Method for simulating and planning of device cluster job based on multi-agent cooperation

Abstract

This invention relates to the field of simulation technology, specifically to a simulation planning method for equipment cluster operations based on multi-agent collaboration. This invention constructs a plot grid model and upper and lower bound channels for production density, uses forward cumulative scanning to determine the capacity critical threshold interval, and generates target points based on the midpoint of the interval and accessibility constraints to guide the collaborative equipment path. A target point offset gating mechanism is introduced to control path replanning triggering, improving path stability and responsiveness. During the operation, the interval boundaries are dynamically corrected based on measured production, and the final meeting point is locked when the interval length and meeting distance conditions are met. After the simulation, row attribution analysis is performed using locking deviation and cumulative displacement indices, and row sequence swapping and iterative resimulation optimization are combined with row accessibility to output the optimal operation plan. This method can improve the efficiency and simulation accuracy of collaborative equipment operations in complex environments.

Description

Technical Field

[0001] This invention relates to the field of simulation technology, specifically to a method for simulating and planning equipment cluster operations based on multi-agent collaboration. Background Technology

[0002] The software simulation execution engine is the core technology for realizing multi-agent collaborative operation planning simulation. Current methods virtualize the behavioral logic, interaction protocols, and environmental models of each agent to simulate the cluster operation process at the software level. In the aforementioned simulation execution engine architecture, the collaborative interaction triggering conditions of each agent are typically injected into the simulation environment as static parameters independent of the path execution module. The path execution simulation module and the collaborative simulation module run decoupled within the engine, driven uniformly by the simulation execution engine to complete the software simulation of the cluster collaborative process.

[0003] However, when the aforementioned software simulation execution engine is applied to the planning simulation of harvesting operations in equipment clusters, a fundamental flaw exists. In the actual physical process of harvesting operations, the triggering time and location of equipment coordination needs are determined by the real-time integral coupling of the equipment's running trajectory and the spatial distribution of output. There is an inseparable bidirectional dependency between the path execution process and the coordination need triggering process. Because existing software simulation execution engines decouple these two processes during the virtualization modeling stage, they cannot reproduce the dynamic evolution process of the aforementioned circular dependency within the engine. This results in a significant deviation between the simulation execution results and the actual cluster operation process, and the generated planning scheme lacks practical feasibility. Summary of the Invention

[0004] The purpose of this invention is to provide a simulation planning method for equipment cluster operations based on multi-agent collaboration. By constructing upper and lower bound channels for output to form a critical capacity interval, and combining target point correction and gated path replanning, dynamic convergence and optimization of collaborative equipment are achieved. Furthermore, the overall collaborative efficiency is improved through iterative swapping of row attribution indicators and row order.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A simulation planning method for equipment cluster operations based on multi-agent collaboration includes: The simulation model constructs upper and lower bound channels for production density for virtual plot grid cells; after the equipment position is updated, the equipment filling amount is used as the initial value, and a forward cumulative scan is performed on the upper and lower bound channels to determine the boundary points and form the capacity critical threshold interval. The target point is obtained by calculating the midpoint of the path space between the two endpoints based on the capacity critical threshold interval and correcting it after drivability verification. When the offset from the target point of the previous step exceeds the gate threshold, the path replanning of the transfer equipment is triggered. Otherwise, the original path is maintained and the transfer equipment is allocated to gradually converge to the center of the interval. When the equipment enters a new grid, the simulation model generates a measured output value and compares it with the boundary point. If it exceeds the upper limit, the earliest capacity boundary point is moved forward; if it is below the lower limit, the latest capacity boundary point is moved backward. When both the interval length and the rendezvous distance of the transfer equipment are satisfied, the center of the interval is locked as the final rendezvous point and written into the result record. After the simulation is completed, the simulation model reads the result record and calculates two indicators, locking position deviation and cumulative displacement, according to the work row attribution. For rows that exceed the standard, the accessibility of adjacent rows is verified. For those that pass, a row sequence swapping scheme is generated to drive the simulation. If the mean value of the indicators decreases, the swapping is accepted; otherwise, the original parameters are restored. The simulation is iterated until the qualified and / or candidate set is exhausted, and the final result is output.

[0006] Preferably, the simulation model includes plot grid objects, operational agent state objects, collaborative agent state objects, shared state storage and result record objects; In the plot grid object, each grid cell independently stores the upper and lower limits of production density, inherent access attributes, and operational status identifiers; the operational agent status object stores the current simulation coordinates, current equipment fill level, current ordered path sequence, and capacity critical threshold interval; the collaborative agent status object stores the current simulation coordinates, current ordered path sequence, current target point coordinates, and scheduling history; the shared state is stored after each simulation step is completed, receiving the status writes from each agent; the result record object stores the final rendezvous point coordinates of all paired relationship locking events, the actual capacity critical position coordinates, the interval width sequence of each step, the number and type of boundary events, and the pairing number.

[0007] Preferably, the steps for forming the capacity critical threshold interval include: after each step of the simulation completes the job agent position update, extracting the remaining ordered grid sequence from the grid where the current coordinates are located to the end of the path from the current path object, with the current device filling amount as the cumulative initial value; Perform an independent forward scan on the upper boundary channel: sequentially take the product of the upper boundary value of the grid production density and the grid area, and accumulate it to the cumulative amount. When the cumulative amount exceeds the rated capacity of the equipment, record the path number and plot coordinates of the current grid as the earliest full-capacity boundary point. Perform an independent forward scan on the lower boundary channel with the same initial value, and similarly accumulate it with the lower boundary value of the production density. When it exceeds the rated capacity, record the current grid as the latest full-capacity boundary point. When the lower boundary channel is scanned to the end of the path and the rated capacity is still not exceeded, the grid at the end of the path is the default value of the latest full-capacity boundary point. Write the path number, plot coordinates and estimated arrival time of the two points into the capacity critical threshold interval state variable. The two scans are independent of each other and together constitute the capacity critical threshold interval.

[0008] Preferably, the process of obtaining the target point based on the capacity critical threshold interval and the accessibility verification is as follows: Take the arithmetic mean of the earliest full-capacity boundary point path number and the latest full-capacity boundary point path number, read the plot coordinates of the corresponding grid, and obtain the candidate target point coordinates; Construct the current accessible area set: Take the union of the set of grid coordinates marked as true for all grids with work status and the set of grid coordinates marked as true for all grids with inherent accessibility attributes; Query whether the grid corresponding to the candidate target point belongs to the current accessible area set: If it belongs, directly use the candidate coordinates as the target point for this step; If it does not belong, read the coordinates of the nearest end point from the current work path object, start from the grid where the candidate target point is located, and step towards the end point grid by grid along the current work path direction. The first grid coordinate that belongs to the current accessible area set is the corrected target point.

[0009] Preferably, the process of triggering path replanning when the offset of the target point from the previous step exceeds the gating threshold, and otherwise maintaining the original path to drive asymptotic convergence, is as follows: The gating threshold parameter is stored in the simulation parameter object; in each step of the simulation, the coordinates of the target point after the correction of the current step and the coordinates of the target point of the previous step in the state object of the cooperative agent are read, and the Euclidean distance between the two points is calculated to obtain the offset of the target point of the current step; the offset of the target point of the previous step is compared with the gating threshold parameter: if it exceeds the threshold, the path planning submodule is called with the current coordinates of the cooperative agent and the corrected target point as input, the output path sequence is overwritten to the current path object of the cooperative agent, and the timestamp and offset of this update are appended to the current step entry of the scheduling history and marked as updated; if it does not exceed the threshold, the original path remains unchanged, the current step entry of the scheduling history is written with the offset and marked as not updated; in both cases, the cooperative agent performs the current step displacement and updates the coordinates according to the current path object, the slight drift of the center of the interval is filtered by gating and does not trigger replanning, the trend displacement triggers the effective path update, and the overall trajectory forms a continuous asymptotic convergence towards the center of the interval.

