Logistics data processing method and device based on multi-agent cooperation, equipment, medium and computer program product

Abstract

The application provides a logistics data processing method and device based on multi-agent cooperation, equipment, medium and computer program product, relating to the technical field of logistics. The method comprises: obtaining user order information, generating a logistics task according to the user order information; decomposing the parsed logistics task into multiple sub-tasks; receiving sub-agent information fed back by each sub-agent; making a decision based on the multiple sub-tasks, the state information of each sub-agent and the task processing type information, and determining a target sub-agent; sending the sub-tasks to each target sub-agent through a first protocol, so that each target sub-agent generates corresponding sub-task execution results based on the sub-tasks and the execution results of the previous sub-tasks in a predetermined order, until the last target sub-agent generates a target sub-task execution result. The embodiments provided by the application can more effectively coordinate and manage logistics resources, improve logistics efficiency, reduce logistics cost and improve logistics service quality.

Description

Logistics data processing methods, devices, equipment, media, and computer program products based on multi-agent collaboration. Technical Field

[0001] This invention relates to the field of logistics technology, and in particular to a logistics data processing method, apparatus, equipment, medium, and computer program product based on multi-agent collaboration. Background Technology

[0002] In the field of modern logistics management, with the rapid development of e-commerce, the number and complexity of logistics orders have increased significantly. This has made logistics order processing more difficult and placed higher demands on logistics order analysis and scheduling systems.

[0003] Existing large-scale logistics models typically consist of only a single agent, making it difficult to handle complex logistics tasks. A single agent struggles to effectively coordinate and manage logistics resources, leading to low efficiency and increased costs. Summary of the Invention

[0004] This invention provides a logistics data processing method, apparatus, equipment, medium, and computer program product based on multi-agent collaboration, which addresses the shortcomings of low logistics transportation efficiency and poor resource coordination in the prior art. It can more effectively coordinate and manage logistics resources, improve logistics efficiency, reduce logistics costs, and enhance the quality of logistics services.

[0005] In a first aspect, the present invention provides a logistics data processing method based on multi-agent collaboration, applied to a master agent, wherein the master agent is connected to any sub-agents via a first protocol, and each sub-agent is connected to each other via a second protocol; the method includes:

[0006] Obtain user order information and generate logistics tasks based on the user order information;

[0007] The logistics task is parsed, and the parsed logistics task is decomposed into multiple sub-tasks;

[0008] Multiple subtasks are broadcast to each sub-agent via the first protocol, and sub-agent information is received from each sub-agent; the sub-agent information includes at least the state information and task processing type information of each sub-agent.

[0009] Decision processing is performed based on multiple sub-tasks, the state information of each sub-agent, and the task processing type information to determine the target sub-agent to receive the corresponding sub-task; among them, multiple target sub-agents are arranged in a preset order, and the first target sub-agent sends the corresponding sub-task execution result to the last target sub-agent through the second protocol, and each target sub-agent executes the corresponding type of sub-task;

[0010] The subtask is sent to each target sub-agent via a first protocol, so that each target sub-agent generates a corresponding subtask execution result according to the preset order based on the subtask and the execution result of the preceding subtask, until the last target sub-agent in the preset order generates the target subtask execution result.

[0011] Preferably, according to the logistics data processing method based on multi-agent collaboration provided by the present invention, the step of sending the sub-task to each target sub-agent through a first protocol, so that each target sub-agent generates a corresponding sub-task execution result according to the preset order based on the sub-task and the execution result of the preceding sub-task, until the last target sub-agent in the preset order generates the target sub-task execution result, includes:

[0012] The first subtask is sent to the first target sub-agent via the first protocol, so that the first target sub-agent executes the corresponding first subtask and generates the corresponding first subtask execution result.

[0013] The second subtask is sent to the second target sub-agent via the first protocol, so that the second target sub-agent generates the execution result of the second subtask based on the execution results of the second subtask and the first subtask; wherein, the execution result of the first subtask is sent to the second target sub-agent by the first target sub-agent via the second protocol;

[0014] The third subtask is sent to the third target sub-agent via the first protocol, so that the third target sub-agent generates the target subtask execution result based on the third subtask, the execution result of the first subtask received from the first target sub-agent via the second protocol, and the execution result of the second subtask received from the second target sub-agent via the second protocol.

[0015] Preferably, in the logistics data processing method based on multi-agent collaboration provided by the present invention, the first target sub-agent is represented as a path planning agent, used to generate the optimal transportation path;

[0016] The step of sending the first subtask to the first target sub-agent via the first protocol, so that the first target sub-agent executes the corresponding first subtask and generates the corresponding first subtask execution result, includes:

[0017] The first subtask is sent to the path planning agent via a first protocol, so that the path planning agent performs optimal path calculation based on the first subtask, current road conditions, weather information, cargo information, timeliness information, and cost information to generate an optimal transportation path; wherein, the optimal transportation path is characterized as the path that minimizes the timeliness, costs, and accuracy of transporting goods from the warehouse to the target delivery location; the optimal transportation path includes at least multiple key trajectory points, which are used to determine whether there is any deviation in the actual transportation trajectory of the transport vehicle.

[0018] Preferably, in the logistics data processing method based on multi-agent collaboration provided by the present invention, the second target sub-agent is represented as a warehouse scheduling agent, used to update warehouse goods entry and exit information;

[0019] The step of sending the second subtask to the second target sub-agent via the first protocol, so that the second target sub-agent generates the execution result of the second subtask based on the execution results of the second subtask and the first subtask, includes:

[0020] The second subtask is sent to the warehouse scheduling agent, so that the warehouse scheduling agent can determine the target warehouse with sufficient goods and the shortest distance from the target delivery location based on the second subtask and the optimal transportation route of the goods, and release the corresponding goods from the target warehouse.

[0021] Preferably, in the logistics data processing method based on multi-agent collaboration provided by the present invention, the third target sub-agent is characterized as a transportation execution agent, used to generate state information of transported goods;

[0022] The step of sending the third subtask to the third target sub-agent via the first protocol, so that the third target sub-agent generates a target subtask execution result based on the third subtask, the execution result of the first subtask received from the first target sub-agent via the second protocol, and the execution result of the second subtask received from the second target sub-agent via the second protocol, includes:

[0023] The third subtask is sent to the transportation execution agent via the first protocol, so that the transportation execution agent determines the target number of target transportation vehicles based on the third subtask, the optimal transportation route, and the goods corresponding to the outbound shipment from the target warehouse. The agent records the movement trajectory and actual transportation time of each target transportation vehicle to the key trajectory point in real time, compares the movement trajectory with the optimal transportation route, and compares the actual transportation time with the predicted transportation time to the key trajectory point. If the movement trajectory does not deviate from the optimal transportation route, the difference between the actual transportation time and the predicted transportation time to the key trajectory point is within a preset range, and all target transportation vehicles of the target number transport the goods to the target delivery location, the agent inspects the goods based on preset inspection rules, generates a goods inspection result, and generates a goods transportation completion result based on the goods inspection result.

[0024] Preferably, according to the multi-agent collaborative logistics data processing method provided by the present invention, after the step of parsing the logistics task and decomposing the parsed logistics task into multiple sub-tasks, the method includes:

[0025] Upon receiving a transportation task query request, the system obtains the execution status of each target sub-agent in executing its corresponding sub-task and displays the execution status of each target sub-agent in executing its corresponding sub-task through a visual interface.

[0026] Secondly, the present invention also provides a logistics data processing device based on multi-agent collaboration, applied to a master agent, wherein the master agent is connected to any sub-agent via a first protocol, and each sub-agent is connected to each other via a second protocol; the device includes:

[0027] The acquisition module is used to acquire user order information and generate logistics tasks based on the user order information;

[0028] The decomposition module is used to parse the logistics task and decompose the parsed logistics task into multiple sub-tasks;

[0029] The receiving module is used to broadcast multiple subtasks to each sub-agent through the first protocol and receive sub-agent information fed back by each sub-agent; the sub-agent information includes at least the state information and task processing type information of each sub-agent;

[0030] The determination module is used to make decisions based on multiple subtasks, the state information of each sub-agent, and the task processing type information to determine the target sub-agent to receive the corresponding subtask. Among them, multiple target sub-agents are arranged in a preset order, and the first target sub-agent sends the corresponding subtask execution result to the last target sub-agent through the second protocol. Each target sub-agent executes the corresponding type of subtask.

[0031] The execution module is used to send the subtask to each target sub-agent through a first protocol, so that each target sub-agent generates a corresponding subtask execution result according to the preset order based on the subtask and the execution result of the preceding subtask, until the last target sub-agent in the preset order generates the target subtask execution result.

[0032] Thirdly, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the logistics data processing method based on multi-agent collaboration as described above.

[0033] Fourthly, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the logistics data processing method based on multi-agent collaboration as described above.

[0034] Fifthly, the present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the logistics data processing method based on multi-agent collaboration as described above.

