneficial effects of this plan This solution integrates task three-stage modeling, heterogeneous AMR attributes, execution time fluctuations, partitioned resources, and time-of-use electricity pricing into the scheduling framework, making the scheduling model more closely aligned with real industrial scenarios such as lights-out factories, smart warehouses, and flexible production lines.

[0187] This scheme describes the task sequence, task partitioning, and AMR assignment simultaneously through joint encoding, avoiding the problem of insufficient scheduling expression caused by optimizing only a single decision variable, and improving the completeness of heterogeneous AMR collaborative scheduling solutions.

[0188] This solution introduces a conflict prediction model driven by interval logic, which extends the nominal execution time to an execution time interval and quantifies the risk of shared resource conflicts by the degree of interval overlap, thereby realizing the transformation from passive obstacle avoidance in the execution stage to early risk suppression in the scheduling stage.

[0189] This solution employs a hybrid optimization mechanism that combines evolutionary search with an improved chameleon-like local rapid response approach. It can search for a better overall scheduling scheme in normal scenarios and quickly take over local tasks in disruptive scenarios such as emergency order insertion, reducing the latency caused by global recalculation.

[0190] This solution supports switching between three scheduling strategies: balanced mode, low-cost mode, and high-speed mode, enabling the system to flexibly adjust its optimization direction based on production cycle, energy consumption cost, and safety risk requirements.

[0191] The device embodiments described above are merely illustrative. The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0192] Through the detailed description of the above embodiments, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, including read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically-Erasable Programmable Read-Only Memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, disk storage, magnetic tape storage, or any other computer-readable medium that can be used to carry or store data.

[0193] Finally, it should be noted that the above embodiments are merely preferred embodiments of the present invention and are only used to illustrate the technical solutions of the present invention, not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A heterogeneous AMR scheduling method based on bidirectional links, characterized in that, The method includes: Acquire historical environmental information; the historical environmental information includes historical AMR body state information and channel occupancy state information in the environment; based on the historical environmental information, obtain the maintenance estimate of the historical AMR; the maintenance estimate is used to describe the overall performance of the AMR in performing tasks; When an emergency task information is received, a preset local scheduling link is initiated; based on the local scheduling link, the target partition AMR information corresponding to the emergency task information is obtained; the target partition AMR information in the preset global scheduling link is marked to obtain a first marked AMR; Based on the first labeled AMR, the emergency task information, and the maintenance estimate, a candidate AMR dataset is obtained; The candidate AMR dataset is input into a preset comprehensive scheduling function to obtain the AMR fitness; the AMR fitness is used to reflect the rationality score of the generated scheduling scheme; based on the AMR fitness, the target AMR scheduling instruction is determined; based on the target AMR scheduling instruction, the target AMR in the global scheduling link is marked to obtain the second marked AMR; A scheduling scheme is determined based on the second marked AMR and the target AMR scheduling instruction.

2. The heterogeneous AMR scheduling method based on bidirectional links according to claim 1, characterized in that, The step of obtaining the maintenance estimate of the historical AMR based on the historical environmental information includes: Based on the historical AMR ontology state information, the estimated resource utilization rate and the estimated task execution latency are obtained; the estimated task execution latency is used to describe the response rate of the AMR. Based on the channel occupancy status information, a conflict risk estimate is obtained; the conflict risk estimate is used to reflect the probability of channel conflicts occurring during task execution. The maintenance estimate is obtained based on the resource utilization rate estimate, the task execution delay estimate, and the conflict risk estimate.

3. The heterogeneous AMR scheduling method based on bidirectional links according to claim 2, characterized in that, The process of obtaining a candidate AMR dataset based on the first labeled AMR, the emergency task information, and the maintenance estimate includes: When it is determined that the task load in the emergency task information is less than or equal to the load of the first marked AMR, and the latest execution time of the AMR in the emergency task information is greater than or equal to the earliest execution time of the first marked AMR, the filtered AMR dataset is obtained. Based on the filtered AMR dataset and the maintenance estimate, the filtered AMR maintenance estimate is obtained; When it is determined that the estimated value of the screened AMR is greater than the preset screening threshold, a candidate AMR dataset is obtained; Furthermore, the comprehensive scheduling function is constructed in the following manner: In the above formula, For the comprehensive scheduling function, Let the resource utilization rate function be... This is the task execution delay function. Let be the interval conflict risk function. The resource utilization coefficient, This represents the task execution delay coefficient. This is the interval conflict coefficient.

4. The heterogeneous AMR scheduling method based on bidirectional links according to claim 3, characterized in that, The step of inputting the candidate AMR dataset into a preset comprehensive scheduling function to obtain the AMR fitness includes: The population size is determined based on the candidate AMR dataset; the population size includes chromosomes corresponding to different candidate AMRs; the chromosomes corresponding to the candidate AMRs are used to represent a scheduling scheme. Based on the population size, determine the first seed size, the second seed size, and the third seed size; When it is determined that the estimated value of the candidate AMR corresponding to the candidate AMR dataset is greater than the preset elite AMR threshold, an elite AMR is obtained; based on the elite AMR and the first seed size, the first AMR chromosome is obtained. Weighted random sampling is performed on the candidate AMR estimates corresponding to the candidate AMR dataset to obtain ordinary AMRs; based on the ordinary AMRs and the second seed size, the second AMR chromosome is obtained; Random sampling is performed on the candidate AMR dataset to obtain random AMRs; based on the random AMRs and the third seed size, the third AMR chromosome is obtained; The total AMR chromosome is obtained based on the first AMR chromosome, the second AMR chromosome, and the third AMR chromosome; The total AMR chromosomes are calculated using the comprehensive scheduling function to obtain a comprehensive scheduling score; the total AMR fitness is obtained based on the comprehensive scheduling score; and the total AMR fitness is calculated using genetic iteration to obtain the AMR fitness.

