A transport vehicle scheduling method and system based on a large language model and a storage medium

Abstract

The application belongs to the field of transportation scheduling, and particularly discloses a transportation vehicle scheduling method and system based on a large language model and a storage medium. The method obtains order information through a large language model based on adjusted order texts; obtains vehicle information required by the orders, and packs the order information and the vehicle information required by the orders into transportation applications; sorts a plurality of current transportation applications, and forms transportation files according to vehicle conditions, with one transportation file corresponding to one vehicle; screens the transportation files to obtain transportation files meeting transportation conditions; combines the screened transportation files to obtain a plurality of transportation schemes, screens the transportation schemes according to a preset target to obtain K transportation schemes; and determines a final transportation scheme from the K transportation schemes through a large language model according to real-time texts. The application can adjust the transportation plan of the transportation vehicle in real time according to the change of the order information in the supply chain, and minimize the impact of the plan adjustment.

Description

Technical Field

[0001] This invention belongs to the field of transportation scheduling, and more specifically, relates to a transportation vehicle scheduling method, system, and storage medium based on a large language model. Background Technology

[0002] For modular integrated buildings, logistics and transportation management is a significant challenge. This includes ensuring timely and cost-effective delivery of prefabricated components and modules, optimizing routes to reduce transportation costs and environmental impact, and coordinating and effectively managing the transportation of large prefabricated modules. This supply chain involves complex interactions among all participants and stakeholders, including component suppliers, MIC factories, carriers, and construction companies. The inherent uncertainties and variability in building logistics necessitate real-time adjustments to logistics and transportation plans to enhance supply chain resilience.

[0003] Currently, many studies have explored the challenges of off-site construction logistics planning, including MIC (Manufacturing Container Registry). These studies primarily focus on individual supply chain stages, addressing issues such as vehicle routing and integrated truck planning. Some researchers have proposed optimization models that integrate different supply chain stages, such as combining production with logistics or logistics with installation. These integrated approaches aim to provide more comprehensive solutions but often overlook uncertainties and dynamic interactions, which are crucial in practical applications. To address these limitations, some researchers have adopted the use of AnyLogic for multi-agent systems to optimize MIC transportation logistics. This method simulates the interactions between agents involved in the MIC process to improve transportation efficiency and reduce costs. Most current methods are only used for planning before transportation logistics begins and cannot discuss and decide on the impact of unforeseen events on the plan during transportation. Even the AnyLogic-based multi-agent system optimization method relies heavily on heuristic rules for its negotiation mechanism, indicating insufficient intelligence in selecting the optimal or suitable machine based on various information about orders and transport vehicles. Large Language Models (LLMs) offer a feasible approach to enhance the capabilities of agents in multi-agent manufacturing systems. Summary of the Invention

[0004] In view of the above-mentioned defects or improvement needs of the existing technology, the present invention provides a transportation vehicle scheduling method, system and storage medium based on a large language model. Its purpose is to realize the real-time updating of the transportation vehicle scheduling scheme based on order information adjustment and reduce operating costs.

[0005] To achieve the above objectives, according to a first aspect of the present invention, a transportation vehicle scheduling method based on a large language model is proposed, comprising the following steps:

[0006] Based on the adjusted order text, order information is obtained through a large language model;

[0007] Obtain the required vehicle information based on the order information, and package the order information and the required vehicle information into a transportation application;

[0008] The current transport applications are sorted and transport documents are generated according to the vehicle information, with one transport document corresponding to one vehicle;

[0009] The transportation documents are filtered to obtain all transportation documents that meet the transportation conditions; the filtered transportation documents are combined to obtain multiple transportation plans; and the transportation plans are filtered according to the preset objectives to obtain K transportation plans.

[0010] Based on the textual instructions of real-time demand, the final transportation plan is determined from K transportation options using a large language model.

[0011] As a further optimization, before using the large language model, a target instruction set including task description text and corresponding vehicle transportation plan is pre-constructed; then the large language model is fine-tuned using the target instruction set, that is, the large language model is trained with the task description text as input and the vehicle transportation plan as output.

[0012] As a further preferred embodiment, the target instruction set is constructed in the following manner:

[0013] Collect task description text and corresponding vehicle transportation plan, format the collected data using natural language task description, and then clean the formatted data to obtain the target instruction set.