[0035] This invention provides a logistics data processing method, apparatus, device, medium, and computer program product based on multi-agent collaboration. Applied to a master agent, the master agent connects to any sub-agents via a first protocol, and each sub-agent connects to each other via a second protocol. The method involves acquiring user order information and generating logistics tasks based on that information; parsing the logistics tasks and decomposing them into multiple sub-tasks; broadcasting these sub-tasks to each sub-agent via the first protocol; and receiving feedback from each sub-agent. The sub-agent information includes at least the state information and task processing type information of each sub-agent. The process involves decision-making based on subtasks, the state information of each sub-agent, and task processing type information to determine the target sub-agent receiving the corresponding subtask. Multiple target sub-agents are arranged in a preset order. The first target sub-agent sends its corresponding subtask execution result to the next target sub-agent via a second protocol. Each target sub-agent executes a subtask of the corresponding type. The subtask is then sent to each target sub-agent via a first protocol, enabling each target sub-agent to generate its corresponding subtask execution result based on the subtask and the execution result of the first subtask, in the preset order, until the last target sub-agent in the preset order generates its target subtask execution result. This approach addresses the shortcomings of low logistics transportation efficiency and poor resource coordination in existing technologies, enabling more effective coordination and management of logistics resources, improving logistics efficiency, reducing logistics costs, and enhancing logistics service quality. Attached Figure Description

[0036] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0037] Figure 1 is a flowchart illustrating the logistics data processing method based on multi-agent collaboration provided by the present invention.

[0038] Figure 2 is a schematic diagram of the structure of the logistics data processing device based on multi-agent collaboration provided by the present invention.

[0039] Figure 3 is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0040] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0041] The following description, in conjunction with Figures 1-3, describes a logistics data processing method, apparatus, equipment, medium, and computer program product based on multi-agent collaboration, which addresses the shortcomings of low logistics transportation efficiency and poor resource coordination in the prior art. It can more effectively coordinate and manage logistics resources, improve logistics efficiency, reduce logistics costs, and enhance the quality of logistics services.

[0042] Figure 1 is a flowchart illustrating a logistics data processing method based on multi-agent collaboration provided by the present invention. As shown in Figure 1, the method may include, but is not limited to, steps S100 to S500:

[0043] S100, Obtain user order information, and generate a logistics task based on the user order information;

[0044] S200, parse the logistics task and decompose the parsed logistics task into multiple sub-tasks;

[0045] S300, broadcast multiple subtasks to each sub-agent through the first protocol, and receive sub-agent information fed back by each sub-agent; the sub-agent information includes at least the state information and task processing type information of each sub-agent;

[0046] S400 performs decision processing based on multiple sub-tasks, the state information of each sub-agent, and the task processing type information to determine the target sub-agent to receive the corresponding sub-task; wherein, multiple target sub-agents are arranged in a preset order, and the first target sub-agent sends the corresponding sub-task execution result to the last target sub-agent through the second protocol, and each target sub-agent executes the corresponding type of sub-task.

[0047] S500, the subtask is sent to each target sub-agent through the first protocol, so that each target sub-agent generates the corresponding subtask execution result according to the preset order based on the subtask and the execution result of the preceding subtask, until the last target sub-agent in the preset order generates the target subtask execution result.

[0048] In step S100 of some embodiments, user order information is obtained, and a logistics task is generated based on the user order information.

[0049] It is understandable that user order information can be obtained through e-commerce platforms, apps, mini-programs, etc., including: product information (SKU, quantity, weight, volume, etc.); recipient information (name, address, contact information); payment status (paid / unpaid / refund in progress); and special requirements (such as scheduled delivery, invoice requirements, etc.).

[0050] Furthermore, data verification of user order information should be performed, including at least: Integrity verification: checking the validity of the shipping address, product inventory, and payment status. Risk verification: identifying abnormal orders (such as frequent orders or high-risk addresses) through anti-fraud models. Inventory verification: querying product inventory in real time; triggering a restocking process or notifying the user of a delayed shipment if an item is out of stock.

[0051] Furthermore, based on user address, warehouse location, inventory distribution, and delivery time requirements, the system selects the optimal shipping warehouse. It also automatically matches the most suitable logistics provider based on order attributes (weight, volume, delivery time requirements) and the logistics provider's service capabilities (delivery range, price, delivery time). If an order contains multiple items or requires split shipments, multiple logistics tasks are generated and associated with the same order number.

[0052] Each logistics task generates a unique tracking number and associates it with the order number, product information, sender and recipient information, etc. It also generates electronic waybills via the logistics provider's API or local printing system, including recipient information, product details, and the logistics provider's logo to create the logistics task.

[0053] The embodiments provided by this invention reduce the time required to generate a logistics task from an order to a minimum, from hours to minutes.

[0054] In step S200 of some embodiments, the logistics task is parsed and the parsed logistics task is decomposed into multiple sub-tasks.

[0055] It is understood that this invention breaks down complex logistics tasks (order processing, transportation execution, route planning, warehouse scheduling, cost accounting, etc.) into multiple sub-tasks, which are handled independently by different intelligent agents (such as intelligent scheduling, cost calculation, and intelligent question answering).

[0056] Based on the entire logistics process (order-warehousing-transportation-delivery-after-sales), key nodes are identified and broken down into sub-task modules that can be performed in parallel or sequentially. Example: A "cross-province delivery order" is broken down into sub-tasks such as "inventory allocation → capacity scheduling → route optimization → cost accounting → anomaly monitoring".

[0057] The intelligent agent includes a master intelligent agent and sub-intelligent agents. The master intelligent agent is connected to any sub-intelligent agent through a first protocol, and each sub-intelligent agent is connected to each other through a second protocol.

[0058] The first protocol refers to the MCP protocol. The main intelligent agent uses the MCP protocol to realize global task decomposition and resource scheduling. For example, the "national warehousing and distribution task" is decomposed into regional sub-tasks and assigned to regional scheduling intelligent agents; the status of each sub-task is synchronized (such as the delay in delivery of goods from a warehouse triggering alternative solutions).

[0059] The second protocol refers to the A2A protocol, which supports low-latency collaboration between sub-agents. For example, the path planning agent can push path change information to the cost calculation agent in real time; the intelligent question answering module can call the anomaly analysis results of the risk warning agent.

[0060] The first protocol (Model Communication Protocol, MCP) is a structured instruction communication protocol for large models and agents, which parses natural language instructions and converts them into an executable format.

[0061] The second protocol (Agent-to-Agent Protocol, A2A) is used for state synchronization, data sharing, and task relay between intelligent agents, enabling distributed collaboration, task transfer, and state reporting.

[0062] In this embodiment of the invention, the main intelligent agent can be a task-assigning intelligent agent used to assign cargo information, vehicle information, order information, etc.

[0063] The sub-agents can be route planning agents, warehouse scheduling agents, and transportation execution agents. Specifically, the first target sub-agent is represented by a route planning agent, used to generate or modify transportation routes. The second target sub-agent is represented by a warehouse scheduling agent, used to manage goods inbound and outbound information. The third target sub-agent is represented by a transportation execution agent, used to control the movement of transportation vehicles to transport goods.

[0064] In some embodiments of the present invention, a layered communication architecture is adopted: the first layer (global control): the main intelligent agent realizes global task decomposition and resource scheduling through the MCP protocol. For example, the "national warehousing and distribution task" is decomposed into regional sub-tasks and assigned to regional scheduling intelligent agents; the status of each sub-task is synchronized (such as the alternative scheme is triggered when a warehouse's shipment is delayed).

[0065] The second layer (point-to-point collaboration): The A2A protocol supports low-latency collaboration between agents, such as: scheduling agents pushing path change information to cost calculation agents in real time; and intelligent question answering modules calling the anomaly analysis results of risk warning agents.

[0066] Protocol bridging module: The main agent calls the capabilities of each agent through MCP instructions, which are internally converted into local task execution (such as microservice calls within a Docker container).

[0067] In step S300 of some embodiments, multiple subtasks are broadcast to each sub-agent via the first protocol, and sub-agent information is received from each sub-agent; the sub-agent information includes at least the state information and task processing type information of each sub-agent.

[0068] Understandably, the main intelligent agent breaks down complex tasks into sub-task modules based on the entire logistics process (order → warehousing → transportation → delivery → after-sales service).

[0069] Task encapsulation: Define a unique identifier for each subtask (e.g., TaskID: Dispatch_001) and associate it with the target agent type (e.g., scheduling, cost calculation).

[0070] The main intelligent agent constructs structured instructions through the MCP protocol, which include: subtask descriptions (such as path optimization requests); input parameters (such as order coordinates and vehicle load limits); and priority fields (such as setting emergency orders to high priority).

[0071] Commands are sent to the corresponding sub-agents via message queues (such as Kafka) or RPC interfaces, so that multiple sub-agents can report their current load status and the processing type information of the sub-tasks they can handle.

[0072] When the main agent receives the status information and task processing type information from each sub-agent, it makes a decision based on the status information and task processing type information of each sub-task and each sub-agent to determine the target sub-agent to receive the corresponding sub-task.

[0073] The main agent aggregates feedback from all subtasks and makes global decisions (such as adjusting scheduling strategies) based on environmental conditions (e.g., traffic congestion index).

