A geotechnical engineering investigation report compliance conflict detection method combined with large model semantic vector enhancement and related device

Abstract

The application discloses a geotechnical engineering investigation report compliance conflict detection method combined with large model semantic vector enhancement, a geotechnical engineering investigation report compliance conflict detection device, a detection equipment and a computer readable storage medium. The method obtains a to-be-detected investigation report and basic data, performs hierarchical analysis on the report and establishes data association; performs engineering object identification and semantic standardization on the analysis content, constructs a report evidence set and a project condition portrait; filters applicable specification clauses according to the project condition, performs semantic analysis and vector enhancement on the report evidence and the specification clauses by using a large language model; generates conflict candidate relationships based on engineering objects, space-time ranges and reference relationships, and determines report internal conflicts and specification applicable conflicts through numerical value, unit, applicable condition and logical relationship review, and outputs associated evidence, specification basis and source positions. The application realizes positioning and tracing of the investigation report compliance conflicts.

Description

Technical Field

[0001] This invention belongs to the field of digital technology of geotechnical engineering investigation results, and more specifically, relates to a method for detecting compliance conflicts in geotechnical engineering investigation reports by combining large model semantic vector enhancement, a device for detecting compliance conflicts in geotechnical engineering investigation reports, detection equipment, and a computer-readable storage medium. Background Technology

[0002] Geotechnical investigation reports are crucial technical documents reflecting site engineering geological conditions, groundwater status, geotechnical parameters, and foundation recommendations. Their content typically draws from various sources, including drilling records, in-situ tests, laboratory experiments, geophysical exploration results, and engineering design conditions, and is presented in different formats such as text, tables, figures, and appendices. As investigation projects expand in scale and report content increases, report quality review has gradually shifted from simple numerical verification to a comprehensive review of cross-chapter data, engineering evaluation conclusions, and the applicability of relevant standards.

[0003] In related technologies, errors in reports are mainly found through manual item-by-item review, keyword retrieval, or fixed field comparison. However, different chapters often use different names, sentence structures, and evaluation perspectives to describe the same strata, parameters, or engineering conditions. Moreover, the standard clauses have specific scope of application and preconditions, making it difficult for traditional methods to establish semantic connections between different contents. This can easily lead to the omission of implicit conflicts such as inconsistencies between internal report data and conclusions, mismatches between engineering recommendations and evaluation basis, and improper application of standard clauses.

[0004] Therefore, there is a need for a detection method that can structurally link survey reports and basic engineering data, select applicable regulatory clauses based on project conditions, and identify conflicts through semantic relationships and engineering constraints, so as to achieve accurate location and traceable review of compliance conflicts. Summary of the Invention

[0005] The purpose of this application is to provide a method, device, and computer-readable storage medium for detecting compliance conflicts in geotechnical engineering investigation reports that combines large-model semantic vector enhancement. This method identifies conflicts through semantic relationships and engineering constraints, thereby achieving accurate location and traceable review of compliance conflicts.

[0006] To address the aforementioned deficiencies or improvement needs of existing technologies, this invention provides a method for detecting compliance conflicts in geotechnical engineering investigation reports by combining large-model semantic vector enhancement, comprising: S1, obtain the geotechnical engineering investigation report to be tested and the corresponding basic engineering data, perform layered analysis on the geotechnical engineering investigation report to be tested, and establish the source association relationship between the basic engineering data and the content of the analyzed report to obtain a set of original content units with source location information; S2, perform geotechnical engineering object identification and engineering semantic standardization processing on the original content unit set, extract engineering objects, object attributes, attribute values ​​or evaluation conclusions, applicable conditions and source information to form a report evidence set, and form a project condition profile based on the engineering basic data; S3. The normative documents on which the report is prepared and reviewed are divided into normative clause units. Based on the matching relationship between the applicable conditions of the normative clause units and the project condition profile, a set of normative clauses applicable to the project is obtained. The report evidence set and the set of normative clauses applicable to the project are subjected to engineering semantic parsing and vectorization processing using a large language model. The engineering object information, object attribute information, project condition information, numerical and measurement unit information, spatiotemporal range information, and semantic polarity information are integrated to obtain a set of enhanced semantic vectors for the report and a set of enhanced semantic vectors for the normative clauses. S4. The report evidence set is grouped according to the project object, the object attribute, the spatiotemporal range and the reference relationship, and a conflict candidate relationship set is generated according to the semantic proximity between the report enhanced semantic vectors and between the report enhanced semantic vectors and the normative clause enhanced semantic vectors. S5, perform consistency checks on the set of conflict candidate relationships in terms of engineering object consistency, numerical and measurement unit consistency, spatiotemporal range consistency, applicable conditions consistency, and logical relationship consistency, determine internal conflicts in the report or conflicts between the report and standard clauses, and output the conflict type, related evidence, standard basis, source location, and review priority.

