A digital main line processing method, medium and system based on an industrial data atlas

Abstract

The application provides a kind of digital main line processing method, medium and system based on industrial data atlas, belongs to industrial data atlas technical field, the application is realized unified access of cross-system data by constructing multi-source data acquisition adapter, carries out cleaning standardization processing to heterogeneous data, constructs integrated data set, abstracts data entity as node, constructs industrial data atlas by abstracting cross-system association as edge, adopts graph index optimization strategy and layered storage architecture to improve cross-system query performance, realizes the dynamic adjustment of atlas structure using atlas self-adaptive evolution model, realizes cross-system depth correlation analysis function by graph traversal algorithm, establishes access control and data security mechanism to complete system protection, solve the technical problem that industrial data processing system is difficult to realize cross-system depth correlation analysis.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of industrial data graph, and in particular, relates to a digital main line processing method, medium and system based on an industrial data graph. BACKGROUND

[0002] In the field of modern industrial intelligent manufacturing, enterprises collect and manage production data through enterprise resource planning systems, manufacturing execution systems, customer relationship management systems and other heterogeneous systems. Traditional data integration technologies mainly use data warehouses, ETL tools, data buses and other solutions to realize unified storage and management of multi-source data. These methods play an important role in production planning management, quality control, supply chain coordination, equipment maintenance and other business scenarios by establishing a unified data model and standardized data interface. They can effectively integrate structured data from different systems and provide basic query and analysis functions. However, traditional data integration solutions have obvious shortcomings when dealing with cross-system data correlation analysis. Since they use a table-based relational data model, they can only perform limited data correlation through predefined foreign key relationships. When faced with complex multi-level correlation relationships, a large number of table join operations are required, resulting in low query efficiency and the risk of Cartesian product problems. They cannot flexibly respond to dynamically changing correlation analysis requirements. In current industrial data analysis applications, due to the structural constraints of traditional relational databases and the limitations of query language expressiveness, enterprises have difficulty implementing deep cross-system data correlation analysis. They cannot effectively mine complex business relationships hidden in multi-source data, which seriously hinders the implementation effect of data-driven decision-making. That is, there is a technical problem in the prior art that industrial data processing systems cannot implement deep cross-system correlation analysis. SUMMARY

[0003] Therefore, the application provides a digital main line processing method, medium and system based on an industrial data graph, which can solve the technical problem that industrial data processing systems cannot implement deep cross-system correlation analysis in the prior art.

[0004] The application is implemented as follows: the first aspect of the application provides a digital main line processing method based on an industrial data graph, an industrial data relationship network is constructed by abstracting data entities as nodes and abstracting association relationships as edges, a graph traversal algorithm is used to realize cross-system deep association analysis, and the following steps are included: a multi-source data collection adapter is constructed, database tables in an enterprise resource planning system, a manufacturing execution system and a customer relationship management system are collected in real time, a data source identification mapping table is established, and the unified access of heterogeneous data sources is completed; the collected multi-source heterogeneous data is cleaned and standardized, an integrated data set is constructed, a data quality evaluation vector is established, and data preprocessing and quality control are completed; an industrial data graph is constructed based on the integrated data set, data entities are abstracted as nodes and association relationships are abstracted as edges, and the attribute information of the nodes and edges is stored by using a graph database; a graph index optimization strategy and a hierarchical storage architecture are constructed, multi-dimensional indexes are established, a hot data caching mechanism and a query result pre-computation system are used; a graph adaptive evolution model is used to monitor business changes in real time, when the Euclidean distance of the evolution feature vector and the benchmark evolution vector in the multi-dimensional space is greater than 0.85, a pattern matching algorithm is started to automatically adjust the graph structure; data tracing and exploration service functions are realized, intelligent data tracing functions are realized based on a graph traversal algorithm, when the query hysteresis effect error is greater than 0.15, a hysteresis compensation algorithm is started, and if the hysteresis effect error is in the range of [0, 0.05], a standard query processing mode is used; a role-based access control and data security guarantee mechanism is established.

[0005] Among them, the multi-source data collection adapter adopts a connection mode based on protocol adaptation, corresponding data access drivers are established for different database types, and multiple data access protocols such as structured query language, open database connection, and application program interface are supported.

[0006] Among them, the data source identification mapping table records the unique identifier, database connection parameters, table structure information and field mapping relationship of each data source, records the access path and update frequency of each system data source, and is used to establish a unified data access interface. The unique identifier is derived from the system code of the enterprise resource planning system, the manufacturing execution system and the customer relationship management system.

[0007] Among them, the data quality evaluation vector includes five dimensions of data integrity coefficient, accuracy coefficient, consistency coefficient, timeliness coefficient and validity coefficient, the data quality evaluation vector is constructed by quality detection and scoring of the original data, and is used to monitor data integrity.

[0008] Among them, the nodes in the industrial data graph represent data entities, and the edges represent the association relationships between entities. Each node and edge contains rich attribute information. The node attributes include entity identification, entity type, business attribute and timestamp information, and the edge attributes include relationship type, relationship strength, relationship direction and relationship weight information.

[0009] The graph index optimization strategy includes three types of node attribute index, relationship index and composite index, the node attribute index is established based on the key attributes of the node, the relationship index is established based on the type and weight of the edge, and the composite index establishes a joint index combining multiple attributes.

[0010] The hierarchical storage architecture stores hot data in the memory layer, warm data in the solid state disk layer, and cold data in the mechanical hard disk layer, and dynamically adjusts the storage hierarchy according to the data access frequency.

[0011] The query performance superposition optimization vector contains five indexes of index hit rate, cache hit rate, query response time, concurrent processing capability and resource utilization, which is used to describe the effect evaluation of query optimization strategy, evaluate the optimization effect, and complete the optimization configuration of query performance.

[0012] The graph adaptive evolution model adopts a multi-scale representation integration mechanism based on hierarchical feature fusion and a sequence interdependence modeling algorithm based on recurrent attention, when detecting changes in business requirements or new system access, the model automatically analyzes data pattern changes, recommends appropriate graph structure adjustment schemes, and ensures the traceability of the evolution process through version control mechanism.

[0013] The evolution feature vector contains five dimensions of node number change rate, edge number change rate, structure density change rate, connectivity change rate and hierarchical depth change rate, and the benchmark evolution vector contains five benchmark values of standard node density, standard edge density, standard structure complexity, standard connectivity and standard hierarchical number.

[0014] The data traceability function is realized based on graph traversal algorithm, after the user inputs the traceability target, the system performs depth-first search or breadth-first search along the association relationship path in the graph, generates complete data traceability link, and adopts path pruning and heuristic search strategy to improve search efficiency in the traceability process.

[0015] The hysteresis compensation algorithm optimizes repeated queries by establishing query history records and result cache mechanism, when the hysteresis effect error is controlled within 0.05, the standard query processing mode is adopted, when the error exceeds the threshold, the compensation algorithm is started to correct the error, eliminate the influence of round trip error, and complete the construction of data traceability link.

[0016] The role-based access control mechanism assigns different data access permissions to different user roles, encrypts the data transmission process, uses the Secure Sockets Layer protocol to prevent data from being stolen or tampered with during transmission, sets up data backup strategies, and completes the security protection and data protection of the digital mainline system.

[0017] The structure of the atlas adaptive evolution model comprises an input layer receiving an evolution feature vector and business change information, a feature extraction layer extracting structural features through a convolutional neural network, an attention layer establishing correlation between features by using a multi-head attention mechanism, a decision layer outputting evolution strategy suggestions through a fully connected neural network, and an output layer generating an atlas structure adjustment scheme.

[0018] The evolution feature fusion function is used for integrating multi-dimensional evolution feature information, the input comprises a structural change feature vector, a business demand change vector and a system performance index vector, and the output is a comprehensive evolution feature vector.

[0019] The query optimization weight function is used for calculating the weight distribution of different optimization strategies, the input comprises a query type identifier, historical performance data and a system resource state, and the output is an optimization strategy weight vector.

[0020] The second aspect of the present application provides a computer readable storage medium, the computer readable storage medium has program instructions stored therein, the program instructions are used for executing the above-mentioned industrial data atlas-based digital main line processing method when running in a computer.

[0021] The third aspect of the present application provides an industrial data atlas-based digital main line processing system, which comprises the above-mentioned computer readable storage medium, the system is any one of a computer, a server and a single-chip microcomputer, the computer readable storage medium is arranged in the system, and the system is provided with a microprocessor for executing the program instructions stored in the computer readable storage medium.

