Intelligent logistics management system based on large language model

Abstract

The application provides a kind of intelligent logistics management system based on large language model, it is related to intelligent logistics management field.The application adopts low rank adaptation technology general large language model in advance to fine-tune, and innovatively the external LLM service obtained by fine-tuning is deeply integrated into task decision process, so that it can realize the deep understanding of user intent, accurate analysis of complex environment situation, and based on this, independently generate optimal distribution scheme.In addition, the city environment management module, logistics management scheduling module, simulation engine module and large language model interface module included in the system do not work independently, but closely cooperate through event-driven and message passing mechanism, thereby forming a complete unmanned aerial vehicle logistics management closed loop from environment perception to decision, execution to event feedback, this architecture can reflect and control unmanned aerial vehicle distribution task in real time, correct the established scheme, thereby ultimately significantly improving the intelligent level and operation efficiency of logistics management.

Description

Technical Field

[0001] This invention relates to the field of intelligent logistics management, and specifically to an intelligent logistics management system based on a large language model. Background Technology

[0002] The application prospects of drones in the logistics industry are very broad, and urban drone delivery systems are gradually becoming an important force driving the transformation of the modern logistics industry.

[0003] In existing technologies, urban drone delivery systems mainly employ the following two methods for task scheduling and resource allocation:

[0004] Rule-based drone delivery expert systems typically use pre-defined rules to statically schedule drone missions. For example, the paper "Research on Urban Drone-Public Transport Cooperative System" (Li Dong & Li Hongtao, 2023) proposes a rule-driven campus delivery model that automates route planning and task execution by setting triggering mechanisms.

[0005] Optimization path scheduling systems based on mathematical modeling typically construct graph structures based on urban road networks, considering multiple constraints such as delivery time windows, power limits, load capacity, and traffic forecasting. They employ linear programming, integer programming, and genetic algorithms to find the optimal path solution. For example, the paper "Research on Route Optimization of Express Delivery Vehicles" (He Linlin et al., 2021) constructs an urban express delivery model based on multiple constraints and uses a genetic algorithm to solve for path optimization.

[0006] However, both rule-based expert systems and mathematical optimization-based decision-making systems have inherent limitations in understanding and coping with complex and ever-changing urban environments. Rule-based systems, due to their hard-coded decision-making logic, exhibit rigidity and poor adaptability when faced with unforeseen or novel environmental situations (e.g., sudden weather changes, temporary no-fly zones, dynamic traffic congestion, and interference from new types of drones), and their rule maintenance costs are high. Mathematical optimization-based systems, on the other hand, heavily rely on explicit assumptions and parameters, resulting in a significant decrease in effectiveness when actual conditions deviate from the model. Furthermore, they struggle to effectively handle unstructured and noisy environmental information (such as breaking news on social media and blurry real-time traffic images), leading to insufficient robustness in complex and ever-changing real-world environments. Summary of the Invention

[0007] (a) Technical problems to be solved

[0008] To address the shortcomings of existing technologies, this invention provides an intelligent logistics management system based on a large language model, which solves the technical problem of how to improve the comprehensive understanding and adaptive decision-making capabilities of complex and dynamic urban environments.

[0009] (II) Technical Solution

[0010] To achieve the above objectives, the present invention provides the following technical solution:

[0011] An intelligent logistics management system based on a large language model includes an urban environment management module, a logistics management and scheduling module, a simulation engine module, and a large language model interface module.

[0012] The urban environment management module is responsible for storing and maintaining in real time environmental situation data, including all drones, all order tasks, and historical events, and formatting it as delivery data;

[0013] The logistics management and scheduling module is used to execute the current decision-making process, including:

[0014] The user instruction data is fused with the delivery data to construct structured LLM request data; a request is sent to an external LLM service through the large language model interface module, and an unstructured text response is received from the external LLM service; the unstructured text response is parsed into LLM decision data, the LLM decision is translated into a series of delivery instruction data, and then sent to the simulation engine module; wherein a low-rank adaptation technique is used in advance to fine-tune the general large language model to obtain the external LLM service;

[0015] The simulation engine module is used to execute delivery instructions, simulate the physical behavior and task execution process of the drone, and trigger corresponding events. The events are fed back to the city environment management module to update the environmental situation data, and to the logistics management and scheduling module to trigger a new round of decision-making process.

[0016] Preferably, the intelligent logistics management system includes a visualization and command interaction module, which includes a simulation monitoring window and a command decision window;

[0017] The simulation monitoring window is used to display, in real time, at least the drone location, order status, and event timeline in the environmental situation data;

[0018] The command and decision window is used to receive user instruction data and send it to the logistics management and scheduling module, display the input prompts and responses of the external LLM service in real time, and display warning information in real time.

[0019] Preferably, the logistics management and scheduling module is used for

[0020] When parsing the unstructured text response into LLM decision data, a preset intelligent resource allocation logic is executed, including:

[0021] Based on the delivery data, a global resource pre-check is performed. If the total available drone resources are insufficient, the allocation of subsequent order tasks is stopped and recorded as a global resource shortage event. If the total resources are sufficient, the global resource pre-check is passed and the optimal solution attempt stage is entered.