[0007] Optionally, in step S1, the geotechnical engineering investigation report to be tested is subjected to layered analysis, including: According to the document hierarchy and content carrier, the geotechnical engineering investigation report to be tested is divided into title unit, paragraph unit, item unit, table unit, figure description unit and attachment reference unit; Record the chapter name, page number, paragraph position, table position, or figure position corresponding to each original content unit; Extract exploration point identifiers, stratigraphic identifiers, sample identifiers, test depths, and observation times from the engineering basic data, and establish source association relationships between the engineering basic data and the corresponding original content units based on the exploration point identifiers, stratigraphic identifiers, sample identifiers, test depths, or observation times.

[0008] Optionally, in step S2, geotechnical engineering object identification and engineering semantic standardization processing are performed on the original set of content units, including: Extract the engineering object, object attributes, attribute values ​​or evaluation conclusions, spatial range, time range, applicable conditions, and data source type from each of the original content units; Establish a mapping relationship between the object name, object abbreviation, object number, and text description corresponding to the same project object; Standardize the names of soil and rock conditions, units of measurement, depth benchmarks, elevation benchmarks, and synonyms; Transform engineering descriptions with spatial, temporal, or observational limitations into report evidence that retains the corresponding limitations.

[0009] Optionally, the project condition profile includes the project location, exploration stage, engineering type, site elevation, geomorphic unit, stratigraphic composition, groundwater conditions, earthquake-related conditions, seasonal permafrost conditions, and special soil and rock conditions. Accordingly, in S3, the matching relationship between the applicable conditions of the standard clause unit and the project condition profile includes: matching the applicable engineering type, applicable survey stage, applicable site conditions and prerequisites corresponding to each standard clause unit with the project condition profile, determining the standard clause unit that meets the matching conditions as the applicable standard clause for the project, and excluding the standard clause unit that does not meet the matching conditions from the conflict detection range.

[0010] Optionally, in step S3, the engineering semantic parsing and vectorization processing using the large language model includes: Using the large language model, the subject, evaluation attributes, status or value, conditions for establishment, spatial range, temporal range, affirmative relationship, negative relationship and conditional relationship are extracted from each of the report evidence and each of the applicable normative clauses of the project to generate a standardized semantic description; The standardized semantic description is vectorized to obtain the basic semantic vector; The report enhanced semantic vector and the specification clause enhanced semantic vector are obtained by fusing the engineering object category, object attribute category, project condition matching result, numerical value and unit of measurement, spatial range, time range, semantic polarity, data source type and chapter function information with the basic semantic vector.

[0011] Optionally, in step S4, the report evidence belonging to the same project object or having a data reference relationship is divided into the same related evidence group according to the project object, the object attribute, the stratigraphic identifier, the exploration point identifier, the spatial range, and the reference relationship; Within each of the aforementioned related evidence groups, based on the semantic proximity between the report-enhanced semantic vectors, candidate relationships for numerical inconsistency, candidate relationships for state description inconsistency, candidate relationships for inconsistency between basic data and parameter statistical results, candidate relationships for inconsistency between parameter statistical results and engineering evaluation conclusions, and candidate relationships for inconsistency between engineering evaluation conclusions and engineering recommendations are generated. Based on the semantic similarity between the enhanced semantic vector of the report and the enhanced semantic vector of the normative clause, candidate relationships of inconsistency between project conditions and applicable conditions of normative clauses, and candidate relationships of inconsistency between report conclusions and preconditions of normative clauses are generated. For report evidence that differs in the engineering object, has no overlap in the spatial range, does not correspond in the time range, or has inconsistent applicable conditions, no conflict candidate relationship is generated.

