Standard implementation process quality evaluation method based on quantification model

Abstract

The application discloses a standard implementation process quality evaluation method based on a quantitative model, which comprises the following steps: performing semantic extraction on standard implementation materials to construct a standard implementation process knowledge graph, extracting standard implementation process data to calculate process entropy and process entropy weight coefficients of each implementation stage, constructing a viewpoint fuzzy number according to an implementation subject and calculating an implementation subject divergence degree, adjusting a rolling evaluation window width to calculate a comprehensive quality index of the standard implementation process data, performing dynamic calibration to obtain a comprehensive calibration quality index and performing quality early warning, mapping a decision event set in the standard implementation process data to the standard implementation process knowledge graph, performing root cause analysis through a graph neural network, and performing quality evaluation and tracing in combination with the process entropy. The method can not only improve the diagnosis efficiency and rectification pertinence of the standard implementation process quality problems, but also reduce the standard implementation management cost and promote the digital transformation of the standardization work.

Description

Technical Field

[0001] This invention relates to the field of intelligent assessment technology for the quality of standard implementation, and in particular to a method for assessing the quality of the standard implementation process based on a quantitative model. Background Technology

[0002] Standardization, as a fundamental institutional arrangement supporting national economic and social development, essentially aims to achieve unified norms and optimal order in production, management, and services through the formulation, publication, and implementation of standards. The effectiveness of standard implementation directly determines the final outcome of standardization work, and most quality deviations in standardization work occur during the implementation process. Therefore, building a quality assessment system covering the entire chain of standard formulation, publication, dissemination, implementation, supervision, and improvement has become a core element in enhancing standardization governance capabilities.

[0003] Existing standards implementation quality assessment techniques primarily rely on manual review and static result comparison, which has significant limitations: First, the assessment dimensions are too singular; traditional methods focus on judging the compliance of implementation results, lacking in-depth semantic analysis of standard implementation materials (standard clause requirements, summary processing of opinions, review meeting minutes, etc.), and failing to uncover the implicit process evolution logic and decision-making connections. Second, the assessment mechanism is static; current technologies mostly adopt a fixed-cycle post-implementation sampling model, which cannot adapt to the dynamic changes in the standard implementation process and is difficult to provide real-time early warning and trend prediction of implementation risks. Finally, existing technologies rely on the experience of the implementing entities for qualitative tracing of quality problems, lacking quantitative analysis methods, and making it difficult to accurately locate the deep-seated root causes such as defects in standard content, disagreements in handling opinions, or errors in review decisions. Therefore, this invention proposes a standards implementation process quality assessment method based on a quantitative model. By establishing a system that integrates knowledge graph construction, process entropy calculation, and deep learning root cause analysis, it achieves dynamic monitoring, precise source tracing, and closed-loop optimization of the implementation process. Summary of the Invention

[0004] The purpose of this invention is to provide a method for quality assessment of standard implementation processes based on a quantitative model.

[0005] To achieve the above objectives, the present invention is implemented according to the following technical solution: This invention includes the following steps: Extract standard implementation materials and input them into a hybrid semantic extraction model for semantic extraction. Construct a knowledge graph of the standard implementation process based on the semantic extraction results. Extract data from the standard implementation process, calculate process entropy and process entropy weight coefficients based on the implementation stage, construct a fuzzy number of viewpoints based on the implementing entities, and calculate the degree of disagreement among the implementing entities; The width of the rolling evaluation window is determined based on the process entropy. The comprehensive quality index of the standard implementation process data within the evaluation window is calculated in combination with the divergence of the implementing entities. Dynamic calibration is performed to obtain the comprehensive calibration quality index and to issue quality warnings. The set of decision events in the standard implementation process data is mapped to the standard implementation process knowledge graph, and root cause analysis is performed through graph neural network, combined with process entropy for quality assessment and tracing. The standard implementation materials include standard clause requirements, summary and processing of comments, review meeting minutes, etc. The data for the standard implementation process includes data on implementation preparation, publicity and training, execution and implementation, supervision and inspection, and evaluation and improvement. The decision-making events include disagreements, event decisions, and event quality.

[0006] Furthermore, the method for constructing a knowledge graph of the standard implementation process includes: The standard implementation materials are extracted using a hybrid semantic extraction model for semantic extraction. The hybrid semantic extraction model includes a process element identification layer, a relational reasoning layer, and a temporal evolution layer; The process element identification layer uses the BERT-BiLSTM-CRF model to extract process entities; the process entities include process nodes and responsible entities; the process nodes correspond to the standard implementation stages; and the responsible entities correspond to the standard implementation materials. The relation reasoning layer employs a document-level relation extraction network based on an attention mechanism to establish the association edges of entities in each process and calculate the edge weights. The temporal evolution layer constructs a dynamic adjacency matrix for the evolution of the standard implementation process knowledge graph based on the transfer function and process time nodes; The process entities extracted from the process element identification layer are used as graph nodes, and a standard implementation process knowledge graph is constructed by combining the associated edges and edge weights of each process entity.

