Method and system for scheduling simulation of agv

Abstract

The application provides an AGV automatic carrying trolley scheduling simulation method and system, relates to the field of AGV automatic carrying trolleys, and comprises the following steps: marking the AGV position to generate an initial path network; extracting a stay event to construct an avoidance channel network; recording a passing time sequence to generate a batch scheduling sequence; and based on a node capacity matrix and a traffic resource distribution diagram, when a task interruption is detected, a capacity-matched standby channel is selected, and the interrupted task is split into multiple subtask packages for time-sharing scheduling. The application can improve the AGV scheduling efficiency, reduce the influence of task interruption, and optimize the resource utilization rate.

Description

Technical Field

[0001] The present invention relates to AGV (Automated Guided Vehicle) (AGV) automatic transport vehicle technology, and in particular to an AGV (Automated Guided Vehicle) automatic transport vehicle scheduling simulation method and system. Background Art

[0002] With the development of intelligent industrial manufacturing, automated guided vehicles (AGVs) are widely used in factory logistics systems. Existing AGV scheduling systems primarily rely on fixed path planning, which is unable to dynamically adjust scheduling strategies based on actual operating conditions. This results in poor system adaptability and easily causes channel congestion when tasks are interrupted.

[0003] Current AGV scheduling simulation methods typically only consider static path planning, ignoring the avoidance requirements of AGVs during actual operation and lacking a reasonable mechanism for evaluating node capacity. This makes it difficult to accurately predict and handle interference between AGVs under complex operating conditions, impacting scheduling efficiency and system stability.

[0004] Existing technologies for handling AGV interruptions rely primarily on simple waiting or rerouting, failing to fully utilize available network resources and failing to consider the impact of node capacity on scheduling decisions. Therefore, a scheduling simulation method for AGVs is urgently needed that can construct an avoidance channel network based on actual operational data and perform time-sharing scheduling based on node capacity characteristics. Summary of the Invention

[0005] The embodiments of the present invention provide an AGV (Automated Guided Vehicle) scheduling simulation method and system, which can solve the problems in the prior art.

[0006] A first aspect of an embodiment of the present invention provides an AGV (Automated Guided Vehicle) scheduling simulation method, comprising:

[0007] Mark the location information of the AGV automatic transport vehicle in the path map to generate the initial path network;

[0008] The AGV's trajectory in the initial path network is collected, and dwell events are extracted. Locations where the dwell time exceeds a preset time threshold are marked as avoidance nodes. Buffer areas are constructed with the avoidance nodes as the center, and adjacent buffer areas are connected to form an avoidance channel network.

[0009] The passing sequence of AGVs is recorded in the avoidance channel network. The time interval between adjacent avoidance nodes is used as a time window. The passing batches are divided according to the distribution pattern of the time window to generate a batch scheduling sequence.

[0010] A node capacity matrix is ​​constructed based on the dwell time of the avoidance node, and the batch scheduling sequence is mapped into a traffic resource distribution map. When the task interruption signal of the AGV is detected, the node capacity value of the interruption location is extracted, and a backup channel with matching capacity is selected from the traffic resource distribution map. The interruption task is split into multiple subtask packages according to the node capacity threshold, and time-sharing scheduling is performed through the backup channel.

[0011] In an optional embodiment,

[0012] Mark the location information of the AGV automatic transport vehicle on the path map to generate the initial path network including:

[0013] Collect laser scanning positioning coordinates and ultra-wideband positioning coordinates, calculate the corresponding weight coefficient based on the laser scanning error value and the ultra-wideband error value, multiply the weight coefficient with the corresponding positioning coordinates and superimpose them to obtain fused position information;

[0014] Collect the fused position information of the AGV automatic transport vehicle and calculate the position change between adjacent sampling periods. When the position change is greater than the preset change threshold, determine the main positioning source based on the change direction of the historical fused position information, use the normal positioning information as position compensation information, and update the fused position information based on the position compensation information;

[0015] Based on the updated fusion position information, a position tracking sequence is constructed, and the position information of the stopping points in the AGV automatic transport vehicle's running path is extracted as the reference point for path construction;

[0016] Calculate the distance and heading angle between the reference points of the path construction, select the reference points that meet the preset distance threshold and the preset angle threshold to establish a connection relationship, use the minimum spanning tree algorithm to traverse and connect the reference points, and generate an initial path network with directional attributes.

[0017] In an optional embodiment,

[0018] The AGV's trajectory in the initial path network is collected, and dwell events are extracted. Locations where the dwell time exceeds a preset time threshold are marked as avoidance nodes. Buffer areas are constructed with the avoidance nodes as the center, and adjacent buffer areas are connected to form an avoidance channel network, including:

[0019] Collect the running trajectory of the AGV automatic transport vehicle in the initial path network, the running trajectory including laser positioning data and ultra-wideband positioning data, perform intelligent fusion processing on the running trajectory, and generate fused trajectory data;

[0020] Based on the fused trajectory data, the stop events of the AGV automatic transport vehicle are identified, the location information and duration of the stop events are extracted, and the stop events are divided into avoidance waiting type and task loading and unloading type;

[0021] Based on the classification results of the dwell events, the avoidance necessity score of the dwell location is calculated. The dynamic score threshold is determined based on the system operation status. The dwell locations with avoidance necessity scores higher than the dynamic score threshold are marked as candidate avoidance nodes. When the distance between adjacent candidate avoidance nodes is less than twice the length of the AGV, they are merged into one avoidance node to obtain the final avoidance node.

[0022] The historical trajectory data of the AGV at the avoidance node location is extracted, and the trajectory density clustering method is used to analyze the avoidance behavior characteristics of the AGV. Based on the avoidance behavior characteristics, the main avoidance direction and avoidance range of each avoidance node are determined.

[0023] An adaptive region growing algorithm is used to construct a buffer area along the main avoidance direction of the avoidance node. The positional relationship and avoidance direction of adjacent avoidance nodes are analyzed, and the trajectory density maximum path algorithm is used to connect adjacent buffer areas to form an avoidance channel network.

[0024] In an optional embodiment,

[0025] Along the main avoidance direction of the avoidance node, the adaptive region growing algorithm is used to construct the buffer area. The position relationship and avoidance direction of adjacent avoidance nodes are analyzed, and the trajectory density maximum path algorithm is used to connect adjacent buffer areas to form an avoidance channel network including:

[0026] The historical avoidance trajectories at the avoidance nodes are vertically projected onto a plane perpendicular to the main avoidance direction to obtain the lateral offset distance of the trajectory points. The kernel density estimation method is used to calculate the probability density distribution of the lateral offset distance to obtain the lateral distribution characteristics of the avoidance trajectory.

[0027] Along the main avoidance direction of the avoidance node, an adaptive threshold function is set based on the lateral distribution characteristics, and a region growing algorithm is used to construct a buffer area. The lateral growth range of the buffer area at different distances is controlled by the adaptive threshold function;

[0028] Extract historical avoidance trajectories between adjacent avoidance nodes, calculate the trajectory point density at each spatial position to obtain a trajectory density field, construct a cost function for the avoidance path based on the trajectory density field, solve the cost function to obtain a time field, and determine the centerline path of the avoidance channel based on the gradient direction of the time field;

[0029] Based on the centerline path and in combination with the lateral extent of the buffer areas at adjacent avoidance nodes, cubic spline interpolation is used to generate the boundary of the avoidance channel, and the buffer areas of all adjacent avoidance nodes are connected to form an avoidance channel network.

[0030] In an optional embodiment,

[0031] The passing sequence of AGVs is recorded in the avoidance channel network. The time interval between adjacent avoidance nodes is used as a time window. The passing batches are divided according to the distribution pattern of the time window. The batch scheduling sequence is generated, including:

[0032] A driving reference point is set at each avoidance node in the avoidance channel network. The reference points between adjacent avoidance nodes are connected to form a motion trajectory chain. The driving deviation of the AGV is determined based on the angle between the velocity vector of the AGV when it passes through the reference point and the trajectory chain. The passing time of the AGV is corrected based on the driving deviation to obtain the corrected passing time sequence.

[0033] Calculate the time intervals between adjacent AGVs in the corrected passing sequence, establish a cumulative distribution curve of the time intervals, identify clustering intervals by detecting the slope change of the cumulative distribution curve, and classify the AGVs in the same clustering interval into the first passing batch;

[0034] Calculate the batch stability index for each first-pass batch. The stability index is determined based on the mean and variance of the driving deviations of the AGVs in the batch. Select batches with stability indexes higher than the preset stability threshold as the second-pass batch. Re-cluster the AGVs that are not included in the second-pass batch at different time intervals.

[0035] The channel occupancy status and AGV automatic transport vehicle distribution density in the avoidance channel network are collected, and the average speed and driving deviation of the AGV automatic transport vehicles in the second passage batch are combined to calculate the avoidance node passing time interval between adjacent batches. The second passage batch is time-series sorted based on the avoidance node passing time interval to generate a batch scheduling sequence.

[0036] In an optional embodiment,

[0037] A node capacity matrix is ​​constructed based on the dwell time of the avoidance node, and the batch scheduling sequence is mapped to a traffic resource distribution map. When the AGV task interruption signal is detected, the node capacity value at the interruption location is extracted, and a backup channel with matching capacity is selected from the traffic resource distribution map. The interruption task is split into multiple subtask packages according to the node capacity threshold. Time-sharing scheduling through the backup channel includes:

[0038] A sliding time window is set for each avoidance node in the avoidance channel network. The operation data of AGVs are collected. The ratio of the AGVs' residence time to the number of passes within the time window is used as the node processing capacity indicator. The node capacity matrix is ​​generated based on the physical space dimensions of each avoidance node.

[0039] The traffic load of the avoidance channel network is calculated based on the node capacity matrix, and the traffic load is discretized into time segments to obtain a load sequence. Based on the load sequence and the topological connection relationship of the avoidance channel network, a traffic resource distribution map is generated.

[0040] When an AGV task interruption signal is detected, the capacity value of the node at the interruption location is obtained, and the capacity difference between adjacent nodes is calculated based on the node capacity matrix to form a capacity gradient. The capacity gradient is used to determine the set of available channels in the traffic resource distribution map, and the channel with the lowest resource occupancy rate is selected from the set of available channels as the backup channel;

[0041] Extract the capacity value of each node on the backup channel, set the node with the smallest capacity value as the task package size constraint, split the interrupt task into multiple task packages based on the constraint, and allocate execution time to each task package based on the real-time capacity value of each node;

[0042] The node load rate is calculated by combining the execution time of the task package with the capacity value of the corresponding node. The execution order of the task package is optimized according to the node load rate to generate a time-sharing scheduling plan.

