CN-121437163-B - State-run assets intelligent supervision method based on large model

CN121437163BCN 121437163 BCN121437163 BCN 121437163BCN-121437163-B

Abstract

The invention relates to the technical field of state-run assets intelligent supervision and discloses a state-run assets intelligent supervision method based on a large model. The method acquires full life cycle circulation data and real-time environment parameters of national assets, constructs state-run assets supervision feature vectors containing space-time dimensions, and realizes comprehensive integration and association of asset data. And merging the historical supervision records and the compliance standard library to generate a multi-level state-run assets supervision index system, and covering supervision requirements of different dimensions and levels. And then, a multi-supervision model collaborative framework containing a dynamic topological interactive structure is designed by utilizing the index system, and the analysis capability is improved through dynamic collaboration among the models. And outputting state-run assets a risk quantification evaluation result based on the dynamic topology interaction structure, generating a dynamic supervision strategy vector according to the risk quantification evaluation result, and finally triggering a supervision instruction execution engine. The method realizes the comprehensiveness, the dynamics and the accuracy of state-run assets supervision, and provides an effective path for intelligent supervision of the national full life cycle of the assets.

Inventors

XIONG YONGGANG
CHEN LI
ZHU YANGYONG
YANG FEI

Assignees

贵阳思普信息技术有限公司

Dates

Publication Date: 20260512
Application Date: 20251230

Claims (8)

1. State-run assets intelligent supervision method based on large model is characterized by comprising the following steps: Acquiring national asset full life cycle circulation data and real-time environment parameters, and constructing state-run assets supervision feature vectors containing space-time dimensions; based on the state-run assets supervision feature vector, a history supervision record and a compliance standard library are fused, and a multi-level state-run assets supervision index system is generated; designing a multi-supervision model collaboration framework by utilizing the multi-level state-run assets supervision index system, wherein the multi-supervision model collaboration framework comprises a dynamic topology interaction structure; outputting state-run assets a risk quantification evaluation result based on the dynamic topological interaction structure; generating a dynamic supervision policy vector according to the state-run assets risk quantification evaluation result; Triggering a supervision instruction execution engine based on the dynamic supervision policy vector; based on the dynamic topology interaction structure, outputting state-run assets the risk quantization evaluation result includes: According to the dynamic topology interaction structure, deploying a large model driven supervision weight distribution mechanism to generate an initial hierarchical supervision model; based on the initial hierarchical supervision model, combining a risk threshold rule base and a compliance constraint condition to construct a double-layer optimized supervision framework; implementing alignment processing of multi-source supervision data through the double-layer optimization supervision framework, and updating supervision model parameters; based on the updated supervision model parameters, performing constraint-driven iterative optimization calculation, and outputting state-run assets risk quantization evaluation results; The construction of the double-layer optimized supervision architecture comprises the following steps: Setting an upper model to minimize state-run assets abnormal fluctuation rate as an objective function; setting a lower model to maximize full-period supervision benefit as an objective function; loading boundary constraint conditions in a risk threshold rule base; Integrating business rule restrictions in compliance with constraint conditions; Binding the objective function and the constraint condition to the initial hierarchical supervision model, and constructing a double-layer optimized supervision framework.
2. The large model-based state-run assets intelligent supervision method according to claim 1, wherein the constructing state-run assets supervision feature vectors containing space-time dimensions includes: Analyzing state-run assets the timestamp information of the circulating data to generate a dynamic supervision time axis and adding event marks to the key supervision nodes; fusing the geographic information system coordinates and the asset position track to construct state-run assets spatially distributed thermodynamic diagrams; Integrating the dynamic supervision time axis and state-run assets space distribution thermodynamic maps, and generating a space-time fusion feature matrix through a knowledge map embedding technology; and performing dimension compression processing on the space-time fusion feature matrix, and outputting state-run assets supervision feature vectors.
3. The large model-based state-run assets intelligent monitoring method as claimed in claim 2, wherein the generating a multi-level state-run assets monitoring metrics system includes: extracting fluctuation mode characteristics of the state-run assets supervision feature vector, and dividing a high-frequency risk index and a low-frequency steady-state index; associating the rule-breaking case characteristics in the history supervision record, and marking the abnormal association rule of the index; Integrating threshold boundary conditions in a compliance standard library, and establishing an index hierarchical mapping relation; and generating a multi-level state-run assets supervision index system according to the high-frequency risk index and the low-frequency steady-state index and combining the index abnormal association rule and the index hierarchical mapping relation.
4. The large model-based state-run assets intelligent supervisory method according to claim 3, wherein the designing a multi-supervisory model collaboration framework includes: Configuring a supervision model interaction node according to the hierarchical structure of the multi-level state-run assets supervision index system; Based on the supervision task dependency relationship, constructing a dynamic topology interaction structure among nodes; Setting a model collaborative updating rule, and defining a parameter transmission path between supervision models; And connecting a plurality of supervision model nodes through the dynamic topology interaction structure to form a multi-supervision model collaborative framework.
5. The large model based state-run assets intelligent supervisory method as set forth in claim 4 wherein said deploying a large model driven supervisory weight allocation mechanism comprises: inputting an index weight initial value of the multi-level state-run assets supervision index system; Analyzing the node connection strength in the dynamic topology interaction structure; Calculating a supervision model relevance coefficient by adopting a large model attention mechanism; generating hierarchical supervision weight distribution according to the supervision model association coefficient and the index weight initial value; And initializing supervision model parameters by applying the hierarchical supervision weight distribution, and outputting an initial hierarchical supervision model.
6. The large model based state-run assets intelligent supervision method according to claim 5, wherein the implementing a multi-source supervision data alignment process includes: Collecting state-run assets transaction flow data and environmental state data in real time, and extracting an aging characteristic vector in the state-run assets transaction flow data; Matching the deviation coefficient of the environmental state data and the supervision time axis, and calculating the characteristic alignment compensation quantity among the multi-source data; and correcting the supervision model parameters based on the characteristic alignment compensation quantity.
7. The large model-based state-run assets intelligent supervision method according to claim 6, wherein the performing constraint-driven iterative optimization calculations includes: Acquiring a current effective value set of the boundary constraint condition, and calculating the deviation gradient between the output of the supervision model and the objective function; And applying the business rule to limit and filter the invalid solution space, performing multiple parameter adjustment and constraint verification, and outputting state-run assets risk quantification assessment results when the double-layer model convergence condition is met.
8. The large model based state-run assets intelligent supervision method according to claim 7, wherein the generating a dynamic supervision policy vector includes: analyzing a risk grade mark in the state-run assets risk quantification assessment result; Based on the risk level indicia, associating the countermeasures in the history handling policy library to generate an executable sequence of supervisory operation instructions; and packaging the supervision operation instruction sequence into a dynamic supervision policy vector.

