Intelligent decision-making system based on value network

Abstract

The invention provides an intelligent decision-making system based on a value network, and aims to realize optimization decision-making of a complex system through multi-module cooperation. The system models a function module or a service node into a multi-dimensional attribute vector through a node creation module; the node association module constructs a multi-dimensional tensor based on the multi-dimensional attribute vector, forms a value network, and describes an association relationship between nodes; the value calculation module calculates the actual contribution value of the node through the multi-dimensional attribute vector and the tensor; the cost aggregation module summarizes the total cost of the system; the dynamic adjustment module accesses a multi-source heterogeneous data stream in real time, and dynamically adjusts node attributes and association relationships; and the optimization decision module generates an optimization scheme based on the contribution value and the total cost and predicts the use effect of the optimization scheme. According to the system, intelligent decision optimization in a complex system is realized through value network modeling and a dynamic adjustment mechanism, and the system is suitable for multi-field service scenes.

Description

Technical Field

[0001] This application relates to the field of intelligent decision-making technology, and particularly relates to an intelligent decision-making system based on a value network. Background Art

[0002] With the development of enterprise scale, correct enterprise decision-making is becoming increasingly important for enterprise development. However, the current enterprise decision-making system still has the following problems: Traditional decision-making systems are mostly based on static index evaluation, and it is difficult to capture the dynamic influence relationship between business nodes; existing value evaluation models often only focus on direct business value and ignore the non-linear synergy effects of technical value and market value; multi-dimensional decision-making factors (such as value, cost, risk, etc.) are usually processed separately, lacking a unified quantitative framework; the decision-making process usually relies on manual experience judgment and is difficult to adapt to the rapidly changing market environment and emergencies.

[0003] Therefore, there is a need for an intelligent decision-making system that can dynamically capture the complex associations between business nodes and comprehensively consider multi-dimensional factors to improve the scientificity, adaptability, and global optimization ability of enterprise decision-making. Summary of the Invention

[0004] In view of the above problems, this application provides an intelligent decision-making system based on a value network, which is used to provide interpretable and trustworthy decision-making suggestions for enterprises and realize the global value optimization in the enterprise decision-making process.

[0005] The specific solution of this application is as follows: An intelligent decision-making system based on a value network, comprising: A node creation module, configured to model each functional module or business node in the system as a multi-dimensional attribute vector , where represents value, represents cost, represents risk, represents technological advancement, represents market demand intensity; A node association module, configured to construct a multi-dimensional tensor based on the multi-dimensional attribute vector to describe the association relationship between nodes; A value calculation module, configured to calculate the actual contribution value of each node in the network based on the multi-dimensional attribute vector and the multi-dimensional tensor ; A cost aggregation module, configured to calculate the total system cost ; A dynamic adjustment module, configured to access multi-source heterogeneous data streams in real time and dynamically adjust the relevant node attribute vectors and association relationships based on the data flow; An optimization decision module, based on the actual contribution value and the total system cost Generate an optimized decision-making plan and predict the usage effect of the optimized decision-making plan.

[0006] Preferably, the system further includes an adaptive module, which dynamically adjusts the relevant parameters and structure of the system through establishing a feedback mechanism, including the multi-dimensional attribute vector and the multi-dimensional tensor.

[0007] The adaptive module dynamically adjusts the relevant parameters and structure of the system (such as the multi-dimensional attribute vector and the multi-dimensional tensor) through the feedback mechanism, enabling the system to automatically optimize according to the actual operation situation, and avoiding the lag and subjectivity of manual intervention. This adaptive ability enhances the flexibility and robustness of the system and can better cope with the complex and changeable business environment.

[0008] Preferably, the feedback mechanism specifically includes: Short-cycle feedback, which is used to monitor key indicators in real time, including the value realization rate, the cost deviation rate, and the risk event frequency, and triggers an alarm when the deviation of the key indicators exceeds the first threshold; Medium-cycle feedback, which collects the difference between the usage results of the actual optimized decision-making plan and the predicted usage effect of the optimized decision-making plan at preset time intervals, and introduces a Bayesian model to dynamically adjust the relevant parameters of the system based on this difference; Long-cycle feedback, which collects the achievement rate of the enterprise's annual assessment business strategic objectives and optimizes the structure of the relevant tensors based on this achievement rate.

