A Structured Extraction Method for Chemical Engineering P&ID Drawings Based on Large Language Models

Abstract

This invention provides a structured extraction method for chemical P&ID drawings based on a large language model, belonging to the field of drawing structuring technology. This method involves image parsing of chemical P&ID drawings to obtain element types, spatial locations, and text annotation relationships. A candidate association set is established based on the spatial adjacency relationships between text annotations and process equipment, pipelines, and instrumentation elements. The text annotations are then split into sentences to form semantic analysis input data containing descriptive objects, operating conditions, and behavioral constraints. A large language model is used for engineering semantic parsing to generate semantic parsing results. Next, the candidate association relationships are semantically consistent and filtered to obtain the final association results after semantic correction. This constructs structured data for chemical P&ID drawings containing process equipment attributes, pipeline connection relationships, and their effective conditions, thereby achieving accurate extraction and structured expression of complex engineering information in the drawings.

Description

Technical Field

[0001] This invention relates to the field of drawing structuring technology, and in particular to a method for structuring and extracting chemical P&ID drawings based on a large language model. Background Technology

[0002] In existing technologies, structured extraction of chemical P&ID drawings typically employs a combination of image recognition and rule-based reasoning. These methods generally identify equipment symbols, pipelines, and instrumentation elements in the drawings using object detection or image segmentation models, and then extract corresponding text annotations using OCR technology. Subsequently, based on pre-defined engineering rules and topological constraints, the identified elements are associated with the text to form structured data such as equipment attributes, pipeline connections, and control loop information. Some improved solutions also introduce graph structure modeling or feature-matching-based relationship inference mechanisms to enhance the handling of complex connections, but the overall extraction process still primarily relies on graphical features and explicit rules.

[0003] However, in practical engineering applications, the aforementioned existing technologies have insurmountable technical defects when dealing with P&ID drawings that contain implicit engineering semantic information. For example, in the detailed design phase of large-scale chemical plants, P&ID drawings often express design intent through textual descriptions, local annotations, or cross-page references, such as "This valve is for use during start-up and shutdown," "This pipeline is only connected under abnormal operating conditions," and "See attached drawing XX for control logic." This type of information does not appear in a clear graphical connection or fixed format, but rather exists scattered in different areas of the drawing in the form of natural language. Existing methods based on image recognition and rule matching can usually only extract the explicit geometric relationships between graphic elements and cannot understand the engineering semantics expressed by the above annotations. As a result, temporary operating condition pipelines, standby equipment, or conditional control loops are misclassified as part of the regular process flow in the structured results. Such misclassification may lead to the long-term presence of process paths that should not be normalized in the model when used for process consistency verification or digital model construction. Moreover, due to the lack of corresponding semantic basis, it is difficult to correct this by simply adding rules or adjusting recognition parameters. Summary of the Invention

[0004] The purpose of this invention is to provide a structured extraction method for chemical P&ID drawings based on a large language model, aiming to solve the problems mentioned in the background art.

[0005] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows: A structured extraction method for chemical P&ID drawings based on a large language model, the method comprising: Obtain chemical P&ID drawings, perform image analysis on the chemical P&ID drawings, and generate primitive recognition results that include the type of each primitive, the location of each primitive, and the spatial pointing relationship of the text annotations; Based on the primitive recognition results, according to the spatial adjacency and pointing relationships between the text annotation information and the process equipment primitives, pipeline primitives and instrument primitives, multiple possible association relationships are established for each text annotation information to generate a candidate association set; The text annotation information in the candidate association set is split into sentences at the level, and each text annotation information is split into multiple semantic fragments describing the object, working conditions and behavioral constraints. The semantic fragments are then combined with their corresponding candidate association relationships to form semantic analysis input data. The semantic analysis input data is input into the large language model, and the semantic fragments are subjected to engineering semantic parsing to obtain semantic parsing results. The semantic parsing results include object matching judgment results for characterizing whether each graphic element process object is consistent with the described object, working condition constraint judgment results for characterizing whether the association relationship is restricted by working conditions, and connection attribute judgment results for characterizing the connection attribute of the association relationship. Based on the semantic parsing results, the candidate association set is subjected to semantic consistency verification and screening to generate the final association results after semantic correction. Based on the final association results, construct structured chemical P&ID data that includes process equipment attributes, pipeline connection relationships, and their effective conditions.

[0006] Preferably, the text annotation information in the candidate association set is split into sentences, each text annotation information is split into multiple semantic fragments describing the object, working conditions, and behavioral constraints, and the semantic fragments are combined with their corresponding candidate association relationships to form semantic analysis input data, including: Based on the text annotation information in the candidate association set, the text is parsed sentence by sentence to identify the object indication content used to indicate the process object, the condition limitation content used to limit the applicable premise, and the behavior description content used to describe the connection or operation, and the text parsing result is generated. Based on the text parsing results, the object indication content is converted into semantic fragments describing the object, the condition constraint content is converted into semantic fragments describing the working condition, and the behavior description content is converted into semantic fragments describing the behavior constraint, thus generating a set of semantic fragments; When the text annotation information contains multiple object indication contents, the text annotation information is split according to the number of object indication contents to generate multiple semantic fragments of description objects corresponding to different objects. Based on the set of semantic fragments, and in accordance with the order in which the semantic fragments appear in the text annotation information, the semantic fragments describing the object, the semantic fragments of the working condition, and the semantic fragments of the behavior limitation are combined to generate a structured semantic sequence. Based on each candidate association in the candidate association set, the structured semantic sequence is bound to the corresponding candidate association to generate semantic analysis input data for engineering semantic parsing.

[0007] Preferably, the engineering semantic parsing process includes: The semantic analysis input data is input into the large language model as an independent semantic unit. The semantic unit includes semantic fragments describing objects, semantic fragments describing working conditions, and semantic fragments defining behaviors. Based on semantic units, the overall engineering semantic parsing is performed on the semantic fragments of the described object, the semantic fragments of the working condition, and the semantic fragments of the behavior constraint, to construct a unified semantic context covering the semantic units and generate semantic association analysis results; Based on the semantic association analysis results, determine whether the process object indicated by the semantic fragment of the description object is consistent with the process equipment elements, pipeline elements or instrument elements in the candidate association relationship in terms of engineering semantics, and generate object matching determination results; Based on the object matching judgment results and semantic association analysis results, combined with the semantic fragments describing the objects, the applicable premises limited by the semantic fragments of the working condition are interpreted, and the working condition constraint judgment results are generated. Based on the object matching judgment results and semantic association analysis results, combined with the semantic fragments describing the object and the semantic fragments of the working condition, the meaning of the connection or operation described by the behavior limiting semantic fragments is interpreted, and the connection attribute judgment results are generated. The object matching determination result, the working condition constraint determination result, and the connection attribute determination result are used as the output of the same semantic parsing process to generate a semantic parsing result for the candidate association relationship.

[0008] Preferably, based on the semantic parsing results, the candidate association set is subjected to semantic consistency verification and filtering to generate the final association results after semantic correction, including: Based on the object matching judgment results in the semantic parsing results, the candidate associations in the candidate association set are screened in the first round to remove associations whose primitive types are inconsistent with the description object and the candidate associations, and the first screening results are generated. Based on the working condition constraint determination results in the semantic parsing results, the associations in the first screening results are marked, and the associations that only exist under specific working conditions are marked as conditional associations, generating working condition marking results; Based on the connection attribute determination results in the semantic parsing results, the working condition labeling results are filtered in the second round to remove the association relationships that do not constitute process flow connections, and the second filtering results are generated. When the same text annotation information corresponds to multiple relationships in the second filtering results, the relationships that are judged to constitute regular process flow connections are retained based on the connection attribute judgment results, and the final association results after semantic correction are generated.

[0009] Preferably, based on semantic units, a holistic engineering semantic parsing is performed on the semantic fragments describing the object, the semantic fragments describing the working conditions, and the semantic fragments defining the behavior, constructing a unified semantic context covering the semantic units and generating semantic association analysis results, including: Based on semantic units, the first stage of parsing is performed on each semantic segment to identify explicit semantic reference relationships between the semantic segments and generate reference relationship results; Based on the referential relationship results, the second stage of parsing is performed on the limiting and functional relationships between each semantic segment to generate the initial semantic relationship structure; Based on the initial semantic relation structure, engineering semantic consistency constraints are introduced to verify and correct semantic relations with multiple or ambiguous pointers, and to generate a corrected semantic relation structure. Based on the corrected semantic relation structure, a unified semantic context covering the entire semantic unit is constructed; Semantic association analysis results are extracted from a unified semantic context to represent stable associations between semantic fragments describing objects, semantic fragments describing operating conditions, and semantic fragments defining behavior.

[0010] Preferably, based on the semantic association analysis results, it is determined whether the process object indicated by the semantic fragment of the described object is consistent with the process equipment elements, pipeline elements, or instrument elements in the candidate association relationship in terms of engineering semantics, and an object matching determination result is generated, including: Based on the semantic association analysis results, object semantic features are extracted from the semantic fragments describing the object to represent the type, functional role and engineering attributes of the process object, and an object semantic feature set is generated. Based on the set of semantic features of objects, extract the corresponding graph element category, connection role and engineering attribute from the candidate association relationship to generate a set of graph element semantic features; The set of semantic features of objects and the set of semantic features of primitives are aligned using engineering semantics to generate an initial set of semantic matching relationships. Based on the initial set of semantic matching relationships, engineering semantic consistency constraints are introduced to eliminate matching relationships that do not meet the consistency of object type or engineering role, and generate a set of valid semantic matching relationships. Based on the set of valid semantic matching relationships, an object matching judgment result is generated to characterize the consistency of the semantic fragments of the described object with the primitives in the candidate association relationship in terms of engineering semantics.

