CN-120911833-B - Industrial chain multi-relation modeling method based on dynamic relation graph

CN120911833BCN 120911833 BCN120911833 BCN 120911833BCN-120911833-B

Abstract

The invention relates to the technical field of knowledge graphs, in particular to an industrial chain multi-relation modeling method based on dynamic relation graphs, which comprises the steps of receiving industrial chain data comprising logic data and management data, constructing a logic graph frame by utilizing logic data, quantifying entity node and event relations in the graph frame by utilizing the management data, endowing dynamic quantification attributes, reasoning by a fusion analysis model, identifying risk conduction paths based on the event graph frame, calculating risk indexes of the paths by combining the dynamic quantification attributes, outputting intermediate results, converting the intermediate results into resource scheduling decision support suggestions by a strategy optimization model, taking the intermediate results as constraint conditions and excitation functions of multi-agent reinforcement learning, and generating resource scheduling suggestions based on heterogeneous intelligent body roles and structural constraints of the entities in the industrial chain. According to the invention, by constructing the dynamic relation graph, the industrial chain risk analysis and the resource optimization scheduling are realized.

Inventors

XU JIAN
Shi Anna
CHEN YINGCHAO
DENG YONG
SUN CHAO
YANG PING
YIN MINGMING
LI WEIGUO
ZHANG GUOQUAN
ZHANG XINYU
XUE JIA

Assignees

数融智联(徐州)信息科技有限公司

Dates

Publication Date: 20260508
Application Date: 20250722

Claims (7)

1. The industrial chain multi-relation modeling method based on the dynamic relation map is characterized by comprising the following steps of: receiving industry chain data, wherein the industry chain data comprises logic data and management data; utilizing the logic data to construct a logic map frame for representing the topological structure among entities and the logical sequence of the events, quantifying the relation between entity nodes and the events in the logic map frame by utilizing the operation data, and endowing the operation health degree and the transaction activity degree with dynamic quantification attributes; Reasoning is carried out through a fusion analysis model, a risk conduction path is identified based on logic map frames, a risk index of the risk conduction path is calculated by combining dynamic quantitative attributes, the whole deduction process is ensured to accord with a preset business rule, and an intermediate result which comprises risk assessment and compliance judgment and is mutually verified is output; The method comprises the steps of converting an intermediate result into resource scheduling decision support suggestions through a strategy optimization model, taking the intermediate result as constraint conditions and an incentive function of multi-agent reinforcement learning, generating resource scheduling suggestions based on heterogeneous agent roles and structural constraints of all entities in an industrial chain, wherein the process of the heterogeneous agent roles and the structural constraints of all the entities in the industrial chain comprises the steps of mapping entity nodes in a dynamic relation map into heterogeneous agent roles according to business functions and industry positioning of the entity nodes, the heterogeneous agent roles comprise core manufacturing enterprises, upstream suppliers, downstream distributors and logistics service providers, configuring a dedicated action space and an observation space for each intelligent agent role, the action space defines resource scheduling operation executable by the heterogeneous agent roles, the observation space defines decision information obtainable by the heterogeneous agent roles, converting the topological connection relation in the industrial chain into interaction constraints in a multi-agent reinforcement learning environment, limiting an interaction mode of information flow and material flow between the intelligent agent, setting a directional channel from an entity A to an intelligent agent B to the intelligent agent, simultaneously enabling a communication channel to be updated to the intelligent agent B, and simultaneously enabling the communication channel to be set up to be in the reinforcement channel from the intelligent agent A to the intelligent agent A.
2. The method for modeling multiple relationships in an industrial chain based on a dynamic relationship graph of claim 1, wherein logic data includes business information, lawsuit information, and contract performance information, and wherein the business data includes business financial statement data, tax declaration data, and supply chain transaction data.
3. The industrial chain multi-relation modeling method based on dynamic relation graphs of claim 1, wherein the process of constructing a logic graph frame representing topological structures among entities and event logic sequences by utilizing logic data comprises the steps of processing logic data by utilizing a pre-trained industrial chain domain language model, synchronously executing event trigger word recognition and entity recognition, extracting event argument based on the event trigger word to obtain structured event information, analyzing the structured event information by utilizing an event association evaluation model, evaluating probability dependency relations among events to construct an event chain representing potential risk conduction logic, taking the identified entity as an entity node, taking the structured event information as an event node, taking the evaluated event chain as a directed edge connected with the event node, and constructing the logic graph frame.
4. The industrial chain multi-relation modeling method based on the dynamic relation map is characterized in that the process of quantifying the relation between entity nodes and events in a map frame by utilizing the operation data comprises the steps of establishing a node index system representing the entity operation state and an edge index system representing the interaction strength of the relation, calculating the operation data according to the node index system and the edge index system to generate a node attribute value and an edge attribute value, and respectively and periodically attaching the generated node attribute value and edge attribute value to the entity node and the event relation in the logic map frame to generate the dynamic relation map.
5. The method for modeling multiple relationships in an industrial chain based on dynamic relationship graphs according to claim 4, wherein the fusion analysis model comprises: the diagram representation layer is used for receiving the dynamic relation map and converting the dynamic relation map into a vectorized representation which can be processed by the model; a path recognition layer for performing graph traversal calculations on the vectorized representation, searching and outputting all potential risk conduction paths; The risk assessment layer is used for carrying out comprehensive risk assessment by combining dynamic quantitative attributes of all nodes on each risk conduction path and outputting a risk index; and the compliance verification layer is used for loading a preset rule base, verifying the risk conduction paths one by one and outputting a compliance judgment result.
6. The method for modeling multiple relationships in an industrial chain based on dynamic relationship graphs according to claim 1, wherein the policy optimization model comprises: the environment perception layer is used for receiving the intermediate result and analyzing the intermediate result into state information, action constraint and rewarding signals which can be observed by the intelligent agent; The value network layer is used for evaluating long-term expected returns of executing different actions according to the current state information and outputting value evaluation data; the policy network layer is used for receiving the state information and the value evaluation data, calculating and outputting the optimal action probability distribution to be executed in the current state; and the action output layer is used for sampling according to the optimized action probability distribution, and determining and outputting the final decision action of the current intelligent agent.
7. The industrial chain multi-relation modeling method based on the dynamic relation graph of claim 1 is characterized in that the process of taking the intermediate result as a constraint condition and an incentive function of multi-agent reinforcement learning comprises the steps of setting a risk conduction path and a compliance judgment result in the intermediate result as constraint conditions for policy evaluation in a training environment containing historical cases, constructing a risk evaluation aggregate value in the intermediate result as an incentive function of the multi-agent reinforcement learning environment by combining a real final result in the historical cases, performing offline learning and policy mining on historical data in a static training environment by utilizing agents representing different roles in an industrial chain, and outputting a policy set which is subjected to learning convergence and has highest association with the historical successful cases as resource scheduling suggestions.

