Intelligent logistics unmanned sorting and distribution system based on multi-agv autonomous cooperation

Abstract

The application discloses an intelligent logistics unmanned sorting and distribution system based on multi-AGV autonomous cooperation, which comprises a logistics task data acquisition module, a logistics state graph construction module, an AI task demand prediction module, a multi-AGV autonomous cooperative scheduling module, an intelligent sorting and distribution execution module, a sorting and distribution state monitoring module and an unmanned operation optimization decision module, wherein the logistics task data acquisition module acquires logistics operation basic data; the logistics state graph construction module constructs a logistics operation global state graph; the AI task demand prediction module predicts sorting and distribution demand through an improved FreTS logistics task demand prediction network; the multi-AGV autonomous cooperative scheduling module generates a multi-AGV cooperative scheduling scheme through a rolling time window priority inheritance backtracking algorithm; the intelligent sorting and distribution execution module controls AGVs to execute sorting and distribution; the sorting and distribution state monitoring module monitors the sorting and distribution state of the AGVs; the unmanned operation optimization decision module updates the multi-AGV cooperative scheduling scheme; and the visualized data storage module visually displays and stores data. The improved FreTS logistics task demand prediction network improves the scheduling foresight of the logistics unmanned sorting and distribution.

Description

Technical Field

[0001] This invention relates to the field of intelligent logistics and unmanned warehouse scheduling technology, and in particular to an intelligent logistics unmanned sorting and delivery system based on multi-AGV autonomous collaboration. Background Technology

[0002] With the increasing demand for intelligent logistics, unmanned warehousing, and multi-AGV collaborative operations, automated scheduling, path planning, and operation monitoring technologies for sorting and delivery scenarios have received widespread attention. Existing intelligent logistics systems mainly rely on order task rules, fixed path planning, or single AGV status feedback for sorting and delivery control, but in practical applications, the following problems are commonly encountered:

[0003] The sources of basic data for logistics operations are complex, and there is a lack of unified correlation between sorting and delivery task data, cargo status data, AGV operation status data, warehouse area status data, and delivery terminal status data. This makes it difficult to form a global status description of the operational relationships between tasks, cargo, AGVs, aisle segments, and delivery terminals. Multi-AGV task allocation typically relies on static sorting based on current task priority or path distance, making it difficult to proactively schedule tasks based on future sorting and delivery needs, AGV resource requirements, and aisle congestion risks. This leads to uneven AGV resource utilization during peak periods, aisle congestion, and increased pressure on delivery terminals. Existing path planning methods mostly address single-vehicle paths or local obstacle avoidance, lacking the ability to collaboratively resolve conflicts such as occupancy of the same aisle segment, conflict at the same intersection, oncoming traffic conflicts, and rear-end collision conflicts. This easily leads to problems such as increased waiting time, repeated path replanning, and decreased sorting and delivery efficiency.

[0004] Therefore, how to provide an intelligent logistics unmanned sorting and delivery system based on multi-AGV autonomous collaboration is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] One objective of this invention is to propose an intelligent unmanned sorting and delivery system based on multi-AGV autonomous collaboration. This invention fully utilizes the improved FreTS logistics task demand prediction network and the rolling window priority inheritance backtracking algorithm, and describes in detail the system processing flow for intelligently realizing logistics task demand prediction, multi-AGV task allocation, candidate driving path generation, and path conflict resolution. It has the advantages of strong scheduling foresight, high path conflict resolution efficiency, good AGV collaborative operation stability, and high unmanned sorting and delivery efficiency.

[0006] The intelligent logistics unmanned sorting and delivery system based on multi-AGV autonomous collaboration according to an embodiment of the present invention includes the following modules:

[0007] Logistics task data acquisition module: used to acquire basic logistics operation data in the target logistics scenario;

[0008] Logistics Status Diagram Construction Module: This module is used to construct a global status diagram of logistics operations based on basic logistics operation data, and to record status snapshots of the global status diagram of logistics operations according to statistical periods, generating a logistics operation status snapshot sequence.

[0009] AI task demand prediction module: used to extract indicators from the snapshot sequence of logistics operation status, and perform frequency domain time series prediction through the improved FreTS logistics task demand prediction network to obtain the sorting and delivery demand prediction results; the improved FreTS logistics task demand prediction network includes a logistics status embedding mapping layer, a phase difference constrained frequency channel learner, a peak and valley suppression frequency time learner, and a prediction projection layer.

[0010] Multi-AGV autonomous collaborative scheduling module: Based on the global state diagram of logistics operations and the results of sorting and delivery demand prediction, it uses a rolling time window priority inheritance backtracking algorithm to perform task allocation, path planning and path conflict resolution for multiple AGVs, and generates a multi-AGV collaborative scheduling scheme.

[0011] Intelligent sorting and delivery execution module: used to control each AGV to perform sorting and delivery according to the multi-AGV collaborative scheduling scheme, and update the execution status;

[0012] Sorting and delivery status monitoring module: used to monitor the sorting and delivery execution status of multiple AGVs and generate sorting and delivery operation status results;

[0013] Unmanned operation optimization decision module: used to update the multi-AGV collaborative scheduling scheme based on the sorting and delivery operation status results;

[0014] Visualization and data storage module: used to visualize the system's operating results and write them into the logistics management database.

[0015] Optionally, the basic logistics operation data includes sorting and delivery task data, cargo status data, AGV operation status data, warehouse area status data, and delivery terminal status data;

[0016] The sorting and delivery task data includes order number, goods identification, task type, task generation time, task priority, pickup location, sorting target compartment, and delivery terminal node;

[0017] The cargo status data includes cargo identification, current cargo location, cargo weight, cargo category, and cargo sorting status;

[0018] The AGV operating status data includes AGV number, AGV current location, AGV power status, AGV load status, AGV task execution status, and AGV availability status;

[0019] The warehouse area status data includes the occupancy status of the sorting area, the occupancy status of the buffer area, the occupancy status of the aisle sections, the occupancy status of the intersection nodes, the connection relationship of the aisle sections, and the passage direction of the aisle sections;

[0020] The delivery terminal status data includes the availability status of the delivery handover area, the availability status of the loading and unloading station, the availability status of the delivery station, and the full load status of the sorting compartment.

[0021] Optionally, the logistics status diagram construction module specifically includes:

[0022] The basic data of logistics operations are parsed and mapped into nodes according to data type to construct sorting and delivery task nodes, cargo status nodes, AGV status nodes, warehouse area status nodes and delivery terminal status nodes.

[0023] Based on the order number and the goods identifier, establish a task-goods binding edge between the sorting and delivery task node and the goods status node;

[0024] Based on the pickup location, current location of goods, sorting target slot and delivery terminal node, establish operation path association edges between sorting and delivery task nodes, goods status nodes, warehouse area status nodes and delivery terminal status nodes;

[0025] Based on the current position of the AGV, the load status of the AGV, the task execution status of the AGV, and the availability status of the AGV, establish AGV task candidate edges between the AGV status nodes and the sorting and delivery task nodes.