[0022] In the optimal solution attempt phase, all available drones are traversed to try to find a single drone that can independently and completely meet the requirements of the current order task. If successful, the current order task is assigned to that single drone; otherwise, the collaborative solution attempt phase is entered.

[0023] In the collaborative solution attempt phase, the current order task requirements are decomposed, and an attempt is made to search for a combination of two or more available drones to collaboratively meet the current order task requirements. If successful, the current order task is assigned to a combination of two or more available drones; otherwise, the assignment of the next order task is initiated.

[0024] After all the above allocation attempts, the final stage begins. The successfully allocated order tasks are translated into the delivery instruction data, and for all order tasks that ultimately fail to be allocated, their status is updated to "insufficient resources, waiting for allocation", and recorded as allocation failure events and their reasons for failure.

[0025] The visualization and command interaction module displays warning messages about global resource shortage events and / or allocation failure events, along with the reasons for their failure.

[0026] Preferably, after receiving LLM request data, the external LLM service identifies the decision type of the current task, including:

[0027] When user command data is input for the first time or a request is made for autonomous system scheduling, the external LLM service identifies it as an initial scheduling decision; when periodic monitoring shows that the task is not completed and the drone has not reached its destination, the external LLM service identifies it as a continuous task allocation or refueling decision; when the drone is interfered with, the external LLM service identifies it as an interference response decision and formulates a new scheduling strategy based on the interference situation.

[0028] Preferably, the logistics management and scheduling module is used for

[0029] After parsing the unstructured text response into LLM decision data, corresponding system prompts and user prompts are constructed based on the decision type.

[0030] Preferably, the events include interference events, and the simulation engine module is used for

[0031] During the simulated drone mission execution, the probability of the drone being interfered with is calculated based on the distance between the drone's current position and the interference source; where the interference source refers to a physical or electronic signal source that interferes with the individual drone, and the interference probability is represented by an attenuation function.

[0032]

[0033] Among them, P interference (d interf ) represents the probability of interfering with an individual drone; d interf P represents the distance between the current location of the drone and the interference source. max d represents the maximum probability of interference that the drone can achieve near the interference source. max β represents the maximum effective range of the interference source, indicating that the drone will not be interfered with beyond this distance; β is the attenuation factor of the interference probability, used to adjust the rate at which the interference intensity decreases with distance.

[0034] When d interf <d max At that time, the drone is considered to be within the interference range and the corresponding interference probability P is calculated. interference (d interf Based on this probability value, random sampling is performed, including:

[0035] Generate a pseudo-random number (Random Number) that is uniformly distributed in the interval [0,1]. If the Random Number is greater than P... interference (d interf If the condition is met, the drone is determined to be uninterrupted; otherwise, the drone is determined to be interfered with, the drone's flight status is set to an interfered state, and an interference event corresponding to the interference source is triggered.

[0036] Preferably, the events include task completion events, task incomplete events, and task failure events, and the simulation engine module is used for

[0037] Based on the drone's flight distance, average speed, and environmental factors, the corrected actual delivery time is obtained and expressed as follows:

[0038]

[0039] EnvironmentalFactor=C1·WindSpeed+C2·LoadRatio+C3·TerrainComplexity

[0040] Wherein T(d) flight ) indicates the actual delivery time of the order task; d flightis the actual flight distance of the drone; v is the average flight speed of the drone; EnvironmentalFactor is the comprehensive correction factor for the time consumption caused by environmental factors; C1, C2, and C3 are the weight coefficients of the corresponding items, WindSpeed ​​is the current wind speed, LoadRatio is the ratio of the drone's current load to its maximum load, and TerrainComplexity is the terrain complexity of the flight path.

[0041] Based on whether the order task was interrupted, and the corrected actual delivery time T(d) flight For the delivery time window of the order task, the following judgment process is performed:

[0042] If the drone was not interfered with and T(d) flight If the order is within or before the delivery time window, the order task status will be updated to completed, and a task completion event containing success information will be triggered.

[0043] If the drone was not interfered with and T(d) flight After the delivery time window, the order task status will be updated to timed out, and a task incomplete event will be triggered.

[0044] If the drone has been interfered with, causing the mission to be aborted or other simulation conditions to fail, the order mission status will be updated to failed, and a mission failure event containing the reason for the failure will be triggered.

[0045] Preferably, the delivery data is a structured snapshot of the environmental situation data, wherein the structured snapshot refers to a complete data set containing the environmental situation data in a predefined format.

[0046] Preferably, the logistics management scheduling module is used to run an LLM worker thread in a separate thread.

[0047] Preferably, the logistics management and scheduling module is used to maintain a task tracking list to manage all ongoing decision-making processes.