[0007] Furthermore, the method for calculating process entropy and process entropy weighting coefficients includes: Extract standard implementation process data and calculate process entropy based on the total number of activity categories in each implementation stage. Then, perform normalization to calculate the process entropy weighting coefficient. The expression is as follows: ; ; in for Implementation phase process entropy The number of phases to be implemented. for Total number of activity categories during the implementation phase for Implementation phase The probability of occurrence of this type of activity for Entropy weighting coefficient during the implementation phase.

[0008] Furthermore, the method for calculating the degree of divergence among implementing entities includes: Based on the feedback documents from various implementing entities at different implementation stages, a fuzzy number of feedback from each implementing entity is constructed. The similarity of viewpoints among the implementing entities is calculated, and the relative consistency of viewpoints among the implementing entities is also calculated. The expression is as follows: ; ; ; in As the implementing entity and Similarity of viewpoints , As the implementing entity and implementing entities The number of ambiguous viewpoints , As the implementing entity and implementing entities exist Handling of opinions during the implementation phase As the implementing entity The relative consistency For the number of implementing entities, As the implementing entity Authority and weight This is an institutional qualification level coefficient, quantified based on the institution's qualification level and industry access qualifications. This is the business relevance coefficient for the organization, quantified based on the degree of matching between the organization's main business and the content of the standard implementation. This is the institution's historical assessment reliability coefficient, quantified based on the proportion of past opinions adopted or historical assessment deviations. Sensitivity to group consensus coefficient The entity that implemented the previous round of assessment The group consensus coefficient; The consensus coefficient of each implementing entity is calculated based on the relative consistency of their viewpoints. The divergence degree between the implementing entities is then calculated using the consensus coefficient and the similarity of their viewpoints. The expression is as follows: ; ; in To determine the degree of disagreement among the implementing entities, , As the implementing entity and implementing entities The group consensus coefficient As a relaxation factor, As the implementing entity Its authority and weight.

[0009] Furthermore, the method for obtaining the comprehensive calibration quality index includes: The width of the rolling evaluation window is determined based on the process entropy. The comprehensive calibration quality index is obtained by dynamically calibrating the comprehensive quality index of the standard implementation process data within the evaluation window, expressed as: ; ; in To evaluate the width of the scrolling evaluation window, , For minimum window width and maximum window width, To comprehensively calibrate the quality index, For smoothing coefficients, The comprehensive calibration quality index is based on the implementation time of the previous round of standards. For the comprehensive quality index, The standard implementation time; The comprehensive quality index is calculated by combining the degree of divergence among implementing entities, and its expression is: ; in For the comprehensive quality index, The time decay coefficient, For standard execution time, To evaluate the information entropy indicator, To determine the degree of disagreement among the implementing entities, , These are the process entropy and the process entropy weighting coefficient, respectively. The maximum process entropy; A quality warning will be issued when the overall calibration quality index falls below the overall quality index threshold. The cosine similarity between the mean of the fuzzy vector of feedback from the implementing entity in the standard implementation process data after the evaluation window is updated and the mean of the fuzzy vector of feedback from the implementing entity corresponding to the risk events in the historical risk case database is calculated. The response strategies corresponding to the top three risk events with the highest similarity are selected as alternative response strategies.

[0010] Furthermore, the method for quality assessment and traceability includes: Based on the process graph nodes of the standard implementation process knowledge graph corresponding to each decision-making matter, and according to the mapping rules, the decision-making events are mapped to the corresponding responsible entity nodes of each process graph node in the standard implementation process knowledge graph, thus forming an enhanced standard implementation process graph: The disagreements are mapped to nodes in the standard clause requirements diagram, and the disagreements and the implementation standards are used as node features in the standard clause requirements diagram. The event decision is mapped to nodes in the opinion summary and processing graph, and the event decision, processing rules, and differences in the viewpoints of the implementing entities are used as the features of the nodes in the opinion summary and processing graph; the differences in the viewpoints of the implementing entities include the standard deviation and skewness of the opinion processing of the implementing entities in the corresponding standard implementation stage. Event quality is mapped to nodes in the review meeting minutes diagram, and event quality, review criteria, and process entropy are used as node features in the review meeting minutes diagram; the event quality is taken as the average score of the implementing entity at the corresponding standard implementation stage; A graph attention network is used to enhance the graph for root cause propagation during the standard implementation process, and the root cause score of each responsible entity's graph node is calculated. The expression is as follows: ; ; in for Standard Implementation Phase Root cause score of the responsible party node. This is the set of neighboring nodes of a node in the responsible entity graph, including the node in the process graph to which it belongs and other associated responsible entity nodes. Attention coefficient, reflecting the node right The contribution of quality impact, For nodes With nodes edge weights, The characteristic transformation matrix, Features of neighboring nodes , This is the process entropy adjustment coefficient. for Entropy of the standard implementation phase For the maximum process entropy, This is a learnable attention parameter vector. For nodes Features; The standard implementation stage where the process entropy is greater than the process entropy threshold is identified as the quality impact stage. The responsible entity node with the highest root cause score in the process diagram node corresponding to the quality impact stage is selected as the standard implementation impact factor to obtain the standard implementation impact factor set. Based on the standard implementation impact factor set, a response strategy is selected from the alternative response strategies.