[0026] Based on the status of channel occupancy, intersection occupancy, sorting grid full load, delivery handover area availability, loading and unloading station availability, and delivery station availability, establish an operation acceptance status edge between the warehouse area status node and the delivery terminal status node.

[0027] The sorting and delivery task nodes, cargo status nodes, AGV status nodes, warehouse area status nodes, delivery terminal status nodes, task cargo binding edges, operation path association edges, AGV task candidate edges, and operation acceptance status edges are organized into a graph structure to construct a global state graph of logistics operations.

[0028] Record the status of graph nodes and edges in the global status graph of logistics operations according to the statistical period, generate a logistics operation status snapshot, and arrange the logistics operation status snapshots in the order of the statistical period to generate a logistics operation status snapshot sequence.

[0029] Optionally, the AI ​​task demand prediction module specifically includes:

[0030] Based on the snapshot sequence of logistics operation status, the status of graph nodes and edge status is read according to the statistical period, and logistics status indicators are extracted from the status of graph nodes and edge status.

[0031] The logistics status indicators are arranged according to the statistical period and the logistics status indicators to generate a logistics operation status tensor.

[0032] The logistics operation state tensor is embedded and mapped by the logistics state embedding mapping layer to generate the logistics state embedding tensor.

[0033] In the phase difference constrained frequency channel learner, the logistics state embedded tensor is subjected to discrete Fourier transform along the logistics state index dimension to obtain the channel complex spectrum tensor.

[0034] Based on the channel complex spectrum tensor, the channel spectrum amplitude tensor and the channel spectrum phase tensor are extracted, and the channel phase difference tensor is generated according to the difference between the channel spectrum phase tensors corresponding to any two logistics status indicators.

[0035] The channel amplitude scaling factor tensor is obtained by performing mean normalization, amplitude linear mapping and sigmoid activation on the channel spectrum amplitude tensor.

[0036] The channel phase difference tensor is subjected to sine and cosine encoding, phase linear mapping and Tanh activation to obtain the channel phase offset tensor.

[0037] The amplitude of the channel spectrum amplitude tensor is adjusted using the channel amplitude scaling factor tensor to obtain the enhanced channel amplitude tensor;

[0038] The channel spectral phase tensor is phase-adjusted using the channel phase offset tensor to obtain the enhanced channel phase tensor;

[0039] The enhanced channel amplitude tensor and enhanced channel phase tensor are converted into complex frequency domain representations containing real and imaginary features to obtain the channel complex frequency enhanced feature tensor;

[0040] Perform an inverse frequency domain transform on the channel complex frequency enhancement feature tensor to obtain the channel enhancement feature tensor;

[0041] In the peak-valley suppression frequency time learner, the channel enhancement feature tensor is subjected to discrete Fourier transform along the statistical period dimension to obtain the time spectrum tensor;

[0042] Based on the amplitude variation intensity of each time frequency component in the time spectrum tensor, the dominant frequency component tensor and the peak-valley perturbation component tensor are determined. The dominant frequency component tensor is subjected to frequency domain enhancement mapping, and the peak-valley perturbation component tensor is subjected to frequency domain suppression mapping to obtain the recalibrated time-frequency domain feature tensor.

[0043] Perform an inverse frequency domain transform on the recalibrated time-frequency domain feature tensor to obtain the time-enhanced feature tensor;

[0044] In the prediction projection layer, the time-enhanced feature tensor is mapped to a multi-head linear prediction to obtain the sorting and delivery demand prediction results.

[0045] The sorting and delivery demand forecast results include the forecast values ​​of sorting task demand, delivery task demand, AGV resource demand, channel congestion risk, and delivery terminal occupancy for future scheduling cycles.

[0046] Optionally, the multi-AGV autonomous collaborative scheduling module specifically includes:

[0047] Construct an AGV task candidate set, which includes multiple AGV task candidate combinations. Each AGV task candidate combination includes AGV number, target task, pickup location, sorting target slot, delivery terminal node, AGV current location, AGV power status, AGV load status, AGV task execution status, and AGV availability status.

[0048] Based on the sorting and delivery demand forecast results, the task scheduling score for each AGV task candidate combination is calculated;

[0049] Based on the task scheduling scores from high to low, the candidate combinations of each AGV task are sorted to obtain the AGV execution priority queue.

[0050] The current scheduling time is used as the starting point of the rolling window, and the current rolling window is generated according to the window length;

[0051] Based on the global state diagram of logistics operations, within the current rolling time window, candidate driving paths for each AGV task candidate combination are generated through A-star search.

[0052] Time window occupancy analysis is performed on each candidate driving route to determine the time window and location of the conflict corresponding to the path conflict, and to obtain the set of path conflicts;

[0053] The time window occupancy analysis includes: identifying occupancy conflicts on the same road segment, occupancy conflicts at the same intersection, oncoming traffic conflicts, and rear-end collision conflicts;

[0054] Based on the AGV execution priority queue, priority inheritance processing is performed on the path conflict set, specifically as follows:

[0055] AGVs that participate in the same path conflict and are ranked higher in the AGV execution priority queue are identified as high-priority AGVs, while AGVs ranked lower are identified as low-priority AGVs.

[0056] Determine whether a low-priority AGV occupies a conflict position in the candidate driving path corresponding to a high-priority AGV within the conflict occurrence time window;

[0057] If a low-priority AGV occupies a conflicting position, the passage priority of the high-priority AGV is transferred to the low-priority AGV, and the avoidance action of the low-priority AGV is updated.

[0058] Based on the updated avoidance action, the candidate driving path of the low-priority AGV is corrected to obtain the priority inheritance processing result.

[0059] If path conflicts still exist in the priority inheritance processing results, the low-priority AGVs will be backtracked and the candidate actions will be verified in the order of waiting, yielding, partial detour and candidate driving path rearrangement to obtain the backtracking processing results that satisfy the channel segment occupancy status, intersection node occupancy status and channel segment travel direction.

[0060] Based on the priority inheritance processing results and the backtracking processing results, the target task, picking location, sorting target slot, delivery terminal node, driving path, arrival time and avoidance actions of each AGV are determined, and a multi-AGV collaborative scheduling scheme is generated by combining them.

[0061] Optionally, the intelligent sorting and delivery execution module specifically includes:

[0062] Based on the multi-AGV collaborative scheduling scheme, the AGV number, target task, pickup location, sorting target slot, delivery terminal node, driving route, arrival time and avoidance actions of each AGV are read.

[0063] Control each AGV to reach the pickup location according to the driving route and perform cargo pickup;

[0064] The AGV that has completed the pickup of goods performs sorting and transfer according to the travel path, and transfers the goods to the target sorting slot.

