Technical and ecological industry collaborative decision-making model construction system of system dynamics

Abstract

The invention relates to the technical field of collaborative decision-making of system dynamics, and discloses a technical and ecological industry collaborative decision-making model construction system of system dynamics. The system comprises a dynamic topology network construction module which is used for establishing an initial dynamic topology network comprising industrial entities, technical element nodes and collaborative relation edges; the data flow driving module is used for injecting real-time industrial and ecological data flow into the network and driving the node state and the edge weight to dynamically evolve; the strategy simulation module responds to network evolution and generates various decision scenes based on the current network topology and the node state; the strategy optimization module is used for carrying out iterative correction on the generated decision scene; and the feedback and learning module is used for feeding back the corrected decision scene to a network triggering structure for adaptive adjustment and recording a whole-process data track to form a decision case library. According to the system, real-time synchronization and dynamic response of the model and real ecology are realized, and continuous learning and optimization of the system are supported through a closed-loop feedback mechanism.

Description

Technical Field

[0001] This invention relates to the field of collaborative decision-making technology in system dynamics, specifically to a system for constructing collaborative decision-making models for the technology ecosystem industry in system dynamics. Background Technology

[0002] Existing technology ecosystem industry collaborative decision-making analysis mostly employs static system dynamics models or network models. These models typically establish fixed parameters and structural relationships based on historical data during the construction phase, simulating scenarios by running pre-defined equations. Once the initial state and interrelationships of the model are determined, they remain unchanged throughout the simulation period, and the analysis results are highly dependent on the accuracy of the assumptions about the system structure and parameter calibration during modeling. This static nature makes it difficult for the models to capture the rapid and continuous systemic changes caused by factors such as technological evolution, market fluctuations, and policy adjustments in the real industrial environment.

[0003] Traditional decision support models exhibit a unidirectional open-loop characteristic in their operational mechanism. Once the model outputs a decision recommendation based on the input parameters, the simulation process ends. The actual effects of the decision scheme cannot be fed back into the model to verify its long-term impact, nor can the model's structure be dynamically modified based on the decision's effectiveness. Each decision analysis is relatively independent, and the model lacks the ability to learn from past decision-making practices and their results. Decision cases, simulation data, and system feedback information cannot be systematically accumulated and integrated, resulting in the model lacking a foundation for self-evolution and optimization.

[0004] The limitations of static modeling and open-loop decision-making mentioned above make it difficult for existing technologies to support an adaptive decision-making system that can evolve synchronously with the real ecosystem and learn from its own decision-making history. A new technological approach is needed to enable the model to respond to changes in the external environment in real time and form a closed-loop learning process in which decision-making and model structure interact. Summary of the Invention

[0005] The purpose of this invention is to provide a system for constructing a collaborative decision-making model for the technology ecosystem industry based on system dynamics, in order to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, this invention provides a system for constructing a collaborative decision-making model for the technology ecosystem industry based on system dynamics. The system includes: The dynamic topology network construction module is used to establish an initial dynamic topology network for the technology ecosystem industry system. The initial dynamic topology network includes nodes representing different industry entities and technological elements, and edges representing the collaborative relationships between nodes. The data flow driving module is used to inject real-time industry operation data flow and ecological monitoring data flow into the initial dynamic topology network, driving the dynamic evolution of network node states and edge weights; The policy simulation module responds to network evolution by activating the policy simulator integrated within the network, generating various decision scenarios based on the current network topology and node states. The strategy optimization module receives the decision scenarios generated by the strategy simulation module and iteratively corrects the decision scenarios. The feedback and learning module is used to feed back the decision scenarios corrected by the strategy optimization module to the initial dynamic topology network, triggering adaptive adjustments to the network structure and recording all data trajectories during the network adjustment process to form a decision case library.

[0007] Preferably, the steps of the dynamic topology network construction module in establishing the initial dynamic topology network of the technology ecosystem industry system include: Define node types, including technology R&D institution nodes, product manufacturing enterprise nodes, ecosystem service provider nodes, and policy and regulatory agency nodes; Define edge relationship types, which include technology transfer relationships, material and energy flow relationships, financial support relationships, and information exchange relationships; Each node is assigned a set of attribute vectors, which include productivity level, resource consumption rate, ecological footprint intensity, and innovation potential index. Each edge is assigned an initial weight value, which is calculated based on historical interaction frequency and association strength; Connect all nodes with edges to form the initial dynamic topology network.

[0008] Preferably, the step of the data flow driving module injecting real-time industry operation data flow and ecological monitoring data flow into the initial dynamic topology network specifically includes: By deploying sensor arrays in industrial entities, real-time production capacity data, raw material inventory data, and energy consumption data are collected to form an industrial operation data stream; Real-time pollutant concentration data, biodiversity index, and carbon sink flow data are collected by monitoring stations deployed in ecological areas to form an ecological monitoring data stream. Perform timestamp alignment and data format standardization on industrial operation data streams and ecological monitoring data streams; The processed data stream is mapped to the corresponding node attribute vectors and edge weight values ​​in the initial dynamic topology network; Based on the mapping results, the values ​​of the node attribute vectors and the edge weights are periodically updated.

[0009] Preferably, the step of activating the policy simulator integrated within the network by the policy simulation module is controlled by network evolution triggering conditions, which are: When the change in the attribute vector of any node exceeds its preset fluctuation threshold, or the change in the weight value of any edge exceeds its preset sensitivity threshold, the network evolution trigger condition is met. Once the strategy simulator is activated, it first scans the entire initial dynamic topology network to capture the current state of all nodes and edges that have changed. The strategy simulator simulates the chain reaction paths that the network may generate under different external policy stimuli or market changes, based on the captured state.

[0010] Preferably, the process by which the strategy simulation module generates multiple decision scenarios based on the current network topology and node states includes: The strategy simulator has a built-in scenario rule template library, which defines a variety of basic intervention modes such as investment tilt, technical standard upgrade, and emission limit adjustment. The strategy simulator matches the captured network state with the scenario rule template, and fills in the specific network nodes and edge parameters for each successfully matched template; Generate a draft decision scenario that includes specific intervention targets, intervention intensity, and intervention time window; The draft decision scenarios are internally consistent, and those with logical inconsistencies are eliminated to form the final multiple decision scenarios.

[0011] Preferably, the step of the strategy optimization module iteratively correcting the decision scenario includes: The policy optimizer receives various decision scenarios sent by the policy simulator; The strategy optimizer calls the historical decision cases stored in the decision case library and calculates the similarity between the current decision scenario and the historical cases; Based on the similarity calculation results, an initial credibility score is assigned to each decision scenario; The policy optimizer initiates a simulated execution loop, virtually applying each decision scenario to the current initial dynamic topology network, and observes the evolution trend of the network state. Based on the results of the simulation, the credibility score of the decision scenario is dynamically adjusted, and the scenario parameters are fine-tuned to complete one iteration correction. Repeat the simulation and parameter fine-tuning until the credibility score of the decision scenario stabilizes.

