Decision strategy determination method and device based on operational research decision large model, electronic equipment and storage medium

Abstract

The embodiment of the invention provides a decision strategy determination method and device based on an operational research decision large model, electronic equipment and a storage medium. The method comprises the steps that decision demand information is acquired; according to a pre-trained operational research decision-making large model, determining a decision-making strategy, the pre-trained operational research decision-making large model at least comprising a decision-making routing model and a decision-making sub-model connected with the decision-making routing model, and the decision-making routing model being used for determining the decision-making sub-model corresponding to the decision-making problem type; the decision sub-model is used for determining a decision strategy corresponding to the decision demand information, the operational research decision large model has higher accuracy, and the pre-trained large language model can learn language laws and knowledge from a large amount of data; therefore, higher accuracy and robustness are shown in information extraction tasks of constraint conditions, targets and the like related to domain problems, and the large language model has better adaptability to complex and unknown input and various different scenes.

Description

Technical Field

[0001] This application relates to the field of computer technology, and more particularly, to a method, apparatus, electronic device, and storage medium for determining a decision-making strategy based on an operations research decision-making large model. Background Art

[0002] Operations research decision-making refers to the process of using the theories and methods of operations research to rationally allocate resources under certain constraints to achieve the optimal selection of specific goals. Traditional operations research decision-making requires professional operations research experts or statisticians, and requires manual definition of constraints and goals. The process is complex and professional. However, if it involves multiple fields, it is impossible to accurately determine the operations research plan. Therefore, how to improve the accuracy of determining the operations research plan is an urgent problem to be solved at present. Summary of the Invention

[0003] Some embodiments of this application aim to provide a method, apparatus, electronic device, and storage medium for determining a decision-making strategy based on an operations research decision-making large model. Through the technical solutions of the embodiments of this application, by obtaining decision-making requirement information, where the decision-making requirement information at least includes the decision problem type; according to a pre-trained operations research decision-making large model, determining a decision-making strategy corresponding to the decision-making requirement information, where the pre-trained operations research decision-making large model at least includes a decision routing model and a decision sub-model connected to the decision routing model, and the decision routing model is used to determine a decision sub-model corresponding to the decision problem type; the decision sub-model is used to determine a decision-making strategy corresponding to the decision-making requirement information. The operations research decision-making large model in the embodiments of this application has higher accuracy. The pre-trained large language model can learn language rules and knowledge from a large amount of data, so it shows higher accuracy and robustness in information extraction tasks related to domain problems such as constraints and goals. The large language model has better adaptability to complex and unknown inputs and various different scenarios.

[0004] In a first aspect, some embodiments of this application provide a method for determining a decision-making strategy based on an operations research decision-making large model, including: Obtaining decision-making requirement information, where the decision-making requirement information at least includes the decision problem type; According to a pre-trained operations research decision-making large model, determining a decision-making strategy corresponding to the decision-making requirement information, including: Using a large language model to parse the decision-making requirement information to obtain a decision problem type corresponding to the decision-making requirement information, and constraint conditions, goals, and resource query information corresponding to the decision problem type; Determining corresponding resource allocation information according to the resource query information; Through the decision routing model, determine a decision sub-model corresponding to the decision requirement information according to the resource configuration information and the decision problem type; Through the decision sub-model, determine a decision strategy corresponding to the decision requirement information; Among them, the pre-trained operations research decision-making large model at least includes a decision routing model and a decision sub-model connected to the decision routing model. The decision routing model is used to determine a decision sub-model corresponding to the decision problem type; the decision sub-model is used to determine a decision strategy corresponding to the decision requirement information.

[0005] Some embodiments of the present application have higher accuracy by setting an operations research decision-making large model. The pre-trained large language model can learn language rules and knowledge from a large amount of data, thus showing higher accuracy and robustness in information extraction tasks related to domain problem constraints, goals, etc. The large language model has better adaptability to complex and unknown inputs and various different scenarios. Some embodiments of the present application allow the model to dynamically balance among conflicting goals such as resource allocation and cost control through a multi-objective weight allocation algorithm, and optimize the expert network selection strategy through reinforcement learning.

