CN-122022608-A - Method and system for simulating and intervening growth of incumbent enterprises based on digital twin

CN122022608ACN 122022608 ACN122022608 ACN 122022608ACN-122022608-A

Abstract

The application relates to the technical field of digital twinning, in particular to a method and a system for simulating and intervening growth of an incubated enterprise based on digital twinning, wherein the method comprises the steps of constructing a digital twinning body of an incubated enterprise of a target and comprising an enterprise multisource state model and a mixed growth mechanism model; the mixed growth mechanism model is coupled with the system dynamics SD model and the intelligent agent basic modeling ABM model, and is provided with a mixed simulation scheduling mechanism in advance, simulation deduction is carried out based on a digital twin body, namely, the current state defined by the enterprise multisource state model is taken as an initial state, the preset mixed simulation scheduling mechanism is operated, the SD model and the ABM model are alternately operated, interaction is carried out through a dynamic calibration module, so that the prediction parameters of the SD model can be adjusted on line based on the result of the ABM model, and a personalized intervention strategy aiming at the target on-hatch enterprise is generated through reinforcement learning of the RL intelligent agent based on the deduction state. Real-time performance and accuracy of simulation deduction are guaranteed.

Inventors

ZHANG HONG
ZHU FA
ZHOU HAIPENG
Zhao Yanhan
DONG CHENCHEN
YANG CHUNLEI
Ren Anran

Assignees

山东浪潮创新创业科技有限公司

Dates

Publication Date: 20260512
Application Date: 20251203

Claims (10)