[0074] In some embodiments, if a sub-agent fails to respond within a timeout period, a backup plan (such as switching to a backup model or a manual intervention interface) is triggered.

[0075] In step S400 of some embodiments, decision processing is performed based on multiple sub-tasks, the state information of each sub-agent, and the task processing type information to determine the target sub-agent to receive the corresponding sub-task; wherein, the multiple target sub-agents are arranged in a preset order, and the first target sub-agent sends the corresponding sub-task execution result to the last target sub-agent through a second protocol, and each target sub-agent executes the corresponding type of sub-task.

[0076] Understandably, the main agent determines the list of target sub-agents based on the sub-task type, sub-agent state, and preset priority rules.

[0077] Furthermore, multiple target sub-agents are arranged in order of priority. After the preceding agent completes its task, it passes the result to the following agent through a dedicated protocol (such as A2A), forming a pipeline-style collaboration.

[0078] Furthermore, the final result is returned to the main intelligent agent, and global decision adjustments are made in conjunction with environmental conditions (such as traffic congestion index) to form a dynamic optimization closed loop.

[0079] It should be noted that the embodiments of the present invention adopt a dynamic task allocation mechanism for decision-making. The specific decision-making logic is as follows: the main intelligent agent selects the target sub-intelligent agent based on the following dimensions: task type matching degree (e.g., path optimization tasks are only assigned to the path planning intelligent agent); sub-intelligent agent status (e.g., the inventory monitoring intelligent agent only receives replenishment instructions when the inventory is sufficient); and preset priority rules (e.g., emergency orders trigger a high-priority scheduling strategy).

[0080] A layered communication architecture (MCP+A2A) is adopted. The main control layer broadcasts task requirements through the MCP protocol, and the collaboration layer transmits real-time status through the A2A protocol.

[0081] It should be further noted that an improved A2A protocol (such as a hybrid HTTP / WebSocket+gRPC mode) is adopted, which supports asynchronous result transmission and synchronous status verification.

[0082] The transmitted content includes structured results (JSON / Protobuf) and metadata (such as task ID and execution time) to ensure that subsequent agents can verify the integrity of the data.

[0083] In some embodiments, each sub-agent maintains an internal state machine (e.g., "standby → processing → completed") and proactively reports its status via heartbeat packets or event notifications.

[0084] If the preceding agent times out and does not return a result, the subsequent agent triggers retry logic or switches to a backup link.

[0085] In some embodiments, a logistics task (such as "inter-provincial fresh food delivery") is received and broken down into sub-tasks (warehousing and picking, cold chain transportation, temperature control monitoring). A target agent is matched according to the task type (e.g., the temperature control monitoring task is assigned to the cold chain agent).

[0086] Check the status of sub-agents (e.g., "Is the cold chain truck idle?") and eliminate unavailable nodes;

[0087] Structured instructions (including task parameters and deadlines) are sent to the target sub-agent according to a preset priority order (e.g., “Transportation” > “Warehousing”).

[0088] This invention reduces task waiting time by prioritizing tasks.

[0089] Furthermore, after the scheduling agent completes path planning, it sends the results (JSON-formatted path data) to the cost calculation agent via the A2A protocol. The cost calculation agent parses the path data and verifies the validity of the coordinates and load limits. After successful verification, it transmits the cost report to subsequent agents (such as the risk warning module) via the A2A protocol.

[0090] In some embodiments, if a step fails (e.g., the path exceeds the vehicle's load capacity), a backtracking mechanism is triggered (e.g., the scheduling model is called back to optimize the path).

[0091] The main agent aggregates the results: summarizing all subtask feedback (such as "total cost $1200" and "estimated time 8 hours").

[0092] Reinforcement learning iteration: Input the results into the simulation environment (parallel universe) to train the VDN / MAAC model to optimize the cooperative strategy.

[0093] Dynamic strategy updates: Learn new scenario data online (such as delivery delays during heavy rain) and adjust priority rules (such as increasing the weight of weather warnings).

[0094] This embodiment achieves the evolution of the logistics big data model from "single-agent serial processing" to "multi-agent parallel optimization" through dynamic task allocation, chain collaboration protocol and reinforcement learning iteration.

[0095] In step S500 of some embodiments, the subtask is sent to each target sub-agent through a first protocol, so that each target sub-agent generates a corresponding subtask execution result according to the preset order based on the subtask and the execution result of the preceding subtask, until the last target sub-agent in the preset order generates the target subtask execution result.

[0096] It is understandable that the specific execution steps may be to send the first subtask to the first target sub-agent through the first protocol, so that the first target sub-agent executes the corresponding first subtask and generates the corresponding first subtask execution result;

[0097] The second subtask is sent to the second target sub-agent via the first protocol, so that the second target sub-agent generates the execution result of the second subtask based on the execution results of the second subtask and the first subtask; wherein, the execution result of the first subtask is sent to the second target sub-agent by the first target sub-agent via the second protocol;

[0098] The third subtask is sent to the third target sub-agent via the first protocol, so that the third target sub-agent generates the target subtask execution result based on the third subtask, the execution result of the first subtask received from the first target sub-agent via the second protocol, and the execution result of the second subtask received from the second target sub-agent via the second protocol.

[0099] It should be noted that the execution steps of the first subtask execution phase are as follows: Main agent task decomposition: The complex task (such as "cross-province fresh food delivery") is decomposed into the first subtask (such as "path optimization"), and broadcast to the first target sub-agent (such as the path planning agent) through the MCP protocol.

[0100] Asynchronous processing by sub-agents: After receiving the task, the path planning agent calls a path planning algorithm (such as Dijkstra + reinforcement learning model) to generate the optimal path based on real-time traffic data (such as congestion index).

[0101] Result return and status update: The path data (JSON format) is sent to the main agent via the A2A protocol, and its own status is updated to "completed". This embodiment only schedules agents to participate in path calculation, avoiding the waste of irrelevant resources.

[0102] The execution steps of the second subtask execution phase are as follows: Main agent dependency distribution: The second subtask (such as "goods outbound warehouse") is sent to the second target sub-agent (such as warehouse scheduling agent) through the MCP protocol, along with the result of the first subtask (path data).

[0103] Dependency Verification and Calculation: The warehouse scheduling agent receives path data through the A2A protocol, verifies the legality of the coordinates and the load limit, calls the warehouse scheduling model to determine the target warehouse with sufficient goods and the shortest distance from the target delivery location, and then dispatches the corresponding goods from the target warehouse.

[0104] Result transmission and status synchronization: The target warehouse with sufficient goods and the shortest distance from the target delivery location is sent to the main intelligent agent via the A2A protocol, and subsequent intelligent agents (such as the transportation execution module) are notified to prepare to receive data.

[0105] The execution steps of the third subtask execution phase are as follows: The main agent aggregates and schedules the third subtask (e.g., "transportation execution risk assessment") via the MCP protocol to the third target sub-agent (e.g., the transportation execution agent), simultaneously triggering it to retrieve path data and the identified target warehouse from the preceding agent via the A2A protocol. The transportation execution agent integrates multi-dimensional data such as path, cost, and weather (e.g., rainstorm warning), and calls upon the transportation execution risk model (e.g., Bayesian network + simulation training) to generate a comprehensive assessment report.

[0106] Final result feedback: The report is sent to the main intelligent agent via the A2A protocol, triggering a global decision (such as adjusting the delivery route or increasing insurance).

[0107] This embodiment can integrate multi-dimensional data to upgrade decision-making (such as avoiding high-risk road sections).

[0108] This embodiment reduces task response time by 30% to 50% (compared to monolithic architecture), improves global resource utilization by 20% to 40%, and supports elastic scheduling in extreme scenarios (such as lockdowns and severe weather).

[0109] In some embodiments of the present invention, the first target sub-agent is characterized as a path planning agent, used to generate the optimal transportation path;

[0110] The step of sending the first subtask to the first target sub-agent via the first protocol, so that the first target sub-agent executes the corresponding first subtask and generates the corresponding first subtask execution result, includes:

[0111] The first subtask is sent to the path planning agent via a first protocol, so that the path planning agent performs optimal path calculation based on the first subtask, current road conditions, weather information, cargo information, timeliness information, and cost information to generate an optimal transportation path; wherein, the optimal transportation path is characterized as the path that minimizes the timeliness, costs, and accuracy of transporting goods from the warehouse to the target delivery location; the optimal transportation path includes at least multiple key trajectory points, which are used to determine whether there is any deviation in the actual transportation trajectory of the transport vehicle.

[0112] Understandably, the main intelligent agent resolves the task by receiving a global task (such as "delivering fresh produce from warehouse A to city B") and breaking it down into path planning sub-tasks.

[0113] MCP Protocol Encapsulation Instructions: Task instructions are sent to the path planning agent via the MCP protocol, along with the following data: Basic Information: Warehouse coordinates, target delivery point coordinates, goods type (e.g., cold chain goods), and delivery time requirements (e.g., 2-hour delivery). Dynamic Data: Real-time traffic conditions (e.g., congestion index), weather (e.g., rainstorm warnings), and cost constraints (e.g., minimum freight cost).

