CN-121981495-A - Production process parameter optimization method, equipment, medium and product based on industrial Internet

CN121981495ACN 121981495 ACN121981495 ACN 121981495ACN-121981495-A

Abstract

The application provides a production process parameter optimization method, equipment, medium and product based on industrial Internet, and relates to the technical field of process optimization. The method comprises the steps of determining target semantic vectors and target values of various physical performance indexes according to task demand information of a target production task, determining current technological parameters corresponding to the target semantic vectors and predicted values of various physical performance indexes corresponding to the current technological parameters, determining mixed rewards based on the semantic vectors corresponding to the current technological parameters, the predicted values of various physical performance indexes, the target semantic vectors and the target values of various physical performance indexes and a plurality of physical constraint functions, determining optimal technological parameters of the target production task by taking the maximized mixed rewards as an optimization target and adopting an optimization algorithm, wherein the method fuses the physical-semantic dual targets to obtain optimal technological parameters with semantic compliance and physical feasibility.

Inventors

LIU DEYAN
CHEN LUCHENG
YANG JIAN
WANG CHAO
LI XIYUAN
WANG LIN
CHEN HUI
LI SHUAI
CHENG JING

Assignees

卡奥斯工业智能研究院(青岛)有限公司

Dates

Publication Date: 20260505
Application Date: 20260401

Claims (10)

1. A method for optimizing production process parameters based on industrial internet, the method comprising: Determining target semantic vectors and target values of various physical performance indexes according to task demand information of a target production task; determining the current technological parameters corresponding to the target semantic vector and the predicted values of various physical performance indexes corresponding to the current technological parameters; Determining a mixed rewarding value based on a semantic vector corresponding to the current process parameter, predicted values of various physical performance indexes, target semantic vectors, target values of various physical performance indexes and a plurality of physical constraint functions, wherein the physical constraint functions are used for limiting the value of the current process parameter; and taking the maximum mixed rewarding value as an optimization target, and adopting an optimization algorithm to determine the optimal technological parameters of the target production task.
2. The method of claim 1, wherein determining the current process parameter corresponding to the target semantic vector and the predicted value of each physical performance indicator corresponding to the current process parameter comprises: Inputting the target semantic vector into a pre-trained industrial large model to obtain a current technological parameter output by the industrial large model based on a current meta-parameter, wherein the current meta-parameter is obtained by training a trial production technological parameter of the target production task, a semantic vector corresponding to the trial production technological parameter and actual values of various physical performance indexes; And inputting the current process parameters into a pre-trained prediction model to obtain predicted values of the physical performance indexes output by the prediction model, wherein the prediction model is obtained by training historical process parameters of a plurality of historical production tasks and actual values of the physical performance indexes corresponding to the historical process parameters.
3. The method of claim 2, wherein said determining optimal process parameters for said target production task using an optimization algorithm with the objective of maximizing said hybrid prize value comprises: After each iteration of the optimization algorithm, updating the meta-parameters and the target semantic vector based on the mixed rewards value corresponding to the current technological parameter obtained after the iteration; inputting the updated target semantic vector into the pre-trained industrial large model to obtain new technological parameters output by the industrial large model based on the updated meta-parameters; and continuing to execute the iteration of the optimization algorithm until a preset convergence condition is reached, so as to obtain the optimal technological parameter.
4. A method according to any one of claims 1-3, wherein said determining a hybrid prize value based on the predicted values of the semantic vector and the physical property indicators corresponding to the current process parameter, the target semantic vector and the target values of the physical property indicators, and a plurality of physical constraint functions comprises: on a pre-constructed semantic Riemann manifold, determining a semantic Riemann distance based on a semantic vector corresponding to the current process parameter and the target semantic vector; Determining a physical Riemann distance on a pre-constructed physical Riemann manifold based on the predicted values of the physical performance indexes, the physical constraint functions and the target values of the physical performance indexes; the hybrid prize value is determined based on the semantic and the physical Riemann distances.
5. The method of claim 4, wherein the determining the hybrid prize value based on the semantic and physical Riemann distances comprises: Determining a plurality of objective functions and a target extremum of each objective function according to the task demand information, wherein the objective functions comprise an energy consumption objective function, a qualification rate objective function and a device loss objective function; based on the predicted values of the physical performance indexes, respectively determining the function value of each objective function; Calculating a normalized loss corresponding to each objective function based on the function value of each objective function and the objective extremum corresponding to the objective function; Carrying out weighted summation on the normalized losses of the plurality of objective functions to obtain multi-objective normalized losses; determining a hybrid Riemann distance according to the semantic Riemann distance and the physical Riemann distance; and determining the mixed rewards value according to the mixed Riemann distance and the multi-objective normalized loss.
6. The method of claim 5, wherein the method further comprises: Determining a bonus gradient of a meta-parameter based on the hybrid bonus value, the physical distance gradient, the semantic distance gradient and the multi-objective loss gradient, the bonus gradient of the meta-parameter being used to update the meta-parameter and the objective semantic vector after each iteration of the optimization algorithm; wherein the physical distance gradient is determined based on the physical Riemann distance, the semantic distance gradient is determined based on the semantic Riemann distance, and the multi-objective loss gradient is determined based on the multi-objective normalized loss.
7. The method of claim 4, wherein said determining a physical Riemann distance based on the predicted values of the physical performance indicators, the plurality of physical constraint functions, the target values of the physical performance indicators on a pre-constructed physical Riemann manifold comprises: For each physical constraint function, determining a function value and a constraint satisfaction weight corresponding to the physical constraint function based on the predicted values of the physical performance indexes; determining a measurement tensor on the physical Riemann manifold based on the function value corresponding to each physical constraint function and the constraint satisfaction degree weight; the physical Riemann distance is determined based on a metric tensor on the physical Riemann manifold, a predicted value of the physical performance indicators, and a target value of the physical performance indicators.
8. An electronic device is characterized by comprising a memory and a processor; the memory stores computer-executable instructions; the processor executing computer-executable instructions stored in the memory, causing the processor to perform the industrial internet-based production process parameter optimization method of any one of claims 1-7.
9. A computer readable storage medium, wherein computer executable instructions are stored in the computer readable storage medium, which when executed by a processor is adapted to implement the industrial internet based production process parameter optimization method according to any one of claims 1-7.
10. A computer program product comprising a computer program which, when executed by a processor, implements the industrial internet-based production process parameter optimization method according to any one of claims 1-7.