[0043] In an optional embodiment,

[0044] The traffic load of the avoidance channel network is calculated based on the node capacity matrix, and the traffic load is discretized into time segments to obtain a load sequence. Based on the load sequence and the topological connection relationship of the avoidance channel network, a traffic resource distribution map is generated, including:

[0045] The operating data of AGVs in the avoidance channel network is collected to obtain the real-time number and dwell time of AGVs at each avoidance node. The ratio of the real-time number to the corresponding node capacity in the node capacity matrix is ​​used as the capacity occupancy rate. The traffic load of the avoidance channel network is obtained by accumulating the duration of the capacity occupancy rate.

[0046] Dividing the time segments according to the change in the traffic load within a unit time, and dynamically adjusting the length of the time segments based on a preset change threshold, converting the traffic load into a load sequence according to the divided time segments;

[0047] Extracting the connection relationship between adjacent nodes in the avoidance channel network to construct a basic topology graph, calculating the load difference between adjacent nodes based on the load sequence, and using the load difference as the weight value of the corresponding edge in the basic topology graph to generate a weighted topology graph;

[0048] The AGV passage frequency of each channel corresponding to each edge in the weighted topology graph is counted, the number of AGV flows between adjacent channels is calculated, and the channel resource occupancy intensity is obtained by multiplying the flow number with the weight value of the corresponding edge. The weight value of the weighted topology graph is updated based on the channel resource occupancy intensity to generate a traffic resource distribution map.

[0049] A second aspect of an embodiment of the present invention provides an AGV automatic transport vehicle scheduling simulation system, comprising:

[0050] The first unit is used to mark the location information of the AGV automatic transport vehicle in the path map to generate an initial path network;

[0051] The second unit is used to collect the running trajectory of the AGV automatic transport vehicle in the initial path network, extract the stop events in the running trajectory, mark the locations where the stop time is longer than the preset time threshold as avoidance nodes, construct buffer areas with the avoidance nodes as the center, and connect adjacent buffer areas to form an avoidance channel network;

[0052] The third unit is used to record the passage sequence of AGVs in the avoidance channel network, use the time intervals between adjacent avoidance nodes as time windows, divide the passage batches according to the distribution pattern of the time windows, and generate a batch scheduling sequence;

[0053] The fourth unit is used to construct a node capacity matrix based on the dwell time of the avoidance node, map the batch scheduling sequence into a traffic resource distribution map, and when a task interruption signal of the AGV automatic transport vehicle is detected, extract the node capacity value of the interruption location, select a backup channel with matching capacity from the traffic resource distribution map, split the interruption task into multiple subtask packages according to the node capacity threshold, and perform time-sharing scheduling through the backup channel.

[0054] According to a third aspect of an embodiment of the present invention, an electronic device is provided, including:

[0055] a processor; a memory for storing instructions executable by the processor;

[0056] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0057] According to a fourth aspect of an embodiment of the present invention, a computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the method described above is implemented.

[0058] In this embodiment, by marking the location information of the AGV automatic transport vehicle in the path map and constructing an avoidance channel network, the accurate identification of the AGV operation trajectory and the automatic construction of the avoidance node are achieved, which effectively reduces the cross-interference of vehicles during operation and improves the operation efficiency and safety of the system. Based on the time window analysis and batch scheduling sequence generation method, an optimized allocation mechanism for AGV traffic resources is established, which makes the path planning when multiple vehicles work together more reasonable, reduces waiting time, and improves the overall throughput and response speed of the logistics system. The node capacity matrix and task interruption processing mechanism are introduced. When an AGV task is abnormal, the task can be intelligently split and the backup channel can be used for time-sharing scheduling, which significantly enhances the fault tolerance and robustness of the system and ensures the continuity and reliability of logistics operations in complex environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 Schematic diagram of the flow of the AGV automatic transport vehicle scheduling simulation method according to an embodiment of the present invention;

[0060] Figure 2 A schematic diagram of the avoidance channel network topology and multi-vehicle coordinated avoidance control according to an embodiment of the present invention;

[0061] Figure 3 Schematic diagram of capacity gradient and backup channel selection according to an embodiment of the present invention. DETAILED DESCRIPTION

[0062] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0063] The following specific embodiments are used to describe the technical solution of the present invention in detail. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0064] Figure 1 FIG. 1 is a flow chart of the AGV automatic transport vehicle scheduling simulation method according to an embodiment of the present invention. Figure 1 As shown, the AGV automatic transport vehicle scheduling simulation method includes:

[0065] Mark the location information of the AGV automatic transport vehicle in the path map to generate the initial path network;

[0066] The AGV's trajectory in the initial path network is collected, and dwell events are extracted. Locations where the dwell time exceeds a preset time threshold are marked as avoidance nodes. Buffer areas are constructed with the avoidance nodes as the center, and adjacent buffer areas are connected to form an avoidance channel network.

[0067] The passing sequence of AGVs is recorded in the avoidance channel network. The time interval between adjacent avoidance nodes is used as a time window. The passing batches are divided according to the distribution pattern of the time window to generate a batch scheduling sequence.

[0068] A node capacity matrix is ​​constructed based on the dwell time of the avoidance node, and the batch scheduling sequence is mapped into a traffic resource distribution map. When the task interruption signal of the AGV is detected, the node capacity value of the interruption location is extracted, and a backup channel with matching capacity is selected from the traffic resource distribution map. The interruption task is split into multiple subtask packages according to the node capacity threshold, and time-sharing scheduling is performed through the backup channel.

[0069] In an optional embodiment,

[0070] Mark the location information of the AGV automatic transport vehicle on the path map to generate the initial path network including:

[0071] Collect laser scanning positioning coordinates and ultra-wideband positioning coordinates, calculate the corresponding weight coefficient based on the laser scanning error value and the ultra-wideband error value, multiply the weight coefficient with the corresponding positioning coordinates and superimpose them to obtain fused position information;

[0072] Collect the fused position information of the AGV automatic transport vehicle and calculate the position change between adjacent sampling periods. When the position change is greater than the preset change threshold, determine the main positioning source based on the change direction of the historical fused position information, use the normal positioning information as position compensation information, and update the fused position information based on the position compensation information;

[0073] Based on the updated fusion position information, a position tracking sequence is constructed, and the position information of the stopping points in the AGV automatic transport vehicle's running path is extracted as the reference point for path construction;

[0074] Calculate the distance and heading angle between the reference points of the path construction, select the reference points that meet the preset distance threshold and the preset angle threshold to establish a connection relationship, use the minimum spanning tree algorithm to traverse and connect the reference points, and generate an initial path network with directional attributes.

[0075] For example, in the process of collecting laser scanning positioning coordinates and ultra-wideband positioning coordinates, the laser scanning device obtains two-dimensional plane coordinates by emitting a laser beam and receiving the reflected signal, and the ultra-wideband positioning device obtains three-dimensional space coordinates through the principle of ranging. Laser scanning positioning has high short-term accuracy but has cumulative errors, and ultra-wideband positioning has stable long-term accuracy but relatively low instantaneous accuracy. When calculating the laser scanning error value, it is determined by comparing the position change rate within the continuous scanning cycle with the degree of deviation from the expected motion model. When the deviation exceeds the set range, the error value increases. The calculation of the ultra-wideband error value is based on signal strength, propagation delay stability, and base station geometric distribution factors. When the base station layout is poor or the signal quality decreases, the error value increases accordingly.

[0076] The weight coefficients are calculated using the inverse proportional distribution principle. The laser scanning weight coefficient is equal to the UWB error value divided by the sum of the two error values, and the UWB weight coefficient is equal to the laser scanning error value divided by the sum of the two error values. The sum of the two weight coefficients is ensured to be 1. The horizontal coordinate of the laser scanning positioning coordinate is multiplied by the laser scanning weight coefficient, and the horizontal coordinate of the UWB positioning coordinate is multiplied by the UWB weight coefficient. The two products are added together to obtain the fused horizontal coordinate. The vertical coordinate is obtained using the same processing method. For example, at a certain moment, the laser scanning coordinate is (1200, 800) with an error value of 0.3, and the UWB coordinate is (1205, 795) with an error value of 0.2. The laser scanning weight coefficient is 0.4, the UWB weight coefficient is 0.6, and the fused position information is (1203, 797).

[0077] During operation, the AGV records and fuses position information at a fixed sampling frequency, typically set between 100 and 500 milliseconds. When calculating the position change between adjacent sampling cycles, the position coordinates of the current cycle are subtracted from the position coordinates of the previous cycle to obtain the lateral and longitudinal changes. The total change is then calculated by taking the square root of the sum of the squares of these two changes. The preset change threshold is determined based on the AGV's maximum operating speed and sampling period. When the AGV's maximum speed is 2 meters per second and the sampling period is 200 milliseconds, the theoretical maximum change is 400 millimeters. To account for a safety margin, the threshold is set to 500 millimeters.

[0078] When the position change exceeds the preset threshold, it indicates that there may be a positioning anomaly, and anomaly detection and compensation processing are required. The change direction of the historical fused position information is determined by analyzing the position change trend in the last 10 sampling cycles, calculating the direction vector between each cycle and counting the main movement direction. The determination of the main positioning source is based on the stability evaluation of historical data. The change continuity of laser scanning positioning and ultra-wideband positioning in the historical time window is compared, and the positioning source with smoother changes and consistent with the movement trend is selected as the main positioning source. Non-abnormal positioning information is judged by the degree of deviation from the main positioning source. When the data of a certain positioning source deviates from the main positioning source by more than the set range, it is marked as abnormal, and the positioning data within the normal range is used as position compensation information.

[0079] The application of position compensation information uses a weighted average method to update the fused position information. The weight of each positioning source is adjusted according to the degree of anomaly. The weight of anomaly positioning sources is temporarily reduced or set to zero, while the weight of normal positioning sources is increased accordingly. For example, when laser scanning positioning is anomaly, its weight is reduced from 0.4 to 0.1, while the weight of ultra-wideband positioning is increased from 0.6 to 0.9, ensuring the reliability of the fused position information.

[0080] A position tracking sequence is constructed based on the updated fused position information, and the position coordinates of each sampling period are arranged in chronological order to form trajectory data. The AGV automatic transport vehicle will stop at loading and unloading points, waiting areas, charging stations, and other locations during the execution of the transport task. The stop point is identified by analyzing the speed change in the position tracking sequence. When the position change in five consecutive sampling periods is less than 50 mm, the AGV is determined to be in a stopped state, and the average position of this period is recorded as the stop point position information. The stop point position information includes the horizontal coordinate, the vertical coordinate, and the stop duration. The position point where the stop duration exceeds 3 seconds is determined as the path construction reference point.

[0081] The Euclidean distance algorithm is used to calculate the distance between path construction benchmarks. The square root of the square root of the square of the coordinate difference between the two benchmarks is used to obtain the straight-line distance. The heading angle is determined by calculating the angle between the line connecting the two benchmarks and the positive direction of the coordinate axis. The angle is calculated using the inverse tangent function and converted to a range of 0 to 360 degrees. The preset distance threshold is set based on the layout characteristics of the work area and is typically set between 50 and 200 meters to ensure a reasonable connection distance between adjacent benchmarks. The preset angle threshold is used to limit directional changes in the connection path and is typically set to within 45 degrees to avoid sharp turns.

