Search

CN-121995883-A - Big data driven intelligent information resource management platform

CN121995883ACN 121995883 ACN121995883 ACN 121995883ACN-121995883-A

Abstract

The invention relates to the technical field of industrial data scheduling and intelligent manufacturing, and provides a big data driven intelligent information resource management platform which adopts a closed loop architecture consisting of five modules of data perception and characterization, knowledge cognition and constraint modeling, on-line dynamic optimization decision, resource execution and coordination, meta cognition and evolution. The method comprises the steps of realizing real-time quantitative evaluation of data quality through a triplet quality vector, automatically converting text business rules into computable mathematical constraint and knowledge maps, adopting a knowledge enhancement reinforcement learning agent fused with a dual rewarding mechanism to carry out dynamic scheduling decision, realizing node-level real-time fine adjustment through an edge executor, and realizing closed-loop self-evolution of a scheduling model and a knowledge base through a meta-cognition mechanism. The invention realizes the cooperative promotion of the dispatching efficiency, the service compliance and the system self-adaptive capacity, and is suitable for the industrial quality inspection data dispatching scene with high real-time performance and high compliance requirement.

Inventors

  • SHANG JINGJING
  • WANG BO
  • WANG LIANGLIANG

Assignees

  • 陕西智星翊泽信息技术服务有限公司

Dates

Publication Date
20260508
Application Date
20260129

Claims (8)

  1. 1. A big data driven intelligent information resource management platform is characterized by comprising: The data perception and characterization module is used for accessing multi-source heterogeneous quality inspection data through a preset multi-protocol adapter, synchronizing the time stamp of the accessed data to a unified time server, attaching a space-time position label to each piece of data based on a preset resource catalog, and further calculating the triplet quality vector of each data stream in real time , For the sake of completeness of the results, For the sake of timeliness, the temperature of the product, Is confidence; The knowledge cognition and constraint modeling module is connected with the data perception and characterization module and is used for generating mathematical constraints which can be directly calculated by analyzing the quality inspection business rule text, constructing an industrial knowledge graph based on historical dispatching data, and adopting a hierarchical constraint propagation network to decompose the global resource constraints into dispatching subunits in a distributed manner; The on-line dynamic optimization decision module is connected with the data perception and characterization module and the knowledge cognition and constraint modeling module and is used for constructing a comprehensive state vector according to the quality vector, the mathematical constraint and the knowledge graph, and inputting the comprehensive state vector into a deep reinforcement learning network model enhanced by knowledge graph information so as to output a scheduling action; The resource execution and coordination module is connected with the online dynamic optimization decision module and is used for converting the scheduling action into a command which can be executed by the bottom layer and issuing the command, and deploying an edge executor at each computing node so as to execute a predefined micro action based on the local context information; and the meta cognition and evolution module is connected with the resource execution and coordination module, the online dynamic optimization decision module and the knowledge cognition and constraint modeling module and is used for collecting scheduling execution results, triggering fine tuning training of the deep reinforcement learning network model and dynamically optimizing the industrial knowledge graph.
  2. 2. The big data driven intelligent information resource management platform of claim 1, wherein the data perception and characterization module calculates timeliness in the triplet quality vector When the method is carried out, the following steps are carried out: Delay in acquiring data from generation to access to the module Preset timeliness qualification benchmark thresholds ; According to the formula Calculating a timeliness score, wherein For the preset attenuation coefficient, the damping coefficient is set, A time difference between the arrival of the data at the module is generated.
  3. 3. The big data driven intelligent information resource management platform of claim 1, wherein the knowledge cognition and constraint modeling module performs the following steps when generating mathematical constraints capable of being directly calculated by parsing business rule text: analyzing the rule text and identifying the key words through a deterministic finite automaton; Mapping rule semantics corresponding to the identified must-class keywords into adaptation types in the position hard constraint and the resource type hard constraint; and mapping rule semantics corresponding to the identified priority class keywords into task delay weight coefficients in the decision model objective function.
  4. 4. The big data driven intelligent information resource management platform of claim 1, wherein when the knowledge cognition and constraint modeling module decomposes global resource constraints by using a hierarchical constraint propagation network, the method comprises the following steps: Initializing Lagrangian multipliers for coordinating global constraints with local targets And broadcast to each scheduling subunit; Local cost function for each scheduling subunit Independent optimization is performed in which In order to be a local cost for the device, Is a local variable related to global constraint and is optimized Reporting; The central node reports according to each subunit Computing global variables And compared with a global constraint threshold, dynamically adjust Is a value of (2); Iteratively performing the above steps until The global constraint is satisfied.
  5. 5. The big data driven intelligent information resource management platform of claim 1, wherein the deep reinforcement learning network model in the online dynamic optimization decision module uses a dual rewarding function for training and deciding, the dual rewarding function is: , Wherein, the For the total prize to be awarded, Efficiency rewards calculated based on average task delay and total resource idle rate; For compliance rewards, the calculation formula is as follows: , Wherein the method comprises the steps of Is the first The penalty weight of the bar constraint, Is the first Degree of violation of the bar constraint, and when In the time-course of which the first and second contact surfaces, According to the formula Performing an adaptive augmentation, wherein For the preset amplification factor, To indicate the function when Time of day When (when) Time of day 。
  6. 6. The big data driven intelligent information resource management platform of claim 1, wherein the edge executor in the resource execution and coordination module periodically performs the following operations: Reading context information of a local computing node, wherein the context information comprises one or even more relevant information of CPU utilization rate, memory occupation and core temperature; Selecting micro-actions to be executed from a group of pre-defined micro-actions aiming at different node operation scenes based on a Topson sampling algorithm; After the selected micro-motion is executed, updating the Beta distribution parameters corresponding to the micro-motion according to the condition of obtaining positive gain feedback 。
  7. 7. The big data driven intelligent information resource management platform of claim 1, wherein the meta-cognition and evolution module triggers the deep reinforcement learning network model fine tuning training conditions comprise at least one of reaching a preset periodic triggering time, and monitoring that an error of a model predictive reward and an actual reward exceeds a preset threshold.
  8. 8. The big data driven intelligent information resource management platform of claim 7, wherein the meta-cognition and evolution module applies importance sampling technique to weight historical interaction data sampled from the experience playback buffer during model fine tuning training.

