AMR cluster dynamic wireless energy supplementing and collision avoidance cooperative scheduling method and device based on prospective congestion prediction, and storage medium

Abstract

The application discloses an AMR cluster dynamic wireless energy supplementing and collision avoidance cooperative scheduling method and device based on prospective congestion prediction and a storage medium, and belongs to the technical field of AMR cluster intelligent scheduling. The technical scheme comprises the following steps: obtaining a candidate scheduling basic scheme of an AMR cluster, the basic scheme comprising task allocation results and task execution sequences of each AMR, and meeting the constraints of unique task allocation, process predecessors, power safety, wireless energy supplementing of fixed charging stations and wireless short-time energy supplementing of work points; performing prospective congestion prediction on the candidate solution of the basic scheme to obtain a nominal scheduling scheme; constructing a discrete space-time execution timeline of each AMR based on the nominal scheduling scheme, performing space-time conflict detection on the execution timeline, and finally outputting a final cooperative scheduling scheme, which comprises task allocation results, task execution sequences, wireless energy supplementing schemes of fixed charging stations, wireless short-time energy supplementing schemes of work points and a collision-free execution timeline of each AMR. The core contribution of the application lies in that, through the prospective congestion prediction and dynamic conflict correction mechanism, the whole-chain cooperative optimization from task allocation to energy supplementing decision and collision-free execution is realized.

Description

Technical Field

[0001] This application relates to the field of intelligent logistics and robot scheduling technology, specifically to a method, device and storage medium for dynamic wireless power replenishment and collision avoidance collaborative scheduling of AMR clusters based on forward-looking congestion prediction. Background Technology

[0002] With the development of intelligent manufacturing and intelligent logistics technologies, autonomous mobile robots (AMRs) have been widely used in material handling, inter-process delivery, and warehousing transfer in production workshops. By replacing manual labor in high-frequency, repetitive transportation tasks, they significantly improve production efficiency and automation levels. In a smart factory environment, AMRs need to frequently travel between areas such as garages, raw material warehouses, workstations, finished product warehouses, and fixed charging stations. Their operation involves multiple stages, including task allocation, path planning, time coordination, and energy management. In existing technologies, AMR scheduling methods often treat task scheduling, path optimization, and energy replenishment as independent problems: one type of method focuses on task allocation and path minimization for multiple AMRs to shorten the maximum completion time or total travel distance; another type of method focuses on a centralized return-to-base charging strategy when the battery is low, triggering fixed charging stations to replenish energy by setting a battery threshold to ensure continuous equipment operation. Although some studies have introduced wireless charging technology and deployed fast wireless charging devices at fixed charging stations, they have not yet broken through the framework of the separation of "task scheduling - charging decision - collision avoidance". They usually only passively handle charging needs after the task planning is completed, or ignore charging constraints only in the path planning stage, resulting in the timing of charging being out of sync with the task sequence.

[0003] Furthermore, traditional AMR systems generally rely on a single fixed charging station for centralized recharging. The AMR must interrupt its current operation before its battery runs out, detour to a dedicated charging point for recharging, and then return to its original mission path. This mode not only increases empty driving distance and downtime, exacerbating congestion at fixed charging stations, but also disrupts the process chain due to mission interruptions, significantly reducing the overall operational efficiency of the cluster. Although wireless charging technology has gradually extended to workstation loading and unloading points in recent years, supporting short-term low-speed recharging during AMR loading or unloading stops, existing scheduling methods have not yet established a collaborative decision-making mechanism between rapid recharging at fixed charging stations and short-term recharging at work points. They cannot dynamically weigh recharging methods based on mission sequence, remaining battery power, stop duration, and subsequent mission energy consumption, resulting in the untapped potential of recharging at work points and persistently high reliance on fixed charging stations.

[0004] More importantly, existing scheduling schemes generally only generate nominal task sequences and path plans, failing to fully consider the actual operational conflicts of multiple AMRs in shared lanes, intersections, workstations, and charging station areas. Most methods only detect static conflicts at nodes or edges, ignoring dynamic execution-level issues such as following vehicle conflicts, long-term blockages at target points, and repeated congestion in local hotspots. When the task scale increases or the number of AMRs increases, although the nominal scheduling scheme may be "feasible" at the mathematical model level, the lack of prediction of spatiotemporal competition at the execution level leads to a large number of AMRs being forced to wait in place for extended periods due to competing for the same node or path, even forming cascading blockages, making it difficult to directly deploy the scheduling scheme in the field. Although some systems introduce fixed priority avoidance mechanisms during the execution phase, they lack dynamic assessment of the cost of conflict impact, easily resulting in phenomena such as repeated yielding by a single AMR, task backlog, and continuous congestion in local areas, ultimately causing the actual completion time to far exceed the theoretical optimization value and a decrease in system utilization.

[0005] In summary, current AMR scheduling technologies generally suffer from poor coordination among task scheduling, wireless power replenishment planning, and collision-free execution; a disconnect between the model and execution; and low on-site executability. As a result, they are difficult to achieve efficient, safe, and stable multi-AMR cluster collaborative operations in complex and dynamic smart factory grid environments.

[0006] The applicant's previously published patent applications CN120601566A and CN122247043A mainly address the integrated modeling problem of task allocation, job sequencing, and wireless recharging decision-making under constraints of battery capacity and task sequence. Their outputs represent tasks and recharging schemes at the nominal scheduling level. This application does not simply add collision avoidance processing after the aforementioned nominal scheduling results. Instead, it transforms potential spatiotemporal competition at the execution layer into evaluation factors for candidate scheduling schemes, allowing the predicted conflict results to directly participate in the selection of task allocation and task execution sequence. Subsequently, using the selected nominal scheduling scheme as input, a collision-free execution timeline is generated through execution layer spatiotemporal occupancy detection and dynamic yielding correction. Therefore, this application solves the problem of inconsistency between nominal scheduling results and actual execution results in multi-vehicle shared spaces, forming a closed-loop relationship where execution risk influences scheduling decisions in reverse. Summary of the Invention

[0007] To address the aforementioned deficiencies in existing technologies, this invention provides a collaborative scheduling method for dynamic wireless power replenishment and collision avoidance of AMR trunking based on forward-looking congestion prediction, comprising: S1. Obtain candidate scheduling basic schemes for the AMR cluster. The basic schemes include the task allocation results and task execution order of each AMR, and satisfy the constraints of unique task allocation, process precursor, power safety, and wireless charging at fixed charging stations and short-term wireless charging at work points. S2. Based on the task allocation results and task execution order in the candidate solutions, a simplified discrete timeline is generated. Under the spatiotemporal occupancy relationship of grid nodes, path edges, and discrete time steps, the number of five types of spatiotemporal conflicts is estimated: node conflict, edge conflict, following vehicle conflict, target point blocking conflict, and repeated hotspot conflict. A congestion penalty term is formed according to the preset weights corresponding to the five types of spatiotemporal conflicts, and the congestion penalty term and the nominal maximum completion time are included in the comprehensive score of the candidate solutions. A nominal scheduling scheme is formed based on the comprehensive score. S3. Based on the nominal scheduling scheme, construct the discrete spatiotemporal execution timeline of each AMR and perform spatiotemporal conflict detection. When a spatiotemporal conflict is detected, select the AMR that needs to be yielded according to the conflict type, the current task execution status of the AMR, and the accumulated avoidance waiting cost. Insert an avoidance waiting event at the anchor point before the conflict occurs, or temporarily migrate the AMR to a preset safe waiting point. After each correction, recalculate the node occupancy, path edge occupancy, parking occupancy, and subsequent event times of the affected AMR, and perform spatiotemporal conflict detection again until a final execution timeline that meets the requirements of no node conflict, no edge conflict, and safe interval is formed. The inserted avoidance event and the time delay it causes are retained in the final execution timeline. S4. Output the final collaborative scheduling scheme, which includes the task allocation results of each AMR, the task execution order, the fixed charging station wireless power replenishment scheme, the work point wireless short-term power replenishment scheme, and the collision-free execution timeline.