[0006] Optionally, the pre-trained operations research decision-making large model is constructed based on the MoE structure.

[0007] Optionally, the pre-trained operations research decision-making large model is obtained through the following steps: Obtain sample data; Process the sample data to obtain training data corresponding to the sample decision problem type, where the sample decision problem type at least includes operations research problems such as path planning or inventory management; Train the large model according to the training data corresponding to the sample decision problem type to obtain a decision sub-model corresponding to the sample decision problem type; Determine the decision routing model according to the decision sub-model corresponding to the sample decision problem type; Determine the operations research decision-making large model according to the decision routing model and the decision sub-model. The pre-trained operations research decision-making large model provided by some embodiments of the present application can efficiently handle complex system problems. The operations research decision-making large model can handle complex scenarios with multiple variables and dynamic constraints by integrating mathematical optimization algorithms (such as linear programming and stochastic integer programming) and deep learning technologies.

[0008] Optionally, the processing of the sample data includes: When the number of bytes of the sample data is greater than a preset value, the MLA technology is used to compress redundant information in the sample data to obtain processed data.

[0009] In some embodiments of the present application, for long sequence inputs (such as supply chain network node data), redundant information is compressed by the MLA technology to reduce computational complexity. Optionally, the method further includes: Adjusting the number of the decision sub-models according to the number of sample decision problems of different types.

[0010] In some embodiments of the present application, the activation ratio of the expert sub-network is dynamically adjusted according to the type of the operations research problem (such as path planning, inventory management) to improve the resource allocation efficiency.

[0011] In a second aspect, some embodiments of the present application provide a device for determining a decision-making strategy based on an operations research decision-making large model, including: An acquisition module, configured to acquire decision-making requirement information, where the decision-making requirement information at least includes a decision problem type; A decision-making module, configured to determine a decision-making strategy corresponding to the decision-making requirement information according to a pre-trained operations research decision-making large model, where the pre-trained operations research decision-making large model at least includes a decision routing model and a decision sub-model connected to the decision routing model, and the decision routing model is configured to determine a decision sub-model corresponding to the decision problem type; the decision sub-model is configured to determine a decision-making strategy corresponding to the decision-making requirement information; specifically: Using a large language model to parse the decision-making requirement information to obtain a decision problem type corresponding to the decision-making requirement information, and constraint conditions, objectives, and resource query information corresponding to the decision problem type; Determining corresponding resource configuration information according to the resource query information; Through the decision routing model, determining a decision sub-model corresponding to the decision-making requirement information according to the resource configuration information and the decision problem type; Determining a decision-making strategy corresponding to the decision-making requirement information through the decision sub-model.

[0012] Some embodiments of the present application set up an operational research decision-making large model, which has higher accuracy. The pre-trained large language model can learn language rules and knowledge from a large amount of data, and thus show higher accuracy and robustness in information extraction tasks related to domain problem constraints, objectives, etc. The large language model has better adaptability to complex and unknown inputs and various different scenarios. Some embodiments of the present application use a multi-objective weight allocation algorithm to allow the model to dynamically balance among conflicting objectives such as resource allocation and cost control, and optimize the expert network selection strategy through reinforcement learning.

[0013] Optionally, the pre-trained operational research decision-making large model is constructed based on the MoE structure.

[0014] Optionally, the device further includes a training module, and the training module is used for: Obtain sample data; Process the sample data to obtain training data corresponding to the sample decision problem type, and the sample decision problem type at least includes an operational research problem of path planning or an operational research problem of inventory management; Train the large model according to the training data corresponding to the sample decision problem type to obtain a decision sub-model corresponding to the sample decision problem type; Determine the decision routing model according to the decision sub-model corresponding to the sample decision problem type; Determine the operational research decision-making large model according to the decision routing model and the decision sub-model. The pre-trained operational research decision-making large model provided by some embodiments of the present application can efficiently process complex system problems. The operational research decision-making large model can process complex scenarios with multiple variables and dynamic constraints by integrating mathematical optimization algorithms (such as linear programming and stochastic integer programming) and deep learning technologies.

