A Multi-Constraint Replanning System for Cold Chain Logistics Routes Based on a Large-Scale Language Model

Abstract

This application discloses a multi-constraint replanning system for cold chain logistics routes based on a large-scale language model. The system includes a cloud-based analysis layer and a vehicle-side execution layer. The cloud-based analysis layer uses a large-scale language model to extract textual elements and generate dictionary data, constructs a three-dimensional spatiotemporal state tensor, and generates a spatiotemporal road network resistance matrix. The vehicle-side execution layer reads sensor data to determine the vehicle's thermophysical parameters, updates the edge weights of the electronic map based on the resistance matrix, and executes a path optimization algorithm that includes expected travel time costs, expected refrigeration energy consumption costs, and chain break risk penalties, outputting control commands. This application enables proactive prediction of unexpected road conditions, ensuring the absolute executability of the replanned route at the physical level and effectively reducing the risk of cold chain breakage and cargo damage.

Description

Technical Field

[0001] This application relates to the field of navigation and route planning, specifically to a multi-constraint replanning system for cold chain logistics routes based on a large language model, which is applied in long-distance cold chain logistics transportation environments and integrates unstructured environmental data analysis with the underlying thermophysical state of vehicles. Background Technology

[0002] Long-distance cold chain logistics transportation places strict requirements on the temperature stability of the environment in which the goods are located. Existing vehicle navigation route planning systems are mostly based on the shortest distance, the shortest time, or structured traffic congestion data for route finding. In actual cold chain transportation scenarios, such general navigation solutions have obvious limitations.

[0003] Currently, existing navigation systems cannot directly process unstructured environmental information, resulting in significant response lag. Temporary traffic control notices issued by highway patrol officers or localized rainstorm and hail warnings issued by meteorological departments are typically in the form of natural language text or meteorological cloud images. Traditional systems rely on manual input or back-end manual translation, making it difficult to anticipate sudden road conditions.

[0004] On the other hand, existing path planning algorithms neglect the physical properties of the vehicles themselves. The energy consumption of the refrigeration unit and the ability to maintain temperature inside the refrigerated truck are directly affected by the external ambient temperature, the vehicle's wind speed, and the aging of the refrigeration unit itself. When a vehicle encounters long-distance traffic jams and is exposed to the scorching heat of asphalt roads in summer, the refrigeration efficiency at idle drops sharply, easily leading to excessive temperature inside the truck and damage to the cargo. Existing navigation path planning does not include the heat leakage rate of the truck compartment, the external temperature field, and the vehicle's dynamic speed as multivariate constraints in the calculation of node cost values. This planning logic, which is detached from the thermodynamic and physical constraints of refrigerated vehicles, easily guides vehicles to high-risk routes and makes it difficult to guarantee the safety of the underlying thermal state. Summary of the Invention

[0005] This application provides a multi-constraint replanning system for cold chain logistics routes based on a large-scale language model, which solves the technical problems of existing navigation systems being unable to parse unstructured emergency event text and route planning being divorced from vehicle thermodynamic and physical constraints.