[0082] Reference points that meet the preset distance and angle thresholds are connected to form a candidate path segment. The distance threshold is determined by the straight-line distance between the two reference points being less than or equal to the preset distance threshold, while the angle threshold is determined by the angle between the connecting path and the AGV's primary direction of travel being less than or equal to the preset angle threshold. For example, if reference point A has coordinates of (1000, 500) and reference point B has coordinates of (1080, 520), the distance is 82.5 meters, and the angle is 14 degrees, then points A and B meet the connection criteria when the distance threshold is set to 100 meters and the angle threshold is set to 30 degrees.

[0083] The application process of the minimum spanning tree algorithm includes four stages: initialization, edge weight calculation, sorting and selection, and loop detection. In the initialization stage, all reference points are treated as independent nodes, with each node forming a connected component. In the edge weight calculation stage, the connection cost is calculated for each pair of reference points that meet the connection conditions. This cost factor comprehensively considers the distance factor and the angle factor. The shorter the distance, the lower the cost, and the smaller the angle deviation, the lower the cost. In the sorting and selection stage, all candidate connection edges are sorted in ascending order of cost, and the edges with the lowest cost are selected and added to the spanning tree. In the loop detection stage, a union-find data structure is used to determine whether a newly added edge will form a loop. If a loop is formed, the edge is skipped and the next edge is selected.

[0084] After the spanning tree is constructed, each connected path is assigned a directional attribute based on the AGV's actual trajectory. By analyzing the order in which reference points are visited in the location tracking sequence, the direction of each path segment is determined. Bidirectional segments are marked with a bidirectional attribute, while unidirectional segments are marked with a corresponding unidirectional attribute. Each node in the path network contains location coordinates, node type (loading and unloading point, waiting area, intersection, etc.), and connection relationship information. Each edge contains attribute information such as the starting node, ending node, path length, direction, and access restrictions.

[0085] In this embodiment, by fusing laser scanning and ultra-wideband positioning data, dynamically adjusting the main positioning source and introducing a position compensation mechanism, the positioning accuracy and robustness of the AGV automatic transport vehicle are effectively improved; the position change analysis is combined to ensure the real-time and accuracy of the path update; in the path construction process, the path reference points are extracted based on the stop points and geometric features, and the minimum spanning tree algorithm is used to efficiently generate a directional initial path network, which can significantly improve the consistency of path modeling and the rationality of navigation scheduling, and meet the stability requirements of AGV path recognition and construction in complex environments.

[0086] In an optional embodiment,

[0087] The AGV's trajectory in the initial path network is collected, and dwell events are extracted. Locations where the dwell time exceeds a preset time threshold are marked as avoidance nodes. Buffer areas are constructed with the avoidance nodes as the center, and adjacent buffer areas are connected to form an avoidance channel network, including:

[0088] Collect the running trajectory of the AGV automatic transport vehicle in the initial path network, the running trajectory including laser positioning data and ultra-wideband positioning data, perform intelligent fusion processing on the running trajectory, and generate fused trajectory data;

[0089] Based on the fused trajectory data, the stop events of the AGV automatic transport vehicle are identified, the location information and duration of the stop events are extracted, and the stop events are divided into avoidance waiting type and task loading and unloading type;

[0090] Based on the classification results of the dwell events, the avoidance necessity score of the dwell location is calculated. The dynamic score threshold is determined based on the system operation status. The dwell locations with avoidance necessity scores higher than the dynamic score threshold are marked as candidate avoidance nodes. When the distance between adjacent candidate avoidance nodes is less than twice the length of the AGV, they are merged into one avoidance node to obtain the final avoidance node.

[0091] The historical trajectory data of the AGV at the avoidance node location is extracted, and the trajectory density clustering method is used to analyze the avoidance behavior characteristics of the AGV. Based on the avoidance behavior characteristics, the main avoidance direction and avoidance range of each avoidance node are determined.

[0092] An adaptive region growing algorithm is used to construct a buffer area along the main avoidance direction of the avoidance node. The positional relationship and avoidance direction of adjacent avoidance nodes are analyzed, and the trajectory density maximum path algorithm is used to connect adjacent buffer areas to form an avoidance channel network.

[0093] As the AGV moves along its initial path network, the laser positioning device continuously scans the surrounding environment to obtain two-dimensional coordinate data, while the ultra-wideband positioning device communicates with a base station to obtain three-dimensional position information. The laser positioning data is typically collected at a frequency of 20 Hz, recording position coordinates every 50 milliseconds, while the ultra-wideband positioning data is collected at a frequency of 10 Hz, recording position information every 100 milliseconds. This trajectory data also includes key parameters such as timestamps, speed, direction angle, and cargo load status to ensure the integrity and traceability of the trajectory information.

[0094] During intelligent fusion processing, positioning data of different frequencies are unified onto the same time base through a time synchronization mechanism. When the timestamps of laser positioning data and ultra-wideband positioning data do not completely correspond, linear interpolation is used to estimate the position coordinates at the corresponding time. The allocation of fusion weights is based on real-time positioning accuracy assessment. In an unobstructed environment, the weight of laser positioning is set to 0.7, and the weight of ultra-wideband positioning is set to 0.3. When laser positioning is interfered with, the weight is adjusted to 0.4, and the weight of ultra-wideband positioning is correspondingly increased to 0.6. The fused trajectory data is generated using a weighted average algorithm. The laser positioning coordinates are multiplied by the corresponding weights, and the ultra-wideband positioning coordinates are multiplied by the corresponding weights. The two results are added together to obtain the fused position coordinates.

[0095] Dwell events are identified by analyzing speed changes in the fused trajectory data. A dwell state is identified when the AGV's position change is less than one-tenth of its vehicle length over five consecutive sampling periods. A dwell event ends when the AGV's movement exceeds half its vehicle length and lasts for at least three sampling periods. Position information is extracted by calculating the center of gravity of all position coordinates during the dwell period, with the duration being the time difference between the start and end of the dwell. For example, if an AGV dwells near coordinates (1500, 800), the recorded position coordinates are (1498, 801), (1501, 799), (1502, 802), and (1499, 798). The center of gravity is calculated to be (1500, 800), and the dwell duration is 15 seconds.

[0096] The classification of dwell events is based on the functional attributes of the dwell location and the dwell duration characteristics. Avoidance and waiting type dwell events usually occur near path intersections, narrow passages, or other AGV operating areas, and the dwell duration is relatively short, generally between 5 and 30 seconds. Task loading and unloading type dwell events occur in operating locations such as shelf areas, loading and unloading platforms, and storage areas, and the dwell duration is longer, usually exceeding 60 seconds. The distance relationship between the dwell location and the predefined functional area is taken into consideration when making classification judgments. When the dwell location is less than 2 meters from the loading and unloading area and the dwell duration exceeds 45 seconds, it is classified as a task loading and unloading type; otherwise, it is classified as an avoidance and waiting type.

[0097] The avoidance necessity score is calculated by comprehensively considering four factors: stop frequency, stop duration, location importance, and traffic density. The stop frequency factor is determined by counting the number of stops at the location in the past 24 hours, with a higher score for a higher number of stops. The stop duration factor is calculated based on the average stop time, with locations with an average stop time of more than 20 seconds receiving a higher score. The location importance factor is determined by the connectivity of the location in the path network, with intersections with a large number of connected paths receiving a higher score. The traffic density factor is calculated by analyzing the frequency of AGV traffic near the location, with locations with a higher traffic frequency receiving a higher score.

[0098] The dynamic scoring threshold is determined and adjusted based on the current system operating status, including three dimensions: the number of online AGVs, task load intensity, and path congestion. When the number of online AGVs is small, the dynamic scoring threshold is set to a higher value, reducing the number of avoidance nodes to improve operational efficiency. Task load intensity is measured by the number of tasks per unit time. When the load intensity is high, the scoring threshold is lowered and avoidance nodes are increased to alleviate congestion. The degree of path congestion is assessed by the average path occupancy rate. When the occupancy rate exceeds 70%, the scoring threshold is lowered by 20% to enhance avoidance capabilities. For example, the scoring threshold is set to 75 points under normal operating conditions, adjusted to 60 points under high load conditions, and adjusted to 85 points under low load conditions.

[0099] The merging of candidate avoidance nodes is based on the principle of spatial proximity. The merge operation is performed when the straight-line distance between two candidate avoidance nodes is less than twice the AGV length. The standard AGV length is typically 1.2 to 1.5 meters, so the merge distance threshold is set at 2.4 to 3.0 meters. The position coordinates of the merged avoidance nodes are calculated using the weighted average of the original node positions. The weight distribution is based on the avoidance necessity score of each node, with nodes with higher scores receiving greater weights. The merged avoidance necessity score takes the maximum value of the original node scores to ensure that important avoidance positions are not weakened.

[0100] The application process of the trajectory density clustering method includes four stages: data preprocessing, density calculation, cluster analysis, and feature extraction. The data preprocessing stage extracts historical trajectory data within a 5-meter radius around the avoidance node, removing anomalous data points and noise points. The density calculation stage divides the extraction area into grids with a side length of 0.2 meters, and the number of trajectory points within each grid is counted as the density value. The cluster analysis stage uses a density-based clustering algorithm to group adjacent grids with density values ​​exceeding a threshold into the same cluster, forming trajectory density areas. The feature extraction stage analyzes the shape and directional characteristics of each cluster area to determine the primary avoidance direction.

[0101] Avoidance behavior characteristics are analyzed by statistically analyzing the direction and distance distribution of avoidance trajectories. Direction distribution uses an angle histogram method, dividing 360 degrees into 36 intervals. Trajectories within each interval are counted, and the directional interval with the highest frequency is determined as the primary avoidance direction. Distance distribution is obtained by analyzing the distance the AGV travels from its stop position to resuming normal operation. The statistical analysis results determine the boundaries of the avoidance range. For example, the primary avoidance direction for a given avoidance node is northeast (between 45 and 75 degrees), with an avoidance range radius of 3.5 meters.

[0102] The adaptive region growing algorithm constructs a buffer zone starting from the avoidance node location and gradually expands the buffer zone along the primary avoidance direction. The seed point for growth is set to the avoidance node coordinates, the growth direction is determined based on the primary avoidance direction, and the growth step size is set to 0.1 meters. Growth stops when reaching the avoidance range boundary, encountering an obstacle, or overlapping with other buffer zones. The growth rate is dynamically adjusted during region growing, with faster growth in areas with high trajectory density and slower growth in areas with low density. The buffer zone after growth is typically elliptical or irregularly polygonal, with the major axis aligned with the primary avoidance direction.