[0008] Optionally, in one possible implementation of the technical solution of the present invention, in step S2, the estimation method of the five types of spatiotemporal conflicts includes: if the same grid node is occupied by two or more AMRs at the same time, it is counted as a node conflict; if the same path edge is occupied by two or more AMRs at the same time, or two AMRs pass towards each other along the same path edge, it is counted as an edge conflict; if the following AMR and the preceding AMR enter the same node or the same road segment in adjacent time steps and do not meet the preset safety time interval, it is counted as a following conflict; if the target node of the AMR is still occupied by other AMRs at its arrival time, it is counted as a target point blocking conflict; if the above conflicts occur repeatedly in the same area within a preset time window, it is counted as a repeated hotspot conflict; the congestion penalty term is obtained by weighted summation of the number of the various spatiotemporal conflicts.

[0009] Optionally, in step S2, the comprehensive scoring of the candidate solution considers multiple factors from the following: nominal maximum completion time, total AMR travel distance, balance of the number of tasks undertaken by each AMR, balance of route time consumption of each AMR, total shortfall of power below the warning threshold after task execution, total revenue from short-term wireless power replenishment at the work point, and a congestion penalty determined by the estimated number of node conflicts, edge conflicts, following vehicle conflicts, target point blocking conflicts, repeated hotspot area conflicts, and the maximum number of single-point hotspot conflicts.

[0010] Optionally, in step S3, selecting the AMR to yield includes: when the task size or number of conflicts is lower than a preset threshold, selecting the AMR to yield using a fixed priority rule; when the task size or number of conflicts reaches the preset threshold, switching to a dynamic priority rule, comprehensively comparing the conflict type, the current task execution status of the AMR, the accumulated yield waiting time of the AMR, the number of remaining tasks to be executed by the AMR, and the estimated impact of yielding different AMRs on the global final completion time, and selecting the AMR with the smaller yield cost.

[0011] Optionally, in step S3, the safe waiting point includes at least one of the following: AMR garage parking spot, preset backup parking spot, fixed charging station queuing spot, and idle grid node that does not affect the passage of the main channel; the selection criteria for the safe waiting point include: the distance between the current location of the AMR and the safe waiting point, the estimated time to travel to and from the conflict location from the safe waiting point, the occupancy status of the safe waiting point within the required waiting time window, and the impact of moving the AMR to the safe waiting point on the passage of other AMRs.

[0012] Optionally, in step S1, the candidate scheduling basic scheme is generated by a hierarchical method combining outer metaheuristic search and inner mathematical programming repair: the outer metaheuristic search adopts an adaptive large neighborhood search, and its destruction operators include at least one of removing a single task, removing consecutive task segments executed by the same AMR, and removing all tasks belonging to the same product process chain; its repair operators include at least one of inserting into the execution position with the minimum increment to the objective function using a greedy strategy and inserting into the execution position with the minimum sum of the connection distances between preceding and following tasks; the inner mathematical programming repair constructs a mixed integer linear programming repair model under a fixed task order to repair the task start and end time, the battery level before and after the task, the selection of fixed charging stations, the energy replenishment and replenishment time at fixed charging stations, and the selection of docking points.

[0013] Optionally, in step S2, the forward congestion prediction and candidate solution comprehensive scoring is performed in each iteration of the outer metaheuristic search. The outer metaheuristic search also includes at least one of the following acceleration strategies: maintaining an elite solution archive and generating new intermediate candidate solutions through path reconnection; adopting a rolling temporal prefix freezing strategy to freeze the task prefix that remains stable in multiple iterations, and only destroying and repairing the remaining task segments.

[0014] Optionally, in step S3, the spatiotemporal conflict detection further includes: statistically analyzing the local hotspot occupancy status of frequently occupied locations that repeatedly appear in narrow channels, intersections, near fixed charging stations, or near work points; in the iterative detection and correction, avoidance waiting events and safe waiting point migration events and the time delays they cause are retained as independent execution events in the final execution timeline and included in the final execution time of AMR.

[0015] In addition, this application also provides a Optionally, in step S2, the outer metaheuristic search further includes at least one of the following acceleration strategies: maintaining an elite solution archive, saving multiple candidate task chain solutions with high quality and appropriate structural differences in the search history, and generating new intermediate candidate solutions between the elite solutions and the current candidate solutions through path reconnection; adopting a rolling temporal prefix freezing strategy, freezing the task prefixes that remain stable and are located at the front end of the schedule during multiple rounds of iterative search, and subsequent searches only perform destruction and repair operations on the remaining task segments that are not frozen or the AMR suffix task segments that are significantly affected.

[0016] Compared with the prior art, the technical solution of the present invention has the following beneficial effects: Existing technologies treat task scheduling, charging decisions, and collision avoidance at the execution layer as three independent problems. This invention, through a collaborative mechanism of S1–S4, integrates task allocation, wireless charging planning, and collision-free execution timeline generation into a unified closed-loop framework: the task allocation result simultaneously constrains the charging decision and execution timeline, and the charging decision in turn influences the task execution order and path arrangement. This avoids the efficiency losses caused by passively handling charging demands or ignoring charging constraints in path planning, as seen in existing methods, and improves the overall coordination and executability of the scheduling scheme.

[0017] This invention explicitly supports the coordinated fulfillment of two types of constraints in S1: rapid wireless charging at fixed charging stations and short-term wireless charging at work sites. The final solution in S4 simultaneously outputs both charging schemes. Compared to existing technologies that rely on centralized charging at a single fixed charging station, this invention dynamically balances charging methods based on task sequence, remaining battery power, dwell time, and energy consumption of subsequent tasks. This fully leverages the short-term charging potential during AMR loading and unloading dwell times, reducing empty driving distances and downtime caused by detouring to dedicated charging points, and alleviating resource congestion at fixed charging stations.

[0018] In step S2 of this invention, congestion prediction is performed on candidate solutions during the nominal scheduling phase. This is achieved by generating a simplified discrete timeline to estimate the number of five types of spatiotemporal conflicts: node conflicts, edge conflicts, following conflicts, target point blocking conflicts, and recurring hotspot conflicts. These are then incorporated as congestion penalties into the comprehensive score of candidate solutions. This mechanism allows the scheduling system to assess the spatiotemporal competition cost of candidate solutions before execution, selecting nominal scheduling schemes with lower congestion penalties. Compared to existing technologies that only generate nominal task sequences and path plans without considering the actual operational conflicts of multiple AMRs, this invention effectively reduces the risk of a large number of AMRs being forced to wait in place for extended periods, or even forming cascading blockages, due to contention for the same node or path during actual operation, thus improving the field deployability of the scheduling scheme.

[0019] In step S3 of this invention, a discrete spatiotemporal execution timeline is constructed and conflict detection is performed. When a spatiotemporal conflict is detected, the AMR that needs to yield is dynamically selected based on the conflict type, the current task execution status of the AMR, and the accumulated yielding waiting cost. A yielding waiting event is inserted at the anchor point before the conflict occurs, or the system is temporarily migrated to a safe waiting point. Compared to existing systems that use a fixed-priority yielding mechanism, this invention comprehensively evaluates the impact cost of each yielding decision, avoiding phenomena such as repeated yielding by a single AMR, task backlog, and continuous congestion in local areas. By iteratively detecting and correcting, a collision-free final execution timeline is generated, which helps reduce the deviation between the nominal maximum completion time and the collision-free final execution time, improving system utilization.

[0020] This invention covers five conflict types in S2 prediction and S3 detection: node conflict, edge conflict, following vehicle conflict, target point blocking conflict, and recurring hotspot conflict. Among these, following vehicle conflict, long-term target point blocking, and recurring local hotspot congestion are dynamic execution layer problems that are generally ignored by existing methods. By extending conflict detection from the static node / edge level to the dynamic spatiotemporal competition level, this invention can more comprehensively identify potential collision risks in shared channels, intersections, workstations, and charging station areas, ensuring the safe and stable collaborative operation of multiple AMR clusters in complex dynamic smart factory grid environments.

[0021] Compared to previous applications that only jointly determine task allocation, execution order, and energy replenishment decisions at the nominal scheduling layer, this application transforms the estimated spatiotemporal conflicts of candidate schemes in the shared space into congestion penalties, which are then fed back to the candidate scheme screening process. This ensures that task allocation and execution order are constrained by estimated congestion risks before entering the execution layer. The execution layer then performs rigorous conflict detection on the screened nominal scheduling schemes, completing the final correction through dynamic yielding and safe waiting point migration. Forward-looking prediction is used to reduce the conflict density and long waiting time risk at the input execution layer, while dynamic correction is used to ensure that the final timeline is collision-free. The two respectively undertake the functions of risk shifting and executability fallback, and are continuously transmitted through the nominal scheduling schemes and their spatiotemporal occupancy information, forming an inseparable collaborative closed loop.