[0010] Preferably, the process of generating measured output values ​​and independently correcting boundary points when the device enters a new grid is as follows: When it is detected that the grid coordinates of the current step size of the working agent are different from those of the previous step size, the output space variation parameter in the simulation parameter object is read. Based on the average of the upper and lower bound values ​​of the current new grid, combined with the output space variation parameter, the measured output density value of the grid is generated. The measured output density value is compared with the upper and lower bound values ​​of the grid in the double-boundary matrix, and independent correction is performed at both ends. When the measured value exceeds the upper limit, the excess mass is calculated. The excess mass is divided by the product of the current grid production density upper limit and the grid area to obtain the forward movement step number. The forward movement step number is subtracted from the path number of the earliest full-cap boundary point. The current position path number is used as the lower limit constraint, and the upper limit deviation event is added to the result record. When the measured value is lower than the lower limit, the missing mass is calculated. The backward movement step number is converted. The backward movement step number is added to the path number of the latest full-cap boundary point. When the value exceeds the path end number, the path end number is used as the upper limit constraint, and the lower limit deviation event is added to the result record.

[0011] Preferably, the process of locking the interval center and writing the result record when both the interval length and the rendezvous distance of the transfer equipment are simultaneously satisfied is as follows: In each step of the deduction, the length of the current interval is calculated and compared with the spatial resolution threshold in the locking condition parameter object to obtain the judgment result of condition one; the Euclidean distance between the current coordinates of the cooperative agent and the coordinates of the current target point is calculated and compared with the rendezvous determination radius parameter to obtain the judgment result of condition two. When both conditions are true, locking is triggered: the coordinates of the plot at the center of the current interval are written as the final rendezvous point, the current pairing relationship is marked as locked, interval updates and target point corrections are stopped, and the final rendezvous point is written as a deterministic target to the state objects of the working agent and the cooperative agent respectively. Write the interval width sequence of each step, the number of events exceeding the upper bound, the number of events falling below the lower bound, the coordinates of the final rendezvous point, the simulation timestamp when locking, and the pairing number to the result record object; the subsequent step size directly advances both sides to the final rendezvous point to perform synchronous updates of the grain box status; if only a single condition is met, continue the deduction until both conditions are met simultaneously and / or the path ends.

[0012] Preferably, the process of calculating two indicators by row attribution after the simulation and driving the row order to swap and iterate for re-simulation is as follows: After the simulation, read the result record objects of all pairing relationships. For each pairing event, use the Euclidean distance between the final rendezvous point coordinates and the actual capacity critical position coordinates as the locking position deviation, and use the sum of the target point offsets of all updated step sizes in the scheduling history as the cumulative displacement. Attribute the two values ​​according to the job row number where the working agent is located at the time of locking, calculate the two averages by row, and the rows that exceed the corresponding qualified thresholds are out-of-standard rows, which are sorted in descending order to form a collaborative adaptability evaluation table; for the first out-of-standard row, read... Adjacent row numbers are used to verify the accessibility of the transition path in the inherent access raster layer of the plot. If it is unreachable, it is skipped. The complete parameters for row sequence swapping are generated and written into the re-simulation parameter object, and a complete re-simulation is performed. After the re-simulation, the mean values ​​of two indicators are extracted: if both mean values ​​decrease, the swap is accepted, the new parameters are used as the baseline parameters, and the row is marked as qualified, and the next out-of-standard row is processed. If neither value decreases, the baseline parameters of the previous round are restored, the swapping direction is marked as invalid, and the next reachable candidate is tried. The iteration continues until all out-of-standard rows in the evaluation table are qualified and / or the candidate set is exhausted. The final row sequence parameters are output as a comparison dataset with the collaborative adaptability indicators of each round of simulation.

[0013] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention constructs upper and lower bound channels for production density and forms a capacity critical threshold interval based on dual-channel forward cumulative scanning, enabling a range-based representation of the equipment's full-load position. It can model the equipment loading process with upper and lower bound constraints under conditions of production space uncertainty and fluctuation, giving the full-load position dynamic envelope characteristics, thereby significantly improving the robustness and fault tolerance of capacity prediction. Especially in scenarios with complex plots and non-uniform production distribution, it can effectively avoid path planning distortion caused by local anomalies, improving overall simulation accuracy and stability.

[0014] 2. This invention introduces a target point offset triggering mechanism based on a gated threshold. By comparing the target point offset over consecutive steps with a threshold, adaptive triggering control for path replanning is achieved. The original path is maintained when there are minor fluctuations in the target point, avoiding the computational overhead and path oscillations caused by frequent replanning. When a trend of offset occurs, the path is updated promptly to ensure that the transport equipment always converges towards the effective rendezvous area. This mechanism achieves a balance between path stability and response sensitivity, significantly reducing scheduling system jitter, improving multi-agent collaborative efficiency, and enhancing the system's real-time decision-making capabilities in dynamic environments.

[0015] 3. After simulation, this invention constructs a collaborative adaptability evaluation and row sequence swapping iterative optimization mechanism based on row attribution. It quantitatively evaluates work rows by locking positional deviation and cumulative displacement as dual indicators, and combines reachability constraints for row sequence reconstruction and re-simulation verification. This method can perform reverse optimization of multi-agent collaborative strategies at a global level, achieving continuous improvement of work paths and collaborative strategies. By iteratively selecting effective swapping schemes and converging to the optimal or suboptimal configuration, it significantly improves overall work efficiency and collaborative consistency. Attached Figure Description

[0016] Figure 1 A schematic diagram of the equipment cluster operation simulation planning method based on multi-agent collaboration provided by the present invention; Figure 2 This is a schematic diagram of the target point offset gating replanning logic flow provided by the present invention; Figure 3 This is a schematic diagram of the logical flow of row attribution analysis and row sequence iterative optimization provided by the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention.

[0018] Example 1: Please see Figure 1 This invention provides a simulation planning method for equipment cluster operations based on multi-agent collaboration. The technical solution is as follows: The simulation model constructs upper and lower bound channels for production density using virtual plot grid cells; after equipment position updates, using the equipment filling amount as the initial value, a forward cumulative scan is performed on the upper and lower bound channels to determine boundary points, forming a capacity critical threshold interval; based on the capacity critical threshold interval, the midpoint of the path space between the two endpoints is calculated and corrected through drivability verification to obtain the target point; when the offset from the target point of the previous step exceeds the gating threshold, the replanning of the transfer equipment path is triggered; otherwise, the original path is maintained, and the transfer equipment is allocated to gradually converge to the center of the interval; let... When entering a new grid, the simulation model generates the measured output value and compares it with the boundary point. If it exceeds the upper limit, the earliest capacity boundary point is moved forward; if it is below the lower limit, the latest capacity boundary point is moved backward. When both the interval length and the rendezvous distance of the transfer equipment are satisfied, the interval center is locked as the final rendezvous point and written into the result record. After the simulation ends, the simulation model reads the result record and calculates the two indicators of locked position deviation and cumulative displacement according to the work row attribution. For rows that exceed the standard, the accessibility of adjacent rows is verified. If they pass, a row sequence swapping scheme is generated to drive the simulation. If the mean value of the indicators decreases, the swapping is accepted; otherwise, the original parameters are restored. The iteration continues until the qualified and / or candidate set is exhausted, and the final result is output.

[0019] Furthermore, the simulation model includes plot grid objects, operational agent state objects, collaborative agent state objects, shared state storage and result record objects; In the plot grid object, each grid cell independently stores the upper limit value of production density, the lower limit value of production density, the inherent passage attribute, and the operation status identifier. The inherent passage attribute remains unchanged throughout the simulation process, the operation status identifier is changed from false to true after the operation passes through, and the grid area is written by multiplying the cut width parameter and the travel direction step size parameter. The operational agent state object stores the current simulation coordinates, current equipment fill level, current ordered path sequence, and capacity critical threshold interval; wherein the capacity critical threshold interval includes the path sequence number, plot coordinates, and estimated arrival step number of the earliest and latest full-capacity boundary points; the collaborative agent state object stores the current simulation coordinates, current ordered path sequence, current target point coordinates, and scheduling history, wherein the scheduling history is indexed by step size, and each step size entry includes three fields: whether an update was triggered, target point offset, and update timestamp; The shared state storage receives the state writes from each agent after each step of the deduction and maintains the current passable area set to achieve step-level data synchronization; the result record object stores the final rendezvous point coordinates, actual capacity critical position coordinates, interval width sequence of each step, number and type of boundary events, and pairing number of all pairing relationship locking events.