[0114] Path planning agent activation: After receiving the task, the internal algorithm engine is started (such as reinforcement learning + Dijkstra algorithm), and road network data and historical transportation features are loaded.

[0115] Technical effects: Multi-source data fusion: Integrates static road network and dynamic environmental data to improve path feasibility; Clearly defines task boundaries through the MCP protocol to avoid invalid calculations.

[0116] Furthermore, the steps for optimal path calculation and key trajectory point generation are as follows: First, perform multi-objective optimization modeling:

[0117] Minimize time efficiency: Dynamically adjust routes based on real-time traffic conditions (such as traffic accidents) to avoid congested sections;

[0118] Cost minimization: comprehensively calculate the distance traveled, fuel consumption, toll fees (such as ETC charges), and time costs;

[0119] Maximize transportation accuracy: Prioritize routes with stable GPS signals and clear road markings to reduce navigation errors.

[0120] Secondly, the algorithm iteration and path generation steps are executed: a graph theory algorithm (such as...) is called. The algorithm generates candidate paths; it simulates vehicle driving using a reinforcement learning model (such as MAAC) to optimize path smoothness (e.g., reducing sharp turns); and it generates the optimal path containing key trajectory points (e.g., "warehouse → highway entrance → service area → urban intersection → destination").

[0121] Path verification and output: Verify whether the path meets timeliness and cost constraints;

[0122] The path data (including the coordinates and sequence of key trajectory points) is sent to the main intelligent agent via the A2A protocol.

[0123] The technical effects achieved by this embodiment are: global optimal solution: multi-objective optimization that balances timeliness, cost and accuracy, which is better than single-index path planning; key trajectory points provide a benchmark for subsequent transportation monitoring and facilitate anomaly detection.

[0124] Furthermore, the execution steps for path execution and deviation determination are as follows:

[0125] The transport execution layer receives the path: the driver or autonomous driving system obtains the optimal path through the A2A protocol and drives in the order of key trajectory points.

[0126] Real-time trajectory reporting and comparison: The transport vehicle reports its real-time location via GPS module; the system matches the actual trajectory with key trajectory points of the planned route (such as determining whether a "service area" node has been missed); if the deviation exceeds the threshold (such as deviating from the route by more than 500 meters), an early warning is triggered (such as replanning or notifying the driver).

[0127] After transportation is completed, the actual trajectory data is fed back to the path planning agent for model iteration (such as updating the weights of congested road segments).

[0128] The technical effects achieved by this embodiment are as follows: By setting a key trajectory point mechanism, the risk of human detours is reduced, thus improving accuracy. Feedback training with actual data improves path planning accuracy (e.g., dynamically adjusting road network weights).

[0129] In some embodiments of the present invention, the second target sub-agent is characterized as a warehouse scheduling agent, used to update warehouse goods entry and exit information;

[0130] The step of sending the second subtask to the second target sub-agent via the first protocol, so that the second target sub-agent generates the execution result of the second subtask based on the execution results of the second subtask and the first subtask, includes:

[0131] The second subtask is sent to the warehouse scheduling agent, so that the warehouse scheduling agent can determine the target warehouse with sufficient goods and the shortest distance from the target delivery location based on the second subtask and the optimal transportation route of the goods, and release the corresponding goods from the target warehouse.

[0132] It is understandable that the main agent sends the warehouse scheduling task (second subtask) to the warehouse scheduling agent through the first protocol (MCP protocol), along with the execution result of the first subtask (i.e. the optimal transportation path generated by the path planning agent).

[0133] Multi-condition warehouse selection: The warehouse scheduling intelligent agent comprehensively calculates and selects the target warehouse (with sufficient goods and closest to the delivery location) based on multi-dimensional data such as goods demand, transportation route, and warehouse inventory, and triggers the corresponding goods outbound operation.

[0134] Dynamic updates and feedback: After the outbound shipment is completed, the warehouse scheduling agent feeds back the execution results (such as outbound quantity and remaining inventory) to the main agent through the second protocol (A2A protocol), and updates the global warehouse status data at the same time, forming a closed-loop optimization.

[0135] Furthermore, in some embodiments, the main intelligent agent receives a global task (such as "deliver 100 boxes of fresh produce to city B"), generates a second sub-task (warehouse scheduling), and sends it to the warehouse scheduling intelligent agent via the MCP protocol.

[0136] Furthermore, for the optimal transportation route from the first subtask (e.g., "Warehouse A → Expressway → Urban Distribution Station"), the real-time inventory of each warehouse (e.g., 150 boxes remaining in Warehouse A, 80 boxes remaining in Warehouse B), outbound costs (e.g., loading and unloading fees, transportation fees), as well as the type of goods (e.g., fresh produce requires refrigerated storage) and timeliness requirements (e.g., outbound within 2 hours) are integrated to avoid decision-making bias caused by a single factor.

[0137] Furthermore, the specific steps for target warehouse selection and outbound execution include:

[0138] Candidate warehouse screening steps: Filter eligible warehouses based on cargo type (e.g., select only cold chain warehouses); then calculate the distance from each warehouse to the target delivery location based on the transportation route (e.g., warehouse A is 15 kilometers away from the urban distribution station, and warehouse B is 25 kilometers away).

[0139] The multi-dimensional assessment and ranking steps are as follows: Inventory adequacy: Exclude warehouses with insufficient inventory (e.g., warehouse B has only 80 boxes left, while the demand is 100 boxes).

[0140] Transportation costs: Prioritize warehouses with shorter distances and lower outbound costs (e.g., warehouse A is closer and has lower loading and unloading fees).

[0141] Time constraints: Ensure that the outbound time meets the delivery requirements (e.g., fresh produce must be outbound within 30 minutes).

[0142] The steps for determining the target warehouse and issuing outbound instructions are as follows: Select the warehouse with the highest overall score (e.g., warehouse A) as the target warehouse. Send an outbound instruction (e.g., "Outbound 100 boxes of fresh produce from warehouse A, high priority") to the warehouse management system via the A2A protocol.

[0143] The technical effects achieved by the embodiments provided by the present invention are as follows: by balancing inventory, distance and cost, the total logistics cost is reduced, and by dynamically screening, the outbound time is shortened, avoiding delays.

[0144] Furthermore, the embodiments provided by this invention are capable of state updates and feedback loops, specifically implementing the following steps:

[0145] Outbound execution monitoring: The warehouse scheduling intelligent agent receives the outbound status from the warehouse management system in real time (such as "100 boxes have been shipped, 50 boxes remain in stock").

[0146] Global status update: The outbound results (such as outbound quantity and remaining inventory) are fed back to the main intelligent agent via the A2A protocol. Warehouse inventory data is updated to the global database for subsequent tasks to access.

[0147] Furthermore, this invention also provides an anomaly handling and compensation mechanism: if the outbound shipment from the target warehouse fails (e.g., due to equipment failure), the backup warehouse selection logic is triggered (e.g., switching to warehouse B). Anomaly information is recorded in the log system for model training and process optimization.

[0148] The embodiments provided by this invention can at least implement an exception handling mechanism to reduce the task failure rate, update the global state in real time, and support accurate decision-making for subsequent tasks.

[0149] In some embodiments of the present invention, the third target sub-agent is characterized as a transportation execution agent, used to generate state information of the transported goods;

[0150] The step of sending the third subtask to the third target sub-agent via the first protocol, so that the third target sub-agent generates a target subtask execution result based on the third subtask, the execution result of the first subtask received from the first target sub-agent via the second protocol, and the execution result of the second subtask received from the second target sub-agent via the second protocol, includes:

[0151] The third subtask is sent to the transportation execution agent via the first protocol, so that the transportation execution agent determines the target number of target transportation vehicles based on the third subtask, the optimal transportation route, and the goods corresponding to the outbound shipment from the target warehouse. The agent records the movement trajectory and actual transportation time of each target transportation vehicle to the key trajectory point in real time, compares the movement trajectory with the optimal transportation route, and compares the actual transportation time with the predicted transportation time to the key trajectory point. If the movement trajectory does not deviate from the optimal transportation route, the difference between the actual transportation time and the predicted transportation time to the key trajectory point is within a preset range, and all target transportation vehicles of the target number transport the goods to the target delivery location, the agent inspects the goods based on preset inspection rules, generates a goods inspection result, and generates a goods transportation completion result based on the goods inspection result.

[0152] Understandably, the main agent issues a task: it sends a third subtask (e.g., "transport goods from warehouse A to city B") to the transportation execution agent via the MCP protocol, with the following inputs: optimal transportation route (from the route planning agent): containing the sequence of key trajectory points (e.g., "warehouse → highway entrance → service area → urban intersection → destination"); and target warehouse outbound information (from the warehouse scheduling agent): including the type of goods, quantity, and outbound time.

[0153] Transportation resource allocation: Based on the cargo volume, vehicle load, and transportation time requirements, the required number of transportation vehicles (e.g., 5 cold chain trucks) is determined through a dynamic scheduling algorithm (e.g., a greedy algorithm); transportation instructions are issued to the vehicle terminals (e.g., "Vehicle V1 travels along path R1, carrying 20 boxes of cargo").