[0022] The present application converts the traditional table format data organization mode into a node-edge graph structure representation by constructing an industrial data relationship network based on a graph structure, so that the data entities across systems can form a complete correlation network through the connection of edges, and the limitations of traditional relational databases in expressing complex correlation relationships are completely broken through. The present application uses a graph traversal algorithm to replace the traditional SQL association query, realizes multi-level association analysis through depth-first search and breadth-first search strategies, avoids the multi-table connection overhead in the traditional method, significantly improves the efficiency and accuracy of cross-system data analysis, and makes the association query that originally needs complex SQL statements become an intuitive graph path search process. The present application ensures that the system can efficiently process large-scale cross-system association analysis tasks through graph index optimization and adaptive evolution mechanism, and has the ability to dynamically adjust the atlas structure, which can adapt to the changing business association requirements, and fundamentally solves the technical limitations of traditional data processing systems in cross-system deep association analysis. In summary, the present application solves the technical problem that the industrial data processing system in the background art is difficult to realize cross-system deep association analysis. BRIEF DESCRIPTION OF DRAWINGS

[0023] Figure 1 is a flow chart of the method of the present application.

[0024] Figure 2 is a graph of the trend of the change in the evolution of the graph structure in Example 2.

[0025] Figure 3 is a comparison graph of the waiting time distribution of different query types in Example 2. DETAILED DESCRIPTION

[0026] In order to make the purposes, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application.

[0027] As shown in Figure 1 is a flow chart of a digital main line processing method based on an industrial data graph provided by the first aspect of the present application, and the method comprises the following steps:

[0028] S01, a multi-source data acquisition adapter is constructed to collect databases in tables in enterprise resource planning systems, manufacturing execution systems and customer relationship management systems in real time, a data source identification mapping table is established to record the access paths and update frequencies of each system data source, and the unified access of heterogeneous data sources is completed;

[0029] S02, the collected multi-source heterogeneous data is cleaned and standardized to eliminate data format differences and noise, an integrated data set in a unified data format is constructed, a data quality evaluation vector is established to monitor data integrity, data preprocessing and quality control are completed;

[0030] S03, an industrial data graph is constructed based on the integrated data set, data entities are abstracted as nodes and association relationships are abstracted as edges, attribute information of the nodes and edges is stored by using a graph database, a complete industrial data relationship network is formed, and the initialization construction of the graph model is completed;

[0031] S04, a graph index optimization strategy and a hierarchical storage architecture are constructed, multi-dimensional indexes are established for the node attributes in the graph database, a hot data caching mechanism and a query result pre-computation system are adopted, a query performance superposition optimization vector is constructed to evaluate the optimization effect, and the optimization configuration of the query performance is completed;

[0032] S05, a graph adaptive evolution model is used to monitor business changes in real time, when the Euclidean distance of the evolution feature vector and the benchmark evolution vector in the multi-dimensional space is >0.85, a pattern matching algorithm is started to automatically adjust the graph structure, and the adaptive evolution of the graph structure is completed;

[0033] S06, realizing data tracing and exploring service function, realizing intelligent data tracing function based on graph traversal algorithm, when query hysteresis effect error > 0.15, starting hysteresis compensation algorithm to eliminate the influence of return error, if hysteresis effect error ∈ [0, 0.05], adopting standard query processing mode, completing the construction of data tracing link;

[0034] S07, establishing role-based access control and data security guarantee mechanism, encrypting data transmission, setting data backup strategy, completing the security protection and data protection of digital mainline system.

[0035] The multi-source data collection adapter adopts a connection mode based on protocol adaptation, establishes corresponding data access driver programs for different database types, and supports multiple data access protocols such as structured query language, open database connection, and application program interface. The data source identifier mapping table records the unique identifier, database connection parameters, table structure information and field mapping relationship of each data source, and is used to establish a unified data access interface. The unique identifier of the data source identifier mapping table is derived from the system code of the enterprise resource planning system, the manufacturing execution system and the customer relationship management system. The database connection parameters are used for subsequent connection establishment of the data collection adapter.

[0036] The data quality evaluation vector includes five dimensions of data integrity coefficient, accuracy coefficient, consistency coefficient, timeliness coefficient and effectiveness coefficient. The data quality evaluation vector is constructed by quality detection and scoring of the original data. The data integrity coefficient represents the completeness of the data record, the accuracy coefficient represents the correctness of the data value, the consistency coefficient represents the uniformity of the data between different data sources, the timeliness coefficient represents the timeliness of the data update, and the effectiveness coefficient represents the degree of data compliance with business rules. The data integrity coefficient is derived from the completeness statistics of the integrated data set. The accuracy coefficient is used for subsequent accuracy verification of the industrial data graph node attribute.

[0037] The nodes in the industrial data graph represent data entities, and the edges represent the association relationship between entities. Each node and edge contains rich attribute information. The node attributes include entity identifier, entity type, business attribute and timestamp information, and the edge attributes include relationship type, relationship strength, relationship direction and relationship weight information. The graph database uses an attribute graph model to store data, supports graph query and traversal operations. The entity identifier is derived from the primary key field in the integrated data set, and the relationship weight is used for subsequent path calculation of the graph traversal algorithm.

[0038] The graph index optimization strategy includes three types of node attribute index, relationship index and composite index. The node attribute index is established based on the key attributes of the nodes, the relationship index is established based on the type and weight of the edges, and the composite index establishes a joint index combining multiple attributes. The hierarchical storage architecture stores hot data in the memory layer, warm data in the solid state disk layer, and cold data in the mechanical hard disk layer, and dynamically adjusts the storage hierarchy according to the data access frequency. The node attribute index is established based on the business attributes of the nodes in the industrial data graph, and the relationship index is used for relationship lookup optimization of subsequent graph traversal algorithms.

[0039] The query performance superposition optimization vector is used to describe the effect evaluation of the query optimization strategy, and includes five indexes of index hit rate, cache hit rate, query response time, concurrent processing capacity and resource utilization rate. The index hit rate represents the proportion of query operations using indexes, the cache hit rate represents the proportion of query results hitting in the cache, the query response time represents the average execution time of query operations, the concurrent processing capacity represents the ability of the system to process query requests at the same time, and the resource utilization rate represents the efficiency of system resources. The index hit rate is derived from the execution statistics of the graph index optimization strategy, and the query response time is used for threshold judgment of the subsequent hysteresis compensation algorithm.

[0040] The graph adaptive evolution model adopts a multi-scale representation integration mechanism based on hierarchical feature fusion and a sequence interdependence modeling algorithm based on recurrent attention, realizing the evolution of the graph structure. When detecting changes in business requirements or new system access, the model automatically analyzes data pattern changes, recommends appropriate graph structure adjustment schemes, and ensures the retroactivity of the evolution process through a version control mechanism. The input of the graph adaptive evolution model comes from business change monitoring data, and the graph structure adjustment scheme is used for subsequent structure update of the industrial data graph.

[0041] The evolution blocking risk matrix is used to evaluate the potential risks in the graph structure evolution process, and the matrix dimensions include structure complexity, data consistency, performance impact, compatibility and rollback difficulty. The structure complexity represents the complexity of the new graph structure, the data consistency represents the degree of data consistency in the evolution process, the performance impact represents the degree of influence of evolution on system performance, the compatibility represents the compatibility of the new structure with the existing system, and the rollback difficulty represents the difficulty of recovering to the original state when evolution fails. The structure complexity is derived from graph structure analysis and calculation, and the rollback difficulty is used for risk assessment of subsequent evolution decisions.

[0042] The query latency distribution vector records the latency distribution characteristics of different types of query operations, including the average latency of simple node queries, relationship traversal queries, multi-hop path queries, aggregation statistical queries and joint queries. The query latency distribution vector is used to analyze the query performance bottleneck of the system and guide the formulation and adjustment of query optimization strategies. The simple node query latency is derived from the query log statistics of the graph database, and the query latency distribution vector is used for subsequent query performance superposition optimization vector calculation.

[0043] The performance gain evaluation matrix is used to quantify the performance improvement effect brought by the graph index optimization strategy. The matrix row represents different query types, the list represents different optimization strategies, and the matrix element represents the performance gain coefficient of the query type under the optimization strategy. The performance gain coefficient is calculated by the ratio of query response time before and after optimization, and is used to evaluate the effectiveness of the optimization strategy. The performance gain coefficient is derived from the comparison of query response time, and the performance gain evaluation matrix is used for subsequent selection and adjustment of query optimization strategies.

[0044] The system load fluctuation monitoring vector records the fluctuation characteristics of system resource usage in real time, including processor usage fluctuation, memory usage fluctuation, disk input / output fluctuation, network bandwidth fluctuation and concurrent connection number fluctuation. The system load fluctuation monitoring vector is used to monitor the system running state, and when the fluctuation amplitude exceeds the preset threshold, the resource adjustment and load balancing mechanism is triggered. The processor usage fluctuation is derived from system monitoring data collection, and the system load fluctuation monitoring vector is used for subsequent dynamic allocation of system resources.

