An interactive emergency logistics distribution automatic optimization modeling method and system

Abstract

The application discloses an interactive emergency logistics distribution automatic optimization modeling method and system; the method converts natural language and business data demand into a mathematical model and executable code based on a sample library through a multi-agent module; if the execution and verification fail, layered elastic repair is performed on the code or model according to the feedback until a satisfactory solution is obtained; expert review feedback is received to update the knowledge base and large model strategy to make it fit the actual constraints; finally, the solution and solving logic are output with interactive interpretation, and the output mode is dynamically adapted according to the scene and user type, and the whole process is recorded to form a traceable file; the application can effectively inhibit modeling illusion, and improve the reliability, automation level and explainability of logistics optimization.

Description

Technical Field

[0001] This invention relates to the field of emergency logistics scheduling and intelligent decision-making technology, specifically to an interactive emergency logistics distribution automatic optimization modeling method and system. Background Technology

[0002] With the continuous iteration and development of Large Language Models (LLMs, or large models for short) technology, its application value in operations research optimization fields such as logistics scheduling and route planning is becoming increasingly prominent, and research on intelligent optimization modeling for emergency logistics distribution scenarios has also attracted much attention. Although traditional emergency logistics distribution modeling and optimization methods have gradually incorporated the concept of large models, there are still many unavoidable practical shortcomings.

[0003] The existing technologies suffer from the following main shortcomings: First, there is the problem of modeling illusion and symbolic bias. Existing methods generally suffer from distorted modeling logic and inaccurate mathematical symbolic representation. Even when building a multi-agent collaborative architecture, human expert experience is only used for post-event result verification, which not only prolongs the response time but also lacks clear and quantifiable human intervention judgment criteria, making it difficult to adapt to emergency delivery scenarios with strict spatiotemporal constraints and frequent dynamic changes. Second, there is the problem of lack of cross-domain knowledge accumulation and reuse. Previous studies have mostly treated various delivery scheduling problems as independent entities, ignoring the accumulation and mining of industry domain rules and historical tasks. When disaster environments change suddenly or delivery constraints are added, insufficient knowledge transfer capabilities make it difficult for models to adapt and reconstruct quickly. Finally, there is the problem of the separation between model building and result interpretation. Traditional methods separate model building and solution solving into independent steps, resulting in obvious semantic barriers for non-professionals. The solution process is regarded as a black box operation, making it impossible to conduct interactive attribution analysis during optimization and deduction, which seriously affects the credibility of emergency decision-making and the feasibility of multi-party collaborative optimization and adjustment.

[0004] To address the numerous shortcomings in current emergency logistics and distribution optimization modeling, the industry urgently needs to construct a new emergency logistics and distribution optimization modeling system that integrates human-machine collaboration, multi-agent linkage, and interactive traceability. This system should be able to achieve end-to-end automatic conversion from natural language requirements to standardized mathematical models and executable code, possess layered elastic repair capabilities based on execution feedback, continuously drive model optimization through expert feedback, and dynamically adapt output according to user scenarios. This will effectively suppress the illusion of large-scale modeling and comprehensively improve the adaptability, reliability, and interpretability of emergency distribution scheduling solutions in dynamic disaster environments. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide an interactive emergency logistics distribution automatic optimization modeling method and system.

[0006] To achieve the above objectives, the present invention provides the following technical solution: This application provides an interactive automatic optimization modeling method for emergency logistics distribution, including: It receives emergency logistics demand information containing natural language descriptions and business data, and through a multi-agent automated modeling module, converts the demand information into a standardized mathematical optimization model and executable code compatible with operations research solvers based on a pre-set sample knowledge base and large model reasoning. The executable code is executed to obtain the solution result, and the solution result is verified. When the verification fails or an error occurs, the executable code or standardized mathematical optimization model is repaired in a layered and flexible manner based on the execution feedback until a delivery plan that meets the preset conditions is obtained or the stopping condition is reached. The system receives feedback from domain experts on the model and delivery plan, and updates the sample knowledge base and the inference strategy of the large model based on the feedback to continuously align with actual logistics and delivery constraints. The delivery plan, solution logic, and optimization basis that meet the preset conditions are converted into interactive explanation results output. The output method and optimization logic are dynamically adapted according to the current emergency logistics stage and user type. At the same time, the entire process operation is recorded to form a traceable process archive.