[0011] Preferably, based on the object matching determination result and the semantic association analysis result, and in conjunction with the semantic fragment describing the object, the applicable premises limited by the semantic fragment of the working condition are interpreted to generate the working condition constraint determination result, including: When the object matching determination result shows that the semantic fragment of the descriptive object and the process object in the candidate association relationship have an engineering semantic basis, the semantic fragment of the working condition is parsed and processed. Based on the semantic association analysis results, determine the corresponding target of the working condition semantic fragment in the unified semantic context, and generate the working condition target determination result. Based on the determination of the object of the working condition, the semantic fragment of the working condition is analyzed by engineering semantics to distinguish it as object state constraint or connection effective condition, and the working condition constraint description result is generated. Based on the description of the working condition constraints, a working condition constraint determination result is generated to characterize whether the candidate association relationship is restricted by the working condition conditions.

[0012] Preferably, based on the object matching determination result and the semantic association analysis result, combined with the semantic fragments describing the object and the semantic fragments describing the working conditions, the meaning of the connection or operation described by the behavior-limiting semantic fragment is interpreted to generate a connection attribute determination result, including: When the object matching determination result indicates that the semantic fragment describing the object has an engineering semantic basis with the process object in the candidate association relationship, the behavior-limited semantic fragment is parsed and processed. Based on the semantic association analysis results, engineering semantic elements that represent the connection direction, connection method and operation type are extracted from the behavior-defined semantic fragments to generate an initial set of connection semantic elements. Based on the initial set of connection semantic elements, and combined with the judgment results of the working condition constraints corresponding to the working condition semantic fragments, the validity of the connection semantic elements is verified, and a set of valid connection semantic elements is generated. Based on the set of valid connection semantic elements, the connection or operation meanings corresponding to candidate associations are merged to generate connection attribute determination results that characterize the connection attributes of candidate associations.

[0013] Preferably, based on the set of semantic fragments, and according to the order in which the semantic fragments appear in the text annotation information, the semantic fragments describing the object, the semantic fragments describing the working conditions, and the semantic fragments defining the behavior are combined to generate a structured semantic sequence, including: Obtain a set of semantic fragments generated for the same text annotation information, the set of semantic fragments including semantic fragments describing objects, semantic fragments of working conditions, and semantic fragments defining behaviors; Based on the original order of appearance of semantic segments in the text annotation information, the semantic segments in the semantic segment set are sorted to generate an ordered semantic segment sequence; In an ordered sequence of semantic fragments, the location of the semantic fragment describing the object is identified, and the semantic fragment describing the object is used as the core node of the semantic sequence. Based on the core node, the working condition semantic fragments and behavior limitation semantic fragments located before or after the semantic fragments of the described object are sequentially associated with the core node to form a structured semantic sequence around the same semantic fragment of the described object. When there are multiple semantic fragments describing objects in an ordered sequence of semantic fragments, each semantic fragment describing an object is used as a core node to generate multiple independent structured semantic sequences.

[0014] Preferably, based on the primitive recognition results, and according to the spatial adjacency and directional relationships between the text annotation information and the process equipment primitives, pipeline primitives, and instrument primitives, multiple possible association relationships are established for each text annotation information, generating a candidate association set, including: Obtain the spatial location of each text annotation information in the graphic element recognition result, as well as the spatial location of process equipment graphic elements, pipeline graphic elements, and instrument graphic elements; Based on the relative positional relationship between the text annotation information and the process equipment elements, pipeline elements and instrument elements, identify the elements that meet the preset adjacency conditions with the text annotation information in space, and generate a set of adjacent elements; Based on the pointing symbols, leaders, or annotation direction information contained in the text annotation information, determine whether there is a pointing relationship between the text annotation information and each graphic element in the adjacent graphic element set, and generate a set of pointing association relationships; By combining the set of adjacent graphic elements with the set of pointing associations, multiple associations are established for each text annotation information, corresponding to different process equipment graphic elements, pipeline graphic elements, or instrument graphic elements, thus generating a candidate association set.

[0015] The above-described solution of the present invention has at least the following beneficial effects: First, by introducing spatial adjacency and directional relationships between text annotations and primitives based on image analysis, a candidate association set is constructed. This prevents text annotations from being directly bound to a single primitive, but instead allows them to enter the subsequent processing in a multi-candidate form. This mechanism avoids prior misjudgments caused by relying solely on geometric distance or simple rules to directly determine associated objects, and preserves sufficient analytical space for correct association in complex annotation scenarios.

[0016] Secondly, by performing sentence-level decomposition on the text annotation information, the annotation content in the form of natural language is subdivided into different semantic fragments such as the description object, working conditions, and behavioral constraints. This makes the originally mixed engineering information explicitly distinguishable in structure, thereby providing clear semantic boundaries for subsequent semantic analysis and solving the problem in the existing technology of not being able to distinguish between "object description" and "conditional description" and "behavioral description".

[0017] Based on this, by constructing a unified semantic context for semantic fragments and generating object matching judgment results, working condition constraint judgment results, and connection attribute judgment results in sequence within this semantic context, the engineering semantics implied in the text can be parsed and verified layer by layer. This not only determines whether the object referred to in the text is consistent with the primitive, but also identifies whether the association relationship is only valid under specific working conditions and whether it constitutes a connection in a regular process flow. This makes up for the shortcomings of existing methods in understanding conditional, stage-based, and explanatory semantics.

[0018] Furthermore, by performing phased semantic consistency verification and screening on the candidate association set based on the above semantic parsing results, associations that are inconsistent with each other, only valid under special working conditions, or do not have process flow significance can be systematically eliminated or marked. This avoids incorporating temporary pipelines, backup equipment, or conditional control logic errors into the conventional process structure, thereby improving the reliability of the structured results at the engineering semantic level.

[0019] Ultimately, by constructing structured data containing process equipment attributes, pipeline connection relationships, and their effective conditions based on semantically corrected association results, the structured results not only reflect the geometric connection relationships in the drawings, but also accurately express the differences in the effectiveness of process connections under different operating conditions. This provides a data foundation that is more in line with the actual engineering semantics for subsequent process analysis, model building, and operation management, thereby overcoming the problems of existing technologies being difficult to correct and prone to long-term misjudgments when processing implicit engineering semantic information. Attached Figure Description

[0020] Figure 1 This is a flowchart of a structured extraction method for chemical P&ID drawings based on a large language model, provided by an embodiment of the present invention. Detailed Implementation

[0021] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0022] like Figure 1 As shown, embodiments of the present invention propose a structured extraction method for chemical P&ID drawings based on a large language model, the method comprising: Obtain chemical P&ID drawings, perform image analysis on the chemical P&ID drawings, and generate primitive recognition results that include the type of each primitive, the location of each primitive, and the spatial pointing relationship of the text annotations; Based on the primitive recognition results, according to the spatial adjacency and pointing relationships between the text annotation information and the process equipment primitives, pipeline primitives and instrument primitives, multiple possible association relationships are established for each text annotation information to generate a candidate association set; The text annotation information in the candidate association set is split into sentences at the level, and each text annotation information is split into multiple semantic fragments describing the object, working conditions and behavioral constraints. The semantic fragments are then combined with their corresponding candidate association relationships to form semantic analysis input data. The semantic analysis input data is input into the large language model, and the semantic fragments are subjected to engineering semantic parsing to obtain semantic parsing results. The semantic parsing results include object matching judgment results for characterizing whether each graphic element process object is consistent with the described object, working condition constraint judgment results for characterizing whether the association relationship is restricted by working conditions, and connection attribute judgment results for characterizing the connection attributes of the association relationship. The connection attributes include connection direction, connection method and operation type. Based on the semantic parsing results, the candidate association set is subjected to semantic consistency verification and screening to generate the final association results after semantic correction. Based on the final association results, construct structured chemical P&ID data that includes process equipment attributes, pipeline connection relationships, and their effective conditions.

[0023] In this embodiment of the invention, by performing image analysis on chemical P&ID drawings, the results of graphic element recognition, including graphic element type, graphic element spatial location, and text annotation spatial pointing relationship, are obtained. This enables the process equipment, pipelines, instruments and their corresponding text annotations in the drawings to be uniformly expressed in a structured form, providing basic data support for the establishment of subsequent association relationships.

[0024] Based on the primitive recognition results, multiple candidate association relationships are established for each text annotation information according to the spatial adjacency and directional relationships between the text annotation information and different types of primitives. This avoids the misjudgment problem caused by directly determining the associated object based on a single spatial relationship, and ensures that the potential correspondence between the text annotation and the process object can be completely preserved and entered into the subsequent analysis process.

[0025] Based on this, the text annotation information is split into sentences, dividing the text content into semantic fragments such as the descriptive object, working conditions, and behavioral constraints. The semantic fragments are then combined with the corresponding candidate associations to form semantic analysis input data, enabling different semantic components in the text annotation to be distinguished and processed, thereby providing a clear semantic structure for subsequent engineering semantic analysis.