[0022] The core contribution of this invention lies in achieving full-chain collaborative optimization from task allocation to power replenishment decision-making and collision-free execution through a forward-looking congestion prediction and dynamic conflict correction mechanism. This overcomes the systemic defects in the prior art, such as poor coordination among task scheduling, wireless power replenishment planning and collision-free execution, disconnect between model and execution, and low on-site executability. Attached Figure Description

[0023] Figure 1 This is a flowchart illustrating the AMR cluster dynamic wireless power replenishment and collision avoidance collaborative scheduling method based on forward-looking congestion prediction in the embodiments of this application. Figure 2 This is a schematic diagram of the 25×25 grid layout of the smart factory in an embodiment of this application; Figure 3 This is a diagram of the collision-free execution path of AMR1 and its travel summary in the embodiments of this application; Figure 4 This is a summary of the collision-free execution path of AMR2 in the embodiments of this application. Figure 5 This is a summary of the collision-free execution path diagram and its journey of AMR3 in the embodiments of this application; Figure 6 This is a summary of the collision-free execution path diagram and its journey for the AMR4 in the embodiments of this application. Figure 7 This is a collision-free execution path diagram for all AMRs in this application embodiment; Figure 8 This is a comparison table of the efficiency of short-term wireless power replenishment at work points for the four AMRs in this application under different task scales. Figure 9 This table compares the operational performance of four AMRs in this application embodiment under different task scales, showing whether or not they have forward-looking congestion prediction capabilities. Detailed Implementation

[0024] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. 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 should fall within the scope of protection of the present invention.

[0025] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0026] As described in the background section, existing AMR scheduling technology separates the three key aspects of task allocation, wireless power replenishment planning, and collision-free execution. The task scheduling model only focuses on the theoretically optimal path and completion time, the power replenishment strategy passively relies on the power threshold to trigger centralized power replenishment at fixed charging stations, and the conflict resolution at the execution layer is carried out in isolation after the scheduling scheme is generated. As a result, although the nominal scheduling scheme is "feasible" in the mathematical model, it fails to predict the dynamic competition relationship between multiple AMRs in the shared space, which ultimately leads to frequent node conflicts, target point blockages, and follow-up congestion in actual operation, reducing the executability on site.

[0027] After long-term research, the inventors discovered that the most fundamental reason for the aforementioned predicament lies in the fact that existing technologies consistently treat "scheduling" as a static planning task focused solely on theoretical optimization, while viewing "execution" as an independent, post-hoc control action, lacking any forward-looking coordination mechanism between the two. This separation leads to three irreconcilable core contradictions: First, the energy replenishment decision is disconnected from the task sequence, ignoring the potential for short-term wireless energy replenishment at work sites and resulting in the over-concentration of fixed charging station resources; second, the scheduling model completely disregards the spatiotemporal competition at the execution layer, inevitably causing cascading congestion in multi-vehicle concurrent scenarios; third, the execution layer's avoidance mechanism relies solely on fixed priorities, failing to dynamically assess the cost of conflict to overall efficiency, leading to a vicious cycle of localized waiting. The root cause is that traditional methods treat "conflict" as an unpredictable disturbance, rather than a systematic variable that can be modeled in advance and incorporated into optimization objectives, thus losing the possibility of proactively suppressing congestion during the planning phase.

[0028] Based on the above findings, the inventors proposed using "prediction of execution-level conflicts" as a direct input for scheduling optimization. This means that when generating candidate scheduling schemes, evaluation criteria are no longer solely based on completion time, travel distance, or number of refueling attempts. Instead, a discrete timeline is generated based on candidate task allocation and task order. The number of node conflicts, edge conflicts, target point blocking conflicts, recurring hotspot area conflicts, and the largest single-point hotspot conflict that may occur during execution is actively calculated. These conflict counts are then directly incorporated into the comprehensive scoring system of candidate solutions as "congestion penalty terms." Through this mechanism, the scheduling process no longer simply pursues theoretical optimality but transforms into a collaborative optimization that balances "theoretical efficiency" and "execution feasibility." Schemes with seemingly short completion times but high expected conflict density will be eliminated due to high penalty scores; while schemes with slight delays but few conflicts and stable execution will be prioritized. Furthermore, after selecting high-scoring candidate solutions, mathematical programming is used to precisely determine the repair time, power consumption, and energy replenishment decisions to form a nominal scheduling scheme. Subsequently, only for this scheme, which has been "screened based on congestion prediction and has a good execution foundation," execution-level conflict detection and dynamic avoidance correction are initiated, thereby achieving closed-loop collaboration of "planning as execution."

[0029] Guided by this creative concept, the inventors proposed a collaborative scheduling method for dynamic wireless power replenishment and collision avoidance of AMR trunking based on forward-looking congestion prediction.

[0030] This invention provides an AMR cluster collaborative scheduling and hybrid power replenishment optimization method for smart factory grid environments. It is used for collaborative planning of a scheduling system that includes multiple AMRs, multiple work tasks, wireless power replenishment at fixed charging stations, short-term wireless power replenishment at workstation loading / unloading points, safety waiting points, and conflict resolution of shared lanes for multiple vehicles. This invention comprehensively considers factors such as task allocation, task execution order, process sequence, time coordination, power propagation, fixed charging decisions, wireless power replenishment at work points, congestion prevention of candidate solutions, and collision-free correction at the execution layer, forming a multi-AMR collision-free operation scheme that balances operational efficiency, energy security, power replenishment efficiency, and on-site feasibility. This invention includes planning at the following four levels: 1) Task Allocation and Job Sequencing Planning. In a smart factory, AMRs (Automatic Mobile Recyclers) are used to perform material handling tasks between raw material warehouses, workstations, and finished product warehouses, transporting materials required for production from a designated starting point to a target workstation, or transporting processed materials from a workstation to a subsequent location. Each job task has a clear starting point, ending point, process type, operation time, and energy consumption, and there are sequential constraints between multiple stages of job tasks corresponding to the same product. Since the number of AMRs is usually less than the number of job tasks, each AMR usually needs to execute multiple tasks continuously. Furthermore, since each AMR has a limited single loading capacity, its execution process generally includes an empty journey from the warehouse, raw material warehouse, or the end point of the previous task to the starting point of the target task, a loaded journey from the starting point to the end point of the task, and corresponding loading, unloading, and parking processes. Therefore, it is necessary to rationally allocate all job tasks among multiple AMRs and determine the execution order of tasks on the same AMR to satisfy process chain sequence constraints and task connection constraints, while minimizing the total time required to complete all tasks and improving the overall operating efficiency of the AMR cluster.

[0031] 2) Coordinated Planning of Wireless Fast Recharging at Fixed Charging Stations and Wireless Short-Term Recharging at Work Sites. Each AMR is equipped with a limited-capacity power battery, whose power level gradually decreases during unloaded movement, load transportation, task execution, and waiting. To ensure that the AMR can continuously complete its assigned tasks, it is necessary to coordinate the AMR's recharging behavior while meeting the safety power limit constraint and the subsequent task accessibility constraint. This invention comprehensively considers two types of wireless recharging scenarios: one is wireless fast recharging at fixed charging stations, where the AMR goes to a fixed charging station for high-speed wireless recharging after completing a task; the other is wireless short-term recharging at workstation loading or unloading points, where the AMR utilizes the parking time for low-speed short-term recharging during loading or unloading parking operations. Wireless charging at fixed charging stations offers a relatively stable and high-rate charging capability, but it consumes additional travel time, charging time, and fixed charging station resources. Short-term wireless charging at work sites can reduce the need for AMRs to travel to fixed charging stations during loading / unloading stoppages, but its charging capacity is limited by stoppage time, wireless charging rate, and task type. Therefore, it is necessary to collaboratively determine when the AMR should travel to a fixed charging station, which fixed charging station should be selected, how much power should be replenished, and how to utilize short-term wireless charging at work sites, based on the AMR's task sequence, remaining battery power, fixed charging station location, work station wireless charging capability, and the accessibility of subsequent tasks. This will balance operational continuity, fixed charging station resource consumption, and energy utilization efficiency.

