Business model construction method and device

Abstract

The invention provides a business model construction method and device, and the method comprises the steps: obtaining a target problem class and structure information of a to-be-constructed or to-be-reconstructed business model, and the structure information indicates N function modules of the to-be-constructed or to-be-reconstructed business model; based on the model structure information and a model library corresponding to the target problem class, constructing a plurality of equivalent models corresponding to the to-be-constructed or to-be-reconstructed business model; determining a target equivalent model from the plurality of equivalent models, wherein the target equivalent model is the equivalent model with the highest solving efficiency in the plurality of equivalent models; and determining a target business model based on the operator subset corresponding to the target equivalent model, the target business model being used for inputting a solver to obtain a corresponding decision when the target business achieves the optimal business target. According to the method, the solving optimization type business model with the aim of shortening the solving time is automatically constructed, so that the solving efficiency of the business model is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, and particularly relates to a method and device for constructing a business model. BACKGROUND

[0002] Operations research mainly uses mathematical methods to study the optimization approach and scheme of various systems, and provides a scientific decision basis for decision makers. Mathematical programming is an important branch of operations research, which mainly studies the optimal solution that can minimize or maximize a certain function in a given region. The application of mathematical programming is extremely common. According to the differences in the nature of the problem and the processing method, mathematical programming can be divided into many different branches, such as linear programming, integer programming, nonlinear programming, combinatorial optimization, multi-objective programming, stochastic programming, dynamic programming, parameter programming and the like. These problems have a large number of applications in many fields such as supply chain, finance, transportation and communication. In order to solve the mathematical programming problem, the industry develops special software systems for linear, integer, mixed integer and various nonlinear programming models, namely mathematical programming solvers.

[0003] There are usually a large number of operations and planning problems in real industrial scenarios that can be modeled as MILP problems. Solving these MILP problems through the branch and bound algorithm described above can often create huge commercial value. However, real-scale MILP problems often need to consider super-large-scale variables and constraints, and have a very high difficulty in solving. However, in real application scenarios, there is often a high requirement for the solving efficiency of the MILP model, and there is a contradiction between the existing MILP modeling scheme and the real solving efficiency requirement. SUMMARY

[0004] Embodiments of the present application provide a method and device for constructing a business model, which constructs a solution-friendly business model and improves the solving efficiency of the business model.

[0005] Firstly, this application provides a method for constructing a business model. The method includes obtaining model information of a business model to be constructed or optimized. The model information includes the target problem class and structural information of the business model to be constructed or reconstructed. The target problem class indicates the category of the target business problem to be optimized by the business model to be constructed or reconstructed. The structural information indicates the N functional modules of the business model to be constructed or reconstructed, where N is a positive integer greater than 1. Based on the model structural information and the model library corresponding to the target problem class, multiple equivalent models corresponding to the business model to be constructed or reconstructed are constructed. The model library includes multiple operators corresponding to each of the M functional modules, where M is a positive integer greater than or equal to N. A target equivalent model is determined from the multiple equivalent models. The target equivalent model is the equivalent model with the highest solution efficiency among the multiple equivalent models. Based on the set of operators corresponding to the target equivalent model, a target business model is determined. The target business model is used as input to a solver to obtain the decision corresponding to the target business when achieving the optimal business objective.

[0006] The business model construction method provided in this application first atomically decomposes the existing business models (e.g., existing expert models) of each problem class to obtain the model library corresponding to each problem class. Using the structure of the business model to be built or rebuilt, multiple equivalent models of the business model to be built or rebuilt are combined from the model library corresponding to its problem class. Then, the target equivalent model with the highest solution efficiency is selected from the multiple equivalent models. Finally, a solution-friendly target business model is constructed based on the target equivalent model to improve the solution efficiency of the target business model.

[0007] In one possible implementation, a specific approach to constructing multiple equivalent models based on model structure information and the model library corresponding to the target problem class is as follows: determine the representation vector of the model structure information; determine the representation graph corresponding to the model library, which includes M layers of nodes and connecting edges. The M layers of nodes correspond to M functional modules, and each layer of nodes includes multiple nodes, which represent multiple operators. Connecting edges connect nodes with dependencies; construct multiple equivalent models based on the representation vector and the representation graph.

[0008] By combining the representation vectors of the business model to be built or rebuilt and the representation graph of the model library, multiple equivalent models of the business model to be built or rebuilt are obtained, so that the optimal equivalent model with the best solution efficiency can be obtained from these multiple equivalent models in the future.

[0009] In another possible implementation, the M functional modules include a basic model class functional module, a target function class functional module, and a constraint class functional module; the basic model class functional module is configured to define the decision variables and the virtual target function, the target function class functional module is configured to define the optimization target of the target service, and the constraint class functional module is configured to constrain the decision variables; each layer of nodes corresponding to the target function class functional module and the constraint class functional module in the representation graph includes a null node; each layer of nodes in the representation graph includes a null node; and based on the representation vector and the representation graph, a specific implementation of constructing the multiple equivalent models is as follows: based on the representation vector, a first functional module and a second functional module in the M functional modules are determined, the first functional module is a functional module possessed by the target service model, and the second functional module is a functional module not possessed by the target service model; based on the dependency relationship between the nodes, the null node corresponding to the second functional module, and the node corresponding to the first functional module, multiple paths are generated by using a preset path generation algorithm, and the multiple paths all pass through the null node corresponding to the second functional module; and based on the operator set corresponding to the nodes passed by each path in the multiple paths, the multiple equivalent models are constructed.

[0010] For example, the structure of the service model to be constructed or reconstructed is parsed into a series of on or off of functional modules, the functional module is on (i.e., the target service model has the functional module), and the functional module is off (i.e., the target service model does not have the functional module), and the representation vector is obtained in this way. The layer corresponding to the index with a value of 0 in the representation vector is fixed as a null node, and multiple s-t paths are generated by using a path generation algorithm (for example, depth-first traversal). For any path, the modeling codes corresponding to the nodes in the path are combined in a string format, and a modeling API is called to construct an equivalent model in combination with the service data.

[0011] The application represents the relationship of the operators in the model library through a general representation graph structure, the representation graph structure takes equivalent local models as nodes, displays the division of problem atoms, and uses paths to represent the complete equivalent model of the target service model, and has the advantages of compactness, high storage efficiency, strong completeness, and easy expansion.

[0012] In another possible implementation, a specific implementation of determining a target equivalent model from the multiple equivalent models is as follows: the multiple equivalent models are converted to a target feature space to obtain multiple feature points corresponding to the multiple equivalent models; a solver is called to solve the equivalent models corresponding to part of the feature points to obtain the solving efficiency of the equivalent models corresponding to the part of the feature points; the multiple feature points are optimized based on the solving efficiency of the equivalent models corresponding to the part of the feature points and a preset optimization algorithm to obtain a target feature point, and the solving efficiency of the equivalent model corresponding to the target feature point is optimal; and the equivalent model corresponding to the target feature point is taken as the target equivalent model.

[0013] The solving efficiency of the partial equivalent model is obtained by multiple invocations of the solver, and then a preset optimization algorithm (also referred to as a black-box optimization algorithm) is used to dynamically predict an equivalent model that is easier to solve, and the target equivalent model with the highest solving efficiency is searched from the target feature space.

[0014] Optionally, the preset optimization algorithm can be a rule-based optimization algorithm or a black-box optimization algorithm. The rule-based optimization algorithm can be a heuristic optimization algorithm, and the black-box optimization algorithm can be a simulated annealing algorithm, an evolutionary algorithm, or a Bayesian optimization algorithm.

[0015] In another possible implementation, based on the operator set corresponding to the target equivalent model, determining a specific implementation of the target business model includes: obtaining business data of the target business; determining parameters of each operator in the operator set corresponding to the target equivalent model based on the business data; and determining the target business model based on the operator set corresponding to the target equivalent model and the parameters of each operator.

[0016] The equivalent model with the highest solving efficiency is obtained from the equivalent model pool by the optimization algorithm, and then the target business model with the highest solving efficiency is constructed using the equivalent model, which greatly improves the solving efficiency of the solver for the target business model.

[0017] In another possible implementation, there are various ways to obtain the structure information of the business model to be constructed or reconstructed. For example, one way to obtain the structure information of the business model to be constructed is to analyze the problem description input by the user to obtain the structure information of the business model to be constructed, and the problem description includes various optimization objectives and various constraint conditions of the target business.

[0018] For example, an input box is set in the interactive page, so that the user can input the problem description of the business optimization problem through the input box, such as various optimization objectives of the business optimization problem and constraint conditions of the business optimization problem, and then the problem description input by the user is analyzed by a natural language processing tool to obtain the structure information of the business model to be constructed. In this way, the dependence on mathematical programming modeling experience is avoided, the operation threshold is reduced, and the intelligent modeling and solving process is facilitated.

[0019] Another way to obtain the structure information of the business model to be constructed is to determine the structure information of the business model to be constructed based on configuration parameters of the business model to be constructed, and the configuration parameters indicate N functional modules configured by the business model to be constructed.

[0020] Configuration parameters can be obtained by parsing a configuration file, or they can be generated based on user interaction data with the configuration page. For example, the interactive page can have a configuration parameter import control. Users can directly import configuration text by clicking this control. Parsing this text yields N functional modules for the business model to be built, thus obtaining the configuration parameters for the business model. In another example, the interactive page can offer multiple functional modules for the user to choose from. The user selects the necessary functional modules for the business model, and the configuration parameters for the business model are generated based on the user's selection.