[0045] The data tracing function is realized based on graph traversal algorithm. After the user inputs the tracing target, the system performs depth-first search or breadth-first search along the association relationship path in the graph to generate complete data tracing link. Path pruning and heuristic search strategy are used in the tracing process to improve search efficiency, while recording the weight and confidence information of the tracing path. The tracing target is derived from user input request, and the data tracing link is used for subsequent business analysis and problem positioning.

[0046] The hysteresis compensation algorithm is used to eliminate the influence of hysteresis error in the graph query process. By establishing query history record and result cache mechanism, repeated queries are optimized. When the hysteresis effect error is controlled within 0.05, the standard query processing mode is adopted, and when the error exceeds the threshold, the compensation algorithm is started to correct the error. The query history record is derived from the operation log of the graph database, and the hysteresis compensation algorithm is used for subsequent accuracy guarantee of query results.

[0047] The role-based access control mechanism assigns different data access permissions to different user roles, ensuring the security and confidentiality of data. The data transmission process is encrypted, using the Secure Sockets Layer protocol to prevent data from being stolen or tampered with during transmission. The data backup strategy regularly backs up the industrial data graph and related data to prevent data loss and disaster recovery. The user roles are derived from the enterprise organizational structure definition, and the data access permissions are used for subsequent data query and operation permission control.

[0048] The evolution feature vector is used to describe the feature information of the graph structure change, including five dimensions of node number change rate, edge number change rate, structure density change rate, connectivity change rate and hierarchical depth change rate. The node number change rate represents the degree of change of the number of nodes in the graph, the edge number change rate represents the degree of change of the number of edges in the graph, the structure density change rate represents the change of the tightness of the graph structure, the connectivity change rate represents the degree of change of the connectivity of the graph, and the hierarchical depth change rate represents the degree of change of the depth of the hierarchical structure of the graph. The node number change rate is derived from the structure statistics of the industrial data graph, and the evolution feature vector is used as input for the subsequent adaptive evolution model of the graph.

[0049] The reference evolution vector is used to provide a reference standard for the evolution of the graph structure, including five reference values of standard node density, standard edge density, standard structure complexity, standard connectivity and standard hierarchical number. The standard node density represents the node distribution density under normal business conditions, the standard edge density represents the relationship distribution density under normal business conditions, the standard structure complexity represents the standard complexity of the graph structure, the standard connectivity represents the standard connectivity level of the graph, and the standard hierarchical number represents the standard hierarchical depth of the graph. The standard node density is derived from statistical analysis of the graph structure in the historical stable period, and the reference evolution vector is used for distance calculation of the subsequent evolution feature vector.

[0050] The structure of the graph adaptive evolution model is that the input layer receives the evolution feature vector and the business change information, the feature extraction layer extracts the structure features through the convolutional neural network, the attention layer establishes the correlation between the features using the multi-head attention mechanism, the decision layer outputs the evolution strategy suggestion through the fully connected neural network, and the output layer generates the graph structure adjustment scheme.

[0051] The training data set establishment step of the graph adaptive evolution model includes collecting historical graph evolution cases, labeling the differences between the graph structures before and after evolution, constructing feature vectors to describe the evolution mode, establishing evolution effect evaluation standards, forming a training sample set containing input features and expected output, and dividing the training set, validation set and test set according to the ratio of 8:1:1.

[0052] The atlas adaptive evolution model training step includes initializing model parameters, using a stochastic gradient descent algorithm for model training, updating network weights through a backpropagation algorithm, evaluating model performance using a validation set, preventing overfitting using an early stopping mechanism, finally verifying the generalization ability of the model on the test set, and saving the trained model parameters for actual deployment.

[0053] The multi-scale representation integration mechanism based on hierarchical feature fusion enhances the model's perception of atlas structure changes by weighting and fusing features from different network levels, combining low-level detail features and high-level semantic features, and improving the accuracy and completeness of the evolution strategy recommendation. The mechanism enables the model to capture both local structural changes and global pattern evolution, providing more accurate analysis and prediction for atlas evolution scenarios.

[0054] The inter-sequence dependence modeling algorithm based on recurrent attention accurately captures long-term dependencies and evolution patterns in the evolution process by establishing a correspondence between evolution sequences through recurrent attention calculations when processing the temporal characteristics of atlas evolution, significantly improving the model's prediction accuracy of evolution trends and its ability to identify abnormal evolution patterns. The algorithm enables the model to understand the internal laws of the evolution process, providing decision support for adaptive evolution.

[0055] The specific implementation of the above steps is described in detail below.

[0056] The specific implementation of step S01 is to realize unified access of heterogeneous data sources by constructing a multi-source data collection adapter. First, a data access driver based on structured query language is established for the enterprise resource planning system. The driver parses the database table structure of the enterprise resource planning system and extracts the metadata information of the table fields. Then, a data access driver based on open database connectivity is established for the manufacturing execution system. The communication with the manufacturing execution system database is realized by configuring the data source name and the connection string. Next, a data access driver based on the application program interface is established for the customer relationship management system. The data interface provided by the customer relationship management system is called to obtain business data. After that, a data source identifier mapping table is created to record the unique identifiers of each system. The unique identifier uses system coding as the primary key to ensure the uniqueness of the mapping relationship. At the same time, the database connection parameters including the host address and the port number, the database name and the user credentials, and the table structure information including the table name, the field name, the data type, the primary key and foreign key relationship, and other attributes are recorded. The field mapping relationship is established to associate the fields with the same business meaning in different systems. Finally, the data update frequency parameters are set. The update frequency of the manufacturing execution system data is set to 1 second to 5 seconds, the update frequency of the enterprise resource planning system data is set to 10 seconds to 30 seconds, and the update frequency of the customer relationship management system data is set to 30 seconds to 60 seconds. The purpose of this step is to solve the protocol differences and inconsistent access methods between different data sources. The transparent access of the data source is realized through the protocol adaptation technology, and a unified data acquisition interface is provided for subsequent data processing.

[0057] The specific implementation of step S02 is to clean and standardize the collected multi-source heterogeneous data. First, the original data is subjected to missing value detection, and the missing proportion of each field is counted. When the missing proportion is less than 5%, the mean filling or mode filling method is used for data completion. When the missing proportion is between 5% and 20%, the prediction filling method based on the regression model is used. When the missing proportion is greater than 20%, the field is marked as low-quality data and its weight is reduced in subsequent processing. Then, the data format is standardized and converted. The date and time format is uniformly converted to the international standardization organization 8601 format. The numerical unit is normalized. The text encoding is uniformly converted to the universal character set conversion format 8-bit encoding. Then, the abnormal value detection is performed. The box plot method is used to identify abnormal data points that are 1.5 times the interquartile range outside the upper and lower quartiles. Whether the identified abnormal values are reasonable extreme values is judged according to the business rules. If it is data noise, it is removed or corrected. Then, the data quality evaluation vector is constructed. The data integrity coefficient is calculated as the number of complete records divided by the total number of records. The accuracy coefficient is calculated by comparing with the reference data set to obtain the correct data proportion. The consistency coefficient is calculated by detecting the data consistency of the same entity between different data sources. The timeliness coefficient is calculated as the real-time data volume divided by the total data volume. The validity coefficient is calculated by verifying whether the data meets the predefined business rules. Finally, the integrated data set in a unified data format is formed. This step uses statistical analysis methods and data mining techniques to improve the quality of the original data, ensuring the reliability of the data basis for subsequent graph construction. When the coefficients of each dimension of the data quality evaluation vector are greater than 0.8, it is considered that the data quality meets the available standard.

[0058] The specific implementation of step S03 is to build an industrial data graph based on the integrated data set. First, data entity recognition is performed, all data records in the integrated data set are traversed, and unique entity identifiers are extracted according to the primary key field. Business objects such as product information, work order information, equipment information, and personnel information are abstracted as graph nodes. An entity type attribute is assigned to each node to distinguish between different categories of business entities. Business attribute information including product number, specification model, production batch, and other field values is recorded. A timestamp attribute is added to record the creation time and update time of the node. Then, the association relationship is extracted. The foreign key constraint relationship between data tables is analyzed to identify the direct association between entities. The implicit association relationship is mined through business rules, such as the supply relationship between products and suppliers and the use relationship between work orders and equipment. The identified association relationship is abstracted as the edge of the graph. The relationship type attribute such as the inclusion relationship, dependency relationship, and association relationship is set for each edge. The relationship strength is calculated by counting the frequency of association. The relationship direction is set to represent the starting node and target node of the association. The relationship weight is calculated using the normalized association strength value. Then, a graph database is selected as the storage engine, and an attribute graph model is used for data organization. The node and edge and their attribute information are written into the graph database. Node index and edge index are established to support efficient graph query operations. Graph query language is used to define graph traversal rules and query patterns. Finally, a complete industrial data relationship network is formed. The purpose of this step is to convert traditional relational data into graph structure data. The complex association relationship between entities is naturally expressed using the topology of the graph. The graph computing foundation is provided for subsequent data tracing and relationship analysis. The uniqueness of the entity identifier is ensured by calculating the digest value of the primary key field using a hash algorithm. The numerical range of the relationship weight is set to 0.1 to 1.0.