[0012] Preferably, the feedback and learning module feeds back the decision scenario corrected by the strategy optimization module to the initial dynamic topology network, triggering adaptive adjustment of the network structure, specifically manifested as follows: The policy optimizer sends the revised and stable decision scenarios to the corresponding nodes in the initial dynamic topology network in the form of virtual instructions; Based on the received virtual instructions, the node rehearses the changes in its own attribute vector and the weights of the connecting edges; The nodes submit the pre-rehearsal results to the network central coordinator in the form of local topology change proposals; The network central coordinator aggregates change proposals from all nodes and assesses the overall network stability; If the evaluation is successful, the network center coordinator approves the proposal, and the initial dynamic topology network updates its topology accordingly.

[0013] Preferably, the step of the feedback and learning module recording all data trajectories during the network adjustment process is implemented as follows: A data trajectory recording agent is built into the initial dynamic topology network. The data trajectory recording agent listens for all node state updates, edge weight changes, policy simulator activation, policy optimizer iteration, and network structure adjustment events in the network. The data trajectory recording agent converts each monitored event into standardized trajectory data points. The trajectory data points include event type, timestamp, involved node or edge identifier, and state before and after the change. All trajectory data points are stored in a time-series database in chronological order to form a decision case library; The decision case library indexes each stored case, with the index key being a summary of the network's global features at the time the case is triggered.

[0014] Preferably, the feedback and learning module forms a decision case library including: Once a new decision-making process is completed and new decision cases are generated and stored in the decision case library, the incremental learning engine is activated. The incremental learning engine extracts the most recently added case data from the decision case library; Incremental learning engines analyze new patterns of network behavior or changes in decision-making rules reflected in newly added case data; Based on the analysis results, the incremental learning engine adjusts the rule weights of the scenario rule template library in the policy simulator, or optimizes the simulation execution loop parameters of the policy optimizer. The incremental learning engine synchronizes the adjusted model parameters to the policy simulator and policy optimizer, completing one incremental learning cycle.

[0015] Preferably, the step of assigning an initial weight value to each edge, wherein the initial weight value is calculated based on historical interaction frequency and association strength, includes: Retrieve all historical interaction event records related to the node pairs connected by the edges from the historical interaction database. The historical interaction event records include interaction type, interaction timestamp, and interaction scale index. Calculating historical interaction frequency involves: counting the total number of interaction events between node pairs within a preset historical time period, and normalizing the total number of interaction events by dividing it by a preset baseline frequency value to obtain a frequency factor. The calculation of historical association strength is specifically as follows: the interaction scale index of each interaction event between node pairs is quantified, and different weight coefficients are assigned according to the interaction type. Then, the weighted sum of the scales of all interaction events is calculated, and the weighted sum is divided by a preset baseline strength value for normalization to obtain the strength factor. The frequency factor and intensity factor are input into a preset weight calculation function, which is a linear combination or a multiplicative combination, and the initial weight value is output. The initial weight values ​​are subject to range constraints to ensure that they fall between the preset minimum and maximum weight values.

[0016] Compared with the prior art, the beneficial effects of the present invention are: By injecting real-time industry operation data streams and ecological monitoring data streams into the initial dynamic topology network, the dynamic evolution of network node states and edge weights is directly driven. This technique frees the model from dependence on fixed parameters and static structures, allowing its internal states to continuously reflect transient changes in the external real world. The model transforms from a simulation system requiring pre-calibration into a dynamic mapping system capable of sensing and responding to ecological data fluctuations in real time, ensuring the synchronization of the analytical baseline with the actual situation, and generating decision scenarios with higher timeliness and relevance.

[0017] By feeding the optimized and corrected decision scenarios back to the initial dynamic topology network, adaptive adjustments to the network structure are triggered, and the entire process's data trajectory is recorded to form a decision case library. This technology constructs a closed loop from decision generation to model correction and experience accumulation. The output of the decision is no longer just the final solution, but serves as input to the model structure, thereby enabling the assessment of the long-term structural impact of the decision on the system topology. Simultaneously, the recording of the entire data trajectory transforms each decision practice into analyzable learning samples. The formation of the case library provides a data foundation based on historical experience for subsequent simulation optimization, enabling the system to possess the ability for continuous iteration and autonomous evolution. Attached Figure Description

[0018] Figure 1 This is a schematic diagram illustrating the working principle of the system for constructing the technological ecosystem industry collaborative decision-making model of system dynamics as described in this invention. Figure 2 Flowchart for constructing dynamic topology networks; Figure 3 A flowchart driven by data flow; Figure 4 A heatmap showing the correlation of core attributes of nodes in the technology ecosystem industry; Figure 5 The average initial weight distribution for different edge relationship types. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Please see Figure 1 This invention provides a system for constructing a collaborative decision-making model for a technology ecosystem industry based on system dynamics. The system includes: a dynamic topology network construction module, a data flow driving module, a strategy simulation module, a strategy optimization module, and a feedback and learning module. The dynamic topology network construction module is responsible for establishing the initial dynamic topology network of the technology ecosystem industry system. This network contains nodes representing different industry entities and technological elements, as well as edges representing the collaborative relationships between nodes. The data flow driving module injects real-time industry operation data streams and ecosystem monitoring data streams into the initial dynamic topology network, driving the dynamic evolution of network node states and edge weights. The strategy simulation module responds to network evolution by activating a strategy simulator integrated within the network, generating multiple decision scenarios based on the current network topology and node states. The strategy optimization module receives the decision scenarios generated by the strategy simulation module and iteratively corrects them. The feedback and learning module feeds back the corrected decision scenarios from the strategy optimization module to the initial dynamic topology network, triggering adaptive adjustments to the network structure and recording all data trajectories during the network adjustment process, forming a decision case library.

[0021] Example 1: See Figure 2In its implementation, the dynamic topology network construction module establishes the initial dynamic topology network of the technology ecosystem industrial system, defining node types, including technology R&D institution nodes, product manufacturing enterprise nodes, ecosystem service provider nodes, and policy and regulatory agency nodes. Technology R&D institution nodes represent entities engaged in core technology research, development, and innovation; product manufacturing enterprise nodes represent entities that transform technology into specific products and carry out production; ecosystem service provider nodes represent entities that provide services such as resource recycling, pollution control, and ecological maintenance; and policy and regulatory agency nodes represent entities that formulate and supervise industrial policies, technical standards, and environmental regulations. Edge relationship types are also defined, including technology transfer relationships, material and energy flow relationships, financial support relationships, and information exchange relationships. Technology transfer relationships represent the transfer of knowledge, patents, or processes from technology R&D institution nodes to product manufacturing enterprise nodes, etc.; material and energy flow relationships represent the physical flow of raw materials, products, by-products, or energy between different nodes; financial support relationships represent the flow of investment, subsidies, or credit between nodes; and information exchange relationships represent the sharing and transmission of market information, policy signals, or monitoring data between nodes.