[0012] Optionally, in step S5, the review of the set of conflict candidate relationships includes: For numerical conflict candidate relationships, the unit of measurement is unified, and the original measured value, statistical value, suggested value and design adopted value are distinguished according to the data source type. After excluding unit conversion, statistical processing, value range and data version differences, numerical conflicts are determined. For state-related conflict candidate relationships, based on the corresponding engineering object, stratum range, depth range and applicable conditions, it is determined whether different state descriptions can be established simultaneously, and when the different state descriptions cannot be established simultaneously under the same conditions, a state description conflict is determined. For candidate relationships of conflicting conclusions, an evidence chain is established from the basic engineering data, the statistical results of the parameters, the engineering evaluation conclusions to the engineering recommendations. When the engineering objects, parameters or applicable conditions cited in the subsequent evidence are inconsistent with the preceding evidence, a data-conclusion conflict is determined. For candidate relationships of normative conflicts, the applicable conditions and prerequisites of the corresponding normative clause units are reviewed according to the project condition profile. When the normative clauses used in the geotechnical engineering investigation report to be tested do not match the project condition profile, or when the report conclusions are inconsistent with the necessary, prohibited, or conditional relationships in the applicable normative clauses of the project, a normative application conflict is determined. The review priority is determined based on the position of each conflict outcome in the chain of evidence and the extent to which it is referenced in subsequent reports.

[0013] This application also provides a geotechnical engineering investigation report compliance conflict detection device that combines large model semantic vector enhancement, including: The data parsing module is used to obtain the geotechnical engineering investigation report to be tested and the corresponding basic engineering data, perform layered parsing on the geotechnical engineering investigation report to be tested, and establish the source association relationship between the basic engineering data and the content of the parsed report to obtain a set of original content units with source location information; The report standardization processing module is used to perform geotechnical engineering object identification and engineering semantic standardization processing on the original content unit set, extract engineering objects, object attributes, attribute values ​​or evaluation conclusions, applicable conditions and source information, form a report evidence set, and form a project condition profile based on the engineering basic data; The report vectorization module is used to divide the normative documents on which the report is prepared and reviewed into normative clause units. Based on the matching relationship between the applicable conditions of the normative clause units and the project condition profile, a set of applicable normative clauses for the project is obtained. The report evidence set and the set of applicable normative clauses for the project are subjected to engineering semantic parsing and vectorization processing using a large language model. The module also integrates engineering object information, object attribute information, project condition information, numerical and measurement unit information, spatiotemporal range information, and semantic polarity information to obtain a report enhanced semantic vector set and a normative clause enhanced semantic vector set. The conflict candidate module is used to group the report evidence set according to the project object, the object attribute, the spatiotemporal range and the reference relationship, and generate a conflict candidate relationship set according to the semantic similarity between the report enhanced semantic vectors and between the report enhanced semantic vectors and the specification clause enhanced semantic vectors; The conflict result output module is used to review the consistency of the candidate conflict relationships in terms of engineering objects, numerical values ​​and units of measurement, spatiotemporal range, applicable conditions, and logical relationships, to determine whether there is an internal conflict in the report or a conflict between the report and the standard clauses, and to output the conflict type, related evidence, standard basis, source location, and review priority.

[0014] This application also provides a testing device, including: Memory, used to store computer programs; A processor is used to implement the steps of the geotechnical engineering investigation report compliance conflict detection method as described above when executing the computer program.

[0015] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the geotechnical engineering investigation report compliance conflict detection method as described above.

