CN-122019529-A - Model matrix-based multi-source heterogeneous industrial data decision method and system

CN122019529ACN 122019529 ACN122019529 ACN 122019529ACN-122019529-A

Abstract

The invention provides a multi-source heterogeneous industrial data decision method and system based on a model matrix in the technical field of industrial data processing and intelligent decision, the method comprises the steps of S1, collecting multi-source industrial data, S2, preprocessing each multi-source industrial data to construct a multi-mode data matrix, S3, constructing the model matrix based on a model layer and an agent layer, setting an API (application program interface) for calling the model layer, S4, acquiring input business requirements, inputting the business requirements to the model matrix, matching the model matrix with corresponding agents from the agent layer based on the type of the requirements of the business requirements, S5, matching the corresponding industrial data from the multi-mode data matrix based on the business requirements, and scheduling collaborative reasoning on the industrial data based on the AI model adapted in the API interface to obtain decision suggestions corresponding to the business requirements. The method has the advantages that the accuracy of industrial data decision making and the intelligent level of the system are greatly improved.

Inventors

LIN XING
ZHUANG HAIJUN

Assignees

福州市数字产业互联科技有限责任公司

Dates

Publication Date: 20260512
Application Date: 20251218

Claims (10)

1. A multi-source heterogeneous industrial data decision method based on a model matrix is characterized by comprising the following steps: s1, collecting multi-source industrial data including industrial structured data, document unstructured data and picture unstructured data; s2, preprocessing each multi-source industrial data to construct a multi-mode data matrix comprising a structured data matrix, a knowledge graph data matrix and a picture data matrix; S3, constructing a model matrix based on a model layer and an agent layer, and setting an API interface for calling the model layer; S4, acquiring an input service demand, and inputting the service demand into a model matrix, wherein the model matrix is used for matching corresponding intelligent agents from the intelligent agent layer based on the demand type of the service demand; And S5, matching corresponding industrial data from a multi-mode data matrix based on service requirements by the agent, carrying out collaborative reasoning on the industrial data based on an AI model adapted in the API interface scheduling model layer to obtain decision suggestions corresponding to the service requirements, and outputting the decision suggestions.
2. The method for making a model matrix-based multi-source heterogeneous industrial data decision as set forth in claim 1, wherein said step S1 is specifically: the server collects multi-source industrial data comprising industrial structured data, document unstructured data and picture unstructured data through a sensor group and a database; The industrial structured data at least comprises industrial equipment operation parameters, equipment operation logs and production orders, wherein the industrial equipment operation parameters at least comprise temperature, pressure and rotating speed; the document unstructured data at least comprises a process manufacturing manual, a technical document and a product design drawing; the picture unstructured data at least comprises a design drawing, a product drawing and a fault scene drawing.
3. The method for making a model matrix-based multi-source heterogeneous industrial data decision as set forth in claim 1, wherein said step S2 is specifically: The server performs preprocessing at least comprising outlier removal and missing value interpolation on each piece of industrial structured data, and stores each piece of preprocessed industrial structured data in a time sequence database by taking a timestamp as an index and taking equipment numbers, data types and specific numerical values as fields so as to construct a structured data matrix; The method comprises the steps that a server performs preprocessing at least comprising word segmentation, part-of-speech tagging and named entity recognition on each piece of document unstructured data, knowledge elements comprising entities, attributes and relations are recognized from each piece of document unstructured data after preprocessing, a knowledge map is constructed after semantic understanding of each knowledge element is enhanced through a pre-trained BERT model, and each piece of document unstructured data and the knowledge map are mapped in a correlated mode to construct a knowledge map data matrix; the method comprises the steps that a server performs at least partial feature extraction, downsampling and feature integration on each piece of picture unstructured data to obtain picture feature vectors with fixed lengths, marks at least comprising category, key design elements and product information on each piece of picture unstructured data to complete preprocessing of each piece of picture unstructured data, and a picture data matrix is constructed based on each piece of picture unstructured data, each picture feature vector and marked labels; And obtaining a multi-mode data matrix based on the constructed structured data matrix, the knowledge graph data matrix and the picture data matrix.
4. The method for making a decision on multi-source heterogeneous industrial data based on model matrix according to claim 1, wherein said step S3 is specifically: the server builds a model matrix based on the model layer and the intelligent agent layer, and sets an API interface for calling the model layer; The model layer at least comprises an AI model with the types of an OCR model, a language model, an embedded model, a CNN model and a multi-mode model, and has the functions of model scheduling, resource management and reasoning, and the intelligent agent layer at least comprises intelligent agents with the types of intelligent customer service, auxiliary design, policy matching and picture marking, and is used as a butt joint bridge of industrial data, business requirements and the AI model.