[0026] By performing engineering semantic parsing on the semantic analysis input data, object matching judgment results, working condition constraint judgment results, and connection attribute judgment results are generated. This enables the determination of whether the process object described in the text annotation is consistent with the graphic element, whether the association relationship is restricted by the working condition, and the connection or operation attributes reflected by the association relationship. In this way, the engineering meaning that was originally mixed in the text is decomposed into multiple judgment results with clear semantic orientation.

[0027] Based on the semantic parsing results, semantic consistency verification and screening are performed on the candidate association set. This can eliminate associations that are inconsistent between the described object and the primitive type, do not hold under specific working conditions, or do not constitute a process flow connection. This ensures that the final retained associations meet the engineering semantic requirements in terms of object consistency, working condition applicability, and connection rationality, thereby reducing erroneous associations caused by textual ambiguity or unclear spatial relationships.

[0028] By constructing structured P&ID data for chemical engineering based on the final association results after semantic correction, process equipment attributes, pipeline connection relationships and their effective conditions can be expressed in a unified data structure, providing a foundation for subsequent process analysis, operation management or data reuse.

[0029] For example, in a specific implementation scenario, a chemical P&ID drawing contains a text annotation that is adjacent to multiple pipelines and equipment elements. This text annotation includes the equipment name, a description of the conditions under a specific operating state, and a description of the connection behavior. Using the method described above, firstly, multiple candidate associations are established for this text annotation, and the text content is semantically segmented. Subsequently, during the engineering semantic parsing process, the consistency between the process object described in the text and the specific equipment element, the constraints of the operating conditions on the association, and the engineering meaning of the connection behavior are determined. Finally, only associations that are consistent with the equipment, valid under the corresponding operating conditions, and constitute a process flow connection are retained, thereby generating structured data that accurately reflects the connection relationship between the equipment and pipelines under specific operating conditions.

[0030] In a preferred embodiment of the present invention, image analysis is performed on chemical P&ID drawings to generate primitive recognition results that include the primitive types, primitive positions, and spatial pointing relationships of text annotations, including: First, obtain electronic image files of the chemical P&ID drawings, and then perform format unification and resolution normalization on the image files to eliminate differences in size, scale and clarity between drawings from different sources, so as to provide a consistent input basis for subsequent image analysis.

[0031] Then, the processed image is analyzed for graphic elements. Based on the predefined symbolic and structural features in the chemical P&ID drawings, process equipment elements, pipeline elements, and instrument elements are identified, and the geometric position range and boundary area of ​​each type of element in the drawing are determined, thus forming a correspondence between element type and element position.

[0032] Next, the text annotation areas in the drawing are identified, the content information of the text annotations and their spatial position in the drawing are extracted, and it is determined whether there are leader lines, arrows or directional marks around the text annotations.

[0033] Based on this, by combining the relative positional relationship between the text annotation and the surrounding graphic elements, the direction of the leader line extension, and the direction of the arrow, the spatial pointing relationship between the text annotation and one or more graphic elements is determined, forming a spatial pointing mapping between the text annotation and the corresponding graphic element.

[0034] Finally, the identified primitive types, primitive locations, and spatial orientation relationships between text annotations and primitives are uniformly organized to generate primitive recognition results for subsequent association analysis.

[0035] In a preferred embodiment of the present invention, based on the final association results, structured chemical P&ID data is constructed, including process equipment attributes, pipeline connection relationships, and their effective conditions, comprising: First, the final association results determined after semantic consistency verification and filtering are obtained. The final association results clarify the semantically established association between text annotations and process equipment elements, pipeline elements, or instrument elements.

[0036] Then, for each process equipment graphic element, based on its corresponding final association result, the equipment name, equipment type and functional attributes reflected by the semantic fragments of the descriptive objects related to the process equipment are collected to form a set of attribute information of the process equipment.

[0037] Next, for pipeline elements, based on the connection attribute determination results determined in the final association results, the connection direction and connection method between the pipeline and the process equipment it connects to are clarified. Furthermore, combined with the operating condition constraint determination results, it is determined under what operating conditions the pipeline connection takes effect, thereby forming pipeline connection information that includes the connection relationship and its effective conditions.

[0038] In this process, for associations marked as conditionally valid, the corresponding operating conditions are recorded as additional attributes of the connection relationship, so that the structured data can reflect the effectiveness differences of the process flow under different operating states.

[0039] Finally, the aforementioned process equipment attribute information, pipeline connection relationship information, and their corresponding effective conditions are uniformly organized to construct a complete chemical P&ID structured data, which can be used for subsequent process analysis, system modeling, or data management applications.

[0040] In a preferred embodiment of the present invention, the text annotation information in the candidate association set is split into sentences, each text annotation information is split into multiple semantic fragments describing the object, working conditions, and behavioral constraints, and the semantic fragments are combined with their corresponding candidate association relationships to form semantic analysis input data, including: Based on the text annotation information in the candidate association set, the text is parsed sentence by sentence to identify the object indication content used to indicate the process object, the condition limitation content used to limit the applicable premise, and the behavior description content used to describe the connection or operation, and the text parsing result is generated. Based on the text parsing results, the object indication content is converted into semantic fragments describing the object, the condition constraint content is converted into semantic fragments describing the working condition, and the behavior description content is converted into semantic fragments describing the behavior constraint, thus generating a set of semantic fragments; When the text annotation information contains multiple object indication contents, the text annotation information is split according to the number of object indication contents to generate multiple semantic fragments of description objects corresponding to different objects. Based on the set of semantic fragments, and in accordance with the order in which the semantic fragments appear in the text annotation information, the semantic fragments describing the object, the semantic fragments of the working condition, and the semantic fragments of the behavior limitation are combined to generate a structured semantic sequence. Based on each candidate association in the candidate association set, the structured semantic sequence is bound to the corresponding candidate association to generate semantic analysis input data for engineering semantic parsing.

[0041] In this embodiment of the invention, the processing of text annotation information involves parsing the text annotation information in the candidate association set sentence by sentence, and splitting the text content into semantic fragments such as descriptive objects, operating conditions, and behavioral constraints, so that different engineering meanings expressed mixed in the text can be clearly distinguished. Furthermore, when there are multiple object indications, the text annotation information is split into multiple semantic fragments corresponding to different process objects, and combined according to the order of appearance in the text to form a structured semantic sequence. This avoids confusion among multiple object information during subsequent parsing, ensuring that each candidate association corresponds to a clear and independent semantic input, providing a stable semantic foundation for engineering semantic parsing.

[0042] In a preferred embodiment of the present invention, text annotation information in the candidate association set is parsed sentence by sentence to identify object indication content for indicating process objects, conditional limitation content for limiting applicable prerequisites, and behavioral description content for describing connections or operations, thereby generating text parsing results, including: First, obtain the complete text content of each text annotation information in the candidate association set, and standardize the text content to unify the character format, punctuation form and language expression structure, and eliminate the ambiguity caused by differences in annotation habits.

[0043] Then, the standardized text annotation information is segmented sentence by sentence according to semantic integrity, so that each sentence contains only a relatively complete engineering semantic expression unit, thereby providing a clear analysis object for subsequent semantic component identification.

[0044] Next, semantic component analysis is performed on each segmented statement to identify the object indication content used to refer to specific process equipment, pipelines or instruments, the condition limitation content used to limit the process status, operating conditions or applicable scope, and the behavioral description content used to describe connection relationships, operation behavior or process actions.

[0045] Finally, the object indications, conditional constraints, and behavioral descriptions identified in each statement are recorded to form a structured text parsing result, providing basic data for subsequent semantic fragment generation.

[0046] In a preferred embodiment of the present invention, based on the text parsing results, the object indication content is converted into a semantic fragment describing the object, the condition constraint content is converted into a semantic fragment describing the working condition, and the behavior description content is converted into a semantic fragment describing the behavior, generating a set of semantic fragments, including: First, the object indication content recorded in the aforementioned text parsing results is obtained, and each object indication content is independently encapsulated into a semantic fragment describing the object, so that it is only used to express the name, category or engineering direction of the process object, without containing condition or behavioral information.

[0047] Then, semantic extraction processing is performed on the condition constraints in the text parsing results, converting information reflecting the operating status, triggering conditions, or applicable prerequisites into semantic fragments of operating conditions, enabling them to independently express the meaning of operating condition constraints.

[0048] Next, the behavioral descriptions in the text parsing results are organized, and the content describing connection methods, operation actions, or process behaviors is converted into behaviorally defined semantic fragments so that they can clearly reflect the characteristics of engineering behavior.

[0049] Based on this, the generated semantic fragments of the description object, the semantic fragments of the working condition, and the semantic fragments of the behavior are uniformly collected to form a set of semantic fragments that correspond one-to-one with the original text annotation information, providing standardized input for the subsequent construction of semantic sequences.

[0050] In a preferred embodiment of the present invention, when the text annotation information contains multiple object indication contents, the text annotation information is split according to the number of object indication contents to generate multiple semantic fragments describing different objects, including: First, examine the text parsing results corresponding to the same text annotation information to determine whether two or more independent object indication contents are identified at the same time, and whether each object indication content points to different process objects in engineering semantics.

[0051] When it is determined that there are multiple object indicators in the text annotation information, the original text annotation information is semantically split according to the number of object indicators, so that each semantic unit after splitting retains only one object indicator as the core expression object.

[0052] During the decomposition process, the conditional constraints and behavioral descriptions that are directly related to the content indicated by the object are retained, so that the semantic expression after decomposition remains complete in terms of engineering semantics.