[0021] Another way to obtain the structural information of the business model to be reconstructed is to determine the structural information of the business model based on the feature extraction results obtained by the feature extraction model. The feature extraction results include the representation vector of the business model to be reconstructed, which indicates the structural information of the business model to be reconstructed. For example, the business model to be reconstructed is used as the input of the feature extraction model, and the output is the representation vector of the business model to be reconstructed, which indicates the structural information of the business model to be reconstructed.

[0022] Optionally, the feature extraction model is a graph neural network. The graph neural network extracts features from the business model to be reconstructed, directly obtaining the representation vector of the business model structure to be reconstructed, which facilitates the workflow of connecting to the solver of the existing business model.

[0023] In another possible implementation, the method for constructing the business model provided in this application further includes: de-identifying the target business model; and storing the de-identified target business model in a modeling scheme library for the target problem class. Thus, when faced with a similar problem-solving task again, the optimal modeling scheme can be retrieved from the scheme library and the problem reconstructed for solving, thereby improving the solver's efficiency.

[0024] Optionally, the target business model is a mixed-integer programming model.

[0025] In another possible implementation, the multiple operators corresponding to each of the M functional modules in the model library are obtained by decomposing the existing business model corresponding to the target problem class. The existing business model includes the expert business model corresponding to the target problem class and / or the historical target business model.

[0026] The system automatically breaks down expert business models from domain literature into multiple local models and inputs them into the model library, so that the optimal modeling scheme (i.e. the final construction scheme of the target business model) can grow with the academic community over a long period of time.

[0027] Secondly, this application also provides a method for determining the optimization strategy of a business. The method includes obtaining a target business model, which is used to optimize the target business. The target business model is constructed based on the business model construction method described in the first aspect. The target business model is solved by a solver to obtain the optimal solution of the target business model. Based on the optimal solution of the target business model, the optimization strategy of the target business is determined. The optimization strategy indicates the decision made by the target business when achieving the optimal business objective.

[0028] The method for determining the optimization strategy of the business provided in this application automatically constructs a target business model for solving optimization by adopting the business model construction method provided in the first aspect, thereby improving the solver's efficiency in solving the target business model and thus improving the efficiency of determining the optimization strategy of the business.

[0029] Thirdly, this application also provides a business model construction apparatus, which includes a first acquisition module, a construction module, a first determination module, and a second determination module. The first acquisition module is used to acquire model information of the business model to be constructed or reconstructed. This model information includes the target problem class and structural information of the business model to be constructed or reconstructed. The target problem class indicates the category of the target business problem to be optimized by the business model to be constructed or reconstructed, and the structural information indicates the N functional modules of the business model to be constructed or reconstructed, where N is a positive integer greater than 1. The construction module is used to determine the model based on the model structural information and the target problem... The model library corresponds to the problem type. It constructs multiple equivalent models corresponding to the business model to be constructed or reconstructed. The model library includes multiple operators corresponding to each of the M functional modules. The M functional modules include N functional modules, where M is a positive integer greater than or equal to N. The first determination module is used to determine the target equivalent model from the multiple equivalent models. The target equivalent model is the equivalent model with the highest solution efficiency among the multiple equivalent models. The second determination module is used to determine the target business model based on the set of operators corresponding to the target equivalent model. The target business model is used as input to the solver to obtain the decision corresponding to the target business when achieving the optimal business objective.

[0030] In another possible implementation, the building module is specifically used for: determining the representation vector of the model structure information; determining the representation graph corresponding to the model library, the representation graph including M layers of nodes and connecting edges, the M layers of nodes respectively corresponding to the M functional modules, each layer of nodes including multiple nodes, the multiple nodes representing the multiple operators, and the connecting edges connecting nodes with dependencies; and constructing multiple equivalent models based on the representation vector and the representation graph.

[0031] In another possible implementation, the M functional modules include a basic model class functional module, an objective function class functional module, and a constraint class functional module. The basic model class functional module defines the decision variables and the virtual objective function, the objective function class functional module defines the optimization objective of the target business, and the constraint class functional module constrains the decision variables. Each layer of nodes corresponding to the objective function class functional module and the constraint class functional module in the representation graph includes empty nodes. A specific implementation of constructing multiple equivalent models based on the representation vector and the representation graph is as follows: Based on the representation vector, a first functional module and a second functional module are determined among the M functional modules. The first functional module is a functional module possessed by the target business model, and the second functional module is a functional module not possessed by the target business model. Based on the dependencies between nodes, the empty nodes corresponding to the second functional module, and the nodes corresponding to the first functional module, multiple paths are generated using a preset path generation algorithm. All multiple paths pass through the empty nodes corresponding to the second functional module. Based on the set of operators corresponding to the nodes passed through by each path in the multiple paths, multiple equivalent models are constructed.

[0032] In another possible implementation, the first determining module is specifically used for: transforming multiple equivalent models into a target feature space to obtain multiple feature points corresponding to the multiple equivalent models; calling a solver to solve the equivalent models corresponding to some of the multiple feature points to obtain the solution efficiency of the equivalent models corresponding to some feature points; optimizing the multiple feature points based on the solution efficiency of the equivalent models corresponding to some feature points and a preset optimization algorithm to obtain target feature points, wherein the solution efficiency of the equivalent models corresponding to the target feature points is optimal; and using the equivalent models corresponding to the target feature points as the target equivalent models.

[0033] In another possible implementation, the second determining module is specifically used to: acquire business data of the target business; determine the parameters of each operator in the operator set corresponding to the target equivalent model based on the business data; and determine the target business model based on the operator set corresponding to the target equivalent model and the parameters of each operator.

[0034] In another possible implementation, the structural information of the business model to be built is obtained by parsing the problem description input by the user. The problem description includes various optimization objectives and various constraints of the target business.

[0035] In another possible implementation, the structural information of the business model to be built is determined based on the configuration parameters of the business model to be built, which indicate the N functional modules configured in the business model to be built.

[0036] In another possible implementation, the configuration parameters are obtained by parsing the configuration file of the business model to be built; or, the configuration parameters are generated based on the interaction data between the user and the configuration page, which includes information on multiple functional modules of the target business model entered by the user through the configuration page.

[0037] In another possible implementation, the structural information of the business model to be reconstructed is determined based on the feature extraction results obtained by the feature extraction model. The feature extraction results include the representation vector of the business model to be reconstructed, which indicates the structural information of the business model to be reconstructed.

[0038] In another possible implementation, the business model construction apparatus provided in this application further includes a desensitization module, which is used to desensitize the target business model and store the desensitized target business model in a modeling scheme library for the target problem class.

[0039] In another possible implementation, the target business model is a mixed-integer programming model.

[0040] In another possible implementation, the multiple operators corresponding to each of the M functional modules in the model library are obtained by decomposing the existing business model corresponding to the target problem class. The existing business model includes the expert business model corresponding to the target problem class and / or the historical target business model.

[0041] Fourthly, this application also provides a device for determining an optimization strategy for a business. The device includes a second acquisition module, a solution module, and a third determination module. The second acquisition module acquires a target business model, which is used to optimize the target business. The target business model is constructed based on the business model construction method described in the first aspect. The solution module solves the target business model using a solver to obtain the optimal solution of the target business model. The third determination module determines an optimization strategy for the target business based on the optimal solution of the target business model. The optimization strategy indicates the decision made by the target business when achieving the optimal business objective.

[0042] Fifthly, embodiments of this application provide a computing device, including a memory and a processor, wherein the memory stores instructions that, when executed by the processor, cause the methods described in the first aspect and / or the second aspect to be implemented.

[0043] In a sixth aspect, embodiments of this application provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, causes the methods described in the first aspect and / or the second aspect to be implemented.

[0044] In a seventh aspect, embodiments of this application also provide a computer program or computer program product, the computer program or computer program product including instructions that, when executed, cause a computer to perform the methods described in the first aspect and / or the second aspect.

[0045] Eighthly, embodiments of this application also provide a chip including at least one processor and a communication interface, the processor being configured to perform the methods described in the first aspect and / or the second aspect. Attached Figure Description

[0046] Figure 1 A schematic diagram illustrating a process for optimizing business operations is shown.

[0047] Figure 2 The illustration shows an application scenario diagram of the business model construction method provided in the embodiments of this application;

[0048] Figure 3 A system architecture diagram of a cloud platform is shown;

[0049] Figure 4 The implementation architecture diagram of a business model construction method provided in this application is shown;

[0050] Figure 5 A schematic diagram of the structure of a characterization diagram is shown;

[0051] Figure 6 A schematic diagram illustrating the implementation process of a Bayesian black-box optimization algorithm is shown.

[0052] Figure 7 A flowchart illustrating a method for constructing a business model provided in an embodiment of this application;

[0053] Figure 8 A flowchart illustrating a method for determining a service optimization strategy provided in an embodiment of this application;

[0054] Figure 9 A schematic diagram of a business model construction apparatus provided in an embodiment of this application;

[0055] Figure 10 A schematic diagram of a service optimization strategy determination device provided in an embodiment of this application;

[0056] Figure 11 A schematic diagram of the structure of a computing device provided in an embodiment of this application. Detailed Implementation

[0057] The term "and / or" used in this article describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent three cases: A alone, A and B simultaneously, and B alone. The symbol " / " in this article indicates that the related objects are in an "or" relationship; for example, A / B means A or B.