[0016] It has the following beneficial effects: By performing layered analysis and source association on the survey report and basic engineering data, the technical content scattered in the main text, tables and appendices is transformed into report evidence with engineering objects, applicable conditions and source locations. Then, by combining project conditions to screen applicable standard clauses, and using large model semantic vectors to establish cross-chapter semantic relationships, conflict candidate relationships are formed. Finally, by reviewing objects, values, spatiotemporal ranges and logical relationships, internal conflicts in the report and conflicts in the application of standards can be accurately identified, and the location, tracing and hierarchical review of conflict results can be achieved. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0018] Figure 1 A flowchart illustrating a method for detecting compliance conflicts in geotechnical engineering investigation reports that combines large model semantic vector enhancement, provided as an embodiment of this application; Figure 2 A schematic diagram of a geotechnical engineering investigation report compliance conflict detection device combining large model semantic vector enhancement provided in this application embodiment; Figure 3 This is a schematic diagram of the structure of a testing device provided in an embodiment of this application. Detailed Implementation

[0019] The purpose of this application is to provide a method, device, and computer-readable storage medium for detecting compliance conflicts in geotechnical engineering investigation reports that combines large-model semantic vector enhancement. This method identifies conflicts through semantic relationships and engineering constraints, thereby achieving accurate location and traceable review of compliance conflicts.

[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0021] The following example illustrates a method for detecting compliance conflicts in geotechnical engineering investigation reports that combines large model semantic vector enhancement, as provided in this application.

[0022] Please refer to Figure 1 , Figure 1 This is a flowchart illustrating a method for detecting compliance conflicts in geotechnical engineering investigation reports that combines large model semantic vector enhancement, as provided in an embodiment of this application.

[0023] In this embodiment, the method may include: S1. Obtain the geotechnical engineering investigation report to be tested and the corresponding basic engineering data. Perform layered analysis on the geotechnical engineering investigation report to be tested, and establish the source correlation between the basic engineering data and the content of the analyzed report to obtain a set of original content units with source location information.

[0024] First, the geotechnical engineering investigation report to be tested and the corresponding basic engineering data are imported into the report compliance conflict detection system. The geotechnical engineering investigation report to be tested can be an editable electronic document or a formatted document converted from a paper report. For formatted documents, the text, tables, and figure descriptions are extracted first; for editable electronic documents, the document title hierarchy, body paragraphs, numbered entries, table structure, and figure reference relationships are directly read.

[0025] The system performs layered parsing of the geotechnical engineering investigation report to be tested, based on document hierarchy and content carrier. The content types generated by the layered parsing include title units, paragraph units, item units, table units, figure description units, and attachment reference units.

[0026] Title sections are used to define the technical aspects of a report's chapter structure and content. For example, the "Site Engineering Geological Conditions" section corresponds to basic geological descriptions, the "Geotechnical Parameter Statistics" section corresponds to data processing results, the "Site Evaluation" section corresponds to engineering analysis conclusions, and the "Foundation Recommendations" section corresponds to design recommendations. By recording chapters, it's possible to differentiate between basic data, statistical results, evaluation conclusions, and engineering recommendations in subsequent processing, avoiding the direct misclassification of content with similar technical meanings but different functions as duplicate data.

[0027] Paragraph units are divided according to natural paragraphs and stratigraphic description boundaries. When a natural paragraph describes multiple engineering geological layers, the system further breaks it down based on stratigraphic number, stratigraphic name, and descriptive conversion terms, ensuring that each paragraph unit corresponds to a single engineering object as closely as possible. When a natural paragraph contains both groundwater observation conditions and groundwater evaluation conclusions, the system retains the intra-sentence relationship between the two to facilitate subsequent determination of whether the evaluation conclusions exceed the observation conditions.

[0028] The entries are divided according to the numbered clauses, sub-recommendations, and precautions in the report. For example, the selection of bearing stratum, recommendations on foundation type, and precautions for foundation pit excavation in the foundation recommendations are each formed as independent entry units.

[0029] The table cells are parsed according to table name, table header, row headings, and column headings. For geotechnical parameter statistics tables, the system records the row and column positions of the stratum number, parameter name, statistical value, and suggested value. For indoor test results tables, the system establishes in-row associations for sample number, sampling depth, stratum number, test item, and test result. This avoids the loss of the corresponding stratum, sample, and test item when only a single value is extracted.

[0030] The drawing description unit includes descriptions of engineering geological profiles, borehole columnar sections, exploration point layout plans, and other textual annotations within the drawings. The appendix reference unit records references to test result tables, borehole records, geophysical reports, and design condition data within the main text.