1. The digital twin-based method for simulating and intervening the growth of the incumbent enterprises is characterized by comprising the following steps: s1, constructing a digital twin body of an object-under-hatch enterprise, wherein the digital twin body comprises an enterprise multi-source state model constructed based on multi-source heterogeneous data and a mixed growth mechanism model, the mixed growth mechanism model is coupled with a system dynamics SD model and an intelligent body basic modeling ABM model, and a mixed simulation scheduling mechanism is preset; S2, carrying out simulation deduction based on the digital twin body, namely, taking the current state defined by an enterprise multisource state model as an initial state, running a preset hybrid simulation scheduling mechanism, enabling the SD model and the ABM model to run alternately and interact through a dynamic calibration module of a hybrid growth mechanism model, so that the prediction parameters of the SD model can be adjusted online based on the result of the ABM model; S3, generating a personalized intervention strategy aiming at the target on-hatch enterprise through reinforcement learning RL intelligent agent based on the deduction state of S2.
2. The method of claim 1, wherein the enterprise multisource state model is used as a unified state representation of the digital twin for the hybrid growth mechanism model to read in S2 for simulation deduction and as input to the state space S of the reinforcement learning RL agent in S3.
3. The method of claim 2, wherein in S1, the step of constructing a digital twin for the target incumbent enterprise comprises: S11, collecting multi-source heterogeneous data from enterprise internal systems, incubator operation data, industry chains, market data and dynamic public opinion data through an API (application program interface), an IoT (internet of things) sensor and a real-time data pipeline, converting all structured data into a uniform serialization format, and storing unstructured data and metadata thereof in an object storage service; S12, detecting and processing abnormal values of the acquired data, and adopting a condition generation countermeasure network CGAN to enhance and fill the missing data; S13, constructing an enterprise multisource state model by quantitatively defining N-dimensional health state vectors of the enterprise; s14, constructing a hybrid growth mechanism model by establishing and coupling a system dynamics SD model and an agent base modeling ABM model, wherein the SD model is used for describing a macroscopic feedback loop, and the ABM model is used for simulating microscopic agent behaviors.
4. The method of digital twinning-based on-incumbent enterprise growth simulation and intervention of claim 3, wherein the step of constructing the enterprise multisource state model in S13 comprises: s13a, quantitatively defining an N-dimensional health state vector of an enterprise, wherein each dimension of the vector represents a comprehensive health index; s13b, mapping a plurality of bottom layer original index data for each comprehensive health index; and S13c, calculating and obtaining the quantized score of each comprehensive health index by adopting a multi-index comprehensive evaluation algorithm based on the bottom layer original index data, thereby forming an enterprise health state vector.
5. The method for simulating and intervening growth of an incumbent enterprise based on digital twinning according to claim 4, wherein the multi-index comprehensive evaluation algorithm in S13c is an entropy weight TOPSIS method, and the step of calculating the quantized score comprises: constructing an evaluation matrix based on the M bottom layer original index data; Performing standardized processing on the evaluation matrix to eliminate the dimensional influence of the original indexes of different bottom layers; calculating objective weight of each item in the M bottom layer original indexes by using an entropy weight method; constructing a weighted standardized evaluation matrix based on the objective weight; Determining a positive ideal solution vector and a negative ideal solution vector, wherein the positive ideal solution vector is composed of optimal values of all hatched enterprises in the same period on corresponding bottom layer indexes, and the negative ideal solution vector is composed of worst values on the corresponding bottom layer indexes; Calculating the distance between each bottom index data of the enterprise in the weighted standardized evaluation matrix and the positive ideal solution vector and the negative ideal solution vector; And calculating the relative closeness of the enterprise and the positive ideal solution based on the distance, and normalizing to a [0,1] interval to serve as the quantized score.
6. The method of claim 5, wherein in S14, the specific steps of establishing and coupling the system dynamics SD model and the agent infrastructure ABM model include: S14a, defining and establishing a core model structure for describing an enterprise operation feedback loop by using a system dynamics modeling tool, wherein the core model structure at least comprises variables for representing enterprise core asset inventory, flow variables for controlling inventory change and feedback loops for connecting the variables; s14b, defining the classes of the creators, clients and competitors in the intelligent agent basic modeling environment, and programming each class of intelligent agents to realize the behavior rules for making decisions based on the environment states; And S14c, establishing a parallel and alternately executed hybrid simulation scheduling logic, and coupling the SD model with the ABM model, wherein the output of the SD model is used as the running environment parameter of the ABM model, and the emerging result of the ABM model is fed back to the SD model for dynamically calibrating the internal parameter of the SD model.
7. The method for simulating and intervening growth of an incumbent enterprise based on digital twinning according to claim 6, wherein in S2, the specific step of running a preset hybrid simulation scheduling mechanism comprises: s20, reading a current health state vector of the enterprise defined by the enterprise multi-source state model, and taking the current health state vector as an initial state of simulation; S21, when a first time period starts, running the SD model based on the initial state, and outputting environment prediction data; s22, in a first time period, operating the ABM model by taking the environment prediction data as input, simulating interaction of microcosmic agents, and developing microcosmic result data; S23, when a second time period starts, microscopic result data emerging from the ABM model is compared with corresponding environment prediction data made by the SD model when the first time period starts, and key parameters in the SD model are adjusted on line through the dynamic calibration module based on a difference value obtained by the comparison.
8. The digital twinning-based on-incumbent enterprise growth simulation and intervention method of claim 7, wherein generating personalized intervention strategies by reinforcement learning RL agents in S3 is based on a training-completed RL agent implementation, and the RL agent deployment and decision process comprises: s31, the RL agent reads the current state of the digital twin body from the deduction result of the S2, wherein the current state is formed by a health state vector defined by the enterprise multisource state model; S32, the RL agent inputs the read current state into an internal strategy network and outputs an action, wherein the action is an intervention strategy or strategy combination selected from a structured intervention strategy library; s33, applying the intervention strategy generated in the S32 to the digital twin body, deducing the influence of the selected strategy on the enterprise growth index in a short-term and long-term range by running a hybrid simulation scheduling mechanism, and calculating a reward signal by the RL intelligent body according to the deduction result and updating a strategy network according to the reward signal so as to optimize the subsequent decision.
9. The digital twinning-based on-incumbent enterprise growth simulation and intervention method of claim 8, wherein the RL agent is trained in a simulation environment by: s30a, defining a digital twin body comprising a mixed growth mechanism model as a reinforcement learning training environment; S30b, MDP modeling: defining a state space S, namely defining an N-dimensional health state vector by the enterprise multisource state model; defining an action space A, namely selecting a strategy identifier from a structured intervention strategy library; defining a reward function R, namely a multi-objective weighting function integrating the estimated increase, the survival probability and the intervention cost of enterprises; s30c, optimizing a PPO algorithm by adopting a near-end strategy, training the RL intelligent agent in a plurality of parallelized simulation environment copies, and learning an intervention strategy through state-action decision interaction; And S30d, regularly evaluating the strategy model in training on a fixed test set, and when the average accumulated rewards are converged and the strategy entropy is stable, selecting the model with the optimal performance to be deployed as the final RL intelligent agent.
10. A digital twinning-based on-incumbent enterprise growth simulation and intervention system for implementing the method of any of claims 1-9, the system comprising: the data acquisition and processing module is used for acquiring heterogeneous data from a multi-source channel and carrying out data cleaning, enhancement, entity analysis and knowledge graph construction; The digital twin body constructing and managing module is in communication connection with the data collecting and processing module and is used for constructing and maintaining a digital twin body of an object under incubation enterprise, and the digital twin body comprises an enterprise multisource state model and a mixed growth mechanism model; the hybrid simulation deduction engine is in communication connection with the digital twin body construction and management module and is used for alternately running the SD model and the ABM model based on a hybrid simulation scheduling mechanism preset by the digital twin body running and interacting through the dynamic calibration module; the intelligent decision and intervention module is in communication connection with the hybrid simulation deduction engine and is used for generating a personalized intervention strategy according to the simulation deduction state through a reinforcement learning RL intelligent body.

