Structured processing method and system for multi-source heterogeneous industrial data

Abstract

The invention provides a structured processing method and system for multi-source heterogeneous industrial data. The method belongs to the technical field of industrial big data management, semantic interoperation and intelligent information fusion. The method comprises the steps of constructing a hierarchical industrial ontology library, generating a multi-dimensional ontology model, performing dynamic semantic annotation on industrial equipment based on the ontology model, and generating an equipment semantic fingerprint library; performing initial semantic association on the original industrial data source according to the equipment semantic fingerprint database to generate an original data stream with an ontology label; by constructing a hierarchical industrial ontology library and combining an equipment operation state and a task target, the method can map multi-source heterogeneous industrial data into computable and associable knowledge graph nodes in real time, greatly improves the semantic standardization degree of the industrial data, enables the data to be converted into information with deep semantics from an original character string, and improves the semantic standardization degree of the industrial data. And powerful support is provided for accurate decision making.

Description

Technical Field

[0001] This invention proposes a structured processing method and system for multi-source heterogeneous industrial data, belonging to the fields of industrial big data governance, semantic interoperability and intelligent information fusion technology. Background Technology

[0002] In the current era of rapid development in intelligent manufacturing, industrial data serves as a core driving force, and its quality and availability directly affect the accuracy and efficiency of production decisions. However, the current structured processing of industrial data faces numerous challenges.

[0003] Traditional methods often employ static schema binding, pre-defining fixed data structures, which struggle to adapt to semantic changes in the same equipment across different production lines and processes in industrial scenarios. For example, "temperature" is a critical parameter in reflow soldering, directly affecting soldering quality, while in a conveyor belt scenario it may only serve as a status monitoring indicator. This difference makes static schemas inflexible.

[0004] Meanwhile, existing technologies only scratch the surface of semantic understanding, relying solely on keyword matching or regular expression extraction, neglecting the deep relationships between industrial entities. Complex causal chains such as "motor overheating, conveyor belt shutdown, and upstream material accumulation" are difficult to capture accurately.

[0005] Furthermore, the structured process is disconnected from the task objectives, making it impossible to enhance semantics as needed for different tasks. Quality analysis requires extracting "tolerance bands," while energy efficiency analysis requires associating "power factors," making it difficult for traditional methods to achieve precise adaptation.

[0006] While some solutions incorporate the OWL ontology, these are primarily used for offline annotation and lack integration with real-time data streams and device operating status, thus hindering dynamic reasoning. Therefore, a novel structured approach with industrial ontology-driven, context-aware, task-adaptive, and dynamic knowledge generation capabilities is urgently needed to enhance the value of industrial data. Summary of the Invention

[0007] This invention provides a structured processing method and system for multi-source heterogeneous industrial data to solve the problems mentioned in the background section above:

[0008] This invention proposes a structured processing method for multi-source heterogeneous industrial data, the method comprising:

[0009] S1. Construct a hierarchical industrial ontology library, generate a multi-dimensional ontology model, perform dynamic semantic annotation on industrial equipment based on the ontology model, and generate an equipment semantic fingerprint library; perform initial semantic association on the original industrial data source according to the equipment semantic fingerprint library, and generate an original data stream with ontology tags.

[0010] S2. Combining real-time device operating status data with task analysis objectives, perform context-aware parsing on the raw data stream with ontology tags; dynamically identify key semantic elements in the current scenario through the device status-task objective joint reasoning engine, and generate context-enhanced semantic units; adaptively adjust the semantic granularity of the context-enhanced semantic units to generate computable semantic nodes;

[0011] S3. Construct a real-time knowledge graph based on computable semantic nodes, realize dynamic association between nodes through an ontology reasoning rule engine, perform conflict detection and consistency verification on the knowledge graph, and generate a verified knowledge graph; iteratively optimize the verified knowledge graph through an ontology self-evolution mechanism to generate a dynamically evolving knowledge graph.

[0012] S4. Based on the task analysis objectives, perform semantic service mapping on the dynamic evolutionary knowledge graph to generate a subgraph structure oriented towards specific tasks. Perform computable transformation on the subgraph structure to generate a structured data interface that supports machine learning model calls. Input the knowledge graph node data into the task analysis model through the structured data interface to generate task execution result data.

[0013] S5. Based on the task execution result data, perform feedback optimization on the hierarchical industrial ontology library to generate an optimized ontology model; based on the optimized ontology model, perform more accurate semantic annotation and structuring processing on subsequent original industrial data sources to generate highly available industrial structured data.

[0014] This invention proposes a system for implementing the structured processing method for multi-source heterogeneous industrial data as described above, the system comprising:

[0015] Semantic association module: Constructs a hierarchical industrial ontology library, generates a multi-dimensional ontology model, performs dynamic semantic annotation on industrial equipment based on the ontology model, and generates a semantic fingerprint library of equipment; performs initial semantic association on the original industrial data source based on the semantic fingerprint library of equipment, and generates an original data stream with ontology tags;

[0016] Element Adjustment Module: Combining real-time device operating status data with task analysis objectives, it performs context-aware parsing on the raw data stream with ontology tags; through the device status-task objective joint reasoning engine, it dynamically identifies key semantic elements in the current scenario and generates context-enhanced semantic units; it adaptively adjusts the semantic granularity of the context-enhanced semantic units to generate computable semantic nodes;

[0017] Iterative optimization module: Constructs a real-time knowledge graph based on computable semantic nodes, realizes dynamic association between nodes through an ontology reasoning rule engine, performs conflict detection and consistency verification on the knowledge graph, and generates a verified knowledge graph; iteratively optimizes the verified knowledge graph through an ontology self-evolution mechanism to generate a dynamically evolving knowledge graph.