[0048] (III) Beneficial Effects

[0049] This invention provides an intelligent logistics management system based on a large language model. Compared with existing technologies, it has the following advantages:

[0050] This invention pre-tunes a general-purpose large language model using low-rank adaptation technology and innovatively integrates the fine-tuned external LLM service into the task decision-making process. This enables the system to deeply understand user intent, accurately analyze complex environmental situations, and autonomously generate optimal delivery plans. Furthermore, the system's urban environment management module, logistics management and scheduling module, simulation engine module, and large language model interface module do not operate independently but collaborate closely through event-driven and message-passing mechanisms. This forms a complete closed loop for UAV logistics management, from environmental perception to decision-making, execution, and event feedback. This architecture can reflect and control UAV delivery tasks in real time, self-correcting predetermined plans, thereby significantly improving the intelligence level and operational efficiency of logistics management. Attached Figure Description

[0051] To more clearly illustrate the technical solutions in the embodiments of the present 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 only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0052] Figure 1 A structural block diagram of an intelligent logistics management system based on a large language model provided in an embodiment of the present invention;

[0053] Figure 2 This is a schematic diagram illustrating the data flow process of an automated closed-loop decision-making and feedback mechanism provided in an embodiment of the present invention. Detailed Implementation

[0054] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention are described clearly and completely. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0055] This application provides an intelligent logistics management system based on a large language model, which solves the technical problem of how to improve the comprehensive understanding and adaptive decision-making ability of complex and dynamic urban environments.

[0056] The technical solution in this application is to solve the above-mentioned technical problems, and the general idea is as follows:

[0057] This invention aims to overcome the shortcomings of existing drone logistics management systems in environmental adaptation (comprehensive understanding and adaptive decision-making capabilities in complex and dynamic urban environments), human-computer interaction, decision transparency, and system iteration and maintenance. It constructs an intelligent drone logistics management system based on a localized, fine-tuned Large Language Model (LLM). This system, through fine-tuning the LLM, enables it to accurately grasp drone scheduling and task planning knowledge. Its core objectives are: to improve real-time environmental perception and task response capabilities, shortening the Observe-Orient-Decide-Act (OODA) cycle; to achieve intelligent scheduling and path planning, supporting natural language commands and automatically generating compliant scheduling schemes; to ensure resource optimization and risk management, maximizing order fulfillment rates, and intelligently selecting the optimal drone combination and path through the LLM.

[0058] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific implementation methods.

[0059] Example 1:

[0060] This invention provides an intelligent logistics management system based on a large language model, such as... Figure 1 As shown, it includes a city environment management module, a logistics management and scheduling module, a simulation engine module, and a large language model interface module.

[0061] The urban environment management module is responsible for storing and maintaining in real time environmental situation data, including all drones, all order tasks, and historical events, and formatting it as delivery data;

[0062] The logistics management and scheduling module is used to execute the current decision-making process, including:

[0063] The user instruction data is fused with the delivery data to construct structured LLM request data; a request is sent to an external LLM service through the large language model interface module, and an unstructured text response is received from the external LLM service; the unstructured text response is parsed into LLM decision data, the LLM decision is translated into a series of delivery instruction data, and then sent to the simulation engine module; wherein a low-rank adaptation technique is used in advance to fine-tune the general large language model to obtain the external LLM service;

[0064] The simulation engine module is used to execute delivery instructions, simulate the physical behavior and task execution process of the drone, and trigger corresponding events. The events are fed back to the city environment management module to update the environmental situation data, and to the logistics management and scheduling module to trigger a new round of decision-making process.

[0065] This system closely integrates LLM with the real-world environment, enabling closed-loop feedback and dynamic adjustments to effectively respond to emergencies, ultimately reducing system costs and significantly improving the intelligence level and operational efficiency of drone logistics management.

[0066] In an optional implementation, the intelligent logistics management system provided in this embodiment of the invention, such as... Figure 1 As shown, it also includes a visualization and command interaction module, which includes a simulation monitoring window and a command decision window.

[0067] The simulation monitoring window is used to display, in real time, at least the drone location, order status, and event timeline in the environmental situation data;

[0068] The command and decision window is used to receive user instruction data and send it to the logistics management and scheduling module, display the input prompts and responses of the external LLM service in real time, and display warning information in real time.

[0069] Understandably, the visualization and command interaction modules enhance the transparency of command interaction and decision-making. This increased transparency allows users to intuitively understand the input basis and output results of AI decisions, thereby improving trust in human-machine collaboration and promoting efficient logistics management.

[0070] The system architecture provided in this invention is sophisticated, forming a complete closed loop for UAV logistics management (i.e., an 'automated closed-loop decision-making and feedback mechanism') from environmental perception to decision-making, execution, and event feedback. The core of this automated closed-loop decision-making and feedback mechanism lies in a coherent, event-driven data flow, the complete flow of which is as follows: Figure 2 As shown, it can be divided into the following core stages:

[0071] Phase 1: Environmental Perception and Command Input

[0072] This stage marks the initiation of the process, with the core focus on collecting all the contextual information required for decision-making. The process can begin when the visualization and command interaction module receives a user instruction data. This user instruction data carries the user's high-level intent. In this embodiment of the invention, the "high-level intent" defines the desired business objective (e.g., "prioritize delivery of all VIP orders"), rather than specific, machine-executable steps (e.g., "command drone A to fly to coordinate point B"). This goal-oriented instruction requires the system to possess semantic understanding and strategy planning capabilities.