[0009] The short-cycle feedback monitors key indicators (such as the value realization rate, the cost deviation rate, and the risk event frequency) in real time, which can quickly discover problems and trigger an alarm, reducing potential losses. The medium-cycle feedback dynamically adjusts parameters through the Bayesian model, optimizes the system based on the difference between the actual and the predicted, and improves the scientificity of decision-making. The long-cycle feedback optimizes the tensor structure through annual assessment, ensuring that the system can long-term adapt to the enterprise strategic objectives and achieving continuous improvement.

[0010] Preferably, the data of the node attribute vector is obtained from the enterprise internal system and the external data interface.

[0011] By integrating the data of the enterprise internal system and the external data interface, the system can obtain comprehensive and real-time information, ensuring the accuracy of the node attribute vector. The access of this multi-source data enhances the data coverage and timeliness of the system and improves the quality of the decision-making basis.

[0012] Preferably, the value calculation module calculates the actual contribution value of the node through the following formula : ; where is the number of iteration steps, and the upper limit of the total number of iterations is ; is the attenuation factor ; is the technological advancement , the market demand intensity , and the risk dimension factor combination; is the node initial contribution value; is the total number of nodes; is the value gain coefficient between node i and node j.

[0013] By calculating the actual contribution value of the node through the above formula, it is possible to comprehensively consider multi-dimensional factors such as technological advancement, market demand intensity, and risk, and dynamically adjust the contribution weight of the node. This calculation method can more accurately reflect the actual value of the node in the network and avoid the one-sidedness of single-dimensional evaluation.

[0014] Preferably, the cost aggregation module calculates the total system cost , the collaboration overhead , and the risk penalty term , and the specific formula is as follows: ; ; where is the cost after collaborative discount of node i; is the node direct cost; is the average value of the cost collaboration coefficient between node i and other nodes; is the discount function, and the specific expression is: ; where, is an adjustable parameter that controls the intensity of the collaborative effect; is the collaboration overhead of node i, ; where, is the overall collaboration cost weight coefficient; represents the intensity of the dependence relationship; represents the intensity of risk conduction; Represents the weighted sum of the dependence relationship strength and the risk conduction strength, and is specifically calculated using the following formula : ; ; ; Among them, 、 are hyperparameters for adjusting the influence degrees of different factors; represents the dependence relationship; represents the risk conduction coefficient.

[0015] is the risk penalty term of node i, λ is the risk penalty coefficient, is the cost of the node with the highest cost in the system.

[0016] Through the comprehensive calculation of the cost after collaborative discount, the collaboration overhead, and the risk penalty term, the system can comprehensively quantify the cost and consider the collaborative effect and risk impact among nodes. This cost aggregation method can more realistically reflect the total cost of the system and provide a reliable basis for optimization decisions.

[0017] Preferably, the dynamic adjustment module includes: A data stream access unit for accessing financial market APIs, supply chain monitoring systems, and policy analysis data; A classification and response unit for constructing a mapping engine and using the accessed data to identify key events and trigger the update of relevant parameters; A dynamic propagation unit for starting a limited-depth propagation update when a key event affects a node.

[0018] By accessing financial market APIs, supply chain monitoring systems, and policy analysis data, the dynamic adjustment module can obtain real-time changes in the external environment. The classification and response unit identifies key events through the mapping engine and triggers parameter updates, and the dynamic propagation unit starts a limited-depth propagation update when a key event affects a node, ensuring that the system can quickly respond to external changes and improve real-time performance and adaptability.

[0019] Preferably, the dynamic propagation unit includes: A direct update layer for directly correcting the attribute vector of the affected node; A first-degree propagation layer for traversing all nodes whose association relationship with the affected node is greater than a second threshold and updating their attribute vectors; A second-degree propagation layer for performing a lightweight update on the nodes whose association relationship with the affected node is greater than the second threshold, and only adjusting the cost and risk-related parameters.

[0020] The direct update layer can quickly correct the attribute vectors of the affected nodes. The first-degree propagation layer traverses the nodes with relatively large correlation relationships and updates their attribute vectors. The second-degree propagation layer performs lightweight updates on the nodes with relatively large correlation relationships. This hierarchical propagation mechanism can effectively control the update scope, avoid unnecessary computational overhead, and ensure timely adjustment of key nodes.