[0018] Data Input Module: Based on the task analysis objectives, the module performs semantic service mapping on the dynamic evolutionary knowledge graph, generates a subgraph structure oriented towards specific tasks, performs computable transformation on the subgraph structure, and generates a structured data interface that supports machine learning model calls; the module then inputs knowledge graph node data into the task analysis model through the structured data interface to generate task execution result data.

[0019] Feedback Optimization Module: Based on the task execution result data, the layered industrial ontology library is optimized to generate an optimized ontology model; based on the optimized ontology model, subsequent original industrial data sources are subjected to more accurate semantic annotation and structuring processing to generate highly available industrial structured data.

[0020] The beneficial effects of this invention are as follows: By constructing a hierarchical industrial ontology library and combining it with equipment operating status and task objectives, this method can map multi-source heterogeneous industrial data into computable and associative knowledge graph nodes in real time. This significantly improves the semantic standardization of industrial data, transforming data from raw strings into information with deep semantics, providing strong support for accurate decision-making. Simultaneously, the semantic granularity adaptation and ontology self-evolution mechanism reduce the risk of data structure failure due to changes in industrial scenarios, enhancing the system's adaptability to complex industrial environments. During processing, context-aware parsing and dynamic reasoning reduce human intervention, avoiding information misjudgment caused by shallow semantic understanding and contextual disconnect in traditional methods. It can enhance key semantics as needed according to different tasks and continuously optimize the ontology model through closed-loop feedback, avoiding the semantic rigidity problem in static solutions. This significantly improves the usability and decision-making value of industrial data, providing a new generation of intelligent data governance paradigm for intelligent manufacturing. Attached Figure Description

[0021] Figure 1 This is a diagram illustrating the steps of the method described in this invention;

[0022] Figure 2 This is a system module diagram of the present invention. Detailed Implementation

[0023] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0024] One embodiment of the present invention, such as Figure 1 As shown, a structured processing method for multi-source heterogeneous industrial data is provided, the method comprising:

[0025] S1. Construct a hierarchical industrial ontology library and generate a multi-dimensional ontology model, which includes an equipment entity layer, a process association layer, and a task target layer; perform dynamic semantic annotation on industrial equipment based on the ontology model to generate an equipment semantic fingerprint library; perform initial semantic association on the original industrial data source according to the equipment semantic fingerprint library to generate an original data stream with ontology tags, which includes logs, alarms, and sensor streams.

[0026] S2. Combining real-time equipment operating status data with task analysis objectives, perform context-aware parsing on the raw data stream with ontology tags; dynamically identify key semantic elements in the current scenario through the equipment status-task objective joint reasoning engine, including tolerance band in the quality analysis scenario and power factor in the energy efficiency analysis scenario, and generate context-enhanced semantic units; adaptively adjust the semantic granularity of the context-enhanced semantic units to generate computable semantic nodes;

[0027] S3. Construct a real-time knowledge graph based on computable semantic nodes, and realize dynamic association between nodes through an ontology reasoning rule engine, such as causal chain reasoning for motor overheating, conveyor belt shutdown, and upstream material stacking; perform conflict detection and consistency verification on the knowledge graph to generate a verified knowledge graph; iteratively optimize the verified knowledge graph through an ontology self-evolution mechanism to generate a dynamically evolving knowledge graph, wherein the ontology self-evolution mechanism includes ontology expansion when new equipment is connected, relationship updates when the process changes, and weight adjustments when the task changes.

[0028] S4. Based on the task analysis objectives, perform semantic service mapping on the dynamic evolutionary knowledge graph to generate a subgraph structure oriented towards specific tasks, including fault prediction, quality traceability, and energy efficiency optimization; perform a computable transformation on the subgraph structure to generate a structured data interface that supports machine learning model calls; input the knowledge graph node data into the task analysis model through the structured data interface to generate task execution result data;

[0029] S5. Based on the task execution result data, the hierarchical industrial ontology library is optimized to generate an optimized ontology model. Based on the optimized ontology model, subsequent original industrial data sources are subjected to more accurate semantic annotation and structured processing to form an intelligent data governance paradigm that enhances the data, knowledge, and decision-making closed loop, and generates highly available industrial structured data.

[0030] The working principle and effects of the above technical solution are as follows: By constructing a hierarchical industrial ontology library and dynamic semantic annotation, the accuracy of semantic association of original industrial data sources is improved, avoiding data processing chaos caused by label conflicts of heterogeneous data sources. Combined with context-aware parsing of equipment status and task objectives, the scenario adaptability of key semantic element identification is enhanced, reducing interference from invalid data on the processing flow. A real-time knowledge graph is constructed based on computable semantic nodes and dynamic associations are achieved, improving the logic and analytical depth of data associations and avoiding industrial scenario analysis biases caused by broken causal relationships. Through ontology self-evolution and closed-loop feedback optimization, the adaptability of the ontology model is continuously improved, enhancing the stability and accuracy of structured industrial data processing. Simultaneously, the structured data interface is adapted to machine learning models, improving the efficiency of task analysis and avoiding model call failures caused by data format incompatibility, thereby improving the overall effectiveness of industrial data governance and ensuring that the generated industrial structured data has high availability and high adaptability.

[0031] In one embodiment of the present invention, S1 includes:

[0032] S11. Integrate industrial internet platform equipment metadata, industry process standards, and enterprise business process specifications to construct a layered industrial ontology library and generate a multi-dimensional ontology model. The multi-dimensional ontology model includes an equipment entity layer, a process association layer, and a task target layer. The equipment entity layer includes equipment model, parameter specifications, and interface protocols. The process association layer includes process logic, material flow, and quality standards. The task target layer includes fault diagnosis, energy efficiency optimization, and production scheduling.

[0033] S12. Based on a multi-dimensional ontology model, combined with real-time operation logs and historical maintenance data of industrial equipment, a dynamic semantic annotation algorithm is used to perform full lifecycle semantic annotation on industrial equipment to generate a device semantic fingerprint database. The dynamic semantic annotation algorithm integrates a hybrid encoding method of TF-IDF and Word2Vec. The device semantic fingerprint database includes the device's unique identifier, functional attributes, association relationships, and operational characteristics.

