CN-122019108-A - Industrial large model task cooperation method, equipment and storage medium

CN122019108ACN 122019108 ACN122019108 ACN 122019108ACN-122019108-A

Abstract

The embodiment of the invention provides a method, equipment and a storage medium for cooperating tasks of an industrial large model, which relate to the technical field of industrial intelligence, and the method comprises the steps of extracting multidimensional features of acquired abnormal events; the method comprises the steps of inputting multidimensional features into a pre-trained classification model for quantitative analysis to obtain event types and discrimination confidence of abnormal events, determining the event types of the abnormal events according to comparison results of the discrimination confidence and a preset confidence threshold, invoking a first system and/or a second system according to the event types of the abnormal events to process the abnormal events, wherein the first system is used for rapidly reasoning the events through an industrial large model, and the second system is used for deeply reasoning the events through the industrial large model. Thus, according to different event types, the independent execution or cooperative processing of the first system and the second system is selected, so that the efficiency and the precision balance of task mixing processing of the industrial large model of a single system architecture in an industrial scene are effectively solved when different abnormal events are faced.

Inventors

XU CHUNXIANG
ZHANG JUN
JIANG XIANFENG

Assignees

成都协鑫数智科技有限责任公司

Dates

Publication Date: 20260512
Application Date: 20260211

Claims (10)

1. An industrial large model task orchestration method, wherein the method comprises: extracting multidimensional features of the obtained abnormal event, wherein the multidimensional features comprise abnormal complexity, emergency degree, standardization degree and resource requirements; Inputting the multidimensional features into a pre-trained classification model for quantitative analysis to obtain the event type and the discrimination confidence of the abnormal event; Determining the event type of the abnormal event according to the comparison result of the discrimination confidence coefficient and the preset confidence coefficient threshold value; and calling a first system and/or a second system according to the event type of the abnormal event to process the abnormal event, wherein the first system is used for rapidly reasoning the event through the industrial large model, and the second system is used for deeply reasoning the event through the industrial large model.
2. The method of claim 1, wherein determining the event type of the abnormal event based on the comparison of the discrimination confidence and the preset confidence threshold comprises: If the comparison result is that the discrimination confidence is greater than or equal to the preset confidence threshold, determining that the event type of the abnormal event is a high-frequency standardized event if the abnormal event is single-parameter abnormal and is matched with the abnormal type in a preset high-frequency abnormal event rule base; And under the condition that the comparison result is that the discrimination confidence is smaller than the preset confidence threshold, determining the event type of the abnormal event as a fuzzy event.
3. The method according to claim 1, wherein invoking the first system and/or the second system to process the exception event according to the event type of the exception event comprises: Under the condition that the event type is a high-frequency standardized event, if the current load rate of the first system is smaller than a first preset load threshold value, the first system is called to process the abnormal event; if the current load rate of the first system is greater than or equal to a first preset load threshold, caching the abnormal event to a queue to be processed of the first system or calling the second system to process the abnormal event.
4. A method according to claim 3, characterized in that the method comprises: And in the process of processing the abnormal event by the first system, if the feature matching degree of the event feature corresponding to the abnormal event extracted by the first system and a preset high-frequency abnormal event rule base is smaller than a preset matching degree threshold value or the confidence degree of the reasoning result of the abnormal event by the first system is smaller than a first preset value, synchronizing the abnormal event and intermediate data obtained by the first system for processing the abnormal event to the second system, and processing the abnormal event by the second system.
5. The method according to claim 1, wherein invoking the first system and/or the second system to process the exception event according to the event type of the exception event comprises: And if the current load rate of the second system is greater than or equal to a second preset load threshold value, calling the first system to process non-core subtasks of the abnormal event, wherein the non-core subtasks are characterized by subtasks which do not need deep mechanism analysis or do not influence abnormal root cause judgment.
6. The method according to claim 5, characterized in that the method comprises: In the process of processing the abnormal event by the second system, if the current load rate of the second system is greater than or equal to a second preset load threshold value and a new low-frequency complex event exists, carrying out gradient division on the new low-frequency complex event to obtain a plurality of gradient complex events; Distributing low-gradient complex events in the plurality of gradient complex events to the first system, and synchronizing basic inference logic of the second system to the first system so that the first system processes the low-gradient complex events based on the basic inference logic of the second system, wherein the low-gradient complex events are abnormal events with the confidence degree of an inference result of the second system being larger than a second preset value.
7. The method of claim 5, wherein after the second system has processed the exception event, the method comprises: Acquiring a judgment rule of the abnormal event; Synchronizing the judging rule to a preset high-frequency abnormal event rule base; And when a new abnormal event is acquired and the judging rule of the new abnormal event can be found in the preset high-frequency abnormal event rule base, calling the first system to process the new abnormal event.
8. The method according to claim 1, wherein invoking the first system and/or the second system to process the exception event according to the event type of the exception event comprises: in the case that the event type is a fuzzy event, decomposing the abnormal event into a plurality of subtasks, wherein the plurality of subtasks comprise a standardized subtask and a complex subtask; invoking a first system to process the standardized subtasks; and calling a second system to process the complex subtasks.
9. An electronic device comprising a processor and a memory, the memory storing a computer program executable by the processor, the processor executable by the computer program to implement the industrial large model task orchestration method according to any one of claims 1-8.
10. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the industrial large model task orchestration method according to any one of claims 1-8.