[0073] This user instruction data, carrying the intentions of high-level management, is transmitted to the logistics management and scheduling module, which acts as the "control center." To translate this intention into action, this module requests delivery data from the city environment management module, which serves as the "digital twin" data center. This delivery data is a structured snapshot of the system's real-time status (i.e., environmental situational data) at a specific point in time. A "structured snapshot" refers to a complete data set organized according to a predefined format (such as JSON) that contains the status of all key entities in the system at that instant (such as all drones, orders, and environmental information), ensuring the temporal consistency and data integrity of the decision-making process.

[0074] The city environment management module then provides the delivery data and transmits it back to the logistics management and scheduling module. Simultaneously, the delivery data is also transmitted in parallel to the visualization and command interaction module to update the interface.

[0075] Phase Two: Intelligent Decision Making and Decision Analysis, Command Issuance

[0076] After understanding the user's instructions and situation, the process enters the intelligent decision-making stage. The logistics management and scheduling module integrates these two elements to construct a structured LLM request data. This LLM request data is a text prompt specifically formatted for the large language model. This "text prompt" not only contains the task data to be processed but may also contain specific instructions, rules, and examples. Its purpose is to guide the LLM to perform reasoning in a specific direction and generate output in a specific format. This request is then sent to the large language model interface module.

[0077] The Large Language Model Interface module acts as a "cognitive bridge," initiating requests to and receiving responses from external LLM services. It then parses the unstructured text response returned by the LLM into LLM decision data. This LLM decision data consists of structured, executable policy instructions (e.g., specifying the allocation relationship between specific drones and tasks) within the system. This decision is then transmitted back to the logistics management and scheduling module.

[0078] Phase 3: Instruction Execution Simulation

[0079] This stage involves translating intelligent decisions into physical actions within a simulated world. Upon receiving the LLM decision, the logistics management and scheduling module translates it into a series of delivery instruction data. This delivery instruction data is a low-level, machine-executable set of commands (e.g., containing specific parameters such as takeoff, waypoints, and landing).

[0080] The delivery instruction is sent to the simulation engine module. Based on the instruction, this module accurately simulates the physical behavior and mission process of the drone, continuously generating new delivery event feedback data in the process. This delivery event feedback data is a discrete, structured message encapsulating specific events (such as "mission completed" or "encountering interference"), and is a key signal for achieving a closed loop.

[0081] Phase 4: Closed-loop feedback adjustment

[0082] This stage represents the closing and restart phase of the closed loop. Newly generated delivery event feedback data is distributed through two parallel paths:

[0083] Firstly, the flow decision is adjusted: delivery event feedback data is directly sent back to the logistics management and scheduling module. This data flow enables the module to trigger a new round of decision-making processes (returning to the second stage) in real time based on the execution results, thus forming a fast and adaptive decision-making closed loop.

[0084] Secondly, the flow of situation updates: Delivery event feedback data is synchronously sent to the urban environment management module for situation updates. In this embodiment of the invention, "situation update" refers to the process of modifying or overwriting the state data in the "digital twin" data center based on event feedback generated by the simulation engine, which represents real changes in the physical world. This data flow ensures that the system's "digital twin" state is always consistent with the latest situation, providing a reliable data foundation for all subsequent decisions, thus forming a fundamental closed loop that guarantees data integrity.

[0085] Through the cyclical data flow of the above four stages, the embodiments of the present invention fully realize closed-loop management of the entire lifecycle from environmental perception to decision-making, execution and event feedback.

[0086] The following sections will provide a detailed introduction to each module of the system:

[0087] The urban environment management module is responsible for storing and maintaining in real time environmental situation data, including all drones, all order tasks, and historical events, and formatting it as delivery data.

[0088] This module is configured as the core data layer in the system, responsible for maintaining and providing a real-time "digital twin" status of the entire city environment and drone delivery missions.

[0089] The environmental situation data is a highly structured dataset containing all drones (including ID, payload, battery level, and status), all drones currently in flight, all order tasks (including ID, destination, cargo information, priority, and time window), and all historical events occurring during the simulation. This module ensures the freshness and accuracy of task information by simulating the real-time injection of external environmental data (such as weather and temporary no-fly zones) and drone status data (such as battery consumption and payload changes). It also manages detailed drone characteristics (such as payload, range, and speed) and aggregates all available drone resources, formatting them into resource status strings that are easy for the LLM to understand, providing the LLM with the resource parameters needed for decision-making.

[0090] In this embodiment of the invention, the "resource status string" refers to a human-readable text specifically designed and formatted for Large Language Models (LLM). Its function is to summarize and condense the complex, scattered, and dynamically changing available resource data within the system into a concise and clear text summary. This formatted text is designed to cater to the text-based reasoning characteristics of LLM, and by precisely controlling its input information, it guides LLM to make more efficient and reasonable decisions, while also optimizing the cost and efficiency of API calls.

[0091] The logistics management and scheduling module is used to execute the current decision-making process, including:

[0092] The user instruction data is fused with the delivery data to construct structured LLM request data; a request is sent to an external LLM service through the large language model interface module, and an unstructured text response is received from the external LLM service; the unstructured text response is parsed into LLM decision data, the LLM decision is translated into a series of delivery instruction data, and sent to the simulation engine module; wherein a low-rank adaptation technique is used in advance to fine-tune the general large language model to obtain the external LLM service.