[0020] Specifically, during the initialization phase, the simulation model reads the input data file, which includes plot boundary parameters, coordinates of each grid cell and their corresponding upper and lower bounds for yield density, descriptions of the plot's inherent traversable area, cut width parameters, and step size parameters for the travel direction. Based on this, the simulation model constructs five core data objects: plot grid objects, operational agent state objects, collaborative agent state objects, shared state storage, and result record objects.

[0021] The plot grid objects are organized as two-dimensional matrices indexed by grid row and column numbers. Each matrix element corresponds to a grid cell, independently storing five attributes: upper limit of yield density, lower limit of yield density, inherent accessibility, operational status flag, and grid area. The inherent accessibility attribute is either true or false. True indicates that the grid corresponds to inherently accessible areas such as field roads, field turning areas, or plot boundary passages within the plot, and remains unchanged throughout the simulation. The operational status flag is initially false, then set to true by the action agent, and remains unchanged thereafter. The grid area is written by multiplying the cutting width parameter by the travel direction step size parameter, and remains unchanged throughout the simulation, serving as the area factor for converting yield density to harvestable grain quality. The upper and lower bounds of yield density are obtained as follows: Before simulation planning, the spatial distribution of yield is predicted for the work plots. Optimistic estimates are written as the upper bounds of each grid, and pessimistic estimates are written as the lower bounds of each grid. The difference between the two bounds reflects the uncertainty of the yield spatial distribution prediction. When historical yield data is sufficient, the upper and lower bounds are respectively obtained by adding or subtracting a deviation covering the main range of variation from the historical mean.

[0022] The operational agent's state object stores four dynamic data items: current simulation coordinates, current grain bin filling level, current ordered path sequence, and full bin location interval. The full bin location interval includes four fields: the earliest full bin boundary point grid coordinates, the latest full bin boundary point grid coordinates, and the estimated arrival simulation step size number corresponding to the two points. The collaborative agent's state object stores four dynamic data items: current simulation coordinates, current ordered path sequence, current scheduling target point coordinates, and scheduling history. The scheduling history is indexed by step size number, and each step size entry includes three fields: whether an update was triggered, the target point offset for this step, and the step size number when the update was triggered. For step size entries that have not triggered an update, the whether an update was triggered flag is false, but the offset field is still written normally.

[0023] The shared state storage receives a complete write of all agent state objects at the end of each simulation step. At the start of the next long simulation step, it provides read interfaces to each functional module. There are no direct calling relationships between modules; all data exchange is relayed through the shared state storage, achieving step-level decoupling and synchronization. The shared state storage also maintains the real-time state of the current passable area set. This set is updated and written after each agent's position is updated by taking the union of the set of grids with a true operational status and the set of grids with true inherent passability attributes.

[0024] The result record object appends a complete lock event record each time a pairing relationship triggers a lock, including the pairing relationship number, the grid coordinates of the final rendezvous point, the grid coordinates of the actual full warehouse position (recorded immediately when the grain bin filling amount of the working agent first exceeds the rated capacity), the lock event step number, the work row number, the width sequence of each step interval, the number of events exceeding the upper boundary and the number of events falling below the lower boundary.

[0025] The path planning submodule receives three inputs: the starting point coordinates, the ending point coordinates, and the current set of passable areas. It constructs a dynamic connectivity graph using the grids in the current set of passable areas as nodes and the connections between adjacent passable grids as edges. It searches for the shortest ordered path from the starting point to the ending point on this graph and outputs the ordered grid coordinate sequence on the path. When the starting point to the ending point is not connected in the current set of passable areas, it returns an unreachable flag. After receiving the unreachable flag, the caller maintains the original path of the cooperative agent and continues to deduce, and writes the unreachable flag into the current step size entry in the scheduling history.

[0026] Furthermore, the process of performing a forward cumulative scan on the upper and lower bound channels to determine the boundary points constituting the capacity critical threshold interval is as follows: After each step of the simulation completes the job agent's position update, the remaining ordered grid sequence from the grid at the current coordinates to the end of the path is extracted from its current path object, with the current device fill amount as the initial cumulative value; Perform an independent forward scan on the upper boundary channel: take the product of the upper boundary value of the production density of each cell and the area of ​​the cell in sequence, and accumulate it to the cumulative amount. When the cumulative amount exceeds the rated capacity of the equipment for the first time, record the path number and plot coordinates of the current cell as the earliest full warehouse boundary point. Perform independent forward scanning of the lower bound channel with the same initial value, and similarly accumulate the production density lower bound value. When the rated capacity is exceeded for the first time, record the current cell as the latest full capacity boundary point. When the lower bound channel is scanned to the end of the path and the rated capacity is not exceeded, use the cell at the end of the path as the default value of the latest full capacity boundary point to ensure the validity of the interval. The path number, plot coordinates, and estimated arrival time of the simulation points are written into the capacity critical threshold interval state variables. The two scans are independent of each other and together constitute the capacity critical threshold interval.

[0027] Specifically, after each simulation completes the update of the agent's position, the simulation controller extracts the remaining ordered grid sequence from the grid at the current coordinates to the end of the path from the agent's current path object, and uses the current device fill amount as the common initial value for the two independent scans.

[0028] The execution process of the upper boundary channel forward cumulative scan is as follows: Starting with the current equipment filling volume as the initial cumulative value, the upper boundary value of the yield density of each corresponding cell is read sequentially according to the path sequence. The upper boundary value of the yield density is multiplied by the grid area to obtain the upper boundary of the harvestable grain mass for that cell. This amount is added to the current cumulative volume. After each accumulation, it is determined whether the cumulative volume exceeds the rated capacity of the grain bin. If it exceeds the rated capacity for the first time, the path cell number and plot coordinates of the current cell are recorded as the earliest full-bin boundary point, and the scan is stopped. The physical meaning of the earliest full-bin boundary point is: under the condition of taking the upper boundary estimate of the yield density, the harvester will fill the grain bin at this position earliest.

[0029] The execution process of the forward cumulative scan of the lower bound channel is the same as that of the upper bound channel. The lower bound value of yield density is read cell by cell with the same initial cumulative value, multiplied by the grid area and accumulated sequentially. When the rated capacity is exceeded for the first time, the current cell is recorded as the latest full-load boundary point. The physical meaning of the latest full-load boundary point is: under the condition of taking the lower bound estimate of yield density, the harvester will fill the grain bin at this position at the latest.

[0030] Boundary Case Handling: When the lower limit of production density is low or the remaining path segment is short, the cumulative amount may still not exceed the rated capacity after the lower limit channel has completely traversed the remaining path sequence. In this case, the path cell number and plot coordinates of the end cell of the path are written as the default value of the latest full-load boundary point into the interval state variable to ensure the validity of the interval; the physical meaning of the default value is that the end of the work row is the latest full-load boundary under the most pessimistic production estimate.

[0031] The estimated arrival step size number is calculated as follows: Using the current step size number as a baseline, the difference between the earliest full-position boundary point path grid number and the current grid path grid number is divided by the standard travel speed of the agent (measured in units of the number of grids traveled per step, kept constant within the same work row). This yields the number of steps required to reach the earliest full-position boundary point. Adding this to the current step size number gives the estimated arrival step size number. The same calculation is performed for the latest full-position boundary point. The estimated arrival step size number is only used for timing reference in result recording and does not participate in the core logic of interval width judgment and lock condition triggering.

[0032] The path grid number of the earliest full-capacity boundary point and the plot coordinates, the path grid number of the latest full-capacity boundary point and the plot coordinates, and the estimated arrival step length numbers of the two points are written into the capacity critical threshold interval state variables of the working agent. The two scans use the same initial value and rated capacity parameters, are independent of each other, and together constitute the capacity critical threshold interval of this step length.

