A method and system for constructing a professional knowledge graph for a hot strip rolling process

Abstract

This invention provides a method and system for constructing a professional knowledge graph for the hot rolling process of strip steel, belonging to the field of industrial monitoring and intelligent manufacturing. The method first acquires multi-source heterogeneous data of the hot rolling process of strip steel, and transforms structured and semi-structured data into unstructured data before merging them into an unstructured dataset; constructs an ontology for the professional knowledge graph, including entity types and entity relationships; divides the unstructured data into training and testing sets; constructs a joint extraction model based on the CasRel framework and multi-hop attention mechanism, and performs training and testing; inputs all unstructured data into the model, identifies and extracts a set of structured triples, processes and stores them to obtain the basic knowledge graph; acquires hot rolling fault data and constructs a fault knowledge graph, and fuses them to obtain the professional knowledge graph for the hot rolling process of strip steel. This invention constructs a professional knowledge graph for the hot rolling process of strip steel, improving the systematic capture capability of hot rolling process data.

Description

Technical Field

[0001] This invention belongs to the field of industrial monitoring and intelligent manufacturing, specifically relating to a method and system for constructing a professional knowledge map for the hot rolling process of strip steel. Background Technology

[0002] Hot rolling of strip steel is a key process in steel production. It involves multiple stages such as heating furnace, roughing, finishing, laminar cooling and coiling. It is characterized by numerous equipment, severe data coupling and complex operating conditions. Each stage has a corresponding impact on the final strip steel performance. Therefore, it is necessary to monitor the hot rolling process throughout its entire life cycle.

[0003] The entire lifecycle of hot-rolled strip steel encompasses design, production, quality inspection, and operation and maintenance, continuously generating massive amounts of multi-source heterogeneous data. However, the current utilization of these data resources is significantly unbalanced and superficial: the use of structured data is mainly concentrated on real-time monitoring and post-event statistics, while the mining and utilization of the deep-level domain knowledge contained in semi-structured and unstructured data is severely insufficient. This results in the valuable expert experience and process knowledge contained within being difficult to capture systematically, manage in a standardized manner, and reuse effectively, and thus cannot be transformed into calculable and reusable digital assets. Summary of the Invention

[0004] To address the aforementioned issues, this invention provides a method and system for constructing a professional knowledge graph for the hot rolling process of strip steel. This method enables the automated and integrated extraction and fusion of multi-source heterogeneous knowledge in the field of hot rolling of strip steel, resulting in a professional knowledge graph that reflects the entire life cycle of the hot rolling process. This achieves the reusability of knowledge in the hot rolling process and the traceability of faults, and forms a dynamically evolving fault propagation graph, providing knowledge support for intelligent diagnosis and decision-making in the production process.

[0005] To achieve the above objectives, the technical solutions adopted in the embodiments of the present invention are as follows:

[0006] In a first aspect, embodiments of the present invention provide a method for constructing a professional knowledge graph for the hot rolling process of strip steel, the method comprising the following steps:

[0007] Step S1: Obtain multi-source heterogeneous data of the hot rolling process of strip steel; the multi-source heterogeneous data includes structured, semi-structured and unstructured data;

[0008] Step S2: Extract domain terms, variable names and value ranges from structured and semi-structured data, convert them into unstructured data, and merge them into the unstructured dataset.

[0009] Step S3: Based on the content of multi-source heterogeneous data, construct the ontology of professional knowledge graph, which includes entity types and entity relationships; divide the unstructured data into training set and test set, and classify and label the unstructured data in the training set according to the ontology;

[0010] Step S4: Construct a joint extraction model based on the CasRel framework and multi-hop attention mechanism, and train and test the joint extraction model using the training set and test set; the joint extraction model uses a pre-trained language model as the encoder and a network structure integrating head entity recognition, multi-hop attention enhancement layer and relation-specific tail entity labeler as the decoder.

[0011] Step S5: Input all unstructured data into the trained joint extraction model, identify and extract the head entity, relation and tail entity of the unstructured data to form a structured set of triples.

[0012] Step S6: Perform data cleaning, entity alignment and conflict resolution on the structured triples of the extracted unstructured data to obtain unified, standardized and non-redundant structured triples of unstructured data, and store them to obtain the basic knowledge graph of the hot rolling process of strip steel.

