Intelligent manufacturing level evaluation method and system based on twin entropy weight and graph network

Abstract

The invention discloses an intelligent manufacturing level evaluation method and system based on a twin entropy weight and a graph network. The method comprises the following steps of: 1, weight dynamic self-adaptive determination: calculating an entropy change rate of each index in a sliding time window based on a real-time data stream of a digital twin model to obtain a real-time basic weight; virtual disturbance is injected into the model and simulated, the influence degree of each index system is obtained to serve as a weight adjustment coefficient, and the two factors are combined to obtain a final dynamic weight. Step 2, index score calculation based on a graph network: constructing an index relation graph according to an index association relation, taking index real-time data and time sequence features as node initial features, and inputting the node initial features into an attention space-time graph convolutional network model; and aggregating spatial neighbor information through a graph convolution layer, weighting time sequence features through a time sequence attention layer, and outputting a final index score fused with a space-time dependency relationship. According to the invention, automatic real-time dynamic adjustment of the evaluation weight is realized, and the intelligent degree of intelligent manufacturing level evaluation is improved.

Description

Technical Field

[0001] This invention relates to a method and system for evaluating the level of intelligent manufacturing based on twin entropy weights and graph networks, belonging to the field of industrial intelligence and data analysis technology. Background Technology

[0002] In advancing intelligent manufacturing, a scientific and dynamic assessment of an enterprise's current level is fundamental to formulating development strategies. Digital twin technology provides an ideal data environment for this. However, existing evaluation methods in digital twins still face two technical challenges: 1. Static and subjective weighting: Traditional evaluation models typically assign weights once by experts (e.g., the AHP method) or calculate them offline based on historical data (e.g., the entropy weight method). This static weighting cannot adapt to dynamically changing production conditions in the factory (e.g., changes in order type or equipment status), resulting in insufficient relevance and accuracy of the evaluation results. Reliance on expert experience also introduces non-technical subjective factors. 2. Isolated and one-sided indicator calculation: Existing methods, when calculating scores for individual indicators, typically only consider their corresponding data flow, ignoring the spatiotemporal correlations that are prevalent between indicators in a complex system like a factory. For example, minor quality fluctuations in upstream materials (spatial correlation) and equipment parameter adjustments made five minutes prior (temporal correlation) can significantly impact the current product qualification rate. Traditional methods fail to capture these complex dependencies, leading to one-sided evaluation scores and a lack of in-depth insight. Therefore, there is an urgent need for a new generation of intelligent evaluation technology that can achieve dynamic adaptive weighting and deeply explore the spatiotemporal correlation between indicators. Summary of the Invention

[0003] The purpose of this invention is to provide a method and system for evaluating the level of intelligent manufacturing based on Siamese entropy weights and graph networks. The evaluation method of this invention achieves automatic, real-time dynamic adjustment of evaluation weights and calculates index scores that integrate spatiotemporal dependencies through an attention-based spatiotemporal graph convolutional network model, thereby improving the intelligence level of the evaluation method.

[0004] The technical solution of this invention: a method for evaluating the level of intelligent manufacturing based on twin entropy weights and graph networks, comprising the following steps:

[0005] Step 1: Dynamic Adaptive Determination of Weights: Based on the real-time data stream from the digital twin model, calculate the entropy change rate of each indicator in the evaluation index system within the sliding time window to generate real-time basic weights; by injecting virtual perturbations into the digital twin model and simulating their impact, calculate the system influence of each indicator to generate weight adjustment coefficients; combine the real-time basic weights with the weight adjustment coefficients to obtain the final dynamic weights of each indicator.

[0006] Step 2: Calculation of index scores based on graph networks: Construct an index relationship graph based on the correlation between each index, and use the real-time data and time series features of each index as the initial features of the corresponding nodes in the graph; input the index relationship graph and its initial node features into the attention spatiotemporal graph convolutional network model; aggregate the spatial neighbor information of the nodes through the graph convolutional layer of the model, and weight the time series features of the nodes through the temporal attention layer to output the final score of each index that integrates the spatiotemporal dependencies.

[0007] The aforementioned intelligent manufacturing level evaluation method based on twin entropy weights and graph networks, wherein the entropy change rate... The calculation formula is:

[0008]

[0009] in, As an indicator At the present moment The data sequence within the sliding window, This is the function for calculating information entropy. This represents the time window step size.

[0010] The aforementioned intelligent manufacturing level evaluation method based on twin entropy weights and graph networks, the system influence degree The calculation formula is:

[0011] ;

[0012] in, To the indicators The virtual perturbation amount injected with the associated parameters. This represents the change in the global evaluation function caused by the disturbance.