[0103] The positional relationship analysis of adjacent avoidance nodes is evaluated by calculating the straight-line distance and connection direction between the nodes. When the distance between two avoidance nodes is less than 15 meters and the angle between the connection direction and the respective main avoidance directions is less than 45 degrees, they are determined to be connectable adjacent nodes. When connecting adjacent buffer areas, the maximum trajectory density path algorithm searches for the path with the highest trajectory density between the two areas as the connecting channel. The search process uses a dynamic programming algorithm, starting from the boundary of the starting buffer area and gradually searching towards the boundary of the target buffer area. At each step, the adjacent grid with the largest density value is selected as the next node. The width of the connection path is determined according to the width of the AGV body and the safety distance, and is usually set to 1.5 to 2 times the body width.

[0104] The formation of an avoidance channel network integrates all buffer areas and connecting paths into a unified network structure. Nodes in the network contain attributes such as avoidance node information, buffer area boundary coordinates, and vehicle capacity. Edges contain attributes such as connecting path geometry, direction of travel, and capacity. The channel network supports two-way traffic but can dynamically adjust to one-way traffic to improve efficiency under high load conditions. The network's topology supports avoidance decision-making for the AGV scheduling system. When potential conflicts are detected, avoidance paths can be quickly planned, ensuring the safety and efficiency of coordinated multi-vehicle operations.

[0105] In this embodiment, by intelligently identifying and classifying stop events in the AGV's running trajectory and combining it with a dynamic scoring mechanism to screen locations with avoidance value, high-frequency conflict areas can be accurately located; trajectory density clustering is used to analyze avoidance behavior, effectively explore the avoidance direction and range, and realize behavioral modeling of avoidance nodes; through adaptive region growing and density maximum path algorithms, an avoidance channel network that matches the actual AGV operation is constructed, which not only improves the rationality and coverage of the avoidance path, but also enhances the traffic efficiency and safety in the process of multi-vehicle collaborative scheduling, and is suitable for dynamic optimization of AGV obstacle avoidance strategies in complex scenarios.

[0106] In an optional embodiment,

[0107] Along the main avoidance direction of the avoidance node, the adaptive region growing algorithm is used to construct the buffer area. The position relationship and avoidance direction of adjacent avoidance nodes are analyzed, and the trajectory density maximum path algorithm is used to connect adjacent buffer areas to form an avoidance channel network including:

[0108] The historical avoidance trajectories at the avoidance nodes are vertically projected onto a plane perpendicular to the main avoidance direction to obtain the lateral offset distance of the trajectory points. The kernel density estimation method is used to calculate the probability density distribution of the lateral offset distance to obtain the lateral distribution characteristics of the avoidance trajectory.

[0109] Along the main avoidance direction of the avoidance node, an adaptive threshold function is set based on the lateral distribution characteristics, and a region growing algorithm is used to construct a buffer area. The lateral growth range of the buffer area at different distances is controlled by the adaptive threshold function;

[0110] Extract historical avoidance trajectories between adjacent avoidance nodes, calculate the trajectory point density at each spatial position to obtain a trajectory density field, construct a cost function for the avoidance path based on the trajectory density field, solve the cost function to obtain a time field, and determine the centerline path of the avoidance channel based on the gradient direction of the time field;

[0111] Based on the centerline path and in combination with the lateral extent of the buffer areas at adjacent avoidance nodes, cubic spline interpolation is used to generate the boundary of the avoidance channel, and the buffer areas of all adjacent avoidance nodes are connected to form an avoidance channel network.

[0112] For example, during the vertical projection processing of the historical avoidance trajectory, a local coordinate system is established with the avoidance node as the origin, the main avoidance direction is set as the positive direction of the longitudinal axis, and the plane perpendicular to the main avoidance direction is used as the projection plane. All historical trajectory data within a radius of 8 meters around the avoidance node are extracted, including the position coordinate sequence of the AGV during the avoidance process. The coordinates of each trajectory point are converted to the local coordinate system, and the vertical distance from the trajectory point to the longitudinal axis is calculated as the lateral offset distance. For example, if a certain avoidance node is located at coordinates (2000, 1500) and the main avoidance direction is 30 degrees east of north, the lateral offset distance of the trajectory point (2003, 1502) after being converted to the local coordinate system is 1.8 meters.

[0113] The statistical analysis of the lateral offset distance covers the offset data of all avoidance trajectories near the avoidance node, which usually contains hundreds to thousands of trajectory points. The kernel density estimation method uses a Gaussian kernel function to calculate the probability density distribution. The bandwidth parameter of the kernel function is adaptively adjusted according to the distribution characteristics of the sample data. When the sample points are relatively concentrated, the bandwidth is set to 0.3 meters, and when the sample points are dispersed, the bandwidth is increased to 0.8 meters. During the probability density calculation process, the range of the lateral offset distance is divided into equally spaced intervals, and the width of each interval is set to 0.1 meters. The density value of each interval is calculated and a continuous probability density function is constructed. The lateral distribution characteristics are obtained by analyzing the peak position, distribution width and skewness coefficient of the probability density function. The peak position indicates the most commonly used avoidance path, and the distribution width reflects the degree of difference in avoidance behavior.

[0114] The setting of the adaptive threshold function is based on the statistical parameters of the lateral distribution characteristics, and the function form is a piecewise function with the distance to the main avoidance direction as the independent variable. In the area near the avoidance node, the threshold function value is larger to accommodate more avoidance trajectories. As the distance increases, the threshold gradually decreases to form a convergent buffer area shape. In the specific setting, the lateral threshold is set to the boundary value of the 95% confidence interval of the probability density distribution in the area within 2 meters from the avoidance node. In the area with a distance of 2 to 5 meters, the threshold linearly decays to the boundary of the 90% confidence interval. In the area with a distance of more than 5 meters, the threshold is fixed at the boundary of the 85% confidence interval. For example, if the half-width of the 95% confidence interval of a certain avoidance node is 2.5 meters, the lateral threshold in the close area is 2.5 meters, and the lateral threshold at 5 meters decays to 2.3 meters.

[0115] The implementation process of the region growing algorithm starts from the avoidance node position and advances forward in steps of 0.2 meters along the main avoidance direction. At each advancement position, the lateral growth range is determined based on the adaptive threshold function value of the current distance, and it expands to the left and right until the threshold boundary is reached. During the growth process, it detects whether an obstacle or path boundary is encountered. If an obstacle is encountered, the growth in that direction is stopped, and the boundary fusion processing is performed when other buffer areas are encountered. The growth termination conditions of the buffer area include reaching the maximum growth distance, encountering an impassable area, or being fully connected with the adjacent buffer area. After the growth is completed, the buffer area is usually spindle-shaped or elliptical, with the major axis along the main avoidance direction and the minor axis width determined by the lateral distribution characteristics.

[0116] Historical avoidance trajectories between adjacent avoidance nodes are extracted using a spatial range screening method. This method determines the boundary of the connecting region between two avoidance nodes and extracts the avoidance trajectory data for all AGVs within that region. The connecting region is determined based on the locations of the two avoidance nodes and their respective primary avoidance directions. A minimum bounding rectangle encompassing the two nodes and their buffer regions is constructed as the search range. Trajectory data screening criteria include the start or end point of the trajectory being within the buffer region of an avoidance node, the trajectory path passing through the connecting region, and the trajectory being generated within the last 30 days to ensure data timeliness.

[0117] Track point density is calculated using a grid-based statistical method. The connected area is divided into square grids with a side length of 0.3 meters. The number of track points within each grid is counted as the density value at that location. The density calculation takes into account the time weight of the track points: track points within the last 7 days are weighted as 1.0, track points between 7 and 15 days are weighted as 0.8, and track points between 15 and 30 days are weighted as 0.6. This ensures that the density field reflects the most recent avoidance behavior patterns. The track density field is constructed by Gaussian smoothing the density values ​​of each grid with a smoothing radius of 0.6 meters to eliminate noise and discontinuities in the density field.

[0118] The cost function is constructed based on the inverse transformation of the trajectory density field. Regions with high density values ​​have low cost values, indicating that the path is more suitable as an avoidance path. The specific form of the cost function is the difference between the maximum density value and the current position density value plus a base cost value, which is set to 1.0 to avoid zero division problems. The cost value of boundary areas and obstacle locations is set to infinity to ensure that these areas are avoided during path planning. The cost function also considers the path curvature penalty, adding an additional cost to paths with large curvatures to encourage the generation of straight avoidance paths.

[0119] The time field is solved using a fast marching method, starting from the buffer zone boundary of the starting avoidance node and gradually advancing toward the buffer zone boundary of the target avoidance node. The time value at each grid position represents the minimum time cost to reach that position from the starting point. The time value is calculated based on the time values ​​and movement costs of adjacent grids. The advancement process uses a heap sort to maintain the priority of the grids to be processed, ensuring that the grid with the smallest time value is always processed. Once the time field is constructed, each grid position has a corresponding time value, forming a continuous time distribution field.

[0120] The gradient direction is calculated by numerically differentiating the time field, using the central difference method to calculate the horizontal and vertical gradient components at each grid location. The gradient direction points in the direction of the fastest increasing time value, and the opposite direction is the shortest time path to the starting point. The centerline path is determined starting from the buffer zone boundary of the target avoidance node and gradually backtracking along the opposite direction of the gradient to the buffer zone boundary of the starting avoidance node, forming the optimal path connecting the two buffer zones. A step size of 0.1 meter is used during the backtracking process to ensure path accuracy and smoothness.

[0121] When applying cubic spline interpolation, each point along the centerline path serves as a control point, and the boundary of the avoidance path is determined by combining the lateral extent of the buffer zones of adjacent avoidance nodes. The lateral extent of each point along the centerline path is calculated using linear interpolation. The lateral extents of the start and end points correspond to the widths of the buffer zones of the two avoidance nodes, respectively. The lateral extent of the intermediate points is interpolated proportionally to the distance between them. Cubic spline interpolation ensures the continuity and smoothness of the boundary curve, avoiding sharp turns or discontinuous boundary shapes.

[0122] The formation of an avoidance channel network integrates the buffer areas and connecting channels of all adjacent avoidance nodes into a unified network structure. Each channel in the network has clear geometric boundaries, traffic directions, and capacity restrictions. It supports two-way traffic for AGVs but can be adjusted to one-way traffic when necessary. The channel network's topological relationships record the connectivity between avoidance nodes, providing the basic data for avoidance path planning in the AGV scheduling algorithm, ensuring that conflicts and deadlocks are effectively avoided during coordinated multi-vehicle operation.

