CN-121980258-A - Production management method and system based on discrete industrial intelligent agent

CN121980258ACN 121980258 ACN121980258 ACN 121980258ACN-121980258-A

Abstract

The invention relates to a production management method and a system based on discrete industrial agents, which relate to the field of production management, and are characterized in that a production prediction sample and quality inspection decision sample data set is acquired, data weight division is carried out on the two data sets to obtain two sample weight sets, a production prediction and quality inspection decision path array, a first prediction accuracy set and a decision accuracy set are obtained after integrated training, the two path arrays are combined to obtain a discrete industrial agent array, combined optimization training is carried out, a second prediction and decision accuracy set is obtained after testing, current production basic data are acquired and input into the agent array, a prediction production yield and decision quality inspection parameter are output, and production management is carried out according to errors of the second prediction and decision accuracy set and the first prediction and decision accuracy set, so that the production management is carried out according to a prediction production yield and decision quality inspection parameter interval, and the technical problem that the production management is difficult to realize data interaction and business collaboration of a plurality of complex independent scenes in discrete manufacturing industry is solved.

Inventors

SU YUJUN
SU YUXUE
ZHANG JIPING
LAI JIE
HU DAOHUA

Assignees

浙江中之杰智能系统有限公司

Dates

Publication Date: 20260505
Application Date: 20251215
Priority Date: 20250403

Claims (10)