[0013] The aforementioned intelligent manufacturing level evaluation method based on twin entropy weights and graph networks constructs the indicator relationship graph based on a pre-constructed indicator knowledge graph. The indicator knowledge graph is generated by mapping a general evaluation indicator system onto a specific system topology graph of an enterprise, so that the nodes in the graph simultaneously carry the dual attributes of enterprise entities and evaluation indicators.

[0014] The aforementioned intelligent manufacturing level evaluation method based on Siamese entropy weights and graph networks uses the following node feature update rule in the graph convolutional layer of the attention-based spatiotemporal graph convolutional network model:

[0015]

[0016] in, For nodes In the Layer feature representation, For nodes The set of neighboring nodes, For a trainable weight matrix, The normalization constant is It is a non-linear activation function.

[0017] The aforementioned intelligent manufacturing level evaluation method based on twin entropy weights and graph networks uses a time-series attention layer to assign different attention weights to the historical time-series data of nodes through a self-attention mechanism in order to capture historical moments that have a key impact on the current evaluation.

[0018] A system for implementing the aforementioned intelligent manufacturing level evaluation method based on twin entropy weights and graph networks includes:

[0019] The digital twin module is used to build and run digital twin models corresponding to physical entities;

[0020] An adaptive evaluation module is communicatively connected to the digital twin module; the adaptive evaluation module includes:

[0021] The indicator knowledge graph database is used to store the indicator relationship graph generated by the fusion of the evaluation indicator system and the enterprise system topology graph;

[0022] The weight adaptive module is used to receive the real-time data stream from the digital twin model and execute step one of claim 1 to output the final dynamic weights of each indicator.

[0023] The scoring module has the attention spatiotemporal graph convolutional network model built in, which is used to receive the real-time data stream, the topology of the indicator relationship graph from the indicator knowledge graph database, and the dynamic weights from the weight adaptation module, and output the indicator score.

[0024] The aforementioned system, wherein the weight adaptive module includes:

[0025] A twin entropy weight calculation unit is used to calculate the entropy change rate and generate the real-time basic weights;

[0026] The disturbance analysis unit is used to perform the virtual disturbance injection, simulation and system impact calculation, and generate the weight adjustment coefficients.

[0027] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, when the processor executes the computer program, it implements the steps of the aforementioned method.

[0028] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the aforementioned method.

[0029] Compared with the prior art, the present invention has the following beneficial effects:

[0030] 1. The method proposed in this invention enables weights to be generated entirely based on real-time data from a digital twin environment. By calculating the entropy change rate of the data stream to perceive unstable elements in the system, and by injecting virtual perturbations to quantify the global impact on the system, online and automatic adjustment of weights is achieved. This not only completely eliminates subjective factors but also allows the evaluation focus to be precisely on the most critical indicators that have the greatest impact on the system, significantly improving the accuracy and relevance of the evaluation.

[0031] 2. This invention constructs the evaluation index system as an index relationship graph and utilizes an attention-based spatiotemporal graph convolutional network for inference. Graph convolution operations enable the score calculation of each index to aggregate information from its spatially related nodes (such as upstream and downstream processes, and mutually influencing equipment parameters); the temporal attention mechanism automatically captures and weights historical moments in the time series that have a key impact on the current state. This mechanism makes the evaluation result of this invention not only a score, but also possesses powerful diagnostic capabilities.

[0032] 3. The attention spatiotemporal graph convolutional network model of the present invention, with its powerful temporal pattern learning ability, can keenly capture key instantaneous events and quantify their impact, while also identifying and synchronously reflecting slow degradation trends, thereby providing early warning.

[0033] 4. This invention not only provides a static score, but also enables virtual scenario simulation (such as simulating emergency orders) through digital twins, and observes the dynamic response pattern of the system evaluation weights, thereby assessing the overall agility and collaborative capabilities of the enterprise. Through virtual scenario stress testing, it allows for in-depth diagnosis of the dynamic collaborative capabilities and structural defects of the enterprise system, achieving a leap from static scoring to dynamic capability assessment. Attached Figure Description

[0034] Figure 1 This is an overall flowchart of the intelligent manufacturing level evaluation method based on twin entropy weights and graph networks of the present invention.

[0035] Figure 2 A comparison of system topology diagrams from different enterprises, including Figure 2 (A) is a topology diagram of a highly interconnected system of manufacturer A; Figure 2 (B) is a sparse and isolated system topology diagram of manufacturer B, which reflects the phenomenon of node data silos caused by the absence of key systems.

[0036] Figure 3 This is a comparison chart of the weighted dynamic responses of the two manufacturers in a virtual emergency order scenario.