[0123] In this embodiment, by performing kernel density estimation on the lateral offset distribution of the avoidance trajectory, the spatial diffusion characteristics of the avoidance behavior are accurately characterized. Combined with the adaptive region growing algorithm, the dynamic expansion of the buffer area in the main avoidance direction is achieved, thereby enhancing the flexibility and environmental adaptability of the avoidance modeling. Further, a cost function is constructed based on the trajectory density field and a time field solution mechanism is introduced to effectively identify the optimal avoidance channel center path, thereby improving the rationality of path planning and traffic efficiency. Finally, the channel boundary is smoothed by cubic spline interpolation to achieve continuous connection and geometric consistency of the avoidance channel network, providing technical support for the construction of a highly reliable avoidance traffic strategy for multiple AGV systems in high-density environments.

[0124] Figure 2 FIG. 1 is a schematic diagram of the avoidance channel network topology and multi-vehicle coordinated avoidance control according to an embodiment of the present invention. Figure 2The figure shows the topology of the avoidance channel network and the multi-vehicle coordinated avoidance control mechanism. A clear legend is provided in the upper right corner of the figure to illustrate the meaning of each element. Orange squares represent avoidance nodes, blue lines represent avoidance channels, and small blue rectangles represent AGVs. In the main body, the gray area represents the main AGV driving path. Six orange dots (nodes 1 to 6) represent key avoidance nodes, which are interconnected by blue avoidance channels to form a complete grid structure. Each channel is labeled with its capacity (the maximum number of vehicles it supports for simultaneous avoidance), for example, 2-3 for vertical channels and 3-4 for horizontal channels. The small blue rectangles (A1 to A9) in the figure represent the nine AGVs currently traveling in the network, distributed across different channels. Through precise channel construction and intelligent coordinated control, this technical solution has improved channel utilization by 18.5%, reduced average travel time by 35.4%, and increased the coordination success rate by 6.9%, effectively resolving bottlenecks in multi-vehicle coordinated avoidance.

[0125] In an optional embodiment,

[0126] The passing sequence of AGVs is recorded in the avoidance channel network. The time interval between adjacent avoidance nodes is used as a time window. The passing batches are divided according to the distribution pattern of the time window. The batch scheduling sequence is generated, including:

[0127] A driving reference point is set at each avoidance node in the avoidance channel network. The reference points between adjacent avoidance nodes are connected to form a motion trajectory chain. The driving deviation of the AGV is determined based on the angle between the velocity vector of the AGV when it passes through the reference point and the trajectory chain. The passing time of the AGV is corrected based on the driving deviation to obtain the corrected passing time sequence.

[0128] Calculate the time intervals between adjacent AGVs in the corrected passing sequence, establish a cumulative distribution curve of the time intervals, identify clustering intervals by detecting the slope change of the cumulative distribution curve, and classify the AGVs in the same clustering interval into the first passing batch;

[0129] Calculate the batch stability index for each first-pass batch. The stability index is determined based on the mean and variance of the driving deviations of the AGVs in the batch. Select batches with stability indexes higher than the preset stability threshold as the second-pass batch. Re-cluster the AGVs that are not included in the second-pass batch at different time intervals.

[0130] The channel occupancy status and AGV automatic transport vehicle distribution density in the avoidance channel network are collected, and the average speed and driving deviation of the AGV automatic transport vehicles in the second passage batch are combined to calculate the avoidance node passing time interval between adjacent batches. The second passage batch is time-series sorted based on the avoidance node passing time interval to generate a batch scheduling sequence.

[0131] The driving reference point for each avoidance node in the avoidance channel network is set using a geometric center positioning method, with the geometric center of the avoidance node's buffer zone serving as the reference point. The coordinates of the reference point are calculated by averaging the coordinates of all vertices along the buffer zone's boundary, ensuring that the reference point is located at the core of the buffer zone. For example, the buffer zone of a particular avoidance node contains eight boundary vertices with coordinates of (1200, 800), (1205, 802), (1208, 805), (1210, 810), (1208, 815), (1205, 818), (1200, 820), and (1195, 815). The calculated reference point coordinates are (1203.9, 810.6). The reference point setting also considers the functional attributes of the avoidance node. For intersection-type avoidance nodes, the reference point position is offset by 0.5 meters toward the primary direction of travel to facilitate AGV path tracking and deviation detection.

[0132] During the construction of the motion trajectory chain, the reference points of adjacent avoidance nodes are connected according to their spatial positional relationships to form a continuous trajectory chain structure. Connection judgment is based on the straight-line distance between avoidance nodes and the connectivity of the avoidance channel. A connection relationship is established when an avoidance channel exists between two avoidance nodes and the straight-line distance is less than 20 meters. The direction of the trajectory chain is determined based on the primary direction of travel of the AGV. The dominant direction is determined by analyzing the distribution of AGV travel directions in historical trajectory data. The geometric representation of the trajectory chain uses a parameterized straight line equation, which contains the starting point coordinates, end point coordinates, and direction vector information, providing a reference for subsequent deviation calculations.

[0133] The velocity vector of the AGV as it passes a reference point is acquired through real-time monitoring by onboard sensors and consists of two components: velocity magnitude and direction of movement. Velocity is calculated by dividing the position change between two consecutive sampling cycles by the time interval. The sampling period is set to 100 milliseconds to ensure accurate and real-time speed measurement. The direction of movement is determined by analyzing the AGV's position change trend and calculating the direction vector formed by three consecutive position points to eliminate the influence of errors in single measurements. The angle between the velocity vector and the trajectory chain is calculated using the vector dot product method. A scalar operation is performed on the velocity vector and the trajectory chain direction vector, and the angle value is obtained using the inverse cosine function.

[0134] Driving deviation is determined by comparing the angle with a deviation threshold. An angle less than 15 degrees indicates normal driving, an angle between 15 and 30 degrees indicates a minor deviation, and an angle exceeding 30 degrees indicates a severe deviation. Deviation determination also considers the AGV's operating state. The deviation threshold for a loaded state is 5 degrees higher than that for an unloaded state to avoid misjudgments due to load influence. For example, when an AGV passes a reference point, the velocity vector direction is 25 degrees north-east, and the trajectory chain direction is 10 degrees north-east. The calculated angle is 15 degrees, resulting in a minor deviation.

[0135] Passing time corrections are made based on the degree of driving deviation and the AGV's motion characteristics. No correction is performed during normal driving conditions. Correction is delayed by 0.2 seconds for minor deviations and by 0.5 seconds for severe deviations. Correction time is determined based on the AGV's steering capability and speed adjustment response time. AGVs with larger turning radii require longer correction times. The corrected passing time series records the precise moment the AGV passes each reference point, with millisecond accuracy, providing accurate data for subsequent time interval analysis.

[0136] The time interval between adjacent AGV passes is calculated by taking the difference between the corrected pass times. The time interval is calculated by subtracting the pass times of two adjacent AGVs at the same reference point. The statistical scope of the time interval data is set to all pass records within the last 24 hours to ensure the adequacy and representativeness of the data. The cumulative distribution curve is constructed using the empirical distribution function method. The time interval data is sorted in ascending order, the cumulative probability corresponding to each value is calculated, and a curve is plotted showing the relationship between the cumulative probability and the time interval. The cumulative distribution curve is smoothed using a moving average method, with a window size set to 5% of the total data volume to eliminate the impact of data fluctuations on the curve shape.

[0137] Slope changes are detected by numerically differentiating the cumulative distribution curve, calculating the tangent slope at each point on the curve, and identifying locations where the slope changes significantly as the boundaries of the clustering interval. The criterion for determining slope changes is based on the absolute value of the slope difference between adjacent points. When the difference exceeds 0.1, a significant change is considered to have occurred. The clustering interval is determined by continuous sections with gentle slopes. Continuous intervals with a slope change of less than 0.05 are identified as a clustering interval. For example, the slope of the cumulative distribution curve is relatively gentle in the three time intervals of 2 to 5 seconds, 8 to 12 seconds, and 15 to 20 seconds. These three intervals correspond to different traffic modes.

[0138] The first batching process groups AGVs within the same clustering interval into the same batch. Each batch contains groups of AGVs that passed through the batch with similar time intervals. The number of AGVs in a batch typically ranges from 3 to 15. Batches with too few AGVs are merged into adjacent batches, while batches with too many AGVs are further subdivided. The time span of a batch is determined by the clustering interval boundaries: the first clustering interval forms the first batch, the second clustering interval forms the second batch, and so on.

[0139] The stability index is calculated based on a statistical analysis of AGV driving deviation data within a batch. It consists of two components: the mean deviation and the variance. The mean deviation reflects the overall driving quality of the batch; a smaller mean indicates more standardized driving behavior of the AGVs within the batch. The variance reflects the consistency of AGV driving within the batch; a smaller variance indicates more uniform driving behavior within the batch. The stability index is calculated by combining the mean deviation and variance, with a weight of 0.6 for the mean and 0.4 for the variance, to comprehensively assess the stability of the batch. The preset stability threshold is determined based on the statistical characteristics of historical operating data and is typically set at the 70th percentile of the stability index distribution for all batches.

[0140] The second batch is selected by comparing the stability index of each batch with a preset threshold. Batches with stability indicators above the threshold are retained as the second batch, while batches below the threshold are disassembled. AGVs in the disassembled batches are re-analyzed for time interval clustering, and the batches are re-divided using more granular clustering criteria to ensure that all AGVs are classified into stable batches. During the re-clustering process, the minimum width requirement for the cluster interval is reduced, allowing batches with shorter time spans but higher stability to be formed.

[0141] Aisle occupancy status is collected by deploying sensors at key locations in the avoidance aisle, monitoring the number and location distribution of AGVs within the aisle in real time. This occupancy status includes information such as the number of AGVs in the aisle, average spacing, and speed distribution, and is updated 10 times per second. AGV density is calculated by counting the number of AGVs per unit length of the aisle. Higher density values ​​indicate more congested aisles, requiring increased time between batches.

[0142] The time interval calculation for the inter-batch avoidance node takes into account four factors: channel occupancy, AGV density, average batch speed, and driving deviation. The time interval is increased when channel occupancy is high, when AGV density is high, when average batch speed is low, and when driving deviation is high. The base time interval is set at 30 seconds, and the adjustment range is between 10 and 60 seconds depending on the impact of each factor. For example, the time interval is adjusted to 45 seconds for high occupancy and 40 seconds for low speed. When multiple factors work together, the adjustment is cumulative.

[0143] The batch scheduling sequence is generated by arranging the second-passing batches according to the calculated time interval, ensuring sufficient safety intervals between adjacent batches. The sorting process takes batch priority into account, with urgent tasks receiving higher priority and being scheduled for a higher transit time. The scheduling sequence includes information such as the estimated transit time for each batch, the list of AGVs included, and the expected transit speed, providing a detailed time plan for AGV scheduling execution.