[0065] Control the AGVs that have completed sorting and transfer to perform delivery handover, and return to the standby position after the delivery handover is completed;

[0066] If an AGV malfunction, path blockage, cargo recognition failure, full sorting slot, or abnormal occupancy of a delivery terminal node is detected, the AGV that is malfunctioning will be identified as an abnormal AGV, the current execution action of the abnormal AGV will be suspended, and the abnormality type and abnormality location will be recorded.

[0067] Based on the anomaly type and location, update the AGV task execution status of the abnormal AGV and the corresponding cargo sorting status, and feed back the anomaly type, anomaly location, AGV number of the abnormal AGV, and corresponding target task to the sorting and delivery status monitoring module.

[0068] Optionally, the sorting and delivery status monitoring module specifically includes:

[0069] Read the AGV task execution status of each AGV, the cargo sorting status of each cargo, the channel occupancy status of each channel segment, and the intersection node occupancy status of each intersection node.

[0070] Based on the AGV task execution status, travel path, and arrival time, determine whether the AGV has deviated from the travel path, timed out of arrival, or stopped executing.

[0071] Based on the cargo sorting status, determine whether the cargo has completed the cargo pickup, sorting and transfer, and delivery handover process;

[0072] Based on the occupancy status of the passageway and intersection, determine whether there is congestion on the passageway or abnormal occupancy at the intersection;

[0073] The sorting and delivery efficiency is statistically analyzed, and the AGV task execution status, cargo sorting status, aisle segment occupancy status, intersection node occupancy status, and sorting and delivery efficiency are combined into a sorting and delivery operation status result.

[0074] Optionally, the unmanned operation optimization decision module specifically includes:

[0075] Based on the sorting and delivery operation status results, determine whether the scheduling update conditions are met;

[0076] If the scheduling update conditions are met, the sorting and delivery operation status results will be fed back to the multi-AGV autonomous collaborative scheduling module to generate an updated multi-AGV collaborative scheduling scheme.

[0077] The scheduling update conditions include: the AGV task execution status of the abnormal AGV is paused, the sorting status of the goods corresponding to the target task is incomplete, the channel segment occupancy status is continuously occupied, the intersection node occupancy status is abnormally occupied, the current rolling window has ended, or new sorting and delivery task data has been received.

[0078] Optionally, the system operation results include sorting and delivery demand prediction results, multi-AGV collaborative scheduling scheme, sorting and delivery operation status results, and updated multi-AGV collaborative scheduling scheme;

[0079] The system operation results are visualized on the scheduling dashboard, and the basic data of logistics operations, the global status diagram of logistics operations, the sequence of logistics operation status snapshots, and the system operation results are written into the logistics management database in batches.

[0080] The beneficial effects of this invention are:

[0081] This invention's system acquires basic logistics operation data from the target logistics scenario and maps sorting and delivery task data, cargo status data, AGV operating status data, warehouse area status data, and delivery terminal status data into nodes and organizes them into a graph structure to construct a global logistics operation status graph. This graph provides a unified expression of the binding relationships, path relationships, candidate relationships, and acceptance relationships between current tasks, cargo, AGVs, warehouse areas, and delivery terminals, improving the global perception capability of multi-source operation status in unmanned sorting and delivery scenarios. Furthermore, it generates a logistics operation status snapshot sequence according to a statistical period and performs frequency domain time-series prediction using an improved FreTS logistics task demand prediction network. The phase difference constraint frequency channel learner characterizes the phase difference relationship between different logistics status indicators, while the peak-valley suppression frequency time learner preserves the main frequency change and suppresses peak-valley disturbances. This allows for advance prediction of sorting task demand, delivery task demand, AGV resource demand, channel congestion risk, and delivery terminal occupancy, improving the foresight and predictive stability of scheduling decisions. During the multi-AGV autonomous collaborative scheduling phase, a rolling time window priority inheritance backtracking algorithm is used to generate an AGV execution priority queue based on the sorting and delivery demand prediction results. Within the current rolling time window, candidate driving paths are generated and path conflict resolution is completed by combining A-star search, time window occupancy analysis, priority inheritance processing, and backtracking candidate action verification. This reduces the impact of occupancy conflicts on the same channel segment, the same intersection node, oncoming traffic conflicts, and rear-end collision conflicts on sorting and delivery efficiency. Simultaneously, an anomaly feedback and scheduling update closed loop is formed through intelligent sorting and delivery execution, sorting and delivery status monitoring, and unmanned operation optimization decisions. Therefore, the system of this invention can improve the accuracy of multi-AGV collaborative scheduling, the efficiency of path conflict resolution, the continuity of unmanned sorting and delivery, and the operational stability of the intelligent logistics system. Attached Figure Description

[0082] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0083] Figure 1 This is a schematic diagram of the modules of the intelligent logistics unmanned sorting and delivery system based on multi-AGV autonomous collaboration proposed in this invention;

[0084] Figure 2 This is a flowchart of the improved FreTS logistics task demand prediction network structure in the intelligent logistics unmanned sorting and delivery system based on multi-AGV autonomous collaboration proposed in this invention.

[0085] Figure 3 This is a flowchart of the rolling time window priority inheritance backtracking algorithm in the intelligent logistics unmanned sorting and delivery system based on multi-AGV autonomous collaboration proposed in this invention. Detailed Implementation

[0086] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0087] refer to Figures 1-3 The intelligent logistics unmanned sorting and delivery system based on multi-AGV autonomous collaboration includes the following modules:

[0088] Logistics task data acquisition module: used to acquire basic logistics operation data in the target logistics scenario;

[0089] Logistics Status Diagram Construction Module: This module is used to construct a global status diagram of logistics operations based on basic logistics operation data, and to record status snapshots of the global status diagram of logistics operations according to statistical periods, generating a logistics operation status snapshot sequence.

[0090] AI task demand prediction module: used to extract indicators from the snapshot sequence of logistics operation status, and to perform frequency domain time series prediction through the improved FreTS logistics task demand prediction network to obtain the sorting and delivery demand prediction results; the improved FreTS logistics task demand prediction network includes a logistics status embedding mapping layer, a phase difference constrained frequency channel learner, a peak and valley suppression frequency time learner, and a prediction projection layer.

[0091] Multi-AGV autonomous collaborative scheduling module: Based on the global state diagram of logistics operations and the results of sorting and delivery demand prediction, it uses a rolling time window priority inheritance backtracking algorithm to perform task allocation, path planning and path conflict resolution for multiple AGVs, and generates a multi-AGV collaborative scheduling scheme.

[0092] Intelligent sorting and delivery execution module: used to control each AGV to perform sorting and delivery according to the multi-AGV collaborative scheduling scheme, and update the execution status;

[0093] Sorting and delivery status monitoring module: used to monitor the sorting and delivery execution status of multiple AGVs and generate sorting and delivery operation status results;

[0094] Unmanned operation optimization decision module: used to update the multi-AGV collaborative scheduling scheme based on the sorting and delivery operation status results;

[0095] Visualization and data storage module: used to visualize the system's operating results and write them into the logistics management database.