[0058] The terms "first" and "second," etc., used in the specification and claims herein are used to distinguish different objects, not to describe a specific order of objects. For example, "first type of functional module" and "second type of functional module," etc., are used to distinguish different functional modules, not to describe a specific order of functional modules.

[0059] In the embodiments of this application, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design that is described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design. Specifically, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0060] In the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more, for example, multiple processing units means two or more processing units, multiple elements means two or more elements, etc.

[0061] To facilitate understanding of the solutions in the embodiments of this application, the technical terms involved in this document will be explained first.

[0062] Operations research and optimization: Operations research is a newly emerging interdisciplinary field that developed in the 1930s and 40s. It mainly studies the use and planning of various resources by humans, aiming to maximize the benefits of limited resources and achieve the overall optimal goal under certain constraints.

[0063] Mathematical programming is a branch of mathematics that studies how to find the optimal solution to a given function that minimizes or maximizes it. Mathematical programming can be divided into many different branches, such as linear programming, integer programming, nonlinear programming, combinatorial optimization, multi-objective programming, stochastic programming, dynamic programming, and parametric programming.

[0064] Mixed integer linear programming (MILP) is a linear programming problem in which some decision variables are restricted to integers.

[0065] Mathematical programming solver: A software system that solves existing linear, integer, mixed integer, and various nonlinear programming models.

[0066] Equivalent model: A mathematical programming model that describes the same optimization problem but has a different form but yields the same solution.

[0067] Black-box optimization: When the intermediate process from decision-making to the final result in an optimization problem cannot be modeled in detail, it is called a black-box problem. Optimization methods that do not rely on process information for this type of problem are called black-box optimization.

[0068] Solving mixed-integer programming problems is the core of mathematical programming solvers. In many real-world scenarios, problems always have integer constraints, such as production scheduling, supply chain problems, production planning, and factory location problems. These problems can usually be modeled as mixed-integer programming problems. The essence of solving mixed-integer programming problems is to obtain the optimal solution by solving a series of linear programming problems. Branch and bound is one of the most commonly used algorithms for solving integer programming problems, and almost all mathematical programming solvers rely on this algorithm framework. It is a search and iterative method that branches by selecting different branch variables and subproblems. Typically, the entire feasible solution space is repeatedly divided into increasingly smaller subsets, called branching; and a target lower bound (for minimum problems) is calculated for the solution set within each subset, called bounding. Simultaneously, branch and bound uses different heuristics to search for feasible integer solutions within the subsets, thus obtaining a target upper bound. When the difference between the target upper and lower bounds obtained in the above search and iteration process converges to 0, it means that the feasible integer solution obtained in the current search is the global optimal solution to the original problem.

[0069] In real-world industrial scenarios, numerous operations research and planning problems can be modeled as MILP problems. Solving these MILP problems using the branch-and-bound algorithm described above can often generate significant commercial value. However, real-scale MILP problems often require consideration of extremely large numbers of variables and constraints, making them very difficult to solve. For example, in real-world power systems, the security-constrained unit commitment (SCUC) problem is modeled as an MILP problem and solved using commercial branch-and-bound solvers, saving approximately $5 billion annually in the US alone. However, real-scale SCUC problems need to consider hundreds or thousands of thermal generators and transmission systems, as well as tens of thousands of buses and lines. Real-world business requirements demand that the algorithm provide a high-quality, feasible solution within 10-15 minutes, placing extremely high demands on the problem-solving efficiency.

[0070] Besides improving the efficiency of algorithm solutions, the modeling of MILP problems also has a significant impact on the difficulty of solving the problem. By adjusting the way variables and constraints are constructed in the MILP model, the "tightness" of the MILP model can be changed, thereby reducing the cost of branch and bound tree search and achieving higher solution efficiency. Therefore, improving the modeling of MILP problems is a very important research direction.

[0071] The modeling methods in related technologies have some problems. For example, one related technology, the modeling optimization method based on expert experience, has the core idea of ​​using the experience of operations research experts to establish a standardized mathematical model. It mainly includes three steps: Step S1, inputting the problem description and user data of the MILP problem; Step S2, constructing a standardized mathematical model of the MILP problem based on expert experience, including variables, objective function, and constraints to be satisfied; Step S3, calling a mathematical programming solver to solve the established model and obtain the solution result. However, it has certain drawbacks. For example, this approach is highly dependent on expert knowledge and can generally only provide one or a few modeling schemes for trial, making it difficult to guarantee the optimal or high-quality modeling scheme within a limited time. Furthermore, this approach lacks interaction with the downstream mathematical programming solver; it only outputs the final modeling scheme through a single construction by the expert, making it impossible to make targeted adjustments to the modeling based on solver feedback. In addition, this approach has a serious problem: expert experience in a single domain cannot be reused in other domains, which greatly increases the cost of modeling and optimizing MILP problems in different domains.

[0072] To address this, this application provides a method for constructing a business model. Based on the general representation of mathematical programming problems, the structure of the variant of the problem to be optimized is decomposed. A large number of equivalent models are constructed using the local models provided by expert models. Then, a black-box optimization algorithm is used to find the model with the shortest solution time, thereby improving the solution efficiency of the called solver for the variant of the problem.

[0073] The technical solutions provided in the embodiments of this application will be further described in detail below with reference to the accompanying drawings and examples.

[0074] Figure 1 This illustrates a process diagram for optimizing business operations. For example... Figure 1 As shown, the actual business data is first modeled into a business model, and then the solver is called to solve the business model to obtain the optimal business decision, that is, the optimal solution for the business.

[0075] Generally, before calling the solver, it is necessary to translate the actual business problem requirements into a business model using expert experience; this process is called modeling. Modeling mainly involves modeling the optimization objective and various constraints. Common business models include linear programming models, mixed integer programming models, integer programming models, nonlinear programming models, and combinatorial optimization models. The quality of the modeling scheme directly affects the solver's efficiency; that is, the quality of the modeled business model directly affects the subsequent solver efficiency. The business model construction method provided in this application can be applied to... Figure 1 The process of building a business model can automatically construct a solution-friendly business model or optimize an existing business model to obtain a solution-friendly business model, thereby improving the efficiency of subsequent solvers in solving the business model.

[0076] It should be explained that the evaluation index of the solver's efficiency in solving the business model can be the solution time when the solver obtains the optimal solution to the business model. In other words, the shorter the solution time of the solver, the higher the efficiency of the solver in solving the business model. Alternatively, the evaluation index of the solver's efficiency in solving the business model can also be the accuracy that the solver can achieve in solving the business model within a fixed time (such as 1 minute). That is, the higher the accuracy that the solver can achieve in solving the business model within a fixed time, the higher the efficiency of the solver in solving the business model.

[0077] Figure 2 This illustration shows an application scenario diagram of the business model construction method provided in the embodiments of this application. For example... Figure 2As shown, the business model construction system 100, which applies the business model construction method provided in this application embodiment, is deployed on a cloud platform. It provides a solution-friendly business model construction service to users (also known as tenants) in the form of a cloud service. Users can remotely access the cloud platform through a client and invoke the solution-friendly business model construction service. For example, a user inputs data and requirements through the client's interactive interface. This data and requirements indicate a business optimization problem. The client sends the user-input data to the business model construction system 100 deployed on the cloud platform. The business model construction system 100 constructs a solution-friendly business model based on the user-input data and then returns the constructed solution-friendly business model to the user. Alternatively, it can invoke a solver to solve the constructed solution-friendly business model, obtain the solution result, and return the solution result to the user. In another example, users can also input an existing business model through the client's interactive interface. The client sends the user-input business model to the business model building system 100 deployed on the cloud platform. The business model building system 100 optimizes the business model, reconstructs a solver-friendly business model, and then returns the constructed solver-friendly business model to the user. Alternatively, it can call the solver to solve the constructed solver-friendly business model, obtain the solution result, and return the solution result to the user.

[0078] It should be noted that, Figure 2 This application merely illustrates one scenario of the business model construction method provided in this embodiment and does not constitute a limitation on the embodiments of this application. The business model construction method provided in this embodiment can also have other application scenarios. For example, the business model construction system can be deployed on a physical server, and users can access the business model construction system deployed on the physical server through a client deployed on a terminal device to construct a solver-friendly business model. Another example is deploying the business model construction system on a physical server, and users can access the business model construction system deployed on the physical server through a webpage to construct a solver-friendly business model. Yet another example is directly deploying the business model construction system on a terminal device, and users can directly complete the construction of a solver-friendly business model on the terminal.

[0079] The business model construction method provided in this application embodiment can be applied to constructing any type of solution-friendly business model, such as linear programming models, mixed integer programming models, integer programming models, nonlinear programming models, and combinatorial optimization models. In particular, it can be applied to the construction of mixed integer programming models, constructing solution-friendly mixed integer programming models, or optimizing existing mixed integer programming models. The optimized mixed integer programming models are easier to solve, improving the solver's efficiency.

[0080] The following example uses a mixed-integer programming model as the business model and the business model construction system deployed on a cloud platform to illustrate the specific implementation of the business model construction method provided in this application.