[0093] This module is configured as the "control center" in the system. It drives the entire decision-making and feedback process by coordinating and invoking the urban environmental management module, the large language model interface module, and the simulation engine module. Its specific functions are as follows:

[0094] It drives the simulation stepping, receives user instruction data, schedules the Large Language Model (LLM) interface module for decision-making, parses and executes the scheduling instructions generated by the LLM, and provides real-time feedback on key events generated by the simulation engine module (especially drone interference and task completion / failure) to the LLM for decision adjustment. Optionally, the logistics management scheduling module also ensures that the asynchronous execution of LLM requests does not block the main UI thread by running an LLM worker thread in an independent thread, and maintains a task tracking list. In this embodiment, the main UI thread refers to the only thread in the system responsible for handling user interface (UI) updates and responding to user operations. To avoid blocking the main UI thread (manifesting as interface lag or unresponsiveness) due to time-consuming network requests to external Large Language Model (LLM) services, such request tasks are handled by an "LLM Worker" running in an independent background thread. The logistics management scheduling module dispatches decision requests to this Worker, which independently performs network communication in the background and then notifies the main thread of the result for further processing. This design, which separates time-consuming operations from interface responses, ensures the smoothness and availability of the system. Meanwhile, the logistics management and scheduling module is used to maintain a task tracking list to manage all ongoing decision-making processes.

[0095] The logistics management and scheduling module implements automated decision-type triggering and management: the system provides real-time delivery data (such as order task status, drone location, and interference events) to the LLM (Logistics Management Provider), which then intelligently summarizes, judges, and identifies the decision type of the current task based on the knowledge learned during its training. In other words, after receiving data requests from the LLM, the external LLM service identifies the decision type of the current task, specifically including:

[0096] 1) When user command data is input for the first time or a request is made for system autonomous scheduling, the external LLM service is identified as the initial scheduling decision;

[0097] 2) When periodic monitoring shows that the task has not been completed and the drone has not reached its destination, the external LLM service identifies it as a continuous task allocation or refueling decision.

[0098] 3) When the drone is interfered with, the external LLM service identifies it as interference response decision and formulates a new scheduling strategy based on the interference situation.

[0099] Building upon this foundation, the logistics management and scheduling module transforms the system's internal structured environmental data and user intent into natural language prompts that the LLM can understand. These prompts are then sent to an external LLM service via network requests. The module also receives and performs preliminary parsing of the decision or feedback text returned by the LLM. Based on different decision task types (such as initial scheduling, ongoing task allocation, and interference response), this module dynamically constructs system and user prompts, precisely guiding the inference direction and output format of the external LLM service. For example, when handling interference response decisions, it injects detailed "UAV interference response rules" as prompts into the external LLM service, guiding the LLM to formulate targeted response strategies.

[0100] Furthermore, when the logistics management and scheduling module parses the unstructured text response into LLM decision data, it executes a preset intelligent resource allocation logic to address complex task requirements and multi-drone collaboration, specifically including:

[0101] Based on the delivery data, a global resource pre-check is performed. If the total available drone resources are insufficient, the allocation of subsequent order tasks is stopped and recorded as a global resource shortage event. If the total resources are sufficient, the global resource pre-check is passed and the optimal solution attempt stage is entered.

[0102] In the optimal solution attempt phase, all available drones are traversed to try to find a single drone that can independently and completely meet the requirements of the current order task. If successful, the current order task is assigned to that single drone; otherwise, the collaborative solution attempt phase is entered.

[0103] In the collaborative solution attempt phase, the current order task requirements are decomposed, and an attempt is made to search for a combination of two or more available drones to collaboratively meet the current order task requirements. If successful, the current order task is assigned to a combination of two or more available drones; otherwise, the assignment of the next order task is initiated.

[0104] After all the above allocation attempts, the final stage begins. The successfully allocated order tasks are translated into the delivery instruction data, and for all order tasks that ultimately fail to be allocated, their status is updated to "insufficient resources, waiting for allocation", and recorded as allocation failure events and their reasons for failure.

[0105] The visualization and command interaction module displays warning messages about global resource shortage events and / or allocation failure events, along with the reasons for their failure.

[0106] The simulation engine module is used to execute delivery instructions, simulate the physical behavior and task execution process of the drone, and trigger corresponding events. The events are fed back to the city environment management module to update the environmental situation data, and the events are fed back to the logistics management and scheduling module to trigger a new round of decision-making process.

[0107] This module is configured as the "physics engine" in the system, responsible for accurately simulating the physical behavior and mission execution process of the drone and triggering corresponding events. It is the core of the evolution of drone flight physics and mission behavior, acting as the "physics engine" of the entire system. It accurately simulates the drone's takeoff, flight, battery consumption, payload transport, judgment of sudden interference (such as signal interference and communication interruption), order completion judgment, delivery time calculation, and dynamic updates of the drone's battery level and payload status. The drone's flight path is planned based on the start and end points, and the flight time is calculated based on flight speed and distance.

[0108] In an optional implementation, the present invention defines the events as including interference events, task completion events, task incomplete events, and task failure events.