[0021] Preferably, the system further includes an anomaly detection module for early detection of potential problems. The anomaly detection module specifically includes: A single-node detection unit for real-time monitoring of the attribute vectors of each node and identifying nodes that deviate from the normal pattern through the local outlier factor algorithm; A relationship network detection unit for discovering mutations in the correlation relationship structure through graph embedding and graph anomaly detection algorithms.

[0022] Through single-node detection and relationship network detection, the anomaly detection module can real-time monitor the anomalies of node attribute vectors and correlation relationship structures, and discover potential problems in advance. This mechanism can help enterprises take timely measures to avoid systemic risks and improve the stability and reliability of the system.

[0023] Preferably, the system is provided with a decision-making system protection mechanism. When an abnormal pattern is detected, the system automatically degrades to a conservative decision-making mode and initiates an isolation impact assessment for the abnormal nodes.

[0024] This protection mechanism can effectively prevent the negative impact of abnormal situations on the overall operation of the system and ensure the stability and security of the system in abnormal situations.

[0025] Compared with the prior art, the beneficial effects of the present application are: Through multi-dimensional modeling and dynamic adjustment mechanisms, the present application can comprehensively quantify multi-dimensional factors such as value, cost, and risk, break through the limitations of traditional fragmented processing, and achieve global optimization. The system introduces an event-driven mechanism to respond in real-time to external changes such as the market, technology, and policies, dynamically adjust decision-making strategies, and significantly improve the timeliness and adaptability of decision-making. At the same time, by finely modeling non-linear cost factors such as collaborative discounts and collaborative overheads, the accuracy of cost prediction is improved. In addition, the system also constructs an anomaly detection mechanism to discover potential problems in advance and trigger early warnings to ensure the stable operation of the system. In short, the present invention provides an enterprise with a scientific, efficient, and reliable intelligent decision-making support solution. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] To more clearly illustrate the technical solutions in the embodiments of the present drawings or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present drawings. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on the structures shown in these drawings.

[0027] Figure 1 This is a schematic diagram of the system principle of the present application. Detailed implementation manners

[0028] In order to make the objectives, technical solutions and advantages of the present application clearer, the following will describe and explain the present application in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application. Based on the embodiments provided in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present application.

[0029] As Figure 1 shown, the present application provides an intelligent decision-making system based on a value network, specifically including the following six modules: A node creation module, which is used to model each functional module or business node in the system as a multi-dimensional attribute vector , where represents value, which may include direct or indirect commercial value and user value; represents cost, including development, maintenance, manpower, funds, etc.; represents risk, which is a comprehensive measure of technical risk and market risk; represents technological advancement, such as technology barrier score, number of patents, etc.; represents the intensity of market demand, which is obtained from indicators such as user research or market growth rate.

[0030] For the technology industry, technological advancement can be comprehensively calculated from indicators such as technology barrier score and number of patents. For other industries, can be replaced with key influencing factors that conform to the characteristics of the industry.

[0031] The data of the node attribute vector is mainly obtained from the enterprise internal system and external data interfaces. The enterprise internal system obtains relevant data through the API to connect to cost data, user value data, technical attributes, and historical risk event libraries. The external data interface integrates the market research API to obtain the intensity of market demand in real time , and extracts industry risk indicators through the policy analysis interface.

[0032] Specifically, for example, a large e-commerce platform plans to develop multiple new functions in the next quarter, including payment system upgrade, improvement of personalized recommendation algorithms, live streaming e-commerce functions, cross-border logistics optimization, and membership points system overhaul. Due to limited resources, it is necessary to determine the optimal priority order for function development. The system can collect historical data of each function module by docking with systems such as ERP and CRM through APIs, as shown in Table 1:

[0033] The node association module constructs a multi-dimensional tensor based on multi-dimensional attribute vectors to describe the association relationship between nodes and nodes.

[0034] For example, constructing a three-dimensional tensor , the first and second dimensions (N×N) represent the relationship between node pairs, and the third dimension (5) represents the type or category of the relationship. is the total number of nodes, and use to describe the influence or coefficient of node i and node j in the k-th type of relationship, where .

[0035] The above tensor contains the following association relationships: Value gain / loss (k = 1): Positive or negative value synergy between functions / businesses; Cost synergy (k = 2): Cost reduction due to shared resources or economies of scale; Risk conduction (k = 3): The transmission coefficient that the risk of a certain node may cause other nodes to be delayed or fail; Dependency relationship (k = 4): The dependency relationship generated by the mutual call of APIs or interfaces between modules; Requirement association (k = 5): The impact on other nodes caused by the change in requirements or user growth of one node.