[0081] Figure 3 A system architecture diagram of a cloud platform is shown. This cloud platform deploys a business model building system based on the business model building method provided in the application embodiments, offering users a solution-friendly model optimization service. Users can purchase this service to automatically build a solution-friendly business model each time a business optimization problem is solved, thereby improving the efficiency of business model solving. For example, a user uploads data via a client on a terminal device. This data can be a description of the business optimization problem and business data, or a pre-built business model. The client uploads this data to the cloud, and the cloud platform calls the solution-friendly model building service (which, for uploaded data that is a pre-built business model, can also be called the solution-friendly model optimization service) to build a solution-friendly business model. Then, it calls the solver service to solve the built solution-friendly business model, quickly obtaining the solution result, which is returned to the user, allowing the user to optimize their business based on the solution result.

[0082] like Figure 3 As shown, both the solver service and the solver-friendly model optimization service can be deployed on the operations research optimization platform provided by the cloud service company. The model optimization service decomposes the expert models in the literature database into multiple local models according to their functional models in the cloud. These multiple local models form a model library. The user-input optimization problem is parsed to obtain the structural information of the model to be built. Based on the structural information and the model library, multiple equivalent models are built. Then, the black-box optimization engine calls the mathematical programming solver multiple times to find the optimal solution and obtain the equivalent model with the highest solution efficiency. The business model is built based on this equivalent model, and the resulting model is stored in the cloud as the solution for the problem class.

[0083] When a user uploads a model and uses the solver service, the model optimization service is provided as an optional feature. When a user selects to use the model optimization service, if a solution is available in the cloud, the solution is read, the model is rebuilt, and then the solver is called to solve the rebuilt model and return the solution result to the user. If no solution is available in the cloud, the model to be optimized (also known as the model to be optimized) is passed in. The model to be optimized is parsed to obtain its structure. Based on this structure, local models are extracted from the model library stored on the platform for optimization to obtain the equivalent model with the highest solution efficiency. The user-uploaded model is rebuilt based on this equivalent model, and then the solver is called to solve the rebuilt model and return the solution result to the user.

[0084] In another example, users can also input a description of the problem to be optimized, such as describing several objectives and constraints of the business to be optimized in natural language. They can then upload this description to the modeling and solving services, with the model optimization service offered as an optional feature. When a user chooses to use the model optimization service, if a modeling solution for this type of optimization problem exists in the cloud, the solution is read and used to build the model. The solver is then called to solve the modeled, solution-friendly model and returns the result to the user. If no solution exists in the cloud, the problem description is parsed to obtain the structure of the model to be built. Based on this structure, local models are extracted from the model library stored on the platform for optimization to obtain the most efficient equivalent model. This equivalent model is then used to model the user-uploaded optimization problem, resulting in a solution-friendly model. Finally, the solver is called to solve the built solution-friendly model and returns the result to the user.

[0085] The categories of problems to be optimized and their corresponding modeling schemes are anonymized and stored on the operations research optimization platform, where they are uniformly stored and managed.

[0086] For example, by decomposing expert models for various problem classes (such as production scheduling optimization, supply chain optimization, production dispatching optimization, factory location optimization, SCUC optimization, etc.) in the literature library, equivalent and modular local model structures are derived, forming a model library for the corresponding problem class. Simultaneously, a general representation structure for the corresponding problem class is constructed. The literature library includes existing publicly available expert models for a specific problem class. The library expands as existing expert models are added, including newly added ones. Optionally, the literature library can be updated in real-time or periodically, including newly added expert models. The model library is updated using the local models derived from these new expert models to ensure its freshness. The business model construction system first parses the problem class to which the user-input optimization problem belongs, and then parses it into a series of functional modules being on or off. The structure of the business model to be constructed is represented as a representation vector, resulting in the structural representation vector of the business model to be constructed. Then, multiple equivalent models are generated based on the structural representation vector of the business model to be constructed, the general representation structure of the problem class, and the model library. Then, the mixed integer programming solver is called to evaluate the solution time of the model. Using a black-box optimization algorithm, the modeling scheme that is easier to solve is dynamically predicted based on the historical solution records. The modeling scheme with high solution efficiency is efficiently searched from the equivalent model space, and then a good solution business model is constructed, which is to obtain the optimal modeling scheme of the user input problem to be optimized.

[0087] Optionally, the solution-friendly business model for the problem class can be anonymized, and then a storage service (such as OBS) can be called to store the anonymized business model in the solution library corresponding to the problem class. This way, when a similar problem is encountered again, the optimal modeling solution can be retrieved from the solution library corresponding to the problem class to increase the solution efficiency of the business model.

[0088] The business model construction method provided in this application decomposes an equivalent, modular local model structure from a mixed-integer model (expert model) of a certain problem class from literature, which is then used to assemble a large number of equivalent models for that problem class. By repeatedly calling the solver to obtain the solution time of some equivalent models, a black-box optimization algorithm can be used to efficiently search for the model with the highest solution efficiency, and this modeling scheme is entered into a scheme pool. When faced with a similar problem-solving task again, the optimal scheme can be retrieved from the scheme pool and the problem can be reconstructed for solving, thereby improving the solution efficiency of the solver.

[0089] Figure 4 This diagram illustrates the implementation architecture of a business model construction method provided in this application. Figure 4 As shown, the implementation architecture mainly includes four processing stages: expert model input stage, user input parsing stage, equivalent model generation stage, and modeling black-box optimization stage. In the expert model input stage, existing expert models are decomposed into local models according to functional modules, and a model library is built. Simultaneously, a general representation structure for the corresponding problem class is constructed. User input parsing determines the problem class to which the user-inputted problem belongs and parses it into a series of functional modules being on or off, representing the structure of the business model to be built as a representation vector. In the equivalent model generation stage, based on the representation vector obtained from the user input parsing module, multiple equivalent models are generated using the representation graph of the corresponding problem class and the model library. The modeling black-box optimization stage calls a mixed-integer programming solver to evaluate the model's solution time. Using a black-box optimization algorithm, it dynamically predicts easier-to-solve modeling schemes based on historical solution records, efficiently searching for high-efficiency modeling schemes from the equivalent model space.

[0090] In one example, the expert model input stage is completed offline, while the other three stages—user input parsing, equivalent model generation, and modeling black-box optimization—are completed online. This means that for user input data, only three steps—user input parsing, equivalent model generation, and modeling black-box optimization—are performed online to complete the construction of a solution-friendly business model.

[0091] In some other examples, the expert model entry phase can also be completed online.

[0092] The main purpose of the expert model input stage is to expand the local model library using expert models from the literature, so that equivalent models with different structures can be combined in subsequent processes (here, equivalent models with different structures refer to the same functional module using different operators). During the expert model input process, local models divided by functional modules are decomposed from expert models belonging to the same optimization problem class, entered into the model library (also called the local model library), and a general representation structure for the problem class is constructed.

[0093] Optionally, the general representation structure of the problem class can be a representation graph. For example, for the problem class of power scenario safety constraint unit combination, a representation graph is constructed based on the operator dependency relationship of each functional module in the model library of the SCUC problem class. This representation graph is the representation graph of the SCUC problem class.

[0094] Figure 5 A schematic diagram of a characterization diagram is shown. For example... Figure 5 As shown, the representation graph includes multiple nodes and connecting edges. The nodes are arranged in layers, with each layer corresponding to a functional module in the model library of the target problem class. These functional modules include three categories: the basic model (P), which includes the definition of decision variables and a virtual objective function. The latter allows multiple optional objective functions to be defined in the form of pure constraints; the objective function (O), which is the scalar that the optimization problem attempts to maximize (minimize); and the constraints (C), which are the specific rules that the solution to the optimization problem must obey. Each node in each layer corresponds to an equivalent model of that functional module, stored in the format of modeling API code, including an empty node. This means the functional module corresponding to this layer is not active. There are two types of connection edges: one type, represented by dashed lines, is used to connect nodes in two layers that have no dependency relationship; the other type, represented by solid lines, is used to connect nodes in two layers that have a dependency relationship. For example... Figure 5 As shown, Figure 5 The dashed edges represent optional connection methods between nodes, and a path from virtual node s to t constitutes an equivalent model for a variation of this problem. In some problems, dependencies may exist between functional modules; for example, the way decision variables are defined may determine the way certain constraints are defined. These can be represented by specifying strong connections (e.g., ...). Figure 5 (Solid line edge).

[0095] Of course, in some other examples, the representation graph of the model library may include only one type of connection edge. The connection edge only connects the nodes corresponding to operators with dependencies, indicating that there is a dependency relationship between the two operators, which is a strong connection relationship. The operators corresponding to the nodes without connection edges have no dependency relationship.

[0096] The expert model input stage can be completed in several ways. One exemplary method is manual input. For instance, experienced modelers can categorize expert models from a literature database, such as by application scenario into different problem categories, including but not limited to production scheduling optimization, supply chain optimization, production dispatching optimization, factory location optimization, and SCUC optimization. Then, the expert models for each problem category are functionally broken down into multiple functional modules. For example, implementing a single objective function, defining a type of constraint, and defining decision variables and virtual objective functions can all be separated into functional modules. These modules are then organized and input into the model database. For instance, functional modules from each expert model can be arranged in a row, and identical functional modules from different expert models can be arranged in a column. Another method is to use natural language processing tools. For example, a large language model can automatically scan, identify, classify, split, and organize expert models before inputting them into the corresponding problem category model database.

[0097] This application does not impose specific limitations on the method of expert model input, and appropriate expert model input methods can be selected according to the actual situation.