[0096] In this embodiment, the basic data for logistics operations includes sorting and delivery task data, cargo status data, AGV operation status data, warehouse area status data, and delivery terminal status data.

[0097] Sorting and delivery task data includes order number, goods identification, task type, task generation time, task priority, pickup location, sorting target compartment, and delivery terminal node;

[0098] Cargo status data includes cargo identification, current cargo location, cargo weight, cargo category, and cargo sorting status;

[0099] AGV operating status data includes AGV number, AGV current position, AGV power status, AGV load status, AGV task execution status, and AGV availability status;

[0100] The warehouse area status data includes the occupancy status of the sorting area, the occupancy status of the buffer area, the occupancy status of the aisle sections, the occupancy status of the intersection nodes, the connection relationship of the aisle sections, and the passage direction of the aisle sections;

[0101] The status data of the delivery terminal includes the availability status of the delivery handover area, the availability status of the loading and unloading station, the availability status of the delivery station, and the full load status of the sorting slot.

[0102] In this embodiment, the logistics status diagram construction module specifically includes:

[0103] The basic data of logistics operations are parsed and mapped into nodes according to data type to construct sorting and delivery task nodes, cargo status nodes, AGV status nodes, warehouse area status nodes and delivery terminal status nodes.

[0104] Based on order number and cargo identifier, establish task-cargo binding edges between sorting and delivery task nodes and cargo status nodes; based on pickup location, current cargo location, sorting target slot, and delivery terminal node, establish operation path association edges between sorting and delivery task nodes, cargo status nodes, warehouse area status nodes, and delivery terminal status nodes; based on AGV current location, AGV load status, AGV task execution status, and AGV availability status, establish AGV task candidate edges between AGV status nodes and sorting and delivery task nodes.

[0105] Based on the status of channel occupancy, intersection occupancy, sorting grid full load, delivery handover area availability, loading and unloading station availability, and delivery station availability, establish an operation acceptance status edge between the warehouse area status node and the delivery terminal status node.

[0106] The sorting and delivery task nodes, cargo status nodes, AGV status nodes, warehouse area status nodes, delivery terminal status nodes, task cargo binding edges, operation path association edges, AGV task candidate edges, and operation acceptance status edges are organized into a graph structure to construct a global state graph of logistics operations.

[0107] Record the status of graph nodes and edges in the global status graph of logistics operations according to the statistical period, generate a logistics operation status snapshot, and arrange the logistics operation status snapshots in the order of the statistical period to generate a logistics operation status snapshot sequence.

[0108] In this embodiment, the AI ​​task demand prediction module specifically includes:

[0109] Based on the snapshot sequence of logistics operation status, the status of graph nodes and edge status is read according to the statistical period, and logistics status indicators are extracted from the status of graph nodes and edge status. The logistics status indicators include the sorting task volume obtained by statistics based on task type, task generation time and sorting target grid; the delivery task volume obtained by statistics based on task type, task generation time and delivery terminal node; the AGV occupancy rate obtained by statistics based on AGV number, AGV task execution status and AGV availability status; the channel congestion rate obtained by statistics based on channel segment occupancy status and intersection node occupancy status; and the delivery terminal occupancy rate obtained by statistics based on the availability status of delivery handover area, loading and unloading station, and delivery station.

[0110] The logistics status indicators are arranged according to the statistical period and the logistics status indicators to generate a logistics operation status tensor.

[0111] The logistics operation state tensor is embedded and mapped by the logistics state embedding mapping layer to generate the logistics state embedding tensor.

[0112] In the phase difference constrained frequency channel learner, the logistics state embedded tensor is subjected to discrete Fourier transform along the logistics state index dimension to obtain the channel complex spectrum tensor.

[0113] Based on the channel complex spectrum tensor, the channel spectrum amplitude tensor and the channel spectrum phase tensor are extracted, and the channel phase difference tensor is generated according to the difference between the channel spectrum phase tensors corresponding to any two logistics status indicators.

[0114] The channel amplitude scaling factor tensor is obtained by normalizing the channel spectrum amplitude tensor, applying amplitude linear mapping and sigmoid activation; the channel phase difference tensor is obtained by applying sine and cosine encoding, phase linear mapping and tanh activation.

[0115] The amplitude of the channel spectrum amplitude tensor is adjusted using the channel amplitude scaling factor tensor to obtain the enhanced channel amplitude tensor; the phase of the channel spectrum phase tensor is adjusted using the channel phase offset tensor to obtain the enhanced channel phase tensor.

[0116] The enhanced channel amplitude tensor and enhanced channel phase tensor are converted into complex frequency domain representations containing real and imaginary features to obtain the channel complex frequency enhanced feature tensor;

[0117] Perform an inverse frequency domain transform on the channel complex frequency enhancement feature tensor to obtain the channel enhancement feature tensor;

[0118] In the peak-valley suppression frequency time learner, the channel enhancement feature tensor is subjected to discrete Fourier transform along the statistical period dimension to obtain the time spectrum tensor;

[0119] Based on the amplitude variation intensity of each time frequency component in the time spectrum tensor, the dominant frequency component tensor and the peak-valley perturbation component tensor are determined. The dominant frequency component tensor is subjected to frequency domain enhancement mapping, and the peak-valley perturbation component tensor is subjected to frequency domain suppression mapping to obtain the recalibrated time-frequency domain feature tensor.

[0120] Perform an inverse frequency domain transform on the recalibrated time-frequency domain feature tensor to obtain the time-enhanced feature tensor;

[0121] In the prediction projection layer, the time-enhanced feature tensor is mapped to a multi-head linear prediction to obtain the sorting and delivery demand prediction results.

[0122] The sorting and delivery demand forecast results include forecasts of sorting task demand, delivery task demand, AGV resource demand, channel congestion risk, and delivery terminal occupancy for future scheduling cycles.

[0123] In this invention, the improved FreTS logistics task demand prediction network has the same main structure as the standard FreTS network. The standard FreTS network typically first embeds or expands the dimensions of the multivariate time series, then converts the time-domain signal into a complex spectral representation using Discrete Fourier Transform, and performs frequency domain learning in both the channel and time dimensions. Specifically, the frequency channel learner learns the channel dependencies between different variables, the frequency time learner learns the time dependencies in the statistical period dimension, the frequency domain MLP maps the real and imaginary features in the complex spectrum, and finally, the prediction result for the future time step is output through a prediction projection layer.

