CN-121997258-A - Industrial data fusion and interoperation processing method, device, equipment and storage medium

CN121997258ACN 121997258 ACN121997258 ACN 121997258ACN-121997258-A

Abstract

The invention discloses an industrial data fusion and interoperation processing method, device, equipment and storage medium. The method comprises the steps of obtaining multi-source industrial data from an industrial field environment, preprocessing the multi-source industrial data by combining a multi-scale time axis reconstruction mechanism to obtain a standardized input data set, carrying out feature extraction and feature mapping on the standardized input data set through a cross-modal coding framework to obtain a multi-modal sample data set, marking samples in the multi-modal sample data set based on a student model to obtain a multi-modal marking data set, constructing an industrial association graph by combining the multi-modal marking data set based on a graph neural network, processing the industrial association graph to obtain a fused unified state representation, carrying out semantic analysis on an industrial protocol by combining an semantic coding model and carrying out cross-protocol mapping on the industrial association graph to obtain an interoperation result, and providing the fused unified state representation and the interoperation result to an upper layer application through a micro-service framework. The method improves the data processing efficiency.

Inventors

GAO YUAN
SHEN BIN
DUAN SHIHUI
LU LIYING
NIU BIAO
YU SICONG
HUANG YING
TANG LIBO
GUO WENSHUANG
HU ZHONGHAO
FU TAO
CHANG HAOLUN
LI XIN

Assignees

中国信息通信研究院

Dates

Publication Date: 20260508
Application Date: 20260121
Priority Date: 20251225

Claims (10)