[0098] In the user input parsing stage, the problem to be optimized is parsed from the user input and its corresponding problem class and the structural information of the business model. For example, the functional modules of the business model. The structural information of the business model is represented in the form of a representation vector, which facilitates subsequent calculations.

[0099] Users can input the problem to be optimized in several ways. One way is to directly input a description of the problem in natural language. For example, if the problem is classified as an SCUC optimization problem, with the optimization objectives being minimum generation cost and minimum start-up cost, and the unit state variables including **, power variables including **, reserve constraint variables including **, generator generation limit constraints being **, ramp constraints being **, start-up and shutdown constraints being **, and power flow constraints being **, then a natural language processing tool is used to process and identify the problem category and the structural information of the business model to be built. This structural information is represented as a representation vector, indicating the functional modules enabled in the business model, i.e., the functional modules that the business model possesses. Optionally, the representation vector can be a fixed-dimensional binary representation vector.

[0100] This input method is more user-friendly for those who lack sufficient knowledge of mathematical programming and modeling, thus lowering the barrier to entry.

[0101] Another input method is to input the configuration file. This method is for users with some mathematical programming modeling knowledge. Users construct a configuration file for the problem to be optimized. This configuration file includes configuration parameters for the business model to be built. These parameters indicate the functional modules of the business model and the problem class to which the optimization problem belongs. Based on the configuration parameters, the representation vector of the business model and the problem class are determined. For example, the client's interface has a configuration file upload control; users can upload the configuration file for the problem to be optimized by clicking this control.

[0102] Another input method is through direct interaction with the configuration interface. For example, a configuration interface for the optimization model to be built is provided. This interface offers multiple problem classes as options. The user selects the problem class to which the problem to be optimized belongs, and then the page redirects to the corresponding functional module configuration interface. This interface provides M functional module options for the model of that problem class. The user selects N functional module options required for the actual business optimization based on actual business needs. Then, the representation vector of the business model to be built is generated based on the user's selected functional modules. For example, if the business problem to be optimized is a production scheduling optimization problem, the M functional modules for production scheduling optimization in the model library include the basic model P, objective functions O1, O2, and O3, and constraints C1, C2, C3, C4, C5, and C6. Based on actual business needs, the target business model that the user needs to build should include the following functional modules: basic model P, objective function O1, objective function O3, and constraints C1, C4, C5, and C6. The user can then select the production scheduling optimization problem from the problem class options provided in the configuration interface, and jump to the corresponding functional module configuration interface. From the options of basic model P, objective function O1, objective function O2, objective function O3, and constraints C1, C2, C3, C4, C5, and C6, the user can select the following: basic model P, objective function O1, objective function O3, and constraints C1, C4, C5, and C6. Then, based on the interaction data between the user and the configuration interface, a representation vector of the business model to be built is generated as 110100111, where 1 represents the functional modules possessed by the business model and 0 represents the functional modules not possessed by the business model. From the representation vector 110100111, it can be seen that the functional modules possessed by the business model to be built are the basic model P, objective functions O1 and O3, and constraints C1, C4, C5, and C6. The functional modules not possessed are objective function O2 and constraints C2 and C3. In this way, the problem class to which the business model to be built belongs and the structural information of the business model to be built (i.e., the functional modules possessed by the business model) can be directly obtained through the interaction data between the user and the configuration interface. This makes it convenient for users to directly define the problem class to which the business model to be built belongs and the functional modules it has, which is more user-friendly.

[0103] Another input method is for users to directly upload pre-modeled business models. For example, a user may have a completed business model of the problem to be optimized, but the modeling scheme is not user-friendly for the solver, resulting in a long solution time and requiring optimization. For instance, the client interface provides a control to upload the business model to be optimized. The user clicks this control to upload the business model's .mps or .lp file. The .mps or .lp file is then parsed to obtain the problem category to which the business model belongs and its structural representation vector. For example, the business model can be used as input to a model classifier, outputting its problem category; or as input to a feature extractor, outputting its structural representation vector, which indicates the functional modules of the business model.

[0104] Optionally, the model classifier can be a first graph neural network, and the feature extractor can be a second graph neural network. After training, the first graph neural network can realize the mapping between the business model and the problem class to which it belongs, and after training, the second graph neural network can realize the mapping between the business model and the structural representation vector.

[0105] The equivalent model generation stage mainly involves dynamically constructing multiple equivalent models from the model library based on the structural representation vectors of the business model to be built or optimized. For example, firstly, the layers corresponding to indices with values ​​of 0 in the representation vector are fixed as empty nodes, and multiple paths from s to t are generated using a path generation algorithm (such as depth-first traversal). One of these paths from s to t is... Figure 5 The path formed by the nodes traversed by the middle arrow. Then, the modeling code corresponding to the nodes in the path each day is merged into a string format, and combined with business data, the modeling API is called to build an equivalent model. In this way, multiple equivalent models are generated that are equivalent to the business model to be built or the business model to be optimized. Multiple equivalent models form an equivalent model pool.

[0106] During the black-box optimization phase of modeling, the equivalent model with the best solution efficiency is selected from the equivalent model pool consisting of multiple equivalent models generated in the equivalent model generation phase.

[0107] Due to the complexity of solving mathematical programming problems, it is often difficult to explicitly characterize the solution efficiency of a mathematical programming model. This means that the efficiency function of different modeling schemes is a typical black-box function. Therefore, after constructing the equivalent model pool, black-box optimization algorithms can be used to search for the optimal modeling scheme for a given mathematical programming solver.

[0108] For example, one process in the black-box optimization stage of modeling is as follows: The equivalent model pool is constructed as a discrete high-dimensional parameter space (i.e., the target feature space), where each parameter point (also called a feature point) represents a complete modeling scheme combining different modeling methods. Based on this, the solver is invoked to evaluate the model solving efficiency, and a preset optimization algorithm is used to search for the optimal modeling scheme by optimizing the parameters in the aforementioned high-dimensional parameter space. Optionally, the preset optimization algorithm can be a rule-based optimization algorithm or a black-box optimization algorithm. Rule-based optimization algorithms can be, for example, heuristic optimization algorithms, while black-box optimization algorithms can be, for example, simulated annealing, evolutionary algorithms, or Bayesian optimization algorithms.

[0109] Taking a Bayesian optimization-based black-box optimization algorithm as an example, the optimization process for the modeling scheme is as follows: Figure 6 As shown:

[0110] 1) Initial random data collection: First, a batch of different modeling schemes are randomly constructed from the equivalent model pool (i.e., a batch of parameter points are randomly sampled from the high-dimensional parameter space to obtain the modeling schemes corresponding to the high parameter points), and a complete business model is constructed based on the business data. Subsequently, the mathematical programming solver is called to solve the obtained business model, and the solution efficiency corresponding to different random modeling schemes is used as its evaluation index to form a modeling scheme-solution efficiency data pair.

[0111] 2) Fitting a surrogate model: Based on this, the algorithm fits a surrogate model (e.g., a parameterized model such as a Gaussian process, random forest, or neural network) according to the collected modeling scheme-solution efficiency data. This surrogate model can fit the mapping from the modeling scheme to the solution efficiency.

[0112] 3) Maximizing the Acquisition Function: Subsequently, the algorithm recommends a new batch of sampling points, i.e., a new batch of modeling schemes, based on the pre-designed acquisition function and the principle of maximizing the acquisition function value. The acquisition function is used to predict a batch of possible optimal parameter points among the unsampled parameter points based on the modeling scheme-solution efficiency data pairs corresponding to the sampled parameter points. The modeling scheme corresponding to the optimal parameter point is the one with the best solution efficiency. In other words, the role of the acquisition function is to recommend the next more likely modeling scheme to bring benefits based on the currently sampled modeling schemes and their solution efficiency. Alternatively, it can be said that the acquisition function predicts the possible optimal parameter points among the remaining unsampled parameter points based on the changing trends guided by the surrogate model, i.e., predicting the equivalent model with the best solution efficiency from the remaining equivalent models in the equivalent model pool.

[0113] Optionally, the input to the acquisition function is the modeling scheme-solution efficiency data pair corresponding to the acquired parameter points, and the output is several optimal parameter points among the unsampled parameter points.

[0114] Understandably, by designing different acquisition functions, users can control the balance between optimizing sampling point performance and exploring the uncertainty of the surrogate model.

[0115] 4) Evaluating the Black-Box Function: Finally, the mathematical programming solver is invoked to evaluate the black-box function of the modeling scheme's solution efficiency (i.e., the surrogate model). This involves evaluating the solution efficiency of the complete business model constructed using the equivalent model and business data collected by the acquisition function on the solver. In other words, the mathematical programming solver is invoked to solve the business model constructed based on the new equivalent model and business data collected by the acquisition function, obtaining the solution efficiency of the newly acquired modeling scheme. The parameters of the surrogate model are adjusted based on the evaluation results until the optimal modeling scheme is found. For example, the solver solves the modeling scheme collected by the acquisition function, obtaining the solution efficiency of that scheme. The parameters of the surrogate model are adjusted based on the newly acquired modeling scheme-solution efficiency data. The acquisition function performs prediction sampling based on the updated surrogate model. If the sampled data is the same as the previous one, the sampled modeling scheme is taken as the optimal modeling scheme; otherwise, the iteration continues.