[0022] In implementation, each node is assigned a set of attribute vectors, including production capacity level, resource consumption rate, ecological footprint intensity, and innovation potential index. The production capacity level attribute quantifies the node's output capacity or service supply capacity per unit time; the resource consumption rate attribute quantifies the node's unit consumption of raw materials, energy, and other resources during operation; the ecological footprint intensity attribute quantifies the pressure or impact of node activities on the ecological environment; and the innovation potential index attribute comprehensively assesses the node's potential capabilities in technological improvement, process innovation, or model innovation. The assignment of attribute vectors relies on the integration of prior knowledge of the industrial system and historical data analysis to ensure that the initial state of each node reasonably reflects its position and function within the system. Each edge is assigned an initial weight value, calculated based on historical interaction frequency and association strength. This calculation requires retrieving all historical interaction event records related to the nodes connected to the edge from the historical interaction database. These historical interaction event records include interaction type, interaction timestamp, and interaction scale index. Interaction type corresponds to edge relationship type. Interaction timestamp records the specific time point when the event occurs. Interaction scale index is a numerical value that quantifies the size of the interaction event, such as the amount of technology transfer contract, the tonnage of material transported, the amount of capital flow, or the frequency of information exchange.

[0023] In practical implementation, the historical interaction frequency is calculated by counting the total number of interaction events between node pairs within a preset historical time period, dividing the total number of interaction events by a preset baseline frequency value, and then normalizing to obtain the frequency factor. The preset historical time period is determined based on the evolution speed of the industry system. For rapidly changing industries, the time period is set shorter, while for mature and stable industries, the time period can be appropriately extended. The baseline frequency value is a reference constant used for normalization, and its value can be determined based on the statistical characteristics of the total number of interaction events between all node pairs in the entire network, such as taking the average or median of the number of interaction events between all node pairs. The historical association strength is calculated by quantifying the interaction scale index of each interaction event between node pairs, assigning different weight coefficients according to the interaction type, then calculating the weighted sum of the scales of all interaction events, and dividing the weighted sum by a preset baseline strength value to obtain the strength factor. The quantification of interaction scale indicators must match the indicator units and dimensions. The weighting coefficients reflect the relative importance of different types of interactions in the association strength calculation. For example, a single interaction involving financial support may have a greater impact on association strength than a single information interaction. The baseline strength value is also a reference constant used for normalization. Its determination method can be similar to that of the baseline frequency value, based on the overall distribution of interaction scale within the network.

[0024] In some embodiments, frequency and intensity factors are input into a preset weight calculation function, which can be a linear combination or a multiplicative combination, and outputs an initial weight value. The linear combination weight calculation function is expressed as the initial weight value equal to coefficient A multiplied by the frequency factor plus coefficient B multiplied by the intensity factor. The sum of coefficients A and B is a constant, and the ratio of the coefficients reflects different emphases on interaction frequency and association strength. The multiplicative combination weight calculation function is expressed as the initial weight value equal to the frequency factor multiplied by the intensity factor multiplied by a scaling factor. The multiplicative combination emphasizes the synergistic effect of interaction frequency and association strength; a low value for either factor will significantly reduce the initial weight value. A range constraint is applied to the initial weight value to ensure it falls between a preset minimum and maximum weight. This range constraint is implemented through a pruning function: if the initial weight value is greater than the maximum weight, it is set to the maximum weight; if the initial weight value is less than the minimum weight, it is set to the minimum weight. After all nodes and edges are defined, attribute-assigned, and weight-calculated, the dynamic topology network construction module performs a connection operation, connecting all nodes and edges to form the initial dynamic topology network. The connection logic between nodes strictly follows the defined edge relationship type. A technology research and development institution node can be connected to one or more product manufacturing enterprise nodes through technology transfer relationships. A product manufacturing enterprise node can be connected to an ecological service provider node through material and energy flow relationships. A policy and regulatory agency node can be connected to all other types of nodes through information exchange relationships.

[0025] In some embodiments, the structured data of the initial dynamic topology network is stored in a graph database system. Nodes in the graph database system store the attribute vectors of their corresponding entities, and edges store the relationship types and calculated initial weight values. After the dynamic topology network construction module completes construction, the network structure is persistently stored in a data serialization format, providing an interface for data mapping and update operations of subsequent data flow-driven modules. The dynamic topology network construction module runs each time the system initializes or receives a network reconstruction command, recalculating node attribute vectors and initial edge weight values ​​based on the latest historical data to generate an updated initial dynamic topology network. In specific implementations, the initial assignment of node attribute vectors not only relies on static historical data but can also incorporate some real-time data snapshots to make the initial network state more closely reflect the current industry operation. The historical interaction database upon which the edge weight calculation process relies needs continuous maintenance and updates to ensure that the historical interaction event records included in the calculation reflect the latest collaboration and association patterns between nodes. Preset weight calculation functions and range constraint parameters can be configurable, allowing users to adjust them according to the characteristics of specific technology ecosystem industry systems to adapt to the network relationship quantification needs of different industries. The dynamic topology network construction module is the cornerstone of the entire collaborative decision-making model construction system. The initial dynamic topology network established therein serves as the operational object and data carrier for subsequent data injection, state evolution, policy simulation, and optimization. The accuracy and rationality of network construction directly affect the reliability of the processing results of subsequent modules. Therefore, in practical implementation, it is necessary to fully demonstrate and verify the scientific validity of the definition of node types, the dimension of attribute vectors, the classification of edge relationship types, and the initial weight calculation method.

[0026] Example 2: See Figure 3In practical implementation, the data flow-driven module injects real-time industry operation data streams and ecological monitoring data streams into the initial dynamic topology network. Real-time production capacity data, raw material inventory data, and energy consumption data are collected by sensor arrays deployed on industrial entities, forming the industry operation data stream. The sensor arrays include sensors installed on production lines for counting or measurement, level sensors installed in raw material warehouses, and smart meters and flow meters installed on energy supply lines. Real-time production capacity data is measured by the quantity of qualified products produced per unit time; raw material inventory data is measured by the real-time storage volume of materials in the warehouse; and energy consumption data is measured by the real-time consumption of electricity, natural gas, or steam. Real-time pollutant concentration data, biodiversity index, and carbon sink flow data are collected by monitoring stations deployed in ecological areas, forming the ecological monitoring data stream. These monitoring stations include chemical substance concentration sensors deployed in rivers, the atmosphere, and soil; infrared cameras and acoustic recorders for observing species activity; and flux towers for measuring vegetation carbon sequestration capacity. Pollutant concentration data are measured by chemical oxygen demand concentration in water bodies and fine particulate matter concentration in the air. Biodiversity index is measured by the species richness and evenness observed per unit area. Carbon sink flow data is measured by the amount of carbon dioxide equivalent absorbed by forests or wetlands per unit time.

