CN-122021726-A - Physical system intelligent decision method and device based on large language model

CN122021726ACN 122021726 ACN122021726 ACN 122021726ACN-122021726-A

Abstract

The application provides a physical system intelligent decision method and device based on a large language model, wherein the method comprises the steps of obtaining multi-source heterogeneous information from a physical system, carrying out data preprocessing on the multi-source heterogeneous information to obtain an information set of the physical system, judging a decision scene type of user input operation intention through the large language model, calling a corresponding intelligent decision process in a collaborative algorithm layer according to the decision scene type by using the large language model through an intention perception network based on the information set to obtain an intelligent decision result, and verifying the intelligent decision result in a digital twin simulation model. The application judges the type of the decision scene of the user input operation intention through the big language model, and further calls the corresponding sub-big language model according to the decision scene, designs a cascading architecture for solving the intelligent decision of the physical system, limits the capacity of each sub-big language model in the good field, and improves the overall risk controllability and the decision effect.

Inventors

XIAO XIN
FU MINGXING
HUANG JIANDONG

Assignees

北京安晨信息技术有限公司

Dates

Publication Date: 20260512
Application Date: 20251218

Claims (10)

1. A physical system intelligent decision method based on a large language model is characterized by comprising the following steps: The method comprises the steps of obtaining multi-source heterogeneous information from a physical system, and carrying out data preprocessing on the multi-source heterogeneous information to obtain an information set of the physical system; Judging a decision scene type of user input operation intention through a large language model, wherein the large language model uses an intention perception network, and calling a corresponding intelligent decision process in a collaborative algorithm layer according to the decision scene type based on the information set to obtain an intelligent decision result; and verifying the intelligent decision result in a digital twin simulation model.
2. The method of claim 1, wherein the obtaining the multi-source heterogeneous information from the physical system, and the performing data preprocessing on the multi-source heterogeneous information specifically comprises: The method comprises the steps of obtaining multi-source heterogeneous information comprising sensor time sequence data, images, videos, sounds, knowledge libraries, knowledge maps and system preset operation rules, and sequentially performing data preprocessing operations comprising cleaning and format conversion on the multi-source heterogeneous information to obtain the information set.
3. The method according to claim 1, wherein the determining, by the large language model, the type of the decision scene of the user input operation intention specifically includes: receiving text content input by a user, embedding the text content into a prompt word template of a large language model to obtain an integrated text, and inputting the integrated text into a dual-channel intention perception network; Analyzing the lexical structure of the integrated text in a first channel to extract words existing in a first preset word list, wherein each word expressing change meanings and corresponding weight are described in the first preset word list, and calculating a logic change index LCI; Analyzing the lexical structure of the integrated text in a second channel to extract words existing in a second preset word list, wherein each word expressing fault meaning and corresponding weight are described in the second preset word list, and calculating a preliminary abnormality index; if the abnormal state index SSI is larger than a first threshold value, judging that the abnormal state index SSI is a fault correction decision scene, otherwise, judging that the abnormal state index SSI is a dynamic adjustment decision scene if the logic change index LCI is larger than a second threshold value, otherwise, judging that the abnormal state index SSI is a task scheduling decision scene.
4. The method of claim 3, wherein the calling, based on the information set, a corresponding intelligent decision flow in a collaborative algorithm layer according to the decision scene type to obtain an intelligent decision result specifically comprises: Invoking the collaborative algorithm layer, wherein the collaborative algorithm layer comprises a fault correction decision scene, a dynamic adjustment decision scene and an intelligent decision flow for arranging a task decision scene; if the user intends to process the fault correction decision scene, a first decision flow is called, and if the user intends to process the dynamic adjustment decision scene, a second decision flow is called; Wherein, each decision process is internally provided with a corresponding sub-large language model.
5. The method of claim 4, wherein invoking the first decision flow if the user intends to handle the fault correction decision scenario specifically comprises: acquiring a pre-established knowledge graph, wherein the knowledge graph comprises a fault name, a fault phenomenon and a cause of the fault; Positioning a corresponding fault name according to a fault phenomenon described in a user intention, or directly extracting the fault name from the user intention, acquiring a corresponding cause of a fault according to the fault name, and generating a diagnosis flow tree; And sequentially calling corresponding data interface verification according to the diagnosis flow tree, and determining the fault reason.
6. The method of claim 4, wherein invoking the second decision flow if the user intends to handle the dynamic adjustment decision scenario specifically comprises: If only one of the targets or the constraint conditions is described in the user intention, judging whether the existing mathematical form model exists in the preset operation rules of the system, and replacing corresponding parameters to update the mathematical form model; and transmitting the mathematical formal model to an optimizer for solving.
7. The method according to claim 4, wherein invoking a third decision flow if the user intends to process the orchestration task decision scene specifically comprises: Acquiring a correlation dependency pattern of a current task from an existing knowledge library, carrying out reasoning decomposition according to the correlation dependency pattern to obtain a workflow composed of subtasks, screening all callable function tools according to the field to which the current task belongs, and distributing corresponding function tools for each subtask; And executing the workflow after the function tool is distributed.
8. A physical system intelligent decision-making device based on a large language model, comprising: The information processing module is used for acquiring multi-source heterogeneous information from a physical system, and carrying out data preprocessing on the multi-source heterogeneous information to obtain an information set of the physical system; The intelligent decision module is used for judging the decision scene type of the user input operation intention through a large language model, wherein the large language model uses an intention perception network, and based on the information set, a corresponding intelligent decision flow in a collaborative algorithm layer is called according to the decision scene type to obtain an intelligent decision result; and the simulation verification module is used for verifying the intelligent decision result in the digital twin simulation model.
9. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the computer program implementing the steps of the large language model based physical system intelligent decision method as claimed in any one of claims 1 to 7 when executed by the processor.
10. A computer-readable storage medium, wherein a program for implementing information transfer is stored on the computer-readable storage medium, and when the program is executed by a processor, the steps of the large language model-based physical system intelligent decision method according to any one of claims 1 to 7 are implemented.