[0015] Optionally, the training module is used for: When the number of bytes of the sample data is greater than a preset value, use the MLA technology to compress the redundant information in the sample data to obtain processed data.

[0016] Some embodiments of the present application compress redundant information for long sequence inputs (such as supply chain network node data) through the MLA technology to reduce the computational complexity. Optionally, the training module is used for: Adjust the number of the decision sub-models according to the number of different types of sample decision problems.

[0017] Some embodiments of the present application dynamically adjust the activation ratio of the expert sub-networks according to the type of operational research problems (such as path planning and inventory management) to improve the resource allocation efficiency.

[0018] In a third aspect, some embodiments of the present application provide an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, the method for determining a decision-making strategy based on an operations research decision-making large model as described in any embodiment of the first aspect can be implemented.

[0019] In a fourth aspect, some embodiments of the present application provide a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, the method for determining a decision-making strategy based on an operations research decision-making large model as described in any embodiment of the first aspect can be implemented.

[0020] In a fifth aspect, some embodiments of the present application provide a computer program product, which includes a computer program. When the computer program is executed by a processor, the method for determining a decision-making strategy based on an operations research decision-making large model as described in any embodiment of the first aspect can be implemented. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] To more clearly illustrate the technical solutions of some embodiments of the present application, the drawings required for some embodiments of the present application will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present application and should not be regarded as limiting the scope. For those of ordinary skill in the art, other related drawings can be obtained based on these drawings without creative efforts.

[0022] Figure 1 It is a schematic flowchart of a method for determining a decision-making strategy based on an operations research decision-making large model provided by an embodiment of the present application; Figure 2 It is a schematic flowchart of another method for determining a decision-making strategy based on an operations research decision-making large model provided by an embodiment of the present application; Figure 3 It is a schematic structural diagram of a device for determining a decision-making strategy based on an operations research decision-making large model provided by an embodiment of the present application; Figure 4 It is a schematic diagram of an electronic device provided by an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0023] The technical solutions in some embodiments of the present application will be described below with reference to the drawings in some embodiments of the present application.

[0024] It should be noted that similar reference numerals and letters denote similar items in the following figures. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. At the same time, in the description of the present application, terms such as "first" and "second" are only used for distinguishing descriptions and cannot be construed as indicating or implying relative importance.

[0025] Operational decision-making refers to the process of using the theories and methods of operations research to rationally allocate resources under certain constraints to achieve the optimal selection of specific goals. Traditional operational decision-making requires professional operations researchers or statisticians, and manual definition of constraints and goals. The process is complex and professional. However, if it involves multiple fields, it is impossible to accurately determine the operational plan. Therefore, how to improve the accuracy of determining the operational plan is an urgent problem to be solved at present. In view of this, some embodiments of the present application provide a method for determining a decision-making strategy based on an operational decision-making large model. The method includes obtaining decision-making requirement information, where the decision-making requirement information at least includes the type of decision-making problem; determining a decision-making strategy corresponding to the decision-making requirement information according to a pre-trained operational decision-making large model, where the pre-trained operational decision-making large model at least includes a decision-making routing model and a decision-making sub-model connected to the decision-making routing model, and the decision-making routing model is used to determine the decision-making sub-model corresponding to the type of decision-making problem; the decision-making sub-model is used to determine the decision-making strategy corresponding to the decision-making requirement information. The operational decision-making large model in the embodiments of the present application has higher accuracy. The pre-trained large language model can learn language rules and knowledge from a large amount of data, and thus shows higher accuracy and robustness in tasks such as extracting information about constraints and goals related to domain problems. The large language model has better adaptability to complex and unknown inputs and various different scenarios.