[0032] 3) Congestion Prevention Planning for Candidate Scheduling Schemes. In areas where multiple AMRs share channels and workstations, optimizing only the nominal maximum completion time (the nominal scheduling indicator used before S2 screening) may lead to multiple AMRs operating in the same workstation, intersection, or local channel area, resulting in a large number of target point blockages, repeated hotspot conflicts, and subsequent avoidance waiting at the execution layer. To reduce the conflict correction cost in the post-processing stage, this invention introduces an execution congestion prevention mechanism during the generation and screening of candidate scheduling schemes. Specifically, when evaluating candidate task allocation and task order using metaheuristic search, in addition to considering nominal completion time, balanced task quantity, task chain integrity, power safety, and fixed charging station replenishment costs, it also estimates node conflicts, edge conflicts, following vehicle conflicts, target point blockage conflicts, repeated hotspot area conflicts, and the maximum number of single-point hotspot conflicts based on the discrete timeline of the candidate schemes, and incorporates these as congestion penalty items into the candidate solution score. In this way, excessive congestion in local areas can be suppressed in advance during the nominal scheduling stage, avoiding the generation of scheduling schemes with small maximum completion times but high avoidance costs at the execution layer.

[0033] 4) Execution Layer Spatiotemporal Conflict Detection and Collision-Free Correction Planning. In the scenario of multiple AMRs sharing a channel, even if task allocation, time coordination, and refueling decisions have been completed, various spatiotemporal conflicts may still occur during the actual execution process. For example, two AMRs may arrive at the same grid node at the same time, forming a node conflict; they may also occupy the same road segment at the same time, or travel towards each other along the same road segment, forming an edge conflict; there may also be following conflicts where the following vehicle closely follows the preceding vehicle into the same local area, target point blockage caused by the long-term occupation of task points or stopping points, and repeated hotspot conflicts formed by multiple conflicts repeatedly concentrating in the same area. The above conflicts will affect the actual executability of the scheduling scheme. Therefore, after obtaining a nominal scheduling scheme that satisfies the task and refueling constraints, this invention further constructs a discrete execution timeline for AMRs to perform spatiotemporal conflict detection on the node occupation, edge occupation, and safety interval status of each AMR at each time.

[0034] When a conflict is detected, this invention selects the AMR that needs to yield based on the conflict type, the current task status of the AMR, the accumulated waiting cost, and the impact on subsequent tasks, using dynamic priority rules. It inserts an avoidance waiting or safe waiting migration event before the conflict occurs. For short-term conflicts, an avoidance waiting event can be inserted at a safe anchor point, postponing its subsequent execution time. For long-term waiting or situations that may block task points, workstations, fixed charging stations, and local passageways, the AMR is temporarily migrated to a safe waiting point to wait until the conflict window ends, at which point it returns to its original execution node. The safe waiting point can be a parking garage spot, a fixed charging station queue point, or other waiting location that does not affect task execution. Time window occupancy checks prevent multiple AMRs from creating new occupancy conflicts at the same safe waiting point. If fixed priority avoidance stagnates in local conflicts or there are too many single-vehicle waits, a dynamic priority locking strategy can be switched to avoid repeatedly delaying a single AMR. By repeatedly executing the process of "conflict detection—yielding object selection—waiting or migration correction—timeline update—re-detection," an execution timeline without node conflicts, edge conflicts, or following vehicle conflicts, and meeting actual operational requirements, is finally generated.

[0035] Based on the collaborative planning at the four levels mentioned above, this invention further establishes a collaborative optimization framework that considers task allocation, task sequencing, process precursors, time propagation, power propagation, wireless fast charging at fixed charging stations, wireless short-term charging at work points, congestion prevention in candidate solutions, terminal docking, and collision-free correction at the execution layer. This framework is then solved using metaheuristic search and mathematical programming repair methods, thereby generating efficient, energy-saving, and directly executable collision-free operation schemes for multiple AMRs in smart factory scenarios. The method steps include: Step 1: Construct a mathematical model for AMR cluster scheduling that considers task allocation, power replenishment coordination, and execution feasibility.

[0036] The model employs a hierarchical objective optimization approach to minimize the risk of battery levels falling below the warning threshold after task execution, while ensuring the overall operational efficiency of the AMR cluster. Specifically, the objective function comprises the following two levels: First-level objective: The goal is to minimize the maximum job completion time of the AMR cluster. (1) in, This represents the maximum completion time of the AMR cluster scheduling scheme. The first-level objective aims to minimize the total time required to complete all job tasks, thereby improving the overall job efficiency of the AMR cluster.

[0037] Second-level objective: Based on the optimization results of the first layer, the objective is to minimize the sum of insufficient warnings after all tasks: (2) in, Represents the set of real tasks. Represents the AMR set, Indicates task By AMR After executing and completing the energy replenishment decision, the energy level is below the warning energy threshold. This insufficient energy level satisfies the following condition: (3) Equations (2) and (3) together constitute the second-level objective, which guides the model to avoid forming a large number of low-warning-power states without compromising the main completion time optimization objective, thereby improving the energy security and operational stability of the scheduling scheme.

[0038] Model constraints include: The garage initialization constraint is used to stipulate that each AMR starts from its corresponding virtual garage task and is injected with a unified initial time and initial power state, as expressed in Equation (4). Through this constraint, the initial position, initial time and initial power of each AMR can be uniformly incorporated into the scheduling model, providing a unified starting point for the time propagation and power propagation of subsequent real tasks.

[0039] (4) The task uniqueness assignment constraint is used to ensure that each real task is executed by one and only one AMR, as expressed in equation (5). This constraint can prevent the same task from being executed repeatedly by multiple AMRs, or a task from being executed by no AMR, thereby ensuring the integrity and uniqueness of the task assignment results.

[0040] (5) The chained scheduling constraint is used to ensure that after each AMR starts from its corresponding virtual garage task, the real tasks it executes can form a continuous, ordered, and non-repeating task chain, as expressed in equations (6) to (9). Equation (6) indicates that after each AMR starts from a virtual garage task, it must and can only connect to one first real task. Equation (7) indicates that each AMR assigned to an AMR... Real mission There must be one and only one precursor, which is either a virtual garage task. Or it could be another real mission. Equation (8) represents each assigned to AMR Real mission Either continue to connect to a subsequent real task, or be marked as the last task of this AMR. Equation (9) represents the consistency constraint between the preceding arc variable and the task assignment variable, that is, the preceding arc variable can only take the value 1 if both the preceding and following tasks are assigned to the same AMR, and a task cannot be its own preceding task. Through equations (6) to (9), it can be guaranteed that the real tasks on each AMR form a complete execution chain starting from the garage and ending at the end of the last task.

[0041] (6) (7) (8) (9) The MW / PW docking point selection constraint is used to uniquely determine the actual docking location for a given task when multiple candidate loading or unloading points exist, as expressed in equations (10) and (11). Equation (10) indicates that each Class A task can only select one MW docking point as the loading start point, and equation (11) indicates that each Class D task can only select one PW docking point as the unloading end point. Through equations (10) and (11), the uniqueness of the task start and end locations under multiple candidate docking point conditions can be guaranteed, providing a definite spatial location basis for subsequent task duration, time propagation, and power propagation constraints.

[0042] (10) (11) The process precursor constraint is used to ensure that tasks with process sequence relationships are executed in a predetermined production order, as expressed in equation (12). Through this constraint, it can be ensured that the tasks at each stage corresponding to the same product proceed in the predetermined process chain order, and to prevent subsequent tasks from being executed prematurely before the precursor tasks are completed.

[0043] (12) The task duration constraint is used to determine the correspondence between the start time, duration, and completion time of each real task, as expressed in equations (13) and (14). Through this constraint, the loading, driving, and unloading stages in the task execution process can be uniformly incorporated into the time modeling, thus providing a basis for subsequent time propagation constraints and maximum completion time constraints.

[0044] (13) (14) The maximum completion time constraint is used to limit the maximum completion time from multiple perspectives, including task completion time, lower bound of single vehicle load, lower bound of system average workload, and lower bound of completion time of a single process chain. The range of values ​​for is expressed as equations (15) to (18). Through equations (15) to (18), we can analyze from different perspectives... Providing an effective lower bound not only ensures the correctness of the maximum completion time definition, but also helps to enhance the relaxation of the lower bound of the model, reduce the search space, and improve the solution efficiency.