[0007] Furthermore, the multi-agent automated modeling module includes: Mathematical modeling agent, used to construct the first mapping function Natural language task space Mapped to mathematical model space ; in, It represents the natural language task space, which includes unstructured text problem descriptions, structured tabular data, and operational parameter information from real business environments; The mathematical model space is composed of standardized optimization models that conform to the norm, including decision variables, objective functions, and constraints. Encode an intelligent agent to construct the second mapping function. The mathematical model space Mapped to executable code space ;in, This represents the executable code space, which consists of a modeling framework and solution code compatible with operations research solvers.

[0008] Furthermore, the sample knowledge base includes emergency logistics operations optimization problem cases, standard mathematical models corresponding to the problem cases, executable solution programs, expert revision opinions, typical error modes, and decision interpretation cases, which are used to provide historical modeling knowledge examples and reusable optimization experience for the automated modeling module.

[0009] Furthermore, the emergency logistics demand information includes unstructured text descriptions, structured tabular data, on-site operation parameters, resource constraint information, delivery task information, and user input information; The entities extracted from the demand information include warehouses, demand points, vehicles, materials, and road nodes, and the attributes include location, quantity, capacity, time, cost, and priority.

[0010] Furthermore, the step of updating the large model inference strategy based on review feedback includes: For different problem categories Construct an error knowledge base, the error knowledge base Represented as ; in, This represents the set of optimization problem categories; This indicates the number of typical errors summarized under this category; Indicates the first A common error pattern; Indicates an error The suggested corrections provided; The automated modeling module is based on the problem category. From the error knowledge base Correction prompts for searching in the middle This information is then injected into subsequent reasoning prompts, forming an error-aware reasoning mechanism.

[0011] Furthermore, the verification of the solution results includes feasibility verification, constraint satisfaction verification, and target performance evaluation. The modeling results are quantitatively evaluated through the performance evaluation module, and its scoring function is defined as follows: in, Indicates the number of verification indicators; Indicates the first Weight of class errors; For indicator functions; Indicates the first Verification conditions.

[0012] Furthermore, the layered resilient repair includes: Execution by mathematical model The generated executable code returns the first... The solution feedback for the next iteration is expressed as follows: in, This indicates the standard output information, including the solution results; It represents the standard error stream, including syntax errors, solver exceptions, and infeasibility report information; When the execution feedback indicates that the error is a code-level error, the execution code self-corrects. When the execution feedback indicates that the error is a model-level error, perform model reconstruction or remodeling. The stopping conditions include obtaining effective delivery optimization results, reaching the preset maximum number of repairs, or determining that the current task cannot continue to be automatically repaired.

[0013] Furthermore, the interactive interpretation results include query, response, and action suggestion generation, which are used to show users delivery plan details, constraint satisfaction, performance bottlenecks, sensitivity analysis results, counterfactual analysis results, and actionable decision suggestions.

[0014] Furthermore, the dynamic adaptation includes: Based on the initial response phase, stabilization and rescue phase, and post-disaster recovery phase of emergency logistics, adjust the modeling optimization objectives, constraint priorities, and result interpretation methods. Furthermore, it adjusts the interaction methods, explanation granularity, and intervention permissions based on the different user groups' operational capabilities, professional backgrounds, and risk preferences.