[0014] As a further optimization, the formatted data is cleaned, including deduplication, correction, and complication reduction, specifically as follows:

[0015] The cosine similarity algorithm is used to calculate the semantic similarity of task description texts within the target instruction set. Some instances with high repetition are deleted or semantically similar instances are merged into new instances. Then, errors in the data are detected and corrected through a syntax analysis model, manual sampling, and script verification. Finally, some simple tasks are rewritten into complex tasks.

[0016] As a further preferred embodiment, the transportation documents include vehicle number, vehicle capacity, vehicle cost, estimated travel time, and the start and end times of two adjacent orders for the target order.

[0017] As a further preferred embodiment, the transportation conditions include vehicle departure time constraints, vehicle estimated arrival time constraints, and latest end time constraints, wherein:

[0018] The vehicle's departure time is subject to the following constraints:

[0019] t ei <t O(i-1) <t li

[0020] The estimated arrival time constraint for the vehicle is:

[0021] t Ei <t O(i-1) +T i <t Li

[0022] The latest end time constraint is:

[0023] t O(i-1) +T i <t S(i+1)

[0024] Among them, t ei t li Let t be the earliest delivery time and the latest delivery time of the i-th order, respectively; Ei t Li Let t be the earliest entry time and the latest entry time of the i-th order, respectively; O(i-1) T represents the estimated end time of the (i-1)th order. i For the i-th order, t is the estimated time to take for the vehicle. S(i+1) This represents the start time of the (i+1)th order.

[0025] As a further optimization, the selected transportation documents are combined to obtain multiple transportation plans, and these plans are then filtered according to preset objectives to obtain K transportation plans, including:

[0026] Based on transportation costs, a preliminary screening of various transportation options obtained from combinations of transportation documents was conducted to identify L transportation options with lower transportation costs.

[0027] Then, the transportation time for each of the L transportation schemes is calculated, and the K transportation schemes with the shortest transportation time are further selected.

[0028] As a further optimization, based on transportation costs, a preliminary screening is conducted on various transportation options obtained from the combination of transportation documents, specifically as follows:

[0029] For the cargo volume, the selected transportation files are combined using the DP matrix, and the loss of each transportation plan is calculated. The top N transportation plans with the smallest loss values ​​are initially selected.

[0030] The values ​​in the DP matrix represent the loss, with the 0th row initialized to the maximum cost, i.e., taking the DP value. 0,j =P, the maximum cost P is defined as greater than the sum of the transportation costs of all vehicles; when l≥1, the value DP in the l-th row and j-th column is...l,j The minimum loss when the first l vehicles are included in the combination and the combination transports j tons of goods is calculated as follows:

[0031] DP l,j =min(DP) l-1,j-pl +C l -P×min(p l ,j),DP l-1,j )

[0032] Where, p l C represents the carrying capacity of the l-th vehicle. l This represents the transportation cost of the l-th vehicle.

[0033] According to a second aspect of the present invention, a system for implementing the above-described transportation vehicle scheduling method based on a large language model is provided, comprising a decision agent, an order agent, a scheduling agent, and a vehicle agent, wherein:

[0034] The decision agent is implemented based on a large language model and is used to generate order information based on the adjusted order text and send it to the order agent;

[0035] The order agent is used to obtain the required vehicle information based on the order information, and package the order information and the required vehicle information into a transportation application and send it to the dispatch agent.

[0036] The dispatch agent is used to sort transportation requests from order agents and generate transportation invitations to send to all vehicle agents;

[0037] The vehicle agent is used to generate a transportation document based on the vehicle information after receiving a transportation invitation and send it to the dispatch agent. One transportation document corresponds to one vehicle.

[0038] The scheduling agent is also used to filter transportation documents to obtain all transportation documents that meet the transportation conditions; combine the filtered transportation documents to obtain multiple transportation plans; and filter the transportation plans according to preset targets to obtain K transportation plans, which are then sent to the decision agent.

[0039] The decision agent is also used to determine the final transportation plan from K transportation plans based on the text instructions of real-time demand, thereby forming a decision and sending it to the dispatch agent and the order agent. The order agent completes its own information update, and the dispatch agent sends the decision to the vehicle agent for execution.