[0045] (15) (16) (17) (18) The AMR workload constraint is used to limit the range of actual tasks undertaken by each AMR, as expressed in equation (19). This constraint can prevent individual AMRs from being idle for a long time or overloaded, and the workload range can be flexibly set according to the task scale, number of AMRs, battery capacity and on-site scheduling requirements, thereby improving the balance and scalability of AMR cluster task allocation.

[0046] (19) The time propagation constraint from the garage to the first task describes the time connection required for the AMR to travel from its initial garage position to the starting point of its first real task, as expressed in equation (20). This constraint ensures that the AMR has sufficient departure and travel time before executing the first real task, thus keeping the task start time consistent with the actual operation process.

[0047] (20) Inter-task time propagation constraints are used to describe the time update relationship when the same AMR performs the next task after completing the previous task, and are expressed as Equation (21) and Equation (22).

[0048] The time propagation relationship between tasks without charging is expressed as Equation (21). This equation describes the time connection relationship when the AMR is directly transferred from the previous task to the next task without recharging.

[0049] Charging status (i.e.) The time propagation relationship between tasks under the condition of charging is expressed as Equation (22). Equations (21) and (22) can be used to characterize the time propagation relationship between tasks in the two cases of no charging and charging, thereby ensuring the feasibility of AMR task sequences in the time dimension.

[0050] (twenty one) (twenty two) The power update constraint under short-term wireless power replenishment at the work site is used to describe the change in the power state of the AMR before and after the task, after it receives wireless power replenishment during the workstation loading and unloading phases. This is expressed by equations (23) to (27). Equation (23) is used to define the task. Intermediate electrical charge after initial standby and recharging; Equation (24) is used to represent the task. The net power update relationship; Equation (25) is an equivalent expression of Equation (24); Equation (26) is used to ensure that the power after the initial parking and recharging does not exceed the upper limit of the battery capacity; Equation (27) is used to ensure the power safety of the AMR during the task execution process. Through Equations (23) to (27), the influence of the working point wireless recharging on the power status during the task execution process can be explicitly incorporated into the model, thereby realizing the collaborative modeling between task execution, energy consumption and short-term wireless recharging.

[0051] (twenty three) (twenty four) (25) (26) (27) The pre- and post-task safety power constraint is used to ensure that the AMR meets the minimum safety power requirement before and after the start of each real task, as expressed in equation (28). This constraint ensures that the AMR has sufficient power at both the start and end of the task, thus avoiding situations where the AMR enters a dangerous low power state before the task has started or immediately after completion.

[0052] (28) The three-stage state propagation constraint for wireless fast charging at fixed charging stations is used to describe the update relationship of the AMR's power state in the three stages of "after task completion", "before arriving at the charging station", and "after charging decision is completed" when it goes to the fixed charging station for wireless fast charging after completing its mission. It is expressed as equations (29) to (32). Equation (29) is used to represent the state propagation constraint of the AMR in the three stages of "after task completion", "before arriving at the charging station", and "after charging decision is completed". After completion, proceed to the next step. When arriving at a fixed charging station, the power state before reaching the charging station is given; Equation (30) is used to represent the power state after the task is completed without charging; Equation (31) is used to represent the power state after the task is completed with charging; Equation (32) is used to give the power boundary constraints in the wireless fast charging process. Through Equations (29) to (32), the process of going to a fixed charging station after the task is completed, choosing whether to charge, and the power recovery process after charging can be uniformly incorporated into the model, thereby realizing the phased state propagation modeling of the wireless fast charging behavior.

[0053] (29) (30) (31) (32) The post-task wireless fast charging decision constraint is used to ensure that fixed charging behavior only occurs when the task is actually executed, and once charging occurs, a fixed charging station must be uniquely selected, and the corresponding charging time is determined by the charging amount, as expressed in equations (33) to (36). Equation (33) describes the consistency relationship between task execution and whether post-task fixed charging is allowed; equation (34) describes the correspondence between charging station selection and post-task charging decision; equation (35) limits the range of values ​​for the wireless charging amount of the post-task fixed charging station; and equation (36) describes the correspondence between the wireless charging time and the charging amount of the post-task charging station. Through equations (33) to (36), it can be ensured that the post-task fixed charging behavior is consistent with the task execution state, charging station selection, charging amount, and charging time, thereby realizing the standardized modeling of the fixed charging decision process.

[0054] (33) (34) (35) (36) The energy propagation constraint from the garage to the first task is used to describe the energy consumption relationship generated by the AMR from the initial garage position to the starting point of the first real task, and is expressed as Equation (37). Through this constraint, the energy consumption of the AMR from the garage to the starting point of the first task is explicitly incorporated into the model, providing an initial energy basis for the energy propagation and refueling decisions of subsequent tasks.

[0055] (37) The inter-task power propagation constraint describes the update relationship of the power level before each task when the same AMR executes the next task after completing the previous one, and is expressed as Equations (38) and (39). The inter-task power propagation relationship under the non-charging condition is expressed as Equation (38); the inter-task power propagation relationship under the charging condition is expressed as Equation (39). Equations (38) and (39) can respectively characterize the inter-task power connection relationship under the two conditions of non-charging and charging, thereby ensuring that the power propagation of the AMR in the process of continuous execution of multiple tasks meets the actual operating rules.

[0056] (38) (39) A one-step forward safety constraint is used to ensure that the AMR can safely connect to subsequent tasks even if it does not immediately perform fixed charging after the current task, and still has the power reserve to reach the fixed charging station after the subsequent task is completed, as expressed in equation (40). Through this constraint, it can be ensured that the AMR will not fall into a state of being unable to reach the fixed charging station after completing the subsequent task if it continues to perform tasks without charging.

[0057] (40) The final mission reachable charging station constraint is used to ensure that after completing the last real mission, the AMR still has the ability to reach the nearest fixed charging station even if it no longer needs to be recharged, as expressed in equation (41). This constraint can prevent the AMR from being unable to safely leave the mission area or reach the recharge location due to insufficient remaining power after completing the last mission.

[0058] (41) The insufficient early warning constraint is used to define the early warning power gap variable in the second-level objective function and to quantify the degree to which the power level falls below the early warning threshold after task execution, as expressed in equations (42) and (43). The risk of the power level being too low after the task but not yet below the safe power level can be explicitly included in the second-level optimization objective, thereby improving the energy security and operational robustness of the scheduling scheme.

[0059] (42) (43) Finally, using the hierarchical objective function formed by equations (1) and (2) as the optimization objective and equations (3)-(43) as the constraints, a hybrid integer programming model for nominal collaborative scheduling of AMR clusters considering fixed charging, wireless power replenishment at the working point and battery capacity constraints is constructed.

[0060] Step 2: Design a hierarchical solution method based on a combination of metaheuristics and mathematical programming.

[0061] In this embodiment, the AMR cluster collaborative scheduling model established in step 1 includes decisions on task allocation, task sequencing, process precursors, time propagation, power propagation, wireless fast charging at fixed charging stations, wireless short-term charging at work points, docking point selection, and terminal docking. To avoid the model becoming too large due to direct overall solution, the work points are workstation loading or unloading points. This embodiment adopts a hierarchical solution method of "outer layer metaheuristic search of task structure, inner layer mathematical programming for repair time and power decision-making".

[0062] First, an initial scheduling solution is constructed based on the product task chain structure, task spatial location, initial position of the AMR, and distance between the task start and end points. The initial scheduling solution includes the task allocation results for each AMR and the task execution order on the same AMR. For a product task chain of the form A→B→C→D, the process sequence within the chain is maintained when generating the initial solution, and task chains or task segments are allocated between different AMRs.

[0063] Then, focusing on task allocation relationships and task execution order as the main optimization objects, an adaptive large neighborhood search method is used for outer-layer search. The outer-layer search removes some tasks, task segments, or product process chains from the current candidate solutions using a disruption operator, and then re-inserts them into suitable AMRs and execution locations using a repair operator, thereby forming a new task chain structure. When evaluating candidate solutions, factors such as nominal maximum completion time, total travel distance, AMR task quantity balance, route time balance, product task chain segment integrity, power safety, and short-term wireless recharging benefits at work sites are comprehensively considered.