[0154] The embodiments provided by this invention achieve precise resource matching, avoid vehicle overloading or empty runs, improve transportation efficiency, and clarify task boundaries through the MCP protocol to reduce ambiguity at the execution layer.

[0155] Furthermore, this embodiment of the invention achieves real-time monitoring and trajectory comparison. Specifically, the vehicle terminal reports the movement trajectory (latitude and longitude, speed) and transportation status (such as cold chain temperature and cargo integrity) in real time through GPS and IoT sensors; and synchronizes the data to the transportation execution intelligent agent through the A2A protocol.

[0156] Furthermore, trajectory compliance verification is performed by comparing the actual trajectory with the key trajectory point sequence of the optimal transportation route (e.g., determining whether the vehicle missed the "service area" node); trajectory matching algorithms (such as Kalman filtering) are used to correct positioning errors and calculate trajectory deviation distance.

[0157] Furthermore, timeliness deviation analysis: compare the actual arrival time at key trajectory points with the predicted transportation timeliness (e.g., the predicted arrival time at the service area is 10:00, and the actual arrival time is 10:05); if the deviation exceeds the preset threshold (e.g., ±5 minutes), trigger an alert (e.g., "Vehicle V1 has exceeded the time limit at the service area").

[0158] The technical effects achieved by this invention are: rapid anomaly detection by identifying risks such as detours, delays, and congestion through dual comparison of trajectory and timeliness; and data-driven decision-making to provide a basis for dynamic scheduling (such as replacing vehicles with backup).

[0159] Furthermore, the cargo acceptance and result generation provided by this invention specifically involves determining the completion of transportation, i.e., all target vehicles have arrived at their destinations and the trajectory and timeliness verification have passed (if the deviation is within the allowable range), thus triggering the cargo acceptance process.

[0160] Furthermore, automated inspection can be achieved, that is, the goods can be inspected based on preset rules (such as image recognition and weighing sensors) (such as "the outer packaging is undamaged and the quantity is consistent"), and the goods inspection results can be generated (such as "20 boxes of fresh produce are intact and the temperature is compliant").

[0161] The transportation completion report generation provided in this embodiment of the invention integrates transportation trajectory, timeliness data, and inspection results to generate a cargo transportation completion result (such as "Task T1 completed, time taken 2 hours, no abnormalities"), which is then fed back to the main intelligent agent through the A2A protocol to update the global state.

[0162] The technical effect provided by the embodiments of the present invention is to achieve an end-to-end closed loop, that is, to automate the entire process from task issuance to acceptance, reduce manual intervention, and achieve quality controllability, that is, to ensure the accuracy of goods delivery through a rule engine (e.g., fresh produce damage rate <1%).

[0163] Furthermore, the present invention also provides a multi-dimensional data model: the transportation status data cube, which integrates multi-source heterogeneous data such as vehicle trajectory (spatial dimension), timeliness (time dimension), and cargo status (attribute dimension).

[0164] And trajectory comparison models, namely, shortest path algorithms based on graph theory (such as Dijkstra) working together with real-time trajectory matching algorithms to support dynamic path correction.

[0165] In some embodiments, the graph-based shortest path algorithm (such as Dijkstra's algorithm) is linked with the real-time trajectory matching algorithm to support dynamic path correction. The specific execution steps are as follows:

[0166] 1. Initialization phase, i.e., building the road network map: Define nodes (intersections, key points) and edges (road segments), and initialize the static weights of the edges (such as distance and average travel time). Support dynamic weight update mechanisms (such as adjusting weights based on real-time traffic data).

[0167] Load the destination and starting point, input the start and end points, and use Dijkstra's algorithm to calculate the initial shortest path.

[0168] The trajectory matching module is started, which involves continuously tracking the vehicle's position using GPS data and initializing the trajectory matching algorithm (such as a Hidden Markov Model (HMM) or particle filter).

[0169] 2. The real-time trajectory matching step involves acquiring real-time GPS coordinate sequences through in-vehicle devices or mobile terminals by collecting GPS data in real time. GPS points are then matched to the road network, mapping them to nodes or edges within the road network to correct positioning errors (e.g., using map topology and direction information). This embodiment of the invention outputs the current vehicle location, road segment, and direction of travel.

[0170] 3. Dynamic weight update steps: In order to obtain real-time traffic information, road conditions (such as congestion, accidents, construction) are obtained through crowdsourced data, traffic cameras, sensors or third-party APIs (such as Gaode / Google Maps traffic layer).

[0171] The edge weight adjustment step involves dynamically adjusting the edge weights based on real-time data (e.g., increasing the weight of travel time on congested road sections and decreasing the weight of travel time on uncongested road sections). Example: If the average speed on a road section drops to 10 km / h, its weight is updated from a static value (e.g., 1 minute) to the actual travel time (e.g., 10 minutes).

[0172] 4. The present invention provides a path recalculation and dynamic correction step, which is triggered when the vehicle moves a certain distance (e.g., every 500 meters), receives new traffic information, or the trajectory matching result changes significantly.

[0173] And recalculate the shortest path, that is, starting from the current vehicle position, using Dijkstra's algorithm (or a more efficient one). The algorithm calculates the new shortest path to the destination in the updated road network.

[0174] Optimization strategy: Utilize heuristic functions (such as...) Speed ​​up the search; prioritize preserving continuous parts of the original path and reduce frequent switching (e.g., branch only when necessary).

[0175] Furthermore, compare the differences between the new path and the original path: if the new path is better (e.g., time reduction >5%) or the original path is not feasible (e.g., severe congestion ahead), then adopt the new path; otherwise, maintain the original path to avoid frequent modifications.

[0176] 5. This invention provides a navigation instruction update step, which involves decomposing the corrected path into step-by-step navigation instructions (such as "turn right in 500 meters") and pushing them to the driver or autonomous driving system. It also enables visualization and log recording, marking the differences between the original and new paths on the interface (e.g., color differentiation); and recording the path correction history (e.g., correction time, reason, time saved) for subsequent analysis.

[0177] 6. Iterative steps: Continue executing steps 2 to 5 until the vehicle reaches its destination.

[0178] The anomaly handling steps are as follows: when GPS signal is lost, the location is estimated by relying on inertial navigation or historical trajectory; when the road network structure changes (such as temporary road closures), the affected edges are dynamically deleted or disabled.

[0179] Furthermore, the present invention also provides a timeliness prediction model, which uses historical transportation data to train an LSTM neural network to predict the arrival time of key trajectory points.

[0180] In some embodiments of the present invention, after the step of parsing the logistics task and decomposing the parsed logistics task into multiple sub-tasks, the method includes:

[0181] Upon receiving a transportation task query request, the system obtains the execution status of each target sub-agent in executing its corresponding sub-task and displays the execution status of each target sub-agent in executing its corresponding sub-task through a visual interface.

[0182] It is understood that, in this embodiment of the invention, the main intelligent agent monitors the execution status of the sub-tasks after the logistics task is decomposed in real time. Upon receiving a transportation task query request, it dynamically summarizes the execution status of each target sub-intelligent agent (such as the route planning agent, warehouse scheduling agent, and transportation execution agent), and displays the overall task overview to the user through a visual interface (such as a web-based dashboard or a mobile app). The purpose is to centrally present the status of the dispersed sub-tasks, assisting managers in quickly grasping the overall progress and making decisions.

[0183] The specific steps for receiving a transportation task query request can be as follows: User triggers query: Enter the unique identifier of the transportation task (such as order number ORD12345) through the web, APP or API interface.

[0184] The main agent resolves the request, which involves retrieving the corresponding list of subtasks (such as path planning, warehouse scheduling, and transportation execution) from the global task database based on the task ID. The status retrieval instruction distribution involves sending a status query request (such as "Please return the task status of the path planning agent") to the relevant target sub-agents via the first protocol (MCP protocol).

[0185] This invention enables precise task location and rapid association of subtasks via task ID, avoiding full data scanning. It also standardizes the protocol; the MCP protocol ensures consistent communication across agents.

[0186] Furthermore, the steps for obtaining the execution status of the sub-agent in this embodiment of the invention are as follows: Sub-agent status reporting: The path planning agent returns the optimal path and key nodes (e.g., "Warehouse A → Expressway → Delivery Station", with a predicted time of 2 hours); The warehouse scheduling agent returns the outbound status (e.g., "50 boxes have been outbound from Warehouse A, and the remaining inventory is sufficient"); The transportation execution agent returns the vehicle trajectory (e.g., "Vehicle V1 has arrived at the expressway entrance, with an estimated delay of 10 minutes").

[0187] The state data integration step involves the main agent aggregating multi-source state data through a second protocol (A2A protocol) and unifying the data format (such as JSON structured logs).

[0188] The technical effects achieved by the embodiments of this invention are: realizing multi-source data fusion, breaking down data silos among sub-agents, and forming a global view; and enabling anomaly exposure, meaning that abnormal states such as delays and insufficient inventory can be directly identified.

[0189] Furthermore, this embodiment of the invention implements a visual interface display. The specific steps are data mapping and rendering, namely, map overlay trajectory: the GPS trajectory of the transport vehicle is overlaid on the map (such as Baidu Map API), and key nodes (such as warehouses and distribution stations) are marked.

