CN-121981585-A - Supplier evaluation method, system, equipment and medium based on material quality classification

Abstract

The invention relates to the technical field of material quality management, in particular to a provider evaluation method, a system, equipment and a medium based on material quality grading, which comprise the steps of obtaining full life cycle multisource data of materials supplied by a provider; the quality classification method comprises the steps of carrying out quality classification on materials by using a preset intelligent classification model to obtain quality grades, constructing a multi-dimensional evaluation index system by combining with provider performance data based on the quality grades of the materials, and finally calculating comprehensive evaluation scores of providers by using the system and classifying the grades according to the comprehensive evaluation scores. According to the invention, through fusion analysis of full life cycle multi-source data, objectification, automation and intellectualization of material quality classification are realized, and subjective bias of manual evaluation is effectively overcome. The multidimensional provider evaluation system constructed based on the intelligent grading result enables the provider performance evaluation to be more comprehensive, accurate and dynamically adjustable, and provides scientific basis for supply chain management.

Inventors

• Kuang Xuelian
• CHEN JIMING
• WANG SHENG
• ZHANG HONGYUAN
• GAO ZEKAI
• DOU HENG
• ZENG NA
• AN WENZHE
• WANG RUIXIN
• WANG YOUJIE
• ZHAI JIQING
• QI LUFENG
• ZHAO HONGWEI
• WANG XIAOMIN
• WANG LI
• ZHANG BAIFU
• XU YANFA

Assignees

• 山东鲁软数字科技有限公司

Dates

Publication Date

20260505

Application Date

20251203

Claims (10)