[0015] An interactive emergency logistics distribution automatic optimization modeling system includes: The multi-agent automated modeling unit receives emergency logistics demand information containing natural language descriptions and business data. Based on a pre-built sample knowledge base, it constructs a first mapping function through mathematical modeling agents. Generate a standardized mathematical optimization model and construct a second mapping function through an coded intelligent agent. Generate executable code compatible with the operations research solver; where, Representing the natural language task space. Represents the space of mathematical models. Indicates the executable code space; The verification and hierarchical repair unit is used to execute the executable code and verify the results. If the verification fails, it is based on the solution feedback. Perform tiered, flexible repair until a delivery plan that meets preset conditions is generated; among which, Indicates standard output information, Represents the standard error stream; An expert-driven optimization unit is used to receive review feedback from domain experts and construct an error knowledge base based on the review feedback. And using the scoring function The modeling results are quantitatively evaluated, and the inference strategy of the large model is updated based on the evaluation results and feedback. in, Indicates the number of typical errors. Indicates the error mode. This indicates a correction prompt. Indicates the number of verification indicators. Indicates the error weight. For indicator functions, Indicate the verification conditions; The interactive explanation and dynamic adaptation unit is used to convert the delivery plan, solution logic and optimization basis into interactive explanation results output, and dynamically adapt the output method and optimization logic according to the emergency scenario status and user object, and record the whole process information to form a traceable process archive.

[0016] Compared with the prior art, this application has the following beneficial effects: This invention proposes an interactive emergency logistics distribution automatic optimization modeling method and system. It receives and parses emergency logistics demand information through a multi-agent automated modeling module, and utilizes a first mapping function. Second mapping function This system achieves end-to-end automatic conversion from natural language requirements to standardized mathematical models and executable code, and significantly reduces the operational threshold for non-professionals by combining it with a pre-built sample knowledge base. Simultaneously, by implementing a hierarchical elastic repair mechanism based on solution feedback when verification fails, it can automatically handle defects at the code and model levels according to error type, effectively suppressing the illusion of large-scale modeling and ensuring that the model strictly conforms to the actual physical constraints of emergency logistics and delivery, avoiding the generation of mathematically feasible but impractical solutions, and significantly improving system robustness and solution reliability. Furthermore, by receiving expert review feedback to build an error knowledge base, expert knowledge is transformed into a driving force for continuous model optimization, effectively bridging the semantic gap between natural language requirements and standardized mathematical modeling, and improving human-machine collaboration efficiency. Finally, the solution results are converted into interactive explanatory output, and the output method and optimization logic are dynamically adapted according to user objects and emergency stages, breaking the limitations of the opaque operation of traditional optimization methods. This allows users to clearly understand the decision-making basis, while retaining traceable process archives, providing solid, transparent, and efficient intelligent support for multi-party collaborative optimization decision-making in high-risk emergency scenarios. Attached Figure Description

[0017] Figure 1 This is a flowchart of an interactive emergency logistics distribution automatic optimization modeling method according to the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and 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.

[0019] Furthermore, in this invention, an element referred to as fixed to or disposed on another element may be directly disposed on the other element, or there may be an intermediate element. When an element is considered to be connected to another element, it may be directly connected to the other element, or there may be an intermediate element present simultaneously. The terms vertical, horizontal, left, right, and similar expressions used herein are for illustrative purposes only and do not represent the only possible implementation.

[0020] Example 1 This application provides an interactive automatic optimization modeling method for emergency logistics distribution, including the following steps A-D: Step A. Receive emergency logistics demand information containing natural language descriptions and business data. Through the multi-agent automated modeling module, based on a pre-built sample knowledge base and large model reasoning, convert the demand information into a standardized mathematical optimization model and executable code compatible with operations research solvers.

[0021] In this embodiment, this stage mainly realizes the conversion from natural language requirements to mathematical optimization models and executable code.