[0109] (1) Interference events

[0110] Specifically, the simulation engine module is used to calculate the probability of interference to the drone during the simulated drone mission execution, based on the distance between the drone's current position and the interference source. The interference source refers to a physical or electronic signal source that interferes with the individual drone (e.g., a high-power radio tower, a complex electromagnetic environment near a communication base station, or an illegal drone jammer). The interference probability is represented by an attenuation function.

[0111]

[0112] Among them, P interference (d interf ) represents the probability of interfering with an individual drone; d interf P represents the distance between the current location of the drone and the interference source. max d represents the maximum probability of interference that the drone can achieve near the interference source. max β represents the maximum effective range of the interference source, indicating that the drone will not be interfered with beyond this distance; β is the attenuation factor of the interference probability, used to adjust the rate at which the interference intensity decreases with distance.

[0113] When d interf <d max At that time, the drone is considered to be within the interference range and the corresponding interference probability P is calculated. interference (d interf Based on this probability value, random sampling is performed, including:

[0114] Generate a pseudo-random number (Random Number) that is uniformly distributed in the interval [0,1]. If the Random Number is greater than P... interference (d interfIf the condition is met, the drone is determined to be uninterrupted; otherwise, the drone is determined to be interfered with, the drone's flight status is set to an interfered state (such as communication interruption or signal loss), and an interference event corresponding to the interference source is triggered.

[0115] (2) Task completion event, task incomplete event, and task failure event

[0116] Specifically, the simulation engine module is used to obtain the corrected actual delivery time based on the drone's flight distance, average speed, and environmental factors, expressed as follows:

[0117]

[0118] EnvironmentalFactor=C1·WindSpeed+C2·LoadRatio+C3·TerrainComplexity

[0119] Wherein T(d) flight ) indicates the actual delivery time of the order task; d flight is the actual flight distance of the drone; v is the average flight speed of the drone; EnvironmentalFactor is the comprehensive correction factor for the time consumption caused by environmental factors (such as weather, wind speed, and load). When EnvironmentalFactor is positive, it indicates that the time consumption increases; when it is negative, it indicates that the time consumption decreases; when it is zero, it indicates that there is no correction; C1, C2, and C3 are the weight coefficients of the corresponding items, WindSpeed ​​is the current wind speed, LoadRatio is the ratio of the drone's current load to its maximum load, and TerrainComplexity is the terrain complexity of the flight path (such as mountains, densely built-up urban areas, etc.).

[0120] It should be noted that the environmental factors here (as opposed to the aforementioned sources of interference) are usually predictable constraints that affect the overall operating environment and are primarily used to correct planning and cost calculations, rather than to trigger sudden anomalies.

[0121] Based on whether the order task was interrupted, and the corrected actual delivery time T(d) flight For the delivery time window of the order task, the following judgment process is performed:

[0122] If the drone was not interfered with and T(d) flight If the order is within or before the delivery time window, the order task status will be updated to completed, and a task completion event containing success information will be triggered.

[0123] If the drone was not interfered with and T(d) flightAfter the delivery time window, the order task status will be updated to timed out, and a task incomplete event will be triggered.

[0124] If the drone has been interfered with, causing the mission to be aborted or other simulation conditions to fail, the order mission status will be updated to failed, and a mission failure event containing the reason for the failure will be triggered.

[0125] The large language model interface module acts as a "cognitive bridge" between the system and external large language model services, responsible for semantic-level interaction. It is called a "cognitive bridge" because this module connects the system's internal structured data world with the external cognitive reasoning world of LLM.

[0126] The visualization and command interaction module is configured as a graphical user interface (GUI) in the system. It mainly includes a simulation monitoring window and a command decision window.

[0127] The simulation monitoring window is used to display, in real time, at least the drone location, order status, and event timeline in the environmental situation data;

[0128] The command and decision window is used to receive user instruction data and send it to the logistics management and scheduling module, display the input prompts and responses of the external LLM service in real time, and display warning information in real time.

[0129] Furthermore, it is necessary to emphasize that the core of the embodiments of the present invention lies in the deep application of LLM, and its successful implementation depends on the following key technical mechanisms:

[0130] 1) Fine-tuning of the large language model

[0131] The general large language model (LLM) is pre-tuned using Low-Rank Adaptation (LoRA) technology to enable it to understand complex logistics scenario data, perform scheduling inference, and generate precise task instructions. This is the core foundation for the system's intelligence. Specifically, LoRA adds two smaller, trainable low-rank matrices as a bypass to the large weight matrix W0 of the pre-trained model (such as the Q, K, V projection matrices in the attention mechanism). and This approach achieves incremental updates, where rank r << min(d,k), i.e., W = W0 + BA. During fine-tuning, only A and B are trained. This method significantly reduces the number of trainable parameters from d×k in the original model to r×(d+k). While maintaining the integrity of the model's original knowledge, it significantly reduces training costs, improves training efficiency, and reduces GPU memory usage during deployment.

[0132] 2) Refined and intelligent allocation and visual management of drone resources

[0133] In terms of unit-level resource management and global aggregation, each drone maintains its independent payload and battery level information. The city environment management module is responsible for aggregating the available resources of all delivery drones and formatting them into "resource status" strings that the LLM can understand. This serves as a crucial input for LLM decision-making, ensuring that the LLM does not exceed actual resource limitations when planning tasks. Regarding intelligent allocation logic, the logistics management scheduling module implements intelligent resource allocation logic when parsing the LLM's scheduling decisions to handle complex task requirements and multi-drone collaboration. This intelligent allocation logic is a progressive process that clearly defines and handles "resource shortages" at different levels. Simultaneously, a visual interface displays resource status and allocation warnings in real time, ensuring optimal and transparent drone resource utilization efficiency.