[0064] To improve the feasibility of candidate solutions, this embodiment introduces nominal spatiotemporal conflict prediction and forward-looking congestion prediction, along with a congestion penalty term, in the outer search phase. Specifically, a simplified discrete timeline is generated based on candidate task allocation and task order. Node conflicts, edge conflicts, following conflicts, target point blocking conflicts, repeated hotspot area conflicts, and the maximum number of single-point hotspot conflicts are predicted and incorporated as congestion penalty terms into the candidate solution score. This approach allows for the suppression of locally excessively congested task structures in advance during the nominal scheduling phase, reducing the cost of subsequent obstacle avoidance corrections at the execution layer.

[0065] During the search process, the selection weights can be dynamically adjusted based on the historical performance of different destruction and repair operators, and an elite solution archive can be maintained to save several high-quality candidate solutions. For candidate solutions in the elite solution archive, new candidate task chain structures can be further generated through path reconnection. For larger-scale task scenarios, a rolling temporal prefix freezing strategy can also be adopted to fix the stable front-end task sequences in multiple rounds of search, so that subsequent searches focus on the remaining task segments or AMR suffix task segments that are more affected by disturbances.

[0066] After the outer layer search provides candidate task allocation and task order, the inner layer constructs a small-scale mixed integer linear programming repair model under a fixed task order. This model repairs and optimizes decisions such as task start time, task completion time, battery level before and after the task, battery level before reaching the fixed charging station, wireless energy replenishment at the fixed charging station, wireless energy replenishment time at the fixed charging station, selection of the fixed charging station, selection of the raw material warehouse docking point, and selection of the finished product warehouse docking point. This ensures that the candidate solutions satisfy time constraints, battery constraints, wireless energy replenishment constraints at the fixed charging station, and short-term wireless energy replenishment constraints at the work point.

[0067] During the repair process, a combination of local and global repair methods can be adopted. For candidate solutions with only local changes, a local repair model is first established for the disturbed task segment or the AMR suffix task segment, while the unchanged task prefix inherits the feasible solution from the previous round; when local repair is not applicable or fails to obtain a feasible solution, the process switches to global repair under a fixed task order. For a small number of infeasible candidate solutions due to insufficient power, a wireless power replenishment insertion repair strategy based on a safe power threshold and the nearest reachable fixed charging station can be used as a supplement.

[0068] After the repair is completed, a comprehensive screening of candidate nominal scheduling schemes is conducted. The comprehensive score includes factors such as nominal maximum completion time, number of fixed charging station refueling attempts, total refueling time at fixed charging stations, final execution time after obstacle avoidance pre-correction, obstacle avoidance waiting cost, number of residual conflicts, task quantity balance, and nominal congestion penalty. For candidate solutions that still have residual conflicts or are difficult to resolve after pre-correction, a penalty weight is increased to avoid selecting schemes with short nominal completion times but poor execution-level feasibility.

[0069] Through the above-mentioned hierarchical solution method, this embodiment can obtain a nominal scheduling scheme that satisfies the time, power and wireless charging logic constraints while maintaining the coupling relationship between task scheduling, fixed charging station wireless fast power replenishment, work point wireless short-term power replenishment and power propagation. It also provides a scheduling foundation with fewer conflicts, less congestion and easier execution for subsequent collision-free correction at the execution layer.

[0070] Step 3: Perform execution layer spatiotemporal conflict detection and avoidance correction.

[0071] Based on the nominal scheduling scheme obtained in step 2, a discrete spatiotemporal execution timeline for each AMR is constructed, and the occupancy relationships of multiple AMRs in shared nodes, shared channels, and parking positions are detected and corrected to generate a final collision-free execution scheme. This step resolves execution-level conflicts in the nominal scheduling scheme without changing the task process sequence or task allocation results.

[0072] First, the actions of each AMR—from departing from the garage, heading to the task starting point unloaded, loading and parking, transporting the load, unloading and parking, heading to the fixed charging station, performing wireless rapid charging at the fixed charging station, performing wireless short-term charging at the work point, returning to the standby position, and docking—are expanded into a time-ordered sequence of execution events. Then, at a discrete time scale, the node occupancy, path edge occupancy, parking position occupancy, and waiting status of each AMR are recorded to form a spatiotemporal occupancy table.

[0073] Subsequently, conflict detection is performed on the spatiotemporal occupancy table. If multiple AMRs occupy the same grid node at the same time, it is determined to be a node conflict; if multiple AMRs occupy the same path edge simultaneously, or two AMRs pass towards each other along the same path edge, it is determined to be an edge conflict; if the target node of an AMR is still occupied by other AMRs when it arrives, it is determined to be a target point occupancy conflict; if the preset safety interval is not met between preceding and following AMRs in adjacent time steps, it is determined to be a following conflict. For frequently occurring occupancy locations that appear repeatedly in narrow passages, intersections, near fixed charging stations, or near work sites, local hotspot occupancy can also be statistically analyzed to identify potential congestion areas.

[0074] When a conflict is detected, the AMR that needs to yield is selected based on the conflict type, the current status of the AMR, the number of subsequent tasks, the accumulated waiting time, and the impact on the global completion time. For scenarios with small task size or few conflicts, a fixed priority rule can be used; for scenarios with large task size, many wireless charging participants, or dense local conflicts, a dynamic priority rule can be used to select the AMR with the lower avoidance cost to yield.

[0075] After determining which AMR should yield, an avoidance waiting event is inserted at the appropriate anchor point before the conflict occurs, and all subsequent execution events are postponed. For short-term conflicts, the AMR can wait at its current location; for longer waiting times and situations where the current location may block the main passage, the AMR can be temporarily moved to a safe waiting point. Safe waiting points include garages, backup parking spots, fixed charging station queuing points, or other idle nodes that do not affect the passage of the main passage. Safe waiting points can be selected based on the current location, conflict location, distance to the waiting point, round-trip time, time window occupancy, and impact on other AMR paths. When a new avoidance waiting event occurs at a work point with wireless charging capability, and the AMR maintains effective wireless coupling during the waiting period, the waiting time that meets the charging conditions is included in the work point's wireless charging time, and the subsequent power status of the affected AMR is updated; when the AMR migrates to a safe waiting point without wireless charging capability, only standby power consumption occurs during the waiting period. The updated power status is used to verify the power safety and fixed charging station accessibility of subsequent tasks.

[0076] After completing one yielding wait or safe waiting point migration, the subsequent timeline of the affected AMR is regenerated, and conflict detection is performed again. If node conflicts, edge conflicts, target point occupancy conflicts, or following conflicts still exist, the process of yielding selection, waiting insertion, or safe waiting point migration continues until a final execution timeline that meets the requirements of no node conflicts, no edge conflicts, and safe following is formed.

[0077] During the execution layer correction process, avoidance waiting events, safe waiting point migration events, and the time delays they cause are all retained as independent execution events in the final scheduling result and included in the final AMR execution time (used for the time when all AMRs complete their tasks after S3 collision avoidance correction). Using this method, major spatiotemporal conflicts in multi-AMR shared channel operation can be eliminated without disrupting the task process sequence and wireless power replenishment logic. This also reduces local congestion caused by prolonged in-situ waiting and improves the field executability of the scheduling scheme.

[0078] Steps S2 and S3 employ mutually mappable discrete spatiotemporal representations. Step S2 generates a simplified timeline based on candidate task allocation and execution order, predicting the spatiotemporal competition risk of candidate solutions with low computational cost, and determining whether a candidate solution enters subsequent precise repair and execution layer processing through congestion penalties. Step S3 performs rigorous spatiotemporal occupancy detection on the complete execution timeline that retains movement, parking, recharging, and waiting events, and ensures the final solution is collision-free through iterative correction. The former is used to proactively reduce potential congestion during the scheduling optimization phase, while the latter is used to eliminate remaining actual execution conflicts. The two are continuously connected through the nominal scheduling scheme and the corresponding node-path edge-time step occupancy relationship.

[0079] Step 4: Output the final collaborative scheduling scheme and operation evaluation results.

[0080] After completing the spatiotemporal conflict detection and collision-free correction at the execution layer in step 3, the final collaborative scheduling scheme is output. The final collaborative scheduling scheme includes at least: the task allocation results of each AMR, the task execution order, the start time and completion time of each task, the power status before and after the task, the fixed charging station selection results, the wireless power replenishment and wireless power replenishment time at the fixed charging station, the short-term wireless power replenishment process at the work point, the terminal docking location, and the collision-free execution timeline corrected by the execution layer.