[0006] On one hand, this application provides a multi-constraint replanning system for cold chain logistics routes based on a large-scale language model, including: a cloud analysis layer; and a vehicle-side execution layer, communicatively connected to the cloud analysis layer; the cloud analysis layer is deployed with a large-scale language model, configured to periodically acquire unstructured traffic and weather forecast texts; the large-scale language model extracts event elements from the unstructured traffic and weather forecast texts and generates structured dictionary data, maps the structured dictionary data to a preset basic road network structure to construct a three-dimensional spatiotemporal state tensor, generates a spatiotemporal road network resistance matrix based on the three-dimensional spatiotemporal state tensor, and sends it to the vehicle-side execution layer; the vehicle-side execution layer includes an onboard edge computing gateway and sensor devices; the onboard edge computing gateway reads the refrigeration compressor status data and vehicle operating status collected by the sensor devices, and based on the refrigeration... The compressor status data and the vehicle operating status determine the vehicle body thermophysical parameters. Based on the received spatiotemporal road network resistance matrix and the vehicle body thermophysical parameters, the edge weights of the preset electronic map are updated, and the path optimization algorithm is executed to output control commands. The path optimization algorithm includes a movement cost function, which includes expected travel time cost, expected cooling energy consumption cost, and temperature chain break risk penalty term. The expected cooling energy consumption cost is calculated by the heat conduction equation through the external forecast temperature, target temperature, convective heat transfer coefficient related to the vehicle's wind speed, and the average heat transfer coefficient of the compartment dynamically fitted based on the vehicle body thermophysical parameters. The temperature chain break risk penalty term is configured to assign a positive penalty value to the current road segment when the theoretical cooling power exceeds a preset threshold of the compressor's rated power, in order to control the path optimization algorithm to abandon the corresponding road segment.

[0007] On the other hand, this application also provides a multi-constraint replanning method for cold chain logistics routes based on a large-scale language model, applied to a system including a cloud analysis layer and a vehicle-side execution layer, comprising: periodically acquiring unstructured traffic and weather forecast texts; extracting event elements from the unstructured traffic and weather forecast texts and generating structured dictionary data; mapping the structured dictionary data to a preset basic road network structure to construct a three-dimensional spatiotemporal state tensor; generating a spatiotemporal road network resistance matrix based on the three-dimensional spatiotemporal state tensor and sending it to the vehicle-side execution layer; reading refrigeration compressor status data and vehicle operating status collected by sensors; determining vehicle body thermophysical parameters based on the refrigeration compressor status data and vehicle operating status; and according to the received data... The spatiotemporal road network resistance matrix and the vehicle body thermophysical parameters update the edge weights of the preset electronic map, and execute a path optimization algorithm including a movement cost function to output control commands. The movement cost function includes expected travel time cost, expected cooling energy consumption cost, and temperature chain break risk penalty term. Based on the external forecast temperature, target temperature, convective heat transfer coefficient associated with the vehicle's windward speed, and the average heat transfer coefficient of the compartment dynamically fitted based on the vehicle body thermophysical parameters, the expected cooling energy consumption cost is calculated through the heat conduction equation. When the theoretical cooling power exceeds a preset threshold of the compressor's rated power, a positive penalty value is assigned to the current road segment through the temperature chain break risk penalty term to control the path optimization algorithm to abandon the corresponding road segment.

[0008] In another aspect, this application also provides an electronic device, including a memory and a processor; the memory stores a computer program; when the computer program is executed, the above-mentioned method for multi-constraint replanning of cold chain logistics paths based on a large language model is implemented.

[0009] In another aspect, this application also provides a computer-readable storage medium having a computer program stored thereon; when the computer program is executed, it implements the above-mentioned method for multi-constraint replanning of cold chain logistics paths based on a large language model.

[0010] The technical solution provided in this application achieves structuring processing of unstructured environmental text through a large-scale language model in the cloud, enabling advance prediction of sudden road conditions. Simultaneously, real-time acquisition and fitting of vehicle-side underlying thermophysical parameters provide a microscopic physical basis for macroscopic environmental prediction. The logical prediction of the large-scale language model and the vehicle's thermophysical state are functionally interlocked, ensuring the absolute executability of replanning the route at the physical level. Furthermore, it maintains underlying safe driving by immediately incorporating vehicle status when the network recovers or communication is limited, effectively reducing the risk of cargo damage due to chain disruptions. Attached Figure Description

[0011] Figure 1This is a structural block diagram of a multi-constraint replanning system for cold chain logistics routes based on a large-scale language model, provided in an embodiment of the present invention.

[0012] Figure 2 This is a flowchart of a multi-constraint replanning method for cold chain logistics routes based on a large language model, provided in an embodiment of the present invention.