1. A method of production management based on discrete industrial agents, the method comprising: Collecting a production prediction sample data set and a quality inspection decision sample data set which comprise a sample production yield set in a discrete production scene and a quality inspection scene; Carrying out data weight division on the production prediction sample data set to obtain a production sample weight set, carrying out integrated training to obtain a production prediction path array and a first prediction accuracy set, carrying out data weight division on the quality inspection decision sample data set to obtain a quality inspection sample weight set, and carrying out integrated training to obtain a quality inspection decision path array and a first decision accuracy set; combining the production prediction path array and the quality inspection decision path array to obtain a discrete industrial intelligent agent array, performing joint optimization training, and testing to obtain a second prediction accuracy set and a second decision accuracy set; And acquiring current production basic data, inputting the discrete industrial intelligent agent array, outputting and obtaining predicted production yield and decision quality inspection parameters, compensating according to errors of the second prediction accuracy set and the second decision accuracy set, the first prediction accuracy set and the first decision accuracy set, obtaining a predicted production yield interval and a decision quality inspection parameter interval, and carrying out production management.
2. The discrete industrial agent-based production management method of claim 1, wherein collecting a production prediction sample data set and a quality control decision sample data set, each comprising a sample production yield set, within a discrete production scenario and a quality control scenario, comprises: In a discrete production scene of a manufacturing factory, collecting a sample production basic data set and a sample production yield set according to production detection data in historical time, and combining to obtain a production prediction sample data set; and in a discrete quality inspection scene, acquiring a sample production yield set and a sample quality inspection parameter set according to quality inspection decision data in the historical time, and combining to obtain a quality inspection decision sample data set.
3. The discrete industrial agent-based production management method of claim 1, wherein performing data weight division on the production prediction sample dataset to obtain a production sample weight set, and integrating training to obtain a production prediction path array and a first prediction accuracy set, comprises: according to the sample production yield set, carrying out weight calculation and distribution to obtain the production sample weight of each production prediction sample, and combining to obtain a production sample weight set, wherein the size of the sample production yield is inversely related to the size of the production sample weight; deep learning is adopted to construct a first production prediction path; Performing supervised training and testing on the first production prediction path by adopting the production prediction sample data set according to the production sample weight set to obtain a production test result set and prediction accuracy, and calculating to obtain an adjusted production sample weight set, wherein the supervised training is performed by configuring training resources according to the production sample weight, and the size of the production sample weight is positively correlated with the training resources of each production prediction sample data; And continuing to construct a second production prediction path, training by adopting the production prediction sample data set according to the adjustment production sample weight set until training is carried out to obtain U production prediction paths, obtaining a production prediction path array, and testing to obtain a first prediction accuracy set, wherein U is a positive integer.
4. The discrete industrial agent-based production management method of claim 3, wherein performing supervisory training and testing on the first production prediction path using the production prediction sample dataset according to the production sample weight set to obtain a production test result set and a prediction accuracy, and calculating to obtain an adjusted production sample weight set, comprises: According to the production sample weight set, carrying out iterative supervision training on the first production prediction path by adopting the production prediction sample data set, and testing each production prediction sample data after convergence to obtain a production test result set, wherein the production test result comprises accurate prediction or inaccurate prediction; calculating the duty ratio of the production test result with accurate prediction in the production test result set, and obtaining the prediction accuracy; According to the production sample weight set, the production test result set and the prediction accuracy, calculating and adjusting the production sample weight set, wherein the formula is as follows: ; Wherein, the Adjusting production sample weights in the t+1th production prediction path training for the ith production prediction sample data, The production sample weight of the ith production prediction sample data in the training of the t-th production prediction path, And C is an indication function, wherein the prediction accuracy of the t-th production prediction path is the prediction accuracy of the t-th production prediction path.
5. The discrete industrial agent-based production management method of claim 1, wherein the performing data weight division on the quality inspection decision sample dataset to obtain a quality inspection sample weight set, and integrating training to obtain a quality inspection decision path array and a first decision accuracy set comprises: Counting the occurrence rate of each sample production yield in the sample production yield set, and carrying out weight distribution calculation on each quality inspection decision sample data to obtain a quality inspection sample weight set, wherein the magnitude of the occurrence rate is positively correlated with the magnitude of the quality inspection sample weight; deep learning is adopted to construct a first quality inspection decision path; According to the quality inspection sample weight set, performing supervision training and testing on the first quality inspection decision path by adopting the quality inspection decision sample data set to obtain a quality inspection test result set and decision accuracy, and calculating to obtain an adjusted quality inspection sample weight set; and continuing to construct a second quality inspection decision path, training by adopting the quality inspection decision sample data set according to the quality inspection sample weight set until training to obtain U production prediction paths, obtaining a quality inspection decision path array, and testing to obtain a first decision accuracy set, wherein U is a positive integer.
6. The discrete industrial agent-based production management method of claim 1, wherein combining the production prediction path array and the quality inspection decision path array to obtain a discrete industrial agent array, performing joint optimization training, and testing to obtain a second prediction accuracy set and a second decision accuracy set, comprises: combining each production prediction path and quality inspection decision path in the production prediction path array and the quality inspection decision path array to form a discrete industrial intelligent agent array; adopting a sample production basic data set and a sample quality inspection parameter set in the production prediction sample data set and the quality inspection decision sample data set to perform optimization training on the discrete industrial intelligent agent array until convergence optimization training times are reached; And testing a plurality of discrete industrial agents in the discrete industrial agent array by adopting the production prediction sample data set and the quality control decision sample data set to obtain a second prediction accuracy set and a second decision accuracy set.
7. The method for production management based on discrete industrial agents according to claim 1, wherein current production basic data is acquired, the discrete industrial agents are input into the array, predicted production yield and decision quality inspection parameters are output, compensation is performed according to errors of the second prediction accuracy set, the second decision accuracy set, the first prediction accuracy set and the first decision accuracy set, a predicted production yield interval and a decision quality inspection parameter interval are obtained, and production management is performed, comprising: Obtaining current production basic data, inputting a plurality of discrete industrial intelligent agents in the discrete industrial intelligent agent array, outputting an output production yield set and an output decision quality parameter set, and respectively calculating average values to obtain a predicted production yield and decision quality inspection parameters; Configuring a prediction compensation coefficient and a decision compensation coefficient according to errors of the second prediction accuracy set and the second decision accuracy set, the first prediction accuracy set and the first decision accuracy set; And carrying out error compensation on the predicted production yield and the decision quality inspection parameters by adopting the predicted compensation coefficient and the decision compensation coefficient to obtain a predicted production yield interval and a decision quality inspection parameter interval, and carrying out production management.
8. The discrete industrial agent-based production management method of claim 7, wherein configuring compensation coefficients based on errors of the second set of prediction accuracies, the second set of decision accuracies and the first set of prediction accuracies, the first set of decision accuracies, comprises: Calculating to obtain a plurality of prediction accuracy error magnitudes according to the second prediction accuracy set and the first prediction accuracy set, and calculating a mean value as a prediction compensation coefficient; And calculating to obtain a plurality of decision accuracy error magnitudes according to the second decision accuracy set and the first decision accuracy set, and calculating a mean value to obtain a decision compensation coefficient.
9. The discrete industrial agent-based production management method of claim 7, wherein obtaining current production base data, inputting a plurality of discrete industrial agents within the array of discrete industrial agents, outputting an obtained output production yield set and an output decision quality parameter set, further comprises: Acquiring current production basic data; Acquiring a historical production basic data sequence in a recent historical time range, and calculating the production representativeness of the production basic data in the historical production basic data sequence; configuring the quantity of the intelligent agents according to the production representativeness; And selecting discrete industrial agents in the discrete industrial agent array according to the number of the agents, inputting the production basic data, and outputting to obtain an output production yield set and an output decision quality parameter set.
10. A discrete industrial agent-based production management system for implementing the discrete industrial agent-based production management method of any one of claims 1-9, the system comprising: the data acquisition module is used for acquiring a production prediction sample data set and a quality inspection decision sample data set which both comprise a sample production yield set in a discrete production scene and a quality inspection scene; The weight dividing module is used for carrying out data weight division on the production prediction sample data set to obtain a production sample weight set, carrying out integrated training to obtain a production prediction path array and a first prediction accuracy set, carrying out data weight division on the quality inspection decision sample data set to obtain a quality inspection sample weight set, and carrying out integrated training to obtain a quality inspection decision path array and a first decision accuracy set; The optimization training module is used for combining the production prediction path array and the quality inspection decision path array to obtain a discrete industrial intelligent agent array, performing joint optimization training, and testing to obtain a second prediction accuracy set and a second decision accuracy set; The production management module is used for acquiring current production basic data, inputting the discrete industrial intelligent agent array, outputting and obtaining predicted production yield and decision quality inspection parameters, compensating according to errors of the second prediction accuracy set, the second decision accuracy set, the first prediction accuracy set and the first decision accuracy set, obtaining a predicted production yield interval and a decision quality inspection parameter interval, and carrying out production management.