[0059] The specific implementation of step S04 is to construct a graph index optimization strategy and a hierarchical storage architecture to improve query performance. First, a multi-dimensional index is established for the nodes in the graph database. A classification index is established for the entity type attribute of the nodes. An exact match index is established for the key fields in the business attribute, such as product number and work order number. A range query index is established for the timestamp attribute to support time period queries. A B-tree index structure is used to realize fast positioning of node attributes. Then, a relationship index is established for the edges. A classification index is established according to the relationship type of the edges to facilitate filtering by relationship type. An ordering index is established according to the relationship weight to support ordering queries by weight. A composite index is established to combine relationship type and weight to establish a joint index to improve complex query efficiency. Next, a hot data caching mechanism is established. Historical query logs are counted to identify frequently accessed nodes and relationships. The top 10% of data by access frequency is marked as hot data and loaded into the memory cache. The least recently used algorithm is used to manage cache space, and the least frequently accessed data is evicted when the cache is full. A query result pre-computation system is implemented. Common query patterns such as fixed path traversal queries are pre-computed and cached. When the same or similar queries are received, the pre-computed results are returned directly. A hierarchical storage architecture is constructed to divide data into hot data, warm data, and cold data based on access temperature. Hot data is stored in the memory layer to achieve sub-millisecond access. Warm data is stored in the solid state disk layer to achieve millisecond access. Cold data is stored in the mechanical hard disk layer to achieve second-level access. The storage level is dynamically adjusted based on data access frequency. When the warm data access frequency exceeds the threshold, it is promoted to the hot data layer. When the hot data access frequency is below the threshold, it is downgraded to the warm data layer. The threshold is set to hot data access frequency greater than 100 times per second, warm data access frequency between 10 and 100 times per second, and cold data access frequency less than 10 times per second. Finally, a query performance superposition optimization vector is constructed for effect evaluation. The index hit rate is calculated as the number of queries using the index divided by the total number of queries. The cache hit rate is calculated as the number of queries returning results from the cache divided by the total number of queries. The query response time is calculated as the average execution time of all queries. The concurrent processing capacity is evaluated by stress testing to obtain the maximum number of queries that the system can handle simultaneously. The resource utilization rate is calculated as the actual amount of resources used divided by the total available amount of resources. This step uses a multi-level caching strategy and hierarchical storage technology to significantly reduce data access latency. Index optimization reduces the query scan range. The overall query performance can be improved by 3 to 5 times compared to the unoptimized state.

[0060] The specific implementation of step S05 is to use the atlas adaptive evolution model to monitor and respond to business changes in real time. First, an evolution feature vector is constructed. The total number of nodes in the current atlas is counted to calculate the node number change rate, which is equal to the current number of nodes minus the baseline number of nodes divided by the baseline number of nodes. The total number of edges in the current atlas is counted to calculate the edge number change rate, which is equal to the current number of edges minus the baseline number of edges divided by the baseline number of edges. The structure density change rate is calculated as the current number of edges divided by the maximum possible number of edges, and the difference between the baseline density is divided by the baseline density. The number of connected components of the graph is analyzed to calculate the connectivity change rate. The maximum depth of the graph is measured to calculate the hierarchical depth change rate. The five-dimensional change rates form the evolution feature vector. Then, the baseline evolution vector is obtained. The standard node density is obtained from the historical stable period atlas structure statistics. The standard edge density is also derived from historical data analysis. The standard structure complexity is obtained by calculating the average loop complexity of the historical atlas. The standard connectivity is the reciprocal of the average connected component of the historical atlas. The standard level number is the average maximum depth of the historical atlas. Then, the Euclidean distance in the five-dimensional space between the evolution feature vector and the baseline evolution vector is calculated. The difference between the corresponding dimensions of the two vectors is squared and summed, and then the square root is taken. When the calculated distance value is greater than 0.85, it indicates that the atlas structure has changed significantly. At this time, the pattern matching algorithm is started to analyze the change pattern. The pattern matching algorithm first extracts the weight of each dimension of the evolution feature vector to identify which dimension has the most significant change. If the node number change rate is the largest, it is judged as the entity expansion mode. If the edge number change rate is the largest, it is judged as the relationship enhancement mode. If the structure density change rate is the largest, it is judged as the structure reorganization mode. According to the identified evolution mode, the appropriate atlas structure adjustment scheme is selected from the pre-defined adjustment scheme library. The adjustment scheme includes adding new node types and new relationship types, modifying node attributes, and adjusting relationship weights. The adjustment scheme is recorded through the version control mechanism to ensure that the evolution process is traceable and rollbackable. The purpose of this step is to enable the atlas to automatically adapt to business changes and avoid the lag of manual maintenance. The multi-scale representation integration mechanism based on hierarchical feature fusion can capture both local detail changes and global structure evolution. The inter-sequence dependency modeling algorithm based on cyclic attention can accurately capture the timing rules of the evolution process. The Euclidean distance threshold of 0.85 is the optimal decision point verified through experiments. Changes below this threshold are normal fluctuations and do not need to be adjusted. Changes above this threshold need to trigger the adaptive evolution mechanism.

[0061] The specific implementation of step S06 is to realize the data traceability and exploration service function. First, the traceability target input by the user is received. The traceability target can be a business entity identifier such as product number, work order number, or batch number. The starting node corresponding to the traceability target is located in the graph database. The depth-first search algorithm or breadth-first search algorithm is used to traverse along the edges of the graph. The depth-first search algorithm is suitable for scenarios with long traceability links, which prioritize exploring a single complete path. The breadth-first search algorithm is suitable for scenarios that require comprehensive exploration, which expands the search range hierarchically. During the traversal, a path pruning strategy is applied. When the traversal depth exceeds the preset maximum depth, the search for that path is stopped. When the cumulative path weight is lower than the threshold, the path is pruned. The cumulative path weight is obtained by multiplying the relation weights of all edges on the path. A heuristic search strategy is used to prioritize exploring edges with higher weights. The nodes to be explored are processed in descending order of edge weight by maintaining a priority queue. All paths, their weights, and confidence information are recorded during the traversal. The confidence is calculated based on the path weight and path length. A complete data traceability link is generated and displayed to the user in a graphical form. Then, query lag effect detection is performed, the response time of the query operation is statistically analyzed, and the lag effect error is calculated to be equal to the actual response time. The response time is calculated by subtracting the theoretical optimal response time from the response time and then dividing by the theoretical optimal response time. When the hysteresis error is greater than 0.15, it indicates that the system has a significant backhaul error. At this point, the hysteresis compensation algorithm is activated. The hysteresis compensation algorithm stores recent query requests and results by establishing a query history table. For duplicate or similar queries, results are returned directly from the history table. A result caching mechanism is established to store the results of high-frequency queries in memory. The cache expiration time is set to 300 to 600 seconds, adjusted according to the data update frequency. For first-time queries or queries where the cache has expired, real-time calculations are performed and the cache is updated. When the hysteresis error is controlled within... A value between 0 and 0.05 indicates good query performance. In this case, the standard query processing mode is used to directly execute the graph traversal operation and return the results. This step utilizes the natural relational structure of the graph to achieve efficient data tracing. Compared with traditional relational database multi-table join queries, the graph traversal algorithm can significantly reduce query complexity. Path pruning and heuristic search further improve tracing efficiency. The hysteresis compensation algorithm effectively eliminates the performance loss caused by repeated queries. The maximum search depth reference value is set to 10 to 15 layers, the path weight threshold is set to 0.1, and the hysteresis effect error threshold of 0.15 corresponds to an additional delay of approximately 150 milliseconds.