[0013] Explanation of reference numerals in the attached figures:

[0014] In the diagram: 100 - Cold chain logistics route multi-constraint replanning system, 101 - Cloud analysis layer, 102 - Vehicle-side execution layer, 103 - Navigation display terminal, 1011 - Large language model, 1012 - Web crawler module, 1013 - Clock synchronization mechanism, 1021 - Vehicle edge computing gateway, 1022 - Sensor devices, 1023 - Power supply and voltage regulation module, 1024 - Communication module. Detailed Implementation

[0015] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0016] Example 1

[0017] like Figure 1 As shown, this application embodiment provides a cold chain logistics route multi-constraint replanning system 100. The system is physically divided into a cloud analysis layer 101 and a vehicle-side execution layer 102. The cloud analysis layer 101 is deployed in a server cluster, carrying a large language model 1011 fine-tuned from cold chain logistics corpus data. The vehicle-side execution layer 102 is installed in the vehicle's cab and maintains a wireless communication connection with the cloud analysis layer 101. Furthermore, the cold chain logistics route multi-constraint replanning system 100 also includes a navigation display terminal 103 for interactive presentation, which transmits instructions to the vehicle-side execution layer 102. Through the collaborative architecture of the cloud and vehicle, the system effectively supports the separation of computing power and the coupling of macro and micro data.

[0018] The cloud analytics layer 101 also includes a built-in web crawler module 1012 and a clock synchronization mechanism 1013. Based on the time reference of the clock synchronization mechanism 1013, the web crawler module 1012 is configured to periodically crawl traffic broadcast text, official road administration announcements, and gridded weather forecast data via an internet interface to obtain unstructured traffic and weather forecast text. Specifically, this unstructured traffic and weather forecast text includes traffic control announcements and local weather warnings. The large-scale language model 1011 receives this unstructured text stream, extracts event elements from the text, and generates structured dictionary data.

[0019] Specifically, the input-output logic for the large-scale language model 1011 is rigorously defined. Input data is encapsulated as a sequence containing prompt word templates. The prompt words are: extracting traffic disruption events and extreme weather events from the following text. The output format is a key-value pair format: [starting coordinates, ending coordinates, disruption start time, estimated recovery time, event type, impact score (0-1)]. After forward propagation, the large-scale language model 1011 outputs structured dictionary data conforming to the above key-value pair format, where the extracted event elements explicitly include event type, occurrence latitude and longitude coordinates, and impact time window.

[0020] The cloud-based analytics layer 101 then maps the generated structured dictionary data to a topology map of a pre-defined basic road network structure to construct a three-dimensional spatiotemporal state tensor. In a preferred embodiment, this tensor is expressed using mathematical notation as M. x,y,t.

[0021] Wherein, variable x represents the longitude dimension of the tensor; variable y represents the latitude dimension of the tensor; and variable t represents the time series dimension of the tensor, used to represent a specific future time window. This three-dimensional spatiotemporal state tensor M... x,y,t Each element in the equation represents the basic drag coefficient for a specific latitude and longitude coordinate segment within a future time window t.

[0022] After constructing the tensor, the cloud analysis layer 101 discretizes the impact time window into a time series containing multiple time steps. For each road segment in the basic road network structure, the cloud analysis layer 101 calculates the road segment's resistance coefficient based on the spatial geometric overlap between the event's latitude and longitude coordinates and the current road segment, as well as the overlap between each time step and the event's impact time window. The spatiotemporal road network resistance matrix is ​​constructed by multiplying the basic traffic resistance by the road segment's resistance coefficient. In one implementation, the basic resistance coefficient is represented by the following mathematical symbol: R base.

[0023] Among them, variable R base This represents the basic resistance coefficient of the physical road segment under conditions unaffected by unforeseen events. After generating the complete spatiotemporal road network resistance matrix, the cloud analysis layer 101 serializes and compresses it before sending it to the vehicle-side execution layer 102 via a wireless network. By using a large language model to parse and generate tensor data with time-series dimensions, the attenuation degree of the spatial and temporal impact of external unforeseen events on the underlying road network can be effectively quantified.