[0036] The initial values of the above association relationships can be obtained based on the statistical analysis of historical data. For the value gain coefficient (k = 1), the A / B test is used to fit the cross-node synergy effect. For example, in the e-commerce scenario, if the GMV of associated function B increases by 15% after function A is launched, then initialize ; or it can be obtained through the expert scoring-fuzzy logic hybrid method. For the association relationship types lacking data (such as risk conduction k = 3), design expert scoring rules: First, experts score the risk conduction intensity between nodes according to {none, weak, medium, strong}; then, the qualitative score is converted into an interval value through triangular fuzzy numbers, such as "weak" corresponding to (0.1, 0.2, 0.3); finally, the initial coefficient is generated by defuzzification.

[0037] Specifically, the system analyzes historical A / B test data to identify key association relationships, such as: The value gain between the payment system and live streaming e-commerce , indicating that the optimization of the payment experience significantly improves the live broadcast conversion rate; Cost synergy between personalized recommendation and membership points , indicating that co - development can save data processing costs; Risk conduction between cross - border logistics and payment systems , indicating that logistics delays have a chain risk on the payment link.

[0038] Value calculation module, calculating the actual contribution value of each node in the network based on multi - dimensional attribute vectors and multi - dimensional tensors , and the specific calculation formula is as follows: ; Among them is the number of iteration steps, and the upper limit of the total number of iterations is set to ; is the attenuation factor , ensuring the convergence of value propagation; When , it terminates, is the third threshold set in advance; is the technological advancement , the intensity of market demand , risk combination of dimensional factors, , used to comprehensively calculate the comprehensive impact of technological advancement, market demand intensity, and risk on value propagation, indicates the amplification or attenuation of technological advancement on the value transfer efficiency, indicates the strengthening effect of market demand intensity on value diffusion, indicates the negative impact of risk on value propagation; is the initial contribution value of node ; is the total number of nodes.

[0039] Cost aggregation module, used to calculate the total system cost .

[0040] When calculating the total system cost, on the basis of considering the costs of each node itself, it is also necessary to consider the cost reduction caused by the cooperation between modules and the additional costs caused by multi - module coupling, risk transfer, etc.

[0041] Specifically, first calculate through the collaborative discount of single - node costs.

[0042] Introduce a discount function to reflect the cost synergy effect between nodes: ; Among them, is an adjustable parameter that controls the intensity of the synergy effect; is the node and the average value of other nodes in the cost synergy coefficient .

[0043] Based on this, the cost after collaborative discount of node is expressed as: , is the direct cost of node .

[0044] For example, when the system calculates the cost after collaborative discount of personalized recommendation, adjust to 0.2, calculate that the average value of personalized recommendation and other nodes in the cost synergy coefficient is 0.35, and the direct cost of personalized recommendation is 0.5 from Table 1. Thus, the cost after collaborative discount of personalized recommendation can be calculated: .

[0045] Secondly, conduct collaborative overhead modeling.

[0046] Define the collaborative overhead of node as: ; Among them, is the overall collaborative cost weight coefficient; represents the intensity of the dependency relationship; represents the intensity of risk conduction; represents the weighted sum of the intensity of the dependency relationship and the intensity of risk conduction. The larger the value, the higher the additional cost caused by collaborative management, communication, integration, risk, etc. Specifically, it is calculated using the following formula : ; ; ; Among them, , are hyperparameters that adjust the influence degree of different factors. If a value is higher, it means that the factor has a greater impact on the collaborative overhead; When , are both 0, , and no additional collaborative cost is added; If any one of the factors increases, it will be even greater, indicating that the collaboration overhead increases accordingly; represents the dependency relationship; represents the risk conduction coefficient.

[0047] For example, set to 0.2, and adjust and to 2. From Table 1, the direct cost of cross-border logistics is 0.8, and through the node association module, and are obtained. From this, the collaboration overhead between cross-border logistics and the payment system can be calculated: .

[0048] Finally, calculate the total system cost , and the specific calculation is carried out using the following formula: ; where is the risk penalty term of node , which is used to prevent a single node from having an excessively high cost and dragging down the whole. λ is the risk penalty coefficient, which is used to adjust the influence degree of the risk penalty term, is the cost of the node with the highest cost in the system.