[0040] According to a third aspect of the present invention, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements a transportation vehicle scheduling method based on a large language model.

[0041] In summary, compared with the prior art, the above-described technical solutions conceived by this invention mainly possess the following technical advantages:

[0042] 1. Existing methods for managing transportation vehicle scheduling assume that the transportation demand for modules and components is fixed, which does not fully reflect the changes and uncertainties in the real world. This invention focuses on the need for real-time adjustments to transportation plans due to order delivery delays caused by variable module sizes and cost fluctuations in supplier and MIC factory projects. Based on a large language model, a vehicle scheduling method is designed. Throughout the transportation vehicle scheduling management process, the transportation plan can be adjusted in real time in response to changes in order information in the supply chain, minimizing the impact of plan adjustments; that is, transportation arrangements can be adjusted in real time by changing order information at any time to reduce the carrier's operating costs.

[0043] 2. This invention designs a transportation scheme screening method based on the DP matrix, which can obtain the optimal vehicle combination loss under different transport capacity and conditions for the same cargo carrying capacity. Thus, the top L transportation schemes with the lowest loss values ​​can be screened out. Furthermore, by combining the screening of the top K transportation schemes with the lowest time consumption, a transportation scheme with lower transportation cost and time consumption can be quickly obtained.

[0044] 3. This invention designs a multi-agent transportation vehicle scheduling system. By defining agents for each object in logistics scheduling, it facilitates the entire process from control to decision-making. Simultaneously, these agents also act as channels between the multi-agent scheduling system and the Large Language Model (LLM), significantly reducing the complexity of modifying other conditions while improving system performance. Attached Figure Description

[0045] Figure 1 This is a schematic diagram of the transportation vehicle scheduling process based on a large language model according to an embodiment of the present invention;

[0046] Figure 2 This is a physical implementation framework diagram of a transportation vehicle scheduling system based on a large language model, according to an embodiment of the present invention.

[0047] Figure 3 This is a flowchart of a transportation vehicle scheduling method based on a large language model, according to an embodiment of the present invention.

[0048] Figure 4 This is a schematic diagram of LLM text indication in an embodiment of the present invention. Detailed Implementation

[0049] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0050] This invention provides a transportation vehicle scheduling system based on a large language model, which determines transportation vehicles for completed order components or modules. The earliest and latest arrival times of the components / modules are requirements of the supplier and the MIC (Manufacturing Component Manufacturer), while the minimum cost and time of the transportation plan reflect the carrier's interests. Furthermore, it is stipulated that transportation can only be carried out in batches after the entire order is completed; components cannot be transported while the order is in production.

[0051] This invention, based on a multi-agent system, incorporates various entities participating in the MIC (Micro-Integrated Building) supply chain process. Each entity is simulated as an agent with specific behaviors and interactions. In this embodiment, for ease of explanation, there are assumed to be N suppliers and M MIC factories. The physical process begins with suppliers producing prefabricated components (such as beams, exterior walls, and laminates). These components are manufactured in several suppliers' factories, which are designed to handle large-scale production with high precision. After production, these components are transported by carriers to several MIC factories, where the prefabricated components are assembled into integrated building modules. The system primarily addresses the transportation phase from suppliers to MIC assembly plants, focusing on solving the problem of combining different transport capacity vehicles to achieve reasonable vehicle costs and transportation times when modifying existing transportation schedules, while meeting the delivery times of MIC factories and construction sites, in order to minimize the operating costs for carriers.

[0052] The transportation vehicle dispatching system based on a large language model includes decision agent, order agent, dispatch agent, and vehicle agent. The functions of each agent are as follows:

[0053] 1. Decision-making agency

[0054] This agent is an instance of a large language model, trained after receiving prompts. It possesses a certain level of human intelligence, capable of receiving and processing textual information, and outputting dialogue through text. Its inherent intelligence comes from the large language model, which is invoked through prompts. The decision-making function is to make decisions based on transportation plans received from the scheduler, select the most suitable transportation plan, and send it to the scheduler for execution. Specifically, it includes the following functions:

[0055] (1) Accept natural language input, convert it into order information data, send it to the order agent to update the order information, and trigger the system operation trigger;

[0056] (2) Receive the set of feasible transport vehicles from the dispatch agent, and judge the value of each transport plan according to the minimum vehicle idle time logic, and select the one with the highest value as the decision and send it to the dispatch agent.