[0024] The vehicle-side execution layer 102 includes an on-board edge computing gateway 1021, sensor components 1022, a power supply and voltage regulation module 1023, and a communication module 1024. The on-board edge computing gateway 1021 is provided with a stable operating voltage by the power supply and voltage regulation module 1023; the communication module 1024 preferably employs a mobile network communication module for receiving data from the cloud. The sensor components 1022 include an on-board diagnostic interface connected to the vehicle controller area network bus, and an array of thermocouple temperature sensors distributed inside and outside the refrigerated compartment.

[0025] The vehicle-mounted edge computing gateway 1021 reads the refrigeration compressor status data and vehicle operating status collected by the sensor device 1022 through the controller local area network bus interface. Specifically, the sensor device 1022 is set to collect refrigeration compressor speed, fuel consumption, vehicle speed, and synchronously received interior and exterior temperatures of the vehicle compartment at a preset high-frequency sampling rate. In this embodiment, the high-frequency sampling rate is configured to 10 Hz to ensure that the underlying physical signal sampling is not distorted. The vehicle-mounted edge computing gateway 1021 determines the vehicle body thermophysical parameters based on these real-time data. In this process, the system fits the refrigeration mechanical efficiency based on the refrigeration compressor speed and fuel consumption over a historical operating period, and dynamically calculates the average heat transfer coefficient of the vehicle compartment by combining the time change rate of the interior and exterior temperatures of the vehicle compartment. It should be noted that the above-mentioned 10 Hz is only a preferred value for high-frequency sampling, and those skilled in the art can also implement this application using high-frequency sampling settings.

[0026] To enable microscopic physical calculations, the vehicle-mounted edge computing gateway 1021 is also equipped with a pre-installed thermodynamic state observer. This observer obtains the corresponding wind speed based on the current vehicle speed and determines the convective heat transfer coefficient by combining it with pre-stored surface roughness parameters of the vehicle body. Subsequently, the thermodynamic state observer inputs the convective heat transfer coefficient, the refrigeration mechanical efficiency, and the current temperature difference between the inside and outside of the vehicle body into a preset energy conservation balance equation, calculates the estimated real-time heat transfer coefficient, and uses this estimate as the current average heat transfer coefficient of the vehicle body.

[0027] The vehicle-mounted edge computing gateway 1021 maintains a directed graph structured electronic map in its local memory. Based on the received spatiotemporal road network resistance matrix and the real-time vehicle thermophysical parameters, the vehicle-mounted edge computing gateway 1021 updates the edge weights of the electronic map. After the update is complete, the system runs a heuristic rule-based node expansion algorithm locally to perform a path optimization algorithm, and outputs control commands to the navigation display terminal 103 to present the route based on the human-machine interaction command transmission protocol.

[0028] The system's innovation focuses on the movement cost function between nodes. This movement cost function includes the expected travel time cost, the expected cooling energy consumption cost, and a penalty for temperature chain disruption risk. Specifically, the actual movement cost from the current node i to the adjacent node j is calculated using the following formula:

[0029]

[0030] Wherein, variable G(i,j) represents the total movement cost from node i to node j; variable T travel Represents the expected travel time cost; variable E cooling Represents the expected energy cost of cooling; variable P risk This represents the penalty term for the risk of chain breakage due to temperature. Variables w1, w2, and w3 represent the proportional weight parameters of the corresponding cost components.