Description

Production management method and system based on discrete industrial intelligent agent Technical Field The invention relates to the field of production management, in particular to a production management method and system based on discrete industrial intelligent agents. Background In current industrial production environments, particularly in discrete manufacturing industries, such as automotive component manufacturing, electronic equipment assembly, and the like, the production process often includes multiple complex and independent scenarios, such as logistics, production, quality, and automation. There is a need for data interaction and business coordination between these scenarios, but it is difficult to implement unified management and intelligent control due to its discreteness. Existing smart manufacturing technologies often rely on an overall smart model in an attempt to manage and optimize the overall production process within a unified framework. However, in discrete manufacturing, such an overall smart model is often difficult to adapt due to the diversity of production scenarios, complexity of data, and real-time requirements. For example, in the manufacturing process, a plurality of scenes such as production, material flow in workshops and quality inspection exist, data dimension and data association characteristics are different among different scenes, so that generalization and applicability of the trained whole intelligent model are low, and the problems of low production efficiency, difficult quality control and the like are caused. Disclosure of Invention Aiming at the technical problems that the data interaction and business coordination of a plurality of complex independent scenes and the defects of intelligent management and optimization are difficult to realize in the discrete manufacturing industry in the prior art, the invention provides a production management method and system based on discrete industrial intelligent bodies. The technical scheme for solving the technical problems is as follows: The invention provides a production management method based on discrete industrial agents, which comprises the steps of collecting a production prediction sample data set and a quality inspection decision sample data set which comprise sample production yield sets in discrete production scenes and quality inspection scenes, carrying out data weight division on the production prediction sample data set to obtain a production sample weight set, carrying out integrated training to obtain a production prediction path array and a first prediction accuracy set, carrying out data weight division on the quality inspection decision sample data set to obtain a quality inspection sample weight set, carrying out integrated training to obtain a quality inspection decision path array and a first decision accuracy set, combining the production prediction path array and the quality inspection decision path array to obtain a discrete industrial agent array, carrying out joint optimization training, carrying out testing to obtain a second prediction accuracy set and a second decision accuracy set, obtaining current production basic data, inputting the discrete industrial agent array, outputting to obtain a prediction production yield and decision quality inspection parameters, carrying out compensation according to the second prediction accuracy set, the second decision accuracy set and the first prediction accuracy set and the error of the first prediction accuracy set, and carrying out quality inspection parameter management in a production interval. The invention provides a production management system based on discrete industrial agents, which comprises a data acquisition module, a weight dividing module, a production management module and a quality control module, wherein the data acquisition module is used for acquiring a production prediction sample data set and a quality control decision sample data set which comprise a sample production yield set in a discrete production scene and a quality control scene, the weight dividing module is used for carrying out data weight division on the production prediction sample data set to obtain a production sample weight set, carrying out integrated training to obtain a production prediction path array and a first prediction accuracy set, carrying out data weight division on the quality control decision sample data set to obtain a quality control sample weight set, carrying out integrated training to obtain a quality control decision path array and a first decision accuracy set, and carrying out optimization training module is used for combining the production prediction path array and the quality control decision path array to obtain a discrete industrial agent array, carrying out joint optimization training to obtain a second prediction accuracy set and a second decision accuracy set, and outputting to obtain a prediction yield and a decision accuracy set, and