[0034] S13. Based on the tag system of the equipment semantic fingerprint library, perform cross-source initial semantic association on the original industrial data source. The original industrial data source includes equipment operation logs, fault alarm information, sensor real-time data streams, and process execution records. Resolve the tag conflicts of heterogeneous data sources through semantic similarity calculation to generate the original data stream with ontology tags.

[0035] S14. Perform data quality preprocessing on the raw data stream with ontology tags to generate a quality-compliant raw data stream with ontology tags.

[0036] The working principle and effects of the above technical solution are as follows: By integrating multiple standards to construct a hierarchical industrial ontology library, the generated multi-dimensional ontology model is more closely aligned with actual industrial scenarios, enhancing the basic adaptability of subsequent semantic processing and avoiding annotation biases caused by an incomplete ontology framework. Dynamic semantic annotation is performed by combining real-time operation logs and historical maintenance data, improving the accuracy of equipment feature characterization, making the equipment semantic fingerprint library more identifiable, and reducing the problems of missing or mislabeled equipment features. Cross-source semantic association resolves label conflicts from heterogeneous data sources, improving the consistency of semantic association of raw data and avoiding data association errors caused by label chaos. Data quality preprocessing further filters qualified data, reducing the interference of invalid data on subsequent processes and preventing unqualified data from entering subsequent stages and affecting processing results. Overall, this lays a solid data foundation for the structured processing of industrial data and improves the reliability and standardization of early data processing.

[0037] In one embodiment of the present invention, S13 includes:

[0038] Extract the core elements of the tag system from the semantic fingerprint database of equipment and generate a standardized tag set; classify and sort the original industrial data sources, distinguish the structural characteristics of different types of data sources, and generate a classified set of original data sources.

[0039] The standardized label set is matched and associated with the original data source set for classification to generate an initial cross-source semantic association candidate set; label conflicts are detected in the initial cross-source semantic association candidate set to generate a conflict label set.

[0040] Perform semantic similarity calculation on the conflicting tag set to resolve tag conflicts between heterogeneous data sources and generate a set of related tags after conflict resolution;

[0041] Integrate the conflict-resolved association set with the conflict-free initial association data to generate a raw data stream with ontology labels.

[0042] The working principle and effects of the above technical solution are as follows: By extracting a standardized set of tags and classifying and organizing the original industrial data sources, the matching and association between tags and data becomes more targeted, improving the initial accuracy of cross-source semantic association and avoiding association deviations caused by blind matching. It actively detects tag conflicts in the initial candidate set of cross-source semantic associations, and then accurately resolves contradictions through semantic similarity calculation, effectively reducing tag conflicts between heterogeneous data sources and preventing data semantic chaos caused by conflicting tags being introduced into subsequent processes. Finally, it integrates conflict-resolved and conflict-free data to generate a raw data stream with ontology tags, further improving the consistency and reliability of data semantic association, laying a solid foundation for subsequent data quality preprocessing, and preventing the overall structured processing efficiency from being affected by improper early semantic association.

[0043] In one embodiment of the present invention, S2 includes:

[0044] S21. Collect real-time operating status data of industrial equipment and current task analysis target, use both as context input, perform context-aware parsing on the original data stream with ontology tag that meets quality compliance, and extract the time sequence features, operating condition features and task association features in the data stream. The real-time operating status data includes sensor parameters such as temperature, pressure and speed; the current task analysis target includes fault prediction, quality traceability and energy efficiency optimization.

[0045] S22. Based on the multi-dimensional features obtained from the analysis, the device status and task target joint reasoning engine is used to dynamically identify key semantic elements in the current business scenario. The key semantic elements include tolerance band threshold and size deviation in the quality analysis scenario, power factor and energy consumption fluctuation rate in the energy efficiency analysis scenario, and vibration frequency and temperature threshold in the fault diagnosis scenario, generating context-enhanced semantic units.

[0046] S23. Based on the requirements of task complexity and data processing efficiency, the semantic granularity of the context-enhanced semantic units is adaptively adjusted. The adaptive adjustment includes fine-grained splitting for complex tasks and coarse-grained merging for efficient processing scenarios. Units are merged and split using semantic clustering algorithms (such as K-Means++) to generate computable semantic nodes with unified structure and adapted granularity.

[0047] S24. Perform attribute completion and relationship definition on the computable semantic nodes, determine the logical association between nodes, and the logical association is the mapping relationship between equipment parameters and process indicators, fault phenomena and fault causes, and generate a standardized set of computable semantic nodes.

[0048] The working principle and effects of the above technical solution are as follows: Context-aware analysis is performed by combining real-time operating status and task objectives, making feature extraction more aligned with business scenarios and improving the accuracy of time-series, working condition, and task-related feature extraction. This avoids invalid features that are detached from actual needs from interfering with subsequent processes. Dynamic identification of key semantic elements through a joint inference engine enhances the scenario adaptability of element recognition and reduces the redundant accumulation of irrelevant semantic information. Adaptive adjustment of semantic granularity flexibly matches different task complexity and processing efficiency requirements, satisfying both the fine-grained analysis requirements of complex tasks and ensuring the response speed of efficient processing scenarios, avoiding inefficient processing or missing core information caused by uniform granularity. Finally, attribute completion and relationship definition are used to organize node logic and generate a standardized, computable semantic node set. This avoids node association confusion or missing attributes affecting the accuracy of subsequent knowledge graph construction, improving the overall reliability and adaptability of the semantic processing stage and laying a solid foundation for subsequent knowledge construction.