5. The method for making a model matrix-based multi-source heterogeneous industrial data decision as set forth in claim 1, wherein said step S5 is specifically: The intelligent agent matches corresponding industrial data from a multi-mode data matrix based on service requirements, wherein the industrial data is industrial structured data, document unstructured data or picture unstructured data; The intelligent agent dynamically allocates computing resources for each AI model based on the AI model adapted in the API interface scheduling model layer, performs collaborative reasoning on industrial data through each AI model, performs data transmission through an API interface in the collaborative reasoning process of each AI model, and finally performs reasoning to obtain decision suggestions corresponding to the service demands; And the agent carries out structural output on the decision advice in a natural language form.
6. A multi-source heterogeneous industrial data decision system based on a model matrix is characterized by comprising the following modules: The multi-source industrial data acquisition module is used for acquiring multi-source industrial data comprising industrial structured data, document unstructured data and picture unstructured data; the multi-modal data matrix construction module is used for preprocessing each multi-source industrial data to construct a multi-modal data matrix comprising a structured data matrix, a knowledge graph data matrix and a picture data matrix; the model matrix construction module is used for constructing a model matrix based on a model layer and an agent layer, and setting an API interface for calling the model layer; The business demand input module is used for acquiring input business demands, inputting the business demands into the model matrix, and matching corresponding intelligent agents from the intelligent agent layer by the model matrix based on the demand types of the business demands; the decision suggestion output module is used for matching corresponding industrial data from the multi-mode data matrix based on the service requirement by the agent, carrying out collaborative reasoning on the industrial data based on the AI model adapted in the API interface scheduling model layer to obtain a decision suggestion corresponding to the service requirement, and outputting the decision suggestion.
7. The multi-source heterogeneous industrial data decision system based on the model matrix of claim 6, wherein the multi-source industrial data acquisition module is specifically configured to: the server collects multi-source industrial data comprising industrial structured data, document unstructured data and picture unstructured data through a sensor group and a database; The industrial structured data at least comprises industrial equipment operation parameters, equipment operation logs and production orders, wherein the industrial equipment operation parameters at least comprise temperature, pressure and rotating speed; the document unstructured data at least comprises a process manufacturing manual, a technical document and a product design drawing; the picture unstructured data at least comprises a design drawing, a product drawing and a fault scene drawing.
8. The multi-source heterogeneous industrial data decision system based on model matrix of claim 6, wherein the multi-modal data matrix construction module is specifically configured to: The server performs preprocessing at least comprising outlier removal and missing value interpolation on each piece of industrial structured data, and stores each piece of preprocessed industrial structured data in a time sequence database by taking a timestamp as an index and taking equipment numbers, data types and specific numerical values as fields so as to construct a structured data matrix; The method comprises the steps that a server performs preprocessing at least comprising word segmentation, part-of-speech tagging and named entity recognition on each piece of document unstructured data, knowledge elements comprising entities, attributes and relations are recognized from each piece of document unstructured data after preprocessing, a knowledge map is constructed after semantic understanding of each knowledge element is enhanced through a pre-trained BERT model, and each piece of document unstructured data and the knowledge map are mapped in a correlated mode to construct a knowledge map data matrix; the method comprises the steps that a server performs at least partial feature extraction, downsampling and feature integration on each piece of picture unstructured data to obtain picture feature vectors with fixed lengths, marks at least comprising category, key design elements and product information on each piece of picture unstructured data to complete preprocessing of each piece of picture unstructured data, and a picture data matrix is constructed based on each piece of picture unstructured data, each picture feature vector and marked labels; And obtaining a multi-mode data matrix based on the constructed structured data matrix, the knowledge graph data matrix and the picture data matrix.
9. The multi-source heterogeneous industrial data decision system based on model matrix according to claim 6, wherein the model matrix construction module is specifically configured to: the server builds a model matrix based on the model layer and the intelligent agent layer, and sets an API interface for calling the model layer; The model layer at least comprises an AI model with the types of an OCR model, a language model, an embedded model, a CNN model and a multi-mode model, and has the functions of model scheduling, resource management and reasoning, and the intelligent agent layer at least comprises intelligent agents with the types of intelligent customer service, auxiliary design, policy matching and picture marking, and is used as a butt joint bridge of industrial data, business requirements and the AI model.
10. The multi-source heterogeneous industrial data decision making system based on model matrix as set forth in claim 6, wherein said decision suggestion output module is specifically configured to: The intelligent agent matches corresponding industrial data from a multi-mode data matrix based on service requirements, wherein the industrial data is industrial structured data, document unstructured data or picture unstructured data; The intelligent agent dynamically allocates computing resources for each AI model based on the AI model adapted in the API interface scheduling model layer, performs collaborative reasoning on industrial data through each AI model, performs data transmission through an API interface in the collaborative reasoning process of each AI model, and finally performs reasoning to obtain decision suggestions corresponding to the service demands; And the agent carries out structural output on the decision advice in a natural language form.