[0037] Figure 4This is a diagram illustrating the correlation between indicators and the diagnosis of root causes.

[0038] Figure 5 This is a comparison chart of real-time data and evaluation scores. Figure 5 (A) Displays a real-time data stream of product pass rates, including "instantaneous anomalies" and "slow degradation zones"; Figure 5 (B) Comparison of evaluation score curves between traditional methods and the method of the present invention. Detailed Implementation

[0039] The present invention will be further described below with reference to the accompanying drawings and embodiments, but this should not be construed as limiting the present invention.

[0040] Example: A method for evaluating the level of intelligent manufacturing based on twin entropy weights and graph networks, such as... Figure 1 As shown, it includes the following steps:

[0041] Step 1: Dynamic adaptive determination of weights, which includes:

[0042] 1.1 For each indicator in the evaluation index system, based on the real-time data stream from the digital twin model, calculate the entropy change rate of each indicator in the evaluation index system within the sliding time window, and generate the real-time basic weight of each indicator according to the entropy change rate.

[0043] In this step, the more unstable the data stream corresponding to an indicator, the more information it contains that needs to be monitored. This invention uses the rate of entropy change to quantify this instability. For any indicator in the indicator system... Its basic weights are determined by the following formula:

[0044]

[0045] in, The normalization coefficient is... It is an indicator At any moment Data sequence within the sliding time window Information entropy, calculated by the formula is: ; This represents the probability of the data appearing within the window. The derivative of entropy with respect to time reflects the degree of change in information entropy, and its calculation formula is:

[0046]

[0047] In the formula, As an indicator At the present moment The data sequence within the sliding window, This is the function for calculating information entropy. This represents the time window step size.

[0048] This step ensures that the system's attention is automatically focused on the most unstable element at present.

[0049] 1.2 Periodically inject a virtual perturbation signal into the parameters associated with a specific indicator in the digital twin model, obtain the impact of the perturbation on the preset global evaluation function through high-speed simulation, calculate the system influence degree of the specific indicator, and generate weight adjustment coefficients accordingly.

[0050] In this step, the importance of an indicator is also reflected in its impact on the entire system. This impact is quantified through proactive virtual experiments conducted within the digital twin. The system periodically performs the following operations:

[0051] Injecting perturbations: In the digital twin model, select an indicator Associated parameters (Such as equipment speed, planned time), apply a very small, risk-free virtual change to it. .

[0052] Observational Response: Through high-speed simulation, the global evaluation function caused by the disturbance is observed. (This could be a simplified total score model) Change ;

[0053] Calculate the impact: Indicator Systemic impact Defined as the ratio of response to disturbance, i.e., the system's sensitivity to this metric:

[0054]

[0055] Weighting adjustment coefficient yes The normalization function. This step ensures that critical nodes that affect the entire system are given higher weights.

[0056] 1.3 Multiply the real-time basic weights by the weight adjustment coefficients to obtain the final dynamic weights of each indicator. :

[0057]

[0058] The final weight is the product of the base weight and the adjustment coefficient, and is normalized to ensure that the sum of all weights is 1.

[0059] Step 2: Calculate the metric score based on the graph network. This step includes:

[0060] 2.1 Based on the physical or logical relationships between the indicators in the aforementioned evaluation index system, construct an index relationship diagram. , where the set of nodes Representing each evaluation indicator, edge set The combination represents the correlation between indicators;

[0061] In this step, the system automatically generates a diagram based on the factory's physical topology (such as equipment connection relationships) and process logic flow. Each node in the diagram This corresponds to a secondary indicator. If the indicator... and indicators If there is a direct physical or logical influence between the nodes (such as upstream and downstream relationships), then at the node... and There is an edge between them .

[0062] (2.2) The real-time data streams and their time-series characteristics corresponding to each indicator are used as the initial feature vectors of the corresponding nodes in the graph. ;

[0063] In this step, each node initial feature vector Composed of its corresponding twin data stream, it is a multi-dimensional vector, for example:

[0064]

[0065] It includes the current value, historical value, first derivative (velocity), and second derivative (acceleration), comprehensively describing the dynamic characteristics of the index.

[0066] 2.3 Input the index relationship graph with initial feature vector into a pre-trained attention spatiotemporal graph convolutional network (AST-GCN) model. The model aggregates spatial neighbor node information through graph convolutional layers and weights the temporal series features of nodes through temporal attention layers, outputting an accurate score for each node that integrates spatiotemporal dependencies.

[0067] The Attention Spatiotemporal Graph Convolutional Network (AST-GCN) model is the computational core of this invention, which alternately contains graph convolutional layers and temporal attention layers.