[0049] In one embodiment of the present invention, step S23 includes:

[0050] Feature extraction is performed on context-enhanced semantic units to generate semantic unit feature sets, which include semantic dimension features, semantic association strength features, and data volume features. Combining task complexity and data processing efficiency requirements, the semantic unit feature sets are quantitatively analyzed to generate semantic granularity adjustment threshold ranges.

[0051] Based on the semantic granularity adjustment threshold range, fine-grained adjustment range and coarse-grained adjustment range are divided to generate semantic granularity adjustment division rules;

[0052] Based on the semantic granularity adjustment partitioning rules, the context-enhanced semantic units are classified to generate a fine-grained set of units to be adjusted and a coarse-grained set of units to be adjusted.

[0053] Semantic decomposition is performed on the fine-grained set of units to be adjusted to generate fine-grained semantic unit fragments; semantic merging is performed on the coarse-grained set of units to be adjusted to generate coarse-grained semantic unit aggregates.

[0054] A semantic clustering algorithm is used to integrate and optimize fine-grained semantic unit fragments and coarse-grained semantic unit aggregates to generate preliminary semantic nodes with a unified structure; the granularity adaptability of the preliminary semantic nodes is verified to generate granularity adaptability verification results.

[0055] Based on the granularity adaptation verification results, the preliminary semantic nodes that failed the verification are re-adjusted, and the preliminary semantic nodes that passed the verification are finalized to generate granularity-adapted computable semantic nodes.

[0056] The working principle and effects of the above technical solution are as follows: By extracting multi-dimensional features of semantic units and performing quantitative analysis in conjunction with task and efficiency requirements, the semantic granularity adjustment threshold is made more aligned with actual needs, improving the rationality of adjustment rules and avoiding adjustment deviations caused by blindly setting thresholds. Semantic units are processed according to rules, accurately distinguishing between fine-grained and coarse-grained adjustment ranges. Splitting and merging operations are more targeted, meeting the needs of complex tasks for fine-grained semantic analysis while ensuring speed in efficient processing scenarios, reducing inefficiencies or omissions of core information caused by improper granularity adaptation. Semantic clustering and integration allows units of different granularities to form structurally unified initial nodes, avoiding fragmentation or chaotic aggregation. Subsequent adaptability verification further filters qualified nodes, and those that fail are readjusted, improving the adaptation accuracy of the final computable semantic nodes and preventing unqualified nodes from entering subsequent processes and affecting the quality and efficiency of knowledge graph construction.

[0057] In one embodiment of the present invention, S3 includes:

[0058] S31. Based on a standardized set of computable semantic nodes, a real-time knowledge graph is constructed using a graph database storage architecture (such as Neo4j). Dynamic association mining between nodes is performed through an ontology reasoning rule engine (based on the SWRL rule language), including causal chain reasoning (such as cooling system failure, spindle temperature exceeding the limit, and decreased machining accuracy) and association rule reasoning (such as equipment load rate > 80% and running time > 4 hours to lubricating oil consumption rate increase by 30%), generating an initial association knowledge graph.

[0059] S32. Perform conflict detection and consistency verification on the initial associated knowledge graph, using multi-dimensional verification indicators, including semantic consistency, logical rationality, and data accuracy. Process contradictory nodes and redundant relationships through conflict resolution algorithms, such as priority-based conflict resolution strategies; generate the verified knowledge graph.

[0060] S33. Activate the ontology self-evolution mechanism. For new equipment access scenarios, automatically expand the equipment entity layer and association relationships in the ontology library; for process change scenarios, update the logical relationships and weight parameters of the process association layer in real time; for task change scenarios, dynamically adjust the priority and association rules of the task target layer and generate a preliminary evolutionary knowledge graph.

[0061] S34. Based on the operational feedback data of the preliminary evolutionary knowledge graph, the operational feedback data includes semantic matching accuracy, inference efficiency, and task suitability. The structure and parameters of the ontology model are iteratively optimized using a reinforcement learning algorithm to generate a dynamic evolutionary knowledge graph.

[0062] The working principle and effects of the above technical solution are as follows: A real-time knowledge graph is constructed through standardized computable semantic nodes. Combined with dynamic association mining, causal chains and association rules are accurately identified, improving the logic and depth of node associations and avoiding scenario analysis biases caused by broken associations between industrial data. Multi-dimensional conflict detection and consistency verification effectively identify and resolve contradictory nodes and redundant relationships, reducing logical confusion in the knowledge graph and preventing erroneous associations from affecting subsequent analysis results. The ontology self-evolution mechanism can flexibly adapt to scenarios such as new equipment access and process changes, automatically expanding and updating the graph content to prevent the graph from becoming invalid due to changes in industrial scenarios. Iterative optimization based on operational feedback data further improves the graph's reasoning efficiency and task adaptability, and the generated dynamically evolving knowledge graph can continuously match the needs of industrial scenarios. Overall, the reliability and timeliness of knowledge construction are enhanced, providing high-quality knowledge support for subsequent task analysis and reducing analysis errors caused by outdated or incomplete knowledge systems.

[0063] In one embodiment of the present invention, step S31 includes:

[0064] Feature extraction and redundancy filtering are performed on the standardized computable semantic node set to remove invalid and duplicate nodes and generate a cleaned computable semantic node set; based on the cleaned computable semantic node set, the graph database storage architecture is initialized, the node storage mode and relation index structure are configured, and the graph storage framework is generated.

[0065] The logic of node association in industrial scenarios is analyzed, semantic matching rules and inference constraints are formulated, and an association mining rule set is generated. The purified computable semantic node set and association mining rule set are input into the ontology inference rule engine to start the node association mining process and generate an initial node association set.

[0066] The initial node association set is classified and processed, and causal chain mining and association rule mining are performed respectively to generate causal association subsets and rule association subsets;

[0067] By integrating causal association subsets, rule association subsets, and a graph storage framework, a preliminary knowledge graph with a complete mapping between nodes and associations is constructed. The preliminary knowledge graph is then subjected to association integrity verification, missing associations are supplemented, and an initial association knowledge graph is generated.