[0144] In this embodiment, by setting reference points and combining them with trajectory chains, refined monitoring and deviation correction of the AGV's passing path are achieved, thereby improving the accuracy of the passing time series data; the introduction of cumulative distribution slope analysis to identify time interval aggregation intervals can effectively divide the passage batches with time correlation; the stability index is used to screen highly consistent batches to improve the controllability and coordination of the scheduling strategy; combined with the real-time occupancy status of the avoidance channel and the AGV operation characteristics, the reasonable intervals between batches are dynamically calculated and sorted to generate a scheduling sequence, further optimizing the AGV passage rhythm, alleviating channel conflicts and congestion, and improving the efficiency and traffic safety of multi-vehicle collaborative scheduling.

[0145] In an optional embodiment,

[0146] A node capacity matrix is ​​constructed based on the dwell time of the avoidance node, and the batch scheduling sequence is mapped to a traffic resource distribution map. When the AGV task interruption signal is detected, the node capacity value at the interruption location is extracted, and a backup channel with matching capacity is selected from the traffic resource distribution map. The interruption task is split into multiple subtask packages according to the node capacity threshold. Time-sharing scheduling through the backup channel includes:

[0147] A sliding time window is set for each avoidance node in the avoidance channel network. The operation data of AGVs are collected. The ratio of the AGVs' residence time to the number of passes within the time window is used as the node processing capacity indicator. The node capacity matrix is ​​generated based on the physical space dimensions of each avoidance node.

[0148] The traffic load of the avoidance channel network is calculated based on the node capacity matrix, and the traffic load is discretized into time segments to obtain a load sequence. Based on the load sequence and the topological connection relationship of the avoidance channel network, a traffic resource distribution map is generated.

[0149] When an AGV task interruption signal is detected, the capacity value of the node at the interruption location is obtained, and the capacity difference between adjacent nodes is calculated based on the node capacity matrix to form a capacity gradient. The capacity gradient is used to determine the set of available channels in the traffic resource distribution map, and the channel with the lowest resource occupancy rate is selected from the set of available channels as the backup channel;

[0150] Extract the capacity value of each node on the backup channel, set the node with the smallest capacity value as the task package size constraint, split the interrupt task into multiple task packages based on the constraint, and allocate execution time to each task package based on the real-time capacity value of each node;

[0151] The node load rate is calculated by combining the execution time of the task package with the capacity value of the corresponding node. The execution order of the task package is optimized according to the node load rate to generate a time-sharing scheduling plan.

[0152] The sliding time window is set using a fixed-length moving window mechanism. Each avoidance node is configured with an independent time window, with a window length set to 30 minutes and a sliding step size set to 5 minutes. The data collected within the window includes information such as the AGV's arrival time, departure time, cargo status, and task type. The dwell time is calculated by subtracting the AGV's arrival time from its departure time, and the total number of AGVs passing through the node within the window is counted. For example, if 12 AGVs pass through a certain avoidance node within a 30-minute time window, with a cumulative dwell time of 180 seconds, the ratio of dwell time to the number of AGVs passing through is 15 seconds per vehicle, representing the average processing time for that node.

[0153] The calculation of the node processing capacity index takes into account the different loading states of AGVs. The weight of the dwell time of an empty AGV is set to 1.0, the weight of the dwell time of a fully loaded AGV is set to 1.2, and the weight of the dwell time of a half-loaded AGV is set to 1.1. The weighted average processing time is obtained by multiplying the dwell time of AGVs in each loading state by the corresponding weight, summing the sum, and then dividing it by the total number of passes. The measurement of physical space dimensions includes the length, width, and shape parameters of the avoidance node buffer area. The length is determined by the maximum span of the main avoidance direction, and the width is determined by the maximum span of the perpendicular avoidance direction. The generation of the node capacity matrix combines the processing capacity index and the physical space dimensions after normalization to form a matrix structure that describes the capacity characteristics of each node.

[0154] The traffic load of the avoidance channel network is calculated based on the real-time occupancy of each avoidance node and the expected traffic demand. The current load is calculated by counting the ratio of the number of AGVs in a node to the node's capacity, with a higher ratio indicating a greater load. The expected load is estimated based on the number of planned but unexecuted tasks, taking into account the expected arrival time and execution duration of the tasks. The combined load is the weighted sum of the current load and the expected load, with the current load weighted at 0.7 and the expected load weighted at 0.3.

[0155] Discretization of the load series divides the continuous time axis into equal-length time segments, each set to 10 minutes. The load value within each time segment is represented by the average load value within that period, eliminating the impact of short-term fluctuations on the overall load trend. The discretized load series contains load forecasts for each time segment within the next two hours, providing time-based reference information for resource allocation.

[0156] The generation of the access resource distribution map combines the topology of the avoidance channel network with the load sequence to form a two-dimensional resource distribution matrix. The row indices of the matrix correspond to different avoidance nodes, the column indices correspond to different time slices, and the matrix element values ​​represent the resource utilization rate of the corresponding node at a specific time. The resource utilization rate is calculated by dividing the load value by the node capacity value. An occupancy rate close to 1.0 indicates that the resource is nearing saturation, while an occupancy rate exceeding 1.0 indicates resource overload. The distribution map also contains information about the connectivity between nodes, indicating which nodes have direct access paths.

[0157] Task interruption signal detection is achieved by monitoring AGV status changes and abnormal events, including various interruption types such as equipment failure, path obstruction, and emergency stop. The interruption location is determined based on the AGV's real-time position information, identifying the avoidance node or avoidance channel the AGV is currently located. Node capacity values ​​are extracted by querying the node capacity matrix for the corresponding node, including information such as the maximum number of AGVs accommodated, average processing time, and space utilization.

[0158] The capacity gradient is calculated by comparing the capacity differences between adjacent nodes. The capacity difference is calculated by subtracting the capacity of the adjacent nodes from the target node's capacity. The capacity gradient vector points toward the node with the greater capacity, and the magnitude of the gradient reflects the degree of capacity difference. The gradient calculation takes into account the distance between nodes, with the contribution of more distant nodes to the gradient being reduced accordingly. For example, if the capacity of the node at the interruption location is 5, the capacity of the adjacent node to the east is 8, and the capacity of the adjacent node to the west is 3, then the eastward gradient is 3, and the westward gradient is -2.

[0159] The set of available channels is determined based on the capacity gradient and occupancy information from the resource distribution map. A search is performed along the capacity gradient to identify reachable avoidance nodes, selecting those with resource occupancy rates less than 0.8 as candidate channel nodes. A path search algorithm is used to check the connectivity of candidate channels, ensuring that a feasible path exists from the interruption point to the candidate node. The set of available channels contains all channels that meet the capacity and connectivity requirements, providing candidates for backup channel selection.

[0160] Backup channels are selected based on the principle of minimizing resource utilization. The average resource utilization of all nodes on each candidate channel is calculated, and the channel with the lowest average utilization is selected as the backup channel. The selection process also considers channel length. When multiple channels have similar utilization rates, the channel with the shorter path length is prioritized. After the backup channels are determined, the traffic resource distribution map is updated to allocate the expected resource usage to the backup channels to avoid resource conflicts.

[0161] The constraints on task splitting are determined by the processing capacity of the node with the lowest capacity on the backup channel, which becomes the bottleneck of the entire backup channel. The task package size is calculated by dividing the total workload of the original task by the processing capacity of the bottleneck node, ensuring that a single task package does not exceed the bottleneck node's processing limit. For example, if the original task requires transporting 20 cargo units, and the bottleneck node can only process 5 cargo units at a time, the task will be split into 4 task packages, each containing 5 cargo units.

[0162] Execution time is allocated based on each node's real-time capacity and the processing requirements of the task package. The execution time of each task package is calculated by dividing the task package size by the corresponding node's processing rate, where the processing rate is the inverse of the node's capacity. Time allocation accounts for node load fluctuations, increasing execution time margin during periods of high load and reducing execution time during periods of low load. The resulting allocation forms a detailed schedule that clearly defines the start and end times and execution nodes for each task package.

[0163] The node load factor is calculated by comparing the task package execution time with the node's capacity. The load factor is equal to the execution time divided by the length of the node's available time window. The load factor calculation takes into account the node's parallel processing capabilities. When a node can process multiple task packages simultaneously, the load factor is the cumulative load factor of all parallel task packages. Nodes with a load factor exceeding 0.9 are marked as overloaded, and the execution order of task packages needs to be optimized to balance load distribution.

[0164] Execution order optimization utilizes a load balancing algorithm, prioritizing task packages on nodes with lower load rates and delaying them on nodes with higher load rates. The optimization process considers dependencies between task packages, ensuring that dependent task packages are executed in the correct order. The generation of a time-sharing scheduling solution combines the optimized execution order with the time schedule to form a complete task package scheduling plan, including the specific execution time, execution node, and resource requirements for each task package, ensuring that interrupted tasks can be efficiently completed through alternate channels.

[0165] In this embodiment, by constructing a capacity matrix that reflects the processing capacity of the avoidance node and dynamically evaluating the channel load status, it is possible to accurately identify traffic bottlenecks and resource redundant areas; combined with AGV operation interruption information, the capacity gradient is used to match the backup channel in the traffic resource distribution map to achieve rapid recovery and path reconstruction after task interruption; by splitting the task package based on the minimum capacity limit of the node and performing time-sharing scheduling based on the real-time capacity value, the impact of sudden tasks on channel resources can be effectively alleviated; finally, the task execution order is optimized through the node load rate, the scheduling flexibility and resource utilization are improved, and the steady-state operation capability and robustness of the AGV system in dynamic abnormal scenarios are enhanced.

[0166] Figure 3 Schematic diagram of capacity gradient and backup channel selection according to an embodiment of the present invention. Figure 3 As shown in the figure, the backup channel selection mechanism based on capacity gradient is demonstrated, which intuitively compares the difference between this technical solution and the traditional shortest distance path algorithm. There are 8 avoidance nodes (A to H) in the figure, each marked with a different capacity value. Node A is the task interruption location (capacity is 5). The traditional shortest distance path algorithm selects the path A→B→D→G, with a total distance of 3, a total capacity of 26 (5+8+7+6), and an average occupancy rate of 0.78; while this technical solution selects the path A→B→E→G based on capacity gradient analysis. Although the distance is the same at 3, the total capacity is increased to 28 (5+8+9+6), and the average occupancy rate is reduced to 0.64. The capacity gradient is intuitively represented by arrows and numerical values. For example, the gradient from A to B is +3, indicating an increase in capacity; the gradient from A to C is -2, indicating a decrease in capacity; and the gradient from E to G is -3, indicating a significant decrease in capacity. This technical solution prioritizes path segments with positive capacity gradients (e.g., A→B, with a gradient of +3) and avoids those with negative capacity gradients (e.g., D→G, with a gradient of -1), thereby ensuring a higher overall path capacity. Specifically, node E (capacity 9) was selected as the intermediate node, rather than node D (capacity 7). This decision-making mechanism, based on capacity rather than pure distance, ensures higher capacity redundancy and lower resource utilization in the backup path, enabling more effective response to mission interruptions and improving system robustness and recovery efficiency.