[0013] Step S7: Obtain hot rolling fault data, and construct a fault knowledge graph representing the fault source, propagation path and constraint relationship based on the hot rolling fault data and the constraint relationship in the basic knowledge graph; merge the basic knowledge graph and the fault knowledge graph to obtain a professional knowledge graph for the hot rolling process of strip steel.

[0014] In a preferred embodiment of the present invention, the entity types in step S3 include transaction data, attribute data, process data, and constraint data, and the entity relationships include subordination, function, description, and specification.

[0015] In a preferred embodiment of the present invention, in step S4, the joint extraction model includes an input representation layer, a deep context encoding layer, a head entity recognizer, a multi-hop attention enhancement layer, and a relation-specific tail entity labeler.

[0016] In a preferred embodiment of the present invention, the joint extraction model performs deep contextual semantic encoding based on a pre-trained language model during the encoding stage; and during the decoding stage, the identified head entities are enhanced with entity representation through a multi-hop attention enhancement layer, and then a relation-specific tail entity annotator is used to achieve joint decoding of entities, attributes and relations.

[0017] In a preferred embodiment of the present invention, the training process of the joint extraction model includes:

[0018] Step S41: Load the pre-trained language model as the encoder; and randomly initialize the decoder parameters; convert the input text sequence into an embedding vector, and feed the embedding vector into the encoder for deep semantic encoding to obtain a context encoding sequence of the same length as the input sequence; in the context encoding sequence, a multi-head attention mechanism is used to enable each position to pay attention to all other positions in the sequence;

[0019] Step S42: Based on the context encoding sequence, identify all possible entities in the text through the head entity recognition layer and obtain the initial representation of each entity;

[0020] Step S43: Construct a fully connected graph for nodes using the initial representation and context encoding sequence. Then, use a K-layer graph attention network to propagate information in multiple rounds, so that the entity representation gradually integrates relevant context information from the global scope to obtain an enhanced entity representation.

[0021] Step S44: The tail entity annotator determines whether the corresponding entity participates in any relation or has any attribute based on the enhanced entity representation; for entities that are determined to have a relation or attribute, the start and end positions of the corresponding object entity or attribute value in the text are further decoded, thereby extracting the complete triples simultaneously.

[0022] Step S45: Based on the difference between the triples and the corresponding real annotations, the total loss of the joint extraction model is defined to include entity recognition loss, relation classification loss and object extraction loss.

[0023] Step S46: The fast gradient method is introduced to generate adversarial examples. The model is trained simultaneously using both original and adversarial examples. This enhances the model's resistance to input noise while maintaining its ability to fit the original data. The Adam optimizer is used to optimize the weighted sum of the original loss and the adversarial loss. A dynamic learning rate adjustment strategy is adopted to obtain the joint extraction model with optimal parameters after training.

[0024] In a preferred embodiment of the present invention, the iteration termination condition during training in step S64 is either the maximum number of training rounds or the validation set performance no longer improves.

[0025] In a preferred embodiment of the present invention, step S6 includes:

[0026] Step S61: Based on the constraint rules defined in the ontology model of hot strip rolling, remove duplicate triples in the structured triple set, correct attribute values ​​that do not conform to the numerical range constraints, and complete the missing necessary attributes.

[0027] Step S62: Extract all entities from the processed triple set. For entities extracted from different data sources that point to the same object or concept in the real world, identify and merge them to eliminate naming ambiguity and referential conflicts. Align entities based on the similarity of entity name, key attributes and contextual relationships.

[0028] Step S63: Merge the attributes and relationships associated with the aligned entities; when conflicts occur, resolve them according to a preset priority strategy or conduct manual review.

[0029] In a preferred embodiment of the present invention, the hot rolling fault data includes fault mechanism documents, expert experience rules, and historical fault case data.

[0030] In a preferred embodiment of the present invention, step S7, constructing a fault knowledge graph, includes:

[0031] Step S71: Extract knowledge causal triples with cause, cause, and result as basic units from fault mechanism documents, expert experience rules, and historical fault case data.

[0032] Step S72: Based on the hot rolling fault data, apply the causal discovery algorithm to mine the potential causal dependencies between process variables and fault events, as well as between process variables themselves, and generate data causal triples.

[0033] Step S73: Align the entities in the knowledge causal triples and the data causal triples with unique identifiers, perform consistency verification and conflict resolution on the causal relationship between the two, and form a unified causal knowledge set that has been verified by knowledge and corroborated by data.

[0034] Step S74: The unified causal knowledge set is augmented in the form of causal edges and stored in the basic knowledge graph to construct a fault knowledge graph.