[0027] In practical implementation, the industrial operation data stream and the ecological monitoring data stream undergo timestamp alignment and data format standardization. Timestamp alignment uses the Network Time Protocol (NTP) to assign a unified time stamp to each batch of collected data. Data format standardization converts heterogeneous data from different sensors and monitoring stations into a unified key-value pair or tabular format according to a predefined data structure. The processed data stream is then mapped to the corresponding node attribute vectors and edge weight values ​​in the initial dynamic topology network. The mapping relationship is based on rules stored in a configuration table, which records the correspondence between data source identifiers and network node identifiers or edge identifiers. For example, real-time production capacity data from a specific manufacturing workshop is mapped to the production capacity level attribute of the product manufacturing enterprise node representing that workshop; pollutant concentration data from the discharge outlet of a specific wastewater treatment plant is mapped to the ecological footprint intensity attribute of the ecological service provider node representing that wastewater treatment plant; and material transport data connecting two nodes is mapped to the weight value of the edge representing the material and energy flow relationship. Based on the mapping results, the values ​​of the node attribute vectors and the edge weight values ​​are periodically updated. The update operation is atomic, completing the assignment of values ​​to all affected attributes and weights within a single update cycle.

[0028] In practice, the data flow-driven module continuously drives the dynamic evolution of network node states and edge weights. The magnitude of change in node attribute vectors is determined by comparing the numerical difference before and after the update with a preset fluctuation threshold, while the magnitude of change in edge weight values ​​is determined by comparing the numerical difference before and after the update with a preset sensitivity threshold. The fluctuation threshold and sensitivity threshold are set independently for different node types and edge relationship types. For example, the innovation potential index fluctuation threshold for technology R&D institution nodes is set higher, allowing for gradual changes over a longer period, while the sensitivity threshold for edge weights in material and energy flow relationships is set lower, enabling them to respond quickly to small fluctuations in material supply. When the magnitude of change in any node attribute vector exceeds its preset fluctuation threshold, or the magnitude of change in any edge weight value exceeds its preset sensitivity threshold, the network evolution trigger condition is met. When the network evolution trigger condition is met, a trigger event carrying the relevant node and edge identifiers is generated and sent to the strategy simulation module.

[0029] In some embodiments, the policy simulation module activates the policy simulator integrated within the network in response to network evolution triggering conditions. Once activated, the policy simulator first scans the entire initial dynamic topology network, capturing the current states of all changed nodes and edges. The scanning process uses a depth-first search algorithm to traverse the network graph, identifying nodes whose attribute vector changes exceed their preset fluctuation thresholds since the last stable state, and edges whose weight changes exceed their preset sensitivity thresholds. Based on the captured states, the policy simulator simulates the potential chain reaction paths that the network may generate under different external policy stimuli or market changes. The simulation process abstracts external policy stimuli or market changes into a series of input perturbation variables. These input perturbation variables act on specific nodes or edges in the network. The system dynamics model built into the policy simulator calculates the propagation effect of the perturbation in the network based on the connections between nodes and edge weights, thereby deduce multiple chain reaction paths. The chain reaction paths are recorded in the form of node state sequences and edge weight change sequences.

[0030] In some embodiments, the setting of the fluctuation threshold and sensitivity threshold considers the historical statistical characteristics of node attributes and edge weights. One optional setting method is based on the standard deviation of historical data multiplied by a coefficient, the mathematical relationship of which can be expressed as: Where: symbol This represents the fluctuation threshold or sensitivity threshold set for a specific node attribute or edge weight. (Symbol) This represents the adjustment coefficient, a constant greater than zero. (Symbol) This represents the standard deviation of the node's attribute value or edge weight value within a preset historical time period. The threshold is calculated using a formula. It exhibits dynamic adaptability, reflecting the scale of changes in attributes or weights themselves. The activation mechanism of the policy simulator is event-driven, initiating computation only when network evolution trigger conditions are met, avoiding meaningless simulation calculations when the system state is stable. Scanning the entire initial dynamic topology network and capturing its state ensures that the policy simulator obtains a globally consistent snapshot of the network at the trigger moment, providing a foundation for generating accurate decision scenarios. Parameters of the system dynamics model, such as the conversion rate between nodes and the decay coefficient of edge weights, need to be calibrated based on knowledge of specific industry domains.

[0031] It is understandable that the data flow-driven module constitutes the perception layer for real-time network status updates, and the deployment density and accuracy of sensor groups and monitoring stations directly affect the completeness and accuracy of the data. Timestamp alignment is a prerequisite for ensuring that industrial operation data and ecological monitoring data are comparable and correlated in the time dimension. The mapping relationship configuration table needs to be updated synchronously with the adjustment of the initial dynamic topology network structure to ensure that data can be accurately injected into the correct network elements. The update cycle needs to be set in a balance between the frequency of data flow generation and the timeliness requirements of decision-making; too short a cycle may lead to excessive system computational load, while too long a cycle may lead to delayed decision response. Fluctuation threshold and sensitivity threshold are key parameters for controlling the sensitivity of the control system, and their settings need to strike a balance between avoiding noise-induced false triggers and preventing important changes from being ignored. The activation and scanning operations of the strategy simulator constitute a bridge from data changes to scenario simulation, and its computational efficiency directly affects the system's ability to respond to dynamic changes.

[0032] Example 3: In specific implementation, the strategy simulation module generates various decision scenarios based on the current network topology and node states. This includes a scenario rule template library built into the strategy simulator, which defines various basic intervention modes such as investment tilt, technology standard upgrade, and emission limit adjustment. The investment tilt template describes the specific rules and parameter framework for concentrating funds or resources to specific types of nodes or regions in the network. The technology standard upgrade template describes the specific rules and parameter framework for improving the access or performance requirements of specific technology paths or process links in the network. The emission limit adjustment template describes the specific rules and parameter framework for modifying the upper limit of pollutant emissions allowed by specific nodes in the network. The strategy simulator matches the captured network state with the scenario rule templates. The matching operation is achieved by calculating the matching degree between the current network state feature vector and the predefined feature triggering conditions of each scenario rule template, and filling in specific network nodes and edge parameters for each successfully matched template. When filling in network nodes and edge parameters, the strategy simulator, according to the template logic, identifies specific nodes that meet the conditions from the current network state as intervention targets, and extracts or calculates specific intervention intensity values ​​and intervention time window ranges from the data stream. A draft decision scenario is generated, containing specific intervention targets, intervention intensity, and intervention time windows. This draft exists as a structured data object, explicitly recording the scenario rule template type used, the target node identifier list, the quantified intervention value, and the start and end times of the intervention's effectiveness. An internal consistency check is performed on the draft decision scenarios, eliminating drafts with logical conflicts. This check primarily examines whether there are contradictory instructions targeting the same node within the same draft decision scenario, or whether the temporal logic of the instructions is infeasible, thus forming multiple final decision scenarios.