[0167] In an optional embodiment,

[0168] The traffic load of the avoidance channel network is calculated based on the node capacity matrix, and the traffic load is discretized into time segments to obtain a load sequence. Based on the load sequence and the topological connection relationship of the avoidance channel network, a traffic resource distribution map is generated, including:

[0169] The operating data of AGVs in the avoidance channel network is collected to obtain the real-time number and dwell time of AGVs at each avoidance node. The ratio of the real-time number to the corresponding node capacity in the node capacity matrix is ​​used as the capacity occupancy rate. The traffic load of the avoidance channel network is obtained by accumulating the duration of the capacity occupancy rate.

[0170] Dividing the time segments according to the change in the traffic load within a unit time, and dynamically adjusting the length of the time segments based on a preset change threshold, converting the traffic load into a load sequence according to the divided time segments;

[0171] Extracting the connection relationship between adjacent nodes in the avoidance channel network to construct a basic topology graph, calculating the load difference between adjacent nodes based on the load sequence, and using the load difference as the weight value of the corresponding edge in the basic topology graph to generate a weighted topology graph;

[0172] The AGV passage frequency of each channel corresponding to each edge in the weighted topology graph is counted, the number of AGV flows between adjacent channels is calculated, and the channel resource occupancy intensity is obtained by multiplying the flow number with the weight value of the corresponding edge. The weight value of the weighted topology graph is updated based on the channel resource occupancy intensity to generate a traffic resource distribution map.

[0173] The collection of AGV automatic transport vehicle operation data is achieved through a sensor array deployed at key locations in the avoidance channel network, including lidar, ultrasonic sensors and visual recognition devices. The sensor scans each avoidance node area with a cycle of 500 milliseconds to identify the number, location and movement status of the AGV. The real-time quantity statistics use a target tracking algorithm to assign a unique identifier to each AGV entering the node area and track its stay in the node until it leaves. The calculation of the stay time starts from the time the AGV enters the node boundary and ends when the AGV completely leaves the node boundary, with time accuracy reaching the millisecond level. For example, at the current moment, an avoidance node detects 3 AGVs, which stayed for 45 seconds, 120 seconds and 80 seconds respectively. The capacity value of the node in the node capacity matrix is ​​5 vehicles, and the capacity occupancy rate is 0.6.

[0174] The capacity occupancy rate is calculated by dividing the real-time number of AGVs in the node by the corresponding capacity value in the node capacity matrix. An occupancy rate value between 0 and 1 indicates normal use, and a value greater than 1 indicates node overload. The accumulation process of duration takes into account the temporal variation characteristics of the occupancy rate. The occupancy rate at different times is multiplied by the corresponding time interval and then summed up using an integral calculation method. The calculation of the traffic load covers the cumulative value of the capacity occupancy rate of all nodes in the avoidance channel network, reflecting the busyness of the entire network. Different weights are assigned to different types of nodes during the accumulation process. The weight of the cross node is set to 1.2, the weight of the loading and unloading node is set to 1.0, and the weight of the waiting node is set to 0.8.

[0175] The change in traffic load is calculated by taking the difference between the load values ​​at adjacent moments. A new time segment is created when the absolute value of the change exceeds a preset change threshold. The change threshold is dynamically adjusted based on network size and operational characteristics. The threshold for small-scale networks is set to 0.1, and for large-scale networks it is set to 0.3. The length of the time segment is dynamically adjusted based on the frequency of load changes. When changes are frequent, the time segment is shortened to 5 minutes, and when changes are gentle, the time segment is extended to 20 minutes. For example, within a 2-hour time window, the traffic load rises from 0.3 to 0.8 and then falls to 0.4. The changes are 0.5 and 0.4, respectively, both exceeding the threshold of 0.3, so the time segment is divided into three time segments.

[0176] The load series conversion uses the average load value within each time segment as the representative value for that segment, forming a discrete numerical sequence. The number of sequence elements is equal to the number of time segments, and the time span of the sequence is typically set to a four-hour forecast window. The load series also includes additional information such as the start and end time of each time segment and the load change trend, providing a time reference for subsequent analysis.

[0177] The construction of the basic topology map extracts all avoidance nodes and their connections within the avoidance corridor network, representing them using an undirected graph structure from graph theory. Nodes are represented by unique identifiers, and edges represent direct connections between nodes. Connections are determined based on the physical connectivity of the avoidance corridor, establishing a connecting edge when an avoidance corridor exists between two avoidance nodes. The topology map also records the geometric properties of each edge, including information such as connection length, corridor width, and directional restrictions.

[0178] Load differences are calculated by comparing the load values ​​of adjacent nodes within the same timeframe. A positive difference is calculated by subtracting the higher load value from the lower load value. The load difference reflects the degree of load imbalance between nodes; a larger difference indicates a more uneven load distribution. Weights are assigned using the load difference as the weight of the corresponding edge. Edges with larger weights indicate a significant load difference between the two connected nodes. Weighted topology generation adds weight attributes to each edge based on the base topology, forming a complete network load distribution model.

[0179] AGV passage frequency statistics use a sliding count window method, recording the number of AGVs passing through each channel in the last hour. The statistics distinguish between different directions of passage, and bidirectional channels count the forward and reverse passage frequencies separately. The passage frequency data is updated every 10 minutes to ensure data timeliness and accuracy. For example, if the channel connecting nodes A and B has 18 AGVs traveling from A to B and 12 AGVs traveling from B to A in the past hour, the total passage frequency for this channel is 30.

[0180] The flow count calculation analyzes AGV transfers between adjacent aisles and counts the number of AGVs entering an adjacent aisle from one channel. This flow count is based on AGV trajectory tracking data, identifying events where an AGV leaves one aisle and enters the next. The flow count calculation takes into account time delays; the time interval between an AGV leaving one aisle and entering the next is typically between 30 seconds and 3 minutes. The flow matrix records the number of AGV flows between all aisle pairs, forming a numerical description of the network flow pattern.

[0181] Channel resource occupancy intensity is calculated by multiplying the number of AGV transactions by the corresponding edge weight, reflecting the actual channel resource usage intensity. Occupancy intensity takes into account both load variation and traffic demand. Channels with large load variations and high traffic frequency have higher occupancy intensities. Normalization of intensity values ​​maps the occupancy intensities of all channels to a range of 0 to 1, facilitating comparative analysis across channels.

[0182] Weight updates are based on channel resource utilization, combining the original weights with the utilization intensity using a weighted average. The updated weights are calculated by multiplying the original weight by 0.6 and the utilization intensity by 0.4, ensuring that the weights reflect both load variations and actual usage. Weight updates are set every 30 minutes to keep the weights synchronized with network performance.

[0183] The generation of a traffic resource distribution map converts the updated weighted topology map into a visual representation of resource distribution, using color coding to display the resource utilization level of different channels. High-weight edges are colored red to indicate resource shortages, low-weight edges are colored green to indicate sufficient resources, and medium-weight edges are colored yellow to indicate moderate resources. The distribution map also indicates the load status of key nodes and the location of bottleneck channels, providing an intuitive reference for resource status for scheduling decisions. The distribution map is updated at the same frequency as weight updates, reflecting real-time changes in resource distribution within the avoidance channel network, supporting dynamic path planning and task scheduling optimization.

[0184] In this embodiment, by integrating the real-time number and residence time of AGVs at each avoidance node and combining the node capacity matrix to dynamically calculate the capacity occupancy rate, a refined modeling of the traffic load is achieved; the time segment division method driven by the load change is adopted to improve the timeliness and adaptability of the load sequence; the topology map and load difference are introduced to construct a weighted network, and the edge weight is further adjusted in combination with the AGV flow frequency and resource occupancy intensity to form a traffic resource distribution map that fits the actual traffic pressure, effectively reflecting the channel bottlenecks and traffic hotspots, and providing highly reliable data support and dynamic control basis for subsequent scheduling optimization, path planning and exception handling.

[0185] In another optional embodiment, the AGV (Automated Guided Vehicle) scheduling simulation method uses a cellular map to represent the work environment, with each cell recording its location coordinates and occupancy status. During implementation, the factory map is divided into a 200×200 grid with a cell size of 0.5×0.5 meters. Path planning utilizes the A* algorithm, with an evaluation function combining physical distance and traffic congestion factor. The congestion factor is dynamically calculated based on the frequency of cell passes within 30 seconds. AGV tasks are broken down into sequences of atomic actions, such as navigation, positioning, and lifting. Each action is completed before the next instruction is received, enabling the AGV to insert avoidance maneuvers between any actions. Traffic conflict management utilizes the Player Killing (PK) algorithm, a decision-making mechanism for resolving AGV traffic conflicts. This algorithm, derived from the competitive mechanism of gaming, is used in an AGV system. When two or more AGVs encounter a path conflict, the system assigns a score to each AGV. High-scoring AGVs are granted right-of-way, while low-scoring AGVs perform avoidance maneuvers, effectively resolving traffic conflicts. The conflicting AGV scores are calculated by comprehensively considering task priority (weight 0.35), distance traveled (0.25), battery level (0.2) and load status (0.2). AGVs with higher scores have priority, while AGVs with lower scores are subject to avoidance. Charging management simulates real battery changes, with an empty load consumption rate of 0.08% / m and a full load consumption rate of 0.12% / m. Charging tasks are automatically generated when the battery level is less than 25%. Traffic conditions are visualized through heat maps, and the frequency of AGV passage is recorded and displayed in levels: 0-5 times green, 6-15 times yellow, 16-30 times orange, and more than 30 times red. The automatic tuning function performs multiple rounds of simulations by increasing the number of AGVs, compares the task completion rate and turnaround time of different configurations, and automatically identifies efficiency inflection points.

[0186] A second aspect of an embodiment of the present invention provides an AGV automatic transport vehicle scheduling simulation system, the system comprising:

[0187] The first unit is used to mark the location information of the AGV automatic transport vehicle in the path map to generate an initial path network;

[0188] The second unit is used to collect the running trajectory of the AGV automatic transport vehicle in the initial path network, extract the stop events in the running trajectory, mark the locations where the stop time is longer than the preset time threshold as avoidance nodes, construct buffer areas with the avoidance nodes as the center, and connect adjacent buffer areas to form an avoidance channel network;

[0189] The third unit is used to record the passage sequence of AGVs in the avoidance channel network, use the time intervals between adjacent avoidance nodes as time windows, divide the passage batches according to the distribution pattern of the time windows, and generate a batch scheduling sequence;

[0190] The fourth unit is used to construct a node capacity matrix based on the dwell time of the avoidance node, map the batch scheduling sequence into a traffic resource distribution map, and when a task interruption signal of the AGV automatic transport vehicle is detected, extract the node capacity value of the interruption location, select a backup channel with matching capacity from the traffic resource distribution map, split the interruption task into multiple subtask packages according to the node capacity threshold, and perform time-sharing scheduling through the backup channel.