[0116] By repeating the above process, we can avoid the inefficient exhaustive search of the high-dimensional parameter space (i.e., the equivalent model pool) and instead efficiently explore the optimal modeling scheme for the solver.

[0117] Figure 7 This is a flowchart illustrating a method for constructing a business model according to an embodiment of this application. This method can be executed by any device, equipment, platform, or cluster of devices with computing capabilities. This application embodiment does not specifically limit the specific computing device executing the method; a suitable computing device can be selected as needed. For example, it can be implemented on a terminal device, meaning the business model construction system is deployed on the terminal device; it can also be implemented on both the terminal device and a cloud device (e.g., a cloud-side server), i.e., using an end-to-cloud collaborative architecture; or it can be completed on a cloud device, providing users with a cloud service for constructing or optimizing a solver-friendly business model. For ease of description, the form of the executing entity will not be distinguished in the following text; all will be described as a business model construction system. Figure 7 As shown, the business model construction method provided in this application embodiment includes at least steps S701 to S704.

[0118] In step S701, the model information of the business model to be built or optimized is obtained.

[0119] The model information of the business model to be built or optimized (hereinafter referred to as the model for convenience) includes the target problem class to which the model belongs and the structural information. The target problem class indicates the category of the target business problem that the model is to optimize, and the structural information indicates the functional modules that the model has. For example, if the model to be built needs to have a basic model P, objective functions O1 and O2, and constraints C1, C2, C3, C4, and C5, then the structural information of the model to be built is that the functional modules of the model to be built need to include the basic model P, objective functions O1 and O2, and constraints C1, C2, C3, C4, and C5.

[0120] Optionally, the structural information of the model can be expressed as a representation vector. For details on how to express the representation vector, please refer to the description of the representation vector above. For the sake of brevity, it will not be repeated here.

[0121] A business model building system can obtain the target problem class and structural information of the model to be built or optimized in various ways. For example, it can parse the user-input description of the business problem to be optimized to obtain the target problem class and structural information of the model. Alternatively, it can parse the user-input configuration file to obtain the target problem class and structural information of the model. Or, it can obtain the target problem class and structural information of the model based on the interaction data between the user and the configuration interface. As another example, a graph neural network can be used to infer the target problem class and structural information of the business model file uploaded by the user.

[0122] The user input of the problem description, configuration parameters, interaction with the configuration interface, and specific implementation of the business model to be optimized can be found in the description above, and will not be repeated here for the sake of brevity.

[0123] In step S702, based on the model structure information and the model library corresponding to the target problem class, multiple equivalent models corresponding to the business model to be built or optimized are constructed.

[0124] After obtaining the structural information and target problem class of the business model to be built or optimized, the business model building system retrieves the model library corresponding to the target problem class. For example, if the target problem class of the business model to be built or optimized is a supply chain optimization problem, the business model building system calls the model library corresponding to the supply chain optimization problem. This module library includes multiple local models obtained by decomposing the expert model corresponding to the supply chain optimization problem. By decomposing multiple expert models into multiple atomic functional modules according to functional modules, it is convenient to combine multiple equivalent models of the target problem class in the future. Then, the system seeks to find the equivalent model with the highest solution efficiency and uses this equivalent model to build the target business model.

[0125] It is understandable that the functional modules of multiple expert models are atomically decomposed into operators. In other words, the model library includes multiple operators corresponding to each functional module. For example, there are N operators obtained by decomposing N expert models under functional module 1, and similarly, there are multiple operators obtained by decomposing multiple expert models under other functional modules.

[0126] By using the structural representation vector of the business model to be built or optimized and the representation graph of the model library corresponding to the target problem class, multiple equivalent models are constructed, and these multiple equivalent models form an equivalent model pool for the business model to be built or optimized.

[0127] Optionally, multiple equivalent models can be represented as a set of operators. For example, if a model library has 8 functional modules, such as functional module 1, functional module 2, functional module 3, functional module 4, functional module 5, functional module 6, functional module 7, and functional module 8, and each functional module has multiple operators, and the model to be constructed or optimized has 6 functional modules, namely functional module 1, functional module 3, functional module 4, functional module 5, functional module 6, and functional module 8, then one of the equivalent models is functional module 1 - operator 1, functional module 3 - operator 3, functional module 4 - operator 2, functional module 5 - operator 7, functional module 6 - operator 6, and functional module 8 - operator 9.

[0128] The model library and its representation graph are constructed offline. Local models in the library can be obtained by decomposing expert models from existing literature or existing business models. Optionally, the model library can be updated in real-time as expert models are added, or periodically, for example, weekly, searching for new expert models and decomposing them to expand the library with the resulting local models. Understandably, as the model library is updated, the corresponding representation graph will also be updated accordingly.

[0129] The construction methods of the model library and the specific construction methods of the representation graph are described above, and will not be repeated here for the sake of brevity.

[0130] In step S703, the target equivalent model is determined from multiple equivalent models.

[0131] A black-box optimization algorithm is used to find the most efficient equivalent model from the pool of equivalent models, and this model is then used as the target equivalent model for the business model to be built or optimized. The black-box optimization algorithm avoids the inefficient exhaustive search of the equivalent model pool, instead efficiently exploring the optimal modeling scheme for the solver. For example, the target equivalent model with the highest solution efficiency obtained from multiple equivalent models using the black-box optimization algorithm is: Functional Module 1 - Operator 3, Functional Module 3 - Operator 2, Functional Module 4 - Operator 1, Functional Module 5 - Operator 7, Functional Module 6 - Operator 6, and Functional Module 8 - Operator 9.

[0132] For details on the specific black-box optimization algorithm, please refer to the description above. For the sake of brevity, it will not be repeated here.

[0133] In step S704, the target business model is determined based on the set of operators corresponding to the target equivalent model.

[0134] Through the above steps, the equivalent model is obtained by finding the optimal combination of the model library. Then, by combining the actual business data, the parameters of each operator in the equivalent model are determined, and the target business model with the highest solution efficiency is constructed. The solver is then called to solve the target business model, and the optimal solution of the target business model can be obtained quickly. The target business can be optimized using the optimal solution.

[0135] In another example, the constructed solution-friendly target business model can be stored in the modeling scheme library of the target problem class. When a similar business optimization problem is encountered again, the target business model can be directly called from the modeling scheme library. The parameters of the target business model can be fine-tuned by combining actual business data, so as to quickly obtain the optimal modeling scheme and save computing power.

[0136] The following practical example illustrates the implementation process of the business model construction method provided in this application embodiment in a specific business optimization scenario.

[0137] This application takes the power scenario safety constraint unit combination problem as an example to introduce the detailed implementation of this application embodiment and the beneficial effects that can be achieved.

[0138] The SCUC problem is a core component of daily generation planning in power grid operation and dispatch. The SCUC problem determines the start-up, shutdown, and output combinations of generating units, and its optimization result is also a key input for intraday economic dispatch, playing a crucial role in the safe operation of the power grid and the structural optimization of generation resources. Mathematically, the SCUC problem is a non-convex, strongly coupled, large-scale mixed-integer optimization problem. Typically, a single round of optimization takes less than 15 minutes to solve, with a gap of less than 0.01%, severely testing the capabilities of modeling and solution algorithms.

[0139] SCUC (Solution-Based Unified Cost) minimizes electricity costs through generator combination while satisfying power security-related constraints. There are approximately a dozen types of constraints in SCUC, including inter-system constraints such as power balance, current grid power flow constraints, and cross-sectional constraints; and generator-related constraints such as ramping constraints, power output limits, and start-stop constraints. The following describes the details of SCUC problem modeling and optimization implementation:

[0140] Step 0: Expert model input, specifically including:

[0141] 1) Construct the power grid configuration file;

[0142] 2) Input the power grid configuration file and use the Python modeling tool EGRET for SCUC problems to generate 14 SCUC expert models.

[0143] Step 1: Scene model analysis, specifically including:

[0144] Based on the modeling principles of expert models, the SCUC model is decomposed into nine modules: unit state variables, power variables, reserve constraint variables, generator generation limit constraints, ramping constraints, start-up and shutdown constraints, power flow constraints, start-up cost (optimization objective), and generation cost (optimization objective). Each expert model contains these nine functional modules, but the modeling methods for each module differ between expert models.

[0145] Step 2: Constructing the equivalent model, specifically including:

[0146] 1) For each functional module, the methods used by the expert model are marked to form the model operator set for that functional module;

[0147] 2) Label the model operator dependency graph to form the operator feasible region. If using functional module 2-model operator 1 requires using functional module 1-model operator 3, then add a directed edge from functional module 1-model operator 3 to functional module 2-model operator 1.

[0148] 3) Input the power grid configuration file from step 0, the generated dependency graph, and the specific method labels for all functional modules. Then, use the modeling tool EGRET to generate an equivalent model.

[0149] Step 3: Black-box optimization modeling, specifically including:

[0150] 1) Set the objective of the black-box optimization acquisition function to minimize the solution time;

[0151] 2) Based on the dependency graph, design an operator dependency-oriented surrogate function that assigns a large penalty coefficient when the generated model violates the dependency.

[0152] 3) Input the power grid configuration file, operator dependency graph, and model operator tags, select the solver, use the black-box optimization algorithm to sample the model operators, and combine them to form the SCUC model;

[0153] 4) Output the operator marker parameters that achieve the highest solution efficiency, and generate the corresponding model based on the operator markers.