[0134] In summary, compared with existing technologies, it has the following beneficial effects:

[0135] 1. Significantly Enhanced Environmental Understanding and Adaptability: This invention overcomes the reliance of traditional rule-based systems on preset scenarios and the limitations of mathematical models in precise modeling. Through LoRA-tuned LLM, the system can learn complex nonlinear environmental patterns and deep scheduling experience from large-scale, high-fidelity simulation data. This allows it to understand and effectively respond to unforeseen, dynamically changing urban environments (such as sudden weather events, temporary no-fly zones, and dynamic obstacles). When facing complex and unknown simulated scenarios, the system can automatically generate more reasonable and efficient response strategies than traditional systems. The flexibility and generalization ability of decision-making are significantly improved, effectively reducing the cost and time of manual maintenance and updating of scheduling rules, and enhancing the flexibility and efficiency of UAV operations.

[0136] 2. Significantly Improved Efficiency and Naturalness of Human-Computer Interaction: This invention enables intuitive and natural language interaction between the user and the system. Users do not need to learn complex command syntax or code; they can clearly express their task intentions directly using everyday language commands. The system can also provide feedback on decision logic and execution status in easily understandable natural language. Operational convenience is significantly improved, effectively reducing the user's cognitive burden and operational complexity, allowing them to focus more on macro-level scheduling strategies and risk assessment, rather than the specific details of scheduling instructions.

[0137] 3. Precisely Ensuring Optimal Order Fulfillment Rate and Operating Costs: The system provided in this embodiment of the invention, while meeting the core service requirement of maintaining a high order fulfillment rate, can accurately select and combine the optimal drone type, quantity, and route based on LLM intelligent reasoning and precise environmental and drone data. This achieves the goal of minimizing total operating costs, effectively avoiding resource waste that may result from traditional manual scheduling or simple rules. Under simulated order volume and resource conditions, the system can complete delivery tasks more efficiently, or optimize the delivery of more orders without additional resource consumption. Resource utilization efficiency is significantly improved, which significantly enhances the overall efficiency and market competitiveness of logistics operations.

[0138] 4. Significantly Enhanced Operational Robustness and Task Continuity: This invention implements a real-time, closed-loop environmental event feedback mechanism, particularly addressing unexpected situations such as interference encountered by drones en route (e.g., communication interruptions, signal jamming). The system can instantly detect interference events, and the LLM (Local Management Module) quickly generates supplementary scheduling strategies based on the "Drone Interference Response Rules." Comparative tests conducted under simulated interference environments effectively guarantee task completion rates. Compared to traditional systems that may experience decision interruptions or delays under interference, this invention can rapidly adjust scheduling, ensuring the continuity and success rate of delivery tasks, thereby enhancing the system's operational adaptability in complex urban environments.

[0139] 5. Significantly Reduced System Iteration and Maintenance Costs: This embodiment of the invention employs LoRA fine-tuning technology, resulting in a minimal number of model training parameters and extremely high training efficiency. When new service demands arise, drone models are upgraded, or the environment changes, the LLM can be quickly fine-tuned using incremental data, without retraining the entire large model or significantly modifying hard-coded rules. This drastically shortens the model update and iteration cycle, significantly reduces system maintenance costs, and enables the system to quickly adapt to the rapidly changing needs and technological advancements in the logistics industry, maintaining its advanced nature and operational competitiveness.

[0140] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0141] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. 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. Such 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. An intelligent logistics management system based on a large language model, characterized in that, It includes modules for urban environmental management, logistics management and scheduling, simulation engine, and large language model interface. The urban environment management module is responsible for storing and maintaining in real time environmental situation data, including all drones, all order tasks, and historical events, and formatting it as delivery data; The logistics management and scheduling module is used to execute the current decision-making process, including: The user instruction data is fused with the delivery data to construct structured LLM request data; a request is sent to an external LLM service through the large language model interface module, and an unstructured text response is received from the external LLM service; the unstructured text response is parsed into LLM decision data, the LLM decision is translated into a series of delivery instruction data, and then sent to the simulation engine module; wherein a low-rank adaptation technique is used in advance to fine-tune the general large language model to obtain the external LLM service; The simulation engine module is used to execute delivery instructions, simulate the physical behavior and task execution process of the drone, and trigger corresponding events. The events are fed back to the urban environment management module to update the environmental situation data, and the events are fed back to the logistics management and scheduling module to trigger a new round of decision-making process. The events include interference events, and the simulation engine module is used for During the simulated drone mission execution, the probability of the drone being interfered with is calculated based on the distance between the drone's current position and the interference source; where the interference source refers to a physical or electronic signal source that interferes with the individual drone, and the interference probability is represented by an attenuation function. Among them, P interference (d interf ) represents the probability of interfering with an individual drone; d interf P represents the distance between the current location of the drone and the interference source. max d represents the maximum probability of interference that the drone can achieve near the interference source. max β represents the maximum effective range of the interference source, indicating that the drone will not be interfered with beyond this distance; β is the attenuation factor of the interference probability, used to adjust the rate at which the interference intensity decreases with distance. When d interf <d max At that time, the drone is considered to be within the interference range and the corresponding interference probability P is calculated. interference (d interf Random sampling determination based on probability values ​​includes: Generate a pseudo-random number (Random Number) that is uniformly distributed in the interval [0,1]. If the Random Number is greater than P... interference (d interf If the condition is met, the drone is determined to be uninterrupted; otherwise, the drone is determined to be interfered with, the drone's flight status is set to an interfered state, and an interference event corresponding to the interference source is triggered.