[0053] Finally, for each split semantic expression, a corresponding semantic fragment of the descriptive object is generated, so that multiple process objects can be independently parsed and independently associated in subsequent processing, avoiding semantic mixing between different process objects.

[0054] In a preferred embodiment of the present invention, the engineering semantic parsing process includes: The semantic analysis input data is input into the large language model as an independent semantic unit. The semantic unit includes semantic fragments describing objects, semantic fragments describing working conditions, and semantic fragments defining behaviors. Based on semantic units, the overall engineering semantic parsing is performed on the semantic fragments of the described object, the semantic fragments of the working condition, and the semantic fragments of the behavior constraint, to construct a unified semantic context covering the semantic units and generate semantic association analysis results; Based on the semantic association analysis results, determine whether the process object indicated by the semantic fragment of the description object is consistent with the process equipment elements, pipeline elements or instrument elements in the candidate association relationship in terms of engineering semantics, and generate object matching determination results; Based on the object matching judgment results and semantic association analysis results, combined with the semantic fragments describing the objects, the applicable premises limited by the semantic fragments of the working condition are interpreted, and the working condition constraint judgment results are generated. Based on the object matching judgment results and semantic association analysis results, combined with the semantic fragments describing the object and the semantic fragments of the working condition, the meaning of the connection or operation described by the behavior limiting semantic fragments is interpreted, and the connection attribute judgment results are generated. The object matching determination result, the working condition constraint determination result, and the connection attribute determination result are used as the output of the same semantic parsing process to generate a semantic parsing result for the candidate association relationship.

[0055] In this embodiment of the invention, by inputting semantic analysis data as independent semantic units and performing overall engineering semantic parsing, the semantic relationships between descriptive objects, operating conditions, and behavioral constraints can be comprehensively analyzed within a unified semantic context. Based on this unified semantic context, object matching judgment results, operating condition constraint judgment results, and connection attribute judgment results are generated sequentially. This ensures that the consistency judgment of process objects, the applicability judgment of operating conditions, and the judgment of connection or operational meaning form a clear causal order, thereby avoiding engineering semantic biases caused by isolated parsing of single semantic fragments and improving the completeness and consistency of semantic parsing results at the process understanding level.

[0056] In a preferred embodiment of the present invention, based on the semantic parsing results, the candidate association set is subjected to semantic consistency verification and filtering to generate a semantically corrected final association result, including: Based on the object matching judgment results in the semantic parsing results, the candidate associations in the candidate association set are screened in the first round to remove associations whose primitive types are inconsistent with the description object and the candidate associations, and the first screening results are generated. Based on the working condition constraint determination results in the semantic parsing results, the associations in the first screening results are marked, and the associations that only exist under specific working conditions are marked as conditional associations, generating working condition marking results; Based on the connection attribute determination results in the semantic parsing results, the working condition labeling results are filtered in the second round to remove the association relationships that do not constitute process flow connections, and the second filtering results are generated. When the same text annotation information corresponds to multiple relationships in the second filtering results, the relationships that are judged to constitute regular process flow connections are retained based on the connection attribute judgment results, and the final association results after semantic correction are generated.

[0057] In this embodiment of the invention, after obtaining the semantic parsing results, the candidate association set is subjected to phased semantic consistency verification and screening. This allows for the gradual elimination of associations that are inconsistent with the objects, only valid under specific operating conditions, or do not constitute a process flow connection. By using the operating condition constraint judgment results to mark conditional associations and further filtering them in conjunction with the connection attribute judgment results, the final retained associations meet the engineering semantic requirements in terms of object matching, operating condition constraints, and process flow rationality. This ensures that the final generated association results accurately reflect the real and valid process connection relationships in the P&ID drawings.

[0058] In a preferred embodiment of the present invention, based on the object matching determination result in the semantic parsing result, a first round of screening is performed on the candidate association relationships in the candidate association set to eliminate association relationships in which the descriptive object and the primitive type are inconsistent, thereby generating a first screening result, including: First, obtain the object matching judgment result corresponding to each candidate association in the candidate association set. The object matching judgment result is used to characterize the consistency of the process object indicated by the semantic fragment of the description object with the primitives in the candidate association in terms of engineering semantics.

[0059] Then, for each candidate association, its corresponding primitive type information is read, and the primitive type is compared with the process object type reflected by the semantic fragment of the described object to determine whether the two belong to the same category or a category that can be reasonably associated at the engineering semantic level.

[0060] When it is determined that the type of process object indicated by the semantic fragment of the described object is inconsistent with the type of primitive in the candidate association relationship in terms of engineering semantics, the corresponding candidate association relationship is removed from the candidate association set.

[0061] Finally, the candidate relationships that were not eliminated are aggregated to generate the first screening results for subsequent processing.

[0062] In a preferred embodiment of the present invention, based on the working condition constraint determination result in the semantic parsing result, the associations in the first screening result are marked, and the associations that are only valid under specific working conditions are marked as conditional associations, generating working condition marking results, including: First, obtain the working condition constraint determination result corresponding to each relationship in the first screening result. The working condition constraint determination result is used to characterize whether the relationship is restricted by working conditions.

[0063] Then, based on the results of the working condition constraint determination, each association is classified and processed. For the association that is determined to be valid under any working condition, it is marked as an unconditional association.

[0064] For associations that are determined to be valid only when specific operating conditions are met, the corresponding operating condition information is recorded as an additional attribute, and the association is marked as a conditional association.

[0065] Finally, a working condition tagging result containing association relationship type tags and corresponding working condition information is generated, which is used for subsequent connection attribute filtering.

[0066] In a preferred embodiment of the present invention, based on the connection attribute determination result in the semantic parsing result, a second round of screening is performed on the working condition marking result to remove associations that do not constitute process flow connections, generating a second screening result, including: First, obtain the connection attribute determination result corresponding to each relationship in the working condition marking result. The connection attribute determination result is used to characterize whether the relationship has a clear process flow connection meaning.

[0067] Then, for each association, determine whether the result of its connection attribute determination indicates that the association constitutes an identifiable connection or operational relationship in the process flow.

[0068] When the connection attribute determination result indicates that the association does not have the meaning of process flow connection, the corresponding association will be removed from the working condition labeling result.

[0069] Finally, the relationships filtered by connection attributes are aggregated to generate a second filtering result for final relationship determination.

[0070] In a preferred embodiment of the present invention, when the same text annotation information corresponds to multiple associations in the second filtering result, based on the connection attribute determination result, the associations determined to constitute conventional process flow connections are retained, and a semantically corrected final association result is generated, including: First, for the same text annotation information, identify the multiple relationships it corresponds to in the second filtering results.

[0071] Then, read the connection attribute judgment results corresponding to the above multiple relationships respectively, and determine whether the connection or operation meaning reflected by each relationship conforms to the connection logic in the conventional process flow.

[0072] When only some of the multiple relationships are determined to constitute a connection in the regular process flow, those relationships are retained and the remaining relationships are removed.

[0073] When there are multiple relationships that all meet the connection conditions of the conventional process flow, all of them are retained as the final relationships that are established in parallel.

[0074] Finally, the retained relationships are aggregated to generate semantically corrected final relationship results, which are used for subsequent structured data construction.

[0075] In a preferred embodiment of the present invention, based on semantic units, a holistic engineering semantic parsing is performed on the semantic fragments describing the object, the semantic fragments of the working condition, and the semantic fragments of the behavior constraint to construct a unified semantic context covering the semantic units and generate semantic association analysis results, including: Based on semantic units, the first stage of parsing is performed on each semantic segment to identify explicit semantic reference relationships between the semantic segments and generate reference relationship results; Based on the referential relationship results, the second stage of parsing is performed on the limiting and functional relationships between each semantic segment to generate the initial semantic relationship structure; Based on the initial semantic relation structure, engineering semantic consistency constraints are introduced to verify and correct semantic relations with multiple or ambiguous pointers, and to generate a corrected semantic relation structure. Based on the corrected semantic relation structure, a unified semantic context covering the entire semantic unit is constructed; Semantic association analysis results are extracted from a unified semantic context to represent stable associations between semantic fragments describing objects, semantic fragments describing operating conditions, and semantic fragments defining behavior.

[0076] In this embodiment of the invention, by parsing each semantic fragment in a semantic unit in stages, explicit semantic referential relationships are first identified, followed by the parsing of limiting and functional relationships. Furthermore, engineering semantic consistency constraints are introduced to verify and correct ambiguous points of reference, ensuring that the semantic relationships between the described object, operating conditions, and behavioral constraints are uniformly organized within the same semantic framework. The resulting unified semantic context stably reflects the actual engineering relationships between the semantic fragments, providing clear and unambiguous semantic support for subsequent object matching, operating condition constraints, and connection attribute determination, thus avoiding misjudgments caused by unclear semantic references.

[0077] In a preferred embodiment of the present invention, based on semantic units, a first-stage parsing is performed on each semantic segment to identify explicit semantic referential relationships between the semantic segments and generate referential relationship results, including: First, the semantic fragments containing the descriptive object, working condition, and behavioral constraints of the semantic unit are obtained, and the text content of each semantic fragment is standardized to unify the referential method and semantic expression form.

[0078] Then, each semantic segment is parsed segment by segment to identify semantic reference information such as clearly appearing object referents, limited objects, or objects of action, and to determine whether the semantic reference clearly points to a certain semantic segment describing an object in the semantic unit.