[0035] Secondly, embodiments of the present invention also provide a professional knowledge graph construction system for the hot rolling process of strip steel. The system includes: a data acquisition module, a data preprocessing module, an ontology construction and mapping module, a joint extraction model construction module, a triplet extraction module, a basic knowledge graph generation module, and a professional knowledge graph construction module; wherein...

[0036] The data acquisition module is used to acquire multi-source heterogeneous data of the hot rolling process of strip steel; the multi-source heterogeneous data includes structured, semi-structured and unstructured data;

[0037] The data preprocessing module is used to extract domain terms, variable names and value ranges from structured and semi-structured data, convert them into unstructured data, and merge them into the unstructured dataset.

[0038] The ontology construction and mapping module is used to construct an ontology of professional knowledge graph based on the content of multi-source heterogeneous data. The ontology includes entity types and entity relationships. Unstructured data is divided into training set and test set, and the unstructured data in the training set is classified and labeled according to the ontology.

[0039] The joint extraction model construction module is used to construct a joint extraction model based on the CasRel framework and multi-hop attention mechanism, and to train and test the joint extraction model using the training set and test set; the joint extraction model uses a pre-trained language model as the encoder and a network structure integrating head entity recognition, multi-hop attention enhancement layer and relation-specific tail entity labeler as the decoder.

[0040] The triple extraction module is used to input all unstructured data into the trained joint extraction model, identify and extract the head entity, relation and tail entity of the unstructured data, and form a structured triple set.

[0041] The basic knowledge graph generation module is used to perform data cleaning, entity alignment and conflict resolution on the structured triples of the extracted unstructured data to obtain unified, standardized and non-redundant structured triples of unstructured data, and store them to obtain the basic knowledge graph of the hot rolling process of strip steel.

[0042] The professional knowledge graph construction module is used to acquire hot rolling fault data, and construct a fault knowledge graph that represents the fault source, propagation path and constraint relationship based on the hot rolling fault data and the constraint relationship in the basic knowledge graph; the basic knowledge graph and the fault knowledge graph are fused to obtain a professional knowledge graph for the hot rolling process of strip steel.

[0043] The method and system for constructing a professional knowledge graph for the hot rolling process of strip steel provided in this invention have the following beneficial effects:

[0044] (1) A standardized domain knowledge framework was constructed. By defining a multi-level ontology model for the hot-rolled strip steel domain, the core concepts, attributes, and their related systems of the domain were systematically sorted out and formalized. This provides a unified and reusable semantic framework and reference standard for the structuring, systematization, integration, and development of knowledge in this domain.

[0045] (2) The integrated entity relationship extraction model is established by adopting a joint extraction model based on the CasRel framework to simultaneously identify entities and relationships, thus avoiding the error propagation of traditional step-by-step extraction. By introducing adversarial training to enhance the robustness of the model and combining it with a multi-hop attention mechanism to capture deep semantics, high-accuracy and high-efficiency knowledge mining of text in the field of hot strip rolling is achieved.

[0046] (3) It realizes dynamic reasoning and knowledge-based fault propagation path. By integrating real-time data and causal models, it can dynamically deduce the complete propagation path of fault from source to phenomenon in the domain knowledge graph, and instantiate the path into traceable and reusable structured knowledge in the graph, providing intuitive and reliable graph support for real-time diagnosis and decision-making.

[0047] Of course, implementing any product or method of the present invention does not necessarily require achieving all of the advantages described above at the same time. Attached Figure Description

[0048] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0049] Figure 1 This is a flowchart of a method for constructing a professional knowledge graph for the hot rolling process of strip steel, provided in an embodiment of the present invention.

[0050] Figure 2 This is an example diagram of the ontology structure of the professional knowledge graph in an embodiment of the present invention;

[0051] Figure 3 This is a schematic diagram illustrating the training principle of the joint extraction model in this embodiment of the invention. Detailed Implementation

[0052] After discovering the aforementioned problems, the inventors of this application conducted a detailed study on the existing issues related to the full lifecycle monitoring of hot-rolled strip steel. The study found that multi-source data with various structures can be presented by constructing knowledge graphs. Knowledge graph technology, with its powerful semantic association and structured organization capabilities, provides an ideal technical path for activating and utilizing these multi-source heterogeneous data and constructing a unified knowledge system in the field of hot-rolled strip steel. In recent years, knowledge graphs have achieved significant results in fields such as financial risk control and medical diagnosis, demonstrating their universal value in integrating domain knowledge and enhancing system cognitive capabilities. However, in the process industry, especially in the specific scenario of hot-rolled strip steel with extremely complex processes and strong data coupling, the in-depth application of knowledge graphs is still in the exploratory stage, and a large-scale, engineered mature solution has not yet been formed.