[0026] As Figure 1 shown, some embodiments of the present application provide a method for determining a decision-making strategy based on an operational decision-making large model. The method includes: S101. Obtain decision-making requirement information, where the decision-making requirement information at least includes the type of decision-making problem; Specifically, the terminal device obtains the decision-making requirement information input by the user and parses the decision-making requirement information to obtain the type of decision-making problem in the decision-making requirement information. For example, the type of decision-making problem may be a path decision-making problem, or branches such as linear programming, nonlinear programming, integer programming, combinatorial programming, graph theory, network flow, decision analysis, queuing theory, reliability mathematics theory, inventory theory, game theory, search theory, and simulation.

[0027] S102. Determine a decision-making strategy corresponding to the decision requirement information according to a pre-trained large-scale operation research decision-making model. The pre-trained large-scale operation research decision-making model at least includes a decision routing model and a decision sub-model connected to the decision routing model. The decision routing model is used to determine a decision sub-model corresponding to the decision problem type, and the decision sub-model is used to determine a decision-making strategy corresponding to the decision requirement information.

[0028] Specifically, a large amount of sample data is obtained on the terminal device, the sample data is parsed to obtain training data corresponding to different sample decision problem types, and the large language model is trained using the training data to obtain a decision routing model and a decision sub-model. The decision routing decision is connected to multiple decision sub-models, that is, a large-scale operation research decision-making model is formed. The decision routing model is used to determine a decision sub-model corresponding to the decision problem type, and the decision sub-model is used to determine a decision-making strategy corresponding to the decision requirement information. The data of the decision sub-model can be determined according to the number of sample decision problem types. The large-scale operation research decision-making model can better understand the user's intention. In terms of user demand parsing, the large model can better understand the user's true needs and intentions.

[0029] After the terminal device obtains the decision problem type, it uses the decision routing model in the pre-trained large-scale operation research decision-making model to determine a decision sub-model corresponding to the decision problem type, and uses the decision sub-model to determine a decision-making strategy corresponding to the decision requirement information, that is, configures according to the resource information of the decision problem type to obtain a target decision-making strategy corresponding to the decision requirement information.

[0030] Some embodiments of the present application have higher accuracy by setting a large-scale operation research decision-making model. The pre-trained large language model can learn language rules and knowledge from a large amount of data, so it shows higher accuracy and robustness in information extraction tasks related to domain problems, such as constraints, objectives, etc. The large language model has better adaptability to complex and unknown inputs and various different scenarios.

[0031] Another embodiment of the present application further supplements and explains the method for determining the decision-making strategy based on the large-scale operation research decision-making model provided in the above embodiment.

[0032] Optionally, determining a decision-making strategy corresponding to the decision requirement information according to the pre-trained large-scale operation research decision-making model includes: Using a large language model to parse the decision requirement information to obtain a decision problem type corresponding to the decision requirement information, and constraint conditions, objectives, and resource query information corresponding to the decision problem type; Determine the corresponding resource configuration information according to the resource query information; Among them, the large model can integrate structured data (such as supply chain inventory) and unstructured data (such as meteorological reports, policy texts), and achieve semantic understanding through natural language processing (NLP) and knowledge graph technology (trees, graph knowledge bases, databases). That is to say, according to the resource query information, the knowledge graph technology can be used to search in the trees, graph knowledge bases, and databases to obtain the resource configuration information corresponding to the resource query information.

[0033] Through the decision routing model, according to the resource configuration information and the type of decision problem, determine the decision sub-model corresponding to the decision requirement information; Through the decision sub-model, determine the decision strategy corresponding to the decision requirement information.

[0034] In the embodiment of this application, the pre-trained operations research decision-making large model in the terminal device performs the following operations: (1) Multi-stage reasoning framework: Split the problem into "modeling - solving - verification" stages, and optimize the search path by combining Monte Carlo tree search (MCTS).

[0035] (2) Combine the lightweight characteristics of edge computing and the large model to propose a regional collaborative reasoning framework. For example, deploy the supply chain network nodes to edge devices, generate sub-problem solutions through the local operations research decision-making model, and the central node integrates the global optimal solution.