[0033] Furthermore, the process of obtaining the target point based on the capacity critical threshold interval and modified through feasibility verification is as follows: Take the arithmetic mean of the earliest full-position boundary point path number and the latest full-position boundary point path number, read the plot coordinates of the corresponding grid, and obtain the coordinates of the candidate target point. Construct the current passable area set: take the union of the set of all grid coordinates with the operation status marked as true and the set of all grid coordinates with the inherent passability attribute as true, and this union is dynamically expanded with each operation step; Query whether the grid corresponding to the candidate target point belongs to the current passable area set: if it does, the candidate coordinates are directly used as the target point for this step; if it does not, the coordinates of the nearest end point are read from the current work path object, and starting from the grid where the candidate target point is located, the grid is moved one grid at a time along the current work path direction towards the end point. The first grid coordinates that belong to the current passable area set are the corrected target point. Write the target point coordinates into the collaborative agent's state object and synchronize them to the shared state storage.

[0034] Specifically, after each step of the simulation completes the update of the capacity critical threshold interval, the scheduling target point calculation process is executed.

[0035] Candidate scheduling target point generation: Read the path grid number of the earliest full-capacity boundary point and the path grid number of the latest full-capacity boundary point in the current capacity critical threshold interval, take the arithmetic mean of the two numbers, and if the mean is not an integer, round it to the end of the path. Read the plot coordinates of the grid corresponding to the number to obtain the coordinates of the candidate scheduling target point.

[0036] Construction of the current passable area set: Read the coordinates of all grid cells with the operation status marked as true in the plot grid objects to form the operation grid set, read the coordinates of all grid cells with the inherent passability attribute marked as true to form the inherent passability grid set, and take the union of the two sets to obtain the current passable area set. This set is dynamically expanded with each operation step.

[0037] Accessibility verification and forward movement correction: Query whether the grid corresponding to the candidate scheduling target point belongs to the current accessible area set. If it does, directly write the candidate coordinates as the scheduling target point for this step into the cooperative agent state object. If it does not belong, read the coordinates of the end points stored in the current work path object, calculate the number of grids from the grid where the candidate target point is located to the end points on both sides, take the end point on the side with fewer grids as the forward movement direction reference, and step grid by grid from the grid where the candidate target point is located along the current work path direction. The first grid coordinates that belong to the current accessible area set are the corrected scheduling target point, which is written into the cooperative agent state object and synchronized to the shared state storage.

[0038] Furthermore, the process of triggering path replanning when the offset from the target point in the previous step exceeds the gating threshold, and otherwise maintaining the original path to drive asymptotic convergence, is referenced. Figure 2 Specifically: The simulation parameter object stores the gating threshold parameter, which is defined as a multiple of the set distance traveled by the cooperative agent in a single step. In each step of the simulation, the target point coordinates after the current step size correction and the target point coordinates of the previous step size in the state object of the cooperative agent are read, and the Euclidean distance between the two points is calculated to obtain the target point offset of the current step size. The target point offset of the previous step is compared with the gating threshold parameter: if it exceeds the threshold, the path planning submodule is called with the current coordinates of the cooperative agent and the corrected target point as input. The output path sequence is overwritten with the current path object of the cooperative agent, and the timestamp and offset of this update are appended to the current step entry of the scheduling history and marked as updated; if it does not exceed the threshold, the original path remains unchanged, the offset is written to the current step entry of the scheduling history and marked as not updated. In both cases, the collaborative agent performs a step-size displacement and updates the coordinates based on the current path object. Minor drifts at the center of the interval are gated and filtered out, preventing replanning. Trend displacements trigger effective path updates, and the overall trajectory forms a continuous and asymptotic convergence toward the center of the interval.

[0039] Specifically, the gating threshold parameter is defined as a multiple of the maximum travel distance of the cooperative agent within a simulation step, and is stored in the simulation parameter object in units of path grids. The set multiple must simultaneously satisfy two constraints: First, the gating threshold must be higher than the typical drift distance between the center of adjacent steps during the interval narrowing and stabilization phase, to ensure that minor drifts do not trigger replanning. The typical drift distance can be estimated before initialization using the coefficient of variation of plot size and yield density. Second, the gating threshold must be lower than the minimum number of grids the boundary point must move forward or backward to trigger a single boundary correction, to ensure that trend migration can trigger effective path updates. The range of multiples satisfying both constraints is determined by the simulation planner based on the plot parameters during the initialization phase, and the median value within this range is written into the simulation parameter object as the initial configuration, remaining unchanged during simulation execution.

[0040] After each step of the simulation completes the calculation of the scheduling target point, the coordinates of the scheduling target point after the current step correction and the coordinates of the scheduling target point of the previous step in the state object of the cooperative agent are read. The straight-line distance between the two points is calculated in terms of the number of path grids to obtain the target point offset of the current step, and then compared with the gating threshold parameter.

[0041] When the offset exceeds the gating threshold: the path planning submodule is called with the current coordinates of the cooperative agent and the corrected scheduling target point as input. The output ordered path sequence is overwritten to the current path object of the cooperative agent. The step size number and offset of this update are appended to the current step size entry in the scheduling history and the update trigger flag is set to true.

[0042] When the offset does not exceed the gating threshold: Do not update the collaborative agent path, write the current step size entry in the scheduling history to the offset and set the whether to trigger the update flag to false.

[0043] In both cases, the cooperative agent performs a step-size displacement based on the current path object and updates the coordinates to the shared state storage. This results in a continuously asymptotically converging trajectory where subtle drifts at the interval center are filtered out and trend displacements trigger path updates.

[0044] Furthermore, the process of generating measured output values ​​and independently correcting boundary points when the equipment enters a new grid is as follows: When the simulation inference controller detects that the grid coordinates of the current step length of the work agent are different from those of the previous step length, it reads the production space variation parameter in the simulation parameter object, and generates the measured production density value of the grid in the simulation engine based on the average of the upper and lower bound values ​​of the current new grid and the production space variation parameter. The measured yield density value is compared with the upper and lower bounds of the corresponding grid in the double-bounded matrix, and a two-end independent correction is performed: When the measured value exceeds the upper limit, calculate the excess mass (the difference between the measured value and the upper limit value multiplied by the grid area), divide the excess mass by the product of the current grid production density upper limit value and the grid area to get the forward movement step number, subtract the forward movement step number from the path number of the earliest full warehouse boundary point, use the current position path number as the lower limit constraint, and add the upper limit deviation event to the result record. When the measured value is lower than the lower bound, calculate the missing quality (the difference between the lower bound value and the measured value multiplied by the grid area), convert the shift step number, increase the shift step number by the latest full-bound boundary point path number, and when it exceeds the path end number, use the path end number as the upper limit constraint, and add the event of being below the lower bound to the result record. The two ends of the correction are executed independently, each writing the path number and plot coordinates of the corresponding boundary point, without affecting each other.

[0045] Specifically, the spatial variation parameter of yield consists of two parts: the variation amplitude parameter and the out-of-bounds probability parameter, which are stored in the simulation parameter object. The variation amplitude parameter describes the maximum deviation of the yield density from the central baseline value in a single cell, and its value is based on the historical yield variation coefficient of the target plot or similar plots. The out-of-bounds probability parameter describes the expected probability that the measured value exceeds the preset upper limit or falls below the preset lower limit, and is determined by the coverage margin of the upper and lower limit setting method in historical data.

[0046] When the simulation controller detects that the coordinates of the current step size of the work agent are different from the coordinates of the previous step size, it triggers the generation of the measured output value: the average of the upper and lower bounds of the current new grid is used as the center base value, the variation amplitude parameter is multiplied by the center base value as the maximum deviation, and the measured output density value of the current step size is generated evenly within the range of the center base value plus or minus the maximum deviation, and each grid is generated independently.

[0047] The measured yield density value is compared with the upper and lower bounds of the corresponding cell in the double-bounded matrix, and a two-end independent correction is performed: When the measured value exceeds the upper limit, calculate the excess mass (the difference between the measured value and the upper limit value multiplied by the area of ​​the grid cell), divide the excess mass by the product of the upper limit value of the cell's production density and the grid area to get the number of cells to move forward, read the cell number of the current path of the earliest full-capacity boundary point and subtract the number of cells to move forward, use the cell number of the current working agent's cell path as the lower limit constraint, write the corrected number to the earliest full-capacity boundary point, and add the upper limit deviation event to the result record object.