[0049] The dynamic adjustment module is used to access multi-source heterogeneous data streams in real time and dynamically adjust the relevant node attribute vectors and association relationships based on this data flow.

[0050] Specifically, the dynamic adjustment module includes: The data stream access unit is used to access financial market APIs, supply chain monitoring systems, and policy analysis data, and capture events such as market value fluctuations, technological breakthroughs, supply chain interruptions, and policy mutations in real time; The classification and response unit is used to build a mapping engine and use the accessed data to identify key events and trigger the update of relevant parameters; for example, build an "event-impact mapping engine" as shown in Table 2:

[0051] The dynamic propagation unit is used to start a limited-depth propagation update when a key event affects a node, specifically including: The direct update layer is used to directly correct the attribute vector of the affected node; The first-degree propagation layer is used to traverse all nodes whose association relationship with the affected node is greater than the second threshold and update their attribute vectors; The secondary propagation layer performs lightweight updates on the nodes whose association relationships with the affected nodes are greater than the second threshold, and only adjusts the parameters related to cost and risk.

[0052] For example, when a strike breaks out at a cross-border logistics node, the risk attribute of the cross-border logistics node increases by 50%. The system will mark this node as high-risk and spread the risk along the conduction path of this node to the payment system node and the live streaming e-commerce node that depend on it, thus triggering the recalculation of the collaboration overhead.

[0053] The optimization decision-making module, based on the actual contribution value and the total system cost generates an optimization decision-making plan and predicts the usage effect of the optimization decision-making plan.

[0054] Specifically, based on the actual contribution value and the total system cost the priority order of function development is determined. For example, through the calculations of the above value calculation module and cost aggregation module, the data in Table 3 are obtained:

[0055] Based on Table 3, the following suggestions for the priority order of function development can be obtained: develop the membership points, live streaming e-commerce, and personalized recommendation functions in sequence, and the remaining payment system and cross-border logistics functions can be flexibly arranged according to the actual resource situation.

[0056] Furthermore, the system also includes an adaptive module. The adaptive module dynamically adjusts the relevant parameters and structure of the system, including the multi-dimensional attribute vector and multi-dimensional tensor, by establishing a feedback mechanism.

[0057] To enable the system to continuously self-optimize, three-layer feedback loops are established, including: Short-cycle feedback, which is used to monitor key indicators in real time, including the value realization rate, cost deviation rate, and risk event frequency. When the deviation of the key indicators exceeds the first threshold, an early warning is triggered; Medium-cycle feedback, which collects the differences between the usage results of the actual optimization decision-making plan and the predicted usage effects of the optimization decision-making plan at preset time intervals, and introduces a Bayesian model to dynamically adjust the relevant parameters of the system based on this difference; Long-cycle feedback, which collects the achievement rate of the enterprise's annual evaluation business strategic objectives and optimizes the structure of the relevant tensors based on this achievement rate.

[0058] Furthermore, the system also includes an anomaly detection module for early discovery of potential problems. The anomaly detection module specifically includes: A single-node detection unit, which is used to monitor the attribute vectors of each node in real time and identify the nodes that deviate from the normal mode through the local outlier factor algorithm; The relationship network detection unit is used to discover mutations in the associated relationship structure through graph embedding and graph anomaly detection algorithms. Typical anomaly patterns include the formation of isolated nodes, the breakage of propagation paths, and abnormal fluctuations in relationship strength.

[0059] The system is set with a decision system protection mechanism. When an abnormal pattern is detected, the system automatically degrades to a conservative decision-making mode, and starts an isolation impact assessment for abnormal nodes to avoid the spread of wrong decisions.

[0060] The intelligent decision-making system based on the value network proposed in this application realizes the global value optimization of the enterprise decision-making process through innovative technologies such as multi-dimensional attribute modeling, tensor association representation, two-way value engine, event-driven dynamic adjustment, and multi-modal decision support. The system can not only accurately capture the complex value associations between business nodes, but also dynamically respond to changes in the internal and external environments, and provide interpretable and trustworthy decision-making suggestions.

[0061] It should be noted that this application is not limited to the above embodiments. The above embodiments are only examples, and embodiments with the same composition and the same effect as the technical idea within the technical solution scope of this application are included in the technical scope of this application. In addition, within the scope of not departing from the main idea of this application, various deformations that can be thought of by those skilled in the art imposed on the embodiments, and other ways constructed by combining some constituent elements in the embodiments are also included in the scope of this application.