[0062] The specific implementation of step S07 is to establish a role-based access control and data security mechanism. First, user roles are defined according to the enterprise organizational structure, and users are divided into different levels such as system administrator role, data analyst role, business operator role, and visitor role. Data access permissions are configured for each role. The system administrator role has read and write permissions for all data and system configuration permissions. The data analyst role has read-only and query permissions for data. The business operator role has read and write permissions for specific business domain data. The visitor role only has read-only permissions for public data. A permission control matrix is established to record the access relationship between roles and data resources. Before a user performs a data operation, permission verification is performed. The current user role is determined to have corresponding operation permissions by querying the permission control matrix. If the permissions are insufficient, the operation is denied and access logs are recorded. Then, encryption is performed on the data transmission process. The Secure Sockets Layer (SSL) protocol or the Transport Layer Security (TLS) protocol is used to establish an encrypted channel. The Advanced Encryption Standard (AES) algorithm is used for symmetric encryption of transmitted data, with a key length of 256 bits to ensure encryption strength. For sensitive data such as passwords and personal information, an asymmetric encryption algorithm is used. The receiver's public key is used to encrypt the data, and the receiver's private key is used to decrypt the original data. At the data storage level, sensitive fields are encrypted and stored. The password is stored using a one-way hash algorithm. Then, a data backup strategy is set. The industrial data graph is backed up in a combination of full backup and incremental backup. Full backup is performed once a week to export and store the complete graph database. Incremental backup is performed once a day to backup only the data added and modified that day. Backup data is stored in a geographically redundant disaster recovery center. The backup retention period is set to 30 days for recent backups and 12 months for historical backups. Backup data exceeding the retention period is automatically cleaned up. A disaster recovery plan is established to quickly recover from backup data when the system fails. Regular recovery drills are conducted to verify the availability of backup data. This step ensures the confidentiality, integrity, and availability of data through multiple layers of security protection mechanisms. Role-based access control enables fine-grained permission management. Encryption technology prevents data from being stolen or tampered with during transmission and storage. Backup strategies ensure data persistence and recoverability. The SSL protocol version is recommended to be 3.0 or later, and the TLS protocol version is recommended to be 1.2 or later. The encryption key for backup data should be managed separately from the production environment.

[0063] It needs to be further explained that the detailed structure of the atlas adaptive evolution model includes five main parts of input layer, feature extraction layer, attention layer, decision layer and output layer. The input layer receives two types of data. The first type is the evolution feature vector containing the number of node change rate, the number of edge change rate, the structure density change rate, the connectivity change rate and the hierarchical depth change rate. The second type is the business change information containing the new system access identifier, the business process change identifier and the data mode change identifier. The input layer normalizes the continuous features to map the values to the interval of 0 to 1, and one-hot encodes the discrete features to represent them as vectors. The feature extraction layer uses a convolutional neural network architecture. The first convolutional layer uses 64 convolutional kernels to extract local features with a kernel size of 3x3 and a stride of 1. The rectified linear unit activation function is used to introduce nonlinearity. The second convolutional layer uses 128 convolutional kernels to extract higher-level features and uses max-pooling operation for down-sampling to reduce feature dimension. The pooling window size is 2x2. The third convolutional layer uses 256 convolutional kernels to extract global semantic features. The residual connection is used to fuse shallow features with deep features to realize multi-scale representation integration based on hierarchical feature fusion. The attention layer uses a multi-head attention mechanism to divide the feature vector into 8 attention heads. Each attention head independently calculates the query vector, key vector and value vector. The scaling dot product attention is used to calculate the correlation weight between features. The output of multiple attention heads is concatenated and linearly transformed to obtain the comprehensive feature representation. The recurrent attention structure is introduced to associate the current time feature with the historical time feature. The gated recurrent unit is used to capture the temporal dependence of the evolution process to realize sequence dependence modeling based on recurrent attention. The decision layer contains three fully connected neural network layers. The first fully connected layer maps the feature vector output by the attention layer to a 512-dimensional hidden representation. The second fully connected layer further maps it to a 256-dimensional representation. The third fully connected layer maps it to the number of evolution strategy categories. The Softmax function is used to convert the output to a probability distribution, representing the recommended probability of different evolution strategies. The output layer generates the atlas structure adjustment scheme. According to the probability distribution output by the decision layer, the evolution strategy with the highest probability is selected. This strategy corresponds to a specific set of adjustment operations including node type addition, relationship type addition, attribute modification and weight adjustment. The adjustment scheme is output in a structured format containing operation type, target object, modification content and execution order.

[0064] The detailed steps of the atlas adaptive evolution model training dataset establishment include four stages of historical data collection, case labeling, feature construction, and dataset division. In the historical data collection stage, the past 2 to 3 years of atlas evolution history records are extracted from the production environment, each record containing the atlas structure snapshot before evolution, evolution trigger event, atlas structure snapshot after evolution, and evolution effect evaluation result. The atlas structure snapshot is stored in a graph serialization format, including node set and edge set and all attribute information. The evolution trigger event record records the business reasons for the change in the atlas, such as the introduction of new product lines, production process optimization, and system function expansion. The evolution effect evaluation result quantitatively evaluates the pros and cons of evolution through system performance indicators such as query response time, data throughput, and resource utilization. In the case labeling stage, each evolution case is manually labeled. First, calculate the structural difference between the atlas before and after evolution, count the number of added, deleted, and modified nodes, and count the number of added, deleted, and modified edge weights. Then, label the evolution type and classify the evolution into four modes: expansion-type evolution, contraction-type evolution, restructuring-type evolution, and optimization-type evolution. Expansion-type evolution corresponds to business scale growth scenarios, contraction-type evolution corresponds to business simplification scenarios, restructuring-type evolution corresponds to process reorganization scenarios, and optimization-type evolution corresponds to performance improvement scenarios. Label the evolution effect according to the degree of system performance improvement after evolution, with four levels: excellent, good, general, and poor. Excellent indicates a performance improvement of more than 30%, good indicates a performance improvement of between 10% and 30%, general indicates a performance improvement of between 0% and 10%, and poor indicates a performance decrease. In the feature construction stage, training samples are constructed based on the labeled evolution cases. For each evolution case, extract the evolution feature vector as the input feature. Calculate the node number change rate as the number of nodes after evolution minus the number of nodes before evolution divided by the number of nodes before evolution. Calculate the edge number change rate as the number of edges after evolution minus the number of edges before evolution divided by the number of edges before evolution. Calculate the structural density change rate by comparing the density indicators of the graphs before and after evolution. Calculate the connectivity change rate by comparing the number of connected components of the graphs before and after evolution. Calculate the level depth change rate by comparing the maximum depth of the graphs before and after evolution. Encode the business change information as a feature vector, and label the evolution type and evolution effect as labels. Establish the correspondence between the input feature vector and the output label to form the training sample. In the dataset division stage, all training samples are randomly divided into training set, validation set, and test set according to the ratio of 8:1:1. The training set is used for model parameter learning and contains 80% of the samples. The validation set is used for hyperparameter tuning and model selection and contains 10% of the samples. The test set is used for final performance evaluation and contains 10% of the samples. Ensure that the sample distribution of the three datasets is consistent. For evolution types with small sample sizes, use data augmentation techniques to expand the samples. Generate similar evolution cases by making minor perturbations to the graph structure, such as randomly adding or deleting a small number of edges. Finally, form a training dataset containing 5000 to 10000 evolution cases.The data set is stored in a standard format to facilitate batch loading during model training, and after the training data set is established, quality inspection is performed to verify the integrity of the input features and the accuracy of the labels, and abnormal samples are removed to ensure the quality of the data set.

[0065] It should be noted that the multi-scale representation integration mechanism based on hierarchical feature fusion integrates the features of different levels of the convolutional neural network by weighted fusion, so that the model obtains local detail information and global semantic information at the same time. Compared with the traditional method of using only single-level features, the mechanism significantly enhances the model's perception ability of graph structure changes. The shallow features retain the fine-grained structure characteristics of the graph, such as the direct connection relationship between nodes, and the deep features extract the high-level semantic features of the graph, such as community structure and modularity features. The fusion of the two enables the model to capture both local minor changes and understand global evolution trends. In complex industrial scenarios, graph evolution often involves changes at multiple scales. The multi-scale representation integration mechanism can comprehensively analyze the evolution patterns and provide more rich feature information for evolution strategy recommendation, improving the accuracy and integrity of evolution decisions.

[0066] In addition, the present application also solves the technical problem of the lack of adaptive evolution capability of the industrial data processing system. The existing industrial data processing system usually adopts static data structure and fixed processing flow. When the enterprise business process is adjusted, new system is integrated or data mode is changed, the system cannot automatically adapt to these changes, and a large amount of manual configuration and system reconstruction work is required, which not only increases the maintenance cost, but also may introduce data inconsistency and system instability problems in the adjustment process. The present application establishes a graph adaptive evolution model, uses an evolution feature vector to monitor graph structure changes, and automatically starts structure adjustment when the Euclidean distance between the evolution feature vector and the benchmark evolution vector exceeds 0.85. The multi-scale representation integration mechanism based on hierarchical feature fusion and the inter-sequence dependence modeling algorithm based on recurrent attention are used to realize intelligent perception of business changes and automatic optimization of graph structure. The version control mechanism ensures the traceability and roll-backability of the evolution process, and completely solves the technical problem of the lack of adaptive capability of the traditional system.

[0067] The second aspect of the present application provides a computer readable storage medium, the computer readable storage medium stores program instructions, the program instructions are run in the computer, and are used for executing the above-mentioned digital main line processing method based on industrial data graph.

[0068] The third aspect of the present application provides a digital main line processing system based on industrial data graph, which comprises the above-mentioned computer readable storage medium. The system is any one of a computer, a server and a single-chip microcomputer. The computer readable storage medium is arranged in the system, and the system is provided with a microprocessor for executing the program instructions stored in the computer readable storage medium.