[0031] Within this formula framework, the expected travel time cost T travel The expected speed is calculated by dividing the base distance of a road segment by the expected speed of the corresponding road segment in the spatiotemporal road network resistance matrix, and is further adjusted in real time based on speed limit information or congestion reduction factors contained in the structured dictionary data. The expected cooling energy consumption cost E... cooling The energy consumption is calculated using the heat conduction equation fitted in real time by the vehicle-mounted edge computing gateway 1021. This equation is determined by a combination of the external forecast temperature, the target temperature, the convective heat transfer coefficient related to the vehicle's windward speed, and the average heat transfer coefficient of the vehicle body based on a dynamic fit of the vehicle's thermophysical parameters. Specifically, this energy consumption calculation is achieved through the following integral formula:

[0032]

[0033] Among them, variable t i and t j The variable represents the upper limit of the integral of the start and end times of the journey on the road segment; variable K is the average heat transfer coefficient of the compartment dynamically fitted by the system based on the historical data of the previous 30 minutes of operation, which characterizes the actual attenuation of the current insulation material of the compartment; variable A is the surface area constant of the compartment; variable T ext (t) represents the external forecast temperature as it changes over time; variable T int\target The set target refrigeration temperature; variables The coefficient represents the convective heat transfer coefficient, and this coefficient is positively correlated with the vehicle's frontal velocity variable v(t); This represents the current refrigeration mechanical efficiency of the refrigeration compressor. By introducing this formula for calculation, the actual energy dissipation physical process of a refrigerated truck under different external temperatures and different vehicle speeds at idle can be accurately reflected.

[0034] To prevent cargo damage caused by refrigeration system failure, the system also uses a temperature chain breakage risk penalty item P. riskA veto mechanism is implemented. This mechanism is configured such that when the theoretical cooling power exceeds a preset threshold of the compressor's rated power, a positive penalty value is assigned to the current road segment. Specifically, this preset threshold is configured to be 85% of the compressor's maximum rated power. When the instantaneous power corresponding to the theoretical cooling energy consumption cost calculated at the node exceeds this 85% threshold, the system assigns a penalty value to P. risk A very large positive penalty threshold (e.g., 1,000,000) is assigned. When the path optimization algorithm expands on this node, the value of the movement cost function corresponding to the current road segment is updated to infinity, thereby controlling the algorithm to abandon the corresponding road segment and preventing this high-risk node from being added to the search open list. Through the above pruning mechanism, the risk of uncontrolled temperature runaway in the carriage caused by overload failure of the refrigeration unit can be forcibly avoided.

[0035] In summary, the system provided in this application provides a microscopic physical basis for cloud-based macro-environment prediction by dynamically fitting the heat leakage rate of the vehicle body and reading the energy efficiency of the refrigeration unit through an on-board edge computing gateway. The logical prediction of the large-scale language model in the cloud and the thermophysical state of the vehicle are functionally interlocked. When network communication is restricted, the vehicle can maintain local safe route planning based on the underlying thermophysical feedback, thereby ensuring the absolute executability of the replanned path at the physical level and reducing cold chain disruptions.

[0036] Example 2

[0037] like Figure 2 As shown in the illustration, this application also provides a multi-constraint replanning method for cold chain logistics routes based on a large-scale language model. The hardware system relied upon by this method is consistent with the system in Embodiment 1, and it is applied to a system environment including a cloud analysis layer and a vehicle-side execution layer. The method mainly includes the following processing steps:

[0038] Execute S201 to acquire unstructured traffic and weather forecast texts at regular intervals.

[0039] S202 involves extracting event elements and generating structured dictionary data from the unstructured traffic and weather forecast text.

[0040] The system uses the mapping data to construct the three-dimensional spatiotemporal state tensor S203. The system maps the structured dictionary data to a preset basic road network structure to construct the three-dimensional spatiotemporal state tensor.

[0041] The system generates a spatiotemporal road network resistance matrix and sends it to the vehicle-side S204. Based on the three-dimensional spatiotemporal state tensor, the system generates a spatiotemporal road network resistance matrix and sends it to the vehicle-side execution layer.

[0042] Execute S205 to read the refrigeration compressor status data and vehicle operating status, which is achieved by reading the refrigeration compressor status data and vehicle operating status collected by the sensor devices through the vehicle bus interface of the vehicle-side execution layer.