[0057] 2. Order Agent

[0058] The parameters are listed below:

[0059] Component order information: Order number i, origin (supplier) number N, destination (MIC factory) number M, earliest delivery time t ei Latest delivery time t li Earliest order entry time t Ei The latest time for order entry t Li Transportation time T N Expected start time t SN Expected end time t ON Transported cargo volume M i And the corresponding transport vehicle number Q.

[0060] Order Information Summary: Order Number i, Quantity of Goods Shipped m i And the corresponding budget vehicle number Q.

[0061] Specifically, it includes the following functions:

[0062] (1) Receive the latest component order information sent by the decision agent and update the component order information list;

[0063] (2) Select the component orders that have changed, organize the packaging and transportation requirements, and send the transportation application to the dispatcher;

[0064] 3. Scheduling Agent

[0065] Specifically, it includes the following functions:

[0066] (1) Accept transportation applications, compile the transportation requirements, and send a transportation invitation containing transportation requirements information to the vehicle agent.

[0067] (2) Organize the transportation documents sent by the vehicle agents, filter them, form a feasible transportation vehicle set consisting of several transportation documents, and send this information set to the decision agent. After receiving the decision from the decision agent, generate a transportation work order and send the work order to all corresponding vehicle agents.

[0068] 4. Vehicle agency

[0069] It consists of all vehicles owned by the carrier, with each vehicle representing an individual agent.

[0070] Parameters: Vehicle number Q, vehicle capacity mQ Vehicle expenses (i.e., transportation costs) C Q Work order task list for vehicle number Q (work order number i, origin number N, destination number M, start time t) Si Estimated time T i Expected end time t Oi ).

[0071] Specifically, it includes the following functions:

[0072] (1) Based on the order information, clear the orders with the same order number in the vehicle work order task list and send the transportation documents to the dispatch agent.

[0073] (2) After accepting the dispatch agent transportation work order, update the work order task list.

[0074] (3) Based on the start point number, end point number, and its own location, the SUMO software can calculate in real time the estimated time T for the start point N and end point MD corresponding to its own work order task. i Expected end time t Oi .

[0075] This invention provides a transportation vehicle scheduling method based on a large language model, such as... Figure 3 As shown, the negotiation process for the agent is as follows: Figure 1 As shown, the specific steps include the following:

[0076] S1: A large-scale language model is pre-tuned, and the fine-tuned language model is used as a decision agent. This includes the following steps:

[0077] S11: Collect data, including task description text and corresponding vehicle transportation plans. The task description text specifically includes component order description text and corresponding component order information parameters, as well as a collection of information on different transportation vehicle plans.

[0078] Specifically, data can be crawled from the internet, and different vehicle transportation schemes can be obtained from papers related to transportation vehicle scheduling and management. The corresponding scheme description texts can be written or modified, and the user input, i.e., the task description text, can be paired with the expected output, i.e., the vehicle transportation scheme, for subsequent training.

[0079] S12: Process the collected data to form a target instruction set, specifically by formatting and cleaning the collected data.

[0080] Specifically, the data crawled from the web and the data filtered from the dataset are formatted using natural language task descriptions to construct fine-tuning instances and instruction sets, followed by data cleaning. Data cleaning involves multi-stage data deduplication, correction, and complication based on the existing fine-tuning data. This includes: first, using a cosine similarity algorithm to calculate the semantic similarity of the task description text within the instruction set, deleting instances with high repetition or merging semantically similar instances into a new fine-tuning instance; then, detecting and correcting errors in the code through a syntax analysis model, manual sampling, and script verification; finally, rewriting a small number of simple tasks into complex tasks to improve the architectural assembly domain knowledge understanding and spatial reasoning capabilities of the large-scale language model.

[0081] S13: Fine-tune the large language model using the target instruction set.

[0082] S2: Based on adjustments to order information by component suppliers and MIC factories, the decision agent sends order information to the order agent, which then sends a transportation request to the scheduling agent. This includes the following steps:

[0083] S21: The decision agent uses a fine-tuned large language model to transform the natural language about order adjustments into component order information (including order number i, earliest order delivery time t). ei Latest delivery time t li Earliest order entry time t Ei The latest time for order entry t Li Transported cargo volume M i The information to be changed includes the earliest delivery time, latest delivery time, earliest arrival time, latest arrival time, and volume of goods transported. After this changed information is sent to the order agent, the order agent updates the order information and generates a shipping request.