[0036] (3) Dynamic termination and backtracking mechanism: Introduce dual-threshold control (confidence threshold + time threshold). When the uncertainty of the output of the operations research decision-making large model is too high, trigger backtracking, and reallocate computing resources to the critical reasoning path to avoid invalid calculations.

[0037] (4) Develop a dedicated prompt word template to structurally express the operations research problem as a "goal - constraint - variable" triple to improve the model's analytical ability for non-linear constraints.

[0038] During the execution of the operations research decision-making large model, the decision routing model in the pre-trained operations research decision-making large model can be used to perform: 1. Problem-aware research and judgment. Facing the user's description and solution requirements for an operations research problem in a certain field, this situation awareness and research and judgment uses the information extraction technology of the large language model to analyze the user's requirements. Obtain a series of information such as constraint conditions, goals, and real-time resources to be queried related to the field problem.

[0039] 2. Precise resource matching. After the situation awareness judges and analyzes the user input and obtains a series of real-time resource query commands, the precise resource matching first combines the code generation ability and tool calling ability of the large language model. Without the need for targeted database query function development, it links different external databases to obtain real-time information of resources. Subsequently, it standardizes different data organizational structures to provide a reference basis for the final solution for callable resources.

[0040] 3. Formulation of the operational decision-making plan. It is mainly achieved by constructing a multi-agent for plan formulation based on the large model. The general planner agent assigns tasks to agents with different responsibilities such as the operational planning officer, code executor, document officer, and evaluator. The agents assist each other to finally complete the plan formulation. The plan formulation system has capabilities such as agent role definition, tool calling, task assignment and coordination, automated execution, self-optimization, and feedback. By parallel processing tasks by agents and optimizing task execution, the efficiency of plan formulation can be improved.

[0041] 4. The decision routing model determines a decision sub-model corresponding to the decision problem type. The decision sub-model is used to evaluate the operational decision-making plan and conduct qualitative and quantitative evaluations on the rationality of the operational decision-making plan, so as to ensure the accuracy of the plan.

[0042] Some embodiments of the present application allow the model to dynamically balance among conflicting objectives such as resource allocation and cost control through a multi-objective weight allocation algorithm, and optimize the expert network selection strategy through reinforcement learning.

[0043] Embodiments of the present application provide a method for solving operational decision-making problems based on a large model. Only by describing the problem in natural language, the large model automatically extracts the constraint conditions and objectives, conducts complex reasoning, and gives an answer. The present invention constructs an operational decision-making large model. Compared with general large language models, such as Figure 2 as shown: Optionally, the pre-trained operational decision-making large model is obtained through the following steps: Obtain sample data; Process the sample data to obtain training data corresponding to the sample decision problem type. The sample decision problem type includes at least operational problems of path planning or operational problems of inventory management; Train the large model according to the training data corresponding to the sample decision problem type to obtain a decision sub-model corresponding to the sample decision problem type; Determine the decision routing model according to the decision sub-model corresponding to the sample decision problem type; Determine the decision routing model and the decision sub-model connected to the decision routing model as the operational decision-making large model. Some embodiments of the present application provide a pre-trained operations research decision-making large model that can efficiently handle complex system problems. The operations research decision-making large model can handle complex scenarios with multiple variables and dynamic constraints by integrating mathematical optimization algorithms (such as linear programming and stochastic integer programming) and deep learning technologies.

[0044] Specifically, the terminal device optimizes the structure of the general large language model, designs an expert sub-network according to the type of operations research problem (such as path planning and inventory management), extracts optimization logics such as linear programming, integer programming, mixed integer programming, and non-linear programming from professional operations research models (such as CPLEX and Gurobi), and performs COT fine-tuning training on the above large language model to generate an operations research large model.