[0124] The improved FreTS logistics task demand prediction network makes structural improvements to the internal frequency domain learning structure of the standard FreTS network, specifically tailored to logistics task demand prediction scenarios. In the frequency channel learner, the standard FreTS method of uniform complex MLP mapping of the channel complex spectrum is improved into a phase difference-constrained frequency channel learning structure. First, the channel spectrum amplitude tensor and channel spectrum phase tensor are extracted from the channel complex spectrum tensor. Then, a channel phase difference tensor is generated based on the difference between the channel spectrum phase tensors corresponding to any two logistics status indicators. Through mean normalization, amplitude linear mapping, Sigmoid activation, sine / cosine coding, phase linear mapping, and Tanh activation, the channel amplitude scaling factor tensor and channel phase offset tensor are obtained, thereby structurally adjusting the channel spectrum amplitude and channel spectrum phase. In the frequency-time learner, the standard FreTS method of performing a unified frequency-domain MLP mapping on the time spectrum is improved into a peak-valley suppressed frequency-time learning structure: based on the amplitude change intensity of each time frequency component in the time spectrum tensor, the main frequency component tensor and the peak-valley perturbation component tensor are distinguished, and frequency-domain enhancement mapping and frequency-domain suppression mapping are performed respectively to obtain the recalibrated time-frequency domain feature tensor.

[0125] By employing a phase difference-constrained frequency channel learner, the improved FreTS logistics task demand prediction network can characterize the phase difference relationship in the frequency domain among sorting task volume, delivery task volume, AGV occupancy rate, channel congestion rate, and delivery terminal occupancy rate. This allows the lagged impact of changes in sorting demand on AGV resource demand, channel congestion risk, and delivery terminal occupancy changes to be incorporated into the prediction process, thereby enhancing the expressive ability of the coupling relationship between multiple logistics status indicators. Through a peak-valley suppression frequency-time learner, the network can retain the dominant frequency variation pattern of logistics operation status within future scheduling cycles, while suppressing peak-valley disturbances caused by sudden order concentrations, short-term channel occupancy, abnormal occupancy at intersection nodes, and short-term full loads at delivery terminals, reducing the interference of abnormal fluctuations on the prediction results. Through these structural improvements, the predicted values ​​of sorting task demand, delivery task demand, AGV resource demand, channel congestion risk, and delivery terminal occupancy can more stably reflect the actual operational pressure of future scheduling cycles.

[0126] In this embodiment, the multi-AGV autonomous collaborative scheduling module specifically includes:

[0127] Construct an AGV task candidate set, which includes multiple AGV task candidate combinations. Each AGV task candidate combination includes AGV number, target task, pickup location, sorting target slot, delivery terminal node, AGV current location, AGV power status, AGV load status, AGV task execution status, and AGV availability status.

[0128] Based on the sorting and delivery demand forecast results, the task scheduling score of each AGV task candidate combination is calculated. Specifically, the sorting task demand forecast, delivery task demand forecast, AGV resource demand forecast, channel congestion risk forecast, and delivery terminal occupancy forecast values ​​of the corresponding work area of ​​each AGV task candidate combination are read. The sorting task demand forecast, delivery task demand forecast, and AGV resource demand forecast values ​​are converted into positive demand scores, and the channel congestion risk forecast and delivery terminal occupancy forecast values ​​are converted into reverse passage acceptance scores. The positive demand scores and reverse passage acceptance scores are then merged to obtain the task scheduling score of each AGV task candidate combination.

[0129] Based on the task scheduling scores from high to low, the candidate combinations of each AGV task are sorted to obtain the AGV execution priority queue.

[0130] The current scheduling time is used as the starting point of the rolling window, and the current rolling window is generated according to the window length;

[0131] Based on the global state diagram of logistics operations, within the current rolling time window, candidate driving paths for each AGV task candidate combination are generated through A-star search.

[0132] Time window occupancy analysis is performed on each candidate driving route to determine the time window and location of the conflict corresponding to the path conflict, and to obtain the set of path conflicts;

[0133] The time window occupancy analysis includes: identifying occupancy conflicts on the same road segment, occupancy conflicts at the same intersection, oncoming traffic conflicts, and rear-end collision conflicts;

[0134] Based on the AGV execution priority queue, priority inheritance processing is performed on the path conflict set, specifically as follows:

[0135] AGVs that participate in the same path conflict and are ranked higher in the AGV execution priority queue are identified as high-priority AGVs, while AGVs ranked lower are identified as low-priority AGVs.

[0136] Determine whether a low-priority AGV occupies a conflict position in the candidate driving path corresponding to a high-priority AGV within the conflict occurrence time window;

[0137] If a low-priority AGV occupies a conflicting position, the passage priority of the high-priority AGV is transferred to the low-priority AGV, and the avoidance action of the low-priority AGV is updated.

[0138] Based on the updated avoidance action, the candidate driving path of the low-priority AGV is corrected to obtain the priority inheritance processing result.

[0139] For example, the current scrolling window is from 10:00:00 to 10:03:00, and the AGV execution priority queue is AGV3, AGV1, AGV5, and AGV2. AGV1's candidate travel path occupies intersection node C6 from 10:01:20 to 10:01:35, and AGV5's candidate travel path occupies intersection node C6 from 10:01:25 to 10:01:40. Time window occupancy analysis determines that the two have a conflict over the same intersection node, with the conflict occurring from 10:01:25 to 10:01:35, and the conflict location being intersection node C6. Since AGV1 has a higher priority position in the AGV execution priority queue, AGV1 is identified as a high-priority AGV, and AGV5 is identified as a low-priority AGV. Since AGV5 occupies the conflict location in AGV1's corresponding candidate travel path within the conflict time window, AGV1's passage priority is transferred to AGV5, and AGV5's avoidance action is updated to waiting, adjusting the time for AGV5 to pass through intersection node C6 to 10:01:35 to 10:01:50, thus obtaining the priority inheritance processing result.

[0140] If path conflicts still exist in the priority inheritance processing results, the low-priority AGVs will be backtracked and the candidate actions will be verified in the order of waiting, yielding, partial detour and candidate driving path rearrangement to obtain the backtracking processing results that satisfy the channel segment occupancy status, intersection node occupancy status and channel segment travel direction.

[0141] Based on the priority inheritance processing results and the backtracking processing results, the target task, picking location, sorting target slot, delivery terminal node, driving path, arrival time and avoidance actions of each AGV are determined, and a multi-AGV collaborative scheduling scheme is generated by combining them.

[0142] In this embodiment, the intelligent sorting and delivery execution module specifically includes:

[0143] Based on the multi-AGV collaborative scheduling scheme, the AGV number, target task, pickup location, sorting target slot, delivery terminal node, driving route, arrival time and avoidance actions of each AGV are read.

[0144] Control each AGV to reach the pickup location according to the driving route and perform cargo pickup;

[0145] The AGV that has completed the pickup of goods performs sorting and transfer according to the travel path, and transfers the goods to the target sorting slot.

[0146] Control the AGVs that have completed sorting and transfer to perform delivery handover, and return to the standby position after the delivery handover is completed;

[0147] If an AGV malfunction, path blockage, cargo recognition failure, full sorting slot, or abnormal occupancy of a delivery terminal node is detected, the AGV that is malfunctioning will be identified as an abnormal AGV, the current execution action of the abnormal AGV will be suspended, and the abnormality type and abnormality location will be recorded.