[0043] S206 is executed to determine the vehicle body thermophysical parameters based on the refrigeration compressor status data and the vehicle operating status.

[0044] S207 updates the edge weights of the electronic map based on the resistance matrix and physical parameters, and updates the preset edge weights of the electronic map based on the received spatiotemporal road network resistance matrix and the vehicle body thermal physical parameters.

[0045] Execute the path optimization algorithm S208, which includes a movement cost function. The algorithm is started and executed based on the updated weights. The movement cost function includes the expected travel time cost, the expected cooling energy consumption cost, and a temperature chain break risk penalty.

[0046] The system executes step S209, which calculates the travel cost using the heat conduction equation. Based on the external forecast temperature, target temperature, convective heat transfer coefficient related to the vehicle's windward speed, and the average heat transfer coefficient of the vehicle body dynamically fitted based on the vehicle's thermophysical parameters, the system calculates the expected cooling energy consumption cost using the heat conduction equation. This results in the expected travel time cost for the corresponding road segment.

[0047] Specifically, regarding the heat conduction calculation details of the expected cooling energy consumption cost, the system obtains the temperature difference between the external forecast temperature and the target temperature at the time of arrival at the road segment; the temperature difference is multiplied by the average heat transfer coefficient and surface area of ​​the compartment to determine the basic heat leakage rate. Convective heat flux density compensation calculation is performed on the basic heat leakage rate based on the convective heat transfer coefficient; finally, the compensated heat leakage rate is integrated over the expected travel time of the road segment, and combined with the refrigeration mechanical efficiency determined by the vehicle-side execution layer, the final expected cooling energy consumption cost is obtained.

[0048] In the convective heat flux density compensation calculation, the system extracts the road segment meteorological wind speed vector contained in the spatiotemporal road network resistance matrix, and performs vector synthesis of the expected driving speed vector and the road segment meteorological wind speed vector to obtain the relative windward speed vector; the dynamic convective heat transfer coefficient is calculated based on the modulus of the relative windward speed vector as the convective heat transfer coefficient, and the dynamic convective heat transfer coefficient is multiplied by the windward area of ​​the vehicle body and the temperature difference to obtain the convective additional heat leakage rate; finally, the convective additional heat leakage rate is superimposed on the base heat leakage rate. Through the above vector synthesis and compensation operations, the heat flux calculation deviation under strong crosswind or headwind conditions in open areas can be effectively eliminated.

[0049] The chain break risk interception operation is executed, and control command S210 is output. When the theoretical cooling power exceeds a preset threshold of the compressor's rated power, a positive penalty value is assigned to the current road segment through the temperature chain break risk penalty term. This is used to control the path optimization algorithm to directly prune and abandon the corresponding road segment, and output a safe control command for navigation.

[0050] Example 3

[0051] This application also provides an electronic device. This electronic device can be implemented as a server entity or an in-vehicle computing unit entity with computing and processing capabilities.

[0052] The electronic device in this embodiment includes a memory and a processor; wherein a computer program is stored on the memory. The memory can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory, electrically erasable programmable read-only memory, erasable programmable read-only memory, programmable read-only memory, read-only memory, magnetic memory, flash memory, magnetic disk, or optical disk.

[0053] When the computer program is executed, it is configured to implement all the steps of the multi-constraint replanning method for cold chain logistics routes based on a large language model as described in Embodiment 2 above. The processor can be a general-purpose processor, digital signal processor, application-specific integrated circuit, field-programmable gate array (FPGA), or other programmable logic device, discrete gate or transistor logic device, or discrete hardware component. By supporting the execution of the above algorithm through this hardware, efficient scheduling of computing resources can be achieved, resulting in the beneficial effect of stable route planning by the system.

[0054] This application also provides a computer-readable storage medium on which a computer program is stored. When executed, the computer program is configured to implement all the steps of the multi-constraint replanning method for cold chain logistics routes based on a large language model as described in Embodiment 2 above. This storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof.