[0053] Currently, in terms of knowledge graph construction technology, mainstream methods mostly adopt a phased pipeline extraction strategy, that is, first performing named entity recognition, and then classifying the relationships between the identified entities. This "entity first, relationship later" paradigm has obvious defects: error propagation, information fragmentation, and the resulting bottlenecks in overall extraction accuracy and efficiency. In addition, facing the challenges of dense technical terms, complex nested entity relationships, and varied sentence structures in texts from the hot strip rolling field, general extraction models often lack generalization ability and robustness. To overcome these limitations, there is an urgent need for a dedicated construction method that can adapt to the characteristics of data and knowledge in the hot strip rolling field. This method needs to achieve automated and integrated conversion from multi-source heterogeneous data to reasonable knowledge, overcome the accuracy and robustness bottlenecks in unstructured text knowledge extraction, and ultimately support dynamic reasoning capabilities for core needs such as fault diagnosis, transforming data resources into core assets driving intelligent production.

[0054] It should be noted that the defects in the above-mentioned prior art solutions are all results obtained by the inventors after practice and careful research. Therefore, the discovery process of the above problems and the solutions proposed by the embodiments of the present invention in the following text should all be contributions made by the inventors to the present invention.

[0055] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. It should be noted that, without conflict, the embodiments and features in the embodiments of the present invention can also be combined with each other.

[0056] It should be noted that similar reference numerals and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. In the description of the embodiments of the present invention, the terms "first," "second," "third," "fourth," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance. In addition, sometimes a subscript such as W1 may be written in a non-subscript form such as W1, and their meanings are consistent unless the distinction is emphasized.

[0057] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0058] Following the above in-depth analysis, this invention provides a method and system for constructing a professional knowledge graph for the hot rolling process of strip steel. The professional knowledge graph is constructed based on the CASREL framework and a multi-hop attention mechanism, while also introducing adversarial training. Then, based on the professional knowledge graph, and using mechanistic rules, process knowledge, and fault data, a causal model of hybrid fault propagation is generated. This provides knowledge such as propagation paths and causal relationships for monitoring the entire lifecycle of hot rolling. For example, when a fault occurs, the entire propagation path of the fault is displayed, including the location of the fault, and then the propagation path of the fault along with the material and information flow during the production process. Furthermore, the final strip steel performance can be predicted or calculated using process parameters throughout the entire lifecycle.

[0059] like Figure 1 As shown, the method for constructing a professional knowledge graph for the hot rolling process of strip steel can be implemented by an electronic device, which can be a terminal or a server. The method includes the following steps:

[0060] Step S1: Obtain multi-source heterogeneous data of the hot rolling process of strip steel; the multi-source heterogeneous data includes structured, semi-structured and unstructured data.

[0061] In this step, the structured data includes production time-series data collected from sensors and actuators in hot rolling control systems such as PLCs and DCSs. The semi-structured data includes semi-structured production events and quality data obtained via APIs from Manufacturing Execution Systems (MES) and Quality Management Systems, such as production schedules, quality inspection records, expert experience reports, technical documents, operating procedures, maintenance logs, and fault records. The unstructured data includes text containing in-depth experience obtained from enterprise document management systems, maintenance logs, and expert reports, including process specification documents, equipment technical manuals, and fault analysis reports.

[0062] Step S2: Extract domain terms, variable names and value ranges from structured and semi-structured data, convert them into unstructured data, and merge them into the unstructured dataset.

[0063] In this step, structured and semi-structured data are uniformly converted into unstructured data. Finally, through processing the unstructured data, a complete professional knowledge map of the hot rolling process is obtained.

[0064] Step S3: Based on the content of multi-source heterogeneous data, construct an ontology of professional knowledge graph, which includes entity types and entity relationships; divide unstructured data into training set and test set, and classify and label the unstructured data in the training set according to the ontology.