[0048] When the measured value is lower than the lower bound, calculate the missing quality (the difference between the lower bound value and the measured value multiplied by the area of ​​the grid cell), obtain the number of cells to be moved back using the same conversion logic, read the current path cell number of the latest full-cap boundary point and add the number of cells to be moved back, use the path cell number of the end cell of the path as the upper limit constraint, write the corrected number to the latest full-cap boundary point, and add the event of being below the lower bound to the result record object.

[0049] When the measured value is between the upper and lower bounds, neither of the boundary points is corrected. The corrections at both ends are performed independently, each only writing the path grid number and plot coordinates of the corresponding endpoint, without affecting each other.

[0050] The aforementioned number of cells to move forward uses the upper limit of the yield density of the currently entering cell as the conversion base, which is an approximate calculation method. In typical farmland scenarios where the yield density spatial distribution is relatively uniform, the approximate accuracy is acceptable. When the spatial variation coefficient of the plot's yield density is large, an alternative implementation method can be used: starting from the earliest full-grain boundary point and moving towards the current position, the product of the actual upper limit density value and the grid area is read cell by cell, and the excess mass is subtracted cell by cell. When the excess mass is exhausted, the current cell number is used as the corrected boundary point. This method has the same accuracy as the cell-by-cell density value.

[0051] Furthermore, the process of locking the interval center and writing the result record when both conditions of interval length and rendezvous distance of transfer equipment are met simultaneously is as follows: In each step of the deduction, the current interval length (the difference between the path number of the latest full position boundary point and the path number of the earliest full position boundary point) is calculated and compared with the spatial resolution threshold in the locking condition parameter object to obtain the judgment result of condition one; the Euclidean distance between the current coordinates of the cooperative agent and the coordinates of the current target point is calculated and compared with the rendezvous judgment radius parameter to obtain the judgment result of condition two. When both conditions are true, locking is triggered: the coordinates of the plot at the center of the current interval are written as the final rendezvous point, the current pairing relationship is marked as locked, interval updates and target point corrections are stopped, and the final rendezvous point is written as the deterministic navigation target to the state objects of the working agent and the cooperative agent respectively. Write the interval width sequence of each step, the number of events exceeding the upper bound, the number of events falling below the lower bound, the coordinates of the final rendezvous point, the simulation timestamp and pairing number when locking to the result record object; the subsequent step size directly advances both sides to the final rendezvous point to perform synchronous updates of the grain box status; if only a single condition is met, continue the deduction until both conditions are met simultaneously or the path ends.

[0052] Specifically, the spatial resolution threshold is stored in the locking condition parameter object in units of path grids. Physically, it represents the maximum interval width corresponding to the narrowing of the uncertainty range of the capacity critical position to acceptable accuracy. The value is determined based on the following: when the interval width is lower than this threshold, the maximum deviation between the final rendezvous point and the actual full-capacity position does not exceed half the interval width. This deviation should be less than the allowable positioning error for collaborative rendezvous. The allowable positioning error is determined by the plot size and the maneuvering adjustment margin of the collaborative agent, and is written into the locking condition parameter object by the simulation planner based on the plot size parameters during simulation initialization.

[0053] The rendezvous determination radius is stored in the locking condition parameter object in the parcel coordinate system distance unit. Physically, it means that when the distance between the collaborative agent and the scheduling target point approaches within this radius, the agent is considered to have the spatial conditions to receive deterministic navigation commands. The value is determined based on the following criteria: the rendezvous determination radius must be greater than the path planning submodule's path endpoint parking accuracy error and less than the typical drift distance between the center of the interval during the final step size of interval narrowing. During simulation initialization, the simulation planner writes this radius into the locking condition parameter object based on the collaborative agent's mobility parameters.

[0054] After each step of the deduction completes the correction of the two-end boundary points and the update of the position of the cooperative agent: calculate the current interval width (the difference between the path grid number of the latest full-position boundary point and the path grid number of the earliest full-position boundary point) and compare it with the spatial resolution threshold. If the width is lower than the threshold, then condition one is true; calculate the straight-line distance between the current coordinates of the cooperative agent and the coordinates of the scheduling target point and compare it with the rendezvous determination radius. If the distance is lower than the radius, then condition two is true.

[0055] When both conditions are true, locking is triggered: the coordinates of the current interval center grid are written as the final rendezvous point; the current pairing relationship is marked as locked; interval updates and scheduling target point corrections are stopped; the final rendezvous point is written as a deterministic navigation target to the state objects of the working agent and the cooperative agent respectively; a complete locking event record is added to the result record object; and the subsequent step size advances both parties to the final rendezvous point to perform a synchronized update of the grain bin status.

[0056] If only condition one is true: Freeze the current interval center as the scheduling target point, stop the interval forward cumulative scan update, but continue to execute the generation of measured output value and double-end boundary correction. In the boundary correction, the lower limit constraint of the earliest full-capacity boundary point is still based on the grid number of the current working agent's path. The corrected data is written to the result record, waiting for condition two to be satisfied. If only condition two is true: Keep the collaborative agent moving along the current path, continue to execute interval narrowing related calculations, waiting for condition one to be satisfied. Both conditions are indispensable; single-condition triggering of locking is not allowed.

[0057] Furthermore, the process of calculating two indicators by row attribution after the simulation and driving iterative resimulation by swapping row order is referenced. Figure 3 Specifically: After the simulation, the post-processing module reads the result record objects of all pairing relationships. For each pairing event, the Euclidean distance between the final rendezvous point coordinates and the actual capacity critical position coordinates is used as the locking position deviation. The cumulative displacement is calculated by summing the target point offsets of all updated step sizes in the cooperative agent scheduling history. The two values ​​are attributed according to the job row number where the working agent is located when locking. The two averages are calculated by row. The rows where both averages exceed the corresponding qualified thresholds are marked as out-of-standard rows. The rows are sorted in descending order to form a cooperative adaptability evaluation table. For the first row that exceeds the standard, read the adjacent row number, verify the reachability of the transition path in the inherent access raster layer of the plot, skip if it is not reachable, and generate row sequence swapping complete parameters for reachable rows and write them into the re-simulation parameter object to drive the simulation execution engine to perform a complete re-simulation. After resimulation, extract the average values ​​of the two indicators for that row: if both average values ​​decrease, accept the swap, use the new parameters as the baseline parameters, mark the row as qualified, and continue processing the next row that exceeds the standard; if neither decreases, restore the baseline parameters of the previous wheel, mark the swap direction as invalid, and continue to try the next reachable candidate. Iterate until all rows in the evaluation table that exceed the standard are deemed qualified or the candidate set is exhausted, and output the final row order parameters and the dataset comparing the collaborative adaptability index of each round of simulation.

[0058] Specifically, after the entire simulation is completed, the post-processing module reads all the result record objects of the pairing relationships, performs two index calculations and iterative resimulation with row order swapping.

[0059] Lock position deviation calculation: For each lock event record, read the coordinates of the final rendezvous point and the coordinates of the actual capacity critical position, calculate the straight-line distance between the two points in the plot coordinate system, obtain the lock position deviation value of this pairing event, and associate it with the work row number at the time of locking.

[0060] Cumulative Displacement Calculation: For each locking event record, retrieve all step size entries for the entire pairing relationship in the collaborative agent scheduling history based on the pairing number. Sum the target point offset values ​​of all step size entries whose update trigger flag is true to obtain the total cumulative displacement of the pairing relationship and associate it with the job line number at the time of locking.

[0061] The qualified threshold parameters are written during simulation initialization: the lock position deviation qualified threshold corresponds to the maximum allowable rendezvous positioning error, expressed as an integer multiple of the typical work row width of the plot; the scheduling cumulative displacement qualified threshold corresponds to the upper limit of the additional empty driving distance acceptable to the cooperative agent, expressed as a certain proportion of the typical side length of the plot. Both thresholds are expressed in plot coordinate system distance units or path grids, stored independently in the simulation parameter object, and are not determined by the simulation results.