[0033] In practical implementation, the strategy optimization module receives multiple decision scenarios generated by the strategy simulation module. The strategy optimizer calls upon historical decision cases stored in the decision case library to calculate the similarity between each current decision scenario and historical cases. When calculating similarity, the strategy optimizer extracts feature vectors from the decision scenarios. These feature vectors include the intervention mode type, the distribution of involved node types, and the average intervention intensity. Simultaneously, it extracts the corresponding feature vectors from the historical decision cases before implementation. A multi-dimensional spatial distance metric is used to calculate the inverse distance between the two sets of feature vectors as the similarity value. Based on the similarity calculation results, an initial credibility score is assigned to each decision scenario. The operation of assigning an initial credibility score can be expressed as the following mathematical relationship: Where: symbol This represents the calculated initial credibility score. (Symbol) and Represents the normalization coefficient. Symbol Represents the weighting coefficient of the similarity factor. (Symbol) This represents the calculated similarity score. (Symbol) The weighting coefficients representing historical effect factors. (Symbol) This represents the final effect score of the corresponding historical decision-making case. The policy optimizer initiates a simulation execution loop, virtually applying each decision scenario to the current initial dynamic topology network and observing the evolution trend of the network state. The simulation execution loop progresses at discrete time steps. Within each time step, the policy optimizer modifies the attribute vectors or edge weights of the corresponding nodes in the network based on the intervention object, intervention intensity, and intervention time window defined in the decision scenario. Then, it calls the system dynamics model shared with the policy simulator to calculate the evolution of the network state to the next time step. Based on the simulation execution results, the credibility score of the decision scenario is dynamically adjusted, and the scenario parameters are fine-tuned to complete one iteration correction. The basis for dynamically adjusting the credibility score is the degree of closeness between the overall network state index and the preset target after the simulation execution. The overall network state index can be comprehensive economic benefits, total ecological footprint, etc. Parameter fine-tuning adopts the gradient trial method, generating new parameter combinations for testing in a small neighborhood of the original parameter values ​​of the decision scenario.

[0034] In some embodiments, simulation execution and parameter fine-tuning are repeated until the credibility score of the decision scenario stabilizes. The criterion for stabilization is that the fluctuation range of the credibility score of the decision scenario is less than a preset stability threshold across multiple iterations. The strategy optimization module outputs the decision scenario or set of decision scenarios with the highest credibility score after multiple iterations. The internal logic of the simulation execution loop ensures that the virtual application of the decision scenario does not actually change the storage state of the initial dynamic topology network; all modifications are performed only on the network copy in memory. In the similarity calculation between the decision scenario and historical cases, the multidimensional spatial distance metric can use Euclidean distance or Manhattan distance. The final effect score of the historical decision cases is also considered. The results are derived from the comparison between historical network state changes recorded in the decision case database and the preset targets.

[0035] In some embodiments, the weighting coefficients in the initial credibility scoring formula and These are parameters that can be adjusted through system configuration. Adjusting the weight coefficients can change the policy optimizer's dependence on historical similarity and historical absolute effects when evaluating decision scenarios. Internal consistency checks, in addition to verifying instruction contradictions and temporal logic, also check whether the intervention intensity exceeds the reasonable range that the target node can bear physically or economically. The setting of the call frequency and discrete time step size of the system dynamics model in the simulation execution loop needs to be balanced between simulation accuracy and computational efficiency. When fine-tuning parameters using the gradient heuristic method, the step size of parameter adjustment can adaptively decrease as the number of iterations increases to facilitate convergence. During the generation of the decision scenario draft, when filling the scenario rule template with specific network nodes and edge parameters, if the number of matched potential nodes is too large, a priority ranking rule can be used to select a specific number of nodes with the highest ranking as the final intervention targets.

[0036] It is understandable that the scenario rule template library within the policy simulator is the basic set of rules for generating decision scenarios. The completeness and rationality of the template library determine the coverage and practicality of the generated scenarios. The process of matching and filling parameters is a key step in instantiating general rules into specific executable solutions. Internal consistency checks are a necessary filtering step to ensure that the generated draft decision scenarios are logically self-consistent and feasible. The policy optimizer provides historical evidence and experience for evaluating decision scenarios by introducing historical decision cases for similarity calculation. The simulation execution loop constructs a safe virtual environment, allowing the effects and impacts of decision scenarios to be repeatedly tested and evaluated without interfering with the real system. The dynamic adjustment mechanism of the credibility score and the parameter fine-tuning process jointly achieve continuous optimization and refinement of the decision scenarios. The iteration termination condition ensures that the optimization process can converge to a relatively stable evaluation result within a finite time.

[0037] Example 4: In specific implementation, the feedback and learning module feeds back the decision scenario corrected by the policy optimization module to the initial dynamic topology network. The policy optimizer sends the corrected and stable decision scenario to the corresponding node in the initial dynamic topology network in the form of virtual instructions. The virtual instruction is a structured data message whose fields include the target node identifier, instruction type code, instruction parameter list, and the time range in which the instruction takes effect. Based on the received virtual instructions, the node pre-simulates the changes in its own attribute vector and the weights of the connected edges. The pre-simulation process is an internal calculation process within the node. The node calls the built-in response function according to the instruction type code. The response function calculates the expected change values ​​of each component in the node's attribute vector and the expected adjustment values ​​of the weights of each edge connected to it according to the instruction parameter list. The node submits the pre-simulation results to the network central coordinator in the form of a local topology change proposal. The local topology change proposal includes the node identifier, the new attribute vector values ​​calculated after the pre-simulation, the new proposed weight values ​​of each edge connected to the node, and a summary of the virtual instructions on which the pre-simulation was based.

[0038] In practice, the network coordinator aggregates all node change proposals and assesses the overall network stability. This assessment involves calculating the differences in multiple network topology metrics before and after applying all change proposals and determining whether these differences are within acceptable tolerance limits. A core stability assessment metric calculated by the network coordinator is network connectivity change, node state consistency, and edge weight balance. Network connectivity change is measured by comparing the change in the largest connected subgraph size or global clustering coefficient before and after proposal application. Node state consistency is assessed by calculating the variance or coefficient of variation of all node attribute vectors after proposal application. Edge weight balance is determined by analyzing the uniformity of edge weight distribution connecting different node types. If the assessment passes, the network coordinator approves the proposal, and the initial dynamic topology network updates its topology accordingly. The update operation includes replacing all node attribute vector values ​​with the new attribute vector values ​​in the proposal, updating all edge weight values ​​with the new suggested weight values ​​in the proposal, and deciding whether to add or delete nodes and edges based on instruction logic. If the evaluation fails, the network central coordinator will reject all or part of the conflicting proposals and send a rejection notice and the reason for rejection to the relevant nodes.