5. The heterogeneous AMR scheduling method based on bidirectional links according to claim 4, characterized in that, The step of performing weighted random sampling on the candidate AMR prediction values ​​corresponding to the candidate AMR dataset to obtain ordinary AMRs includes: Based on the candidate AMR prediction values ​​corresponding to the candidate AMR dataset, the extraction weights corresponding to the candidate AMR dataset are obtained; all the extraction weights are normalized to obtain the candidate AMR probability corresponding to each candidate AMR in the candidate AMR dataset; the candidate AMR probabilities are sampled using roulette wheel sampling to obtain ordinary AMRs; And, obtaining the second AMR chromosome based on the ordinary AMR and the second seed size includes: Based on the second seed size, a preset time delay is randomly assigned to each of the ordinary AMRs to obtain the second sub-AMR chromosome; based on all the second sub-AMR chromosomes, the second AMR chromosome is obtained.

6. The heterogeneous AMR scheduling method based on bidirectional links according to claim 5, characterized in that, The step of performing genetic iteration calculations on the total AMR fitness to obtain the AMR fitness includes: When the total AMR fitness is determined to be greater than or equal to a preset fitness threshold, the retained AMR fitness is obtained; based on the retained AMR fitness, the first offspring AMR chromosome corresponding to the retained AMR fitness is obtained; Based on the total fitness of all AMRs, a variety of AMR chromosomes are obtained by roulette wheel sampling; and a mixed AMR chromosome is obtained based on the variety of AMR chromosomes and the first progeny AMR chromosome. The hybrid AMR chromosomes are cross-matched to obtain cross-matched AMR chromosome pairs; the cross-matched AMR chromosomes are used for progeny calculation, and a preset encoding correction algorithm is used to constrain them during the calculation process to obtain the second progeny AMR chromosome; the second progeny AMR chromosomes are adjusted using a preset mutation algorithm to obtain the third progeny AMR chromosome. All the third-generation AMR chromosomes are calculated using the comprehensive scheduling function to obtain an iterative comprehensive scheduling score; the iterative AMR fitness is obtained based on the iterative comprehensive scheduling score. When it is determined that the fitness of the iterative AMR meets the preset iteration requirements, the fitness of the AMR is obtained; the fitness of the AMR is the maximum fitness obtained after sorting all the fitness of the iterative AMR.

7. The heterogeneous AMR scheduling method based on bidirectional links according to claim 6, characterized in that, The interval conflict risk function is constructed in the following way: Based on the historical environment information, the total time window corresponding to resolving historical emergency tasks is obtained; the total time window is divided into several time slice windows according to preset time slices; based on the historical environment information, overcapacity judgment is performed on the channels in each time slice window to obtain the overcapacity AMR number; Based on the time slice window and the overcapacity AMR number, an interval conflict risk function is constructed. The specific formula for the interval conflict risk function is as follows: In the above formula, The number of time slice windows where conflicts occur. The number of overcapacity AMRs within the time slice window where the conflict occurs. This represents the total number of time slice windows. This represents the number of overcapacity AMRs in the total time slice window.

8. A heterogeneous AMR scheduling device based on bidirectional links, characterized in that, The device includes: The valuation acquisition module is used to acquire historical environmental information, including historical AMR body status information and channel occupancy status information in the environment; based on the historical environmental information, the maintenance estimate of the historical AMR is obtained; the maintenance estimate is used to describe the overall performance of the AMR in performing tasks. The data marking module is used to activate a preset local scheduling link when it is determined that an emergency task information has been received; based on the local scheduling link, obtain the target partition AMR information corresponding to the emergency task information; and mark the target partition AMR information in the preset global scheduling link to obtain a first marked AMR. The scheme scoring module is used to obtain a candidate AMR dataset based on the first labeled AMR, the emergency task information, and the maintenance estimate; input the candidate AMR dataset into a preset comprehensive scheduling function to obtain the AMR fitness; the AMR fitness is used to reflect the rationality score of the generated scheduling scheme. The scheme determination module is used to determine the target AMR scheduling instruction based on the AMR fitness; mark the target AMR in the global scheduling link based on the target AMR scheduling instruction to obtain the second marked AMR; and determine the scheduling scheme based on the second marked AMR and the target AMR scheduling instruction.

9. An apparatus comprising a memory and a processor, characterized in that, The device includes: Memory containing executable program code; A processor coupled to the memory; The processor calls the executable program code stored in the memory to execute the heterogeneous AMR scheduling method based on bidirectional links as described in any one of claims 1-7.

10. A computer storage medium, characterized in that, The computer storage medium stores computer instructions, which, when invoked by the processor, are used to execute the heterogeneous AMR scheduling method based on bidirectional links as described in any one of claims 1-7.

Images