[0022] First, the system constructs a seed sample library for emergency logistics operations research optimization and initializes a multi-agent automated modeling system. The seed sample library stores historical problems, standard modeling schemes, solution codes, expert revisions, typical error patterns, and decision interpretation cases related to emergency logistics distribution optimization modeling, providing historical knowledge examples and reusable optimization experience for automated modeling. The sample library contains at least the following data types: emergency logistics problem cases (covering transportation scheduling, route selection, load allocation, resource scheduling, demand prioritization, multi-point distribution in disaster areas, warehouse location selection, and activation of temporary distribution centers); standard mathematical models (including formally expressed decision variables, objective functions, constraints, and model assumptions); executable solution codes (modeling programs and output programs compatible with operations research solvers); expert review data (revisions, error annotations, and applicability judgments); and interpretation cases (interpretable reports, sensitivity and counterfactual analysis results). Multi-agent communication is achieved through standardized messages (including task number, problem category, input data, model description, code file, solution status, error type, and expert feedback), forming a closed-loop pipeline.

[0023] Secondly, the mathematical modeling agent receives natural language demand descriptions for emergency logistics (which can be input by frontline responders, dispatchers, emergency management personnel, or external business systems), parses the demands, extracts entities such as warehouses, temporary distribution points, demand points, and vehicles, as well as attributes such as location, quantity, capacity, time, and cost, and establishes relationships such as supply and demand, routes, and vehicle tasks. The function of the mathematical modeling agent is defined as a mapping function. ;in, It represents the natural language task space, which includes unstructured text problem descriptions, structured tabular data, and operational parameters from real business environments in data-driven decision-making scenarios. It represents the mathematical model space, which consists of standardized optimization models that conform to the specifications, including core elements such as decision variables, objective functions, and constraints.

[0024] Subsequently, the coded agent maps the mathematical model into an executable solver; the function of the coded agent is defined as the mapping function. ;in, This is a standardized mathematical model space containing optimization problems that have already been formally expressed; The executable code space mainly consists of program code based on the Pyomo modeling framework and compatible with mainstream operations research and optimization solvers such as Gurobi. The coded agent generates a data reading module (for reading data from tables, databases, interfaces, or manually entered data), a model building module, an objective function module, a constraint generation module, a solution execution module, and a result output module.

[0025] Step B. Execute the executable code to obtain the solution result and verify the solution result; when the verification fails or an error occurs, perform hierarchical elastic repair on the executable code or standardized mathematical optimization model based on the execution feedback until a delivery plan that meets the preset conditions is obtained or the stopping condition is reached.

[0026] In this embodiment, this stage mainly realizes the automatic execution of executable code, result verification, and intelligent repair when errors occur.

[0027] The system executes the executable code to obtain the solution result and verifies it; the verification process includes feasibility verification, constraint satisfaction verification, and target performance verification of the standardized mathematical optimization model and the executable solution program; At the same time, the system uses a performance evaluation module to quantitatively evaluate the modeling results, and its scoring function is defined as: ;in, Indicates the number of verification indicators; Indicates the first Weight of class errors; For indicator functions; Indicates the first Verification conditions.

[0028] When verification fails or an error occurs during execution (such as code execution failure, solution exception, model infeasibility, constraint violation, etc.), an elastic repair process is initiated. The elastic repair process adopts a state machine-driven hierarchical repair mechanism, dynamically switching between the execution state and the repair state, and achieving automatic error correction and model reconstruction through solver feedback. The system first executes the executable code generated by the mathematical model M(t) and returns the solution feedback for the t-th iteration: ;in, This indicates the standard output information, including the solution results; It represents the standard error stream, including information such as syntax errors, solver exceptions, and infeasibility reports.