[0190] Status Indicator Lights: Use color to indicate task status (green = normal, red = abnormal, yellow = delayed). Progress Bar and Timeline: Displays the time spent on each subtask and the overall task progress (e.g., "Route planning completed, warehouse outbound in progress, transportation progress 30%"). Interactive Query: Users can click on subtasks to view detailed information (e.g., click "Warehouse A Outbound" to view outbound order details).

[0191] The technical effects achieved by the embodiments of this invention are as follows: It is intuitive, lowering the barrier to information understanding through visualization (e.g., non-technical personnel can also determine whether a task is normal). It provides decision support, highlighting abnormal states to help quickly locate problems (e.g., adjusting delivery routes due to vehicle delays).

[0192] The following describes the logistics data processing device based on multi-agent collaboration provided by the present invention. The logistics data processing device based on multi-agent collaboration described below and the logistics data processing method based on multi-agent collaboration described above can be referred to in correspondence.

[0193] In some embodiments of the present invention, when multiple agents collaboratively process logistics data, the process further includes a step of synchronizing the states of each target sub-agent and the master agent, and improving the collaborative efficiency of each target sub-agent through an event-driven or feedback closed-loop mechanism.

[0194] It's important to note that the master agent, acting as the central brain, has the core responsibility of real-time system state awareness, unified task scheduling, and allocation of sub-tasks among multiple sub-agents. First, the master agent parses external input, such as an order or scheduling request. Then, it uses a task graph to decompose this input into multiple dependent, ordered sub-tasks. Based on the current load, resource status, and historical execution efficiency of each sub-agent, it constructs a multi-objective optimization problem to minimize the total execution cost and latency while satisfying resource constraints. This process is typically modeled as an integer programming or reinforcement learning scheduling problem, considering capacity matching, cost budgeting, and time window constraints in the objective function.

[0195] The main agent then broadcasts the task status to each sub-agent via the MCP protocol, while collecting and synchronizing status receipts to construct a global task state graph.

[0196] In the MCP protocol, the master agent periodically triggers a state synchronization polling mechanism to fuse the local states of all sub-agents. Bayesian updates or graph attention networks are commonly used to achieve cross-node perception fusion, thereby adjusting the scheduling strategy in a timely manner during task execution and dynamically responding to task failures or resource bottlenecks.

[0197] The main intelligent agent MCP implementation steps are: Task Perception: Receive external instructions or system change events, and identify new tasks or anomalies.

[0198] Global Resource Awareness: Broadcasts a request for the current system status (via the A2A protocol or middleware bus); summarizes the status data of all sub-agents.

[0199] Hierarchical Task Allocation: Intelligent matching is performed using a hybrid optimizer (such as MILP + genetic algorithm).

[0200] Task assignment and confirmation mechanism: Assign subtasks → Send confirmation request → Wait for ACK confirmation; if no confirmation is received, proceed to the fault tolerance branch strategy.

[0201] State synchronization: Periodically broadcast a global state summary; synchronize replicas using the Gossip protocol or consensus algorithm (such as Raft).

[0202] The objective function for multi-objective combinatorial optimization is:

[0203] In the formula, Represented as a task Whether by an intelligent agent implement, Indicates cost, Indicates time, Indicates resource consumption. These are expressed as weighting coefficients, reflecting the focus of system optimization.

[0204] Constraints:

[0205] This is used to constrain the requirement that each task must be assigned. This is used to constrain the resource capacity.

[0206] Furthermore, the master agent not only sends task requests but also periodically requests UpdateStatus from the sub-agents via the MCP protocol to update the current state of each agent (such as load, resource usage, etc.). The sub-agents periodically return status information based on the actual task execution.

[0207] Status information content: Load: The current load status of the agent (e.g., computational load, resource usage). Execution status: e.g., "Idle," "Executing," "Completed," etc. Resource consumption: Resources consumed during task execution (e.g., time, bandwidth, computing power, etc.). Timestamp: The time of status feedback, ensuring timely feedback.

[0208] State sharing: State information among sub-agents can be exchanged periodically via an A2A protocol. For example, when an agent completes a task or undergoes a state change, it can send a state update message to other agents, ensuring that all relevant agents have a consistent understanding of the system state. This information includes task progress, fault reports, resource usage, etc.

[0209] Collaborative Computing and Scheduling: During task execution, if multiple agents need to work collaboratively (e.g., multiple agents participating in the same large task), they can exchange data in real time through the A2A protocol to coordinate resource usage, task splitting, etc. For example, if one agent's computational load is too high, it can request computing resources from other agents to share the task.

[0210] In some embodiments, adaptive scheduling is achieved by combining reinforcement learning algorithms such as Q-learning to enhance the system's ability to allocate and optimize resources in complex scenarios.

[0211] Reinforcement learning method: Centralized Training, Distributed Execution (CTDE). Training phase: All agents' policies are trained under centralized control, allowing them to share information and optimize the global objective. Execution phase: Each agent makes decisions based on its own independent perception and execution environment, without needing to communicate with each other, but still maintaining cooperative behavior.

[0212] Multi-agent reinforcement learning (MARL) includes at least the following:

[0213] Independent Q-learning: Each agent performs reinforcement learning as an independent entity, using Q-learning or other RL methods to determine its behavior. Each agent's reward is based solely on its own actions and states.

[0214] Joint Q-learning: Multiple agents share a global reward function, or jointly optimize a task through information sharing and cooperation. For example, a task can optimize its global policy based on the interactions and resource sharing of agents.

[0215] MADDPG (Multi-Agent Deep Deterministic Policy Gradient) is a deep reinforcement learning method that can be extended to multi-agent systems, suitable for multi-agent systems with continuous action spaces. Each agent learns its policy by considering the policies of other agents, thus enabling multi-agent cooperation.

[0216] Furthermore, the collaboration between reinforcement learning methods and agents is a cooperative game model, specifically:

[0217] A shared reward mechanism is implemented, where all agents share a global reward function. Task objectives can include minimizing cost, maximizing resource utilization, and optimizing task completion time. Through shared rewards, agents can learn collaborative strategies during training. For example, the system can encourage agents to work collaboratively by sharing task completion rates, efficiency, and resource consumption as reward functions.

[0218] In task scheduling scenarios, deep reinforcement learning algorithms such as Deep Q-Network (DQN) or Proximal Policy Optimization (PPO) can be used to train agents to learn how to dynamically adjust task allocation based on the current environment and task requirements.

[0219] Collaborative DQN involves multiple agents using a centralized Q-network to evaluate the optimal execution strategy for each task and coordinating resource allocation through a reward function.

[0220] Asynchronous execution and state updates ensure real-time state updates between agents through A2A or MCP protocols. Agents periodically synchronize their states by sharing perception, thereby adjusting their strategies. This approach is also suitable for using the Actor-Critic method for multi-agent state synchronization and decision optimization.

[0221] Furthermore, the steps for training a reinforcement learning model are as follows:

[0222] Set up the training environment: the current task status (e.g., task progress, resource usage, task priority, etc.). The resource status of each agent (e.g., computing power, bandwidth, task execution capability, etc.). The shared environment status (e.g., overall system load, global task completion rate, etc.).

[0223] State space and motion space design steps:

[0224] The state space design for each agent can include the current task execution status, resource utilization status, target distance, execution status, etc.

[0225] The action space is designed so that each agent can take actions such as selecting tasks, choosing execution paths, adjusting resource allocation, and cooperating with other agents.

[0226] Reward Function Design: The design of the reward function is crucial to ensuring the agent learns the optimal policy. Rewards can be set based on the following criteria:

[0227] Task Completion: Whether each agent's task was successfully executed or failed. Resource Utilization: Whether the agent effectively utilizes resources, such as reducing energy consumption and improving transportation efficiency. Collaboration Efficiency: Whether the agent coordinates with other agents to achieve a common goal (e.g., reducing overall execution time and lowering costs).

[0228] Reward functions include the following forms: Positive rewards: task success, goal achievement, efficient resource utilization, etc. Negative rewards: task failure, resource waste, execution delay, etc.

[0229] Furthermore, the training steps for model training:

[0230] Initialize the model: Initialize the reinforcement learning model (such as DQN, PPO, etc.) according to the task requirements.

[0231] Training process: The strategy of each agent is continuously updated through interaction with the simulated environment. Each agent makes decisions by combining exploration and exploitation.

[0232] Gain experience: Improve training efficiency and reduce data correlation through experience replay pooling or batch sampling.

[0233] Policy update: The policy is updated using a reinforcement learning algorithm based on the agent's performance in the environment.

[0234] Performance evaluation: Real-time evaluation is performed during training to check metrics such as task completion and resource utilization.

[0235] Furthermore, it also includes testing and verification steps:

[0236] Single agent testing: During training, the behavior of a single agent can be tested first, and its performance under different task conditions can be verified step by step.

[0237] Multi-agent testing: In multi-agent collaboration, the collaborative capabilities, resource sharing, and effectiveness of optimization strategies among agents are verified by simulating a multi-agent environment.