[0084] Specifically, this step receives information about changed component orders. The orders are placed in a queue, indicating that multiple orders can be managed simultaneously. The purpose of this step is to update the changed orders and sort them in the order list (prioritizing orders by delivery time) to facilitate component production and delivery.

[0085] S22: The order agent will index the received change order in its order information master table and obtain the original component order information of this order; then, it will integrate the change component order information and the required vehicle information into a package and send it to the dispatch agent as a transportation application.

[0086] Specifically, in this step, the order number of the changed order will be indexed in the master order information table to obtain the corresponding transport vehicle number Q, and then integrated with the changed order information to package the order's transport requirements (order number i, origin number N, destination number M, earliest delivery time t). ei Latest delivery time t li Earliest order entry time t Ei The latest time for order entry t Li Transported cargo volume M i ), and send it to the dispatch agent as a transportation request.

[0087] S3: After compiling the transportation application information, the dispatch agent sends transportation invitations to the vehicle agents. Based on the transportation invitations, the vehicle agents package the vehicle information and send the transportation documents to the dispatch agent. This includes the following steps:

[0088] S31: After receiving a transportation request from the order agent, the dispatch agent will process the transportation request in a first-come, first-served order, that is, package the information as a transportation invitation and send it to all vehicle agents.

[0089] Specifically, in this step, transport requests are placed in a queue, indicating that multiple transport requests can be managed simultaneously. The purpose of this step is to sort the received requests in a first-come, first-served order, and then send the order information from the transport requests as transport conditions to the vehicle agents, inviting each vehicle agent to send the transport documents.

[0090] S32: Upon receiving a transportation invitation, the vehicle agent will first check if there is a task in its work order task list that matches the order number of this transportation request. If they match, the task will be cancelled to free up work time. Then, the vehicle agent will determine whether the required vehicle type is met. If the conditions are met, the vehicle agent will respond to the transportation invitation by compiling and packaging the vehicle information into a transportation file (one transportation file corresponds to one vehicle) and sending it to the dispatch agent.

[0091] Specifically, in this step, transport invitations sent to the vehicle agent are placed in a queue, indicating that multiple transport invitations can be managed simultaneously. The agent sorts the received invitations in a first-come, first-served order. Next, it checks the order information in the sent transport invitations, clears the work order tasks in its own work order task list that have the same order number i, and restores its idle time for that period. Then, it integrates the vehicle information, including vehicle number Q, vehicle capacity m... Q Vehicle expenses C Q Expected time T i The start times t of the two adjacent orders of target order i O(i-1) Stop time t S(i+1)Package the documents as shipping documents and send them to the dispatch representative.

[0092] S4: The dispatch agent filters transportation documents according to preset screening criteria, forms a set of feasible transportation vehicles, and sends it to the decision agent. This includes the following steps:

[0093] S41: Based on the parameters of the corresponding vehicle contained in the transport document: vehicle number Q, estimated travel time T. i The start time t of the two adjacent orders of target order i O(i-1) Stop time t S(i+1) The transportation documents are initially screened. The screening criteria are as follows:

[0094] (1) Vehicle departure time:

[0095] t ei <t O(i-1) <t li

[0096] (2) Estimated arrival time of the vehicle:

[0097] t Ei <t O(i-1) +T i <t Li

[0098] (3) Latest end time:

[0099] t O(i-1) +T i <t S(i+1)

[0100] Only if all three conditions above are met can the transport vehicle pass the initial screening and form a candidate transport vehicle set X.

[0101] S42: In the set of candidate transport vehicles X, combine the transport documents, and perform preliminary screening of the obtained transport plans based on the transport cost to obtain L transport plans.

[0102] Specifically, transportation documents can be combined to obtain multiple transportation solutions that meet transportation needs, and then L transportation solutions with lower transportation costs can be selected from them.