[0011] The beneficial effects of this invention are: This invention is a standard implementation process quality assessment method based on a quantitative model. Compared with existing technologies, this invention has the following technical advantages: This invention transforms scattered unstructured documents such as standard clause requirements, opinion summaries, and review meeting minutes into temporal graph structure data by constructing a hybrid semantic extraction model and a knowledge graph of the standard implementation process. This achieves semantic deconstruction and knowledge flow modeling of the standard implementation process, providing a process data source for quality assessment. This invention applies information entropy to the orderly assessment of standard implementation processes by calculating process entropy and the degree of divergence among implementing entities, and integrates trapezoidal fuzzy numbers to handle the heterogeneity of opinions among implementing entities. This overcomes the one-sidedness of traditional assessments that only focus on result deviations, and provides a computable mathematical basis for dynamic risk assessment. This invention establishes an adaptive logic of "the more chaotic the process, the more intensive the monitoring" by determining the width of the rolling evaluation window and the state calibration mechanism of the exponential weighted moving average, thereby realizing real-time smooth tracking of the quality index and dynamic allocation of early warning resources. This invention maps a set of decision events to a knowledge graph and introduces root cause analysis using graph attention networks with entropy enhancement. This enables precise identification of key responsible parties, tracing quality issues back to specific clauses in standard requirements, opinion summaries, or review meeting minutes, and triggering corresponding adjustment strategies. This forms a technical closed loop of "assessment-tracing-optimization," significantly improving the diagnostic efficiency and targeted rectification of standard implementation quality issues. It has significant practical value for reducing standardization management costs and promoting the digital transformation of standardization work. Attached Figure Description

[0012] Figure 1 This is a flowchart illustrating the steps of the standard implementation process quality assessment method based on a quantitative model, as described in this invention. Detailed Implementation

[0013] The present invention will be further described below through specific embodiments. The illustrative embodiments and descriptions herein are used to explain the present invention, but are not intended to limit the present invention.

[0014] The present invention provides a standard implementation process quality assessment method based on a quantitative model, comprising the following steps: like Figure 1 As shown, this embodiment includes the following steps: Extract standard implementation materials and input them into a hybrid semantic extraction model for semantic extraction. Construct a knowledge graph of the standard implementation process based on the semantic extraction results. Extract data from the standard implementation process, calculate process entropy and process entropy weight coefficients based on the implementation stage, construct a fuzzy number of viewpoints based on the implementing entities, and calculate the degree of disagreement among the implementing entities; The width of the rolling evaluation window is determined based on the process entropy. The comprehensive quality index of the standard implementation process data within the evaluation window is calculated in combination with the divergence of the implementing entities. Dynamic calibration is performed to obtain the comprehensive calibration quality index and to issue quality warnings. The set of decision events in the standard implementation process data is mapped to the standard implementation process knowledge graph, and root cause analysis is performed through graph neural network, combined with process entropy for quality assessment and tracing. The standard implementation materials include standard clause requirements, summary and processing of comments, review meeting minutes, etc. The data for the standard implementation process includes data on release preparation, dissemination and training, implementation and execution, supervision and inspection, and evaluation and improvement. The decision-making events include disagreements, event decisions, and event quality.