[0079] When there is a semantic expression in the semantic fragment of working condition or the semantic fragment of behavior that clearly points to the semantic fragment of the described object, the corresponding referential relationship is recorded; when there is no clear referential relationship, the semantic fragment is marked as having no explicit referential relationship yet.

[0080] Finally, the referential relationships identified between the semantic segments are aggregated to generate referential relationship results for subsequent semantic relationship analysis.

[0081] In a preferred embodiment of the present invention, based on the referential relationship results, a second-stage parsing is performed on the limiting and functional relationships between each semantic segment to generate an initial semantic relationship structure, including: First, obtain the aforementioned referential relationship results, and use the semantic fragments of the described objects as the core nodes for semantic relationship analysis.

[0082] Then, for the semantic fragments of working conditions that have a referential relationship with the core node, we analyze their limiting role in engineering semantics, determine whether they are used to limit the state, attributes or applicable conditions of the described object, and establish the corresponding limiting relationship.

[0083] At the same time, for behavioral semantic fragments that have a referential relationship with core nodes, we analyze their role and meaning in engineering semantics, determine whether they are used to describe the connection method or operation behavior between the described objects, and establish the corresponding role relationship.

[0084] During this process, semantic fragments that cannot be directly established as limiting or action relationships through referential relationships will be temporarily retained as independent nodes.

[0085] Finally, the semantic fragments describing the object, the semantic fragments describing the working conditions, and the semantic fragments defining the behavior, as well as the limiting and interaction relationships established between them, are organized in a unified manner to generate an initial semantic relationship structure.

[0086] In a preferred embodiment of the present invention, based on the initial semantic relationship structure, engineering semantic consistency constraints are introduced to verify and correct semantic relationships with multiple or ambiguous pointers, generating a corrected semantic relationship structure, including: First, the initial semantic relationship structure is traversed to identify multiple pointing cases where a semantic fragment points to multiple semantic fragments of the descriptive object, or ambiguous pointing cases where the semantic pointing is unclear.

[0087] Then, engineering semantic consistency constraints are introduced to verify the above-mentioned multiple or ambiguous directions. The engineering semantic consistency constraints include consistency of process object type, rationality of engineering role, and rationality of semantic scope.

[0088] When a semantic relationship does not meet the engineering semantic consistency constraint, the semantic relationship is released, and the semantic fragment is re-evaluated to see if a reasonable semantic relationship can be established between the semantic fragment and other semantic fragments of the described object.

[0089] When a semantic relation satisfies both the engineering semantic consistency constraint and the referential relation constraint, the semantic relation is retained as a valid relation.

[0090] Finally, the initial semantic relation structure is verified and corrected to generate the corrected semantic relation structure.

[0091] In a preferred embodiment of the present invention, a unified semantic context covering the entire semantic unit is constructed based on the corrected semantic relation structure, including: First, the corrected semantic relationship structure is obtained, and the semantic fragments describing the objects in it are used as the core semantic nodes of the unified semantic context.

[0092] Then, the conditional semantic fragments that have a valid limiting relationship with the core semantic node, as well as the behavioral limiting semantic fragments that have a valid action relationship, are sequentially incorporated into the unified semantic context according to their semantic relationship hierarchy.

[0093] During the construction process, ensure that each semantic fragment retains only one explicit semantic reference in a unified semantic context, and that its semantic scope matches the engineering semantic consistency constraints.

[0094] Ultimately, a unified semantic context is formed that can fully reflect the internal semantic structure and engineering meaning of the semantic unit.

[0095] In a preferred embodiment of the present invention, semantic association analysis results, extracted from a unified semantic context, are used to characterize stable associations among semantic fragments describing objects, semantic fragments describing operating conditions, and semantic fragments defining behaviors. These results include: First, the unified semantic context is parsed to identify the limiting and functional relationships that have passed the engineering semantic consistency constraint verification.

[0096] Then, semantic relationships that remain stable in a unified semantic context and do not change with semantic ambiguity are extracted and used as stable associations between semantic fragments describing objects, semantic fragments describing working conditions, and semantic fragments describing behaviors.

[0097] Next, the extracted stable relationships are structured to clearly reflect the engineering semantic relationships between each semantic segment.

[0098] Finally, the above-mentioned structured stable relationships are output as semantic association analysis results, which are used for subsequent object matching, working condition constraint determination, and connection attribute determination.

[0099] In a preferred embodiment of the present invention, based on the semantic association analysis results, it is determined whether the process object indicated by the semantic fragment of the described object is consistent with the process equipment elements, pipeline elements, or instrument elements in the candidate association relationship in terms of engineering semantics, and an object matching determination result is generated, including: Based on the semantic association analysis results, object semantic features are extracted from the semantic fragments describing the object to represent the type, functional role and engineering attributes of the process object, and an object semantic feature set is generated. Based on the set of semantic features of objects, extract the corresponding graph element category, connection role and engineering attribute from the candidate association relationship to generate a set of graph element semantic features; The set of semantic features of objects and the set of semantic features of primitives are aligned using engineering semantics to generate an initial set of semantic matching relationships. Based on the initial set of semantic matching relationships, engineering semantic consistency constraints are introduced to eliminate matching relationships that do not meet the consistency of object type or engineering role, and generate a set of valid semantic matching relationships. Based on the set of valid semantic matching relationships, an object matching judgment result is generated to characterize the consistency of the semantic fragments of the described object with the primitives in the candidate association relationship in terms of engineering semantics.

[0100] In this embodiment of the invention, by extracting the process object type, functional role, and engineering attributes from the semantic fragments of the described object, and aligning them with the semantic features corresponding to the primitives in the candidate association relationships, the object matching process is transformed from single-name matching to multi-dimensional engineering semantic matching. Based on this, an engineering semantic consistency constraint is introduced to eliminate matching relationships that do not satisfy the consistency of object type or engineering role, thereby generating an object matching judgment result that truly reflects the consistency between the described object and the primitives at the engineering semantic level, ensuring that subsequent semantic parsing is based on reliable object correspondences.

[0101] In a preferred embodiment of the present invention, based on the semantic association analysis results, object semantic features for characterizing the process object type, functional role, and engineering attributes are extracted from the semantic fragments describing the object, generating an object semantic feature set, including: First, semantic relationship information directly related to the semantic fragments of the described object is obtained from the semantic association analysis results, which is used to clarify the semantic positioning of the described object in a unified semantic context.

[0102] Then, engineering semantic parsing is performed on the text content describing the semantic fragments of the objects to identify information used to indicate the categories of process objects, such as equipment, pipeline, or instrument categories, and to identify information used to reflect process functions or process roles.

[0103] Based on this, the engineering attribute information contained in the semantic fragments of the object is further analyzed, such as function, purpose, media type or operating characteristics, and the object type, functional role and engineering attribute identified above are recorded as object semantic features respectively.

[0104] Finally, the semantic features of various objects extracted from the semantic fragments of the same descriptive object are aggregated to generate a set of object semantic features, which is used for subsequent semantic matching processing.

[0105] In a preferred embodiment of the present invention, based on the object semantic feature set, the corresponding graph element category, connection role, and project attribute are extracted from the candidate association relationship to generate a graph element semantic feature set, including: First, obtain the candidate association relationships corresponding to the semantic fragments of the described object, and read the element type information of the process equipment elements, pipeline elements, or instrument elements involved in the candidate association relationships.

[0106] Then, based on the structural features and connection relationships of the graphic element in the chemical P&ID drawing, determine the connection role of the graphic element in the process flow, such as as a medium inflow end, outflow end or control node.

[0107] Next, engineering attribute information related to the graphic element is extracted from the graphic element recognition results and its associated engineering information, such as the functional attributes or technological uses of the object represented by the graphic element.

[0108] Finally, the extracted primitive categories, connection roles, and engineering attributes are uniformly organized to generate a primitive semantic feature set, which is used for subsequent alignment analysis with the object semantic feature set.

[0109] In a preferred embodiment of the present invention, the object semantic feature set and the primitive semantic feature set are subjected to engineering semantic alignment processing to generate an initial semantic matching relationship set, including: First, each semantic feature in the object semantic feature set and the primitive semantic feature set is compared item by item to determine whether the object type corresponds to the primitive category in engineering semantics.

[0110] Then, a correspondence analysis is performed on the functional roles in the semantic features of the object and the connection roles in the semantic features of the primitive to determine whether the semantic roles of the two have a reasonable correspondence in the process flow.

[0111] During this process, the consistency of the engineering attributes in the semantic features of the object and the semantic features of the primitive is judged to confirm whether there are any conflicts or inconsistencies between the two at the level of engineering attributes.

[0112] When the semantic features of an object and the semantic features of a graph primitive satisfy the semantic correspondence of engineering at the three levels of object type, engineering role, and engineering attribute, the semantic matching relationship between the two is recorded.

[0113] Finally, all semantic matching results that satisfy the above correspondence are aggregated to generate an initial set of semantic matching relationships.

[0114] In a preferred embodiment of the present invention, based on the initial semantic matching relationship set, an engineering semantic consistency constraint is introduced to eliminate matching relationships that do not satisfy object type consistency or engineering role consistency, thereby generating a valid semantic matching relationship set, including: First, the initial set of semantic matching relationships is traversed, and the consistency of object types in each semantic matching relationship is checked one by one to determine whether the described object and the graphic element belong to the same category or a category of process objects that are allowed to be associated in terms of engineering semantics.