[0081] The collision-free execution timeline includes execution events for each AMR at discrete time scales, such as node occupancy, path edge occupancy, loading / unloading parking, wireless fast charging at fixed charging stations, wireless short-term charging at work points, obstacle avoidance waiting, safe waiting point migration, and terminal docking. For obstacle avoidance waiting events and safe waiting point migration events inserted during the execution layer correction process, their occurrence time, duration, associated AMR, and corresponding conflict type are retained in the final scheme to ensure that the scheduling results can be completely parsed and executed.

[0082] The final collaborative scheduling scheme determines the operating paths, task connections, wireless power replenishment behaviors, obstacle avoidance behaviors, and docking states of multiple AMRs throughout the entire operation cycle, ensuring that the output simultaneously meets task process constraints, time constraints, power constraints, hierarchical wireless power replenishment constraints, and collision-free execution requirements for multiple AMRs. Therefore, this scheme can serve as a direct basis for the operation scheduling and field control of multi-AMR systems in smart factories. The method flow can be expressed as follows: Figure 1 .

[0083] Figure 2 This is a schematic diagram of a 25×25 grid layout for a smart factory, used to represent the AMR operating environment in this embodiment of the invention. The map uses grid nodes of unit length to form passable areas; gray dots represent feasible nodes that the AMR can traverse. The blue area in the lower left corner is the raw material warehouse, the yellow area in the upper left corner is the finished product warehouse, and the orange area in the lower right corner is the AMR parking area. Green areas represent workstations, indicating loading and unloading locations for different process tasks. Red areas represent fixed charging stations for rapid wireless recharging of the AMR. Black areas represent obstacles that the AMR must not cross during path planning and scheduling. The small dots of different colors in the diagram represent loading points in the raw material warehouse, unloading points in the finished product warehouse, loading points in the workstation, unloading points in the workstation, fixed charging points, and AMR docking points, respectively, used to define the specific node locations that the AMR can choose when performing tasks, recharging, and docking.

[0084] Figure 3 , Figure 4 , Figure 5 , Figure 6 and Figure 7 This illustration demonstrates a typical collision-free execution path for multiple AMRs in a smart factory grid environment. The accompanying figures include the complete execution paths for each of AMR1, AMR2, AMR3, and AMR4, as well as the overall scheduling execution path for the four AMRs stacked together.

[0085] Each vehicle's path map illustrates the entire process of its corresponding AMR, from its departure from the garage to the task starting point, performing loading and transportation tasks, arriving at the workstation or warehouse node, receiving wireless charging at a fixed charging station or short-term wireless charging at the work point as needed, completing obstacle avoidance and waiting, and finally returning to the standby position. Different colors are used to distinguish the operating trajectories of different AMRs: solid lines represent task execution paths, dashed or dotted lines represent non-loaded operation processes such as departure from the garage, empty transfer, and terminal standby, and locations marked with crosses indicate obstacle avoidance and waiting events inserted during the execution layer conflict resolution process.

[0086] As can be seen from the overall path diagram of the four AMRs, each AMR coordinates its path and time according to the execution layer's spatiotemporal conflict detection results in shared channels, intersection nodes, near workstations, and near fixed charging stations. After avoidance waiting, safety interval control, and timeline extension, the node and edge occupancy of each AMR does not overlap, avoiding problems such as simultaneous occupancy of the same node, conflicts in the same or opposite directions on the same road segment, and target point occupation. The operation summary on the right further provides the number of tasks, running distance, final completion time, power changes, and detailed execution event sequences for each AMR, illustrating that the scheduling scheme can be fully parsed and used for on-site execution control.

[0087] The accompanying figures show that the scheduling results generated by this invention can not only determine the task allocation and path execution order of each AMR, but also form a collision-free execution timeline based on consideration of refueling behavior and multi-vehicle shared space constraints, thereby improving the actual executability of multi-AMR collaborative operation schemes.

[0088] Figure 8 The table shows a comparison of the final execution times of four AMRs under different task chain scales, with and without the use of on-site wireless short-term power replenishment. The number of task chains in the table are 4, 6, 8, 10, and 12, corresponding to different task scales. "No on-site wireless short-term power replenishment" means that the AMR relies solely on fixed charging stations for wireless fast power replenishment to complete the task. "With on-site wireless short-term power replenishment" means that the AMR can not only go to fixed charging stations for wireless fast power replenishment, but also perform on-site wireless short-term power replenishment during loading and unloading stops at the workstation.

[0089] Depend on Figure 8 It can be seen that, under the same number of AMRs and the same task scale, the final execution time for each task scale is reduced after introducing short-term wireless power replenishment at the work site. Specifically, the final execution time decreased from 173 to 136 with 4 task chains, an efficiency improvement of 21.38%; from 321 to 178 with 6 task chains, an efficiency improvement of 44.55%; from 478 to 298 with 8 task chains, an efficiency improvement of 37.66%; from 527 to 432 with 10 task chains, an efficiency improvement of 18.02%; and from 743 to 535 with 12 task chains, an efficiency improvement of 27.99%. These results indicate that short-term wireless power replenishment at the work site can effectively utilize the loading / unloading time of AMRs or the downtime of workstations to replenish power, reducing the empty running, waiting, and downtime caused by dedicated trips to fixed charging stations for power replenishment, thereby improving the operational continuity and overall scheduling efficiency of multi-AMR systems under different task scales.

[0090] Figure 9The table shows a performance comparison of four AMRs operating under different task chain scales, with and without forward-looking congestion prediction enabled. To ensure consistency of the comparison results, both sets of experiments used the same smart factory map, number of AMRs, task instances, random seeds, battery and recharging parameters, and execution-layer dynamic obstacle avoidance strategy. The only difference was whether the forward-looking congestion prediction mechanism in step S2 participated in the generation and selection of candidate scheduling schemes. The potential spatiotemporal conflict number in the table is the sum of the number of node conflicts and edge conflicts detected before the execution-layer dynamic obstacle avoidance correction; the total waiting time includes nominal scheduling waiting and the obstacle avoidance waiting added by the execution layer; the final execution time is the time required for all AMRs to complete the corresponding tasks after the dynamic obstacle avoidance correction in step S3. All times in the table are simulation time units.

[0091] Depend on Figure 9 The results show that after implementing forward-looking congestion prediction, the average number of potential spatiotemporal conflicts under the five task scales decreased from 107 to 33, a decrease of 69.07%; the average total waiting time decreased from 166.6 to 123.6, a decrease of 25.81%; the average maximum waiting time decreased from 65.6 to 39.4, a decrease of 39.94%; and the average final execution time decreased from 356.8 to 341.6, an overall improvement of 4.26%. These results indicate that forward-looking congestion prediction has a significant inhibitory effect on potential spatiotemporal competition and long waiting times, and can improve the overall final execution efficiency of multi-AMR systems.

[0092] In scenarios with 8, 10, and 12 task chains where congestion pressure is more pronounced, the role of proactive congestion prediction becomes even more significant. Specifically, in the 8-task-chain scenario, the number of potential spatiotemporal conflicts decreased from 58 to 13, a reduction of 77.59%, the maximum waiting time decreased from 22 to 9, and the final execution time decreased from 297 to 286, an improvement of 3.70%; in the 10-task-chain scenario, the number of potential spatiotemporal conflicts decreased from 118 to 31, a reduction of 73.73%, the maximum waiting time decreased from 49 to 27, and the final execution time decreased from 418 to 394, an improvement of 5.74%; and in the 12-task-chain scenario, the number of potential spatiotemporal conflicts decreased from 228 to 57, a reduction of 75.00%, the maximum waiting time decreased from 239 to 144, and the final execution time decreased from 681 to 632, an improvement of 7.20%. This indicates that when proactive congestion prediction identifies concentrated node contention, path contention, or local hotspot occupancy in candidate scheduling schemes, it can guide task allocation and task execution order to avoid spatiotemporal combinations with high congestion risk, thereby reducing the complexity of subsequent dynamic obstacle avoidance correction and the risk of long-term blockage.