[0148] Based on the anomaly type and location, update the AGV task execution status of the abnormal AGV and the corresponding cargo sorting status, and feed back the anomaly type, anomaly location, AGV number of the abnormal AGV, and corresponding target task to the sorting and delivery status monitoring module.

[0149] In this embodiment, the sorting and delivery status monitoring module specifically includes:

[0150] Read the AGV task execution status of each AGV, the cargo sorting status of each cargo, the channel occupancy status of each channel segment, and the intersection node occupancy status of each intersection node.

[0151] Based on the AGV task execution status, travel path, and arrival time, determine whether the AGV has deviated from the travel path, timed out, or stopped; based on the cargo sorting status, determine whether the cargo has completed the pickup, sorting, transfer, and delivery handover; based on the channel segment occupancy status and intersection node occupancy status, determine whether there is channel segment congestion or intersection node occupancy anomalies.

[0152] The sorting and delivery efficiency is statistically analyzed, and the AGV task execution status, cargo sorting status, aisle segment occupancy status, intersection node occupancy status, and sorting and delivery efficiency are combined into a sorting and delivery operation status result.

[0153] In this embodiment, the unmanned operation optimization decision module specifically includes:

[0154] Based on the sorting and delivery operation status results, it is determined whether the scheduling update conditions are met. If the scheduling update conditions are met, the sorting and delivery operation status results are fed back to the multi-AGV autonomous collaborative scheduling module to generate an updated multi-AGV collaborative scheduling scheme. The scheduling update conditions include: the AGV task execution status of the abnormal AGV is paused, the sorting status of the goods corresponding to the target task is incomplete, the channel segment occupancy status is continuously occupied, the intersection node occupancy status is abnormally occupied, the current rolling time window has ended, or new sorting and delivery task data has been received.

[0155] In this embodiment, the system operation results include sorting and delivery demand prediction results, multi-AGV collaborative scheduling scheme, sorting and delivery operation status results, and updated multi-AGV collaborative scheduling scheme;

[0156] The system operation results are visualized on the scheduling dashboard, and the basic data of logistics operations, the global status diagram of logistics operations, the sequence of logistics operation status snapshots, and the system operation results are written into the logistics management database in batches.

[0157] Example 1: To verify the feasibility of this invention in practice, it was applied to an unmanned sorting and delivery operation scenario in an e-commerce warehousing and distribution center. This center includes a sorting area, a buffer area, main aisle sections, intersection nodes, a delivery handover area, and multiple delivery terminal nodes. It is equipped with 48 AGVs and processes approximately 42,000 orders daily. Existing operations typically allocate AGVs according to the order generation sequence or a fixed priority. When orders arrive in concentrated bursts, sorting slots are near full capacity, and intersection nodes are continuously occupied, problems such as aisle segment contention, conflicting traffic flow, excessively long AGV waiting times, and uneven delivery terminal load can easily occur.

[0158] During the implementation of this invention system, the logistics task data acquisition module acquires sorting and delivery task data, cargo status data, AGV operation status data, warehouse area status data, and delivery terminal status data. The logistics status graph construction module performs field parsing and node mapping on the basic logistics operation data, constructing sorting and delivery task nodes, cargo status nodes, AGV status nodes, warehouse area status nodes, and delivery terminal status nodes. It also establishes task-cargo binding edges, operation path association edges, AGV task candidate edges, and operation acceptance status edges, forming a global logistics operation status graph. The system records the graph node and edge states at a 3-minute statistical cycle, generating a logistics operation status snapshot sequence.

[0159] The AI ​​task demand prediction module extracts sorting task volume, delivery task volume, AGV occupancy rate, channel congestion rate, and delivery terminal occupancy rate from the logistics operation status snapshot sequence, generates a logistics operation status tensor, and performs frequency domain time series prediction through an improved FreTS logistics task demand prediction network to obtain predicted values ​​for sorting task demand, delivery task demand, AGV resource demand, channel congestion risk, and delivery terminal occupancy. The multi-AGV autonomous collaborative scheduling module calculates the task scheduling score of AGV task candidate combinations based on the sorting and delivery demand prediction results, generates an AGV execution priority queue, and generates candidate driving paths through A* search within the current rolling time window.

[0160] During path conflict resolution, the system performs time window occupancy analysis on candidate travel paths, identifying occupancy conflicts on the same passage segment, the same intersection node, oncoming traffic conflicts, and rear-end collision conflicts. When AGV12 and AGV27 collide at intersection N6, the system determines AGV12 as a high-priority AGV and AGV27 as a low-priority AGV based on the AGV execution priority queue, and transfers AGV12's passage priority to AGV27, causing AGV27 to either wait or partially detour. If path conflicts still exist, the system continues to backtrack and verify candidate actions in the order of waiting, yielding, partial detour, and candidate path rearrangement. Subsequently, the intelligent sorting and delivery execution module controls each AGV to perform sorting and delivery, the sorting and delivery status monitoring module monitors the sorting and delivery execution status, and the unmanned operation optimization decision module triggers scheduling updates when the current rolling time window ends or an abnormal AGV suspends execution.

[0161] To compare and analyze the implementation effect of the system of the present invention, the system of the present invention is compared and analyzed with the fixed priority scheduling scheme and the standard A-star path planning scheme. The comparison effect is shown in Table 1.

[0162] Table 1. Comparison of the effects of different scheduling schemes on peak-hour operations in warehousing and distribution centers.

[0163] Accuracy of sorting task demand forecasting in the next 15 minutes 78.6% 80.1% 91.8% Accuracy of predicting traffic congestion risk in the next 15 minutes 72.4% 74.9% 89.6% Average task completion time 18.7 minutes 16.9 minutes 12.4 minutes AGV average waiting time 5.8 minutes 4.6 minutes 2.1 minutes AGV empty running rate 21.3% 18.7% 11.5% Number of path conflicts 326 trains / trips 241 times / train 96 times / train Path conflict resolution success rate 82.5% 87.9% 96.7% Task delays caused by sorting compartments being fully loaded 184 orders / shifts 139 orders / shift 52 orders / shift Average occupancy imbalance rate of delivery terminals 27.8% 23.5% 13.2% Order volume completed per shift 13760 orders 14620 orders 16390 orders

[0164] As can be seen from Table 1, under the same warehousing and distribution center, the same number of AGVs, and the same peak order pressure, the system of the present invention has better task demand prediction ability, channel congestion prediction ability, and multi-AGV collaborative scheduling effect compared with the fixed priority scheduling scheme and the standard A-star path planning scheme.

[0165] The system of this invention achieved a 91.8% accuracy rate in predicting sorting task demand within the next 15 minutes and an 89.6% accuracy rate in predicting channel congestion risk within the next 15 minutes. This demonstrates that the improved FreTS logistics task demand prediction network can more accurately capture the changing relationships between sorting and delivery tasks, AGV resources, and channel operating status. Simultaneously, the system reduced the average task completion time to 12.4 minutes, the average AGV waiting time to 2.1 minutes, and the AGV empty-run rate to 11.5%. This indicates that generating an AGV execution priority queue based on the sorting and delivery demand prediction results can reduce AGV ineffective waiting and empty runs.