Description

State-run assets intelligent supervision method based on large model Technical Field The invention relates to the technical field of state-run assets intelligent supervision, in particular to a state-run assets intelligent supervision method based on a large model. Background The national assets are used as important material bases for the national economy development, and the supervision efficiency of the national assets is directly related to the economic operation quality and the social public benefit. Along with the continuous deepening of state-owned enterprise reforms and the continuous expansion of asset scale, the state asset forms are increasingly diversified, a plurality of categories such as fixed asset, financial asset, intangible asset and the like are covered, a plurality of links such as investment, operation, disposal and the like are involved in the circulation process, and the complexity of full life cycle management is obviously improved. The asset data are stored in information systems of different departments in a scattered mode, data islanding phenomenon exists, so that asset circulation tracks are difficult to track in the whole process, real-time sharing of part of key data such as asset states, rights and interests changes, value fluctuation and the like cannot be achieved, deep mining of historical supervision experience and dynamic adaptation of compliance standards are lacking, and when an asset operation environment changes or a novel business mode appears, potential risks cannot be identified in time by existing indexes. Most traditional supervision models are single-function models, each of the traditional supervision models operates independently, and an effective cooperative mechanism is lacked. For example, the risk assessment model, the compliance verification model and the like respectively process supervision tasks with different dimensionalities, information interaction among the models is insufficient, and correlation analysis is difficult to be performed by integrating multidimensional data, so that part of cross-domain and cross-link compound risks are difficult to be found in time. In addition, the risk assessment results are mainly described qualitatively and lack quantitative analysis support. Under the tide of digital transformation, state-run assets is increasingly in need of intelligent technology. Although some areas have been tried to introduce informatization means to improve the supervision efficiency, the existing system stays at the data recording and simple statistics level, the data processing capability and the deep learning advantage of a large model are not fully utilized, intelligent perception, dynamic evaluation and accurate regulation and control on the full life cycle of the asset cannot be realized, and the requirements of the new era state-run assets supervision on comprehensiveness, instantaneity and accuracy are difficult to meet. Disclosure of Invention The invention aims to provide a state-run assets intelligent supervision method based on a large model, so as to solve the problems in the background technology. In order to achieve the above object, the present invention provides a state-run assets intelligent supervision method based on a large model, the method comprising: Acquiring national asset full life cycle circulation data and real-time environment parameters, and constructing state-run assets supervision feature vectors containing space-time dimensions; based on the state-run assets supervision feature vector, a history supervision record and a compliance standard library are fused, and a multi-level state-run assets supervision index system is generated; designing a multi-supervision model collaboration framework by utilizing the multi-level state-run assets supervision index system, wherein the multi-supervision model collaboration framework comprises a dynamic topology interaction structure; outputting state-run assets a risk quantification evaluation result based on the dynamic topological interaction structure; generating a dynamic supervision policy vector according to the state-run assets risk quantification evaluation result; And triggering a supervision instruction execution engine based on the dynamic supervision policy vector. Preferably, the outputting state-run assets the risk quantification assessment result based on the dynamic topological interaction structure includes: According to the dynamic topology interaction structure, deploying a large model driven supervision weight distribution mechanism to generate an initial hierarchical supervision model; based on the initial hierarchical supervision model, combining a risk threshold rule base and a compliance constraint condition to construct a double-layer optimized supervision framework; implementing alignment processing of multi-source supervision data through the double-layer optimization supervision framework, and updating supervision model parameters; Based on t