[0068] Spatial Aggregation (SAG) layer: Each graph convolution operation fuses the feature vectors of each node with the feature information of all its first-order neighbor nodes. After convolutional layers, each node can perceive the distance to itself in the graph. By analyzing the states of all nodes within a step range, spatial correlation is modeled.

[0069] Temporal Attention: Before feature aggregation, this layer processes the historical sequence of nodes. It processes data by automatically assigning higher weights to historical moments that are most important for the current evaluation through a self-attention mechanism, thus modeling the time-criticality of events.

[0070] Output: The entire graph network ultimately outputs a scalar value for each indicator node, which is the final accurate score of the indicator after incorporating complex spatiotemporal dependencies.

[0071] The invention will be further illustrated below with specific application examples.

[0072] Step 1: System initialization and indicator knowledge graph construction;

[0073] This step aims to build a computable knowledge base that integrates a general evaluation model with enterprise-specific architecture for subsequent adaptive evaluation algorithms. This step mainly includes the following three automated processes:

[0074] Evaluation index system loading: First, a pre-set evaluation index system model as the evaluation benchmark is loaded. In a preferred embodiment, the model includes six primary dimensions (such as intelligent production, intelligent management, etc.) and twenty-eight secondary indicators.

[0075] Enterprise System Topology Construction: By scanning and parsing the enterprise's system integration architecture diagram and process flow documents, an enterprise system topology diagram that objectively reflects the current level of integration between the enterprise's information technology and physical equipment is automatically constructed. The graph includes key system entities (such as ERP, MES, etc.) as nodes, and the data flow between them as edges.

[0076] Appendix Figure 2 It demonstrates the structural differences in system topology diagrams built for different enterprises. Figure 2 A represents a highly interconnected network topology for manufacturer A, while... Figure 2 B is a sparse and isolated topology of manufacturer B.

[0077] Generation and mapping of indicator knowledge graphs:

[0078] This is one of the key steps in this invention. The system merges the results of the first two steps to generate a final indicator knowledge graph for use by subsequent algorithms. .

[0079] Mapping process: Based on predefined rules, the system "maps" and "mounts" each secondary indicator in the evaluation indicator system model to the enterprise system topology. The most relevant entity nodes. For example, the secondary indicator "Manufacturing Execution System Application" will be mapped to the "MES" entity node in the topology diagram.

[0080] Final Result: Generated Indicator Knowledge Graph Its topology is similar to The same applies, but now each node in the diagram simultaneously carries the dual attributes of "enterprise entity" and "evaluation indicator".

[0081] The "indicator knowledge graph" generated in this step provides the most critical and information-rich input for the AST-GCN model in a subsequent score calculation module (220), enabling the abstract evaluation indicators to be intelligently bound to the specific enterprise architecture.

[0082] Appendix Figure 2 The structural differences between the system topology diagrams automatically generated by two different companies are shown. Figure 2 A shows the topology diagram generated for manufacturer A. Due to the deep integration between its various information systems (PLM, ERP, MES, etc.) and physical devices (intelligent equipment, etc.), the entity nodes in the diagram are tightly connected by multiple data flow edges, forming a highly interconnected network topology. Figure 2 B shows the topology diagram generated for manufacturer B. Due to the lack of many of its critical information systems (such as MES and PLM), most entity nodes in the diagram become data silos, network connections are sparse, and the main data flow exists only between the ERP and a few devices.

[0083] Step 2: Dynamic adaptive determination of weights (comparison between companies A and B);

[0084] This step details how the weighted adaptive module (210) uses the method of the present invention to perform a differentiated in-depth diagnosis of the dynamic response capabilities of two companies when the same virtual scene pressure is applied.

[0085] For Manufacturer A (Agile Response Diagnostic Process):

[0086] Scenario: Simulate an urgent, highly customized order placement event in the digital twin model of manufacturer A.

[0087] TEDA algorithm diagnostic process:

[0088] Real-time base weights Generation: Monitoring revealed that frequent adjustments to production plans due to urgent orders caused significant fluctuations in twin data streams (such as work order adjustment frequency) associated with multiple secondary indicators belonging to the "Smart Manufacturing" dimension (e.g., "Manufacturing Execution System Application," "Production Scheduling Management," etc.). According to the formula, the overall entropy change rate of this dimension... The increase is significant, thus automatically boosting the real-time base weight of the "intelligent production" dimension and shifting the evaluation focus primarily to the production execution phase.

[0089] .