[0029] When execution feedback indicates that the error is a code-level error (such as syntax error, incorrect variable name, index call error, data reading error, solver interface call error, etc.), the system enters code repair state. The elastic repair agent regenerates or partially repairs the code based on the error information, but does not change the underlying mathematical model logic. When execution feedback indicates that the error is a model-level error (such as constraint conflict, incomplete variable definition, inconsistency between objective function and constraints, data conditions causing model infeasibility, omission of business constraints, etc.), the system enters model repair state. The elastic repair agent re-examines the variable definitions, objective function, and constraints in the mathematical model and generates a repaired model. After each repair, the execution verification agent re-executes the verification process. The system sets a hard limit of up to 5 repair attempts until a delivery plan that meets the preset conditions is obtained or the stopping condition is reached.

[0030] Step C. Receive feedback from domain experts on the model and delivery plan, and update the inference strategy of the sample knowledge base and the large model based on the feedback to continuously align with actual logistics and delivery constraints. In this embodiment, this stage mainly involves domain experts reviewing the model and delivery plan, as well as updating the system knowledge base.

[0031] The expert-driven improvement module receives review comments from domain experts on modeling schemes, solution results, or repair schemes. Domain experts review the actual operational constraints of emergency logistics, traffic conditions in disaster areas, material priorities, vehicle dispatching rules, etc. The review content includes: whether the mathematical model accurately reflects the actual delivery task; whether the objective function meets the decision-making objectives of the current emergency stage; whether the constraints omit key business restrictions; whether the vehicle routes and loading schemes are feasible on site; and whether the model output has mathematically feasible but unexecutable problems in reality.

[0032] Once the system is operational, emergency logistics experts will address different problem categories. (in The model output is systematically reviewed (representing a set of optimization problem categories) to identify typical error patterns and build an error knowledge base; for each problem category, experts generate a set of error summaries. : ;in, This indicates the number of typical errors summarized under this category; Indicates the first A common error pattern; Indicates an error The system provides correction suggestions; these expert annotations are stored in a knowledge base, and during the reasoning stage, prompt words are automatically retrieved and injected according to the problem category, forming an error-aware reasoning mechanism that drives the system to continuously align with actual logistics and delivery constraints.

[0033] Step D. Convert the delivery plan, solution logic and optimization basis that meet the preset conditions into an interactive explanation result output, and dynamically adapt the output method and optimization logic according to the current emergency logistics stage and user type. At the same time, record the entire process operation to form a traceable process archive.

[0034] This stage mainly focuses on achieving interpretable output of optimization results, dynamic adaptation, and full-process recording.

[0035] The decision-maker explanation module converts the delivery plan, solution logic, and optimization basis that meet preset conditions into interactive explanation results output. The interactive explanation methods include query, response, and action suggestion generation. Users can query delivery plan details, bottleneck causes, constraint impacts, sensitivity changes, etc. through natural language. Based on the query call model results, solution logs, constraint analysis results, and historical cases, the system generates delivery plan overview (allocation relationships of various vehicles, warehouses, demand points, and material categories), route arrangement description, constraint satisfaction status (capacity, time window, inventory, etc.), performance bottleneck analysis, sensitivity analysis, and counterfactual analysis results.

[0036] The dynamic adaptation module dynamically adapts the output method and optimization logic based on the current emergency logistics stage and user type. On one hand, it adapts according to the stage: prioritizing timeliness in the initial response stage, balancing speed, cost, and allocation fairness in the stable rescue stage, and focusing on long-term resilience and cost-effectiveness in the post-disaster recovery stage. On the other hand, it adapts according to the user group: providing natural language interaction and visual causal explanations for frontline responders; providing domain experts with deep intervention interfaces for custom constraint injection and model logic revision; and providing decision-makers with multi-solution comparison, robustness assurance, and extreme scenario assessment. Simultaneously, the system records the entire process of this task, including requirement inputs, model version, code version, execution logs, repair records, expert feedback, and explanation reports, forming a traceable process archive to support rapid retrieval and iterative optimization for subsequent tasks.