[0065] In this step, the entity types include transaction data, attribute data, process data, and constraint data, and the entity relationships include subordination, function, description, and specification. For example... Figure 2 The diagram shown is an example of the ontology structure of the constructed professional knowledge graph. In this example diagram, entity data includes equipment, material, and indicator objects in the production process; attribute data includes physical, state, performance, and time attributes, describing the quantitative characteristics of entities and processes; process data includes operation and procedure data, representing production activities and operations over time; constraint data includes parameter constraints, operational constraints, and quality constraints, specifying the parameter limits and rules that must be followed in the production process. The ontology defines four types of data relationships: subordinate relationships describe the hierarchical structure between entities, action relationships characterize the dynamic interaction between entities and processes, descriptive relationships associate attributes with entities, and normative relationships define processes and results.

[0066] The ontology of the professional knowledge graph described in this step is a multi-level concept classification system constructed using the OWL2 Web Ontology language. Specifically, the formal coding described above is completed using an ontology editing tool; logical consistency is checked through a built-in inference engine to ensure that all definitions and constraints are consistent, generating an ontology for the hot-rolled strip steel domain; finally, the verified ontology model is serialized and output in standard RDF format as the core schema definition for the entire knowledge graph project.

[0067] Step S4: Construct a joint extraction model based on the CasRel framework and multi-hop attention mechanism, and train and test the joint extraction model using the training set and test set; the joint extraction model uses a pre-trained language model as the encoder and a network structure integrating head entity recognition, multi-hop attention enhancement layer and relation-specific tail entity labeler as the decoder.

[0068] In this step, the joint extraction model includes an input representation layer, a deep context encoding layer, a head entity recognizer, a multi-hop attention enhancement layer, and a relation-specific tail entity annotator. During the encoding phase, the joint extraction model performs deep context semantic encoding based on a pre-trained language model; during the decoding phase, the recognized head entities are enhanced with entity representation through the multi-hop attention enhancement layer, and then the relation-specific tail entity annotator is used to achieve joint decoding of entities, attributes, and relations.

[0069] like Figure 3 As shown, the training process of the joint extraction model includes:

[0070] Step S41: Load the pre-trained language model as the encoder; and randomly initialize the decoder parameters; convert the input text sequence into an embedding vector, and feed the embedding vector into the encoder for deep semantic encoding to obtain a context encoding sequence of the same length as the input sequence; in the context encoding sequence, a multi-head attention mechanism is used to enable each position to pay attention to all other positions in the sequence.

[0071] In this step, the predicted language model uses a pre-trained BERT model; the embedding vector is formed by concatenating word embedding vectors and position embedding vectors. Each layer of the encoder includes multi-head attention and a feedforward neural network to enable each position to pay attention to all other positions in the sequence through the multi-head attention mechanism. Preferably, the encoder in this step is an L-layer Transformer encoder, and after L layers of encoding, the final context-encoded sequence is obtained.

[0072] Step S42: Based on the context encoding sequence, identify all possible entities in the text through the head entity recognition layer and obtain the initial representation of each entity.

[0073] Step S43: Construct a fully connected graph for nodes using the initial representation and context encoding sequence. Then, use a K-layer graph attention network to propagate information in multiple rounds, so that the entity representation gradually integrates relevant context information from the global scope to obtain an enhanced entity representation.

[0074] In step S44, the tail entity annotator determines whether the corresponding entity participates in any relation or has any attribute based on the enhanced entity representation; for entities that are determined to have a relation or attribute, the start and end positions of the corresponding object entity or attribute value in the text are further decoded, thereby simultaneously extracting the complete triplet.

[0075] In this step, entities that do not have any relationships or attributes will not be decoded and will be skipped.

[0076] Step S45: Based on the difference between the triples and the corresponding ground truth labels, the total loss of the joint extraction model is defined to include entity recognition loss, relation classification loss, and object extraction loss.

[0077] Step S46: The Fast Gradient Method (FGM) is introduced to generate adversarial examples. The model is trained simultaneously using both original and adversarial examples. This enhances the model's resistance to input noise while maintaining its ability to fit the original data. The Adam optimizer is used to optimize the weighted sum of the original loss and the adversarial loss. A dynamic learning rate adjustment strategy is employed to obtain the joint extraction model with optimal parameters after training.

[0078] In this step, the iteration termination condition during training can be the maximum number of training rounds or the validation set performance no longer improving.

[0079] Step S5: Input all unstructured data into the trained joint extraction model, identify and extract the head entities, relations and tail entities of the unstructured data to form a set of structured triples.

[0080] Step S6 involves cleaning, aligning entities, and resolving conflicts in the extracted unstructured data's structured triples to obtain unified, standardized, and non-redundant unstructured data's structured triples, and storing them to obtain the basic knowledge graph of the hot rolling process of strip steel.