[0154] Experimental Design: To verify the effectiveness of the business model construction method provided in this application, this experiment uses the Hebo black-box parameter tuning tool as a black-box optimization tool and the Tianchou solver as a solution-friendly mixed integer programming solver. The test platform is Ubuntu 18.04.4, Intel(R) Xeon(R) Gold6244 CPU @ 3.60GHz, 8 threads.

[0155] Dataset: The publicly available power grid dataset RTS area was used as the prototype for the basic power grid topology and generator attributes. Fourteen SCUC expert models compiled by EGRET were imported as the expert model library. These 14 expert models also served as the comparison scheme for this experiment. The evaluation metrics were the relative gap (gap) between the feasible solution and the dual solution after a fixed solution time of 3600 seconds, and the solution time up to a given gap of 0.01%.

[0156] Results: Through operator combinations in the model operator library generated by scenario analysis, 14 expert models can theoretically generate over 100,000 equivalent models. Using a black-box optimization method, only 240 models were generated to find an equivalent model with significantly improved solution efficiency, which is 8.64 times faster than the minimum solution time of the expert model.

[0157] This application also provides a method for determining a service optimization strategy.

[0158] Figure 8 This is a flowchart illustrating a method for determining a business optimization strategy, provided in an embodiment of this application. This method can be executed by any device, equipment, platform, or cluster of devices with computing capabilities. This application embodiment does not specifically limit the specific computing device executing the method; a suitable computing device can be selected as needed. For example, it can be implemented on a terminal device, meaning the business optimization strategy determination system is deployed on the terminal device; it can also be implemented on both the terminal device and a cloud device (e.g., a cloud-side server), i.e., using an end-to-cloud collaborative architecture; or it can be completed on a cloud device, providing the business optimization strategy determination service to users in the form of a cloud service. For ease of description, the form of the executing entity will not be distinguished in the following text; all instances will be described as a business optimization strategy determination system. Figure 8As shown, the service optimization strategy determination method provided in this application embodiment includes at least steps S801 to S803.

[0159] In S801, obtain the target business model.

[0160] The target business model is a solution-friendly business model constructed using the business model construction method provided in this application embodiment. The target business model is uploaded to the business optimization strategy determination system, which receives the uploaded target business model and obtains the target business model.

[0161] In S802, the target business model is solved by the solver to obtain the optimal solution of the target business model.

[0162] The business optimization strategy determines the system call the solver to solve the target business model, quickly obtaining the optimal solution of the target business model, that is, the optimal solution of each decision variable in the target business model.

[0163] In S803, the optimization strategy for the target business is determined based on the optimal solution of the target business model.

[0164] The optimal solution of the target business model, i.e., the values ​​of each decision variable in the target business model, is used to determine the optimization strategy for the target business. This optimization strategy is then applied to optimize the target business and improve its efficiency.

[0165] Based on the same concept as the aforementioned embodiment of a business model construction method, this application also provides a business model construction apparatus 900. This apparatus 900 can be deployed in a terminal device or cloud server to construct a solution-friendly business model or optimize an existing business model, thereby improving the solution efficiency of the optimized business model. The business model construction apparatus 900 includes components for implementing... Figures 4-7 The units or modules of each step in the construction method of the business model shown.

[0166] Figure 9 This is a schematic diagram of a business model construction apparatus provided in an embodiment of this application. Figure 9As shown, the business model construction device 900 includes at least a first acquisition module 901, a construction module 902, a first determination module 903, and a second determination module 904. The first acquisition module 901 acquires model information of the business model to be constructed or optimized. This model information includes the target problem class and structural information of the business model. The target problem class indicates the category of the target business problem to be optimized by the business model, and the structural information indicates the N functional modules of the business model, where N is a positive integer greater than 1. The construction module 902 is used to acquire model information based on the model structure information and... The model library corresponding to the target problem class constructs multiple equivalent models corresponding to the business model to be constructed or optimized. The model library includes multiple operators corresponding to each of the M functional modules, where M is a positive integer greater than or equal to N. The first determining module 903 is used to determine the target equivalent model from the multiple equivalent models. The target equivalent model is the equivalent model with the highest solution efficiency among the multiple equivalent models. The second determining module 904 is used to determine the target business model based on the set of operators corresponding to the target equivalent model. The target business model is used as input to the solver to obtain the decision corresponding to the target business when achieving the optimal business objective.

[0167] In another possible implementation, the construction module 902 is specifically used for: determining the representation vector of the model structure information; determining the representation graph corresponding to the model library, the representation graph including M layers of nodes and connecting edges, the M layers of nodes respectively corresponding to the M functional modules, each layer of nodes including multiple nodes, the multiple nodes representing the multiple operators, and the connecting edges connecting nodes with dependencies; and constructing multiple equivalent models based on the representation vector and the representation graph.

[0168] In another possible implementation, the M functional modules include a basic model class functional module, an objective function class functional module, and a constraint class functional module. The basic model class functional module defines the decision variables and the virtual objective function, the objective function class functional module defines the optimization objective of the target business, and the constraint class functional module constrains the decision variables. Each layer of nodes corresponding to the objective function class functional module and the constraint class functional module in the representation graph includes empty nodes. A specific implementation of constructing multiple equivalent models based on the representation vector and the representation graph is as follows: Based on the representation vector, a first functional module and a second functional module are determined among the M functional modules. The first functional module is a functional module possessed by the target business model, and the second functional module is a functional module not possessed by the target business model. Based on the dependencies between nodes, the empty nodes corresponding to the second functional module, and the nodes corresponding to the first functional module, multiple paths are generated using a preset path generation algorithm. All multiple paths pass through the empty nodes corresponding to the second functional module. Based on the set of operators corresponding to the nodes passed through by each path in the multiple paths, multiple equivalent models are constructed.

[0169] In another possible implementation, the first determining module 903 is specifically used to: transform multiple equivalent models into a target feature space to obtain multiple feature points corresponding to the multiple equivalent models; call a solver to solve the equivalent models corresponding to some of the multiple feature points to obtain the solution efficiency of the equivalent models corresponding to some feature points; optimize the multiple feature points based on the solution efficiency of the equivalent models corresponding to some feature points and a preset optimization algorithm to obtain target feature points, and the solution efficiency of the equivalent models corresponding to the target feature points is optimal; and use the equivalent models corresponding to the target feature points as the target equivalent models.

[0170] In another possible implementation, the second determining module 904 is specifically used to: acquire business data of the target business; determine the parameters of each operator in the operator set corresponding to the target equivalent model based on the business data; and determine the target business model based on the operator set corresponding to the target equivalent model and the parameters of each operator.

[0171] In another possible implementation, the structural information of the business model to be built is obtained by parsing the problem description input by the user. The problem description includes various optimization objectives and various constraints of the target business.

[0172] In another possible implementation, the structural information of the business model to be built is determined based on the configuration parameters of the business model to be built, which indicate the N functional modules configured in the business model to be built.

[0173] In another possible implementation, the configuration parameters are obtained by parsing the configuration file of the business model to be built; or, the configuration parameters are generated based on the interaction data between the user and the configuration page, which includes information on multiple functional modules of the target business model entered by the user through the configuration page.

[0174] In another possible implementation, the structural information of the business model to be optimized is determined based on the feature extraction results obtained by the feature extraction model. The feature extraction results include the representation vector of the business model to be optimized, which indicates the structural information of the business model to be optimized.

[0175] In another possible implementation, the business model construction apparatus 900 provided in this application further includes a desensitization module 905, which is used to desensitize the target business model and store the desensitized target business model in a modeling scheme library for the target problem class.

[0176] In another possible implementation, the target business model is a mixed-integer programming model.

[0177] In another possible implementation, the multiple operators corresponding to each of the M functional modules in the model library are obtained by decomposing the existing business model corresponding to the target problem class. The existing business model includes the expert business model corresponding to the target problem class and / or the historical target business model.

[0178] The business model construction apparatus 900 provided according to the embodiments of this application can correspond to the execution of the methods described in the embodiments of this application, and the above and other operations and / or functions of each module in the business model construction apparatus 900 are respectively for implementing Figures 4-7 For the sake of brevity, the corresponding processes of each method in the code will not be elaborated here.

[0179] Based on the same concept as the aforementioned embodiment of a service optimization strategy determination method, this application also provides a service optimization strategy determination device 1000. This device 1000 can be deployed in a terminal device or a cloud server to quickly determine the service optimization strategy. The service optimization strategy determination device 1000 includes components for implementing... Figure 8 The units or modules in the method for determining the optimization strategy of the business shown.

[0180] Figure 10 This is a schematic diagram of a service optimization strategy determination device provided in an embodiment of this application. Figure 10 As shown, the optimization strategy determination device 1000 for a certain type of business includes at least a second acquisition module 1001, a solution module 1002, and a third determination module 1003. The second acquisition module 1001 is used to acquire a target business model, which is used to optimize the target business. The target business model is constructed based on the business model construction method provided in the embodiments of this application. The solution module 1002 is used to solve the target business model using a solver to obtain the optimal solution of the target business model. The third determination module 1003 is used to determine the optimization strategy of the target business based on the optimal solution of the target business model. The optimization strategy indicates the decision corresponding to the target business when achieving the optimal business objective.