[0191] According to a third aspect of an embodiment of the present invention, an electronic device is provided, including:

[0192] processor;

[0193] a memory for storing processor-executable instructions;

[0194] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0195] According to a fourth aspect of an embodiment of the present invention, a computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the method described above is implemented.

[0196] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0197] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. AGV automatic transport vehicle scheduling simulation method, characterized in that: include: Mark the location information of the AGV automatic transport vehicle in the path map to generate the initial path network; The AGV's trajectory in the initial path network is collected, and dwell events are extracted. Locations where the dwell time exceeds a preset time threshold are marked as avoidance nodes. Buffer areas are constructed with the avoidance nodes as the center, and adjacent buffer areas are connected to form an avoidance channel network. The passing sequence of AGVs is recorded in the avoidance channel network. The time interval between adjacent avoidance nodes is used as a time window. The passing batches are divided according to the distribution pattern of the time window to generate a batch scheduling sequence. A node capacity matrix is ​​constructed based on the dwell time of the avoidance nodes, and the batch scheduling sequence is mapped into a traffic resource distribution diagram. This includes: collecting the operating data of AGVs in the avoidance channel network, obtaining the real-time number and dwell time of AGVs at each avoidance node, and taking the ratio of the real-time number to the corresponding node capacity in the node capacity matrix as the capacity occupancy rate. The traffic load of the avoidance channel network is obtained by accumulating the duration of the capacity occupancy rate. Dividing the time segments according to the change in the traffic load within a unit time, and dynamically adjusting the length of the time segments based on a preset change threshold, converting the traffic load into a load sequence according to the divided time segments; Extracting the connection relationship between adjacent nodes in the avoidance channel network to construct a basic topology graph, calculating the load difference between adjacent nodes based on the load sequence, and using the load difference as the weight value of the corresponding edge in the basic topology graph to generate a weighted topology graph; Count the AGV passage frequency of each channel corresponding to each edge in the weighted topology graph, calculate the number of AGV flows between adjacent channels, multiply the flow number by the weight value of the corresponding edge to obtain the channel resource occupancy intensity, and update the weight value of the weighted topology graph based on the channel resource occupancy intensity to generate a traffic resource distribution map; When an AGV task interruption signal is detected, the node capacity value of the interruption location is extracted, and a backup channel with matching capacity is selected from the traffic resource distribution map. The interruption task is split into multiple subtask packages according to the node capacity threshold, and time-sharing scheduling is performed through the backup channel.

2. The method according to claim 1, characterized in that Mark the location information of the AGV automatic transport vehicle on the path map to generate the initial path network including: Collect laser scanning positioning coordinates and ultra-wideband positioning coordinates, calculate the corresponding weight coefficient based on the laser scanning error value and the ultra-wideband error value, multiply the weight coefficient with the corresponding positioning coordinates and superimpose them to obtain fused position information; Collect the fused position information of the AGV automatic transport vehicle and calculate the position change between adjacent sampling periods. When the position change is greater than the preset change threshold, determine the main positioning source based on the change direction of the historical fused position information, use the normal positioning information as position compensation information, and update the fused position information based on the position compensation information; Based on the updated fusion position information, a position tracking sequence is constructed, and the position information of the stopping points in the AGV automatic transport vehicle's running path is extracted as the reference point for path construction; Calculate the distance and heading angle between the reference points of the path construction, select the reference points that meet the preset distance threshold and the preset angle threshold to establish a connection relationship, use the minimum spanning tree algorithm to traverse and connect the reference points, and generate an initial path network with directional attributes.

3. The method according to claim 1, characterized in that The AGV's trajectory in the initial path network is collected, and dwell events are extracted. Locations where the dwell time exceeds a preset time threshold are marked as avoidance nodes. Buffer areas are constructed with the avoidance nodes as the center, and adjacent buffer areas are connected to form an avoidance channel network, including: Collect the running trajectory of the AGV automatic transport vehicle in the initial path network, the running trajectory including laser positioning data and ultra-wideband positioning data, perform intelligent fusion processing on the running trajectory, and generate fused trajectory data; Based on the fused trajectory data, the stop events of the AGV automatic transport vehicle are identified, the location information and duration of the stop events are extracted, and the stop events are divided into avoidance waiting type and task loading and unloading type; Based on the classification results of the dwell events, the avoidance necessity score of the dwell location is calculated. The dynamic score threshold is determined based on the system operation status. The dwell locations with avoidance necessity scores higher than the dynamic score threshold are marked as candidate avoidance nodes. When the distance between adjacent candidate avoidance nodes is less than twice the length of the AGV, they are merged into one avoidance node to obtain the final avoidance node. The historical trajectory data of the AGV at the avoidance node location is extracted, and the trajectory density clustering method is used to analyze the avoidance behavior characteristics of the AGV. Based on the avoidance behavior characteristics, the main avoidance direction and avoidance range of each avoidance node are determined. An adaptive region growing algorithm is used to construct a buffer area along the main avoidance direction of the avoidance node. The positional relationship and avoidance direction of adjacent avoidance nodes are analyzed, and the trajectory density maximum path algorithm is used to connect adjacent buffer areas to form an avoidance channel network.

4. The method according to claim 3, characterized in that Along the main avoidance direction of the avoidance node, the adaptive region growing algorithm is used to construct the buffer area. The position relationship and avoidance direction of adjacent avoidance nodes are analyzed, and the trajectory density maximum path algorithm is used to connect adjacent buffer areas to form an avoidance channel network including: The historical avoidance trajectories at the avoidance nodes are vertically projected onto a plane perpendicular to the main avoidance direction to obtain the lateral offset distance of the trajectory points. The kernel density estimation method is used to calculate the probability density distribution of the lateral offset distance to obtain the lateral distribution characteristics of the avoidance trajectory. Along the main avoidance direction of the avoidance node, an adaptive threshold function is set based on the lateral distribution characteristics, and a region growing algorithm is used to construct a buffer area. The lateral growth range of the buffer area at different distances is controlled by the adaptive threshold function; Extract historical avoidance trajectories between adjacent avoidance nodes, calculate the trajectory point density at each spatial position to obtain a trajectory density field, construct a cost function for the avoidance path based on the trajectory density field, solve the cost function to obtain a time field, and determine the centerline path of the avoidance channel based on the gradient direction of the time field; Based on the centerline path and in combination with the lateral extent of the buffer areas at adjacent avoidance nodes, cubic spline interpolation is used to generate the boundary of the avoidance channel, and the buffer areas of all adjacent avoidance nodes are connected to form an avoidance channel network.

5. The method according to claim 1, wherein The passing sequence of AGVs is recorded in the avoidance channel network. The time interval between adjacent avoidance nodes is used as a time window. The passing batches are divided according to the distribution pattern of the time window. The batch scheduling sequence is generated, including: A driving reference point is set at each avoidance node in the avoidance channel network. The reference points between adjacent avoidance nodes are connected to form a motion trajectory chain. The driving deviation of the AGV is determined based on the angle between the velocity vector of the AGV when it passes through the reference point and the trajectory chain. The passing time of the AGV is corrected based on the driving deviation to obtain the corrected passing time sequence. Calculate the time intervals between adjacent AGVs in the corrected passing sequence, establish a cumulative distribution curve of the time intervals, identify clustering intervals by detecting the slope change of the cumulative distribution curve, and classify the AGVs in the same clustering interval into the first passing batch; Calculate the batch stability index for each first-pass batch. The stability index is determined based on the mean and variance of the driving deviations of the AGVs in the batch. Select batches with stability indexes higher than the preset stability threshold as the second-pass batch. Re-cluster the AGVs that are not included in the second-pass batch at different time intervals. The channel occupancy status and AGV automatic transport vehicle distribution density in the avoidance channel network are collected, and the average speed and driving deviation of the AGV automatic transport vehicles in the second passage batch are combined to calculate the avoidance node passing time interval between adjacent batches. The second passage batch is time-series sorted based on the avoidance node passing time interval to generate a batch scheduling sequence.

6. The method according to claim 1, characterized in that A node capacity matrix is ​​constructed based on the dwell time of the avoidance node, and the batch scheduling sequence is mapped to a traffic resource distribution map. When the AGV task interruption signal is detected, the node capacity value at the interruption location is extracted, and a backup channel with matching capacity is selected from the traffic resource distribution map. The interruption task is split into multiple subtask packages according to the node capacity threshold. Time-sharing scheduling through the backup channel includes: A sliding time window is set for each avoidance node in the avoidance channel network. The operation data of AGVs are collected. The ratio of the AGVs' residence time to the number of passes within the time window is used as the node processing capacity indicator. The node capacity matrix is ​​generated based on the physical space dimensions of each avoidance node. The traffic load of the avoidance channel network is calculated based on the node capacity matrix, and the traffic load is discretized into time segments to obtain a load sequence. Based on the load sequence and the topological connection relationship of the avoidance channel network, a traffic resource distribution map is generated. When an AGV task interruption signal is detected, the capacity value of the node at the interruption location is obtained, and the capacity difference between adjacent nodes is calculated based on the node capacity matrix to form a capacity gradient. The capacity gradient is used to determine the set of available channels in the traffic resource distribution map, and the channel with the lowest resource occupancy rate is selected from the set of available channels as the backup channel; Extract the capacity value of each node on the backup channel, set the node with the smallest capacity value as the task package size constraint, split the interrupt task into multiple task packages based on the constraint, and allocate execution time to each task package based on the real-time capacity value of each node; The node load rate is calculated by combining the execution time of the task package with the capacity value of the corresponding node. The execution order of the task package is optimized according to the node load rate to generate a time-sharing scheduling plan.

7. An AGV automatic transport vehicle scheduling simulation system, used to implement the method according to any one of claims 1 to 6, characterized in that: include: The first unit is used to mark the location information of the AGV automatic transport vehicle in the path map to generate an initial path network; The second unit is used to collect the running trajectory of the AGV automatic transport vehicle in the initial path network, extract the stop events in the running trajectory, mark the locations where the stop time is longer than the preset time threshold as avoidance nodes, construct buffer areas with the avoidance nodes as the center, and connect adjacent buffer areas to form an avoidance channel network; The third unit is used to record the passage sequence of AGVs in the avoidance channel network, use the time intervals between adjacent avoidance nodes as time windows, divide the passage batches according to the distribution pattern of the time windows, and generate a batch scheduling sequence; The fourth unit is used to construct a node capacity matrix based on the dwell time of the avoidance node, map the batch scheduling sequence into a traffic resource distribution map, and when a task interruption signal of the AGV automatic transport vehicle is detected, extract the node capacity value of the interruption location, select a backup channel with matching capacity from the traffic resource distribution map, split the interruption task into multiple subtask packages according to the node capacity threshold, and perform time-sharing scheduling through the backup channel.

8. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images