[0037] Example 2 This embodiment provides an interactive emergency logistics distribution automatic optimization modeling system, which includes an LLM initialization module, a multi-agent collaboration module, a verification and resilient repair module, an expert-driven improvement module, a decision-maker interpretation module, and a dynamic adaptation module.

[0038] The LLM initialization module is used to build a seed sample library for emergency logistics distribution optimization modeling, and to perform feasibility verification, constraint satisfaction detection, and target performance evaluation of the initial scheme through a generative code pipeline. The multi-agent collaboration module is used to decompose the modeling process into specialized sub-agent tasks such as requirement analysis, model building, code execution, and iterative debugging. The expert-driven improvement module combines the actual operational constraints of emergency logistics and transforms expert revision opinions into the basis of the multi-agent evolution framework. The decision-maker interpretation module is used to retrieve details of the distribution plan, identify performance bottlenecks, test the sensitivity of core assumptions, and summarize feasible emergency decision-making suggestions. The dynamic adaptation module dynamically adapts to different stages of emergency logistics and different user types, and records the process and feeds it back to the sample library for experience reuse.

[0039] Example 3 Based on the aforementioned Embodiments 1 and 2, this embodiment combines Figure 1 The specific implementation of this application is explained as follows: The specific steps are as follows: Step S101: Construct a seed sample library and build a multi-agent system; Step S102 uses a mathematical modeling agent to parse natural language requirements and generate an optimized model; Step S103 involves the automatic generation of solution code by the coded intelligent agent; Step S104: Experts review the plan, provide correction instructions, and the framework evolves automatically; Step S105: Execute the solution code and determine whether it was successful; If successful, proceed to step S106 to output the explanation results and decision suggestions; if unsuccessful, depending on the error type, proceed to code self-repair (post-step process) or model reconstruction, then update the model and code and re-enter the execution loop; after execution, proceed to step S107 to perform scene and user dynamic adaptation and record the process.

[0040] These specific sub-steps fully cover the entire automated process from natural language interactive input, model and code generation, execution verification and layered repair, to final interpretation output.

[0041] Application examples The following example illustrates the implementation process of this invention using an emergency supplies delivery task in a sudden disaster scenario. Assume a sudden disaster occurs in a certain area, and emergency management personnel input the following natural language requirements: There are two material warehouses, several disaster relief sites, and multiple transport vehicles with different load capacities. It is necessary to prioritize the delivery of drinking water, food, and medical supplies to high-priority relief sites within a limited time. At the same time, some roads are damaged, and some vehicles cannot enter high-risk areas. The system is required to generate an executable delivery plan and explain the basis for the plan.

[0042] After receiving the request, the request parsing agent identifies task elements such as warehouses, resettlement sites, vehicles, materials, roads, risk areas, time constraints, and delivery priorities, and organizes the relevant tabular data and business parameters into structured input.

[0043] The mathematical modeling agent generates a standardized mathematical optimization model based on the analysis results. The model uses vehicle scheduling routes, material loading quantity, and delivery service relationships as decision variables, with the optimization objectives of maximizing emergency demand coverage and minimizing delivery delays. It also sets constraints such as vehicle capacity, warehouse inventory, road accessibility, time windows, risk area restrictions, and demand satisfaction.

[0044] The coding agent converts the standardized mathematical optimization model into executable solution code and generates data reading and result output interfaces. After the execution verification agent runs the code, if the solver returns a feasible solution, it outputs the vehicle route, material allocation, and estimated arrival time. If it returns a syntax error, runtime exception, or model infeasibility, it initiates a resilient repair process. When the execution feedback shows an error in referencing a vehicle capacity, the resilient repair agent determines that the error is at the code level, executes code repair, and reruns the solution. When the execution feedback shows that all high-priority demand points are unreachable due to road closures, the resilient repair agent determines that the problem is at the model or business constraint level, reconstructs alternative path constraints, or suggests the need to activate temporary distribution points. If multiple repairs still fail to yield effective results, the system outputs the reason for the failure and suggests expert intervention.