[0081] This step specifically includes:

[0082] Step S61: Based on the constraint rules defined in the ontology model of hot strip rolling, remove duplicate triples in the structured triple set, correct attribute values ​​that do not conform to the numerical range constraints, and complete the missing necessary attributes.

[0083] Step S62: Extract all entities from the processed triple set. For entities extracted from different data sources that point to the same object or concept in the real world, identify and merge them to eliminate naming ambiguity and referential conflicts. Align entities based on the similarity of entity name, key attributes and contextual relationships.

[0084] Step S63: Merge the attributes and relationships associated with the aligned entities; when conflicts occur, resolve them according to a preset priority strategy or conduct manual review.

[0085] Step S7: Obtain hot rolling fault data, and construct a fault knowledge graph representing the fault source, propagation path and constraint relationship based on the hot rolling fault data and the constraint relationship in the basic knowledge graph; merge the basic knowledge graph and the fault knowledge graph to obtain a professional knowledge graph for the hot rolling process of strip steel.

[0086] In this step, the hot-rolling fault data includes fault mechanism documents, expert experience rules, and historical fault case data. Constructing a fault knowledge graph includes the following steps:

[0087] Step S71: Extract knowledge causal triples with cause, cause, and result as basic units from fault mechanism documents, expert experience rules, and historical fault case data.

[0088] Step S72: Based on the hot rolling fault data, apply the causal discovery algorithm to mine the potential causal dependencies between process variables and fault events, as well as between process variables themselves, and generate data causal triples.

[0089] Step S73: Align the entities in the knowledge causal triples and the data causal triples with unique identifiers, perform consistency verification and conflict resolution on the causal relationship between the two, and form a unified causal knowledge set that has been verified by knowledge and corroborated by data.

[0090] Step S74: The unified causal knowledge set is augmented in the form of causal edges and stored in the basic knowledge graph to construct a fault knowledge graph.

[0091] Based on the same approach, this invention also provides a professional knowledge graph and its construction system for the hot rolling process of strip steel. The system includes: a data acquisition module, a data preprocessing module, an ontology construction and mapping module, a joint extraction model construction module, a triplet extraction module, a basic knowledge graph generation module, and a professional knowledge graph construction module.

[0092] The data acquisition module is used to acquire multi-source heterogeneous data of the hot rolling process of strip steel; the multi-source heterogeneous data includes structured, semi-structured and unstructured data;

[0093] The data preprocessing module is used to extract domain terms, variable names and value ranges from structured and semi-structured data, convert them into unstructured data, and merge them into the unstructured dataset.

[0094] The ontology construction and mapping module is used to construct an ontology of professional knowledge graph based on the content of multi-source heterogeneous data. The ontology includes entity types and entity relationships. Unstructured data is divided into training set and test set, and the unstructured data in the training set is classified and labeled according to the ontology.

[0095] The joint extraction model construction module is used to construct a joint extraction model based on the CasRel framework and multi-hop attention mechanism, and to train and test the joint extraction model using the training set and test set; the joint extraction model uses a pre-trained language model as the encoder and a network structure integrating head entity recognition, multi-hop attention enhancement layer and relation-specific tail entity labeler as the decoder.

[0096] The triple extraction module is used to input all unstructured data into the trained joint extraction model, identify and extract the head entity, relation and tail entity of the unstructured data, and form a structured triple set.

[0097] The basic knowledge graph generation module is used to perform data cleaning, entity alignment and conflict resolution on the structured triples of the extracted unstructured data to obtain unified, standardized and non-redundant structured triples of unstructured data, and store them to obtain the basic knowledge graph of the hot rolling process of strip steel.

[0098] The professional knowledge graph construction module is used to acquire hot rolling fault data, and construct a fault knowledge graph that represents the fault source, propagation path and constraint relationship based on the hot rolling fault data and the constraint relationship in the basic knowledge graph; the basic knowledge graph and the fault knowledge graph are fused to obtain a professional knowledge graph for the hot rolling process of strip steel.