[0055] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A multi-constraint replanning system for cold chain logistics routes based on a large-scale language model, characterized in that, include: Cloud analytics layer; The vehicle-side execution layer is communicatively connected to the cloud-based analysis layer. The cloud-based analytics layer is equipped with a large language model configured to periodically acquire unstructured traffic and weather forecast texts. The large language model extracts event elements from the unstructured traffic and weather forecast texts and generates structured dictionary data. The structured dictionary data is mapped to a preset basic road network structure to construct a three-dimensional spatiotemporal state tensor. Based on the three-dimensional spatiotemporal state tensor, a spatiotemporal road network resistance matrix is ​​generated and sent to the vehicle-side execution layer. The vehicle-side execution layer includes an in-vehicle edge computing gateway and sensor devices; The vehicle edge computing gateway reads the refrigeration compressor status data and vehicle operating status collected by the sensors, determines the vehicle body thermophysical parameters based on the refrigeration compressor status data and vehicle operating status, updates the edge weights of the preset electronic map according to the received spatiotemporal road network resistance matrix and the vehicle body thermophysical parameters, and executes the path optimization algorithm to output control commands. The path optimization algorithm includes a movement cost function, which includes expected travel time cost, expected cooling energy consumption cost, and temperature chain break risk penalty term; The expected cooling energy consumption cost is calculated by the heat conduction equation based on the external forecast temperature, target temperature, convective heat transfer coefficient related to the vehicle's wind speed, and the average heat transfer coefficient of the compartment based on the dynamic fitting of the vehicle's thermophysical parameters. The temperature chain break risk penalty item is configured to assign a positive penalty value to the current road segment when the theoretical cooling power exceeds a preset threshold of the compressor's rated power, in order to control the path optimization algorithm to abandon the corresponding road segment.

2. The multi-constraint replanning system for cold chain logistics routes based on a large-scale language model as described in claim 1, characterized in that, The unstructured traffic and weather forecast text includes traffic control notices and local weather warnings; the event elements include event type, latitude and longitude coordinates of occurrence, and time window of impact; the dimensions of the three-dimensional spatiotemporal state tensor include longitude dimension, latitude dimension, and time series dimension.

3. The multi-constraint replanning system for cold chain logistics routes based on a large-scale language model as described in claim 2, characterized in that, The cloud-based analysis layer discretizes the impact time window into a time series containing multiple time steps. For each road segment in the basic road network structure, the cloud-based analysis layer calculates the road segment obstruction coefficient based on the spatial geometric overlap between the occurrence latitude and longitude coordinates and the current road segment, as well as the overlap between each time step and the impact time window. The spatiotemporal road network resistance matrix is ​​constructed by multiplying the basic traffic resistance by the road segment obstruction coefficient.

4. The multi-constraint replanning system for cold chain logistics routes based on a large-scale language model as described in claim 1, characterized in that, The sensor collects the refrigeration compressor speed, fuel consumption, vehicle speed, and interior and exterior temperatures of the compartment at a preset high-frequency sampling rate as the refrigeration compressor status data and the vehicle operating status; the high-frequency sampling rate is configured to 10 Hz; the vehicle edge computing gateway fits the refrigeration mechanical efficiency based on the refrigeration compressor speed and fuel consumption in the historical operating interval, and dynamically calculates the average heat transfer coefficient of the compartment by combining the time change rate of the interior and exterior temperatures of the compartment.

5. The multi-constraint replanning system for cold chain logistics routes based on a large-scale language model as described in claim 4, characterized in that, The vehicle-mounted edge computing gateway has a built-in thermodynamic state observer; the thermodynamic state observer obtains the corresponding wind speed based on the current vehicle speed, and determines the convective heat transfer coefficient by combining it with the pre-stored surface roughness parameters of the compartment; the thermodynamic state observer inputs the convective heat transfer coefficient, the refrigeration mechanical efficiency, and the current temperature difference between the inside and outside of the compartment into a preset energy conservation balance equation to solve for the estimated real-time heat transfer coefficient as the average heat transfer coefficient of the compartment.