Description

Industrial large model task cooperation method, equipment and storage medium Technical Field The invention relates to the technical field of industrial intelligence, in particular to an industrial large model task cooperation method, equipment and a storage medium. Background Along with the advanced intelligent upgrade of intelligent manufacturing, an industrial large model becomes a core technical support for promoting industry transformation, and is widely applied to key scenes such as industrial quality inspection, intelligent interaction with a tool body, PLC program generation and the like. The existing industrial large model generally adopts a single system architecture design, such as a full-parameter large model based on a Transformer, covers all task types with the same set of model capacity, adopts unified and isomorphic structural units and shared parameters/resources, and processes various tasks on the premise of not changing the underlying architecture by inputting identification, calling strategy or compiling optimization and the like. However, for high-frequency and standardized tasks, the industrial large model of a single system architecture lacks the processing capability of complex tasks, while for low-frequency and complex tasks, the processing precision can be guaranteed, but the problems of long processing time and high resource occupation exist, so that the efficiency and precision balance of task mixing processing in industrial scenes are difficult to solve when different tasks are faced. Disclosure of Invention Therefore, the present invention is directed to a task collaboration method for large industrial models, which selects a first system and/or a second system for processing according to different event types, so as to effectively solve the problem of balance of efficiency and precision of task mixing processing of large industrial models with a single system architecture in industrial scenes when different abnormal events are faced. In order to achieve the above purpose, according to a first aspect, an embodiment of the present invention provides an industrial large model task collaboration method, which includes extracting multidimensional features of an obtained abnormal event, where the multidimensional features include abnormal complexity, urgency, standardization and resource requirements, inputting the multidimensional features into a pre-trained classification model to perform quantitative analysis to obtain an event type and a discrimination confidence coefficient of the abnormal event, determining the event type of the abnormal event according to a comparison result of the discrimination confidence coefficient and the preset confidence coefficient threshold, invoking a first system and/or a second system according to the event type of the abnormal event to process the abnormal event, where the first system is used to rapidly infer the event through an industrial large model, and the second system is used to deeply infer the event through the industrial large model. In this embodiment, the multi-dimensional feature extraction of the abnormal event is used to facilitate the classification model to quantitatively analyze the event type and the discrimination confidence level of the abnormal event, so as to facilitate the determination of the event type of the abnormal event according to the comparison between the discrimination confidence level and the preset confidence level threshold, and then select the first system processing and/or the second system processing according to the type of the abnormal event. Thus, according to different event types, the independent execution or cooperative processing of the first system and the second system is selected, so that the efficiency and the precision balance of task mixing processing of the industrial large model of a single system architecture in an industrial scene are effectively solved when different abnormal events are faced. In some embodiments, determining the event type of the abnormal event according to the comparison result of the discrimination confidence coefficient and the preset confidence coefficient threshold value comprises determining that the event type of the abnormal event is a high-frequency standardized event if the discrimination confidence coefficient is greater than or equal to the preset confidence coefficient threshold value and is matched with the abnormal type in a preset high-frequency abnormal event rule base, determining that the event type of the abnormal event is a low-frequency complex event if the abnormal event is a multi-parameter abnormal event and has unknown data characteristics, and determining that the event type of the abnormal event is a fuzzy event if the comparison result is that the discrimination confidence coefficient is less than the preset confidence coefficient threshold value. The method is arranged so as to determine the event type of the abnormal event as a high-frequen