[0068] The working principle and effects of the above technical solution are as follows: By extracting features and filtering redundancy from a standardized set of computable semantic nodes, invalid and duplicate nodes are removed, improving the purity of the node set, reducing the interference of redundant data on subsequent processes, and avoiding biases in association mining caused by invalid nodes. Initializing the graph database storage architecture and configuring a reasonable storage and index structure makes the graph storage more standardized, improving the efficiency of subsequent data queries and association calls, and avoiding processing bottlenecks caused by a chaotic storage architecture. Analyzing the node association logic in industrial scenarios and formulating mining rules makes association mining more directional, improving the accuracy of node associations, and avoiding invalid associations generated by blind mining. Categorizing the initial node association set, clarifying the different dimensions of causal chains and association rules, enhancing the logic and orderliness of association relationships, and avoiding analytical confusion caused by mixed types of associations. Finally, merging the association subset with the storage framework and verifying its completeness, supplementing missing associations, improving the completeness and reliability of the initial association knowledge graph, laying a solid foundation for subsequent conflict detection and optimization, and reducing the impact of missing associations on the usability of the knowledge graph.

[0069] In one embodiment of the present invention, step S4 includes:

[0070] S41. Based on the specific task analysis objectives, such as fault prediction, quality traceability, energy efficiency optimization, and production scheduling, semantic service mapping is performed on the dynamic evolutionary knowledge graph. Nodes and relationships strongly related to the task are filtered through a subgraph extraction algorithm. The subgraph extraction algorithm is a subgraph pruning strategy based on breadth-first search, generating a subgraph structure oriented towards the specific task.

[0071] S42. Perform a computable transformation on the subgraph structure, including data format standardization, feature engineering encapsulation, and dimensionality reduction optimization. The data format standardization is used to convert it into machine learning model compatible formats such as JSON and CSV. The feature engineering encapsulation is used to extract node attribute features, relationship features, and time series features. The dimensionality reduction optimization is used to reduce the data dimensionality using algorithms such as PCA and LDA to generate a structured data interface that supports machine learning model calls.

[0072] S43. Perform secondary quality verification on the input data through the structured data interface. The secondary quality verification includes verification of data integrity, format standardization, and feature validity. Input the knowledge graph node data that meets the requirements into the corresponding task analysis model. For example, fault prediction uses an LSTM model, quality traceability uses a graph neural network model, and energy efficiency optimization uses a reinforcement learning model.

[0073] S44. Based on the output of the task analysis model, the results are analyzed and optimized in combination with the business rules of the industrial scenario. The business rules of the industrial scenario include fault level judgment criteria, quality qualification threshold and energy efficiency optimization target, to generate task execution result data that accurately matches business needs.

[0074] The working principle and effects of the above technical solution are as follows: By extracting relevant subgraphs based on specific task objectives, nodes and relationships strongly related to the task are accurately selected, reducing redundant interference from irrelevant data and avoiding inefficient processing or obscuring of core information due to complex data. The subgraphs undergo computable transformation, achieving data format standardization, feature encapsulation, and dimensional optimization, making the data more compatible with machine learning models and avoiding model call failures caused by format incompatibility. Simultaneously, reducing data dimensionality alleviates the computational burden on the model, improving analysis efficiency. Secondary data quality verification filters out unqualified data, preventing erroneous or invalid data from entering the model and affecting analysis results, ensuring the reliability of input data. Combining industrial scenario business rule parsing to optimize model output makes task execution results more aligned with actual business needs, improving the accuracy and practicality of the results and avoiding a disconnect between theoretical analysis and practical application. The overall process achieves efficient transformation from knowledge graphs to task results, enhancing the relevance and reliability of task analysis and providing precise data support for industrial decision-making.

[0075] In one embodiment of the present invention, S43 includes:

[0076] The knowledge graph node data to be input is extracted through a structured data interface to generate an unverified data set; verification dimensions are constructed based on data integrity, format standardization, and feature validity to generate a multi-dimensional verification rule system.

[0077] Based on a multi-dimensional verification rule system, each item of the unverified dataset is verified to generate data verification results and problematic data identifiers; problematic data in the data verification results are filtered out, and data that meets the verification standards is retained to generate a qualified dataset.

[0078] The input data requirements for various task analysis models are reviewed, and model input adaptation specifications are generated. The data format and features of qualified datasets are adjusted according to the model input adaptation specifications to generate model adaptation data.

[0079] The model-adapted data is then input into the matching task analysis model to complete the data input process.

[0080] The working principle and effects of the above technical solution are as follows: By extracting the input data and constructing a multi-dimensional verification rule system, data verification becomes more comprehensive and detailed, improving the accuracy of identifying problematic data and avoiding the omission of potential data defects by single-dimensional verification. Verifying and identifying problematic data item by item based on rules accurately filters out unqualified data, reducing the entry of erroneous, incomplete, or abnormally formatted data into subsequent processes, and preventing such problematic data from interfering with model analysis results. The model input requirements are streamlined and adapted to ensure that qualified data accurately matches the needs of various task analysis models, avoiding model lag or analysis bias caused by incompatible data formats or features. Finally, the adapted data is input into the matching model, ensuring the smoothness and accuracy of the data input process, improving the overall reliability of model input data, laying a solid foundation for the accuracy of subsequent task analysis results, and reducing task analysis failures or result distortions caused by data problems.

[0081] In one embodiment of the present invention, step S5 includes:

[0082] S51. Based on the task execution result data, analyze the shortcomings of the current hierarchical industrial ontology library through the effect evaluation index system, and extract the optimization requirements of the ontology library. The effect evaluation index system includes data processing accuracy, task execution efficiency, and decision support value. The shortcomings of the current hierarchical industrial ontology library include insufficient semantic tag accuracy, missing ontology relationships, and poor task adaptability.