6. A multi-constraint replanning method for cold chain logistics routes based on a large-scale language model, characterized in that, Systems applied to both cloud-based analytics and vehicle-side execution layers include: Timely acquisition of unstructured traffic and weather forecast texts; Extract event elements from the unstructured traffic and weather forecast texts and generate structured dictionary data; The structured dictionary data is mapped to a preset basic road network structure to construct a three-dimensional spatiotemporal state tensor; A spatiotemporal road network resistance matrix is ​​generated based on the three-dimensional spatiotemporal state tensor and sent to the vehicle-side execution layer; Read the status data of the refrigeration compressor and the vehicle operating status collected by the sensors; The vehicle body thermophysical parameters are determined based on the refrigeration compressor status data and the vehicle operating status. The edge weights of the preset electronic map are updated based on the received spatiotemporal road network resistance matrix and the vehicle body thermophysical parameters, and a path optimization algorithm including a movement cost function is executed to output control commands. The mobility cost function includes the expected travel time cost, the expected cooling energy consumption cost, and the temperature chain breakage risk penalty term. The expected cooling energy consumption cost is calculated using the heat conduction equation based on the external forecast temperature, target temperature, convective heat transfer coefficient associated with the vehicle's windward speed, and the average heat transfer coefficient of the compartment dynamically fitted based on the vehicle's thermophysical parameters. When the theoretical cooling power exceeds a preset threshold of the compressor's rated power, a positive penalty value is assigned to the current road segment through the temperature chain breakage risk penalty term to control the path optimization algorithm to abandon the corresponding road segment.

7. The multi-constraint replanning method for cold chain logistics routes based on a large-scale language model as described in claim 6, characterized in that, The calculation of the expected cooling energy consumption cost based on the external forecast temperature, target temperature, convective heat transfer coefficient related to the vehicle's windward speed, and the average heat transfer coefficient of the compartment dynamically fitted based on the vehicle's thermophysical parameters, using the heat conduction equation, includes: The temperature difference between the external forecast temperature and the target temperature at the time of arrival of the road segment is obtained; The temperature difference is multiplied by the average heat transfer coefficient of the compartment and the surface area of ​​the compartment to determine the basic heat leakage rate. The convective heat flux density compensation calculation is performed on the basic heat leakage rate based on the convective heat transfer coefficient. The compensated heat leakage rate is integrated over the expected travel time of the road segment, and the expected cooling energy consumption cost is calculated by combining it with the refrigeration mechanical efficiency determined by the vehicle-side execution layer.

8. The multi-constraint replanning method for cold chain logistics routes based on a large-scale language model as described in claim 7, characterized in that, The calculation of convective heat flux density compensation for the basic heat loss rate based on the convective heat transfer coefficient includes: Extract the meteorological wind speed vectors of road segments contained in the spatiotemporal road network resistance matrix; The expected driving speed vector is vector-synthesized with the meteorological wind speed vector of the road section to obtain the relative windward speed vector. The dynamic convective heat transfer coefficient is calculated based on the modulus of the relative windward velocity vector and is used as the convective heat transfer coefficient. The convective heat transfer coefficient is multiplied by the windward area of ​​the compartment and the temperature difference to obtain the convective additional heat leakage rate. The convective additional heat loss rate is superimposed on the base heat loss rate.

9. An electronic device, characterized in that, Including memory and processor; The memory stores computer programs; When the computer program is executed, the multi-constraint replanning method for cold chain logistics paths based on a large language model as described in claim 6 is implemented.

10. A computer-readable storage medium, characterized in that, It contains computer programs; When the computer program is executed, it implements the multi-constraint replanning method for cold chain logistics paths based on a large language model as described in claim 6.

Images