[0103] Alternatively, preliminary screening can be performed based on the DP matrix. This invention is a dynamic programming algorithm derived from the traditional 0 / 1 knapsack problem through adjustments and modifications. Its core is the calculated DP matrix (i.e., Dynamic Programming). The algorithm's computational approach is as follows: Faced with a fixed quantity of goods and transport vehicles with different capacities, different vehicle combinations are considered as "states," represented by rows l and columns j of the matrix. l represents the first l vehicles in set X included in the combination range according to their permutation order, and j represents the situation where the vehicle combination carries j tons of goods (j can be greater than the quantity M of goods in order number i). i The corresponding element value in the matrix represents the loss in that state, indicating the incurred losses from using the vehicle combination. Starting with the state containing only the first vehicle, the algorithm calculates several possible states. The algorithm selects the state with the lowest loss as the next state. Through continuous recursion of these states, the optimal vehicle combination loss for the same cargo volume under different transport capacities and conditions can be obtained. This allows for the selection of the top L lowest-loss transport schemes.

[0104] Specifically, let the set of vehicles X contain n vehicles, Q be the vehicle number, and C be the vehicle number. Q This refers to the vehicle's expenses (i.e., transportation costs), m Q This refers to the vehicle's carrying capacity (in tons, and in integers), i.e.:

[0105] X = {(C1,m1),(C2,m2),…,(C i ,m i ),…,(C n ,m n )}

[0106] The target cargo volume is M i Tonnage, by combining vehicles in set X, the sum of the transport capacity corresponding to the combination can be obtained as ∑m. If M i For every ton of cargo remaining (>∑m), a loss of P is imposed, where P is a value greater than the sum of all vehicle expenses. This means that even with only 1 ton of cargo remaining, the loss is far greater than the total expenses incurred by all vehicles used. The loss is the sum of vehicle expenses and the penalty for the remaining cargo.

[0107] Initialize the DP matrix, with the 0th row initialized to the maximum cost. Let l represent the current state with l vehicles, and j represent the cargo capacity of the current vehicle combination (j can be greater than M). i ), DP l,j This represents the minimum loss incurred when the first l vehicles or the combined vehicles transport j tons of goods.

[0108]

[0109] DP 0,j =P

[0110] For the l-th vehicle, when l≥0:

[0111]

[0112] This can be understood as follows: for the l-th vehicle, there are two decisions: use it or not use it; the loss when using it is... The loss when not in use is DP l-1,j Take the minimum of the two.

[0113] Iterative calculation yields a carrying capacity of M. i At that time, calculate the loss value of each transportation plan; sort the plans by loss from smallest to largest and obtain the top L transportation plans.

[0114] S43: Based on the earliest and latest transport times of the vehicles in each of the L transport plans, the transport time T for each plan can be calculated. N Expected start time t SN Expected end time t ON By arranging the vehicles, we can obtain the top K transportation schemes with the shortest transportation time, forming a set of feasible transportation vehicles.

[0115] Once a feasible fleet of vehicles is assembled, it will be sent to the decision-making agent. If no transportation plan can be formed, a "cannot be dispatched" message will be sent to the decision-making agent.

[0116] S5: The decision-making agent assesses the value of each transportation option, formulates a decision, and sends it to the dispatch agent and order agent. The order agent updates its own information, and the dispatch agent sends the decision to the vehicle agent for execution. This includes the following steps:

[0117] S51: The decision agent makes the final decision by invoking the LLM through prompts. During the invocation process, text instructions are used to determine the LLM's solution selection, and the decision-making process influences the LLM's selection from multiple perspectives in real time through instruction constraint text, including the estimated time, losses, estimated start time, and estimated end time of the transportation vehicle solution, such as... Figure 4 As shown, the system uses prompt text to evaluate the value of different transportation options and selects the optimal option as the final decision.

[0118] Specifically, to fully utilize the functionality of LLM, Markdown format is used to describe LLM behavior from multiple perspectives, including agent goal prioritization, output information, and instruction constraints. Agent goal prioritization determines the direction of the agent's selection of transportation solutions; it is described in natural language for easy understanding by the LLM. Output information defines the information relevant to the agent's response. For example, to reduce costs, the decision-making agent can output only the estimated time and losses for each route segment. To better control time nodes, the estimated start and end times can be output. However, to fully utilize the agent's cognitive abilities, this invention employs the widely used "thinking chain," guiding the agent through a step-by-step decision-making process, using multiple instruction constraints to gradually narrow down the range of solutions and select the high-value solution that meets management requirements.