[0045] In the embodiments of the present application, the activation ratio of the expert sub-network is dynamically adjusted according to the type of operations research problem (such as path planning and inventory management) to implement a dynamic MoE activation strategy and improve the resource allocation efficiency. A multi-objective weight allocation algorithm under a MoE architecture provided in the embodiments of the present application allows the model to dynamically balance conflicting objectives such as resource allocation and cost control, and optimizes the expert network selection strategy through reinforcement learning. MOE (Mixture of Experts) is a technical architecture used to improve the performance and efficiency of deep learning models. Its core idea is to combine multiple specialized sub-models (called "experts") and use a gating network to dynamically select the most suitable expert to process the input data, thereby improving the flexibility and computational efficiency of the model.

[0046] Using CoT (Chain of Thought) data distillation technology, extract optimization logics from professional operations research models (such as CPLEX and Gurobi) to enhance the domain adaptability of the operations research decision-making large model.

[0047] Optionally, processing the sample data includes: When the number of bytes of the sample data is greater than a preset value, use the MLA technology to compress the redundant information in the sample data to obtain the processed data.

[0048] Some embodiments of the present application compress redundant information through the MLA technology for long sequence inputs (such as supply chain network node data) to reduce the computational complexity. Optionally, the method further includes: Adjust the number of decision sub-models according to the number of sample decision problems of different types.

[0049] Specifically, the pre-trained operational decision-making large model in this application is based on the MOE (Mixture of Experts) structure, and can be adjusted according to the number of sample decision problems of different types. Weights can also be set for the decision sub-models corresponding to different types of sample decision problems. For example, in logistics distribution optimization, the model can simultaneously consider dozens of variables such as route planning, vehicle capacity, and time windows to generate a globally optimal solution. Compared with traditional operations research models, the large model reduces the limitations of manual modeling through end-to-end learning and improves the solution efficiency.

[0050] Some embodiments of this application dynamically adjust the activation ratio of the expert sub-network according to the type of operations research problem (such as route planning, inventory management) to improve the resource allocation efficiency.

[0051] It should be noted that each implementable manner in this embodiment can be implemented independently, or can be combined in any combination manner without conflict. This application does not make any limitations.

[0052] Another embodiment of this application provides a device for determining a decision-making strategy based on an operational decision-making large model, which is used to execute the method for determining a decision-making strategy based on an operational decision-making large model provided in the above embodiment.

[0053] As Figure 3 shown, it is a schematic structural diagram of a device for determining a decision-making strategy based on an operational decision-making large model provided in an embodiment of this application. The device for determining a decision-making strategy based on an operational decision-making large model includes an acquisition module 301 and a decision-making module 302, where: The acquisition module 301 is used to acquire decision requirement information, where the decision requirement information at least includes the decision problem type; The decision-making module 302 is used to determine a decision-making strategy corresponding to the decision requirement information according to the pre-trained operational decision-making large model. The pre-trained operational decision-making large model at least includes a decision routing model and a decision sub-model connected to the decision routing model. The decision routing model is used to determine the decision sub-model corresponding to the decision problem type; the decision sub-model is used to determine the decision-making strategy corresponding to the decision requirement information.

[0054] Regarding the device in this embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment related to the method, and will not be elaborated here.

[0055] Some embodiments of the present application have higher accuracy by setting up an operational research decision-making large model. The pre-trained large language model can learn language rules and knowledge from a large amount of data, thus showing higher accuracy and robustness in information extraction tasks related to domain problem constraints, objectives, etc. The large language model has better adaptability to complex and unknown inputs and various different scenarios.

[0056] Another embodiment of the present application further supplements the device for determining the decision-making strategy based on the operational research decision-making large model provided in the above embodiment.

[0057] Optionally, the decision-making module is used for: Using a large language model to analyze the decision-making requirement information to obtain the decision-making problem type corresponding to the decision-making requirement information, as well as the constraint conditions, objectives, and resource query information corresponding to the decision-making problem type; Determine the corresponding resource allocation information according to the resource query information; Through the decision-making routing model, determine the decision-making sub-model corresponding to the decision-making requirement information according to the resource allocation information and the decision-making problem type; Determine the decision-making strategy corresponding to the decision-making requirement information through the decision-making sub-model.