[0039] In some embodiments, referring to Table 1, the feedback and learning module records all data trajectories during the network adjustment process. A data trajectory recording agent is built into the initial dynamic topology network. This agent monitors all node state updates, edge weight changes, policy simulator activation, policy optimizer iteration, and network structure adjustment events in the network. The agent converts each monitored event into a standardized trajectory data point. Each trajectory data point includes the event type, timestamp, involved node or edge identifier, and the state before and after the change. The event type distinguishes different types of operations, the timestamp is accurate to milliseconds, the involved node or edge identifier locates specific elements in the network, and the state before and after the change is recorded in a serialized format, showing the complete attributes or weight values ​​of the element before and after the event. All trajectory data points are stored in a time-series database in chronological order to form a decision case library. The data model of the time-series database is optimized for fast writing and retrieval of time-series data. The decision case library indexes each stored case, with the index key being the network global feature summary at the time the case was triggered. The network global feature summary is a fixed-length string obtained by extracting features and hashing the node attribute vectors and edge weights of the entire network at the trigger time, and is used to quickly locate historical cases with similar backgrounds during queries.

[0040] Table 1: Trajectory Data Point Structure Table Field Name Data types describe Event Type String Identify the event type, such as "NodeAttributeUpdate" or "EdgeWeightChange". Timestamp Long plastic surgery The absolute time of the event, such as 1720321200000 Involving logos String A unique identifier for the event subject, such as node IDN_203 or edges IDE_N101-N205. State before change Text / JSON A snapshot of the state of the element corresponding to this identifier before the event occurred. After the change Text / JSON A snapshot of the state of the element corresponding to this identifier after the event occurs. In some embodiments, the process by which the network central coordinator assesses overall network stability can be aided by a quantitative stability assessment function, which can be expressed as: Where: symbol This represents the calculated network stability evaluation factor; the closer its value is to 1, the more stable the network. (Symbol) and This represents a weighting coefficient, used to adjust the importance of different indicators in the evaluation. (Symbol) The mean absolute change in node degree centrality reflects the degree of change in the network connectivity structure. (Symbol) The change in variance representing edge weights reflects the change in the concentration of network connectivity distribution. The network central coordinator sets a stability threshold. When the calculated stability evaluation factor Greater than or equal to the stability threshold At that time, the evaluation result was passed. When a node rehearses changes in its own attribute vector and connection edge weights, the built-in response function used needs to be predefined and configured according to different node types and instruction types. The data trajectory recording agent's listening mechanism adopts a publish-subscribe pattern; any state change event in the network is published to an internal message bus, and the data trajectory recording agent subscribes to all event types of interest on this bus.

[0041] It is understandable that the policy optimizer sending virtual instructions is the starting point for the decision results to act on the network model. The virtual instructions themselves do not immediately change the network storage state, but rather trigger a proposal and coordination process. The pre-calculation performed by the nodes is a local computation, only considering the impact of the instructions on the nodes themselves and their directly connected edges. This distributed computation method helps reduce the computational burden on the network central coordinator. Local topology change proposals are change suggestions put forward by nodes based on local views, which may potentially conflict with proposals from other nodes. The global stability assessment performed by the network central coordinator is a centralized decision from the perspective of the entire network, ensuring that any structural adjustments will not lead to network dysfunction or collapse. The monitoring and recording functions of the data trajectory recording agent constitute the system's "black box," completely preserving the digital footprint of network evolution and decision-making processes. The choice of a time-series database ensures efficient storage of massive trajectory data and rapid retrieval by time dimension. Establishing an index for the decision case library with the network global feature summary as the key greatly optimizes the efficiency of the policy optimization module in retrieving similar historical cases in subsequent decisions. The entire process realizes a closed loop from decision output to network structure adjustment, and then to process data recording, enabling the network model to adapt based on historical feedback.

[0042] It is understandable that when evaluating proposals, the network coordinator may use not only quantitative formulas but also a series of business rules for logical judgment. The standardized trajectory data points generated by the data trajectory recorder are structurally designed to ensure that subsequent analysis programs can accurately parse the event content. The continuous accumulation of the decision case library provides a data foundation for the system's incremental learning. The formula includes the average absolute change in node degree centrality. The calculation requires the network adjacency matrix before and after the proposal application, and the change in variance of the edge weights. The calculation requires taking the variance of the weight value sequence of all edges and then comparing the differences.

[0043] See Figure 4In the node attribute analysis of the technology ecosystem industry collaborative decision-making model, the heatmap presents the distribution of correlation coefficients among four core node attributes: production capacity level, resource consumption rate, ecological footprint intensity, and innovation potential index. Specifically, the correlation coefficient of each attribute itself is 1.00 (diagonal area), representing a completely positive correlation; production capacity level and resource consumption rate show a weak negative correlation (coefficient -0.09), while the correlation with ecological footprint intensity and innovation potential index is close to 0; resource consumption rate and ecological footprint intensity show a weak negative correlation (coefficient -0.14), and the correlation with innovation potential index is relatively weak (coefficient -0.06); ecological footprint intensity and innovation potential index show a weak positive correlation (coefficient 0.26). These correlation results reflect that the overall degree of correlation among the core node attributes in the technology ecosystem industry system is low, with only ecological footprint intensity and innovation potential index showing a certain positive correlation. This can provide a quantitative basis for attribute correlation for subsequent node state evolution and decision scenario simulation.

[0044] Example 5: In specific implementation, after the feedback and learning module forms a decision case library, when a new decision-making process is completed and new decision cases are generated and stored in the decision case library, the incremental learning engine is started. The incremental learning engine is an independent computing process, and its activation is triggered by the data writing event of the decision case library. Specifically, when the cumulative number of newly added cases in the decision case library reaches a preset batch size, or when a preset time interval has elapsed since the last run of the incremental learning engine, the incremental learning engine will automatically start. The incremental learning engine extracts the most recent batch of newly added case data from the decision case library. The extraction operation is based on the timestamp of the case storage and filters out all case records stored within a specified time window. These case records contain complete event trajectory data from the activation of the policy simulator to the completion of network structure adjustment. The incremental learning engine analyzes the new network behavior patterns or changes in decision-making rules reflected in the newly added case data. The analysis process first preprocesses the newly added case data, including data cleaning, feature extraction, and sequence alignment. Then, unsupervised learning algorithms such as cluster analysis or association rule mining are applied to identify recurring patterns in the case data, potential correlations between different decision scenarios and the final network evolution results, and frequent sequences of network element state changes.