[0015] In this embodiment, the method for constructing a knowledge graph of the standard implementation process includes: The standard implementation materials are extracted using a hybrid semantic extraction model for semantic extraction. The hybrid semantic extraction model includes a process element identification layer, a relational reasoning layer, and a temporal evolution layer; The process element identification layer uses the BERT-BiLSTM-CRF model to extract process entities; the process entities include process nodes and responsible entities; the process nodes correspond to the standard implementation stages; and the responsible entities correspond to the standard implementation materials. The relation reasoning layer employs a document-level relation extraction network based on an attention mechanism to establish the association edges of entities in each process and calculate the edge weights. The temporal evolution layer constructs a dynamic adjacency matrix for the evolution of the standard implementation process knowledge graph based on the transfer function and process time nodes; The process entities extracted from the process element identification layer are used as graph nodes, and a standard implementation process knowledge graph is constructed by combining the associated edges and edge weights of each process entity. In actual evaluation, in the process element identification layer, BERT (Bidirectional Encoder Based on Transformer) is first used to encode the semantics of the standard implementation materials, then BiLSTM (Bidirectional Long Short-Term Memory Network) is used to perform sequence modeling of the semantic encoding, and finally the process entity is obtained by CRF (Conditional Random Field) label decoding; process nodes correspond to each stage of standard implementation; the same process diagram node includes three responsible entity diagram nodes; In the relational reasoning layer, the mixing coefficient is determined to be 0.3 through validation set optimization. The word frequency statistical similarity is calculated based on word frequency-inverse document frequency, reflecting the degree of lexical overlap on the document surface. The semantic embedding similarity is a deep semantic similarity calculated based on the semantic vectors output by the BERT model through cosine similarity, used to capture implicit semantic associations. The document fragment is a text unit of standard implementation materials after sentence or segmentation processing (such as a certain clause requirement or opinion processing in the standard implementation materials), corresponding to a process entity. The edge weight expression is: ; in For document fragments and The strength of the process association between the two documents, i.e., the edge weights of the process entities in the two documents. The mixing coefficient, For word frequency statistics similarity, For semantic embedding similarity; The dynamic adjacency matrix output by the temporal evolution layer is used to characterize the knowledge flow at each stage of standard implementation. The matrix elements represent the connection weights between nodes at each time step. The previous temporal hidden state is the node hidden state matrix of the previous temporal graph neural network, which encodes the knowledge flow information of the historical stage of standard development. The expression for the dynamic adjacency matrix is: ; in for The knowledge graph adjacency matrix of the implementation process of the time standard. This is the time-series weight matrix. This is the previous hidden state. The current input features are the semantic features of the document corresponding to the time series. This is the timing bias vector.

[0016] In this embodiment, the method for calculating process entropy and process entropy weighting coefficients includes: Extract standard implementation process data and calculate process entropy based on the total number of activity categories in each implementation stage. Then, perform normalization to calculate the process entropy weighting coefficient. The expression is as follows: ; ; in for Implementation phase process entropy The number of phases to be implemented. for Total number of activity categories during the implementation phase for Implementation phase The probability of occurrence of this type of activity for Entropy weighting coefficient during the implementation phase; In actual evaluation, the process entropy of each implementation stage is inversely proportional to the process entropy weight coefficient (the quality indicators of the low-entropy stage have small coefficients of variation and high data confidence, and should be given higher decision weights, that is, the smaller the process entropy, the more orderly the activities are, the more concentrated the activities in the implementation stage are, the obvious quality characteristics of this stage are, and the greater the contribution to the discrimination of the overall evaluation, the larger the corresponding process entropy weight coefficient).

[0017] In this embodiment, the method for calculating the degree of divergence among implementing entities includes: Based on the feedback documents from various implementing entities at different implementation stages, a fuzzy number of feedback from each implementing entity is constructed. The similarity of viewpoints among the implementing entities is calculated, and the relative consistency of viewpoints among the implementing entities is also calculated. The expression is as follows: ; ; in As the implementing entity and Similarity of viewpoints , As the implementing entity and implementing entities The number of ambiguous viewpoints , As the implementing entity and implementing entities exist Handling of opinions during the implementation phase As the implementing entity The relative consistency For the number of implementing entities, As the implementing entity Authority and weight This is an institutional qualification level coefficient, quantified based on the institution's qualification level and industry access qualifications. This is the business relevance coefficient for the organization, quantified based on the degree of matching between the organization's main business and the content of the standard implementation. This is the institution's historical assessment reliability coefficient, quantified based on the proportion of past opinions adopted or historical assessment deviations. Sensitivity to group consensus coefficient The entity that implemented the previous round of assessment The group consensus coefficient; The consensus coefficient of each implementing entity is calculated based on the relative consistency of their viewpoints. The divergence degree between the implementing entities is then calculated using the consensus coefficient and the similarity of their viewpoints. The expression is as follows: ; ; in To determine the degree of disagreement among the implementing entities, , As the implementing entity and implementing entities The group consensus coefficient As a relaxation factor, As the implementing entity Authority and weight; In actual assessments, the consensus coefficient of the implementing entities from the previous round of assessments is used. If an implementing entity's judgment consistently deviates from the consensus in a certain area, then... This measure corrects for the drift in authority weights caused by the drift in opinions from implementing entities. Implementing entities include various types of organizations such as manufacturing / service companies (e.g., OEMs, parts suppliers), testing and inspection agencies, certification bodies, and industry regulatory departments. The weights are comprehensively quantified based on the inherent attributes of the organizations, such as their qualification level, business relevance, and historical assessment reliability. The weights for feedback opinions from different types of implementing entities can be set according to their qualification level, business relevance to the standard, and historical assessment accuracy.