Claims

1. An intelligent decision-making system based on value network, characterized in that: include: Node creation module, used to model each functional module or business node in the system as a multi-dimensional attribute vector ,in, Indicates value, Indicates the cost, Indicates risk, Indicates technological advancement. Indicates the intensity of market demand; The node association module builds a multidimensional tensor based on the multidimensional attribute vector to describe the association relationship between nodes; The value calculation module calculates the actual contribution value of each node in the network based on multi-dimensional attribute vectors and multi-dimensional tensors ; Cost aggregation module to calculate the total system cost ; A dynamic adjustment module is used to access multi-source heterogeneous data streams in real time and dynamically adjust the attribute vectors and association relationships of related nodes based on the data streams; Optimize decision-making modules based on actual contribution value and total system cost Generate an optimized decision plan and predict the effect of using the optimized decision plan.

2. The intelligent decision-making system according to claim 1, characterized in that: The system further comprises an adaptive module, which dynamically adjusts relevant parameters and structures of the system, including the multi-dimensional attribute vector and the multi-dimensional tensor, by establishing a feedback mechanism.

3. The intelligent decision-making system according to claim 2, characterized in that: The feedback mechanism specifically includes: Short-loop feedback is used to monitor key indicators in real time, including value realization rate, cost deviation rate, and risk event frequency. When the deviation of key indicators exceeds the first threshold, an early warning is triggered; The middle loop feedback collects the difference between the actual use results of the optimization decision-making scheme and the predicted use results of the optimization decision-making scheme at a preset time, and introduces a Bayesian model to dynamically adjust the relevant parameters of the system based on the difference; Long-loop feedback collects the achievement rate of the company's annual business strategy goals and optimizes the structure of related tensors based on the achievement rate.

4. The intelligent decision-making system according to claim 1, characterized in that: The data of the node attribute vector is obtained from the enterprise internal system and the external data interface.

5. The intelligent decision-making system according to claim 1, characterized in that: The value calculation module calculates the actual contribution value of the node through the following formula : ； in is the number of iteration steps, and the upper limit of the total number of iterations is ; is the attenuation factor ; For technological advancement , Market demand intensity , Risk Dimension Factor combination of; For Node The initial contribution value of is the total number of nodes; is the value gain coefficient between node i and node j.

6. The intelligent decision-making system according to claim 1, characterized in that: The cost aggregation module is based on the collaborative discounted cost , collaboration expenses and risk penalty items Calculate total system cost , the specific formula is as follows: ； in is the discounted coordination cost of node i; is the discount function; is the direct cost of node i; is the average value of the cost coordination coefficient between node i and other nodes; is the cooperation cost of node i; is the risk penalty term for node i, λ is the risk penalty coefficient, is the cost of the most expensive node in the system.

7. The intelligent decision-making system according to claim 1, characterized in that: The dynamic adjustment module comprises: Data stream access unit, used to access financial market APIs, supply chain monitoring systems and policy analysis data; The classification and response unit is used to build a mapping engine and use the accessed data to identify key events and trigger the update of related parameters; Dynamic propagation unit, used to initiate limited depth propagation updates when critical events affect nodes.

8. The intelligent decision-making system according to claim 7, characterized in that: The dynamic propagation unit comprises: Direct update layer, used to directly modify the attribute vector of the affected node; The first propagation layer is used to traverse all nodes whose association relationship with the affected node is greater than the second threshold and update their attribute vectors; The second-degree propagation layer performs a lightweight update on the nodes whose association relationship with the affected nodes is greater than a second threshold, and only adjusts the cost and risk related parameters.

9. The intelligent decision-making system according to claim 1, characterized in that: The system also includes an anomaly detection module for detecting potential problems in advance, and the anomaly detection module specifically includes: Single node detection unit, used to monitor the attribute vector of each node in real time and identify nodes that deviate from the normal mode through the local anomaly factor algorithm; The relational network detection unit is used to discover mutations in the relational structure through graph embedding and graph anomaly detection algorithms.

10. The intelligent decision-making system according to claim 9, characterized in that: The system is provided with a decision system protection mechanism. When an abnormal mode is detected, the system automatically downgrades to a conservative decision mode and initiates an isolation impact assessment for the abnormal node.

Images