[0238] Performance evaluation metrics: The performance of reinforcement learning models is evaluated using metrics such as task completion time, resource consumption, and system throughput.

[0239] In some embodiments of the present invention, a protocol bridging module is also included:

[0240] MCP Protocol Module: The main intelligent agent uses this module for task scheduling, resource control, and global state monitoring.

[0241] The bridging module is responsible for converting MCP protocol messages into A2A protocol requests that sub-agents can understand, and for completing task distribution and result feedback.

[0242] A2A Protocol: Sub-agents use the A2A protocol to perform tasks and provide status feedback to the master agent or other agents.

[0243] After setting up the protocol bridging module, rigorous verification and testing are required to ensure that the master agent can invoke the capabilities of each agent via MCP commands and provide rapid feedback on task results through the A2A protocol. The testing process typically includes the following aspects:

[0244] Functional Testing: Verify that the primary agent can correctly send task requests to the bridging module via the MCP protocol. Check that the bridging module can correctly forward task requests to the child agents and ensure that the data format and protocol conversion are correct. Ensure that the child agents can execute tasks according to the task requests and provide feedback on the task execution results via the A2A protocol.

[0245] Furthermore, it also includes a message passing verification step: testing the task scheduling process from the main agent to the sub-agent to ensure that information such as task ID, task type, and parameters can be accurately transmitted.

[0246] Verify whether the sub-agent has performed the task according to the task request, and return the result to the master agent via the A2A protocol.

[0247] Furthermore, it also includes latency and response time testing steps: measuring the transmission latency of task requests from the main agent to the sub-agents, and ensuring that the response time meets the system's real-time requirements.

[0248] Test whether the task feedback of the A2A protocol can be returned quickly, and ensure that the results of task execution can be fed back to the main intelligent agent in a timely manner.

[0249] Furthermore, it also includes state synchronization and consistency checks: testing whether the main agent can obtain the latest state of each child agent through the UpdateStatus request.

[0250] Verify the state synchronization mechanism between the agents to ensure the consistency and correctness of state information.

[0251] Furthermore, it also includes fault tolerance and error handling steps: simulating situations where the sub-agent fails or malfunctions, ensuring that the bridging module can capture errors and report them to the master agent via the MCP protocol.

[0252] Test how the master agent reschedules tasks when a sub-agent fails to ensure the robustness of the system.

[0253] Figure 2 shows a schematic diagram of the logistics data processing device based on multi-agent collaboration provided by the present invention. The device is applied to a master agent, which is connected to any sub-agent via a first protocol, and each sub-agent is connected to the others via a second protocol. The device includes:

[0254] The acquisition module 210 is used to acquire user order information and generate logistics tasks based on the user order information;

[0255] The decomposition module 220 is used to parse the logistics task and decompose the parsed logistics task into multiple sub-tasks;

[0256] The receiving module 230 is used to broadcast multiple subtasks to each sub-agent through the first protocol and receive sub-agent information fed back by each sub-agent; the sub-agent information includes at least the state information and task processing type information of each sub-agent;

[0257] The determination module 240 is used to make decisions based on multiple sub-tasks, the state information of each sub-agent, and the task processing type information to determine the target sub-agent that receives the corresponding sub-task; wherein, the multiple target sub-agents are arranged in a preset order, and the first target sub-agent sends the corresponding sub-task execution result to the last target sub-agent through the second protocol, and each target sub-agent executes the corresponding type of sub-task;

[0258] The execution module 250 is used to send the subtask to each target sub-agent through a first protocol, so that each target sub-agent generates a corresponding subtask execution result according to the preset order based on the subtask and the execution result of the preceding subtask, until the last target sub-agent in the preset order generates the target subtask execution result.

[0259] Preferably, the logistics data processing device for multi-agent collaboration provided by the present invention is specifically used to send a first sub-task to a first target sub-agent through a first protocol, so that the first target sub-agent executes the corresponding first sub-task and generates the corresponding first sub-task execution result;

[0260] The second subtask is sent to the second target sub-agent via the first protocol, so that the second target sub-agent generates the execution result of the second subtask based on the execution results of the second subtask and the first subtask; wherein, the execution result of the first subtask is sent to the second target sub-agent by the first target sub-agent via the second protocol;

[0261] The third subtask is sent to the third target sub-agent via the first protocol, so that the third target sub-agent generates the target subtask execution result based on the third subtask, the execution result of the first subtask received from the first target sub-agent via the second protocol, and the execution result of the second subtask received from the second target sub-agent via the second protocol.

[0262] Preferably, the logistics data processing device for multi-agent collaboration provided by the present invention is specifically used to represent the first target sub-agent as a path planning agent, for generating the optimal transportation path;

[0263] The first subtask is sent to the path planning agent via a first protocol, so that the path planning agent performs optimal path calculation based on the first subtask, current road conditions, weather information, cargo information, timeliness information, and cost information to generate an optimal transportation path; wherein, the optimal transportation path is characterized as the path that minimizes the timeliness, costs, and accuracy of transporting goods from the warehouse to the target delivery location; the optimal transportation path includes at least multiple key trajectory points, which are used to determine whether there is any deviation in the actual transportation trajectory of the transport vehicle.

[0264] Preferably, the logistics data processing device for multi-agent collaboration provided by the present invention is specifically used to represent the second target sub-agent as a warehouse scheduling agent, for updating warehouse goods entry and exit information;

[0265] The second subtask is sent to the warehouse scheduling agent, so that the warehouse scheduling agent can determine the target warehouse with sufficient goods and the shortest distance from the target delivery location based on the second subtask and the optimal transportation route of the goods, and release the corresponding goods from the target warehouse.

[0266] Preferably, the logistics data processing device for multi-agent collaboration provided by the present invention is specifically used to represent the third target sub-agent as a transportation execution agent, and to generate state information of transported goods;

[0267] The third subtask is sent to the transportation execution agent via the first protocol, so that the transportation execution agent determines the target number of target transportation vehicles based on the third subtask, the optimal transportation route, and the goods corresponding to the outbound shipment from the target warehouse. The agent records the movement trajectory and actual transportation time of each target transportation vehicle to the key trajectory point in real time, compares the movement trajectory with the optimal transportation route, and compares the actual transportation time with the predicted transportation time to the key trajectory point. If the movement trajectory does not deviate from the optimal transportation route, the difference between the actual transportation time and the predicted transportation time to the key trajectory point is within a preset range, and all target transportation vehicles of the target number transport the goods to the target delivery location, the agent inspects the goods based on preset inspection rules, generates a goods inspection result, and generates a goods transportation completion result based on the goods inspection result.

[0268] Preferably, the logistics data processing device for multi-agent collaboration provided by the present invention is specifically used to, upon receiving a transportation task query request, obtain the execution status of each target sub-agent executing the corresponding sub-task, and display the execution status of each target sub-agent executing the corresponding sub-task through a visual interface.

[0269] This invention provides a logistics data processing method, apparatus, device, medium, and computer program product based on multi-agent collaboration. Applied to a master agent, the master agent connects to any sub-agents via a first protocol, and each sub-agent connects to each other via a second protocol. The method involves acquiring user order information and generating logistics tasks based on that information; parsing the logistics tasks and decomposing them into multiple sub-tasks; broadcasting these sub-tasks to each sub-agent via the first protocol; and receiving feedback from each sub-agent. The sub-agent information includes at least the state information and task processing type information of each sub-agent. The process involves decision-making based on subtasks, the state information of each sub-agent, and task processing type information to determine the target sub-agent receiving the corresponding subtask. Multiple target sub-agents are arranged in a preset order. The first target sub-agent sends its corresponding subtask execution result to the next target sub-agent via a second protocol. Each target sub-agent executes a subtask of the corresponding type. The subtask is then sent to each target sub-agent via a first protocol, enabling each target sub-agent to generate its corresponding subtask execution result based on the subtask and the execution result of the first subtask, in the preset order, until the last target sub-agent in the preset order generates its target subtask execution result. This approach addresses the shortcomings of low logistics transportation efficiency and poor resource coordination in existing technologies, enabling more effective coordination and management of logistics resources, improving logistics efficiency, reducing logistics costs, and enhancing logistics service quality.

[0270] Figure 3 illustrates a schematic diagram of the physical structure of an electronic device. As shown in Figure 3, the electronic device may include: a processor 310, a communication interface 320, a memory 330, and a communication bus 340. The processor 310, the communication interface 320, and the memory 330 communicate with each other through the communication bus 340. The processor 310 can call logical instructions in the memory 330 to execute a logistics data processing method based on multi-agent collaboration. This method includes: acquiring user order information and generating a logistics task based on the user order information; parsing the logistics task and decomposing the parsed logistics task into multiple sub-tasks; broadcasting the multiple sub-tasks to each sub-agent via a first protocol and receiving sub-agent information fed back by each sub-agent; the sub-agent information includes at least the state information and task processing type information of each sub-agent; performing decision processing based on the multiple sub-tasks, the state information of each sub-agent, and the task processing type information to determine the target sub-agent receiving the corresponding sub-task; wherein the multiple target sub-agents are arranged in a preset order, and the first target sub-agent sends the corresponding sub-task execution result to the next target sub-agent via a second protocol, and each target sub-agent executes a sub-task of the corresponding type; the sub-tasks are sent to each target sub-agent via the first protocol, so that each target sub-agent generates a corresponding sub-task execution result based on the sub-tasks and the execution result of the first sub-task in the preset order, until the last target sub-agent in the preset order generates the target sub-task execution result.