[0083] S52. Based on the optimization requirements, structural adjustments and parameter updates are performed on the equipment entity layer, process association layer, and task target layer of the hierarchical industrial ontology library. The structural adjustments and parameter updates include adding new equipment types, improving process logic relationships, optimizing task weight allocation, and generating an optimized ontology model.

[0084] S53. The optimized ontology model is used to perform more accurate semantic annotation, context-aware parsing and knowledge graph construction on the raw industrial data sources that are subsequently connected. The data processing effect is tracked in real time through a closed-loop monitoring mechanism, and the processing parameters are dynamically adjusted.

[0085] S54. Continuously iterate the data access, semantic processing, knowledge construction, task execution, and ontology optimization processes to form an intelligent data governance paradigm that enhances the closed loop of data, knowledge, and decision-making, generating highly available, highly adaptable, and highly accurate industrial structured data.

[0086] The working principle and effects of the above technical solution are as follows: By analyzing the deficiencies of the ontology library based on task execution results, problems in semantic tags, ontology relationships, and task adaptation can be accurately located, avoiding resource waste or inadequate optimization caused by blind optimization. Targeted adjustments and updates to the structure and parameters of each level of the ontology library enhance the adaptability of the ontology model and reduce issues such as semantic annotation bias and missing relationships. The optimized model is used to process subsequent data, and parameters are dynamically adjusted in conjunction with closed-loop monitoring to improve the accuracy of semantic annotation, parsing, and knowledge graph construction, avoiding unstable processing results. Continuous iteration forms a closed-loop governance paradigm, allowing the data processing process to continuously improve itself. This ensures both the high quality of current data processing and adaptability to the dynamic changes in industrial scenarios, improving the usability, adaptability, and accuracy of industrial structured data, avoiding performance degradation caused by a lagging ontology library, and providing a guarantee for long-term stable industrial data governance.

[0087] In one embodiment of the present invention, S53 includes:

[0088] Collect raw industrial data sources for subsequent access and generate a set of raw data sources to be processed; use the optimized ontology model to perform semantic annotation on the set of raw data sources to be processed and generate annotated raw data sources.

[0089] Context-aware parsing is performed based on the annotated original data source to generate parsed semantic data; knowledge graph construction is advanced based on the parsed semantic data to generate an updated knowledge graph;

[0090] A closed-loop monitoring mechanism is initiated to collect processing indicators such as semantic annotation accuracy, parsing accuracy, and graph construction completeness, and generate data processing effect monitoring data; the data processing effect monitoring data is analyzed to determine the deviation between the current processing effect and the expected standard, and effect deviation analysis results are generated.

[0091] Based on the results of the performance deviation analysis, the processing parameters such as semantic annotation threshold, parsing rules, and graph construction parameters are dynamically adjusted to complete parameter optimization and adaptation.

[0092] The working principle and effects of the above technical solution are as follows: Using an optimized ontology model to process subsequent raw data significantly improves the accuracy of semantic annotation, making the parsed semantic data more aligned with actual business scenarios and reducing annotation bias and parsing errors caused by insufficient initial ontology model data. The updated knowledge graph built upon accurately parsed data has stronger completeness and reliability, avoiding issues such as missing relationships or logical inconsistencies. A closed-loop monitoring mechanism collects multiple processing indicators in real time, promptly capturing deviations in annotation accuracy, parsing accuracy, etc., preventing the accumulation of problems from affecting subsequent task execution. Processing parameters are dynamically adjusted based on deviation analysis results, allowing the data processing workflow to continuously adapt to changes in industrial data and scenarios, improving the stability of processing effects, avoiding the decline in adaptability caused by parameter solidification, providing solid support for the continuous optimization of the entire closed-loop governance paradigm, and ensuring the high-quality advancement of subsequent data processing.

[0093] One embodiment of the present invention, such as Figure 2 As shown, a system for implementing the structured processing method for multi-source heterogeneous industrial data as described above is provided, the system comprising:

[0094] Semantic association module: Constructs a hierarchical industrial ontology library and generates a multi-dimensional ontology model, which includes an equipment entity layer, a process association layer, and a task target layer; performs dynamic semantic annotation on industrial equipment based on the ontology model to generate an equipment semantic fingerprint library; performs initial semantic association on the original industrial data source according to the equipment semantic fingerprint library to generate an original data stream with ontology tags, which includes logs, alarms, and sensor streams.

[0095] Element Adjustment Module: Combining real-time equipment operating status data with task analysis objectives, the module performs context-aware parsing on the raw data stream with ontology tags; through the joint inference engine of equipment status and task objectives, it dynamically identifies key semantic elements in the current scenario, including tolerance band in the quality analysis scenario and power factor in the energy efficiency analysis scenario, and generates context-enhanced semantic units; the module then adaptively adjusts the semantic granularity of the context-enhanced semantic units to generate computable semantic nodes.

[0096] Iterative optimization module: Constructs a real-time knowledge graph based on computable semantic nodes, and realizes dynamic association between nodes through an ontology reasoning rule engine, such as causal chain reasoning for motor overheating, conveyor belt shutdown, and upstream material accumulation; performs conflict detection and consistency verification on the knowledge graph to generate a verified knowledge graph; iteratively optimizes the verified knowledge graph through an ontology self-evolution mechanism to generate a dynamically evolving knowledge graph. The ontology self-evolution mechanism includes ontology expansion when new equipment is connected, relationship updates when processes change, and weight adjustments when tasks change.

[0097] Data Input Module: Based on the task analysis objectives, the module performs semantic service mapping on the dynamic evolutionary knowledge graph to generate subgraph structures oriented towards specific tasks, including fault prediction, quality traceability, and energy efficiency optimization; it performs computable transformation on the subgraph structures to generate structured data interfaces that support machine learning model calls; and it inputs knowledge graph node data into the task analysis model through the structured data interfaces to generate task execution result data.