[0045] In practice, the incremental learning engine adjusts the rule weights in the scenario rule template library within the policy simulator or optimizes the simulation execution loop parameters of the policy optimizer based on the analysis results. When adjusting the rule weights in the scenario rule template library, the incremental learning engine calculates the frequency of each basic intervention mode in new cases and the degree of network state improvement achieved by using that mode in the decision scenario. Then, it reassigns the weight coefficients of the corresponding rules in the template library according to an update formula. When optimizing the simulation execution loop parameters of the policy optimizer, the incremental learning engine analyzes the relationship between the number of iterations in the simulation execution loop, the parameter fine-tuning step size, and the convergence speed of the final decision scenario credibility score in new cases. This allows it to adjust control parameters such as the maximum number of iterations limit and the parameter fine-tuning step size decay rate in the simulation execution loop. The incremental learning engine synchronizes the adjusted model parameters to the policy simulator and the policy optimizer, completing one incremental learning cycle. The synchronization operation uses inter-process communication or shared memory to pass the updated rule weight configuration file or simulation loop parameter configuration file to the running instances of the policy simulator and the policy optimizer. Upon receiving the new configuration, the policy simulator and the policy optimizer dynamically load and apply these parameters without restarting the entire system.

[0046] In some embodiments, when the incremental learning engine analyzes new patterns of network behavior or changes in decision-making patterns reflected in newly added case data, association rule mining can be used to discover patterns such as "when the innovation potential index of the technology R&D institution node significantly increases and there is simultaneous investment tilt intervention towards the product manufacturing enterprise node, the overall network capacity level increases in the subsequent three evolution cycles." Cluster analysis can group cases with similar initial network state characteristics and similar decision interventions into one category, thereby identifying effective decision-making patterns for specific types of network problems. The update operation of adjusting the rule weights of the scenario rule template library can be formally represented as: Where: symbol Representative Rules Updated weight values. (Symbol) Representative Rules Weight values ​​before update. (Symbol) Representing the learning rate, it is a constant between 0 and 1 used to control the magnitude of weight updates. (Symbol) Representative Rules Normalized values ​​of frequency of occurrence in the newly added case dataset. Symbol Representative application rules The normalized value of the average performance score of the decision-making cases. The performance score is derived from the network state improvement index recorded in the decision-making case library. The incremental learning engine analyzes the new case data periodically and automatically, without manual intervention. The analysis results are directly used to optimize the internal parameters of the core decision-making components. The ability of the policy simulator and policy optimizer to dynamically load new parameters ensures that the system can continuously improve its decision-making performance without interrupting service.

[0047] In some embodiments, when the incremental learning engine extracts the most recently added case data from the decision case library, the preset batch size or time interval is a configurable system parameter. A smaller batch size or shorter time interval allows for more frequent learning and faster adaptation to changes, but at a higher computational cost; a larger batch size or longer time interval reduces the learning frequency, which is beneficial for discovering more stable patterns, but the adaptation may be slightly slower. The data cleaning step mainly handles missing or outlier values ​​that may exist in the case data, while the feature extraction step converts complex case event sequences into numerical feature vectors that can be processed by machine learning algorithms. The choice of unsupervised learning algorithm can be adjusted according to specific needs and data types. For example, in addition to clustering and association rules, principal component analysis can also be used to reduce dimensionality and discover the main direction of change. The learning rate in the rule weight update formula... It can be set to a fixed value, or it can be designed to adjust adaptively, for example, gradually decreasing as the number of incremental learning cycles increases, to achieve rapid initial learning and fine-tuning later. The simulation execution loop parameter optimization of the policy optimizer is also based on data-driven methods, such as establishing a relationship model between the number of iterations, convergence speed, and network complexity through regression analysis, and then adjusting the parameters.

[0048] The introduction of the incremental learning engine enables the entire collaborative decision-making model construction system to learn autonomously from historical decision-making experience. The system no longer relies solely on initially set rules and parameters but can continuously optimize itself based on case data accumulated during actual operation. The decision case library is the knowledge source for the incremental learning engine, and its data quality and completeness directly affect the learning effect. Analyzing new patterns of network behavior or changes in decision-making patterns is the core objective of the unsupervised learning process, aiming to discover hidden, unprogrammed knowledge within the data. Adjusting the rule weights in the scenario rule template library essentially adjusts the priority or preference of different decision-making models in the strategy generation stage, making the system more inclined to adopt intervention models that have proven effective historically. Optimizing the simulation execution loop parameters of the policy optimizer improves the efficiency and effectiveness of the decision scenario optimization process, enabling the system to obtain better decision solutions with less computational resources. The parameter synchronization mechanism ensures that the learned knowledge can be applied to subsequent decision-making processes in a timely manner, forming a complete closed loop of "decision-execution-recording-learning-optimization." This incremental learning mechanism enables the system to adapt to the dynamic evolution of the technological ecosystem and maintain the advanced nature and effectiveness of its decisions in the long term.

[0049] See Figure 5 In the analysis of edge weight composition and evolution during the data flow-driven phase, the average initial weight distribution of different edge relationship types reflects the basic correlation strength of various collaborative relationships within the technology ecosystem industry system. Specifically, the figure presents the average initial weights of six types of edge relationships: collaborative R&D, policy constraints, financial support, information exchange, technology transfer, and material and energy flow. Collaborative R&D and policy constraints have relatively high average initial weights, reflecting a stronger basic interaction strength between industrial entities in the initial system topology network, indicating strong R&D collaboration and policy-oriented correlation. Financial support and information exchange have the next highest weights, reflecting their core position as carriers of resources and information flow. Technology transfer and material and energy flow have relatively low weights, corresponding to their interaction frequency and correlation strength characteristics in the initial stage.

[0050] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0051] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A system for constructing a collaborative decision-making model for a technology ecosystem industry based on system dynamics, characterized in that, Includes the following modules: The dynamic topology network construction module is used to establish an initial dynamic topology network for the technology ecosystem industry system. The initial dynamic topology network includes nodes representing different industry entities and technological elements, and edges representing the collaborative relationships between nodes. The data flow driving module is used to inject real-time industry operation data flow and ecological monitoring data flow into the initial dynamic topology network, driving the dynamic evolution of network node states and edge weights; The policy simulation module responds to network evolution by activating the policy simulator integrated within the network, generating various decision scenarios based on the current network topology and node states. The strategy optimization module receives the decision scenarios generated by the strategy simulation module and iteratively corrects the decision scenarios. The feedback and learning module is used to feed back the decision scenarios corrected by the strategy optimization module to the initial dynamic topology network, triggering adaptive adjustments to the network structure and recording all data trajectories during the network adjustment process to form a decision case library.