[0069] Specifically, the principle of the present application is that the present application can solve the core technical problem that the industrial data processing system is difficult to realize cross-system deep correlation analysis. The fundamental principle of the present application is to use a graph data model to reorganize and manage cross-system industrial data. The traditional relational database uses a two-dimensional table structure to store data, and expresses the correlation between data through foreign key constraints. Although this structured organization method has advantages in data management within a single system, it faces serious limitations when dealing with complex correlation analysis across systems. It needs to use multi-table connection operations to obtain correlation information, which not only has high query complexity, but also is difficult to express multi-level indirect correlation. The present application abstracts data entities from different systems as nodes in a graph and abstracts cross-system correlation as edges in a graph, and constructs a unified industrial data relationship network. This graph structure is naturally suitable for expressing and processing complex correlation, so that cross-system deep correlation analysis becomes a path search and traversal problem in the graph. The application of graph traversal algorithm is the key technical means for the present application to realize cross-system deep correlation analysis. Through depth-first search, indirect correlation paths between data entities can be found, and through breadth-first search, the correlation range and influence degree of data entities can be analyzed. These analysis capabilities are difficult to achieve by traditional relational queries. The graph index optimization strategy of the present application greatly improves the efficiency of graph traversal by establishing node attribute index and relationship index, making large-scale cross-system correlation analysis possible. The adaptive evolution model ensures that the graph structure can be dynamically adjusted as the cross-system business relationship changes, maintaining the system's ability to adapt to complex correlation analysis requirements. This data organization and analysis method based on graph atlas solves the fundamental defects of traditional methods from the data model level, so the technical solution of the present application is logically scientific and reasonable.

[0070] A specific embodiment 1 of the present application is provided below, and the specific implementation of each step in embodiment 1 is described in detail as follows.

[0071] The specific implementation of step S01 is to realize the unified access of heterogeneous data sources by constructing a multi-source data acquisition adapter, wherein the data update frequency optimization function is expressed as follows:

[0072] ;

[0073] In the formula, is a real-time data demand coefficient, which is calculated by counting the number of data queries per unit time, is a standard data update reference value, which is usually 1, is the system response time, in seconds, is the reference response time, in seconds, which is usually 1 second, The system type weighting coefficient is 0.8 for Manufacturing Execution Systems, 0.5 for Enterprise Resource Planning Systems, and 0.3 for Customer Relationship Management Systems. This is the latency sensitivity factor, with a default value of 0.6. This is the network fluctuation error term, with an empirical value of 0.05 to 0.15.

[0074] The specific implementation of step S02 involves cleaning and standardizing the collected multi-source heterogeneous data. The formula for calculating the data quality assessment vector is as follows:

[0075] ;

[0076] Among them, data integrity coefficient Accuracy coefficient Consistency coefficient Timeliness coefficient , effectiveness coefficient In the formula, For the complete record count, This represents the total number of records. For the correct number of data, To verify the total number of data, For the first Data values ​​from one data source, For reference data values, For the largest difference in data values, For the number of data sources, For real-time data volume, To comply with business rules regarding data volume, To check the total amount of data.

[0077] The specific implementation of step S03 is to construct an industrial data map based on the integrated dataset. The formula for calculating the relationship weight is as follows:

[0078] ;

[0079] In the formula, For the frequency of occurrence of association, For the maximum association frequency, This represents the relation type weight coefficient: 0.9 for containment relations, 0.7 for dependency relations, and 0.5 for association relations. The semantic distance between entities is calculated using the cosine distance of word vectors. This is the distance threshold, typically set to 5. This is a semantic similarity enhancement term, with an empirical value of 0.1 to 0.3.

[0080] The specific implementation of step S04 involves constructing a graph index optimization strategy and a hierarchical storage architecture. The formula for calculating the query performance superposition optimization vector is as follows:

[0081] ;

[0082] Among them, index hit rate Cache hit rate Normalized query response time Normalized concurrent processing capability resource utilization rate In the formula, The number of queries that used the index. This represents the total number of queries. To cache the number of hit queries, For the first The execution time of this query is in milliseconds. This represents the total number of queries. This is the baseline response time, measured in milliseconds, typically 10 milliseconds. To the maximum number of concurrent queries, This is the baseline concurrency level, typically set to 100. This represents the amount of resources already used. This represents the total amount of available resources.

[0083] The specific implementation of step S05 involves using a graph adaptive evolution model to monitor business changes in real time. The formulas for calculating the rate of change of each dimension are as follows:

[0084] Rate of change in the number of nodes ;

[0085] Rate of change of edge number ;

[0086] Structural density change rate ;

[0087] rate of change of connectivity ;

[0088] Hierarchical depth change rate ;

[0089] In the formula, This represents the current node number. The number of reference nodes, This is the current edge number. As the baseline number of edges, For an undirected graph, the maximum possible number of edges is... For directed graphs , As the baseline density, the number of current connected components, the number of reference connected components, the current maximum depth, the reference maximum depth. The Euclidean distance between the evolution feature vector and the reference evolution vector is calculated as follows:

[0090]

[0091] In the formula, is the first dimensional evolution feature change rate, in order , , , , , is the historical maximum value of the first dimensional change rate, is the reference evolution value of the first dimensional, in order reference node density, reference edge density, reference structural complexity, reference connectivity, reference hierarchical number, is the maximum value of the first dimensional reference evolution. The evolution feature fusion function is expressed as follows:

[0092]

[0093] In the formula, is the structural change feature vector, and the length of its module , is the business demand change vector, and the length of its module , wherein is the new system access index, is the business process change index, is the function expansion index, is the system performance index vector, and the length of its module , wherein is the processor performance index, is the memory performance index, is the input / output performance index, , , are the historical maximum values of the corresponding vector module lengths, , , are weight coefficients and , the experience value , , .

[0094] ​​The specific implementation of step S06 is to realize the data traceability and exploration service function, wherein the path weight cumulative calculation formula is expressed as follows:

[0095] ;

[0096] The confidence calculation formula is expressed as follows:

[0097] ;

[0098] In the formula, is the relationship weight of the i-th edge in the path, is the number of edges contained in the path, is the path length, which is equivalent to the number of edges , , is the standard path length, which is usually 5, is the semantic consistency enhancement item, and the experience value is 0.05 to 0.2. The hysteresis error calculation formula is expressed as follows:

[0099] ;

[0100] In the formula, is the actual response time, in milliseconds, is the theoretical optimal response time, in milliseconds, is the calculation complexity coefficient, which is calculated by the number of nodes and the number of edges involved in the query , wherein is the number of nodes involved in the query, is the number of edges involved in the query, and are complexity weight coefficients, and the experience values are 0.3 and 0.7 respectively, is the standard complexity reference value, which is usually 100, is the path complexity influence factor, which is determined according to the path depth and branch number, and the experience value is 0.8 to 1.2. The query optimization weight function is expressed as follows:

[0101] ;

[0102] In the formula, is the query type coefficient, which is 0.3 for simple node query, 0.6 for relationship traversal query, and 0.9 for multi-hop path query, is the historical performance coefficient, and the calculation method is , wherein is the historical average response time, is the current average response time, is the resource state coefficient, and the calculation method is , is the current response time in milliseconds, is the response time threshold in milliseconds, and the default value is 100 milliseconds.

[0103] The specific implementation of step S07 is the same as the foregoing, and will not be described in detail here.

[0104] The evolutionary blocking risk matrix calculation formula is expressed as follows:

[0105] ;

[0106] In the formula, is the structural complexity, which is obtained by calculating the number of loops and the variance of node degrees of the graph, is the data consistency coefficient, is the performance impact coefficient, is the compatibility coefficient, is the rollback difficulty coefficient, 、 、 、 、 are the historical maximum values of the corresponding indicators, respectively.

[0107] The query waiting time distribution vector calculation formula is expressed as follows:

[0108] ;

[0109] In the formula, is the average waiting time for simple node queries, is the average waiting time for relationship traversal queries, is the average waiting time for multi-hop path queries, is the average waiting time for aggregation statistical queries, is the average waiting time for joint queries, and all units are milliseconds, is the baseline waiting time, which is usually 10 milliseconds.

[0110] The performance gain evaluation matrix calculation formula is expressed as follows:

[0111] ;

[0112] wherein the performance gain coefficient , is the average response time before optimization of the th query type, is the average response time of the th query type under the th optimization strategy, is the number of query types, is the number of optimization strategies.

[0113] System load fluctuation monitoring vector calculation formula is expressed as follows:

[0114] ;

[0115] In the formula, 、 、 、 、 respectively, the standard deviation of the processor usage rate, memory usage rate, disk input / output rate, network bandwidth usage rate, and concurrent connection number, 、 、 、 、 respectively, the mean of the corresponding indicators.