[0271] Furthermore, the logical instructions in the aforementioned memory 330 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0272] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the logistics data processing method based on multi-agent collaboration provided by the above methods. The method includes: acquiring user order information and generating a logistics task based on the user order information; parsing the logistics task and decomposing the parsed logistics task into multiple sub-tasks; broadcasting the multiple sub-tasks to each sub-agent through the first protocol; and receiving sub-agent information fed back by each sub-agent; the sub-agent information includes at least the state information and task of each sub-agent. The system processes type information; based on multiple subtasks, the state information of each sub-agent, and task processing type information, it performs decision processing to determine the target sub-agent to receive the corresponding subtask; wherein, multiple target sub-agents are arranged in a preset order, and the first target sub-agent sends the corresponding subtask execution result to the next target sub-agent through a second protocol, and each target sub-agent executes the corresponding type of subtask; the subtask is sent to each target sub-agent through a first protocol, so that each target sub-agent generates the corresponding subtask execution result according to the preset order based on the subtask and the execution result of the first subtask, until the last target sub-agent in the preset order generates the target subtask execution result.

[0273] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements the logistics data processing method based on multi-agent collaboration provided by the above methods. This method includes: acquiring user order information; generating a logistics task based on the user order information; parsing the logistics task and decomposing the parsed logistics task into multiple sub-tasks; broadcasting the multiple sub-tasks to each sub-agent via a first protocol; and receiving sub-agent information fed back by each sub-agent; the sub-agent information includes at least the state information and task processing type information of each sub-agent; based on multiple sub-tasks, The state information and task processing type information of each sub-agent are used for decision processing to determine the target sub-agent to receive the corresponding sub-task. Multiple target sub-agents are arranged in a preset order. The first target sub-agent sends the corresponding sub-task execution result to the next target sub-agent via a second protocol. Each target sub-agent executes a sub-task of the corresponding type. The sub-task is then sent to each target sub-agent via a first protocol, so that each target sub-agent generates a corresponding sub-task execution result based on the sub-task and the execution result of the first sub-task in the preset order, until the last target sub-agent in the preset order generates the target sub-task execution result.

[0274] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0275] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0276] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A logistics data processing method based on multi-agent collaboration, characterized in that, This method is applied to a master agent, which connects to any sub-agents via a first protocol, and each sub-agent connects to each other via a second protocol. The method includes: acquiring user order information; generating a logistics task based on the user order information; parsing the logistics task and decomposing the parsed logistics task into multiple sub-tasks; broadcasting the multiple sub-tasks to each sub-agent via the first protocol; and receiving sub-agent information from each sub-agent; the sub-agent information includes at least the state information and task processing type information of each sub-agent; and making decisions based on the multiple sub-tasks, the state information of each sub-agent, and the task processing type information. The process involves determining the target sub-agent to receive the corresponding sub-task; wherein multiple target sub-agents are arranged in a preset order, and the first target sub-agent sends the corresponding sub-task execution result to the last target sub-agent via a second protocol, with each target sub-agent executing a sub-task of a corresponding type; the sub-task is then sent to each target sub-agent via a first protocol, so that each target sub-agent generates a corresponding sub-task execution result based on the sub-task and the execution results of all the preceding sub-tasks in the preset order, until the last target sub-agent in the preset order generates the target sub-task execution result; the sub-task is then sent to each target sub-agent via the first protocol. The protocol is sent to each target sub-agent so that each target sub-agent generates a corresponding sub-task execution result according to the preset order based on the sub-task and the execution result of the preceding sub-task, until the last target sub-agent in the preset order generates the target sub-task execution result. This includes: sending a first sub-task to a first target sub-agent via a first protocol, so that the first target sub-agent executes the corresponding first sub-task and generates a corresponding first sub-task execution result; sending a second sub-task to a second target sub-agent via the first protocol, so that the second target sub-agent generates a second sub-task execution result based on the second sub-task and the execution result of the first sub-task; wherein the first sub-task execution result is sent by the first target sub-agent to the second target sub-agent via the second protocol; and sending a third sub-task to a third target sub-agent via the first protocol, so that the third target sub-agent generates the target sub-task execution result based on the third sub-task, the first sub-task execution result received from the first target sub-agent via the second protocol, and the second sub-task execution result received from the second target sub-agent via the second protocol; the first target sub-agent is represented as a path planning agent used to generate the optimal transportation path.The step of sending the first subtask to the first target sub-agent via the first protocol, so that the first target sub-agent executes the corresponding first subtask and generates the corresponding first subtask execution result, includes: sending the first subtask to the path planning agent via the first protocol, so that the path planning agent performs optimal path calculation processing based on the first subtask, current road condition information, weather information, cargo information, timeliness information, and cost information to generate an optimal transportation path; wherein, the optimal transportation path is characterized as the path that minimizes the timeliness, costs, and transportation accuracy of transporting goods from the warehouse to the target delivery location; the optimal transportation path includes at least multiple key trajectory points, the... Key trajectory points are used to determine whether there is a deviation in the actual transportation trajectory of the transport vehicle; the second target sub-agent is represented as a warehouse scheduling agent, used to update warehouse goods entry and exit information; the step of sending the second sub-task to the second target sub-agent through the first protocol, so that the second target sub-agent generates the second sub-task execution result based on the second sub-task and the first sub-task execution result, includes: sending the second sub-task to the warehouse scheduling agent, so that the warehouse scheduling agent, based on the second sub-task and the optimal transportation path of the goods, determines the target warehouse with sufficient goods and the shortest distance from the target delivery location, and retrieves the corresponding goods from the target warehouse; The third target sub-agent is represented as a transportation execution agent, used to generate status information of transported goods. The step of sending the third sub-task to the third target sub-agent via the first protocol, so that the third target sub-agent generates a target sub-task execution result based on the third sub-task, the execution result of the first sub-task received from the first target sub-agent via the second protocol, and the execution result of the second sub-task received from the second target sub-agent via the second protocol, includes: sending the third sub-task to the transportation execution agent via the first protocol, so that the transportation execution agent, based on the third sub-task, the optimal transportation route, and the goods corresponding to the outbound shipment from the target warehouse,... A target number of target transport vehicles are determined, and the movement trajectory and actual transport time to key trajectory points of each target transport vehicle are recorded in real time. The movement trajectory is compared with the optimal transport route, and the actual transport time is compared with the predicted transport time to the key trajectory point. If the movement trajectory does not deviate from the optimal transport route, the difference between the actual and predicted transport time to the key trajectory point is within a preset range, and all target transport vehicles transport the goods to the target destination, the goods are inspected based on preset inspection rules, a goods inspection result is generated, and a goods transport completion result is generated based on the goods inspection result.

2. The logistics data processing method based on multi-agent collaboration according to claim 1, characterized in that, After the steps of parsing the logistics task and decomposing the parsed logistics task into multiple sub-tasks, the method includes: upon receiving a transportation task query request, obtaining the execution status of each target sub-agent executing the corresponding sub-task, and displaying the execution status of each target sub-agent executing the corresponding sub-task through a visual interface.

3. A logistics data processing device based on multi-agent collaboration, applied to the logistics data processing method based on multi-agent collaboration as described in claim 1 or 2, characterized in that, The device is applied to a main intelligent agent, which connects to any sub-intelligent agents via a first protocol, and each sub-intelligent agent connects to each other via a second protocol. The device includes: an acquisition module for acquiring user order information and generating a logistics task based on the user order information; a decomposition module for parsing the logistics task and decomposing the parsed logistics task into multiple sub-tasks; a receiving module for broadcasting the multiple sub-tasks to each sub-intelligent agent via the first protocol and receiving sub-intelligent agent information fed back by each sub-intelligent agent; the sub-intelligent agent information includes at least the state information and task processing type information of each sub-intelligent agent; and a determination module for determining the sub-tasks based on the multiple sub-tasks and each sub-intelligent agent. The system performs decision processing on the state information and task processing type information of the target intelligent agent to determine the target intelligent agent to receive the corresponding sub-task. Multiple target intelligent agents are arranged in a preset order. The first target intelligent agent sends the corresponding sub-task execution result to the next target intelligent agent via a second protocol. Each target intelligent agent executes a sub-task of the corresponding type. The execution module is used to send the sub-task to each target intelligent agent via a first protocol, so that each target intelligent agent generates a corresponding sub-task execution result based on the sub-task and the execution result of the first sub-task in the preset order, until the last target intelligent agent in the preset order generates the target sub-task execution result.

4. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the program, it implements the logistics data processing method based on multi-agent collaboration as described in claim 1 or 2.

5. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the logistics data processing method based on multi-agent collaboration as described in claim 1 or 2.

6. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the logistics data processing method based on multi-agent collaboration as described in claim 1 or 2.

Images