[0099] The system or device for executing the method in this embodiment of the invention can be a terminal or a server. The system includes a processor, a memory, and / or a transceiver, etc., and is connected via a communication bus. Each module can be implemented by a processor, a memory, and / or a transceiver, etc. The processor can be, but is not limited to, one or more microprocessors (MPUs), central processing units (CPUs), network processors (NPs), digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), and other programmable logic devices, discrete gates, transistor logic devices, discrete hardware components, etc., or can be configured to implement one or more integrated circuits of this invention. The processor can perform various functions by running or executing software programs in the memory and calling data in the memory. The memory includes Random Access Memory (RAM), Read-Only Memory (ROM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Compact Disc Read-Only Memory (CD-ROM), and / or Non-Volatile Memory (NVM), etc. The transceiver is used to communicate with network devices or terminal devices, and includes a receiver and a transmitter. The memory and transceiver can be integrated with the processor or exist independently.

[0100] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means.

[0101] It should also be noted that the professional knowledge graph construction system for the hot strip rolling process described in this embodiment corresponds to the professional knowledge graph construction method for the hot strip rolling process. The description and limitations of the method also apply to the system, and will not be repeated here.

[0102] It should be understood that, in various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0103] The above description is merely a preferred embodiment of the present invention and an explanation of the technical principles employed, and is not intended to limit the scope of the claimed invention, but merely to illustrate preferred embodiments of the invention. Those skilled in the art should understand that the scope of the invention is not limited to the specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the inventive concept. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

Claims

1. A method for constructing a professional knowledge graph for the hot rolling process of strip steel, characterized in that, The method includes the following steps: Step S1: Obtain multi-source heterogeneous data of the hot rolling process of strip steel; the multi-source heterogeneous data includes structured, semi-structured and unstructured data; Step S2: Extract domain terms, variable names and value ranges from structured and semi-structured data, convert them into unstructured data, and merge them into the unstructured dataset. Step S3: Based on the content of multi-source heterogeneous data, construct the ontology of professional knowledge graph, which includes entity types and entity relationships; divide the unstructured data into training set and test set, and classify and label the unstructured data in the training set according to the ontology; Step S4: Construct a joint extraction model based on the CasRel framework and multi-hop attention mechanism, and train and test the joint extraction model using the training set and test set; the joint extraction model uses a pre-trained language model as the encoder and a network structure integrating head entity recognition, multi-hop attention enhancement layer and relation-specific tail entity labeler as the decoder. Step S5: Input all unstructured data into the trained joint extraction model, identify and extract the head entity, relation and tail entity of the unstructured data to form a structured set of triples. Step S6: Perform data cleaning, entity alignment and conflict resolution on the structured triples of the extracted unstructured data to obtain unified, standardized and non-redundant structured triples of unstructured data, and store them to obtain the basic knowledge graph of the hot rolling process of strip steel. Step S7: Obtain hot rolling fault data, and construct a fault knowledge graph representing the fault source, propagation path and constraint relationship based on the hot rolling fault data and the constraint relationship in the basic knowledge graph; merge the basic knowledge graph and the fault knowledge graph to obtain a professional knowledge graph for the hot rolling process of strip steel.

2. The method according to claim 1, characterized in that, The entity types mentioned in step S3 include transaction data, attribute data, process data, and constraint data, and the entity relationships include subordination, function, description, and specification.

3. The method according to claim 1, characterized in that, In step S4, the joint extraction model includes an input representation layer, a deep context encoding layer, a head entity recognizer, a multi-hop attention enhancement layer, and a relation-specific tail entity annotator.

4. The method according to claim 1, characterized in that, In the encoding stage, the joint extraction model performs deep contextual semantic encoding based on a pre-trained language model; in the decoding stage, the identified head entities are enhanced with entity representation through a multi-hop attention enhancement layer, and then a relation-specific tail entity annotator is used to achieve joint decoding of entities, attributes and relations.

5. The method according to claim 1, characterized in that, The training process of the joint extraction model includes: Step S41: Load the pre-trained language model as the encoder; and randomly initialize the decoder parameters; convert the input text sequence into an embedding vector, and feed the embedding vector into the encoder for deep semantic encoding to obtain a context encoding sequence of the same length as the input sequence; in the context encoding sequence, a multi-head attention mechanism is used to enable each position to pay attention to all other positions in the sequence; Step S42: Based on the context encoding sequence, identify all possible entities in the text through the head entity recognition layer and obtain the initial representation of each entity; Step S43: Construct a fully connected graph for nodes using the initial representation and context encoding sequence. Then, use a K-layer graph attention network to propagate information in multiple rounds, so that the entity representation gradually integrates relevant context information from the global scope to obtain an enhanced entity representation. Step S44: The tail entity annotator determines whether the corresponding entity participates in any relation or has any attribute based on the enhanced entity representation; for entities that are determined to have a relation or attribute, the start and end positions of the corresponding object entity or attribute value in the text are further decoded, thereby extracting the complete triples simultaneously. Step S45: Based on the difference between the triples and the corresponding real annotations, the total loss of the joint extraction model is defined to include entity recognition loss, relation classification loss and object extraction loss. Step S46: The fast gradient method is introduced to generate adversarial examples. The model is trained simultaneously using both original and adversarial examples. This enhances the model's resistance to input noise while maintaining its ability to fit the original data. The Adam optimizer is used to optimize the weighted sum of the original loss and the adversarial loss. A dynamic learning rate adjustment strategy is adopted to obtain the joint extraction model with optimal parameters after training.