[0058] Some embodiments of the present application allow the model to dynamically balance among conflicting objectives such as resource allocation and cost control through a multi-objective weight allocation algorithm, and optimize the expert network selection strategy through reinforcement learning.

[0059] Optionally, the device further includes a training module, and the training module is used for: Obtain sample data; Process the sample data to obtain training data corresponding to the sample decision-making problem type, and the sample decision-making problem type at least includes operational research problems such as path planning or inventory management; Train the large model according to the training data corresponding to the sample decision-making problem type to obtain a decision-making sub-model corresponding to the sample decision-making problem type; Determine the decision-making routing model according to the decision-making sub-model corresponding to the sample decision-making problem type; Determine the operational research decision-making large model according to the decision-making routing model and the decision-making sub-model. The pre-trained operational research decision-making large model provided by some embodiments of the present application can efficiently process complex system problems. The operational research decision-making large model can handle complex scenarios with multiple variables and dynamic constraints by integrating mathematical optimization algorithms (such as linear programming and stochastic integer programming) and deep learning technologies.

[0060] Optionally, the training module is used for: When the number of bytes of the sample data is greater than a preset value, the MLA technology is used to compress the redundant information in the sample data to obtain the processed data.

[0061] In some embodiments of the present application, for long sequence inputs (such as supply chain network node data), the MLA technology is used to compress redundant information and reduce the computational complexity. Optionally, the training module is used to: Adjust the number of decision sub-models according to the number of sample decision problems of different types.

[0062] In some embodiments of the present application, the activation ratio of the expert sub-network is dynamically adjusted according to the type of the operations research problem (such as path planning, inventory management) to improve the resource allocation efficiency.

[0063] Regarding the device in this embodiment, the specific manner in which each module performs operations has been described in detail in the embodiments related to the method, and will not be elaborated here.

[0064] It should be noted that the various implementable ways in this embodiment can be implemented separately or in any combination manner without conflict. The present application makes no limitation.

[0065] The embodiment of the present application also provides a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, it can implement the operations corresponding to any of the methods in the method for determining the decision-making strategy based on the operations research decision-making large model provided in the above embodiments.

[0066] The embodiment of the present application also provides a computer program product. The computer program product includes a computer program, and when the computer program is executed by a processor, it can implement the operations corresponding to any of the methods in the method for determining the decision-making strategy based on the operations research decision-making large model provided in the above embodiments.

[0067] As Figure 4 As shown, some embodiments of the present application provide an electronic device 400, which includes: a memory 410, a processor 420, and a computer program stored on the memory 410 and executable on the processor 420. When the processor 420 reads the program from the memory 410 through the bus 430 and executes the program, it can implement the methods of any of the embodiments included in the above method for determining the decision-making strategy based on the operations research decision-making large model.

[0068] The processor 420 can process digital signals and can include various computing architectures. For example, a complex instruction set computer architecture, a reduced instruction set computer architecture, or an architecture that implements a combination of multiple instruction sets. In some examples, the processor 420 can be a microprocessor.

[0069] The memory 410 can be used to store instructions executed by the processor 420 or data related to the instruction execution process. These instructions and / or data can include code for implementing some or all of the functions of one or more modules described in the embodiments of the present application. The processor 420 of the embodiments of the present disclosure can be used to execute the instructions in the memory 410 to implement the methods shown above. The memory 410 includes dynamic random access memory, static random access memory, flash memory, optical memory, or other memories well known to those skilled in the art.

[0070] The above are only embodiments of the present application and are not used to limit the protection scope of the present application. For those skilled in the art, various changes and modifications can be made to the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application. It should be noted that similar reference numerals and letters indicate similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings.

[0071] As mentioned above, the above are only the specific implementation manners of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or replacements within the technical scope disclosed by the present application, and all should be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0072] It should be noted that in this document, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "including a..." does not exclude the existence of additional identical elements in the process, method, article or device including the said element.