[0090] Weighting adjustment coefficient Generation: Building upon the evaluation focus already on the "intelligent manufacturing" dimension, further identification of key control points within this dimension is undertaken. For example, a small perturbation of -0.5% is applied to parameters related to the secondary indicator "flexible production capacity," which also belongs to the "intelligent manufacturing" dimension (such as "virtual changeover time for bottleneck processes"). High-speed simulations revealed that, under the current urgent order conditions, this slight changeover delay will affect the global evaluation function. A sharp drop of 1.8% ( ), calculate the systemic impact of this indicator according to the formula. Extremely high.

[0091] ;

[0092] Final dynamic weights Synthesis: Based on the formula, by combining high "entropy change rate" and "system impact", the weight of the first-level dimension "intelligent production" was significantly and dynamically increased in a concentrated manner.

[0093] .

[0094] For Manufacturer B (a static and rigid diagnostic process):

[0095] Scenario: Simulate placing the exact same emergency order in the digital twin model of manufacturer B.

[0096] TEDA algorithm diagnostic process:

[0097] Real-time base weights Generation: Due to its indicator relationship diagram (as attached) Figure 2 Due to the sparsity shown in Figure B, order information cannot be effectively transmitted between systems. Monitoring revealed that each data stream operates in isolation, without generating systemic interconnected fluctuations; therefore, the entropy change rate of all indicators... All are at low levels, and the basic weights have not changed significantly.

[0098] Weighting adjustment coefficient Generation: When performing virtual disturbance analysis, due to the lack of effective correlation paths in the graph, any local disturbance cannot propagate to the global extent, resulting in a decrease in the systemic impact of all indicators. All approach zero.

[0099] Final dynamic weights Synthesis: Ultimately, its weights remain almost unchanged throughout the entire event.

[0100] Figure 3 The above dual diagnostic process is summarized visually. For Manufacturer A, as shown by the thick black line in the figure, its "Intelligent Production" weight (thick dashed line) showed a rapid and concentrated significant increase after the event was triggered. This is a quantitative manifestation of the invention's method, which diagnoses its healthy and dynamic system coordination capabilities through two consecutive steps of analysis (first identifying the fluctuation dimension, then finding the sensitive points within that dimension). Meanwhile, the weights of other dimensions, such as "Intelligent Management" (thick dotted line), decreased slightly or remained stable. In contrast, for Manufacturer B, as shown by the thin gray line in the figure, its corresponding "Intelligent Production" (gray dashed line) and "Intelligent Management" (gray dotted line) weights remained almost level throughout the process. This is direct evidence that the invention diagnosed the company as lacking system linkage and unable to respond effectively and dynamically to external pressures. Figure 3 The results show that this invention can not only assess an enterprise's static foundation, but also accurately diagnose an enterprise's true system agility, dynamic collaboration capabilities, and deep-seated structural defects by conducting virtual, high-intensity scenario simulations in a digital twin environment and analyzing the dynamic response patterns of its weight curves.

[0101] Step 3: Calculate the index score based on the AST-GCN model;

[0102] This step details how to use the AST-GCN model to achieve deep diagnosis and accurate calculation.

[0103] In-depth diagnosis of spatial correlation:

[0104] Diagnostic Process: For Manufacturer A, diagnosed as having "agile dynamic response capabilities" in the second step, the algorithm of this invention can perform a deeper root cause diagnosis. In this example, when the model detects an abnormal decline in the score of "production line efficiency," a key sub-indicator belonging to the "intelligent manufacturing" dimension, the AST-GCN model utilizes the enterprise system topology diagram generated for it in the first step (as shown in the attached diagram). Figure 2 As shown in Figure A), and combined with the evaluation metrics mapped onto it, it automatically performs backpropagation tracing. It aggregates information from all upstream neighbor nodes pointing to "production line efficiency" (such as "equipment A status", "raw material quality", etc.) and quantifies the "contribution" of each node to the current anomaly.

[0105] Diagnostic results presented: Figure 4 The results of the above diagnostic process are shown. Each box in the figure represents a node in the indicator relationship diagram, and the directed arrows between nodes represent the causal relationships between indicators. Figure 4 The demonstrated technical effect is that when an abnormal decline in the scores of downstream indicators such as "product qualification rate" or "order delivery rate" is detected, traditional isolated calculation methods cannot find the cause. However, the Attention Spatiotemporal Graph Convolutional Network (AST-GCN) model of this invention can utilize the topological structure of a pre-constructed indicator relationship graph to perform backpropagation tracing. The AST-GCN model analyzes upstream level by level along the influence path (represented by thick dashed lines), ultimately locking the "root cause" of the problem onto the node of "device A status" (shown by a thick circle). This figure demonstrates that this invention can deeply mine the spatial correlation between indicators, achieving intelligent diagnosis from "discovering the phenomenon" to "locating the root cause," providing a precise basis for subsequent optimization decisions.