[0062] Average by row: Using the job row number as the grouping key, calculate the average locking position deviation and the average cumulative scheduling displacement of all locking events in the same job row, and compare them with their respective qualified thresholds. If both averages exceed the standard, the row is marked as exceeding the standard. The rows are sorted in descending order of the sum of the two averages to form a collaborative adaptability evaluation table.

[0063] Row order swapping iteration: Take the top-ranked out-of-standard row in the evaluation table, read the numbers of its two adjacent rows, and verify the reachability of the operation agent from the end cell of the out-of-standard row path to the beginning cell of the adjacent row path in the inherent access grid layer of the plot using the path planning submodule. If it is unreachable, skip it; if it is reachable, generate complete row order swapping parameters and write them into the re-simulation parameter object, and call the simulation execution interface to perform a complete re-simulation. After the re-simulation, extract the mean values ​​of two indicators for that row: if both mean values ​​decrease, accept the swap, use the new parameters as the baseline parameters, and mark the row as qualified to continue processing the next out-of-standard row; if neither decreases, restore the baseline parameters of the previous round, mark the swapping direction as invalid, and continue trying the next candidate; if all candidates are exhausted, mark the candidate set for that row as exhausted and skip it. Iterate until all out-of-standard rows are qualified or the candidate set is exhausted, and output the final row order parameters and the comparison dataset of the collaborative adaptability indicators of each round of simulation.

[0064] Example 2: This embodiment, based on the basic simulation process, data object definition, and gated path replanning mechanism described in Embodiment 1, further expands the interval center drift prediction and multi-machine resource arbitration mechanism. This embodiment achieves greater adaptability to complex operating environments by adding parameters such as drift rate calculation window length, compensation amount upper limit, and spatial conflict determination distance to the simulation parameter object, while maintaining the same basic data flow as Embodiment 1.

[0065] The simulation model constructs upper and lower bound channels for yield density based on virtual plot grid cells. In each step of the simulation, it performs forward cumulative scanning, target point accessibility correction, dual-condition locking, and row order swapping iteration. The simulation parameter object stores gating threshold parameters. In each step, the target point offset is calculated; if the offset exceeds the gating threshold, path replanning of the collaborative agent is triggered; otherwise, the original path is maintained and converged asymptotically. The above process is the pre-execution step of the advance compensation mechanism in this embodiment. After accessibility verification, the compensated scheduling target point is replaced by the gate offset comparison logic; the two are sequentially connected, and the interface is the corrected scheduling target point coordinates.

[0066] Before triggering route replanning for transfer equipment or maintaining the original route, the simulation model also performs the steps of predicting the interval center drift rate and generating the advance compensation scheduling target point, specifically: The simulation parameter object stores the drift rate calculation window length parameter, which is in units of step size; After each step of the simulation completes the verification and correction of the drivability of the scheduling target point, the historical coordinates of the scheduling target points in the most recent consecutive steps of the current pairing relationship are read from the shared state storage. The component displacements of the scheduling target points in the row and column directions between adjacent steps are calculated in order of step number. The arithmetic mean of the component displacements in all directions of all adjacent steps is combined into a drift rate vector, which is stored in the intermediate calculation variable of the current step in units of path grids per step. Using the number of path grids from the current coordinates of the cooperative agent to the current corrected scheduling target point as the numerator and the number of grids traveled by the cooperative agent in a single step as the denominator, calculate the estimated number of steps required for the cooperative agent to reach the current scheduling target point. Multiply the estimated number of steps by the row and column components of the drift rate vector to obtain the row and column advance amounts, respectively. Add the two to the corresponding coordinate components of the current corrected scheduling target point to obtain the coordinates of the compensated scheduling target point. The compensation scheduling target point is executed, its accessibility is verified and shifted, and the corrected compensation scheduling target point replaces the current corrected scheduling target point in the gating offset comparison and path replanning call; after the lock event is triggered, the advance compensation calculation is stopped, and the subsequent step size is directly executed with the final rendezvous point as the deterministic target.

[0067] When performing advance compensation, boundary constraints must be followed: if the number of steps already performed is insufficient, the actual step size shall be used instead of the window length; if the straight-line distance between the compensated coordinates and the target point before correction exceeds the upper limit parameter of the compensation amount (e.g., exceeding half the row width), the offset vector shall be truncated proportionally to prevent the compensation point from jumping due to sudden changes in production.

[0068] Specifically, the drift rate calculation window length parameter is stored in the simulation parameter object in units of step sizes, and its value is no less than two step sizes. The value is determined by dividing the typical work row length by the single-step walking distance of the work agent to obtain the total number of single-row extrapolation steps. One-tenth of this total is written as the window length parameter into the simulation parameter object. This value corresponds to the sampling density of approximately ten independent displacement estimates completed by the center of the interval within a single-row extrapolation cycle. In typical farmland scenarios where the spatial variation coefficient of yield density does not exceed 30%, the drift rate estimation variance is acceptable.

[0069] After each step of the simulation completes the verification and correction of the drivability of the scheduling target point, the historical coordinates of the scheduling target points within the most recent consecutive step lengths of the current pairing relationship are read from the shared state storage. The number of records is based on the window length parameter. The coordinates of the scheduling target points of adjacent step lengths are read in order of step length number. The differences in row and column coordinates are calculated separately. The arithmetic mean of the row and column differences for all adjacent step length pairs is taken. These two directional averages are used as the row and column components of the drift rate vector, respectively. They are stored in the intermediate calculation variable of the current step length per step length, with the number of path grids as the unit. The two directional components are calculated independently and do not interfere with each other.

[0070] The estimated step size is calculated as follows: The current path object of the cooperative agent is read, and the number of path cells from the current coordinate cell to the current corrected scheduling target cell is counted. This number of path cells is divided by the standard number of cells traveled per step by the cooperative agent, and the result is rounded to obtain the estimated step size. This estimated step size is a static approximation, assuming that the interval center remains stationary during the estimation period. Its accuracy is highest when the interval center drift stabilizes at the end of the interval narrowing phase. In the initial stage of simulation, when interval migration is drastic, the estimation deviation is large. Combining this with the upper limit parameter of the compensation amount can limit compensation runaway caused by excessive deviation. The estimated step size is multiplied by the row and column direction components of the drift rate vector to obtain the row and column direction advance amounts. The row direction advance amount is superimposed on the row coordinates of the current corrected scheduling target point, and the column direction advance amount is superimposed on the column coordinates to obtain the coordinates of the compensated scheduling target point.

[0071] After the coordinates of the compensation scheduling target point are determined, accessibility verification and forward correction are performed: The grid corresponding to the compensation target point is checked to see if it belongs to the current accessible area set. If it does, it is used directly; otherwise, the grid is moved one grid at a time along the work path towards the nearest endpoint to the first accessible grid, and the coordinates of that grid are used as the corrected compensation scheduling target point. The corrected compensation scheduling target point replaces the current corrected scheduling target point and participates in gating offset calculation and path replanning. After a lock event is triggered, compensation calculation stops, and subsequent steps are executed with the final rendezvous point as the deterministic navigation target.

[0072] The calculation of the interval center step size level drift rate and lead time also includes the following boundary constraint handling: When the number of steps executed in the current pairing relationship is lower than the drift rate calculation window length parameter, the drift rate calculation is performed using the actual number of steps executed instead of the window length parameter, ensuring that the drift rate calculation within the step size in the initial stage of pairing relationship scheduling does not fail due to insufficient historical data; When the calculated interval center step-level drift rate is zero, the advance is zero, the compensation scheduling target point is the same as the current corrected scheduling target point coordinates, and the current step-level gating judgment is directly based on the current corrected scheduling target point. When the distance between the corrected coordinates of the compensation scheduling target point after the accessibility verification and the current corrected coordinates of the scheduling target point exceeds the preset compensation amount upper limit parameter, the lead amount is truncated by the coordinate offset corresponding to the compensation amount upper limit parameter to prevent the compensation target point from shifting to outside the range of the current pairing operation row when the center drift rate of the interval is abnormally large; the compensation amount upper limit parameter is represented by an integer multiple of the typical row width of the plot, and is written into the simulation parameter object by the simulation planner during simulation initialization.