[0181] The construction apparatus 1000, which determines the service optimization strategy according to the embodiments of this application, can correspond to executing the method described in the embodiments of this application, and the above and other operations and / or functions of each module in the service optimization strategy determination apparatus 1000 are respectively for implementing Figure 8 For the sake of brevity, the corresponding processes of each method in the code will not be elaborated here.

[0182] This application embodiment also provides a computing device, including at least one processor, a memory, and a communication interface, wherein the processor is used to execute... Figures 1-8 The method described.

[0183] Figure 11 A schematic diagram of the structure of a computing device provided in an embodiment of this application.

[0184] like Figure 11 As shown, the computing device 1100 includes at least one processor 1101, a memory 1102, and a communication interface 1103. The processor 1101, memory 1102, and communication interface 1103 are communicatively connected, which can be achieved via a wired (e.g., bus) or wireless connection. The communication interface 1103 is used to send and / or receive data from other devices. The memory 1102 stores computer instructions, which the processor 1101 executes to perform the methods described in the preceding method embodiments, thereby constructing a solution-friendly business model and improving the efficiency of solving the business model.

[0185] It should be understood that, in the embodiments of this application, the processor 1101 may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0186] The memory 1102 may include read-only memory and random access memory, and provides instructions and data to the processor 1101. The memory 1102 may also include non-volatile random access memory.

[0187] The memory 1102 can be volatile memory or non-volatile memory, or it can include both. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous linked dynamic random access memory (SLDRAM), and direct rambus RAM (DR RAM).

[0188] It should be understood that the computing device 1100 according to the embodiments of this application can perform the implementation of the embodiments of this application. Figures 1-8 The method shown is described in detail above, and will not be repeated here for the sake of brevity.

[0189] Embodiments of this application provide a computer-readable storage medium having a computer program stored thereon, wherein when the computer instructions are executed by a processor, the aforementioned method is implemented.

[0190] An embodiment of this application provides a chip including at least one processor and an interface, wherein the at least one processor determines program instructions or data through the interface; the at least one processor is used to execute the program instructions to implement the method mentioned above.

[0191] Embodiments of this application provide a computer program or computer program product that includes instructions that, when executed, cause a computer to perform the methods mentioned above.

[0192] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0193] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented using hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0194] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of this application. It should be understood that the above description is only a specific embodiment of this application and is not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for constructing a business model, characterized in that, include: Obtain model information of the business model to be built or rebuilt. The model information includes the target problem class and structure information of the business model to be built or rebuilt. The target problem class indicates the category of the target business problem to be optimized by the business model to be built or optimized. The structure information indicates the N functional modules of the business model to be built or rebuilt, where N is a positive integer greater than 1. Based on the model structure information and the model library corresponding to the target problem class, multiple equivalent models corresponding to the business model to be built or rebuilt are constructed. The model library includes multiple operators corresponding to each of the M functional modules. The M functional modules include the N functional modules, and M is a positive integer greater than or equal to N. A target equivalent model is determined from the plurality of equivalent models, wherein the target equivalent model is the equivalent model with the highest solution efficiency among the plurality of equivalent models; Based on the set of operators corresponding to the target equivalent model, the target business model is determined. The target business model is used as input to the solver to obtain the decision corresponding to the target business when achieving the optimal business objective.

2. The method according to claim 1, characterized in that, Based on the model structure information and the model library corresponding to the target problem class, multiple equivalent models are constructed, including: Determine the representation vector of the model structure information; Determine the representation graph corresponding to the model library. The representation graph includes M layers of nodes and connecting edges. The M layers of nodes correspond to the M functional modules respectively. Each layer of nodes includes multiple nodes, which represent the multiple operators. The connecting edges connect nodes with dependencies. Based on the representation vector and the representation graph, the multiple equivalent models are constructed.

3. The method according to claim 2, characterized in that, The M functional modules include a basic model class functional module, an objective function class functional module, and a constraint class functional module; wherein, the basic model class functional module is used to define decision variables and virtual objective functions, the objective function class functional module is used to define the optimization objective of the target business, and the constraint class functional module is used to constrain the decision variables; Each layer of nodes corresponding to the objective function class functional module and the constraint class functional module in the representation diagram includes empty nodes; The construction of the multiple equivalent models based on the representation vector and the representation graph includes: Based on the representation vector, a first functional module and a second functional module are determined among the M functional modules. The first functional module is a functional module that the business model to be built or rebuilt has, and the second functional module is a functional module that the business model to be built or rebuilt does not have. Based on the dependencies between the nodes, the empty node corresponding to the second functional module, and the node corresponding to the first functional module, multiple paths are generated using a preset path generation algorithm, and all of the multiple paths pass through the empty node corresponding to the second functional module. Based on the set of operators corresponding to the nodes traversed by each of the multiple paths, the multiple equivalent models are constructed.

4. The method according to any one of claims 1-3, characterized in that, The step of determining the target equivalent model from the plurality of equivalent models includes: The multiple equivalent models are transformed into the target feature space to obtain multiple feature points corresponding to the multiple equivalent models; The solver is invoked to solve the equivalent model corresponding to some of the multiple feature points, and the solution efficiency of the equivalent model corresponding to the partial feature points is obtained. Based on the solution efficiency of the equivalent model corresponding to the aforementioned feature points and a preset optimization algorithm, the multiple feature points are optimized to obtain the target feature point, and the solution efficiency of the equivalent model corresponding to the target feature point is optimal. The equivalent model corresponding to the target feature point is taken as the target equivalent model.

5. The method according to any one of claims 1-4, characterized in that, The step of determining the target business model based on the set of operators corresponding to the target equivalent model includes: Obtain the business data of the target service; Based on the business data, determine the parameters of each operator in the operator set corresponding to the target equivalent model; Based on the set of operators corresponding to the target equivalent model and the parameters of each operator, the target business model is constructed.

6. The method according to any one of claims 1-5, characterized in that, The structural information of the business model to be constructed is obtained by parsing the problem description input by the user. The problem description includes various optimization objectives and various constraints of the target business.

7. The method according to any one of claims 1-5, characterized in that, The structural information of the business model to be built is determined based on the configuration parameters of the business model to be built, and the configuration parameters indicate the N functional modules configured in the business model to be built.

8. The method according to claim 7, characterized in that, The configuration parameters are obtained by parsing the configuration file of the business model to be built; Alternatively, the configuration parameters may be generated based on user interaction data with the configuration page, the interaction data including information on multiple functional modules of the business model to be built, input by the user through the configuration page.

9. The method according to any one of claims 1-5, characterized in that, The structural information of the business model to be reconstructed is determined based on the feature extraction results obtained by performing feature extraction on the business model to be reconstructed using a feature extraction model. The feature extraction results include the representation vector of the business model to be reconstructed, and the representation vector indicates the structural information of the business model to be reconstructed.

10. The method according to any one of claims 1-9, characterized in that, Also includes: The target business model is anonymized. The anonymized target business model is stored in a modeling scheme library for the target problem class.

11. The method according to any one of claims 1-10, characterized in that, The target business model is a mixed integer programming model.

12. The method according to any one of claims 1-11, characterized in that, The multiple operators corresponding to each of the M functional modules in the model library are obtained by decomposing the existing business model corresponding to the target problem class. The existing business model includes the expert business model corresponding to the target problem class and / or the historical target business model.

13. A method for determining an optimization strategy for a business, characterized in that, include: Obtain a target business model, the target business model being used to optimize a target business, the target business model being constructed based on the business model construction method according to any one of claims 1-12; The target business model is solved by a solver to obtain the optimal solution of the target business model; Based on the optimal solution of the target business model, an optimization strategy for the target business is determined, and the optimization strategy indicates the decision-making process for the target business to achieve the optimal business objective.

14. A business model construction apparatus, characterized in that, include: The first acquisition module is used to acquire model information of the business model to be built or rebuilt. The model information includes the target problem class and structure information of the business model to be built or rebuilt. The target problem class indicates the category of the target business problem to be optimized by the business model to be built or rebuilt. The structure information indicates the N functional modules of the business model to be built or rebuilt, where N is a positive integer greater than 1. The construction module is used to construct multiple equivalent models corresponding to the business model to be constructed or rebuilt based on the model structure information and the model library corresponding to the target problem class. The model library includes multiple operators corresponding to each of the M functional modules. The M functional modules include the N functional modules, and M is a positive integer greater than or equal to N. The first determining module is used to determine a target equivalent model from the plurality of equivalent models, wherein the target equivalent model is the equivalent model with the highest solution efficiency among the plurality of equivalent models; The second determining module is used to determine the target business model based on the set of operators corresponding to the target equivalent model. The target business model is used as input to the solver to obtain the decision corresponding to the target business when achieving the optimal business objective.

15. A device for determining optimization strategies for a business, characterized in that, include: The second acquisition module is used to acquire a target business model, which is used to optimize a target business. The target business model is constructed based on the business model construction method according to any one of claims 1-12. The solution module is used to solve the target business model through a solver to obtain the optimal solution of the target business model; The third determining module is used to determine the optimization strategy of the target business based on the optimal solution of the target business model. The optimization strategy indicates the decision corresponding to the target business when achieving the optimal business objective.

16. A computing device, comprising a memory and a processor, characterized in that, The memory stores instructions that, when executed by a processor, cause the method described in any one of claims 1-13 to be implemented.

17. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it causes the method as described in any one of claims 1-13 to be implemented.

18. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, causes the method as described in any one of claims 1-13 to be implemented.

Images