Description

Physical system intelligent decision method and device based on large language model Technical Field The invention relates to the technical field of intelligent decision making, in particular to a physical system intelligent decision making method and device based on a large language model. Background The physical system generally refers to an integral body formed by interaction of physical components such as a sensor, an actuator, a mechanical structure, a circuit and the like through physical laws such as mechanics, electromagnetism, thermodynamics and the like, and is combined with practical applications such as a wind driven generator, a room temperature regulating system, an automatic driving automobile and the like, and the decision refers to a process of executing an operation in order to achieve a specific target. In the related technology, the traditional method is seriously dependent on an accurate mathematical model, when the mathematical model is different from the physical system, such as equipment aging, certain dynamic characteristics which are not modeled and the like, the control decision performance can be drastically reduced, the large language model LLM can make up for the defect of model accuracy by utilizing knowledge and reasoning, has strong generalization capability and analogy reasoning capability, but in the existing scheme for realizing auxiliary decision by using the large language model, a single model is often used for simultaneously mastering a plurality of thinking modes with huge differences, such as planning, fault diagnosis, flow arrangement and the like, knowledge interference is extremely easy to be caused, and once an output result is poor, a link with a specific problem is difficult to be positioned, and a prompting word of the large language model can become huge and complex. By integrating the analysis of the development condition in the technical field, the prior art lacks a scheme for judging the type of the decision scene of the user input operation intention based on the large language model, and further calls the corresponding sub-large language model according to the decision scene so as to realize intelligent decision of the physical system. Disclosure of Invention The invention aims to provide a physical system intelligent decision method and device based on a large language model, and aims to solve the problems in the prior art. According to a first aspect of an embodiment of the present invention, there is provided a physical system intelligent decision method based on a large language model, including: the method comprises the steps of obtaining multi-source heterogeneous information from a physical system, and carrying out data preprocessing on the multi-source heterogeneous information to obtain an information set of the physical system; judging the decision scene type of the user input operation intention through a large language model, wherein the large language model uses an intention perception network, and calling a corresponding intelligent decision process in a collaborative algorithm layer according to the decision scene type based on an information set to obtain an intelligent decision result; And verifying the intelligent decision result in the digital twin simulation model. According to a second aspect of the embodiment of the present invention, there is provided a physical system intelligent decision device based on a large language model, including: the information processing module is used for acquiring multi-source heterogeneous information from the physical system, and carrying out data preprocessing on the multi-source heterogeneous information to obtain an information set of the physical system; The intelligent decision module is used for judging the decision scene type of the user input operation intention through a large language model, wherein the large language model uses an intention perception network, and based on an information set, a corresponding intelligent decision flow in a collaborative algorithm layer is called according to the decision scene type to obtain an intelligent decision result; and the simulation verification module is used for verifying the intelligent decision result in the digital twin simulation model. According to a third aspect of embodiments of the present invention there is provided an electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, which when executed by the processor implements the steps of the large language model based physical system intelligent decision method as provided in the first aspect of the present disclosure. According to a fourth aspect of the embodiments of the present invention, there is provided a computer readable storage medium having stored thereon a program for implementing information transfer, which when executed by a processor, implements the steps of the large language model based physical system