[0098] Feedback Optimization Module: Based on the task execution result data, the layered industrial ontology library is optimized to generate an optimized ontology model. Based on the optimized ontology model, subsequent original industrial data sources are subjected to more accurate semantic annotation and structured processing, forming an intelligent data governance paradigm that enhances the data, knowledge, and decision-making closed loop, and generating highly available industrial structured data.

[0099] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for structured processing of multi-source heterogeneous industrial data, characterized in that, The method includes: S1. Construct a hierarchical industrial ontology library, generate a multi-dimensional ontology model, perform dynamic semantic annotation on industrial equipment based on the ontology model, and generate an equipment semantic fingerprint library; perform initial semantic association on the original industrial data source according to the equipment semantic fingerprint library, and generate an original data stream with ontology tags. S2. Combining real-time device operating status data with task analysis objectives, perform context-aware parsing on the raw data stream with ontology tags; dynamically identify key semantic elements in the current scenario through the device status-task objective joint reasoning engine, and generate context-enhanced semantic units; adaptively adjust the semantic granularity of the context-enhanced semantic units to generate computable semantic nodes; S3. Construct a real-time knowledge graph based on computable semantic nodes, realize dynamic association between nodes through an ontology reasoning rule engine, perform conflict detection and consistency verification on the knowledge graph, and generate a verified knowledge graph; iteratively optimize the verified knowledge graph through an ontology self-evolution mechanism to generate a dynamically evolving knowledge graph. S4. Based on the task analysis objectives, perform semantic service mapping on the dynamic evolutionary knowledge graph to generate a subgraph structure oriented towards specific tasks. Perform computable transformation on the subgraph structure to generate a structured data interface that supports machine learning model calls. Input the knowledge graph node data into the task analysis model through the structured data interface to generate task execution result data. S5. Based on the task execution result data, perform feedback optimization on the hierarchical industrial ontology library to generate an optimized ontology model; based on the optimized ontology model, perform more accurate semantic annotation and structuring processing on subsequent original industrial data sources to generate highly available industrial structured data.

2. The method for structured processing of multi-source heterogeneous industrial data according to claim 1, characterized in that, S1 includes: S11. Integrate industrial internet platform equipment metadata, industry process standards, and enterprise business process specifications to build a hierarchical industrial ontology library and generate a multi-dimensional ontology model. S12. Based on a multi-dimensional ontology model, combined with real-time operation logs and historical maintenance data of industrial equipment, a dynamic semantic annotation algorithm is used to perform full lifecycle semantic annotation of industrial equipment to generate a semantic fingerprint database of equipment. S13. Based on the labeling system of the device semantic fingerprint database, perform cross-source initial semantic association on the original industrial data source, resolve the label conflict of heterogeneous data sources through semantic similarity calculation, and generate the original data stream with ontology label. S14. Perform data quality preprocessing on the raw data stream with ontology tags to generate a quality-compliant raw data stream with ontology tags.

3. The method for structured processing of multi-source heterogeneous industrial data according to claim 2, characterized in that, S13 includes: Extract the core elements of the tag system from the semantic fingerprint database of equipment and generate a standardized tag set; classify and sort the original industrial data sources, distinguish the structural characteristics of different types of data sources, and generate a classified set of original data sources. The standardized label set is matched and associated with the original data source set for classification to generate an initial cross-source semantic association candidate set; label conflicts are detected in the initial cross-source semantic association candidate set to generate a conflict label set. Perform semantic similarity calculation on the conflicting tag set to resolve tag conflicts between heterogeneous data sources and generate a set of related tags after conflict resolution; Integrate the conflict-resolved association set with the conflict-free initial association data to generate a raw data stream with ontology labels.

4. The method for structured processing of multi-source heterogeneous industrial data according to claim 1, characterized in that, The S2 includes: S21. Collect real-time operating status data of industrial equipment and current task analysis target, use both as context input, perform context-aware parsing on the original data stream with ontology label that meets quality compliance, and extract the time sequence features, operating condition features and task-related features in the data stream. S22. Based on the multi-dimensional features obtained from the analysis, the key semantic elements in the current business scenario are dynamically identified through the joint reasoning engine of device status and task objectives, and context-enhanced semantic units are generated. S23. Based on the requirements of task complexity and data processing efficiency, the semantic granularity of the context-enhanced semantic unit is adaptively adjusted. The unit is merged and split through the semantic clustering algorithm to generate computable semantic nodes with unified structure and adapted granularity. S24. Perform attribute completion and relationship definition on the computable semantic nodes, determine the logical associations between nodes, and generate a standardized set of computable semantic nodes.

5. The method for structured processing of multi-source heterogeneous industrial data according to claim 4, characterized in that, S23 includes: Feature extraction is performed on context-enhanced semantic units to generate semantic unit feature sets. Combining task complexity and data processing efficiency requirements, the semantic unit feature sets are quantitatively analyzed to generate semantic granularity adjustment threshold ranges. Based on the semantic granularity adjustment threshold range, fine-grained adjustment range and coarse-grained adjustment range are divided to generate semantic granularity adjustment division rules; Based on the semantic granularity adjustment partitioning rules, the context-enhanced semantic units are classified to generate a fine-grained set of units to be adjusted and a coarse-grained set of units to be adjusted. Semantic decomposition is performed on the set of fine-grained units to be adjusted to generate fine-grained semantic unit fragments; semantic merging is performed on the set of coarse-grained units to be adjusted to generate coarse-grained semantic unit aggregates. A semantic clustering algorithm is used to integrate and optimize fine-grained semantic unit fragments and coarse-grained semantic unit aggregates to generate preliminary semantic nodes with a unified structure; the granularity adaptability of the preliminary semantic nodes is verified to generate granularity adaptability verification results. Based on the granularity adaptation verification results, the preliminary semantic nodes that failed the verification are re-adjusted, and the preliminary semantic nodes that passed the verification are finalized to generate granularity-adapted computable semantic nodes.