[0093] In the four-task-chain scenario, the same final execution result was obtained before and after enabling forward congestion prediction, indicating that when there are no prominent congestion bottlenecks in the candidate scheduling scheme that require further improvement, forward congestion prediction will not forcibly change the original feasible solution. In the six-task-chain scenario, after enabling forward congestion prediction, the number of potential spatiotemporal conflicts decreased from 83 to 16, a decrease of 80.72%, and the maximum waiting time decreased from 13 to 12. Although the final execution time changed slightly from 221 to 229, this result reflects the coordinated trade-off between completion efficiency and congestion risk in the comprehensive score of candidate solutions, that is, a smaller fluctuation in execution time is exchanged for a more significant reduction in potential conflict risk.

[0094] After the dynamic obstacle avoidance correction at the execution layer in step S3, all the above-mentioned schemes ultimately achieve a strict spatiotemporal conflict count of zero, while simultaneously satisfying battery safety constraints and charging resource constraints. Therefore, the forward-looking congestion prediction in step S2 does not replace the dynamic conflict detection and avoidance correction in step S3, but rather moves some conflict management from the execution stage to the scheduling optimization stage, giving the nominal scheduling scheme input to the execution layer a lower potential congestion risk and a better execution foundation. This demonstrates that the present invention, through the synergistic effect of forward-looking congestion prediction and dynamic conflict correction, can reduce the risks of spatiotemporal competition, long waiting times, and local cascading blockages in the operation of multiple AMRs in a shared space, thereby improving the field executability of the scheduling scheme and the stability of system operation.

[0095] The above descriptions are merely preferred embodiments of the present invention and are not intended to limit the scope of the invention. These embodiments are simply examples of the principles and applications of the invention. Therefore, it should be understood that many modifications can be made to the exemplary embodiments, and other arrangements can be designed, as long as they do not depart from the spirit and scope of the invention as defined by the appended claims. It should be understood that different dependent claims and features described herein can be combined in ways different from those described in the original claims. It is also understood that any simple, equivalent changes and modifications made based on the claims and description of this application fall within the scope of protection of this patent.

Claims

1. A collaborative scheduling method for dynamic wireless power replenishment and collision avoidance of AMR trunking based on forward-looking congestion prediction, characterized in that, include: S1. Obtain candidate scheduling basic schemes for the AMR cluster. The basic schemes include the task allocation results and task execution order of each AMR, and satisfy the constraints of unique task allocation, process precursor, power safety, and wireless charging at fixed charging stations and short-term wireless charging at work points. S2. Based on the task allocation results and task execution order in the candidate solutions, a simplified discrete timeline is generated. Under the spatiotemporal occupancy relationship of grid nodes, path edges, and discrete time steps, the number of five types of spatiotemporal conflicts is estimated: node conflict, edge conflict, following vehicle conflict, target point blocking conflict, and repeated hotspot conflict. A congestion penalty term is formed according to the preset weights corresponding to the five types of spatiotemporal conflicts, and the congestion penalty term and the nominal maximum completion time are included in the comprehensive score of the candidate solutions. A nominal scheduling scheme is formed based on the comprehensive score. S3. Construct the discrete spatiotemporal execution timeline of each AMR based on the nominal scheduling scheme and perform spatiotemporal conflict detection; when a spatiotemporal conflict is detected, select the AMR that needs to be yielded according to the conflict type, the current task execution status of the AMR and the accumulated yield waiting cost, insert the yield waiting event at the anchor point before the conflict occurs, or temporarily migrate the AMR to the preset safe waiting point. After each correction, the node occupancy, path edge occupancy, parking occupancy, and subsequent event times of the affected AMR are recalculated, and spatiotemporal conflict detection is performed again until a final execution timeline that meets the requirements of no node conflict, no edge conflict, and safe interval is formed; the inserted avoidance events and the time delays they cause are retained in the final execution timeline. S4. Output the final collaborative scheduling scheme, which includes the task allocation results of each AMR, the task execution order, the wireless power replenishment scheme for fixed charging stations, the wireless short-term power replenishment scheme for work points, and the collision-free execution timeline.

2. The method according to claim 1, characterized in that, In step S2, the estimation methods for the five types of spatiotemporal conflicts include: if the same grid node is occupied by two or more AMRs at the same time, it is counted as a node conflict; if the same path edge is occupied by two or more AMRs at the same time, or if two AMRs pass each other along the same path edge, it is counted as an edge conflict; if the following AMR and the preceding AMR enter the same node or the same road segment in adjacent time steps and do not meet the preset safety time interval, it is counted as a following conflict; if the target node of the AMR is still occupied by other AMRs when it arrives, it is counted as a target point blocking conflict; if the above conflicts occur repeatedly in the same area within a preset time window, it is counted as a repeated hotspot conflict; the congestion penalty term is obtained by weighted summation of the number of the various spatiotemporal conflicts.

3. The method according to claim 1, characterized in that, In step S2, the comprehensive scoring of the candidate solution simultaneously considers multiple factors from the following: nominal maximum completion time, total AMR travel distance, balance of the number of tasks undertaken by each AMR, balance of route time consumption of each AMR, total shortfall of power below the warning threshold after task execution, total revenue from short-term wireless power replenishment at the work point, and a congestion penalty item determined by the estimated number of node conflicts, edge conflicts, following vehicle conflicts, target point blocking conflicts, repeated hotspot area conflicts, and the maximum number of single-point hotspot conflicts.

4. The method according to claim 1, characterized in that, In step S3, the selection of AMRs to yield includes: when the task size or number of conflicts is lower than a preset threshold, a fixed priority rule is used to select the AMR to yield; when the task size or number of conflicts reaches the preset threshold, a dynamic priority rule is switched to comprehensively compare the conflict type, the current task execution status of the AMR, the accumulated avoidance waiting time of the AMR, the number of remaining tasks to be executed by the AMR, and the estimated impact of yielding different AMRs on the global final completion time, and select the AMR with the smaller yielding cost.

5. The method according to claim 1, characterized in that, In step S3, the safe waiting point includes at least one of the following: AMR garage parking point, preset backup parking point, fixed charging station queuing point, and idle grid node that does not affect the passage of the main channel; Safety The selection criteria for waiting points include: the distance between the current location of the AMR and the safe waiting point, the estimated time to travel to and from the conflict location from the safe waiting point, the occupancy status of the safe waiting point within the required waiting time window, and the impact of moving the AMR to the safe waiting point on the passage of other AMR paths.

6. The method according to claim 1, characterized in that, In step S1, the candidate scheduling base scheme is generated by a hierarchical method combining outer metaheuristic search and inner mathematical programming repair: the outer metaheuristic search adopts an adaptive large neighborhood search, and its destruction operators include at least one of removing a single task, removing consecutive task segments executed by the same AMR, and removing all tasks belonging to the same product process chain; its repair operators include at least one of inserting into the execution position with the minimum increment to the objective function using a greedy strategy and inserting into the execution position with the minimum sum of the connection distances between preceding and following tasks; the inner mathematical programming repair constructs a mixed integer linear programming repair model under a fixed task order to repair the task start and end time, the battery level before and after the task, the selection of fixed charging stations, the energy replenishment and replenishment time at fixed charging stations, and the selection of docking points.

7. The method according to claim 1, characterized in that, In step S2, the forward congestion prediction and candidate solution comprehensive scoring are performed in each iteration of the outer metaheuristic search. The outer metaheuristic search also includes at least one of the following acceleration strategies: maintaining an elite solution archive and generating new intermediate candidate solutions through path reconnection; adopting a rolling temporal prefix freezing strategy to freeze the task prefix that remains stable in multiple iterations, and only destroying and repairing the remaining task segments.

8. The method according to claim 1, characterized in that, In step S3, the spatiotemporal conflict detection further includes: statistically analyzing the local hotspot occupancy of frequently occurring occupancy locations such as narrow channels, intersections, near fixed charging stations, or near work points; in the iterative detection and correction, avoidance waiting events and safe waiting point migration events and the time delays they cause are retained as independent execution events in the final execution timeline and included in the final execution time of AMR.

9. A collaborative scheduling device for dynamic wireless power replenishment and collision avoidance of AMR trunking based on forward-looking congestion prediction, characterized in that, The method includes a processor and a memory, the memory storing at least one instruction, which is loaded and executed by the processor to implement the method as described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that, The system stores at least one instruction that, when executed by a processor, implements the method as described in any one of claims 1 to 8.

Images