[0018] The institutional qualification level coefficient reflects the legal status and professional credibility of the implementing entity in the industry; the institutional business relevance coefficient reflects the strength of the relevance between its business scope and the implementation stage of the standard; and the institutional historical evaluation reliability coefficient reflects the accuracy and stability of the institution's opinions in previous standardization evaluation work.

[0019] When calculating the degree of disagreement among implementing entities, the relaxation factor is set to 0.5. The higher the degree of disagreement among implementing entities, the more serious the differences of opinion among implementing entities, and the higher the implementation risk.

[0020] In this embodiment, the method for obtaining the comprehensive calibration quality index includes: The width of the rolling evaluation window is determined based on the process entropy. The comprehensive calibration quality index is obtained by dynamically calibrating the comprehensive quality index of the standard implementation process data within the evaluation window, expressed as: ; ; in To evaluate the width of the scrolling evaluation window, , For minimum window width and maximum window width, To comprehensively calibrate the quality index, For smoothing coefficients, The comprehensive calibration quality index is based on the implementation time of the previous round of standards. For the comprehensive quality index, The standard implementation time; The comprehensive quality index is calculated by combining the degree of divergence among implementing entities, and its expression is: ; in For the comprehensive quality index, The time decay coefficient, For standard execution time, To evaluate the information entropy indicator, To determine the degree of disagreement among the implementing entities, , These are the process entropy and the process entropy weighting coefficient, respectively. The maximum process entropy; A quality warning will be issued when the overall calibration quality index falls below the overall quality index threshold. Calculate the cosine similarity between the mean of the fuzzy vector of the implementation entity feedback in the standard implementation process data after the evaluation window is updated and the mean of the fuzzy vector of the implementation entity feedback corresponding to the risk events in the historical risk case database. Take the response strategies corresponding to the top three risk events with the highest similarity as alternative response strategies. In the actual evaluation, the rolling evaluation window width of the previous round of evaluation is used to extract the standard implementation process data to calculate the process entropy, and the rolling evaluation window width is updated according to the process entropy. The standard implementation process data is then extracted again to calculate the comprehensive calibration quality index (when the comprehensive quality index of three consecutive windows shows a downward trend, the dynamic calibration protocol is triggered, and the smoothing coefficient is 0.7). When calculating the overall quality index, The balance coefficient is adaptively calculated using the entropy weight method, and the time decay coefficient is set to 0.15. When the comprehensive quality index is lower than the comprehensive quality index threshold (0.65), a quality warning is issued, and the cosine similarity between the mean of the fuzzy vector of the implementation subject feedback of the standard implementation process data after the evaluation window is updated and the mean of the fuzzy vector of the implementation subject feedback of the risk event in the historical risk case library is calculated. The response strategies corresponding to the top three risk events with the highest similarity are selected as alternative response strategies.

[0021] In this embodiment, the method for quality assessment and traceability includes: Based on the process graph nodes of the standard implementation process knowledge graph corresponding to each decision event, and according to the mapping rules, the decision events are mapped to the corresponding responsible entity nodes of each process graph node in the standard implementation process knowledge graph, thus forming an enhanced standard implementation process graph: Map the disagreements to nodes in the standard clause requirements diagram, and treat the disagreements as features of the standard clause requirements diagram nodes. The event decision is mapped to nodes in the opinion summary and processing graph, and the event decision, processing rules, and differences in the viewpoints of the implementing entities are used as the features of the nodes in the opinion summary and processing graph; the differences in the viewpoints of the implementing entities include the standard deviation and skewness of the opinion processing of the implementing entities in the corresponding standard implementation stage. Event quality is mapped to nodes in the review meeting minutes diagram, and event quality, review criteria, and process entropy are used as node features in the review meeting minutes diagram; the event quality is taken as the average score of the main implementing group at the corresponding standard implementation stage; A graph attention network is used to enhance the graph for root cause propagation during the standard implementation process, and the root cause score of each responsible entity's graph node is calculated. The expression is as follows: ; ; in for Standard Implementation Phase Root cause score of the responsible party node. This is the set of neighboring nodes of a node in the responsible entity graph, including the node in the process graph to which it belongs and other associated responsible entity nodes. Attention coefficient, reflecting the node right The contribution of quality impact, For nodes With nodes edge weights, The characteristic transformation matrix, Features of neighboring nodes , This is the process entropy adjustment coefficient. for Entropy of the standard implementation phase For the maximum process entropy, This is a learnable attention parameter vector. For nodes Features; The standard implementation stage where the process entropy is greater than the process entropy threshold is identified as the quality impact stage. The responsible entity node with the highest root cause score in the process diagram node corresponding to the quality impact stage is selected as the standard implementation impact factor to obtain the standard implementation impact factor set. Based on the standard implementation impact factor set, a response strategy is selected from the alternative response strategies. In actual evaluation, when calculating the importance of neighboring nodes, if the current stage has high process entropy (implementation chaos), the root cause score of all related edges in that stage is enhanced, making root cause analysis focus more on the interactive effects of the chaotic stage. Even if the responsible party in the high entropy stage has low absolute attention, it may still obtain a high root cause score due to process chaos, which is consistent with the technical assumption that "chaotic stages are prone to generating the root causes of quality problems". Through root cause analysis, the influencing factors of multiple different standard implementation stages can be identified, which together form a set of standard implementation influencing factors, and the appropriate response strategy can be selected from the alternative response strategies.