Claims

1. A method for determining a decision strategy based on a large operations research decision model, characterized in that: The method comprises: Acquiring decision demand information, wherein the decision demand information at least includes a decision problem type; According to the pre-trained operational decision model, a decision strategy corresponding to the decision requirement information is determined, including: Using a large language model, the decision requirement information is parsed to obtain a decision problem type corresponding to the decision requirement information, and constraint conditions, objectives, and resource query information corresponding to the decision problem type; Determine corresponding resource configuration information according to the resource query information; Determine, through a decision routing model, a decision sub-model corresponding to the decision requirement information according to the resource configuration information and the decision problem type; The decision strategy corresponding to the decision requirement information is determined through the decision sub-model; wherein the pre-trained operations research decision model at least includes the decision routing model and a decision sub-model connected to the decision routing model, and the decision routing model is used to determine the decision sub-model corresponding to the decision problem type; the decision sub-model is used to determine the decision strategy corresponding to the decision requirement information.

2. The method for determining a decision strategy based on a large operations research decision model according to claim 1, characterized in that: The pre-trained operations research decision-making model is constructed based on the MoE structure.

3. The method for determining a decision strategy based on a large operations research decision model according to claim 1, characterized in that: The pre-trained operations research decision model is obtained in the following way: Get sample data; Processing the sample data to obtain training data corresponding to the sample decision problem type, wherein the sample decision problem type at least includes a route planning operations research problem or an inventory management operations research problem; Training the large model according to the training data corresponding to the sample decision problem type to obtain a decision sub-model corresponding to the sample decision problem type; Determining the decision routing model according to the decision sub-model corresponding to the sample decision problem type; The operations research decision model is determined according to the decision routing model and the decision sub-model.

4. The method for determining a decision strategy based on a large operations research decision model according to claim 3 is characterized in that: The processing of the sample data includes: When the number of bytes of the sample data is greater than a preset value, the MLA technology is used to compress the redundant information in the sample data to obtain processed data.

5. The method for determining a decision strategy based on a large operations research decision model according to claim 3, characterized in that: The method further comprises: The number of the decision sub-models is adjusted according to the number of different types of sample decision problems.

6. A device for determining a decision strategy based on a large operational decision model, characterized in that: The device comprises: An acquisition module, used to acquire decision demand information, wherein the decision demand information at least includes a decision problem type; The decision module is used to determine the decision strategy corresponding to the decision demand information according to the pre-trained operational decision model, including: Using a large language model, the decision requirement information is parsed to obtain a decision problem type corresponding to the decision requirement information, and constraint conditions, objectives, and resource query information corresponding to the decision problem type; Determine corresponding resource configuration information according to the resource query information; Determine, through a decision routing model, a decision sub-model corresponding to the decision requirement information according to the resource configuration information and the decision problem type; The decision strategy corresponding to the decision requirement information is determined through the decision sub-model; wherein the pre-trained operations research decision model at least includes the decision routing model and a decision sub-model connected to the decision routing model, and the decision routing model is used to determine the decision sub-model corresponding to the decision problem type; the decision sub-model is used to determine the decision strategy corresponding to the decision requirement information.

7. The device for determining decision strategies based on the large operational decision model according to claim 6, characterized in that: The pre-trained operations research decision-making model is constructed based on the MoE structure.

8. The device for determining decision strategies based on the large operational decision model according to claim 7, characterized in that: The device also includes a training module, which is used to: Get sample data; Processing the sample data to obtain training data corresponding to the sample decision problem type, wherein the sample decision problem type at least includes a route planning operations research problem or an inventory management operations research problem; Training the large model according to the training data corresponding to the sample decision problem type to obtain a decision sub-model corresponding to the sample decision problem type; Determining the decision routing model according to the decision sub-model corresponding to the sample decision problem type; The operations research decision model is determined according to the decision routing model and the decision sub-model.

9. An electronic device, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor can implement the method for determining a decision strategy based on a large operations research decision model as described in any one of claims 1 to 5 when executing the program.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which is characterized in that when the program is executed by a processor, it can implement the method for determining a decision strategy based on an operations research decision model as described in any one of claims 1 to 5.

Images