Description

Industrial chain multi-relation modeling method based on dynamic relation graph Technical Field The invention relates to the technical field of knowledge graphs, in particular to an industrial chain multi-relation modeling method based on a dynamic relation graph. Background Modern industry chain structures are increasingly complex, and networking and dynamic characteristics are presented. The fluctuations of the operations of any entity on the chain may be conducted through various relationship paths such as supply chain, guarantee chain, and fund chain, and cause systematic risks. Therefore, it is important to perform efficient analysis and risk management on the industrial chain. Existing industry chain analysis methods often have limitations. On one hand, the traditional method depends on static financial reports or isolated operation data, and is difficult to describe the complex staggered dynamic relationship among entities, so that the prediction capability of risk conduction paths and probabilities is insufficient. On the other hand, although the prior art adopts a knowledge graph to represent the relationship between entities, the prior art is mostly limited to static relationship display, and can not effectively fuse the operation data with strong variability. In addition, the attempt of risk prediction by partially adopting an artificial intelligent model also causes that the analysis result is not strong in interpretation and is difficult to be used for guiding actual collaborative decisions due to the lack of understanding of the internal logic of the industrial chain by the model. Therefore, an industrial chain multi-relation modeling method based on dynamic relation graphs is provided. Disclosure of Invention The invention aims to provide an industrial chain multi-relation modeling method based on a dynamic relation map, which realizes industrial chain risk analysis and resource optimization scheduling by constructing the dynamic relation map. The method comprises the steps of receiving industrial chain data comprising logic data and management data, constructing logic map frames by utilizing logic data, quantifying the relation between entity nodes and events in the map frames by utilizing the management data, endowing dynamic quantification attributes, reasoning by means of a fusion analysis model, identifying risk conduction paths based on the event map frames, calculating risk indexes of the paths by combining the dynamic quantification attributes, outputting intermediate results, converting the intermediate results into resource scheduling decision support suggestions by means of a strategy optimization model, taking the intermediate results as constraint conditions and excitation functions of multi-agent reinforcement learning, and generating resource scheduling suggestions based on heterogeneous intelligent body roles and structural constraints of the entities in the industrial chain. In order to achieve the above purpose, the present invention provides the following technical solutions: an industrial chain multi-relation modeling method based on a dynamic relation graph comprises the following steps: receiving industry chain data, wherein the industry chain data comprises logic data and management data; utilizing the logic data to construct a logic map frame for representing the topological structure among entities and the logical sequence of the events, quantifying the relation between entity nodes and the events in the logic map frame by utilizing the operation data, and endowing the operation health degree and the transaction activity degree with dynamic quantification attributes; Reasoning is carried out through a fusion analysis model, a risk conduction path is identified based on an event map frame of the map, a risk index of the risk conduction path is calculated by combining dynamic quantitative attributes, the whole deduction process is ensured to accord with a preset business rule, and an intermediate result which comprises risk assessment and compliance judgment and is mutually verified is output; And converting the intermediate result into a resource scheduling decision support suggestion through a strategy optimization model, taking the intermediate result as a constraint condition and an incentive function of multi-agent reinforcement learning, and generating the resource scheduling suggestion based on heterogeneous agent roles and structural constraints of each entity in an industrial chain. Preferably, the logic data includes enterprise business information, lawsuit information and contract performance information, and the management data includes enterprise financial statement data, tax declaration data and supply chain transaction data. Preferably, the process of constructing the logic atlas framework for representing the topological structure and the event logic sequence among entities by utilizing logic data comprises the steps of processing logic data by utilizing