[0022] Taking the quality assessment of the implementation process of the standard "Limits and Measurement Methods for Pollutant Emissions from Light-Duty Vehicles (National X Stage VI)" in the automotive field as an example, a hybrid semantic extraction model is used to identify six process nodes: preparation for release, publicity and training, implementation, supervision and inspection, information feedback, and evaluation and improvement. Each process node corresponds to three responsible entity nodes: standard clause requirements (drafting group), opinion summary and processing (automotive industry association / car company representatives, drafting group), and review meeting minutes (standardization technical committee). Edge relationships include intra-process edges (standard clause requirements, opinion summary and processing, and review meeting minutes have mutual influence relationships) and cross-process temporal edges (adjacent processes have influence relationships between similar responsible entity nodes). The edge relationship weights of each responsible entity node are calculated to construct a knowledge graph of the standard implementation process. The standard implementation phase includes the release preparation phase (activity categories include document drafting, technical verification, departmental coordination, and implementation entity verification), the publicity and training phase (activity categories include textbook compilation, enterprise training, and institutional training), the implementation phase (activity categories include production line transformation, inventory digestion, new vehicle testing, compliance announcement, and after-sales monitoring), the supervision and inspection phase (activity categories include on-site road inspection, data comparison, violation penalties, and rectification review), and the evaluation and improvement phase (activity categories include effect evaluation, problem summary, and revision suggestions). Based on the probability of occurrence of activities in the implementation phase, the number of implementation phases, and the total number of activity categories in the implementation phase, the process entropy and process entropy weight coefficients are calculated to be 0.597 / 0.668 / 0.589 / 0.782 / 0.873 and 0.270 / 0.223 / 0.276 / 0.146 / 0.085, respectively. In the evaluation of the standard implementation process, taking the evaluation of implementing entities A and B as examples, implementing entity A is a vehicle manufacturer with a consensus coefficient of 0.8 in the previous round of evaluation and implementation difficulty scores of (8.5, 7.2, 9.1, 6.8, 7.5) for the five implementation stages. Implementing entity B is an inspection and testing institution with a consensus coefficient of 0.7 in the previous round of evaluation and implementation difficulty scores of (7.8, 7.5, 8.5, 7, 7.2) for the five implementation stages. The authority weights of implementing entities A and B are calculated to be 0.3 / 0.3, and the similarity of viewpoints is 0.58. Similarly, the authority weights and mutual similarity of viewpoints of all implementing entities are calculated. The group consensus coefficients of implementing entities A and B are 0.4 and 0.4, respectively. Similarly, the group consensus coefficients of all implementing entities (the group consensus coefficients of implementing entities A and B are 0.4 and 0.4, respectively) are calculated. The divergence degree of implementing entities is calculated to be 0.42. Calculate the entropy of the weighted process Based on the maximum process entropy of 0.873 and the minimum / maximum window width of 3 / 10, the rolling evaluation window width is calculated to be 8.3 (rounded to 8 time units, calculated by month), and the corresponding comprehensive quality index is calculated to be 0.592 (the information entropy of the evaluation index is taken as 0.2). At this time, the comprehensive quality index of three consecutive windows shows a downward trend (0.65→0.63→0.592). Based on the comprehensive calibration quality index of 0.65 from the previous standard implementation time, dynamic calibration is performed to obtain a comprehensive calibration quality index of 0.609. At this time, the comprehensive calibration quality index of 0.609 is less than the comprehensive quality index threshold of 0.65, and a quality warning is issued. Based on the mean vector of all implementing entities' viewpoints (8.15, 7.35, 8.80, 6.90, 7.35), risk events in the historical risk case database are matched, and the corresponding risk event response strategies are used as alternative response strategies. Based on the standard implementation status corresponding to each decision event, match the process graph nodes of the standard implementation process knowledge graph, and map the decision events to the corresponding responsible entity nodes of each process graph node in the standard implementation process knowledge graph according to the mapping rules, forming an enhanced standard implementation process graph and updating the graph node feature vectors; Taking the implementation phase as an example, the process entropy of 0.589 in the implementation phase is greater than the process entropy threshold of 0.55. This phase is considered the quality impact phase. Attention coefficients, edge weights, and feature vectors are obtained through graph attention network learning. The process entropy adjustment coefficients λλ and ββ are set to 0.2 / 0.3. Root cause analysis is performed to calculate the root cause scores of the three responsible entity graph nodes in the implementation phase: 0.78 / 0.649 / 0.657. The standard requirements of the implementation phase are determined as a factor influencing standard implementation (i.e., the actual driving pollutant emission (RDE) limit (4.5 × 10⁻⁶)). 12 The "particles / km" standard was affected by atmospheric pressure at high altitudes (>2000m) during actual implementation, leading to a high number of abnormal detection data during the implementation phase. This made it difficult and less effective to implement according to the standard requirements. The following adjustment strategy was selected from the alternative response strategies: modify the "RDE limit correction coefficient for high altitude areas", supplement the altitude gradient correction model, and similarly calculate the root cause scores of the responsible entity graph nodes in all standard implementation phases for quality assessment and tracing.