[0045] After domain experts review the delivery plans generated by the system, they can point out that some roads, although not completely closed, only allow small vehicles to pass, or that certain types of medical supplies must be prioritized for delivery to specific resettlement points. The expert-driven improvement module transforms this opinion into structured rules and feeds it back into the mathematical modeling agent and the error knowledge base. In subsequent similar tasks, the system automatically retrieves this type of expert knowledge and uses it for modeling.

[0046] The decision-maker interpretation module generates an interpretation report based on the final solution results. The interpretation report can explain: which vehicles departed from which warehouses, transported which materials, and arrived at which demand points; which demand points were prioritized; which road or vehicle constraints affected the delivery results; and how the delivery coverage would change if an additional vehicle was added, a temporary warehouse was activated, or some non-critical time restrictions were relaxed.

[0047] For frontline responders, the system provides concise action suggestions, such as vehicle A departing from warehouse 1 to prioritize delivering medical supplies to resettlement site 3, with an estimated arrival time of a certain period. For domain experts, the system provides model constraints, variable relationships, and repair records. For decision-makers, the system provides multi-option comparisons, robustness analysis, and extreme scenario assessments.

[0048] The system ultimately writes the task's requirements, model version, code version, execution logs, repair records, expert feedback, explanation reports, and final delivery plan into a traceable process archive; this process archive is used for rapid retrieval, model reuse, and iterative optimization of subsequent similar emergency logistics tasks.

[0049] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0050] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. An interactive automatic optimization modeling method for emergency logistics distribution, characterized in that, include: The system receives emergency logistics demand information containing natural language descriptions and business data. Through a multi-agent automated modeling module, based on a pre-built sample knowledge base and large model reasoning, the demand information is converted into a standardized mathematical optimization model and executable code compatible with operations research solvers. The executable code is executed to obtain the solution result, and the solution result is verified. When verification fails or an error occurs during execution, the executable code or standardized mathematical optimization model is repaired in a layered and flexible manner based on the execution feedback until a delivery plan that meets the preset conditions is obtained or the stopping condition is reached. The system receives feedback from domain experts on the model and delivery plan, and updates the sample knowledge base and the inference strategy of the large model based on the feedback to continuously align with actual logistics and delivery constraints. The delivery plan, solution logic, and optimization basis that meet the preset conditions are converted into interactive explanation results output. The output method and optimization logic are dynamically adapted according to the current emergency logistics stage and user type. At the same time, the entire process operation is recorded to form a traceable process archive.

2. The interactive emergency logistics distribution automatic optimization modeling method according to claim 1, characterized in that, The multi-agent automated modeling module includes: Mathematical modeling agent, used to construct the first mapping function Natural language task space Mapped to mathematical model space ; in, It represents the natural language task space, which includes unstructured text problem descriptions, structured tabular data, and operational parameter information from real business environments; The mathematical model space is composed of standardized optimization models that conform to the norm, including decision variables, objective functions, and constraints. Encode an intelligent agent to construct the second mapping function. The mathematical model space Mapped to executable code space ;in, This represents the executable code space, which consists of a modeling framework and solution code compatible with operations research solvers.

3. The interactive emergency logistics distribution automatic optimization modeling method according to claim 1, characterized in that, The sample knowledge base includes emergency logistics operation and optimization problem cases, standard mathematical models corresponding to the problem cases, executable solution programs, expert revision opinions, typical error modes and decision interpretation cases, which are used to provide historical modeling knowledge examples and reusable optimization experience for the automated modeling module.

4. The interactive emergency logistics distribution automatic optimization modeling method according to claim 1, characterized in that, The emergency logistics demand information includes unstructured text descriptions, structured tabular data, on-site operation parameters, resource constraint information, delivery task information, and user input information; The entities extracted from the demand information include warehouses, demand points, vehicles, materials, and road nodes, and the attributes include location, quantity, capacity, time, cost, and priority.