6. The method according to claim 5, characterized in that, In step S64, the iteration termination condition during training is either the maximum number of training rounds or the validation set performance no longer improves.

7. The method according to claim 1, characterized in that, Step S6 includes: Step S61: Based on the constraint rules defined in the ontology model of hot strip rolling, remove duplicate triples in the structured triple set, correct attribute values ​​that do not conform to the numerical range constraints, and complete the missing necessary attributes. Step S62: Extract all entities from the processed triple set. For entities extracted from different data sources that point to the same object or concept in the real world, identify and merge them to eliminate naming ambiguity and referential conflicts. Align entities based on the similarity of entity name, key attributes and contextual relationships. Step S63: Merge the attributes and relationships associated with the aligned entities; when conflicts occur, resolve them according to a preset priority strategy or conduct manual review.

8. The method according to claim 1, characterized in that, The hot rolling fault data includes fault mechanism documents, expert experience rules, and historical fault case data.

9. The method according to claim 1, characterized in that, Step S7 constructs a fault knowledge graph, including: Step S71: Extract knowledge causal triples with cause, cause, and result as basic units from fault mechanism documents, expert experience rules, and historical fault case data. Step S72: Based on the hot rolling fault data, apply the causal discovery algorithm to mine the potential causal dependencies between process variables and fault events, as well as between process variables themselves, and generate data causal triples. Step S73: Align the entities in the knowledge causal triples and the data causal triples with unique identifiers, perform consistency verification and conflict resolution on the causal relationship between the two, and form a unified causal knowledge set that has been verified by knowledge and corroborated by data. Step S74: The unified causal knowledge set is augmented in the form of causal edges and stored in the basic knowledge graph to construct a fault knowledge graph.

10. A professional knowledge graph construction system for hot strip rolling processes, characterized in that, The system includes: a data acquisition module, a data preprocessing module, an ontology construction and mapping module, a joint extraction model construction module, a triple extraction module, a basic knowledge graph generation module, and a professional knowledge graph construction module; wherein... The data acquisition module is used to acquire multi-source heterogeneous data of the hot rolling process of strip steel; the multi-source heterogeneous data includes structured, semi-structured and unstructured data; The data preprocessing module is used to extract domain terms, variable names and value ranges from structured and semi-structured data, convert them into unstructured data, and merge them into the unstructured dataset. The ontology construction and mapping module is used to construct an ontology of professional knowledge graph based on the content of multi-source heterogeneous data. The ontology includes entity types and entity relationships. Unstructured data is divided into training set and test set, and the unstructured data in the training set is classified and labeled according to the ontology. The joint extraction model construction module is used to construct a joint extraction model based on the CasRel framework and multi-hop attention mechanism, and to train and test the joint extraction model using the training set and test set; the joint extraction model uses a pre-trained language model as the encoder and a network structure integrating head entity recognition, multi-hop attention enhancement layer and relation-specific tail entity labeler as the decoder. The triple extraction module is used to input all unstructured data into the trained joint extraction model, identify and extract the head entity, relation and tail entity of the unstructured data, and form a structured triple set. The basic knowledge graph generation module is used to perform data cleaning, entity alignment and conflict resolution on the structured triples of the extracted unstructured data to obtain unified, standardized and non-redundant structured triples of unstructured data, and store them to obtain the basic knowledge graph of the hot rolling process of strip steel. The professional knowledge graph construction module is used to acquire hot rolling fault data, and construct a fault knowledge graph that represents the fault source, propagation path and constraint relationship based on the hot rolling fault data and the constraint relationship in the basic knowledge graph; the basic knowledge graph and the fault knowledge graph are fused to obtain a professional knowledge graph for the hot rolling process of strip steel.

Images