2. The system for constructing a collaborative decision-making model for the technology ecosystem industry based on system dynamics according to claim 1, characterized in that, The steps for establishing the initial dynamic topology network of the technology ecosystem industrial system by the dynamic topology network construction module include: Define node types, including technology R&D institution nodes, product manufacturing enterprise nodes, ecosystem service provider nodes, and policy and regulatory agency nodes; Define edge relationship types, which include technology transfer relationships, material and energy flow relationships, financial support relationships, and information exchange relationships; Each node is assigned a set of attribute vectors, which include productivity level, resource consumption rate, ecological footprint intensity, and innovation potential index. Each edge is assigned an initial weight value, which is calculated based on historical interaction frequency and association strength; Connect all nodes with edges to form the initial dynamic topology network.

3. The system for constructing a collaborative decision-making model for the technology ecosystem industry based on system dynamics according to claim 2, characterized in that, The specific steps of the data flow driving module injecting real-time industry operation data flow and ecological monitoring data flow into the initial dynamic topology network include: By deploying sensor arrays in industrial entities, real-time production capacity data, raw material inventory data, and energy consumption data are collected to form an industrial operation data stream; Real-time pollutant concentration data, biodiversity index, and carbon sink flow data are collected by monitoring stations deployed in ecological areas to form an ecological monitoring data stream. Perform timestamp alignment and data format standardization on industrial operation data streams and ecological monitoring data streams; The processed data stream is mapped to the corresponding node attribute vectors and edge weight values ​​in the initial dynamic topology network; Based on the mapping results, the values ​​of the node attribute vectors and the edge weights are periodically updated.

4. The system for constructing a collaborative decision-making model for the technology ecosystem industry based on system dynamics according to claim 3, characterized in that, The step of activating the policy simulator integrated within the network by the policy simulation module is controlled by network evolution triggering conditions, which are: When the change in the attribute vector of any node exceeds its preset fluctuation threshold, or the change in the weight value of any edge exceeds its preset sensitivity threshold, the network evolution trigger condition is met. Once the strategy simulator is activated, it first scans the entire initial dynamic topology network to capture the current state of all nodes and edges that have changed. The strategy simulator simulates the chain reaction paths that the network may generate under different external policy stimuli or market changes, based on the captured state.

5. The system for constructing a collaborative decision-making model for the technology ecosystem industry based on system dynamics according to claim 4, characterized in that, The process by which the strategy simulation module generates multiple decision scenarios based on the current network topology and node states includes: The strategy simulator has a built-in scenario rule template library, which defines a variety of basic intervention modes such as investment tilt, technical standard upgrade, and emission limit adjustment. The strategy simulator matches the captured network state with the scenario rule template, and fills in the specific network nodes and edge parameters for each successfully matched template; Generate a draft decision scenario that includes specific intervention targets, intervention intensity, and intervention time window; The draft decision scenarios are internally consistent, and those with logical inconsistencies are eliminated to form the final multiple decision scenarios.

6. The system for constructing a collaborative decision-making model for the technology ecosystem industry based on system dynamics according to claim 5, characterized in that, The steps of the strategy optimization module in iteratively correcting the decision scenario include: The policy optimizer receives various decision scenarios sent by the policy simulator; The strategy optimizer calls the historical decision cases stored in the decision case library and calculates the similarity between the current decision scenario and the historical cases; Based on the similarity calculation results, an initial credibility score is assigned to each decision scenario; The policy optimizer initiates a simulated execution loop, virtually applying each decision scenario to the current initial dynamic topology network, and observes the evolution trend of the network state. Based on the results of the simulation, the credibility score of the decision scenario is dynamically adjusted, and the scenario parameters are fine-tuned to complete one iteration correction. Repeat the simulation and parameter fine-tuning until the credibility score of the decision scenario stabilizes.

7. The system for constructing a collaborative decision-making model for the technology ecosystem industry based on system dynamics according to claim 6, characterized in that, The feedback and learning module feeds back the decision scenario, corrected by the strategy optimization module, to the initial dynamic topology network, triggering adaptive adjustments to the network structure. Specifically, this is manifested as follows: The policy optimizer sends the revised and stable decision scenarios to the corresponding nodes in the initial dynamic topology network in the form of virtual instructions; Based on the received virtual instructions, the node rehearses the changes in its own attribute vector and the weights of the connecting edges; The nodes submit the pre-rehearsal results to the network central coordinator in the form of local topology change proposals; The network central coordinator aggregates change proposals from all nodes and assesses the overall network stability; If the evaluation is successful, the network center coordinator approves the proposal, and the initial dynamic topology network updates its topology accordingly.

8. The system for constructing a collaborative decision-making model for the technology ecosystem industry based on system dynamics according to claim 7, characterized in that, The feedback and learning module records all data trajectories during the network adjustment process, and its detailed implementation is as follows: A data trajectory recording agent is built into the initial dynamic topology network. The data trajectory recording agent listens for all node state updates, edge weight changes, policy simulator activation, policy optimizer iteration, and network structure adjustment events in the network. The data trajectory recording agent converts each monitored event into standardized trajectory data points. The trajectory data points include event type, timestamp, involved node or edge identifier, and state before and after the change. All trajectory data points are stored in a time-series database in chronological order to form a decision case library; The decision case library indexes each stored case, with the index key being a summary of the network's global features at the time the case is triggered.

9. The system for constructing a collaborative decision-making model for the technology ecosystem industry based on system dynamics according to claim 8, characterized in that, The feedback and learning module forms a decision case library, including: Once a new decision-making process is completed and new decision cases are generated and stored in the decision case library, the incremental learning engine is activated. The incremental learning engine extracts the most recently added case data from the decision case library; Incremental learning engines analyze new patterns of network behavior or changes in decision-making rules reflected in newly added case data; Based on the analysis results, the incremental learning engine adjusts the rule weights of the scenario rule template library in the policy simulator, or optimizes the simulation execution loop parameters of the policy optimizer. The incremental learning engine synchronizes the adjusted model parameters to the policy simulator and policy optimizer, completing one incremental learning cycle.

10. The system for constructing a collaborative decision-making model for the technology ecosystem industry based on system dynamics according to claim 2, characterized in that, The step of assigning an initial weight value to each edge, wherein the initial weight value is calculated based on historical interaction frequency and association strength, includes: Retrieve all historical interaction event records related to the node pairs connected by the edges from the historical interaction database. The historical interaction event records include interaction type, interaction timestamp, and interaction scale index. Calculating historical interaction frequency involves: counting the total number of interaction events between node pairs within a preset historical time period, and normalizing the total number of interaction events by dividing it by a preset baseline frequency value to obtain a frequency factor. The calculation of historical association strength is specifically as follows: the interaction scale index of each interaction event between node pairs is quantified, and different weight coefficients are assigned according to the interaction type. Then, the weighted sum of the scales of all interaction events is calculated, and the weighted sum is divided by a preset baseline strength value for normalization to obtain the strength factor. The frequency factor and intensity factor are input into a preset weight calculation function, which is a linear combination or a multiplicative combination, and the initial weight value is output. The initial weight values ​​are subject to range constraints to ensure that they fall between the preset minimum and maximum weight values.

Images