[0119] S52: After a decision is made, the decision will be sent to the order agent. The order agent that receives the decision will update its own order information master table in preparation for the information retrieval of the next event.

[0120] S53: After a decision is made, the decision will be sent to the dispatch agent, which will then send the decision to the vehicle agent with the corresponding vehicle number. The vehicle agent that receives the decision will add the transportation task to its own work order task list, thus completing the decision execution.

[0121] The physical implementation framework of the above scheduling method and system is as follows: Figure 2 As shown, it includes the physical layer, data layer, proxy negotiation layer, and engine layer, wherein:

[0122] 1. Physical Layer: The physical layer includes all transport vehicles dispatched by the carrier for this project. It primarily utilizes intelligent networking to monitor the real-time status of these vehicles and uses their data to determine when they will deliver the goods, i.e., calculating the time it takes for the vehicle to complete the current transport order. In an intelligent networked environment, the completion time of a vehicle's transport order is measured using the following methods:

[0123] GPS positioning: The GPS positioning system records the vehicle's location and speed in real time, and can accurately calculate the vehicle's delay based on the actual driving conditions on the road.

[0124] Vehicle-mounted sensors: By using vehicle-mounted sensors, data such as vehicle acceleration, speed, and steering angle can be monitored in real time to determine whether the vehicle is in a congested state and further calculate the vehicle delay.

[0125] Vehicle-to-everything (V2X) data: By utilizing V2X technology, information can be exchanged between vehicles, including road conditions and vehicle driving status, thereby enabling real-time calculation of vehicle delays.

[0126] 2. Data Layer: The data layer divides the data obtained from the physical layer into vehicle physical information and vehicle semantic information. Using this information, the "Urban Traffic Simulation" (SUMO) software is used to simulate transportation conditions. SUMO is an open-source, highly portable, microscopic, and continuous traffic simulation software package designed to handle large networks. The above method enables real-time calculation of the current work order's transportation status for each vehicle. Finally, the information is compiled to form vehicle agent information representing each vehicle, which then flows as vehicle transportation information throughout the multi-agent system.

[0127] 3. Agent Negotiation Layer; The agent negotiation layer mainly consists of order agents and scheduling agents, and is the core component of this system. Its function is to organize the information from the data layer to form an information set that the decision-making agent can identify and value-judge. In this layer, after the order agent obtains the changed component order information, it drives the scheduling agent to collect vehicle transportation information from various vehicle agents. With the changed order information and vehicle transportation information, the scheduling agent forms a set of feasible transportation vehicles from several transportation plans for the decision-making agent to evaluate and select.

[0128] 4. Engine Layer; The engine layer provides the LLM engine. LLM inference and training require large graphics processing units (GPUs). An application programming interface (API) is needed to establish communication between this system and the LLM deployed in a computer center enhanced with multiple GPUs, enabling the agents in the system to run more efficiently. However, this will inevitably impact agent performance. Here, Alibaba's Tongyi Qianwen Big Prediction Model can be used for fine-tuning. The first step is to select a task-related dataset and preprocess the data, including cleaning, word segmentation, and encoding. The second step is to set fine-tuning parameters, including hyperparameters such as learning rate, training epochs, and batch size, and setting other hyperparameters as needed, such as weight decay and gradient clipping. The third step is to train the reward model, which is used to evaluate the quality of text sequences. The training data typically consists of text sequences generated by multiple language models, which are either manually evaluated or scored using other large models. The fourth step is to fine-tune the large model using supervised data. The process involves loading the pre-trained model and weights; making necessary modifications to the model according to the task requirements, such as changing the output layer; selecting a suitable loss function and optimizer; and fine-tuning the training using the selected dataset, including forward propagation, loss calculation, backpropagation, and weight updates.