5. The interactive emergency logistics distribution automatic optimization modeling method according to claim 1, characterized in that, The strategy for updating the large model inference based on review feedback includes: For different problem categories Construct an error knowledge base, the error knowledge base Represented as ; in, Represents the set of optimization problem categories; This indicates the number of typical errors summarized under this category; Indicates the first A common error pattern; Indicates an error The suggested corrections provided; The automated modeling module is based on the problem category. From the error knowledge base Correction prompts for searching in the middle This information is then injected into subsequent reasoning prompts, forming an error-aware reasoning mechanism.

6. The interactive emergency logistics distribution automatic optimization modeling method according to claim 1, characterized in that, The verification of the solution results includes feasibility verification, constraint satisfaction verification, and target performance evaluation. The modeling results are quantitatively evaluated through the performance evaluation module, and its scoring function is defined as follows: in, Indicates the number of verification indicators; Indicates the first Weight of class errors; For indicator functions; Indicates the first Verification conditions.

7. The interactive emergency logistics distribution automatic optimization modeling method according to claim 1, characterized in that, The layered elastic repair includes: Execution by mathematical model The generated executable code, and returns the first... The solution feedback for the next iteration is expressed as follows: in, This indicates the standard output information, including the solution results; It represents the standard error stream, including syntax errors, solver exceptions, and infeasibility report information; When the execution feedback indicates that the error is a code-level error, the execution code self-corrects. When the execution feedback indicates that the error is a model-level error, perform model reconstruction or remodeling. The stopping conditions include obtaining effective delivery optimization results, reaching the preset maximum number of repairs, or determining that the current task cannot continue to be automatically repaired.

8. The interactive emergency logistics distribution automatic optimization modeling method according to claim 1, characterized in that, The interactive explanation results include query, response, and action suggestion generation, which are used to show users delivery plan details, constraint satisfaction, performance bottlenecks, sensitivity analysis results, counterfactual analysis results, and actionable decision suggestions.

9. The interactive emergency logistics distribution automatic optimization modeling method according to claim 1, characterized in that, The dynamic adaptation includes: Based on the initial response phase, stabilization and rescue phase, and post-disaster recovery phase of emergency logistics, adjust the modeling optimization objectives, constraint priorities, and result interpretation methods. Furthermore, it adjusts the interaction methods, explanation granularity, and intervention permissions based on the different user groups' operational capabilities, professional backgrounds, and risk preferences.

10. An interactive emergency logistics distribution automatic optimization modeling system, characterized in that, include: The multi-agent automated modeling unit receives emergency logistics demand information containing natural language descriptions and business data. Based on a pre-built sample knowledge base, it constructs a first mapping function through mathematical modeling agents. Generate a standardized mathematical optimization model and construct a second mapping function through an coded intelligent agent. Generate executable code compatible with the operations research solver; where, Representing the natural language task space. Represents the space of mathematical models. Indicates the executable code space; The verification and hierarchical repair unit is used to execute the executable code and verify the results. If the verification fails, it is based on the solution feedback. Perform tiered, flexible repair until a delivery plan that meets preset conditions is generated; among which, Indicates standard output information, Represents the standard error stream; An expert-driven optimization unit is used to receive review feedback from domain experts and construct an error knowledge base based on the review feedback. and using the scoring function The modeling results are quantitatively evaluated, and the inference strategy of the large model is updated based on the evaluation results and feedback; among them, Indicates the number of typical errors. Indicates the error mode. This indicates a correction prompt. Indicates the number of verification indicators. Indicates the error weight. For indicator functions, Indicate the verification conditions; The interactive explanation and dynamic adaptation unit is used to convert the delivery plan, solution logic and optimization basis into interactive explanation results output, and dynamically adapt the output method and optimization logic according to the emergency scenario status and user object, and record the whole process information to form a traceable process archive.

Images