[0115] Then, the consistency of engineering roles in each semantic matching relationship is verified to determine whether the functional roles reflected by the semantic features of the object match the connection roles reflected by the semantic features of the primitives in the process flow.

[0116] When a semantic matching relationship does not meet the requirements of object type consistency or project role consistency, the semantic matching relationship is removed from the initial semantic matching relationship set.

[0117] Finally, the semantic matching relationships verified by the engineering semantic consistency constraints will be aggregated to generate a set of valid semantic matching relationships, which will be used to generate the final object matching judgment result.

[0118] In a preferred embodiment of the present invention, an object matching determination result is generated based on a set of valid semantic matching relationships to characterize the consistency of semantic fragments of the described object with primitives in candidate association relationships in terms of engineering semantics, including: First, obtain the semantic matching relationships retained in the set of valid semantic matching relationships, and confirm the engineering semantic correspondence between the semantic fragments of the description objects and the primitives corresponding to each matching relationship.

[0119] Then, based on the completeness of the set of valid semantic matching relationships, it is determined whether the semantic fragments of the described object can completely correspond to the target primitives in the candidate association relationship at the engineering semantic level.

[0120] When there is a clear and unique valid semantic matching relationship between the semantic fragment of the described object and the primitive, an object matching judgment result is generated to characterize the consistency between the two in engineering semantics.

[0121] When there is no valid semantic matching relationship that satisfies the engineering semantic consistency constraint, the corresponding candidate association relationship is judged as object inconsistency and reflected in the object matching judgment result.

[0122] Finally, the output object matching determination result serves as an important input for subsequent working condition constraint determination and connection attribute determination.

[0123] In a preferred embodiment of the present invention, based on the object matching determination result and the semantic association analysis result, and in conjunction with the semantic fragment describing the object, the applicable premises limited by the semantic fragment of the working condition are interpreted to generate a working condition constraint determination result, including: When the object matching determination result shows that the semantic fragment of the descriptive object and the process object in the candidate association relationship have an engineering semantic basis, the semantic fragment of the working condition is parsed and processed. Based on the semantic association analysis results, determine the corresponding target of the working condition semantic fragment in the unified semantic context, and generate the working condition target determination result. Based on the determination of the object of the working condition, the semantic fragment of the working condition is analyzed by engineering semantics to distinguish it as object state constraint or connection effective condition, and the working condition constraint description result is generated. Based on the description of the working condition constraints, a working condition constraint determination result is generated to characterize whether the candidate association relationship is restricted by the working condition conditions.

[0124] In this embodiment of the invention, given a clear object matching relationship, the target of the semantic fragment of the working condition is determined by combining a unified semantic context, and the engineering meaning of the working condition is interpreted. This allows for the distinction between whether the working condition restricts the state of a process object or restricts the conditions for connection or operation to take effect. The resulting working condition constraint determination result not only clarifies whether candidate associations are restricted by working conditions, but also clarifies the specific way in which the restriction functions in the engineering semantics, thus providing a clear basis for working condition constraints for subsequent association screening and structured data construction.

[0125] In a preferred embodiment of the present invention, when the object matching determination result indicates that the semantic fragment describing the object and the process object in the candidate association relationship have an engineering semantic basis, the semantic fragment of the working condition is parsed and processed, including: First, obtain the object matching determination result and determine whether the determination result indicates that there is a valid correspondence between the semantic fragment describing the object and the process equipment elements, pipeline elements or instrument elements in the candidate association relationship in terms of engineering semantics.

[0126] When the object matching result indicates that the two do not have a basis for association in engineering semantics, further interpretation and processing of the corresponding working condition semantic fragments are terminated, and the candidate association is marked as unaffected by the working condition parsing.

[0127] When the object matching determination result shows that the two have a basis of association in engineering semantics, the semantic fragment of the working condition corresponding to the candidate association relationship is extracted and used as the processing object for subsequent working condition constraint interpretation.

[0128] In a preferred embodiment of the present invention, based on the semantic association analysis results, the corresponding action object of the working condition semantic fragment in a unified semantic context is determined, and a working condition action object determination result is generated, including: First, obtain the semantic association information between semantic fragments related to working conditions, semantic fragments related to the described objects, and semantic fragments related to behavioral constraints from the semantic association analysis results.

[0129] Then, based on the established limiting and function relationships in the unified semantic context, it is determined whether the semantic fragment of the working condition is used in engineering semantics to limit the process object indicated by the described object, or to limit the connection relationship between the described objects.

[0130] When a semantic fragment of a working condition is determined to be a semantic fragment that directly defines a certain descriptive object in the semantic association analysis results, the process object corresponding to that semantic fragment of the descriptive object is determined as the working condition action object.

[0131] When a conditional semantic fragment is determined to limit a certain connection or operation behavior, the candidate association corresponding to that connection or operation is identified as the object of the condition action.

[0132] Finally, the output is the result of determining the object of action of the working condition, which is used to characterize the semantic fragment of the working condition.

[0133] In a preferred embodiment of the present invention, based on the determination result of the object acting on the working condition, engineering semantic parsing is performed on the semantic fragment of the working condition to distinguish it as an object state constraint or a connection activation condition, and a working condition constraint description result is generated, including: First, obtain the result of the determination of the object affected by the working condition, and determine the object type limited by the semantic fragment of the working condition.

[0134] When the result of the determination of the working condition object indicates that the semantic fragment of the working condition condition limits a specific process object, the semantic fragment of the working condition condition is semantically interpreted and interpreted as a state constraint on the operating state, working mode or applicable scope of the process object.

[0135] When the result of the determination of the object of operation condition indicates that the semantic fragment of the operation condition defines a candidate association relationship or connection behavior, the semantic fragment of the operation condition is semantically interpreted as an effective condition for the connection or operation to take effect under specific conditions. The existence and semantic location of the connection have been determined in a unified semantic context through semantic association analysis.

[0136] After completing the above distinction, the engineering semantic parsing results of the working condition semantic fragments are organized into structured working condition constraint description results.

[0137] In a preferred embodiment of the present invention, based on the working condition constraint description result, a working condition constraint determination result is generated to characterize whether the candidate association relationship is restricted by working condition conditions, including: First, obtain the working condition constraint description results generated for the candidate association relationship, and determine whether they contain explicit working condition limitation information.

[0138] When the result of the working condition constraint description indicates that the candidate association is valid under any working condition, the working condition constraint judgment result corresponding to the candidate association is marked as not being restricted by the working condition conditions.

[0139] When the result of the working condition constraint description indicates that the candidate association relationship is only valid when the specific working condition conditions are met, the corresponding working condition information is associated with the candidate association relationship and the working condition constraint determination result is marked as being restricted by the working condition conditions.

[0140] Finally, the output is the result of the working condition constraint judgment for subsequent semantic consistency verification and filtering.

[0141] In a preferred embodiment of the present invention, based on the object matching determination result and the semantic association analysis result, and combined with the semantic fragments describing the object and the working condition, the meaning of the connection or operation described by the behavior-limiting semantic fragment is interpreted to generate a connection attribute determination result, including: When the object matching determination result indicates that the semantic fragment describing the object has an engineering semantic basis with the process object in the candidate association relationship, the behavior-limited semantic fragment is parsed and processed. Based on the semantic association analysis results, engineering semantic elements that represent the connection direction, connection method and operation type are extracted from the behavior-defined semantic fragments to generate an initial set of connection semantic elements. Based on the initial set of connection semantic elements, and combined with the judgment results of the working condition constraints corresponding to the working condition semantic fragments, the validity of the connection semantic elements is verified, and a set of valid connection semantic elements is generated. Based on the set of valid connection semantic elements, the connection or operation meanings corresponding to candidate associations are merged to generate connection attribute determination results that characterize the connection attributes of candidate associations.

[0142] In this embodiment of the invention, based on object matching and semantic association analysis, engineering semantic elements such as connection direction, connection method, and operation type contained in the behaviorally defined semantic fragment are extracted item by item. The validity of these semantic elements is then verified by combining the constraint results corresponding to the operating conditions. This ensures that the interpretation of the connection or operation meaning no longer relies solely on the textual description itself, but is simultaneously constrained by both object consistency and operating condition constraints. The resulting connection attribute determination result accurately reflects the true connection attributes corresponding to the candidate association at the engineering semantic level, avoiding the misidentification of behaviors that do not conform to operating conditions or object relationships as valid process connections.

[0143] In a preferred embodiment of the present invention, when the object matching determination result indicates that the semantic fragment describing the object and the process object in the candidate association relationship have an engineering semantic basis, the behavior-limited semantic fragment is parsed and processed, including: First, obtain the object matching judgment result and determine whether the semantic fragment describing the object and the corresponding process equipment elements, pipeline elements or instrument elements in the candidate association form a valid correspondence at the engineering semantic level.

[0144] When the object matching result indicates that there is no basis for association between the two in terms of engineering semantics, further parsing and processing of the corresponding behavioral semantic fragments is terminated, and the candidate association is marked as having no valid behavioral interpretation premise.

[0145] When the object matching determination result indicates that the two have a basis of association in engineering semantics, the behavioral limiting semantic fragment corresponding to the candidate association relationship is extracted and used as the parsing object for subsequent connection attribute interpretation.

[0146] In a preferred embodiment of the present invention, based on the semantic association analysis results, engineering semantic elements for characterizing connection direction, connection mode, and operation type are extracted from the behavior-defined semantic fragments to generate an initial set of connection semantic elements, including: First, obtain the semantic association information between the behavioral constraint semantic fragments, the descriptive object semantic fragments, and the working condition semantic fragments from the semantic association analysis results, in order to limit the scope of the behavioral semantics.