[0073] Specifically, when the number of steps executed in the current pairing relationship is lower than the window length parameter, the drift rate calculation is performed using the actual number of steps executed instead of the window length parameter, and only the accumulated historical coordinate records are used to ensure that the compensation step is not skipped in the initial stage of pairing relationship scheduling due to insufficient historical data.

[0074] When both the row and column components of the calculated drift rate vector are zero, the advance value is zero, the coordinates of the compensation scheduling target point are the same as the current corrected scheduling target point, and the current step size gating judgment is directly based on the current corrected scheduling target point without introducing additional offset.

[0075] The upper limit parameter for compensation is represented as an integer multiple of the typical row width of the plot and is written into the simulation parameter object during simulation initialization. The value is determined as follows: half the distance between the centerlines of any two adjacent work rows of the plot is used as the reference upper limit. Within this reference upper limit, the maximum value among integer multiples of the row width is written, and this value is verified to be greater than the product of the maximum expected value of the spatial variation coefficient of production density, obtained by multiplying the maximum displacement of the interval center within the window length parameter step by the estimated step size. If this is not satisfied, the reference upper limit is increased to a nearby larger integer multiple of the row width for re-verification. When the straight-line distance between the corrected coordinates of the compensation scheduling target point after drivability verification and the current corrected scheduling target point exceeds the upper limit parameter for compensation, the upper limit parameter is used as the upper limit of the modulus. The modulus of the offset vector composed of the row direction advance and the column direction advance is truncated to the upper limit of compensation, maintaining the proportional relationship between the two components (i.e., each component is multiplied by the ratio of the upper limit of compensation to the original modulus). The truncated two components are superimposed on the corrected scheduling target point coordinates to obtain the final compensation target point. After performing drivability verification again, it participates in subsequent gating judgments.

[0076] In a multi-harvester agent collaborative scenario where multiple pairings share a passable area, after each simulation step completes the update of the full-load position intervals for all active pairings, the simulation model also performs the following interval spatial overlap detection and dynamic arbitration of scheduling resources: For all active pairings, calculate the Euclidean distance between the center plot coordinates of the full-position intervals of each pair and compare it with the spatial conflict determination distance parameter stored in the simulation parameter object. The spatial conflict determination distance parameter is represented as an integer multiple of the typical channel width of the plot and is written into the simulation parameter object during simulation initialization. When the distance between the center of any two active pairings is less than the spatial conflict determination distance, the absolute value of the width change of each full-position interval within the most recent consecutive step size calculated based on the drift rate window length parameter is used as the narrowing rate index. When the number of steps executed for the current pairing is less than the window length parameter, the actual number of steps executed is used instead. Pairings with a larger narrowing rate index are marked as scheduling priority pairings, and pairings with a smaller narrowing rate index are marked as deferred pairings. When the narrowing rate indices of two pairings are equal, the pairing with the smaller pairing number is given priority. Collaborative agents with priority scheduling relationships will perform gating judgment and path replanning normally; collaborative agents with deferred pairing relationships will maintain the original path and perform displacement in this step, without performing scheduling target point update and gating judgment, and the current step entry in the scheduling history will be written with a deferred flag. When the center distance between two pairing intervals exceeds the spatial conflict determination distance again, or after the priority pairing relationship is locked, the temporary state is automatically released; the temporary flag of the temporary step is written into the result record, and the target point offset of the temporary step is not included in the total cumulative displacement of the scheduling when the post-processing module calculates the cumulative displacement.

[0077] Specifically, the spatial conflict determination distance parameter is represented as an integer multiple of the typical channel width of the plot and is written into the simulation parameter object during simulation initialization. The value is determined as follows: the center-line distance between any two adjacent work rows is read from the plot initialization parameters. One-third of this distance's lower limit is used as the lower limit of the parameter, and half of this lower limit is used as the upper limit. The lower limit value within this range is then written into the simulation parameter object. This value ensures that conflict detection is triggered only when the center-to-center distance between two paired intervals is less than half the single-row spacing, without interfering with normally distributed paired relationships.

[0078] After each step of the simulation completes the update of the full-position intervals of all active pairings, the Euclidean distance between the coordinates of the center plots of each pair of active pairings is calculated and compared with the spatial conflict determination distance parameter. When the interval center distance between any two active pairings is lower than this parameter, the narrowing rate index of each pairing is calculated: the interval width sequence of each pairing is read within the most recent consecutive step (the step size is based on the window length parameter; if the executed step size is insufficient, the actual executed step size is used instead), and the average absolute value of the difference between the interval widths of adjacent step sizes is calculated as the narrowing rate index of that pairing. When the interval width of a pairing is lower than the spatial resolution threshold in the locking condition parameter object but the locking has not yet been triggered, the pairing is given the highest priority because its interval width is lower than the spatial resolution threshold. It is not included in the narrowing rate index comparison and is directly marked as a scheduling priority pairing. The remaining pairings are then sorted according to the narrowing rate index. For pairing relationships involved in the sorting, the pairing relationship with the larger narrowing rate index is marked as the scheduling priority pairing relationship, and the pairing relationship with the smaller narrowing rate index is marked as the postponed pairing relationship; when the narrowing rate indices of the two are equal, the pairing relationship with the smaller index is the scheduling priority pairing relationship.

[0079] Cooperative agents with priority scheduling relationships execute compensation scheduling target point generation, gating judgment, and path replanning according to the normal process. Cooperative agents with deferred pairing relationships maintain the original path and execute displacement within the current step, without updating scheduling target points, calculating compensation, or performing gating judgment. The current step entry in the scheduling history is written with a deferred flag, and the target point offset field is written with a negative special flag value. When the center distance between the two pairing relationships exceeds the spatial conflict judgment distance again, or when the priority scheduling pairing relationship is locked, the deferred state is automatically released, and the deferred pairing relationship resumes the normal scheduling process from the next step. When calculating the cumulative displacement of deferred pairing relationships, the post-processing module skips all step entries containing deferred flags or target point offset fields with negative special flag values. It only sums the target point offsets of step entries with true update flags and no deferred flags to obtain the effective cumulative displacement, ensuring that the row attribution index calculation results are not affected by the forced deferred step size.

[0080] In summary, this invention, by constructing dual-channel capacity constraint modeling and a row attribution and row sequence swapping mechanism, can effectively narrow the uncertainty range of the capacity critical position in environments with uneven production space. Compared with single-channel planning, this method significantly improves the spatial consistency between the final rendezvous point and the actual full-capacity position of the equipment, reduces path oscillations and ineffective empty runs during the coordination process, thereby improving simulation accuracy and the operational efficiency of the actual scheme.