1. 1. A method for evaluating suppliers based on quality classification of materials, comprising: acquiring full life cycle multi-source data of materials supplied by a supplier, wherein the full life cycle multi-source data comprises at least one of purchasing data, production data, transportation data, storage data, use data and external environment data; based on the full life cycle multi-source data, carrying out quality classification on the materials by utilizing a preset intelligent classification model to obtain quality grades of the materials; Based on the quality grade of the materials, combining performance data of the suppliers to construct a multi-dimensional evaluation index system of the suppliers; And calculating the comprehensive evaluation score of the supplier by using the multi-dimensional evaluation index system, and grading the supplier based on the comprehensive evaluation score.
2. 2. The method of claim 1, wherein quality grading the material using a preset intelligent grading model based on the full life cycle multisource data to obtain a quality grade of the material comprises: Constructing a dynamic causal graph, wherein the dynamic causal graph is obtained by processing the full life cycle multi-source data through a causal discovery algorithm and is used for representing causal relations among variables affecting the quality of materials; Constructing a digital twin simulation environment based on the dynamic causal graph; Training an intelligent agent by adopting a multi-intelligent-agent reinforcement learning algorithm in the digital twin simulation environment so as to learn to obtain an intelligent agent strategy for classifying the quality of materials, thereby forming the intelligent classification model; And analyzing the multidimensional material data input in real time by using the intelligent grading model after training, and outputting the quality grade of the material.
3. 3. The method of claim 2, wherein constructing a dynamic causal graph comprises: Performing space-time alignment and feature engineering processing on the full-life-cycle multi-source data to generate a unified time sequence feature matrix, wherein the feature engineering processing comprises generation of statistical features and causal inference guiding features, and the causal inference guiding features comprise approximate entropy of a time sequence, accumulated time exceeding a preset threshold value and fluctuation rates of quality parameters among different batches; A causal discovery algorithm based on continuous optimization is adopted, the time sequence feature matrix is taken as input, and a sparse adjacency matrix representing causal relation among variables is solved by minimizing a loss function combined with L1 regularization, wherein the causal discovery algorithm is NOTEARS algorithm; The method comprises the steps of embedding domain knowledge into an optimization process as hard constraint in a solving process, wherein the domain knowledge comprises deterministic causal rules defined by expert experience, and the hard constraint is realized by forcedly setting zero corresponding elements in the adjacency matrix, which violate the deterministic causal rules.
4. 4. The method of claim 2, wherein constructing a digital twin simulation environment based on the dynamic causal graph comprises: for each variable in the dynamic causal graph Based on all the parent node variable sets And (3) establishing a structural equation: Wherein the said Is a preset mapping function for quantization For a pair of Is due to the causal influence of (a) To act on variables The structural equations of all variables jointly form a structural causal model which is used as a dynamic evolution rule of the digital twin simulation environment; wherein the structural causal model provides dynamic evolution rules for the digital twin simulation environment, the rules being defined as, for any variable in the causal graph The value of which is represented by a function And generating, wherein, the method comprises the steps of, Representing variables And the digital twin simulation environment is used for simulating the dynamic change process of state variables of each link of the supply chain under the action intervention of multiple agents.
5. 5. The method of claim 2, wherein training agents in the digital twin simulation environment using a multi-agent reinforcement learning algorithm to learn an agent strategy for quality classification of materials, forming the intelligent classification model, comprises: the method comprises the steps of constructing a multi-agent system comprising a quality evaluation agent and a spot check strategy agent, and respectively configuring independent strategy networks and value networks for the quality evaluation agent and the spot check strategy agent; Designing a causal driving rewarding function, wherein the expression of the function is as follows: wherein In return for the end of the environment, For causal treatment effects of agent actions on the final return Y estimated based on the dual machine learning method, Is a trade-off coefficient; in the training process, a centralized criticizer architecture is adopted, namely, the value network of each intelligent agent is updated based on the state and action information of all intelligent agents; iteratively updating strategy network parameters of each agent by using a deterministic strategy gradient method by maximizing the causal driving reward function until the strategy converges to form the agent strategy for classifying the quality of materials; The quality evaluation intelligent agent and the sampling inspection strategy intelligent agent are constructed and updated by adopting a deterministic strategy gradient-based algorithm, wherein the quality evaluation intelligent agent's strategy network takes the material multidimensional quality data perceived by the quality evaluation intelligent agent as an input state and outputs an action representing the material quality grade, and the sampling inspection strategy intelligent agent's strategy network takes at least the material quality grade output by the quality evaluation intelligent agent as a part of the input state and outputs an action representing the dynamic sampling inspection rate.
6. 6. The method of claim 1, wherein constructing a multi-dimensional evaluation index system for a provider in combination with performance data for the provider based on the quality level of the material comprises: establishing a plurality of evaluation dimensions including quality, cost, delivery, service, and risk; Setting the evaluation basis of the quality dimension as a weighted average value of the quality grades of all materials supplied by the supplier and output by the intelligent grading model; setting an evaluation basis of the cost dimension as a hidden loss index based on the quality cost of the spot check and disqualification of the spot check; setting an evaluation basis of delivery dimensions as a delivery coordination index for evaluating delivery time rate and spot check batch qualification rate based on time sequence analysis; Setting an evaluation basis of service dimension as a service coordination index based on communication response timeliness and unqualified material processing efficiency; Setting the evaluation basis of the risk dimension as the accessed third party financial data and the negative emotion intensity obtained through the network public opinion analysis; and dynamically calculating and distributing the weight of each dimension based on the distribution of the index data of each dimension by adopting an entropy weight method so as to construct the multi-dimension evaluation index system.
7. 7. The method of claim 1, wherein calculating a composite rating score for a provider using the multi-dimensional rating index system and ranking the provider based on the composite rating score comprises: Carrying out weighted summation on the quantized data of each dimension in the multi-dimensional evaluation index system and the corresponding dimension weight determined by an entropy weight method, and calculating to obtain a comprehensive evaluation score of a provider; Presetting a provider grade corresponding to the comprehensive evaluation score range, wherein the grade at least comprises a strategic provider, a good provider, a qualified provider and a disqualified provider; the comprehensive evaluation score of the provider is mapped to a corresponding rank.
8. 8. A supplier evaluation system based on quality grading of materials, comprising: The system comprises a data acquisition module, a data processing module and a data processing module, wherein the data acquisition module is used for acquiring full life cycle multi-source data of materials supplied by suppliers, and the full life cycle multi-source data comprises at least one of purchasing data, production data, transportation data, storage data, use data and external environment data; The material grading module is used for grading the quality of the material by utilizing a preset intelligent grading model based on the full life cycle multi-source data to obtain the quality grade of the material; The index construction module is used for constructing a multi-dimensional evaluation index system of the supplier based on the quality grade of the materials and combining performance data of the supplier; And the grade mapping module is used for calculating the comprehensive evaluation score of the supplier by utilizing the multi-dimensional evaluation index system and grading the supplier based on the comprehensive evaluation score.
9. 9. A provider assessment apparatus based on quality grading of materials, comprising: A memory for storing a supplier evaluation program based on the quality classification of the material; A processor for implementing the steps of the material quality grading based supplier assessment method according to any one of claims 1-7 when executing the material quality grading based supplier assessment program.
10. 10. A computer readable medium storing a computer program, characterized in that the readable medium has stored thereon a quality-grading based supplier evaluation program, which, when executed by a processor, implements the steps of the quality-grading based supplier evaluation method according to any one of claims 1-7.