[0106] For Manufacturer B: Since it was diagnosed as "static rigidity" in the second step, its sparse index relationship graph (attached) Figure 2 B) This type of cross-node correlation analysis cannot be supported, therefore this in-depth diagnostic cannot be performed.

[0107] Precise capture of time-critical moments:

[0108] Diagnostic Process: The temporal attention mechanism of the AST-GCN model enables in-depth analysis of the time-series data of the "root cause" node (i.e., "equipment A state") to uncover the transient causes leading to the anomaly. Example: In manufacturer A's factory, sensor data for a critical piece of equipment showed an abnormal current spike 10 minutes ago, but subsequently returned to normal. When processing the time-series features of the "equipment A state" node, the model's self-attention mechanism, through learned patterns, can identify this brief "current spike" as a key historical event affecting subsequent production quality and efficiency. Therefore, when calculating the comprehensive evaluation score at the current moment, the model automatically assigns a very high attention weight to this anomalous data point from 10 minutes ago.

[0109] Diagnostic results presented:

[0110] Figure 5 The advantages of the method of this invention compared with those of traditional methods are shown. Figure 5 A raw data stream of "product pass rate" from a digital twin. This data stream has a "transient anomaly" (a brief numerical spike) at a certain moment, and then enters a "slow degradation zone" (the data shows a continuous, slow downward trend) in subsequent periods. Figure 5 B compares the score curves of the two methods. The traditional method (dotted lines + circle markers) uses a hard threshold for judgment. The method of this invention (thick solid lines + square markers) uses the AST-GCN model for calculation. Figure 5This invention demonstrates that it can accurately capture and quantify the impact of key instantaneous events in the time dimension, as detailed below:

[0111] For "transient anomalies": the score curve of traditional methods remains unchanged because the outlier does not fall below its set threshold. However, the score curve of the method in this invention shows a brief but significant drop, achieving a risk warning. This is because the model's temporal attention mechanism captures this key anomalous data point and considers it a potential indication of risk. This demonstrates that the invention possesses a high sensitivity to critical events.

[0112] For the "slow degradation zone": traditional methods maintain high scores until the data finally falls below the threshold, resulting in a sharp drop and a severely delayed response. In contrast, the score curve of the method in this invention declines smoothly and synchronously from the very beginning of the degradation trend. This demonstrates that the invention can effectively identify and quantify dynamic trends, possesses strong early warning capabilities, and can provide decision support for predictive maintenance.

[0113] Final score output:

[0114] Manufacturer A's "production line efficiency" score is a comprehensive reasoning result that integrates information from multiple spatially related nodes and key time points. It is not only accurate but also has in-depth diagnostic capabilities.

[0115] Manufacturer B's score is a relatively one-sided and lagging value because it is unable to conduct effective spatiotemporal correlation analysis.

[0116] Step 4: Presentation and summary of evaluation results;

[0117] After comprehensive calculations through the above steps, the final evaluation result is output. The advancement of this invention lies not only in obtaining the final score, but also in endowing the score with profound and traceable diagnostic connotations.

[0118] For manufacturer A:

[0119] Diagnostic conclusion:

[0120] Dynamic response capability diagnosis: as attached Figure 3 As shown, the TEDA algorithm of the weight adaptive module (210) successfully diagnosed that when faced with a virtual emergency order scenario, the company's evaluation weights can make agile adaptive adjustments focused on the "smart production" dimension.

[0121] System correlation and risk prediction capability diagnosis: as attached Figure 4 and attached Figure 5As shown, the AST-GCN model successfully diagnosed that there are effective spatiotemporal correlations between the various system entities of the enterprise, enabling it to perform in-depth root cause tracing and provide real-time warnings of system risks that may be caused by "transient anomalies".

[0122] Quantitative Results and Optimization Space Analysis: The system integrates the above dynamic diagnostic conclusions and determines that Manufacturer A possesses strong dynamic adaptability and risk foresight capabilities. However, [attached text] Figure 3 The dynamic weight change curves also reveal its optimization potential: the response curve exhibits a certain ramp-up time, indicating that the agility of its resource mobilization and process coordination still needs improvement. Therefore, the final comprehensive score of 71.5 points is an objective quantification of its current dynamic operational capabilities, accurately positioning it at the "intermediate" level where further optimization is needed.

[0123] For Manufacturer B:

[0124] Diagnostic conclusion: When evaluating Manufacturer B, the algorithm of this invention yielded a dual diagnosis of dynamic and static defects.

[0125] Dynamic response defect diagnosis: as attached Figure 3 As shown by the weight curve, when faced with the same external pressure event, its evaluation weight remains unchanged. Based on this, the weight adaptive module (210) diagnoses that the enterprise system lacks dynamic response capability and resource coordination mechanism.