2. The intelligent logistics management system as described in claim 1, characterized in that, It includes a visualization and command interaction module, which includes a simulation monitoring window and a command decision window; The simulation monitoring window is used to display, in real time, at least the drone location, order status, and event timeline in the environmental situation data; The command and decision window is used to receive user instruction data and send it to the logistics management and scheduling module, display the input prompts and responses of the external LLM service in real time, and display warning information in real time.

3. The intelligent logistics management system as described in claim 2, characterized in that, The logistics management and scheduling module is used for When parsing the unstructured text response into LLM decision data, a preset intelligent resource allocation logic is executed, including: Based on the delivery data, a global resource pre-check is performed. If the total available drone resources are insufficient, the allocation of subsequent order tasks is stopped and recorded as a global resource shortage event. If the total resources are sufficient, the global resource pre-check is passed and the optimal solution attempt stage is entered. In the optimal solution attempt phase, all available drones are traversed to try to find a single drone that can independently and completely meet the requirements of the current order task. If successful, the current order task is assigned to that single drone; otherwise, the collaborative solution attempt phase is entered. In the collaborative solution attempt phase, the current order task requirements are decomposed, and an attempt is made to search for a combination of multiple available drones to collaboratively meet the current order task requirements. If successful, the current order task is assigned to the combination of multiple available drones; otherwise, the assignment of the next order task is initiated. After all the above allocation attempts, the final stage begins. The successfully allocated order tasks are translated into the delivery instruction data, and for all order tasks that ultimately fail to be allocated, their status is updated to "insufficient resources, waiting for allocation", and recorded as allocation failure events and their reasons for failure. The visualization and command interaction module displays warning messages about global resource shortage events and / or allocation failure events, along with the reasons for their failure.

4. The intelligent logistics management system as described in claim 1, characterized in that, After receiving the LLM request data, the external LLM service identifies the decision type of the current task, including: When user command data is input for the first time or a request is made for autonomous system scheduling, the external LLM service identifies it as an initial scheduling decision; when periodic monitoring shows that the task is not completed and the drone has not reached its destination, the external LLM service identifies it as a continuous task allocation or refueling decision; when the drone is interfered with, the external LLM service identifies it as an interference response decision and formulates a new scheduling strategy based on the interference situation.

5. The intelligent logistics management system as described in claim 4, characterized in that, The logistics management and scheduling module is used for After parsing the unstructured text response into LLM decision data, corresponding system prompts and user prompts are constructed based on the decision type.

6. The intelligent logistics management system as described in claim 1, characterized in that, The events include task completion events, task incomplete events, and task failure events. The simulation engine module is used for... Based on the drone's flight distance, average speed, and environmental factors, the corrected actual delivery time is obtained and expressed as follows: EnvironmentalFactor=C1·WindSpeed+C2·LoadRatio+C3·TerrainComplexity Wherein T(d) flight ) indicates the actual delivery time of the order task; d flight is the actual flight distance of the drone; v is the average flight speed of the drone; EnvironmentalFactor is the comprehensive correction factor for the time consumption caused by environmental factors; C1, C2, and C3 are the weight coefficients of the corresponding items, WindSpeed ​​is the current wind speed, LoadRatio is the ratio of the drone's current load to its maximum load, and TerrainComplexity is the terrain complexity of the flight path. Based on whether the order task was interrupted, and the corrected actual delivery time T(d) flight For the delivery time window of the order task, the following judgment process is performed: If the drone was not interfered with and T(d) flight If the order is within or before the delivery time window, the order task status will be updated to completed, and a task completion event containing success information will be triggered. If the drone was not interfered with and T(d) flight After the delivery time window, the order task status will be updated to timed out, and a task incomplete event will be triggered. If the drone has been interfered with, causing the mission to be aborted or other simulation conditions to fail, the order mission status will be updated to failed, and a mission failure event containing the reason for the failure will be triggered.

7. The intelligent logistics management system according to any one of claims 1 to 6, characterized in that, The delivery data is a structured snapshot of the environmental situation data, and the structured snapshot refers to a complete data set containing the environmental situation data in a predefined format.

8. The intelligent logistics management system as described in any one of claims 1 to 6, characterized in that, The logistics management scheduling module is used to run LLM worker threads in a separate thread.

9. The intelligent logistics management system according to any one of claims 1 to 6, characterized in that, The logistics management and scheduling module is used to maintain a task tracking list to manage all ongoing decision-making processes.

Images