[0116] The principle and effect of the above formula are explained as follows: data update frequency optimization function By comprehensively considering real-time data demand, system response characteristics, and network fluctuation influence, dynamic adjustment of the update frequency of different data sources is realized. Compared with the fixed frequency update mode, this function can adaptively adjust according to the actual business demand and system performance, effectively reduce system resource consumption while ensuring data timeliness, and on average, improve data collection efficiency by 25% to 35%. The data quality evaluation vector quantifies the data quality status in multiple dimensions, wherein the consistency coefficient The normalized Manhattan distance is used to measure the difference between different data sources, and the abstract data quality concept is converted into a calculable numerical indicator, providing a quality benchmark for subsequent data processing. Compared with traditional subjective evaluation methods, this vector calculation method has objectivity and reproducibility, and can monitor the data quality change trend in real time, providing accurate quality guidance for data cleaning and preprocessing. The relationship weight calculation formula combines multiple factors such as association frequency, relationship type, and semantic distance, wherein the exponential decay term Ensures that the greater the distance between entities, the smaller the entity relationship weight, accurately quantifies the association strength between entities, and compared with the simple frequency statistical method, this formula introduces a semantic distance decay term and a semantic similarity enhancement term, which can more accurately reflect the real association degree between entities, providing accurate path weight information for graph traversal algorithms, and significantly improving the accuracy and efficiency of data tracing. The query performance superposition optimization vector comprehensively evaluates multiple performance indicators to reflect the effect of query optimization strategy. The dimensions of this vector are related to each other to form a closed-loop feedback mechanism for performance optimization, which can identify system performance bottlenecks and guide optimization strategy adjustment. Compared with single performance indicator evaluation, this vector method can avoid the overall performance decline caused by local optimization, and achieve comprehensive improvement of query performance. The evolution feature change rate calculation formula group calculates the standardized relative change amount The five dimensions quantitatively measure the degree of change in the atlas structure. The formulas adopt a relative change rate form to eliminate the influence of the reference value difference, ensure the comparability of changes in different dimensions, and provide accurate quantitative basis for evolution decision-making. Euclidean distance calculation formula The degree of change in the atlas structure is quantified by the geometric distance in the five-dimensional space to provide accurate judgment criteria for adaptive evolution triggering. The formula adopts normalization processing to eliminate the influence of dimensional differences in different dimensions. Compared with the simple threshold comparison method, the geometric distance can comprehensively consider the combined effect of multi-dimensional changes, improve the accuracy and timeliness of evolution decision-making. Evolution feature fusion function The integrated evolution feature is represented by weighted fusion of structure change, business demand and system performance. The function adopts vector length calculation and normalization processing to balance the influence of different types of features and provide comprehensive input information for the adaptive evolution model. Compared with single feature input, the fusion feature can improve the accuracy and adaptability of evolution strategy recommendation. Path weight accumulation formula The influence of path weight and path length is considered, and the exponential decay term suppresses the confidence of long paths to ensure the reliability of the tracing result. Compared with the simple weight average method, the formula can more accurately evaluate the credibility of the tracing path. Hysteresis effect error calculation formula By quantifying the backhaul error in the query process, the formula provides trigger conditions and compensation basis for the hysteresis compensation algorithm. The formula considers the influence of actual response time deviation and calculation complexity , which can accurately identify system performance anomalies. Compared with the fixed threshold determination method, the formula considers the dynamic influence of path complexity and can provide accurate performance evaluation in different query scenarios to effectively improve the stability of query response. Query optimization weight function adopts an exponential decay form , which combines the influence of multiple factors to realize dynamic allocation of the weight of different optimization strategies. The exponential term in the function can effectively suppress the weight of high response time queries and prioritize the performance of low latency queries. Compared with the linear weight allocation method, the exponential decay mechanism can better balance the relationship between query performance and resource utilization to improve the overall query service quality. Other matrix and vector formulas

[0117] convert complex system state and performance indicators into computable and comparable numerical forms to provide quantitative basis for system monitoring, performance evaluation and decision support. Compared with qualitative analysis methods, these mathematical models have accuracy and operability, which can support automated system management and optimization adjustment.

[0118] An enterprise has a complete production and manufacturing system, including enterprise resource planning system (ERP), manufacturing execution system (MES) and customer relationship management system (CRM). The technical team adopts the technical solution of the application to build a complete industrial data graph digital main line processing system.

[0119] Firstly, the technical team builds a multi-source data collection adapter to collect real-time data from the enterprise's existing three core systems. The ERP system contains 2847 database tables, the MES system contains 1925 database tables, and the CRM system contains 856 database tables. The technical team establishes a data source identification mapping table to record the access path and update frequency of each system. The data update frequency of the ERP system is set to once an hour, the real-time production data update frequency of the MES system is once every 5 minutes, and the customer data update frequency of the CRM system is once every 30 minutes. The multi-source data collection adapter supports SQL, ODBC, API and other data access protocols to ensure the unified access of heterogeneous data sources.

[0120] In the data preprocessing stage, the technical team cleans and standardizes the collected multi-source heterogeneous data. The total amount of raw data reaches 67TB, and after data cleaning, the effective data amount is 52TB. The technical team builds a data quality evaluation vector, which contains five dimensions of quality indicators. As shown in Table 1:

[0121] Table 1 Data quality evaluation vector index table

[0122]

[0123] The data quality evaluation results show that all indicators meet the preset threshold standard, laying a good foundation for subsequent industrial data graph construction.

[0124] Based on the integrated data set, the technical team builds an industrial data graph. The graph contains 156892 data nodes and 284567 associated edges. The node types include product entity nodes, process flow nodes, equipment state nodes, quality detection nodes and order information nodes, etc. The edge types include process association, equipment dependency, quality correlation, time sequence relationship, etc. Each node contains entity identification, entity type, business attribute and timestamp information, and each edge contains relationship type, relationship strength, relationship direction and relationship weight information. The graph database uses Neo4j attribute graph model to store data, supporting complex graph query and traversal operations.

[0125] To optimize query performance, the technical team implemented a graph index optimization strategy and a tiered storage architecture. They established 89 node attribute indexes, 156 relationship indexes, and 43 composite indexes. The tiered storage architecture stores frequently accessed data (higher than 80% access frequency) in a 256GB memory layer, warm data (accessed between 20% and 80% access frequency) in an 8TB SSD layer, and cold data (accessed less than 20% access frequency) in a 32TB hard disk layer. The five metrics for optimizing query performance are shown in Table 2.

[0126] Table 2 Query Performance Optimization Metrics

[0127]

[0128] The technical team implemented an adaptive graph evolution model. This model employs a multi-scale representation integration mechanism based on hierarchical feature fusion and a sequence dependency modeling algorithm based on recurrent attention. The evolutionary feature vector includes five dimensions: the rate of change of node number, the rate of change of edge number, the rate of change of structural density, the rate of change of connectivity, and the rate of change of layer depth. The baseline evolutionary vector is defined with standard node density of 0.745, standard edge density of 1.287, standard structural complexity of 2.156, standard connectivity of 0.923, and standard number of layers of 4.2. When the Euclidean distance between the evolutionary feature vector and the baseline evolutionary vector exceeds 0.85, the system automatically initiates a pattern matching algorithm to adjust the graph structure.

[0129] During system operation, the technical team detected a significant business change event. The company added an automated production line, causing a significant change in the evolutionary feature vector. For example... Figure 2 As shown, the rate of change of the number of nodes increased from 0.123 to 0.456, the rate of change of the number of edges increased from 0.089 to 0.378, and the rate of change of the structure density increased from 0.067 to 0.234. The calculated Euclidean distance between the evolved eigenvector and the baseline evolved vector was 0.912, exceeding the set threshold of 0.85. The system automatically initiated graph structure adjustment, adding 2847 nodes and 4692 edges, completing the adaptive evolution of the graph structure.

[0130] The data traceability and exploration service is implemented based on a graph traversal algorithm. The technical team designed an intelligent data traceability function that supports tracing from any data node along related paths. In a quality issue traceability case, starting from the non-conforming product node, the system used a depth-first search algorithm to traverse 156 relevant nodes and 289 related edges, generating a complete data traceability link. During the traceability process, a query lag effect error of 0.187 was found, exceeding the set threshold of 0.15. The system automatically activated a lag compensation algorithm, eliminating the impact of backhaul errors through query history and result caching mechanisms, ultimately reducing the lag effect error to 0.043.

[0131] As shown in Figure 3 , the query latency distribution vector of the system records the latency characteristics of different types of query operations. The average latency of simple node queries is 23.7 , the latency of relationship traversal queries is 89.4 , the latency of multi-hop path queries is 167.8 , the latency of aggregate statistical queries is 245.3 , and the latency of joint queries is 312.6 . These data provide important reference for query performance optimization.