Description

Model matrix-based multi-source heterogeneous industrial data decision method and system Technical Field The invention relates to the technical field of industrial data processing and intelligent decision making, in particular to a multi-source heterogeneous industrial data decision making method and system based on a model matrix. Background In the background of the deep advancement of industry 4.0, industrial data presents explosive growth, and the type of the industrial data not only comprises structured sensor data such as temperature, pressure, rotating speed and the like, but also comprises unstructured information such as a large number of process manufacturing manuals, technical documents, design drawings, production site pictures and the like. The realization of efficient integration and intelligent analysis of the multi-source heterogeneous data has become a key link for promoting the intelligent development of industry, and is directly related to the accuracy of industrial decision and the intelligent level of a production system. At present, the industrial field mainly has the following practical mode in the aspects of data processing and model construction that the traditional structured database and data warehouse technology are depended on in the data integration layer, the unstructured data are usually stored independently or simply indexed, and a unified semantic association and depth fusion mechanism is lacked, so that the information island phenomenon is obvious. At the model construction level, a special model oriented to a single data type is usually adopted, or a plurality of models are simply connected in series at the task level, so that the cross-modal semantic alignment and joint reasoning capability are difficult to realize. Taking equipment fault diagnosis as an example, the existing method is mainly used for carrying out threshold judgment or simple time sequence analysis only according to the sensor value, and unstructured information such as equipment history records, maintenance reports, field pictures and the like cannot be effectively fused, so that the diagnosis process is not comprehensive enough and the traceability is limited. In a product design link, design drawings and technical documents are often stored respectively, visual information and text knowledge are mutually split, and the design optimization integrated process and efficiency improvement are restricted. In a combined view, the prior art mainly has the following problems: 1. The data integration is difficult, the semantic level fusion of the multi-source heterogeneous data is difficult to realize by the traditional data processing method, the structured data and the unstructured data are mutually independent, the information correlation is weak, a unified data view cannot be constructed, and the exertion of the comprehensive value of the data is limited. 2. The model has poor adaptability, and is faced with complex and changeable industrial scenes, most of the existing models are in a single mode or simple combination form, and the lack of joint semantic modeling capability on cross-mode data such as texts, images and the like leads to large semantic understanding deviation and limited prediction precision, so that the intelligent analysis requirement of complex industrial tasks is difficult to support. 3. The scene application is limited in that the existing method can not realize the effective coordination of the multi-source information in specific scenes such as fault diagnosis, product design and the like. For example, unstructured information is difficult to integrate into an analysis flow in the fault diagnosis process, so that the diagnosis efficiency is low and the misjudgment rate is high, and graphic information is split in the product design, so that the comprehensiveness of design optimization is affected, the research and development period is prolonged, and the development cost is increased. Therefore, how to provide a multi-source heterogeneous industrial data decision method and system based on a model matrix, so as to improve the accuracy of industrial data decision and the intelligent level of the system, is a technical problem to be solved. Disclosure of Invention The invention aims to solve the technical problem of providing a multi-source heterogeneous industrial data decision method and a multi-source heterogeneous industrial data decision system based on a model matrix, so that the accuracy of industrial data decision and the intelligent level of the system are improved. In a first aspect, the present invention provides a model matrix-based multi-source heterogeneous industrial data decision method, comprising the steps of: s1, collecting multi-source industrial data including industrial structured data, document unstructured data and picture unstructured data; s2, preprocessing each multi-source industrial data to construct a multi-mode data matrix comprising a structured data matri