Description

Production process parameter optimization method, equipment, medium and product based on industrial Internet Technical Field The application relates to the technical field of process optimization, in particular to a production process parameter optimization method, equipment, medium and product based on industrial Internet. Background Production tasks in the modern manufacturing industry exhibit the core feature of "multi-species, small lot, fast switching". This means that the production line needs to replace different products frequently, only a small amount of trial production data is needed in the initial stage of each new product, and the production replacement time is required to be very short. In the prior art, a process Parameter generating method based on a large language model (Large Language Model, LLM) is used, specifically, a Parameter-efficient fine Tuning technique (Parameter-EFFICIENT FINE-Tuning, PEFT) is adopted to optimize local parameters of the model, so that the model is adapted to a specific production scene to generate process parameters under different production demands. However, the above large language model based on the parameter efficient fine tuning technique uses the semantic loss of the minimum generated process parameters as the optimization target, and cannot sense the actual physical safety boundary, so the generated process parameters may violate the physical rule. Disclosure of Invention The embodiment of the application provides a production process parameter optimization method, equipment, medium and product based on industrial Internet, which are used for solving the problem that the generated process parameter may violate a physical rule because the actual physical safety boundary cannot be perceived by the existing process parameter generation method. In a first aspect, an embodiment of the present application provides a method for optimizing production process parameters based on industrial internet, including: Determining target semantic vectors and target values of various physical performance indexes according to task demand information of a target production task; determining the current technological parameters corresponding to the target semantic vector and the predicted values of various physical performance indexes corresponding to the current technological parameters; Determining a mixed rewarding value based on a semantic vector corresponding to the current process parameter, predicted values of various physical performance indexes, target semantic vectors, target values of various physical performance indexes and a plurality of physical constraint functions, wherein the physical constraint functions are used for limiting the value of the current process parameter; and taking the maximum mixed rewarding value as an optimization target, and adopting an optimization algorithm to determine the optimal technological parameters of the target production task. Optionally, the determining the current process parameter corresponding to the target semantic vector and the predicted value of each physical performance index corresponding to the current process parameter includes: Inputting the target semantic vector into a pre-trained industrial large model to obtain a current technological parameter output by the industrial large model based on a current meta-parameter, wherein the current meta-parameter is obtained by training a trial production technological parameter of the target production task, a semantic vector corresponding to the trial production technological parameter and actual values of various physical performance indexes; And inputting the current process parameters into a pre-trained prediction model to obtain predicted values of the physical performance indexes output by the prediction model, wherein the prediction model is obtained by training historical process parameters of a plurality of historical production tasks and actual values of the physical performance indexes corresponding to the historical process parameters. Optionally, the optimizing algorithm is used to determine the optimal technological parameters of the target production task by taking the maximum mixed rewards value as an optimizing target, and the optimizing algorithm comprises the following steps: After each iteration of the optimization algorithm, updating the meta-parameters and the target semantic vector based on the mixed rewards value corresponding to the current technological parameter obtained after the iteration; inputting the updated target semantic vector into the pre-trained industrial large model to obtain new technological parameters output by the industrial large model based on the updated meta-parameters; and continuing to execute the iteration of the optimization algorithm until a preset convergence condition is reached, so as to obtain the optimal technological parameter. Optionally, the determining the hybrid prize value based on the semantic vector correspon