[0166] Furthermore, the system of this invention reduces the number of path conflicts to 96 times per shift, increases the success rate of path conflict resolution to 96.7%, and reduces the task delay caused by full capacity of sorting slots to 52 orders per shift. This indicates that the rolling window priority inheritance backtracking algorithm can effectively handle conflicts such as occupancy of the same channel segment, occupancy of the same intersection node, oncoming traffic conflicts, and rear-end collision conflicts. Therefore, this invention can improve the accuracy of task prediction, the stability of multi-AGV collaborative scheduling, the efficiency of path conflict resolution, and the order processing capacity per shift in unmanned sorting and delivery scenarios.

[0167] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. An intelligent logistics unmanned sorting and delivery system based on multi-AGV autonomous collaboration, characterized in that, include: Logistics task data acquisition module: used to acquire basic logistics operation data in the target logistics scenario; Logistics Status Diagram Construction Module: This module is used to construct a global status diagram of logistics operations based on basic logistics operation data, and to record status snapshots of the global status diagram of logistics operations according to statistical periods, generating a logistics operation status snapshot sequence. AI task demand prediction module: used to extract indicators from the snapshot sequence of logistics operation status, and perform frequency domain time series prediction through the improved FreTS logistics task demand prediction network to obtain the sorting and delivery demand prediction results; the improved FreTS logistics task demand prediction network includes a logistics status embedding mapping layer, a phase difference constrained frequency channel learner, a peak and valley suppression frequency time learner, and a prediction projection layer. Multi-AGV autonomous collaborative scheduling module: Based on the global state diagram of logistics operations and the results of sorting and delivery demand prediction, it uses a rolling time window priority inheritance backtracking algorithm to perform task allocation, path planning and path conflict resolution for multiple AGVs, and generates a multi-AGV collaborative scheduling scheme. Intelligent sorting and delivery execution module: used to control each AGV to perform sorting and delivery according to the multi-AGV collaborative scheduling scheme, and update the execution status; Sorting and delivery status monitoring module: used to monitor the sorting and delivery execution status of multiple AGVs and generate sorting and delivery operation status results; Unmanned operation optimization decision module: used to update the multi-AGV collaborative scheduling scheme based on the sorting and delivery operation status results; Visualization and data storage module: used to visualize the system's operating results and write them into the logistics management database.

2. The intelligent logistics unmanned sorting and delivery system based on multi-AGV autonomous collaboration as described in claim 1, characterized in that, The basic data for logistics operations includes sorting and delivery task data, cargo status data, AGV operation status data, warehouse area status data, and delivery terminal status data. The sorting and delivery task data includes order number, goods identification, task type, task generation time, task priority, pickup location, sorting target compartment, and delivery terminal node; The cargo status data includes cargo identification, current cargo location, cargo weight, cargo category, and cargo sorting status; The AGV operating status data includes AGV number, AGV current location, AGV power status, AGV load status, AGV task execution status, and AGV availability status; The warehouse area status data includes the occupancy status of the sorting area, the occupancy status of the buffer area, the occupancy status of the aisle sections, the occupancy status of the intersection nodes, the connection relationship of the aisle sections, and the passage direction of the aisle sections; The delivery terminal status data includes the availability status of the delivery handover area, the availability status of the loading and unloading station, the availability status of the delivery station, and the full load status of the sorting compartment.

3. The intelligent logistics unmanned sorting and delivery system based on multi-AGV autonomous collaboration according to claim 1, characterized in that, The logistics status diagram construction module specifically includes: The basic data of logistics operations are parsed and mapped into nodes according to data type to construct sorting and delivery task nodes, cargo status nodes, AGV status nodes, warehouse area status nodes and delivery terminal status nodes. Based on the order number and the goods identifier, establish a task-goods binding edge between the sorting and delivery task node and the goods status node; Based on the pickup location, current location of goods, sorting target slot and delivery terminal node, establish operation path association edges between sorting and delivery task nodes, goods status nodes, warehouse area status nodes and delivery terminal status nodes; Based on the current position of the AGV, the load status of the AGV, the task execution status of the AGV, and the availability status of the AGV, establish AGV task candidate edges between the AGV status nodes and the sorting and delivery task nodes. Based on the status of channel occupancy, intersection occupancy, sorting grid full load, delivery handover area availability, loading and unloading station availability, and delivery station availability, establish an operation acceptance status edge between the warehouse area status node and the delivery terminal status node. The sorting and delivery task nodes, cargo status nodes, AGV status nodes, warehouse area status nodes, delivery terminal status nodes, task cargo binding edges, operation path association edges, AGV task candidate edges, and operation acceptance status edges are organized into a graph structure to construct a global state graph of logistics operations. Record the status of graph nodes and edges in the global status graph of logistics operations according to the statistical period, generate a logistics operation status snapshot, and arrange the logistics operation status snapshots in the order of the statistical period to generate a logistics operation status snapshot sequence.

4. The intelligent logistics unmanned sorting and delivery system based on multi-AGV autonomous collaboration according to claim 1, characterized in that, The AI ​​task demand prediction module specifically includes: Based on the snapshot sequence of logistics operation status, the status of graph nodes and edge status is read according to the statistical period, and logistics status indicators are extracted from the status of graph nodes and edge status. The logistics status indicators are arranged according to the statistical period and the logistics status indicators to generate a logistics operation status tensor. The logistics operation state tensor is embedded and mapped by the logistics state embedding mapping layer to generate the logistics state embedding tensor. In the phase difference constrained frequency channel learner, the logistics state embedded tensor is subjected to discrete Fourier transform along the logistics state index dimension to obtain the channel complex spectrum tensor. Based on the channel complex spectrum tensor, the channel spectrum amplitude tensor and the channel spectrum phase tensor are extracted, and the channel phase difference tensor is generated according to the difference between the channel spectrum phase tensors corresponding to any two logistics status indicators. The channel amplitude scaling factor tensor is obtained by performing mean normalization, amplitude linear mapping and sigmoid activation on the channel spectrum amplitude tensor. The channel phase difference tensor is subjected to sine and cosine encoding, phase linear mapping and Tanh activation to obtain the channel phase offset tensor. The amplitude of the channel spectrum amplitude tensor is adjusted using the channel amplitude scaling factor tensor to obtain the enhanced channel amplitude tensor; The channel spectral phase tensor is phase-adjusted using the channel phase offset tensor to obtain the enhanced channel phase tensor; The enhanced channel amplitude tensor and enhanced channel phase tensor are converted into complex frequency domain representations containing real and imaginary features to obtain the channel complex frequency enhanced feature tensor; Perform an inverse frequency domain transform on the channel complex frequency enhancement feature tensor to obtain the channel enhancement feature tensor; In the peak-valley suppression frequency time learner, the channel enhancement feature tensor is subjected to discrete Fourier transform along the statistical period dimension to obtain the time spectrum tensor; Based on the amplitude variation intensity of each time frequency component in the time spectrum tensor, the dominant frequency component tensor and the peak-valley perturbation component tensor are determined. The dominant frequency component tensor is subjected to frequency domain enhancement mapping, and the peak-valley perturbation component tensor is subjected to frequency domain suppression mapping to obtain the recalibrated time-frequency domain feature tensor. Perform an inverse frequency domain transform on the recalibrated time-frequency domain feature tensor to obtain the time-enhanced feature tensor; In the prediction projection layer, the time-enhanced feature tensor is mapped to a multi-head linear prediction to obtain the sorting and delivery demand prediction results. The sorting and delivery demand forecast results include the forecast values ​​of sorting task demand, delivery task demand, AGV resource demand, channel congestion risk, and delivery terminal occupancy for future scheduling cycles.