[0081] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A device cluster job simulation planning method based on multi-agent collaboration, characterized in that, include: The simulation model constructs upper and lower bound channels for yield density for virtual plot grid cells; After the device location is updated, the device filling amount is used as the initial value. A forward cumulative scan is performed on the upper and lower bound channels to determine the boundary points and form the capacity critical threshold range. The steps to form the capacity critical threshold range include: after each step of the simulation completes the job agent position update, extracting the remaining ordered grid sequence from the grid where the current coordinates are located to the end of the path from the current path object, with the current device filling amount as the cumulative initial value; Perform an independent forward scan on the upper boundary channel: sequentially take the product of the upper boundary value of the grid production density and the grid area, and accumulate it to the cumulative amount. When the cumulative amount exceeds the rated capacity of the equipment, record the path number and plot coordinates of the current grid as the earliest full capacity boundary point. Perform an independent forward scan on the lower boundary channel with the same initial value, and similarly accumulate it with the lower boundary value of the production density. When it exceeds the rated capacity, record the current grid as the latest full capacity boundary point. When the lower boundary channel is scanned to the end of the path and still has not exceeded the rated capacity, the grid at the end of the path is the default value of the latest full capacity boundary point. Write the path number, plot coordinates and estimated arrival time of the two points into the capacity critical threshold interval state variable. The two scans are independent of each other and together constitute the capacity critical threshold interval. The target point is obtained by calculating the midpoint of the path space between the two endpoints based on the capacity critical threshold interval and correcting it after drivability verification. When the offset from the target point of the previous step exceeds the gate threshold, the path replanning of the transfer equipment is triggered. Otherwise, the original path is maintained and the transfer equipment is allocated to gradually converge to the center of the interval. When the equipment enters a new grid, the simulation model generates the measured output value and compares it with the boundary point. If it exceeds the upper limit, the earliest full-load boundary point is moved forward; if it is below the lower limit, the latest full-load boundary point is moved backward. When both the interval length and the rendezvous distance of the transfer equipment are satisfied, the center of the interval is locked as the final rendezvous point and written into the result record. After the simulation, the simulation model reads the result records and calculates two indicators, locking position deviation and cumulative displacement, based on the job row attribution. For rows exceeding the standard, it verifies the reachability of adjacent rows to switch to other rows. If they pass, it generates a row order swapping scheme to drive the simulation. If the mean of the indicators decreases, the swapping is accepted; otherwise, the original parameters are restored. The simulation iterates until the qualified and / or candidate set is exhausted, and then outputs the final result. The specific process of calculating the two indicators based on the row attribution and driving the row order swapping iteration and resimulation after the simulation is as follows: After the simulation, it reads the result record objects of all pairing relationships. For each pairing event, it uses the Euclidean distance between the final rendezvous point coordinates and the actual capacity critical position coordinates as the locking position deviation, and the sum of the target point offsets of all updated step sizes in the scheduling history as the cumulative displacement. The two values ​​are then combined according to the job row where the job agent is located at the time of locking. The algorithm assigns the cause by number, calculates the mean of two values ​​for each row, and identifies rows exceeding the corresponding pass threshold as out-of-range rows. These rows are then sorted in descending order to form a collaborative adaptability evaluation table. For the first out-of-range row, the adjacent row numbers are read, and the accessibility of the transition path is verified in the plot's inherent access raster layer. If the path is unreachable, it is skipped, and the complete parameters for row order swapping are generated and written into the resimulation parameter object. A complete resimulation is then performed. After the resimulation, the mean of two indicators is extracted: if both mean values ​​decrease, the swap is accepted, the new parameters are used as the baseline parameters, and the row is marked as passable, and the next out-of-range row is processed. If neither value decreases, the baseline parameters from the previous round are restored, the swap direction is marked as invalid, and the next reachable candidate is tried. The iteration continues until all out-of-range rows in the evaluation table are considered passable and / or the candidate set is exhausted. Finally, the row order parameters are output as a comparison dataset with the collaborative adaptability indicators from each round of simulation.

2. The method of claim 1, wherein the method is characterized by: The simulation model includes plot grid objects, operational agent state objects, collaborative agent state objects, shared state storage and result record objects; In the plot grid object, the grid cell independently stores the upper limit value of production density, the lower limit value of production density, the inherent access attributes, and the operation status identifier; the operation agent status object stores the current simulation coordinates, the current equipment filling amount, the current ordered path sequence, and the capacity critical threshold range. The collaborative agent state object stores the current simulation coordinates, the current ordered path sequence, the current target point coordinates, and the scheduling history; the shared state is stored after each simulation step is completed, receiving the state writes from each agent; the result record object stores the final rendezvous point coordinates of all pairing relationship locking events, the actual capacity critical position coordinates, the interval width sequence of each step, the number and type of boundary events, and the pairing number.

3. The method of claim 1, wherein the method is characterized by: The process of obtaining target points based on capacity critical threshold intervals and passability verification is as follows: take the arithmetic mean of the path number of the earliest full-capacity boundary point and the path number of the latest full-capacity boundary point, read the plot coordinates of the corresponding grid, and obtain the coordinates of the candidate target points; construct the current passable area set: take the union of the set of grid coordinates marked as true for all grids with working status and the set of grid coordinates marked as true for all grids with inherent passability attributes. Query whether the grid corresponding to the candidate target point belongs to the current passable area set: if it does, directly use the candidate coordinates as the target point for this step; If it does not belong to the target area, the coordinates of the nearest end point are read from the current work path object. Starting from the grid cell where the candidate target point is located, the target point is moved grid by grid along the current work path towards the end point. The first grid cell coordinates that belong to the current passable area set are the corrected target point.

4. The method of claim 1, wherein the method further comprises: The process of triggering path replanning when the offset of the target point from the previous step exceeds the gating threshold, and otherwise maintaining the original path to drive asymptotic convergence, is as follows: The gating threshold parameter is stored in the simulation parameter object; in each step of the simulation, the coordinates of the target point after the correction of the current step and the coordinates of the target point of the previous step in the cooperative agent's state object are read, and the Euclidean distance between the two points is calculated to obtain the offset of the target point of the current step; the offset of the target point of the previous step is compared with the gating threshold parameter: if it exceeds the threshold, the path planning submodule is called with the current coordinates of the cooperative agent and the corrected target point as input, the output path sequence is overwritten to the current path object of the cooperative agent, and the timestamp and offset of this update are appended to the current step entry in the scheduling history and marked as updated; if it does not exceed the threshold, the original path remains unchanged, the current step entry in the scheduling history is written with the offset and marked as not updated; in both cases, the cooperative agent performs the current step displacement and updates the coordinates according to the current path object. The slight drift of the center of the interval is filtered by gating and does not trigger replanning, while the trend displacement triggers the effective path update, and the overall trajectory forms a continuous asymptotic convergence towards the center of the interval.

5. The equipment cluster operation simulation planning method based on multi-agent collaboration according to claim 1, characterized in that: The process of generating measured output values ​​and independently correcting boundary points when the equipment enters a new grid is as follows: When it is detected that the grid coordinates of the current step size of the work agent are different from those of the previous step size, the output space variation parameter in the simulation parameter object is read. Based on the average of the upper and lower bound values ​​of the current new grid, combined with the output space variation parameter, the measured output density value of the grid is generated. The measured output density value is compared with the upper and lower bound values ​​of the grid in the double-boundary matrix, and independent correction is performed on both ends. When the measured value exceeds the upper limit, the excess mass is calculated. The excess mass is divided by the product of the current grid production density upper limit and the grid area to obtain the forward movement step number. The forward movement step number is subtracted from the path number of the earliest full-cap boundary point. The current position path number is used as the lower limit constraint, and the upper limit deviation event is added to the result record. When the measured value is lower than the lower limit, the missing mass is calculated. The backward movement step number is converted. The backward movement step number is added to the path number of the latest full-cap boundary point. When the value exceeds the path end number, the path end number is used as the upper limit constraint, and the lower limit deviation event is added to the result record.

6. The equipment cluster operation simulation planning method based on multi-agent collaboration according to claim 1, characterized in that: The process of locking the interval center and writing the result record when both the interval length and the rendezvous distance of the transfer equipment are met simultaneously is as follows: In each step of the deduction, the length of the current interval is calculated and compared with the spatial resolution threshold in the locking condition parameter object to obtain the judgment result of condition one; the Euclidean distance between the current coordinates of the cooperative agent and the coordinates of the current target point is calculated and compared with the rendezvous determination radius parameter to obtain the judgment result of condition two. When both conditions are true, locking is triggered: the coordinates of the plot at the center of the current interval are written as the final rendezvous point, the current pairing relationship is marked as locked, interval updates and target point corrections are stopped, and the final rendezvous point is written as a deterministic target to the state objects of the working agent and the cooperative agent respectively. Write the interval width sequence of each step, the number of events exceeding the upper bound, the number of events falling below the lower bound, the coordinates of the final rendezvous point, the simulation timestamp when locking, and the pairing number to the result record object; the subsequent step size directly advances both sides to the final rendezvous point to perform synchronous updates of the grain box status; if only a single condition is met, continue the deduction until both conditions are met simultaneously and / or the path ends.

Images