1. A method for industrial data fusion and interoperation processing, the method comprising: Acquiring multi-source industrial data from an industrial field environment; Preprocessing the multi-source industrial data by combining a multi-scale time axis reconstruction mechanism to obtain a standardized input data set; feature extraction is carried out on the standardized input data set through a cross-modal coding framework, and features of different modalities are mapped to a unified semantic space to obtain a uniformly-characterized multi-modal sample data set; Automatically labeling samples in the multi-modal sample data set based on a trained student model to obtain a multi-modal labeling data set; Constructing an industrial association graph based on a graph neural network and combining the multi-mode labeling data set, and processing the industrial association graph to obtain a unified state representation after fusion; Carrying out semantic analysis on the industrial protocol through a semantic coding model, and carrying out cross-protocol mapping by combining the industrial association graph to obtain an interoperation result; And providing the fused unified state representation and the interoperation result to an upper-layer application through a micro-service architecture.
2. The method of claim 1, wherein the multi-source industrial data comprises at least high frequency data, low frequency data, event class data, and image class data, and wherein the preprocessing of the multi-source industrial data in combination with the multi-scale timeline reconstruction mechanism to obtain the standardized input dataset comprises: Window aggregation and low-pass filtering are carried out on the high frequency data to obtain a multi-scale time slice; Restoring the missing segment in the low-frequency data through interpolation, migration interpolation and morphological smoothing processing to obtain restored low-frequency data; Time alignment is carried out on the low-frequency data and the multi-scale time slices, so that aligned low-frequency data are obtained; Performing time stamp normalization and context binding processing on the event data, and inserting the event data into the multi-scale time slice to obtain embedded event data; generating an associated index of the working condition and the variable state corresponding to the moment in the multi-scale time slice for the image class data to obtain the image class data with the associated index; And taking the multi-scale time slices, the aligned low-frequency data, the embedded event class data and the image class data with the associated index as data in a standardized input data set.
3. The method of claim 1, wherein the cross-modal encoding framework comprises a text encoding sub-model, an image encoding sub-model, a time sequence encoding sub-model and a cross-modal alignment network, the data types in the standardized input data set comprise a text type, an image type and a time sequence type, and the method correspondingly comprises the steps of performing feature extraction on the standardized input data set through the cross-modal encoding framework, mapping features of different modalities to a unified semantic space, and obtaining a uniformly-characterized multi-modal sample data set, and comprises: semantic extraction is carried out on the text type data through the text coding sub-model, so that text characteristics are obtained; extracting spatial features of the image type data through the image coding sub-model; acquiring time sequence characteristics of the time sequence type data through the time sequence coding sub-model; Mapping the text features, the space features and the time sequence features to a unified semantic space through the cross-modal alignment network to obtain a uniformly-characterized multi-modal sample data set; The text coding sub-model is a large language model for secondary training on industrial corpus, the image coding sub-model adopts a depth convolution network and visual transform fusion structure, and the time sequence coding sub-model adopts a two-way long-short-term memory network, time sequence transform and frequency domain convolution combined structure.
4. The method of claim 1, wherein automatically labeling the samples in the multi-modal sample dataset based on the trained student model to obtain a multi-modal labeling dataset, comprising: Determining a candidate tag set, tag confidence distribution, multi-modal consistency scores and semantic similarity vectors of each sample in the multi-modal sample data set based on the trained student model; based on a candidate label set, label confidence distribution, multi-modal consistency scores and semantic similarity vectors of all samples, taking samples meeting a first preset condition in the multi-modal sample data set as data in the multi-modal labeling data set; The first preset condition is that the confidence coefficient of the sample is in a first confidence coefficient interval and passes cross-mode consistency check, or the confidence coefficient of the sample is in a second confidence coefficient interval and passes manual check, and the confidence coefficient in the first confidence coefficient interval is higher than the confidence coefficient in the second confidence coefficient interval.
5. The method of claim 4, wherein the cross-modality consistency check includes one or more of checking whether an image defect corresponds to a time series vibration anomaly, whether an alarm text matches an event in an equipment log, whether a protocol field trend is consistent with a process state, and whether a process stage label corresponds to a production lot progress.
6. The method of claim 1, wherein the constructing an industrial association graph based on the graph neural network in combination with the multi-modal labeling dataset, and processing the industrial association graph to obtain a fused unified state representation, comprises: constructing an industrial association graph based on the multi-mode labeling data set, wherein graph nodes of the industrial association graph are field devices, sensor nodes, station procedures, process parameters, quality detection results and log events, and graph edges are physical connection relations, control logic dependencies, time sequence relations, semantic dependencies and working condition dynamic dependencies; The industrial association graph is subjected to cross-modal feature aggregation through a multi-layer graph attention network to obtain aggregated features; and optimizing the aggregated features through a joint loss function to obtain a unified state representation after fusion.
7. The method of claim 1, wherein the performing semantic parsing of the industrial protocol by the semantic coding model and performing cross-protocol mapping in combination with the industrial association graph to obtain the interoperation result comprises: Carrying out semantic analysis on a data model description of an industrial protocol through a semantic coding model to generate a protocol semantic embedded representation, wherein the protocol semantic embedded representation comprises semantic information of a new protocol variable; combining entity nodes and a relation structure in the industrial association graph, and carrying out entity alignment and relation linking on the new protocol variable to form a protocol view under a unified semantic space; calculating the similarity and semantic distance of the embedded vectors of the protocol semantic embedded representation, and the process constraint matching degree and the historical mapping credibility of the new protocol variable; Generating mapping relations among different protocol variables based on the similarity of the embedded vectors, the semantic distance, the process constraint matching degree and the historical mapping credibility to obtain an executable mapping table, a conversion rule set and an automatically generated conversion function, wherein the mapping relations comprise equivalent mapping, similar mapping, combined mapping and rule mapping; and taking the executable mapping table, the conversion rule set and the automatically generated conversion function as an interoperation result.
8. An industrial data fusion and interoperation processing device, the device comprising: The acquisition module is used for acquiring multi-source industrial data from an industrial field environment; The processing module is used for preprocessing the multi-source industrial data by combining a multi-scale time axis reconstruction mechanism to obtain a standardized input data set; The extraction module is used for extracting the characteristics of the standardized input data set through a cross-modal coding framework, mapping the characteristics of different modalities to a unified semantic space and obtaining a uniformly-characterized multi-modal sample data set; the labeling module is used for automatically labeling the samples in the multi-mode sample data set based on a trained student model to obtain the multi-mode labeling data set; The construction module is used for constructing an industrial association graph based on the graph neural network and combining the multi-mode labeling data set, and processing the industrial association graph to obtain a unified state representation after fusion; the analysis module is used for carrying out semantic analysis on the industrial protocol through the semantic coding model and carrying out cross-protocol mapping by combining the industrial association graph to obtain an interoperation result; and the providing module is used for providing the fused unified state representation and the interoperation result to an upper layer application through a micro-service architecture.
9. An electronic device, the device comprising: At least one processor, and A memory communicatively coupled to the at least one processor, wherein, The memory stores a computer program executable by the at least one processor to enable the at least one processor to perform the industrial data fusion and interoperation process method of any of claims 1-7.
10. A computer readable storage medium storing computer instructions for causing a processor to perform the industrial data fusion and interoperation process method of any of claims 1-7.