[0132] The technical team established a role-based access control and data security mechanism. The system defines five user roles, including system administrator, data analyst, production engineer, quality inspector and ordinary user. Different roles have different data access permissions. The system administrator has read and write permissions for all data, the data analyst has read permission for statistical analysis data, the production engineer has read and write permissions for production-related data, the quality inspector has read and write permissions for quality data, and the ordinary user can only view basic statistical information. The data transmission process uses SSL encryption protocol with 256-bit encryption strength. The data backup strategy is set to daily incremental backup and weekly full backup, and the backup data is stored in a separate storage system.

[0133] The system load fluctuation monitoring vector records the system resource usage in real time. The processor usage fluctuation range is between 45% and 78%, the memory usage fluctuation range is between 52% and 85%, the disk I / O fluctuation range is between 156 and 892MB / s, the network bandwidth fluctuation range is between 23 and 156 , and the concurrent connection number fluctuation range is between 89 and 267. When the system load fluctuation exceeds the preset threshold, the system automatically triggers resource adjustment and load balancing mechanism to ensure stable operation of the system.

[0134] The evolution blocking risk matrix evaluates the potential risks in the evolution process of the graph structure. The structural complexity risk coefficient is 0.234, the data consistency risk coefficient is 0.156, the performance impact risk coefficient is 0.089, the compatibility risk coefficient is 0.067, and the rollback difficulty risk coefficient is 0.123. The comprehensive risk assessment result is 0.134, which belongs to the low risk range, ensuring the safety of the evolution process.

[0135] The performance gain evaluation matrix quantifies the performance improvement effect brought by different optimization strategies. The performance gain coefficient of the node attribute index strategy for simple node queries is 2.34, and the performance gain coefficient for relationship traversal queries is 1.67. The performance gain coefficient of the relationship index strategy for multi-hop path queries is 3.12, and the performance gain coefficient for aggregation statistical queries is 2.78. The performance gain coefficient of the composite index strategy for joint queries is 4.25, which significantly improves the execution efficiency of complex queries.

[0136] In practical applications, the evolutionary feature fusion function integrates multi-dimensional evolutionary feature information. The maximum value of the structural change feature vector is 2.456, the maximum value of the business demand change vector is 1.789, and the maximum value of the system performance index vector is 3.123. The weight coefficients are set to =0.4, =0.35, and =0.25, which satisfy the constraint condition. The calculation result of the comprehensive evolutionary feature vector is 0.867, which provides a quantitative basis for graph evolution decision-making.

[0137] The query optimization weight function calculates the weight distribution of different optimization strategies. The query type coefficient is divided into 5 levels according to the query complexity, the historical performance coefficient is calculated based on the performance statistics of the past 30 days, and the resource state coefficient reflects the current system resource usage. The response time threshold is set to 200, and when the query response time exceeds the threshold, the optimization strategy weight decays exponentially.

[0138] The present application has brought significant technical progress compared to traditional data processing methods. The traditional method usually uses a relational database to store multi-source data, and the complex association relationship between data is difficult to express and query efficiently. The present application stores data entities and association relationships in the form of a graph structure by constructing an industrial data graph, which can intuitively express the complex relationship network between data and significantly improve the efficiency of association query and data traceability. The traditional method needs to manually adjust the data model and query logic when facing business changes, while the adaptive evolution model of the present application can automatically perceive business changes and adjust the graph structure, significantly improving the adaptive ability of the system. The query optimization of the traditional method mainly relies on database indexing and SQL optimization, while the present application forms a multi-dimensional performance optimization system through graph index optimization strategy, hierarchical storage architecture and query performance superposition optimization, which can better cope with complex industrial data query requirements.

[0139] ​​​​​​​​​​​​​It should be noted that the variables involved in the present application are explained in detail as shown in the following Tables 3, 4, and 5.

[0140] Table 3 Variable Explanation Table (First Part)

[0141]

[0142] Table 4 Variable Explanation Table (Second Part)

[0143]

[0144] Table 5 Variable Explanation Table (Third Part)

[0145]

[0146] The above description is merely a specific implementation of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application.

Claims

1. A digital thread processing method based on industrial data maps, characterized in that, An industrial data relationship network is constructed by abstracting data entities as nodes and relationships as edges. A graph traversal algorithm is used to achieve deep cross-system correlation analysis, including the following steps: Constructing a multi-source data acquisition adapter to collect data in real time from database tables in Enterprise Resource Planning (ERP), Manufacturing Execution System (MES), and Customer Relationship Management (CRM) systems; establishing a data source identifier mapping table to achieve unified access to heterogeneous data sources; cleaning and standardizing the collected multi-source heterogeneous data to construct an integrated dataset; establishing a data quality assessment vector to complete data preprocessing and quality control; and constructing an industrial data graph based on the integrated dataset, abstracting data entities as nodes and relationships as edges, and using a graph database to store nodes and edges. Edge attribute information; construct graph index optimization strategies and hierarchical storage architecture, establish multi-dimensional indexes, adopt hot data caching mechanisms and query result pre-computation systems; utilize graph adaptive evolution models to monitor business changes in real time, and when the Euclidean distance between the evolution feature vector and the baseline evolution vector in multi-dimensional space is >0.85, activate the pattern matching algorithm to automatically adjust the graph structure; implement data traceability and exploration service functions, implement intelligent data traceability functions based on graph traversal algorithms, and activate the hysteresis compensation algorithm when the query hysteresis error is >0.15, and adopt the standard query processing mode if the hysteresis error ∈ [0, 0.05]; establish role-based access control and data security protection mechanisms; The evolutionary feature vector is used to describe the feature information of graph structure changes, including five dimensions: the rate of change of the number of nodes, the rate of change of the number of edges, the rate of change of structural density, the rate of change of connectivity, and the rate of change of level depth; the baseline evolution vector is used to provide a reference standard for graph structure evolution, including five baseline values: standard node density, standard edge density, standard structural complexity, standard connectivity, and standard number of levels. The graph adaptive evolution model has the following structure: the input layer receives evolutionary feature vectors and business change information; the feature extraction layer extracts structural features through a convolutional neural network; the attention layer establishes inter-feature relationships using a multi-head attention mechanism; the decision layer outputs evolutionary strategy suggestions through a fully connected neural network; and the output layer generates a graph structure adjustment scheme. The graph adaptive evolution model adopts a multi-scale representation integration mechanism based on hierarchical feature fusion and a sequence dependency modeling algorithm based on recurrent attention to realize the evolution of the graph structure.

2. The method according to claim 1, characterized in that, The multi-source data acquisition adapter adopts a protocol-adaptive connection method, establishes corresponding data access drivers for different database types, and supports multiple data access protocols such as Structured Query Language, Open Database Connectivity, and Application Programming Interface.

3. The method according to claim 2, characterized in that, The data source identifier mapping table records the unique identifier, database connection parameters, table structure information, and field mapping relationships of each data source. It also records the access path and update frequency of each system's data source, and is used to establish a unified data access interface. The unique identifier is derived from the system codes of the Enterprise Resource Planning System, Manufacturing Execution System, and Customer Relationship Management System.

4. The method according to claim 3, characterized in that, The data quality assessment vector comprises five dimensions: data integrity coefficient, accuracy coefficient, consistency coefficient, timeliness coefficient, and validity coefficient. It is constructed by performing quality checks and scoring on the original data to monitor data integrity.

5. The method according to claim 4, characterized in that, In an industrial data graph, nodes represent data entities, and edges represent the relationships between entities. Each node and edge contains rich attribute information. Node attributes include entity identifier, entity type, business attributes, and timestamp, while edge attributes include relationship type, relationship strength, relationship direction, and relationship weight.

6. The method according to claim 5, characterized in that, Graph index optimization strategies include three types: node attribute index, relation index, and composite index. Node attribute index is built based on the key attributes of nodes, relation index is built based on the type and weight of edges, and composite index combines multiple attributes to create a joint index.

7. The method according to claim 6, characterized in that, The tiered storage architecture stores hot data in the memory tier, warm data in the solid-state drive tier, and cold data in the hard disk drive tier, dynamically adjusting the storage tiers based on data access frequency.

8. The method according to claim 7, characterized in that, The query performance optimization vector includes five metrics: index hit rate, cache hit rate, query response time, concurrency processing capability, and resource utilization. These metrics are used to describe the effectiveness of query optimization strategies, evaluate the optimization results, and complete the configuration for query performance optimization.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores program instructions, which, when executed in a computer, are used to perform the digital mainline processing method based on industrial data maps as described in any one of claims 1-8.

10. A digital mainline processing system based on industrial data maps, characterized in that, The system includes the computer-readable storage medium of claim 9, wherein the system is any one of a computer, a server, or a microcontroller, the computer-readable storage medium is disposed within the system, and the system is provided with a microprocessor that executes the program instructions stored in the computer-readable storage medium.

Images