[0023] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A standard implementation process quality assessment method based on a quantitative model, characterized in that, Includes the following steps: S1. Extract the standard implementation materials and input them into the hybrid semantic extraction model for semantic extraction. Construct a knowledge graph of the standard implementation process based on the semantic extraction results. S2. Extract standard implementation process data, calculate process entropy and process entropy weight coefficient according to the implementation stage, construct viewpoint fuzzy number according to the implementation subject, and calculate the degree of divergence of the implementation subject; S3. Determine the width of the rolling evaluation window based on the process entropy, calculate the comprehensive quality index of the standard implementation process data within the evaluation window in combination with the divergence of the implementing entities, perform dynamic calibration to obtain the comprehensive calibration quality index and issue quality warnings. S4. Map the set of decision events in the standard implementation process data to the standard implementation process knowledge graph, perform root cause analysis through graph neural network, and combine process entropy for quality assessment and tracing. The standard implementation materials include standard clause requirements, summary and processing of comments, review meeting minutes, etc. The data for the standard implementation process includes data on release preparation, dissemination and training, implementation and execution, supervision and inspection, and evaluation and improvement. The decision-making events include disagreements, event decisions, and event quality.

2. The standard implementation process quality assessment method based on a quantitative model according to claim 1, characterized in that, The method for constructing a knowledge graph of the standard implementation process includes: The standard implementation materials are extracted using a hybrid semantic extraction model for semantic extraction. The hybrid semantic extraction model includes a process element identification layer, a relational reasoning layer, and a temporal evolution layer; The process element identification layer uses the BERT-BiLSTM-CRF model to extract process entities; the process entities include process nodes and responsible entities; the process nodes correspond to the standard implementation stages; and the responsible entities correspond to the standard implementation materials. The relation reasoning layer employs a document-level relation extraction network based on an attention mechanism to establish the association edges of entities in each process and calculate the edge weights. The temporal evolution layer constructs a dynamic adjacency matrix for the evolution of the standard implementation process knowledge graph based on the transfer function and process time nodes; The process entities extracted from the process element identification layer are used as graph nodes, and a standard implementation process knowledge graph is constructed by combining the associated edges and edge weights of each process entity.

3. The standard implementation process quality assessment method based on a quantitative model according to claim 1, characterized in that, The method for calculating process entropy and process entropy weighting coefficients includes: Extract standard implementation process data and calculate process entropy based on the total number of activity categories in each implementation stage. Then, perform normalization to calculate the process entropy weighting coefficient. The expression is as follows: ； ； in for Implementation phase process entropy The number of phases to be implemented. for Total number of activity categories during the implementation phase for Implementation phase The probability of occurrence of this type of activity for Entropy weighting coefficient during the implementation phase.