6. The method for structured processing of multi-source heterogeneous industrial data according to claim 1, characterized in that, The S3 includes: S31. Based on a standardized set of computable semantic nodes, a real-time knowledge graph is constructed using a graph database storage architecture. An ontology reasoning rule engine is used to mine dynamic associations between nodes and generate an initial association knowledge graph. S32. Perform conflict detection and consistency verification on the initial associated knowledge graph, using multi-dimensional verification indicators and conflict resolution algorithms to handle contradictory nodes and redundant relationships; generate the verified knowledge graph. S33. Activate the ontology self-evolution mechanism. For new equipment access scenarios, automatically expand the equipment entity layer and association relationships in the ontology library; for process change scenarios, update the logical relationships and weight parameters of the process association layer in real time; for task change scenarios, dynamically adjust the priority and association rules of the task target layer and generate a preliminary evolutionary knowledge graph. S34. Based on the operational feedback data of the preliminary evolutionary knowledge graph, the structure and parameters of the ontology model are iteratively optimized using a reinforcement learning algorithm to generate a dynamic evolutionary knowledge graph.

7. The method for structured processing of multi-source heterogeneous industrial data according to claim 6, characterized in that, S31 includes: Feature extraction and redundancy filtering are performed on the standardized computable semantic node set to generate a cleaned computable semantic node set; based on the cleaned computable semantic node set, the graph database storage architecture is initialized, the node storage mode and relation index structure are configured, and the graph storage framework is generated. The logic of node association in industrial scenarios is analyzed, semantic matching rules and inference constraints are formulated, and an association mining rule set is generated. The purified computable semantic node set and association mining rule set are input into the ontology inference rule engine to start the node association mining process and generate an initial node association set. The initial node association set is classified and processed, and causal chain mining and association rule mining are performed respectively to generate causal association subsets and rule association subsets; By integrating causal association subsets, rule association subsets, and a graph storage framework, a preliminary knowledge graph with a complete mapping between nodes and associations is constructed. The preliminary knowledge graph is then subjected to association integrity verification, missing associations are supplemented, and an initial association knowledge graph is generated.

8. The method for structured processing of multi-source heterogeneous industrial data according to claim 1, characterized in that, The S4 includes: S41. Based on the specific task analysis objectives, perform semantic service mapping on the dynamic evolutionary knowledge graph, and use the subgraph extraction algorithm to filter nodes and relationships that are strongly related to the task to generate a subgraph structure oriented towards the specific task. S42. Perform a computable transformation on the subgraph structure to generate a structured data interface that supports machine learning model calls; S43. Perform secondary quality verification on the input data through the structured data interface, and input the knowledge graph node data that meets the requirements into the corresponding task analysis model; S44. Based on the output of the task analysis model, the results are analyzed and optimized in combination with the business rules of the industrial scenario to generate task execution result data that accurately matches business needs.

9. The method for structured processing of multi-source heterogeneous industrial data according to claim 1, characterized in that, The S5 includes: S51. Based on the task execution result data, analyze the shortcomings of the current hierarchical industrial ontology library through the effect evaluation index system, and extract the ontology library optimization requirements. S52. Based on the optimization requirements, the structure of the equipment entity layer, process association layer, and task target layer of the hierarchical industrial ontology library is adjusted and the parameters are updated to generate an optimized ontology model. S53. The optimized ontology model is used to perform semantic annotation, context-aware parsing and knowledge graph construction on the subsequently connected original industrial data sources. The data processing effect is tracked in real time through a closed-loop monitoring mechanism, and the processing parameters are dynamically adjusted. S54. The continuous iterative processing flow of data access, semantic processing, knowledge construction, task execution, and ontology optimization forms an intelligent data governance paradigm that enhances the closed loop of data, knowledge, and decision-making, generating industrial structured data.

10. A system for implementing the structured processing method for multi-source heterogeneous industrial data as described in claim 1, characterized in that, The system includes: Semantic association module: Constructs a hierarchical industrial ontology library, generates a multi-dimensional ontology model, performs dynamic semantic annotation on industrial equipment based on the ontology model, and generates a semantic fingerprint library of equipment; performs initial semantic association on the original industrial data source based on the semantic fingerprint library of equipment, and generates an original data stream with ontology tags; Element Adjustment Module: Combining real-time device operating status data with task analysis objectives, it performs context-aware parsing on the raw data stream with ontology tags; through the device status-task objective joint reasoning engine, it dynamically identifies key semantic elements in the current scenario and generates context-enhanced semantic units; it adaptively adjusts the semantic granularity of the context-enhanced semantic units to generate computable semantic nodes; Iterative optimization module: Constructs a real-time knowledge graph based on computable semantic nodes, realizes dynamic association between nodes through an ontology reasoning rule engine, performs conflict detection and consistency verification on the knowledge graph, and generates a verified knowledge graph; iteratively optimizes the verified knowledge graph through an ontology self-evolution mechanism to generate a dynamically evolving knowledge graph. Data Input Module: Based on the task analysis objectives, the module performs semantic service mapping on the dynamic evolutionary knowledge graph, generates a subgraph structure oriented towards specific tasks, performs computable transformation on the subgraph structure, and generates a structured data interface that supports machine learning model calls; the module then inputs knowledge graph node data into the task analysis model through the structured data interface to generate task execution result data. Feedback Optimization Module: Based on the task execution result data, the layered industrial ontology library is optimized to generate an optimized ontology model; based on the optimized ontology model, subsequent original industrial data sources are subjected to more accurate semantic annotation and structuring processing to generate highly available industrial structured data.

Images