Description

Big data driven intelligent information resource management platform Technical Field The invention relates to the technical field of industrial data scheduling and intelligent manufacturing, in particular to a big data driven intelligent information resource management platform. Background In the current industrial manufacturing intelligent transformation process, efficient collaborative scheduling of massive multi-source heterogeneous data (e.g., real-time video, equipment status, production work orders) becomes a key challenge. The traditional scheduling system generally adopts a rule engine or a static optimization algorithm, and has the technical problems that firstly, a multi-source data format is different, quality is poor, a unified quantitative evaluation system is lacked, so that scheduling input is unreliable, secondly, business rules (such as safety compliance and energy efficiency constraint) exist in a text or configuration mode, and are difficult to directly integrate into a real-time decision model, so that scheduling efficiency and business compliance are difficult to consider, thirdly, the system is mainly designed in an open loop, dynamic optimization and knowledge update cannot be carried out according to execution feedback, adaptability is insufficient when facing complex disturbance, and long-term operation performance is easy to degrade. The intelligent information resource management platform driven by big data is provided, firstly, unified access of multi-source data is realized through multi-protocol adaptation and space-time tagging, and real-time quantification of data quality is creatively provided by a triplet quality vector, secondly, text rules are automatically converted into computable mathematical constraint and structured knowledge maps, and distributed decomposition of global constraint is realized through a hierarchical constraint propagation mechanism, furthermore, knowledge-enhanced dual rewarding reinforcement learning intelligent bodies are adopted, efficiency and compliance targets are synchronously optimized in scheduling decisions, meanwhile, node-level real-time fine adjustment is realized by depending on an edge executor, accurate execution of instructions is guaranteed, and finally, model self-optimization and knowledge base dynamic update are realized through a meta-cognition module, so that the platform has continuous evolution capability. Disclosure of Invention Aiming at the technical problems that the quality of multi-source data in the existing industrial scheduling system is uncontrollable, business rules are difficult to integrate into real-time decisions, and the system lacks self-adaptive optimization capability, the invention provides a big data driven intelligent information resource management platform. In order to achieve the above-mentioned purpose, the present invention provides a big data driven intelligent information resource management platform, comprising: The data perception and characterization module is used for accessing multi-source heterogeneous quality inspection data through a preset multi-protocol adapter, synchronizing the time stamp of the accessed data to a unified time server, attaching a space-time position label to each piece of data based on a preset resource catalog, and further calculating the triplet quality vector of each data stream in real time ,For the sake of completeness of the results,For the sake of timeliness, the temperature of the product,Is the confidence level. The knowledge cognition and constraint modeling module is connected with the data perception and characterization module and is used for generating mathematical constraints which can be directly calculated by analyzing quality inspection business rule texts, constructing an industrial knowledge graph based on historical dispatching data and adopting a hierarchical constraint propagation network to decompose the global resource constraints into dispatching subunits in a distributed mode. The on-line dynamic optimization decision module is connected with the data perception and characterization module and the knowledge cognition and constraint modeling module and is used for constructing a comprehensive state vector according to the quality vector, the mathematical constraint and the knowledge graph, and inputting the comprehensive state vector into a deep reinforcement learning network model enhanced by knowledge graph information so as to output a scheduling action; And the resource execution and coordination module is connected with the online dynamic optimization decision module and is used for converting the scheduling action into a command which can be executed by the bottom layer and issuing the command, and deploying an edge executor on each computing node so as to execute the predefined micro action based on the local context information. And the meta cognition and evolution module is connected with the resource execution and coordination module