Description

Method and system for simulating and intervening growth of incumbent enterprises based on digital twin Technical Field The application relates to the technical field of digital twinning, in particular to a method and a system for simulating and intervening growth of an incumbent enterprise based on digital twinning. Background Growth management in incubators (i.e., the initial enterprises in the incubator) is a key element in promoting innovation development. However, conventional enterprise assessment and support modes, such as relying on static financial reports, expert experience interviews, and tools for SWOT analysis, have limitations. Firstly, the methods have hysteresis and statics, and cannot describe dynamic and nonlinear feedback loops and delay effects among key elements such as research and development investment, product competitiveness, market share, income and the like in enterprises, so that future nonlinear growth and emergent behaviors of the enterprises cannot be effectively predicted. Second, the prior art lacks a mature framework to effectively fuse both the System Dynamics (SD) model and the agent-based modeling (ABM) to capture both macroscopic feedback and microscopic heterogeneity. In addition, at the data level, the hatched enterprises generally face the challenges of sparse historical data and serious loss of key indexes, so that most data-driven models cannot be effectively trained. At the decision level, the existing hatching supporting strategy is an empirical mode, the capability of dynamic adjustment according to the real-time state of an enterprise is lacking, and the decision is usually directly executed in the real world, so that the cost is high and the risk is high. Therefore, there is a strong need in the art for a systematic solution that can fuse multi-source data, dynamically simulate the whole process of enterprise growth, and provide personalized, verifiable intelligent intervention strategies to overcome the above-mentioned technical bottlenecks. Disclosure of Invention In order to solve the above problems, the present invention provides a method and a system for simulating and intervening growth of an incumbent enterprise based on digital twinning. In a first aspect, the present invention provides a method for simulating and intervening growth of an incumbent enterprise based on digital twinning, comprising the steps of: s1, constructing a digital twin body of an object-under-hatch enterprise, wherein the digital twin body comprises an enterprise multi-source state model constructed based on multi-source heterogeneous data and a mixed growth mechanism model, the mixed growth mechanism model is coupled with a system dynamics SD model and an intelligent body basic modeling ABM model, and a mixed simulation scheduling mechanism is preset; S2, carrying out simulation deduction based on the digital twin body, namely, taking the current state defined by an enterprise multisource state model as an initial state, running a preset hybrid simulation scheduling mechanism, enabling the SD model and the ABM model to run alternately and interact through a dynamic calibration module of a hybrid growth mechanism model, so that the prediction parameters of the SD model can be adjusted online based on the result of the ABM model; S3, generating a personalized intervention strategy aiming at the target on-hatch enterprise through reinforcement learning RL intelligent agent based on the deduction state of S2. As a further limitation of the technical scheme of the invention, the enterprise multisource state model is used as the unified state representation of the digital twin body, is read by the hybrid growth mechanism model in S2 to carry out simulation deduction, and is used as the input of the state space S of the reinforcement learning RL intelligent body in S3. And the unified state characterization function of the enterprise multisource state model is clarified, and the data intercommunication generated by the simulation deduction and the intervention strategy is realized. And a consistent initial input basis is provided for the hybrid growth mechanism model, the continuity of simulation logic is ensured, meanwhile, a clear state input dimension is built for the RL intelligent agent, and the accurate matching of the intervention strategy generation and the actual state of the enterprise is ensured. As a further limitation of the present invention, in S1, the step of constructing a digital twin of the object in an hatched enterprise includes: S11, collecting multi-source heterogeneous data from enterprise internal systems, incubator operation data, industry chains, market data and dynamic public opinion data through an API (application program interface), an IoT (internet of things) sensor and a real-time data pipeline, converting all structured data into a uniform serialization format, and storing unstructured data and metadata thereof in an object storage service; S12, detect