[0126] Static correlation defect diagnosis: its sparse enterprise system topology diagram (attached) Figure 2 B) This directly causes the score calculation module (220) to be unable to establish a valid spatiotemporal correlation. Figure 4 and attached Figure 5 The sluggish response pattern of traditional Chinese methods is precisely a reflection of the inevitable result of the company's inability to conduct correlation analysis and risk foresight.

[0127] Quantitative Results: Based on the aforementioned dual-deficiency diagnosis, the system determined that the enterprise is currently operating in an isolated state, unresponsive to external factors and unconnected internally. Therefore, the final comprehensive score of 46.8 points is an objective quantification of its basic module's independent operation level, accurately positioning it at the "entry-level" level.

[0128] This embodiment fully demonstrates that the present invention, through its original algorithm, achieves a fundamental shift from static scoring to dynamic diagnosis. It not only assesses how well a company is performing, but also deeply diagnoses the bottlenecks in the company's system operation and the structural defects present. Its depth of evaluation and level of intelligence surpass existing technologies.

[0129] Example 2: This example provides a system for implementing the method described in Example 1, such as... Figure 2As shown, it includes:

[0130] The digital twin module uses real-time multi-source data from physical entities / factories to build and run digital twin models corresponding to the physical entities;

[0131] An adaptive evaluation module is communicatively connected to the digital twin module; the adaptive evaluation module includes:

[0132] The indicator knowledge graph database is used to store the indicator relationship graph generated by the fusion of the evaluation indicator system and the enterprise system topology graph. The indicator knowledge graph database is generated by merging the general "evaluation indicator system" and the enterprise-specific "system topology graph", providing other modules with necessary prior knowledge, such as indicator definitions, correlation relationships and the topological structure of the indicator relationship graph G.

[0133] The weight adaptive module is used to receive the real-time data stream from the digital twin model, execute step one in Example 1, and output the final dynamic weights of each indicator; the weight adaptive module includes:

[0134] A twin entropy weight calculation unit is used to calculate the entropy change rate and generate the real-time basic weights;

[0135] The disturbance analysis unit is used to perform the virtual disturbance injection, simulation and system impact calculation, and generate the weight adjustment coefficients.

[0136] The twin entropy weight calculation unit and the perturbation analysis unit work together. The twin entropy weight calculation unit directly receives the real-time data stream from the digital twin model and, in conjunction with the indicator definitions from the indicator knowledge graph database, generates real-time basic weights, which are then sent to the perturbation analysis unit. The perturbation analysis unit also calls upon the digital twin model and the indicator knowledge graph database, ultimately outputting dynamic weights to the score calculation module.

[0137] The scoring module has the attention spatiotemporal graph convolutional network model built in, which is used to receive the real-time data stream, the topology of the indicator relationship graph from the indicator knowledge graph database, and the dynamic weights from the weight adaptation module, and output the indicator score.

[0138] Ultimately, the evaluation results are sent to the comprehensive evaluation and visualization module at the application layer, and can be fed back to the physical entity / factory, forming a complete intelligent closed loop.

[0139] Example 3: This example provides an electronic device that implements the method of Example 1. The electronic device is in the form of a general-purpose computing device. The components of the electronic device may include, but are not limited to: at least one processing unit, at least one storage unit, a bus connecting different system components (including the storage unit and the processing unit), and a display unit.

[0140] The storage unit stores program code, which can be executed by the processing unit to perform the steps described in the method section of Embodiment 1 above, according to various exemplary embodiments of the present invention.

[0141] The storage unit may include readable media in the form of volatile storage units, such as random access memory (RAM) and / or cache storage units, and may further include read-only memory (ROM).

[0142] The storage unit may also include a program / utility having a set (at least one) of program modules, including but not limited to: an operating system, one or more application programs, other program modules, and program data, each or some combination of these examples may include an implementation of a network environment.

[0143] A bus can represent one or more of several types of bus structures, including a memory cell bus or memory cell controller, a peripheral bus, a graphics acceleration port, a processing unit, or a local bus that uses any of the various bus structures.