[0147] Then, engineering semantic analysis is performed on the text content of the behavior-defined semantic fragments to identify information used to describe the direction of medium flow, the order of connection, or the direction of action, as semantic elements related to the connection direction.

[0148] At the same time, information in the semantic fragments that define the behavior to describe the form of physical connection or control method, such as direct connection, indirect connection or connection through control elements, is analyzed and treated as semantic elements related to the connection method.

[0149] Furthermore, the content used to describe the meaning of process actions or operations in the behaviorally defined semantic fragments is analyzed and classified into semantic elements related to operation types.

[0150] Finally, the engineering semantic elements related to the connection direction, connection method, and operation type identified above are uniformly organized to generate an initial set of connection semantic elements.

[0151] In a preferred embodiment of the present invention, based on the initial set of connection semantic elements and combined with the determination result of the working condition constraint corresponding to the working condition semantic fragment, the connection semantic elements are validated to generate a set of valid connection semantic elements, including: First, obtain the initial set of semantic elements for connection, and then read the semantic elements related to connection direction, connection method, and operation type one by one.

[0152] Then, based on the working condition constraint judgment results of the corresponding candidate associations, it is determined whether each connection semantic element has the premise for establishment under the current working condition.

[0153] If a connection semantic element is only valid when a specific operating condition is met, and that operating condition is not supported by the operating condition constraint determination result, the connection semantic element is determined to be an invalid semantic element.

[0154] When a connection semantic element is consistent with the result of the working condition constraint at the engineering semantic level, the connection semantic element is determined to be a valid semantic element.

[0155] Finally, the connection semantic elements that have passed the validity check will be aggregated to generate a set of valid connection semantic elements.

[0156] In a preferred embodiment of the present invention, based on the set of valid connection semantic elements, the connection or operation meanings corresponding to candidate associations are merged to generate a connection attribute determination result for characterizing the connection attributes of candidate associations, including: First, obtain a set of valid connection semantic elements, and then organize the connection or operational meanings reflected by each semantic element in a unified manner.

[0157] Then, semantic elements belonging to the same connection or operation behavior are merged to form a consistent expression of connection direction, connection method and operation type at the engineering semantic level.

[0158] During the merging process, conflicting or inconsistent semantic elements of connection are excluded, and only connections or operational meanings that can be established together in terms of engineering semantics are retained.

[0159] Finally, the meaning of the merged connections or operations is output as the connection attribute determination result of the candidate association, which is used for subsequent semantic consistency verification and structured data construction.

[0160] In a preferred embodiment of the present invention, based on a set of semantic segments, and according to the order in which the semantic segments appear in the text annotation information, the semantic segments describing the object, the semantic segments describing the working conditions, and the semantic segments defining the behavior are combined to generate a structured semantic sequence, including: Obtain a set of semantic fragments generated for the same text annotation information, the set of semantic fragments including semantic fragments describing objects, semantic fragments of working conditions, and semantic fragments defining behaviors; Based on the original order of appearance of semantic segments in the text annotation information, the semantic segments in the semantic segment set are sorted to generate an ordered semantic segment sequence; In an ordered sequence of semantic fragments, the location of the semantic fragment describing the object is identified, and the semantic fragment describing the object is used as the core node of the semantic sequence. Based on the core node, the working condition semantic fragments and behavior limitation semantic fragments located before or after the semantic fragments of the described object are sequentially associated with the core node to form a structured semantic sequence around the same semantic fragment of the described object. When there are multiple semantic fragments describing objects in an ordered sequence of semantic fragments, each semantic fragment describing an object is used as a core node to generate multiple independent structured semantic sequences.

[0161] In this embodiment of the invention, the semantic fragments representing descriptive objects, operating conditions, and behavioral constraints generated from the same text annotation information are sorted according to their order of appearance in the original text. The descriptive object semantic fragments are then used as core nodes to structurally associate other semantic fragments, thus explicitly expressing the implicit semantic hierarchy in the text annotation. When the text annotation contains multiple descriptive objects, independent semantic sequences are generated for each, thereby avoiding semantic mixing between different process objects. This structured semantic sequence provides a clear and stable input structure for subsequent semantic parsing, helping to improve the accuracy and consistency of semantic parsing results in complex annotation scenarios.

[0162] In a preferred embodiment of the present invention, based on the graphic element recognition result, and according to the spatial adjacency and directional relationships between the text annotation information and process equipment graphic elements, pipeline graphic elements, and instrument graphic elements, multiple possible association relationships are established for each text annotation information to generate a candidate association set, including: Obtain the spatial location of each text annotation information in the graphic element recognition result, as well as the spatial location of process equipment graphic elements, pipeline graphic elements, and instrument graphic elements; Based on the relative positional relationship between the text annotation information and the process equipment elements, pipeline elements and instrument elements, identify the elements that meet the preset adjacency conditions with the text annotation information in space, and generate a set of adjacent elements; Based on the pointing symbols, leaders, or annotation direction information contained in the text annotation information, determine whether there is a pointing relationship between the text annotation information and each graphic element in the adjacent graphic element set, and generate a set of pointing association relationships; By combining the set of adjacent graphic elements with the set of pointing associations, multiple associations are established for each text annotation information, corresponding to different process equipment graphic elements, pipeline graphic elements, or instrument graphic elements, thus generating a candidate association set.

[0163] In this embodiment of the invention, by comprehensively utilizing text annotation information and the spatial positional relationships of various graphic elements in drawings, as well as the directional symbols, leader lines, or annotation direction information contained in the text annotations, a joint analysis is performed on the spatial adjacency and directional relationships between text annotations and process equipment elements, pipeline elements, and instrument elements. This ensures that the establishment of candidate associations possesses both spatial and directional rationality. The resulting candidate association set can comprehensively cover the potential process objects pointed to by the text annotations at the spatial level, providing a sufficient and structurally clear candidate basis for subsequent screening based on engineering semantics, and avoiding the omission of effective associations due to insufficient handling of spatial relationships.

[0164] In a preferred embodiment of the present invention, a preset adjacency condition is used to limit whether a valid spatial adjacency relationship exists between the text annotation information and process equipment elements, pipeline elements, or instrument elements. The method for setting this condition is as follows: First, based on the primitive recognition results, the spatial location range of each text annotation information and the spatial location range of each process equipment primitive, pipeline primitive and instrument primitive are obtained respectively, wherein the spatial location range is characterized by the boundary area of ​​the primitive in the drawing.

[0165] Then, using the spatial location range of the text annotation information as a reference area, a spatial search range for adjacency determination is determined. This spatial search range is obtained by extending the spatial location of the text annotation outwards by a preset distance. This preset distance is determined based on the overall scale of the drawing, the average size of the graphic elements, and common annotation habits between text annotations and graphic elements, ensuring that the spatial search range covers the graphic element area that the text annotation might point to under normal annotation conditions.

[0166] Based on this, it is determined whether at least a portion of the spatial location range of each process equipment element, pipeline element, or instrument element falls within the spatial search range. When the spatial location range of a certain element overlaps with or is adjacent to the spatial search range, it is determined that the element meets the adjacency condition in space and is included in the candidate adjacent elements.

[0167] Furthermore, to avoid excessively wide adjacency ranges due to dense local areas on the drawings, a relative position and orientation check is performed on the graphic elements that meet the preliminary adjacency conditions. This orientation check determines whether the orientation of the graphic element relative to the text annotation information is within the main directional direction of the text annotation or its reasonable offset range, thus excluding graphic elements located in obviously opposite or irrelevant directions.

[0168] Simultaneously, for pipeline elements, in addition to satisfying the aforementioned spatial adjacency and direction verification conditions, it is further determined whether there is a reasonable minimum spatial interval relationship between the text annotation information and the pipeline element. This determination is achieved by comparing whether the distance between the text annotation and the nearest position of the pipeline is less than a preset pipeline adjacency threshold, which is set based on the pipeline line width, drawing scale, and conventional annotation distance.

[0169] After making a comprehensive judgment on the spatial range, directional consistency and distance rationality, the process equipment elements, pipeline elements or instrument elements that simultaneously meet the above conditions are identified as elements that meet the preset adjacency conditions with the text annotation information, forming a corresponding set of adjacency elements.

[0170] The pre-defined adjacency conditions set in the above manner can comprehensively consider the spatial positional relationship, directional rationality, and engineering annotation habits between text annotations and graphic elements without relying on a single distance or direction judgment. This makes the adjacency judgment not only have spatial constraint certainty, but also adapt to different drawing layouts and annotation densities, thus providing a stable and feasible spatial foundation for subsequent directional relationship judgment and candidate association relationship construction.