4. The standard implementation process quality assessment method based on a quantitative model according to claim 1, characterized in that, The method for calculating the degree of divergence among implementing entities includes: Based on the feedback documents from various implementing entities at different implementation stages, a fuzzy number of feedback from each implementing entity is constructed. The similarity of viewpoints among the implementing entities is calculated, and the relative consistency of viewpoints among the implementing entities is also calculated. The expression is as follows: ； ； ； in As the implementing entity and Similarity of viewpoints , As the implementing entity and implementing entities The number of ambiguous viewpoints , As the implementing entity and implementing entities exist Handling of opinions during the implementation phase As the implementing entity The relative consistency For the number of implementing entities, As the implementing entity Authority and weight This is an institutional qualification level coefficient, quantified based on the institution's qualification level and industry access qualifications. This is the business relevance coefficient for the organization, quantified based on the degree of matching between the organization's main business and the content of the standard implementation. This is the institution's historical assessment reliability coefficient, quantified based on the proportion of past opinions adopted or historical assessment deviations. Sensitivity to group consensus coefficient The entity that implemented the previous round of assessment The group consensus coefficient; The consensus coefficient of each implementing entity is calculated based on the relative consistency of their viewpoints. The divergence degree between the implementing entities is then calculated using the consensus coefficient and the similarity of their viewpoints. The expression is as follows: ； ； in To determine the degree of disagreement among the implementing entities, , As the implementing entity and implementing entities The group consensus coefficient As a relaxation factor, As the implementing entity Its authority and weight.

5. The standard implementation process quality assessment method based on a quantitative model according to claim 1, characterized in that, The method for obtaining the comprehensive calibration quality index includes: The width of the rolling evaluation window is determined based on the process entropy. The comprehensive calibration quality index is obtained by dynamically calibrating the comprehensive quality index of the standard implementation process data within the evaluation window, expressed as: ； ； in To scroll the evaluation window width, , For minimum window width and maximum window width, To comprehensively calibrate the quality index, For smoothing coefficients, The comprehensive calibration quality index is based on the implementation time of the previous round of standards. For the comprehensive quality index, The standard implementation time; The comprehensive quality index is calculated by combining the degree of divergence among implementing entities, and its expression is: ； in For the comprehensive quality index, The time decay coefficient, For standard execution time, To evaluate the information entropy indicator, To determine the degree of disagreement among the implementing entities, , These are the process entropy and the process entropy weighting coefficient, respectively. The maximum process entropy; A quality warning will be issued when the overall calibration quality index falls below the overall quality index threshold. The cosine similarity between the mean of the fuzzy vector of feedback from the implementing entity in the standard implementation process data after the evaluation window is updated and the mean of the fuzzy vector of feedback from the implementing entity corresponding to the risk events in the historical risk case database is calculated. The response strategies corresponding to the top three risk events with the highest similarity are selected as alternative response strategies.

6. The standard implementation process quality assessment method based on a quantitative model according to claim 1, characterized in that, The method for quality assessment and traceability includes: Based on the process graph nodes of the standard implementation process knowledge graph corresponding to each decision event, and according to the mapping rules, the decision events are mapped to the corresponding responsible entity nodes of each process graph node in the standard implementation process knowledge graph, thus forming an enhanced standard implementation process graph: The disagreements are mapped to nodes in the standard clause requirements diagram, and the disagreements and the implementation standards are used as node features in the standard clause requirements diagram. The event decision is mapped to nodes in the opinion summary and processing graph, and the event decision, processing rules, and differences in the viewpoints of the implementing entities are used as the features of the nodes in the opinion summary and processing graph; the differences in the viewpoints of the implementing entities include the standard deviation and skewness of the opinion processing of the implementing entities in the corresponding standard implementation stage. Event quality is mapped to nodes in the review meeting minutes diagram, and event quality, review criteria, and process entropy are used as node features in the review meeting minutes diagram; the event quality is taken as the average score of the implementing entity at the corresponding standard implementation stage; Graph attention networks are used to enhance the graph for root cause propagation during standard implementation, and the root cause score of each responsible entity's graph node is calculated. The expression is as follows: ； ； in for Standard Implementation Phase Root cause score of the responsible party node. This is the set of neighboring nodes of a node in the responsible entity graph, including the node in the process graph to which it belongs and other associated responsible entity nodes. Attention coefficient, reflecting the node right The contribution of quality impact, For nodes With nodes edge weights, The characteristic transformation matrix, Features of neighboring nodes , This is the process entropy adjustment coefficient. for Entropy of the standard implementation phase For the maximum process entropy, This is a learnable attention parameter vector. For nodes Features; The standard implementation stage where the process entropy is greater than the process entropy threshold is identified as the quality impact stage. The responsible entity node with the highest root cause score in the process diagram node corresponding to the quality impact stage is selected as the standard implementation impact factor to obtain the standard implementation impact factor set. Based on the standard implementation impact factor set, a response strategy is selected from the alternative response strategies.

Images