[0129] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A transportation vehicle scheduling method based on a large language model, characterized in that, Includes the following steps: Based on the adjusted order text, the order information is obtained through a large language model. Before using the large language model, a target instruction set including task description text and corresponding vehicle transportation plan is pre-constructed. Then, the large language model is fine-tuned through the target instruction set, that is, the large language model is trained with the task description text as input and the vehicle transportation plan as output. Obtain the required vehicle information based on the order information, and package the order information and the required vehicle information into a transportation application; The current number of transportation applications are sorted and transportation documents are generated according to the vehicle information. Each transportation document corresponds to one vehicle. The transportation document includes the vehicle number, vehicle capacity, vehicle transportation cost, estimated time, and the start and end times of two adjacent orders of the target order. Based on the parameters of the corresponding vehicles contained in the transportation documents, the transportation documents are filtered to obtain all transportation documents that simultaneously meet three transportation conditions: vehicle departure time constraint, vehicle estimated arrival time constraint, and latest end time constraint. The vehicle's departure time is subject to the following constraints: The estimated arrival time constraint for the vehicle is: The latest end time constraint is: in, , The first i The earliest and latest delivery times for each order; , The first i The earliest and latest entry times for each order; For the first i- The estimated end time of one order. For the first i The estimated time for this vehicle to be delivered for this order. For the first i+ Start time of one order; The selected transportation documents are combined to obtain multiple transportation plans, and these plans are then filtered according to preset objectives to obtain K transportation plans; including: Based on transportation costs, a preliminary screening is performed on various transportation options obtained from the combination of transportation documents to obtain L transportation options with lower transportation costs; then the transportation time of each of the L transportation options is calculated, and further screening is performed to obtain K transportation options with shorter transportation times. Based on the textual instructions of real-time demand, the final transportation plan is determined from K transportation options using a large language model.

2. The transportation vehicle scheduling method based on a large language model as described in claim 1, characterized in that, The target instruction set is constructed as follows: Collect task description text and corresponding vehicle transportation plan, format the collected data using natural language task description, and then clean the formatted data to obtain the target instruction set.

3. The transportation vehicle scheduling method based on a large language model as described in claim 2, characterized in that, The formatted data is cleaned, including deduplication, correction, and complication, specifically as follows: The cosine similarity algorithm is used to calculate the semantic similarity of task description texts within the target instruction set. Some instances with high repetition are deleted or semantically similar instances are merged into new instances. Then, errors in the data are detected and corrected through a syntax analysis model, manual sampling, and script verification. Finally, some simple tasks are rewritten into complex tasks.

4. The transportation vehicle scheduling method based on a large language model as described in claim 1, characterized in that, Based on transportation costs, a preliminary screening of various transportation options obtained from combining transportation documents is conducted, specifically as follows: For the cargo volume, the selected transportation files are combined using the DP matrix, and the loss of each transportation plan is calculated. The top N transportation plans with the smallest loss values ​​are initially selected. The values ​​in the DP matrix represent the loss, with the 0th row initialized to the maximum cost, i.e., taking... The greatest cost P The value is defined as being greater than the sum of the transportation costs of all vehicles; l When ≥1, the first l Line 1 j Column values Indicates the preceding l Vehicles included in the combination scope, the combined transportation of the vehicles j The minimum loss per ton of cargo is calculated using the following formula: in, , Indicates the first l The transportation cost of the vehicle.

5. A system for implementing the transportation vehicle scheduling method based on a large language model as described in any one of claims 1-4, characterized in that, This includes decision-making agents, order agents, dispatch agents, and vehicle agents, among which: The decision agent is implemented based on a large language model and is used to generate order information based on the adjusted order text and send it to the order agent; The order agent is used to obtain the required vehicle information based on the order information, and package the order information and the required vehicle information into a transportation application and send it to the dispatch agent. The dispatch agent is used to sort transportation requests from order agents and generate transportation invitations to send to all vehicle agents; The vehicle agent is used to generate a transportation document based on the vehicle information after receiving a transportation invitation and send it to the dispatch agent. One transportation document corresponds to one vehicle. The scheduling agent is also used to filter transportation documents to obtain all transportation documents that meet the transportation conditions; combine the filtered transportation documents to obtain multiple transportation plans; and filter the transportation plans according to preset targets to obtain K transportation plans, which are then sent to the decision agent. The decision agent is also used to determine the final transportation plan from K transportation plans based on the text instructions of real-time demand, thereby forming a decision and sending it to the dispatch agent and the order agent. The order agent completes its own information update, and the dispatch agent sends the decision to the vehicle agent for execution.

6. A 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 transportation vehicle scheduling method based on a large language model as described in any one of claims 1-4.

Images