[0171] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A chemical P&ID drawing structural extraction method based on a large language model, characterized in that, The method includes: Obtain chemical P&ID drawings, perform image analysis on the chemical P&ID drawings, and generate primitive recognition results that include the type of each primitive, the location of each primitive, and the spatial pointing relationship of the text annotations; Based on the primitive recognition results, according to the spatial adjacency and pointing relationships between the text annotation information and the process equipment primitives, pipeline primitives and instrument primitives, multiple possible association relationships are established for each text annotation information to generate a candidate association set; The text annotation information in the candidate association set is split into sentences at the level, and each text annotation information is split into multiple semantic fragments describing the object, working conditions and behavioral constraints. The semantic fragments are then combined with their corresponding candidate association relationships to form semantic analysis input data. The semantic analysis input data is input into the large language model, and the semantic fragments are subjected to engineering semantic parsing to obtain semantic parsing results. The semantic parsing results include object matching judgment results for characterizing whether each graphic element process object is consistent with the described object, working condition constraint judgment results for characterizing whether the association relationship is restricted by working conditions, and connection attribute judgment results for characterizing the connection attribute of the association relationship. Based on the semantic parsing results, the candidate association set is subjected to semantic consistency verification and screening to generate the final association results after semantic correction. Based on the final association results, construct structured chemical P&ID data containing process equipment attributes, pipeline connection relationships, and their effective conditions; The engineering semantic parsing process includes: Perform overall engineering semantic parsing on semantic fragments describing objects, semantic fragments of working conditions, and semantic fragments of behavioral constraints, construct a unified semantic context covering semantic units, and generate semantic association analysis results; Based on the semantic association analysis results, determine whether the process object indicated by the semantic fragment of the description object is consistent with the process equipment elements, pipeline elements or instrument elements in the candidate association relationship in terms of engineering semantics, and generate object matching determination results; Perform engineering semantic parsing on the semantic fragments of working condition conditions, distinguish them as object state constraints or connection activation conditions, and generate working condition constraint description results. Based on the description of the working condition constraints, a working condition constraint determination result is generated to characterize whether the candidate association relationship is restricted by the working condition conditions. Based on the object matching judgment results and semantic association analysis results, combined with the semantic fragments describing the object and the semantic fragments of the working condition, the meaning of the connection or operation described by the behavior limiting semantic fragments is interpreted, and the connection attribute judgment results are generated. Perform semantic consistency verification and filtering on the candidate association set, including: Based on the object matching judgment results in the semantic parsing results, the candidate associations in the candidate association set are screened in the first round to remove associations whose primitive types are inconsistent with the description object and the candidate associations, and the first screening results are generated. Based on the working condition constraint determination results in the semantic parsing results, the associations in the first screening results are marked, and the associations that only exist under specific working conditions are marked as conditional associations, generating working condition marking results; Based on the connection attribute determination results in the semantic parsing results, the working condition labeling results are filtered in the second round to remove the association relationships that do not constitute process flow connections, and the second filtering results are generated. When the same text annotation information corresponds to multiple relationships in the second filtering results, the relationships that are judged to constitute regular process flow connections are retained based on the connection attribute judgment results, and the final association results after semantic correction are generated.

2. The method for structured extraction of chemical P&ID drawings based on a large language model according to claim 1, characterized in that, The text annotation information in the candidate association set is split into sentences at the level, and each text annotation information is broken down into multiple semantic fragments describing the object, working conditions, and behavioral constraints. The semantic fragments are then combined with their corresponding candidate association relationships to form semantic analysis input data, including: Based on the text annotation information in the candidate association set, the text is parsed sentence by sentence to identify the object indication content used to indicate the process object, the condition limitation content used to limit the applicable premise, and the behavior description content used to describe the connection or operation, and the text parsing result is generated. Based on the text parsing results, the object indication content is converted into semantic fragments describing the object, the condition constraint content is converted into semantic fragments describing the working condition, and the behavior description content is converted into semantic fragments describing the behavior constraint, thus generating a set of semantic fragments; When the text annotation information contains multiple object indication contents, the text annotation information is split according to the number of object indication contents to generate multiple semantic fragments of description objects corresponding to different objects. Based on the set of semantic fragments, and in accordance with the order in which the semantic fragments appear in the text annotation information, the semantic fragments describing the object, the semantic fragments of the working condition, and the semantic fragments of the behavior limitation are combined to generate a structured semantic sequence. Based on each candidate association in the candidate association set, the structured semantic sequence is bound to the corresponding candidate association to generate semantic analysis input data for engineering semantic parsing.

3. The method for structured extraction of chemical P&ID drawings based on a large language model according to claim 1, characterized in that, Based on semantic units, a holistic engineering semantic parsing is performed on the semantic fragments describing the object, the semantic fragments of the working condition, and the semantic fragments of the behavior constraint. A unified semantic context covering the semantic units is constructed, and semantic association analysis results are generated, including: Based on semantic units, the first stage of parsing is performed on each semantic segment to identify explicit semantic reference relationships between the semantic segments and generate reference relationship results; Based on the referential relationship results, the second stage of parsing is performed on the limiting and functional relationships between each semantic segment to generate the initial semantic relationship structure; Based on the initial semantic relation structure, engineering semantic consistency constraints are introduced to verify and correct semantic relations with multiple or ambiguous pointers, and to generate a corrected semantic relation structure. Based on the corrected semantic relation structure, a unified semantic context covering the entire semantic unit is constructed; Semantic association analysis results are extracted from a unified semantic context to represent stable associations between semantic fragments describing objects, semantic fragments describing operating conditions, and semantic fragments defining behavior.

4. The method for structured extraction of chemical P&ID drawings based on a large language model according to claim 1, characterized in that, Based on the semantic association analysis results, it is determined whether the process object indicated by the semantic fragment of the described object is consistent with the process equipment elements, pipeline elements, or instrument elements in the candidate association relationship in terms of engineering semantics, and object matching determination results are generated, including: Based on the semantic association analysis results, object semantic features are extracted from the semantic fragments describing the object to represent the type, functional role and engineering attributes of the process object, and an object semantic feature set is generated. Based on the set of semantic features of objects, extract the corresponding graph element category, connection role and engineering attribute from the candidate association relationship to generate a set of graph element semantic features; The set of semantic features of objects and the set of semantic features of primitives are aligned using engineering semantics to generate an initial set of semantic matching relationships. Based on the initial set of semantic matching relationships, engineering semantic consistency constraints are introduced to eliminate matching relationships that do not meet the consistency of object type or engineering role, and generate a set of valid semantic matching relationships. Based on the set of valid semantic matching relationships, an object matching judgment result is generated to characterize the consistency of the semantic fragments of the described object with the primitives in the candidate association relationship in terms of engineering semantics.

5. The method for structured extraction of chemical P&ID drawings based on a large language model according to claim 1, characterized in that, Based on the object matching determination results and semantic association analysis results, and combined with the semantic fragments describing the object and the semantic fragments describing the working conditions, the meaning of the connection or operation described by the behavioral limitation semantic fragments is interpreted, and connection attribute determination results are generated, including: When the object matching determination result indicates that the semantic fragment describing the object has an engineering semantic basis with the process object in the candidate association relationship, the behavior-limited semantic fragment is parsed and processed. Based on the semantic association analysis results, engineering semantic elements that represent the connection direction, connection method and operation type are extracted from the behavior-defined semantic fragments to generate an initial set of connection semantic elements. Based on the initial set of connection semantic elements, and combined with the judgment results of the working condition constraints corresponding to the working condition semantic fragments, the validity of the connection semantic elements is verified, and a set of valid connection semantic elements is generated. Based on the set of valid connection semantic elements, the connection or operation meanings corresponding to candidate associations are merged to generate connection attribute determination results that characterize the connection attributes of candidate associations.

6. The method for structured extraction of chemical P&ID drawings based on a large language model according to claim 2, characterized in that, Based on the set of semantic fragments, and according to the order in which the semantic fragments appear in the text annotation information, the semantic fragments describing the object, the semantic fragments of the working condition, and the semantic fragments of the behavior constraint are combined to generate a structured semantic sequence, including: Obtain a set of semantic fragments generated for the same text annotation information, the set of semantic fragments including semantic fragments describing objects, semantic fragments of working conditions, and semantic fragments defining behaviors; Based on the original order of appearance of semantic segments in the text annotation information, the semantic segments in the semantic segment set are sorted to generate an ordered semantic segment sequence; In an ordered sequence of semantic fragments, the location of the semantic fragment describing the object is identified, and the semantic fragment describing the object is used as the core node of the semantic sequence. Based on the core node, the working condition semantic fragments and behavior limitation semantic fragments located before or after the semantic fragments of the described object are sequentially associated with the core node to form a structured semantic sequence around the same semantic fragment of the described object. When there are multiple semantic fragments describing objects in an ordered sequence of semantic fragments, each semantic fragment describing an object is used as a core node to generate multiple independent structured semantic sequences.

7. The method for structured extraction of chemical P&ID drawings based on a large language model according to claim 1, characterized in that, Based on the primitive recognition results, and according to the spatial adjacency and directional relationships between the text annotation information and process equipment primitives, pipeline primitives, and instrument primitives, multiple possible association relationships are established for each text annotation information, generating a candidate association set, including: Obtain the spatial location of each text annotation information in the graphic element recognition result, as well as the spatial location of process equipment graphic elements, pipeline graphic elements, and instrument graphic elements; Based on the relative positional relationship between the text annotation information and the process equipment elements, pipeline elements and instrument elements, identify the elements that meet the preset adjacency conditions with the text annotation information in space, and generate a set of adjacent elements; Based on the pointing symbols, leaders, or annotation direction information contained in the text annotation information, determine whether there is a pointing relationship between the text annotation information and each graphic element in the adjacent graphic element set, and generate a set of pointing association relationships; By combining the set of adjacent graphic elements with the set of pointing associations, multiple associations are established for each text annotation information, corresponding to different process equipment graphic elements, pipeline graphic elements, or instrument graphic elements, thus generating a candidate association set.

Images