Description

Supplier evaluation method, system, equipment and medium based on material quality classification Technical Field The invention belongs to the technical field of material quality management, and particularly relates to a provider evaluation method, a system, equipment and a medium based on material quality classification. Background In current supply chain management, the assessment of suppliers generally relies on human experience and scattered data. The traditional method is generally limited to limited dimensions such as purchase price, delivery timing rate and the like, and data are isolated from different systems such as ERP, WMS, quality inspection and the like, so that the data are difficult to integrate and utilize. The quality evaluation of materials is carried out by manual spot inspection, the subjectivity is strong, the efficiency is low, and the full link data of production, transportation, use and the like cannot be associated to trace back the root cause. This results in one-sided and lagged evaluation results, lack of prospective risk early warning, and difficulty in dynamically reflecting the actual performance and potential risk of the suppliers. Enterprises cannot accurately identify high-quality strategic partners, and high-risk suppliers are difficult to effectively manage and control, so that the toughness and the overall quality of a supply chain are restricted. The prior art lacks an automatic method capable of deeply fusing multi-source data, realizing intelligent objective classification of material quality and constructing a dynamic supplier evaluation system according to the intelligent objective classification. Disclosure of Invention The invention provides a supplier evaluation method, a system, equipment and a medium based on material quality classification, aiming at the defects of the prior art, so as to solve the technical problems. In a first aspect, the present invention provides a method for evaluating suppliers based on quality classification of materials, comprising: acquiring full life cycle multi-source data of materials supplied by a supplier, wherein the full life cycle multi-source data comprises at least one of purchasing data, production data, transportation data, storage data, use data and external environment data; based on the full life cycle multi-source data, carrying out quality classification on the materials by utilizing a preset intelligent classification model to obtain quality grades of the materials; Based on the quality grade of the materials, combining performance data of the suppliers to construct a multi-dimensional evaluation index system of the suppliers; And calculating the comprehensive evaluation score of the supplier by using the multi-dimensional evaluation index system, and grading the supplier based on the comprehensive evaluation score. In an optional embodiment, based on the full life cycle multisource data, performing quality classification on the materials by using a preset intelligent classification model to obtain quality grades of the materials, including: Constructing a dynamic causal graph, wherein the dynamic causal graph is obtained by processing the full life cycle multi-source data through a causal discovery algorithm and is used for representing causal relations among variables affecting the quality of materials; Constructing a digital twin simulation environment based on the dynamic causal graph; Training an intelligent agent by adopting a multi-intelligent-agent reinforcement learning algorithm in the digital twin simulation environment so as to learn to obtain an intelligent agent strategy for classifying the quality of materials, thereby forming the intelligent classification model; And analyzing the multidimensional material data input in real time by using the intelligent grading model after training, and outputting the quality grade of the material. In an alternative embodiment, constructing the dynamic causal graph comprises: Performing space-time alignment and feature engineering processing on the full-life-cycle multi-source data to generate a unified time sequence feature matrix, wherein the feature engineering processing comprises generation of statistical features and causal inference guiding features, and the causal inference guiding features comprise approximate entropy of a time sequence, accumulated time exceeding a preset threshold value and fluctuation rates of quality parameters among different batches; A causal discovery algorithm based on continuous optimization is adopted, the time sequence feature matrix is taken as input, and a sparse adjacency matrix representing causal relation among variables is solved by minimizing a loss function combined with L1 regularization, wherein the causal discovery algorithm is NOTEARS algorithm; The method comprises the steps of embedding domain knowledge into an optimization process as hard constraint in a solving process, wherein the domain knowledge comprises deter