Description

Industrial data fusion and interoperation processing method, device, equipment and storage medium Technical Field The embodiment of the invention relates to the technical field of industrial Internet, in particular to an industrial data fusion and interoperation processing method, device and equipment and a storage medium. Background Along with the continuous improvement of the digitization and intellectualization level of the manufacturing industry, the scale of data generated by industrial enterprises in various business links such as production operation, process control, quality management, equipment maintenance and supply chain coordination is continuously increased, and the data types are also expanded from the traditional structured field measurement data to multi-mode forms such as images, videos, time sequence signals, alarm logs, text records, protocol messages and the like. Industrial data exhibit typical characteristics in terms of quantity, speed, diversity, and value density, which increasingly play a role in process optimization, condition diagnostics, predictive maintenance, quality control, and even production decisions. However, unlike the relatively standardized, format-converged data in the internet domain, the industrial data is deeply coupled to the equipment, process, plant, scene and enterprise's own logic, which natural heterogeneity and high specificity present great challenges for fusion processing and interoperability. In practical application scenes, industrial data are often in various quantities and various types, so that a great deal of manpower is often required to rearrange and explain the industrial data when the industrial data are processed in the prior art, and the industrial data are low in processing efficiency and long in period. Disclosure of Invention The invention provides an industrial data fusion and interoperation processing method, device, equipment and storage medium, which are used for solving the problems of low efficiency and long period when industrial data are processed in the prior art. According to an aspect of the present invention, there is provided an industrial data fusion and interoperation processing method, the method comprising: Acquiring multi-source industrial data from an industrial field environment; Preprocessing the multi-source industrial data by combining a multi-scale time axis reconstruction mechanism to obtain a standardized input data set; feature extraction is carried out on the standardized input data set through a cross-modal coding framework, and features of different modalities are mapped to a unified semantic space to obtain a uniformly-characterized multi-modal sample data set; Automatically labeling samples in the multi-modal sample data set based on a trained student model to obtain a multi-modal labeling data set; Constructing an industrial association graph based on a graph neural network and combining the multi-mode labeling data set, and processing the industrial association graph to obtain a unified state representation after fusion; Carrying out semantic analysis on the industrial protocol through a semantic coding model, and carrying out cross-protocol mapping by combining the industrial association graph to obtain an interoperation result; And providing the fused unified state representation and the interoperation result to an upper-layer application through a micro-service architecture. According to another aspect of the present invention, there is provided an industrial data fusion and interoperation processing device, the device comprising: The acquisition module is used for acquiring multi-source industrial data from an industrial field environment; The processing module is used for preprocessing the multi-source industrial data by combining a multi-scale time axis reconstruction mechanism to obtain a standardized input data set; The extraction module is used for extracting the characteristics of the standardized input data set through a cross-modal coding framework, mapping the characteristics of different modalities to a unified semantic space and obtaining a uniformly-characterized multi-modal sample data set; the labeling module is used for automatically labeling the samples in the multi-mode sample data set based on a trained student model to obtain the multi-mode labeling data set; The construction module is used for constructing an industrial association graph based on the graph neural network and combining the multi-mode labeling data set, and processing the industrial association graph to obtain a unified state representation after fusion; the analysis module is used for carrying out semantic analysis on the industrial protocol through the semantic coding model and carrying out cross-protocol mapping by combining the industrial association graph to obtain an interoperation result; and the providing module is used for providing the fused unified state representation and the interoperation result to an up