5. The intelligent logistics unmanned sorting and delivery system based on multi-AGV autonomous collaboration according to claim 1, characterized in that, The multi-AGV autonomous collaborative scheduling module specifically includes: Construct an AGV task candidate set, which includes multiple AGV task candidate combinations. Each AGV task candidate combination includes AGV number, target task, pickup location, sorting target slot, delivery terminal node, AGV current location, AGV power status, AGV load status, AGV task execution status, and AGV availability status. Based on the sorting and delivery demand forecast results, the task scheduling score for each AGV task candidate combination is calculated; Based on the task scheduling scores from high to low, the candidate combinations of each AGV task are sorted to obtain the AGV execution priority queue. The current scheduling time is used as the starting point of the rolling window, and the current rolling window is generated according to the window length; Based on the global state diagram of logistics operations, within the current rolling time window, candidate driving paths for each AGV task candidate combination are generated through A-star search. Time window occupancy analysis is performed on each candidate driving route to determine the time window and location of the conflict corresponding to the path conflict, and to obtain the set of path conflicts; The time window occupancy analysis includes: identifying occupancy conflicts on the same road segment, occupancy conflicts at the same intersection, oncoming traffic conflicts, and rear-end collision conflicts; Based on the AGV execution priority queue, priority inheritance processing is performed on the path conflict set, specifically as follows: AGVs that participate in the same path conflict and are ranked higher in the AGV execution priority queue are identified as high-priority AGVs, while AGVs ranked lower are identified as low-priority AGVs. Determine whether a low-priority AGV occupies a conflict position in the candidate driving path corresponding to a high-priority AGV within the conflict occurrence time window; If a low-priority AGV occupies a conflicting position, the passage priority of the high-priority AGV is transferred to the low-priority AGV, and the avoidance action of the low-priority AGV is updated. Based on the updated avoidance action, the candidate driving path of the low-priority AGV is corrected to obtain the priority inheritance processing result. If path conflicts still exist in the priority inheritance processing results, the low-priority AGVs will be backtracked and the candidate actions will be verified in the order of waiting, yielding, partial detour and candidate driving path rearrangement to obtain the backtracking processing results that satisfy the channel segment occupancy status, intersection node occupancy status and channel segment travel direction. Based on the priority inheritance processing results and the backtracking processing results, the target task, picking location, sorting target slot, delivery terminal node, driving path, arrival time and avoidance actions of each AGV are determined, and a multi-AGV collaborative scheduling scheme is generated by combining them.

6. The intelligent logistics unmanned sorting and delivery system based on multi-AGV autonomous collaboration according to claim 1, characterized in that, The intelligent sorting and delivery execution module specifically includes: Based on the multi-AGV collaborative scheduling scheme, the AGV number, target task, pickup location, sorting target slot, delivery terminal node, driving route, arrival time and avoidance actions of each AGV are read. Control each AGV to reach the pickup location according to the driving route and perform cargo pickup; The AGV that has completed the pickup of goods performs sorting and transfer according to the travel path, and transfers the goods to the target sorting slot. Control the AGVs that have completed sorting and transfer to perform delivery handover, and return to the standby position after the delivery handover is completed; If an AGV malfunction, path blockage, cargo recognition failure, full sorting slot, or abnormal occupancy of a delivery terminal node is detected, the AGV that is malfunctioning will be identified as an abnormal AGV, the current execution action of the abnormal AGV will be suspended, and the abnormality type and abnormality location will be recorded. Based on the anomaly type and location, update the AGV task execution status of the abnormal AGV and the corresponding cargo sorting status, and feed back the anomaly type, anomaly location, AGV number of the abnormal AGV, and corresponding target task to the sorting and delivery status monitoring module.

7. The intelligent logistics unmanned sorting and delivery system based on multi-AGV autonomous collaboration according to claim 1, characterized in that, The sorting and delivery status monitoring module specifically includes: Read the AGV task execution status of each AGV, the cargo sorting status of each cargo, the channel occupancy status of each channel segment, and the intersection node occupancy status of each intersection node. Based on the AGV task execution status, travel path, and arrival time, determine whether the AGV has deviated from the travel path, timed out of arrival, or stopped executing. Based on the cargo sorting status, determine whether the cargo has completed the cargo pickup, sorting and transfer, and delivery handover process; Based on the occupancy status of the passageway and intersection, determine whether there is congestion on the passageway or abnormal occupancy at the intersection; The sorting and delivery efficiency is statistically analyzed, and the AGV task execution status, cargo sorting status, aisle segment occupancy status, intersection node occupancy status, and sorting and delivery efficiency are combined into a sorting and delivery operation status result.

8. The intelligent logistics unmanned sorting and delivery system based on multi-AGV autonomous collaboration according to claim 1, characterized in that, The unmanned operation optimization decision module specifically includes: Based on the sorting and delivery operation status results, determine whether the scheduling update conditions are met; If the scheduling update conditions are met, the sorting and delivery operation status results will be fed back to the multi-AGV autonomous collaborative scheduling module to generate an updated multi-AGV collaborative scheduling scheme. The scheduling update conditions include: the AGV task execution status of the abnormal AGV is paused, the sorting status of the goods corresponding to the target task is incomplete, the channel segment occupancy status is continuously occupied, the intersection node occupancy status is abnormally occupied, the current rolling window has ended, or new sorting and delivery task data has been received.

9. The intelligent logistics unmanned sorting and delivery system based on multi-AGV autonomous collaboration according to claim 1, characterized in that, The system operation results include sorting and delivery demand prediction results, multi-AGV collaborative scheduling scheme, sorting and delivery operation status results, and updated multi-AGV collaborative scheduling scheme. The system operation results are visualized on the scheduling dashboard, and the basic data of logistics operations, the global status diagram of logistics operations, the sequence of logistics operation status snapshots, and the system operation results are written into the logistics management database in batches.

Images