[0144] The electronic device can also communicate with one or more external devices (e.g., keyboards, pointing devices, Bluetooth devices, etc.), one or more devices that enable a user to interact with the electronic device, and / or any device that enables the electronic device to communicate with one or more other computing devices (e.g., routers, modems, etc.). This communication can be performed via input / output (I / O) interfaces. Furthermore, the electronic device can communicate with one or more networks (e.g., local area networks (LANs), wide area networks (WANs), and / or public networks, such as the Internet) via a network adapter. The network adapter communicates with other modules of the electronic device via a bus. Further, other hardware and / or software modules can be used in conjunction with the electronic device, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0145] From the above description of the embodiments, those skilled in the art will readily understand that the embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solutions according to the embodiments of this disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, external hard drive, etc.) or on a network, including several instructions to cause a computing device (such as a personal computer, server, terminal device, or network device, etc.) to execute the methods according to the embodiments of this disclosure.

[0146] In exemplary embodiments of this disclosure, a computer-readable storage medium is also provided, on which a program product capable of implementing the methods described above is stored. In some possible embodiments, various aspects of the invention may also be implemented as a program product comprising program code that, when the program product is run on a terminal device, causes the terminal device to perform the steps of the various exemplary embodiments of the invention described in the "Exemplary Methods" section above.

Claims

1. A method for evaluating the level of intelligent manufacturing based on twin entropy weights and graph networks, characterized by: Includes the following steps: Step 1: Dynamic Adaptive Determination of Weights: Based on the real-time data stream from the digital twin model, calculate the entropy change rate of each indicator in the evaluation index system within the sliding time window to generate real-time basic weights; by injecting virtual perturbations into the digital twin model and simulating their impact, calculate the system influence of each indicator to generate weight adjustment coefficients; combine the real-time basic weights with the weight adjustment coefficients to obtain the final dynamic weights of each indicator. Step 2, Calculation steps of indicator scores based on graph network: Construct an indicator relationship graph based on the correlation between indicators, and use the real-time data and time series features of each indicator as the initial features of the corresponding nodes in the graph; The index relationship graph and its initial node features are input into the attention spatiotemporal graph convolutional network model; the spatial neighbor information of the nodes is aggregated through the graph convolutional layer of the model, and the temporal series features of the nodes are weighted through the temporal attention layer to output the final score of each index that integrates the spatiotemporal dependencies.

2. The intelligent manufacturing level evaluation method based on twin entropy weights and graph networks according to claim 1, characterized in that: The entropy change rate The calculation formula is: in, As an indicator At the present moment The data sequence within the sliding window, This is the function for calculating information entropy. This represents the time window step size.

3. The intelligent manufacturing level evaluation method based on twin entropy weights and graph networks according to claim 1, characterized in that: The system influence The calculation formula is: ； in, To the indicators The virtual perturbation injected with the associated parameters. This represents the change in the global evaluation function caused by the disturbance.

4. The intelligent manufacturing level evaluation method based on twin entropy weights and graph networks according to claim 1, characterized in that: The basis for constructing the indicator relationship graph is a pre-constructed indicator knowledge graph; the indicator knowledge graph is generated by mapping a general evaluation indicator system onto a specific system topology graph of an enterprise, so that the nodes in the graph simultaneously carry the dual attributes of enterprise entities and evaluation indicators.

5. The intelligent manufacturing level evaluation method based on twin entropy weights and graph networks according to claim 1, characterized in that: The node feature update rule for the graph convolutional layer in the attention-spatiotemporal graph convolutional network model is as follows: in, For nodes In the Layer feature representation, For nodes The set of neighboring nodes, For a trainable weight matrix, The normalization constant is It is a non-linear activation function.

6. The intelligent manufacturing level evaluation method based on twin entropy weights and graph networks according to claim 5, characterized in that: The temporal attention layer assigns different attention weights to the historical time series data of nodes through a self-attention mechanism in order to capture historical moments that have a critical impact on the current evaluation.

7. A system for implementing the intelligent manufacturing level evaluation method based on twin entropy weights and graph networks as described in any one of claims 1 to 6, characterized in that, include: The digital twin module is used to build and run digital twin models corresponding to physical entities; An adaptive evaluation module is communicatively connected to the digital twin module; The adaptive evaluation module includes: The indicator knowledge graph database is used to store the indicator relationship graph generated by the fusion of the evaluation indicator system and the enterprise system topology graph; The weight adaptive module is used to receive the real-time data stream from the digital twin model and execute step one of claim 1 to output the final dynamic weights of each indicator. The scoring module has the attention spatiotemporal graph convolutional network model built in, which is used to receive the real-time data stream, the topology of the indicator relationship graph from the indicator knowledge graph database, and the dynamic weights from the weight adaptation module, and output the indicator score.

8. The system according to claim 7, characterized in that, The weight adaptive module includes: A twin entropy weight calculation unit is used to calculate the entropy change rate and generate the real-time basic weights; The disturbance analysis unit is used to perform the virtual disturbance injection, simulation and system impact calculation, and generate the weight adjustment coefficients.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 8.

Images