[0031] After the content is divided, the system records the source location information for each original content unit. This source location information includes the document name, chapter name, page number, paragraph position, table name, table row and column position, figure name, figure label position, and original text. For content further subdivided within the same paragraph, the system also records the start and end positions of that content within the original paragraph.

[0032] Subsequently, the system reads the basic engineering data and extracts the exploration point identifier, stratigraphic identifier, sample identifier, test depth, and observation time. The exploration point identifier can be the borehole number, well number, or in-situ test point number; the stratigraphic identifier can be the engineering geological layer number, stratigraphic code, or stratigraphic name; the sample identifier can be the soil sample number, rock sample number, or water sample number; the test depth includes the sampling depth, test depth, groundwater observation depth, and geophysical interpretation depth; the observation time includes the drilling date, groundwater observation date, in-situ test date, and laboratory test date.

[0033] The system establishes source relationships between basic engineering data and report content based on exploration point identifiers, stratigraphic identifiers, sample identifiers, test depths, or observation times. For example, the condition description of a silty clay layer in the report text is linked to borehole records and laboratory test results belonging to that stratigraphic layer; the recommended value of the compression modulus in the geotechnical parameter statistics table is linked to the sample data used to form that recommended value; and the groundwater assessment conclusions are linked to the water level observation records, observation depths, and observation dates of the corresponding exploration points.

[0034] When the report text and basic engineering data use different numbering methods, the system uses exploration point coordinates, sampling depths, stratigraphic names, and test items for auxiliary matching. Source associations are established only when the engineering object and location conditions correspond. Through this process, a set of original content units with source location information is obtained, providing a data foundation for subsequent engineering object identification and conflict tracing.

[0035] S2 performs geotechnical engineering object identification and engineering semantic standardization processing on the original content unit set, extracts engineering objects, object attributes, attribute values ​​or evaluation conclusions, applicable conditions and source information, forms a report evidence set, and forms a project condition profile based on basic engineering data.

[0036] In this step, the system processes the original content unit set in sequence, extracting the project object, object attributes, attribute values ​​or evaluation conclusions, spatial range, time range, applicable conditions and data source type from each original content unit.

[0037] The engineering objects include the project, site, exploration points, strata, rock mass, groundwater, samples, test items, engineering parameters, bearing stratum, slopes, foundation pits, and proposed buildings and structures. Object attributes include stratum name, soil and rock condition, burial depth, thickness, elevation, water content, density, weathering degree, bearing capacity, compressibility, corrosivity, frost heave, and engineering suitability.

[0038] For stratigraphic descriptions, the system extracts stratigraphic number, stratigraphic name, distribution range, top elevation, bottom elevation, burial depth, thickness, state, compaction degree, water-bearing state, and engineering properties. For groundwater information, the system extracts groundwater type, initial water level, stable water level, observation time, observation depth, influence range, and evaluation conclusions. For experimental results, the system extracts sample number, sampling depth, stratigraphic stratum, test items, test results, and units of measurement. For engineering recommendations, the system extracts the recommended bearing stratum, foundation type, parameter values, applicable buildings and structures, and construction control conditions.

[0039] The system establishes a mapping relationship between object name, object abbreviation, object number, and text description for the same engineering object. For example, the terms "Layer ③ Silty Clay," "Layer ③ Silty Clay," and "Silty Clay Layer" appearing in the report, along with their corresponding stratigraphic codes, are uniformly mapped to the same engineering object if they can be identified as belonging to the same stratigraphic layer through their location, depth range, and relationship with adjacent layers. If stratigraphic layers with the same name but different spatial distributions exist in different exploration areas, the area identifier is retained, and the two are not merged.

[0040] The system standardizes the names of soil and rock conditions, units of measurement, depth benchmarks, elevation benchmarks, and synonyms. Soil and rock condition names are unified according to engineering semantics, but the original condition terms and their defined scopes are retained. For example, a condition that is generally plastic and locally stiff plastic cannot be simply categorized as a stiff plastic condition; instead, the overall scope, local scope, and corresponding condition should be recorded separately. Similarly, the transition from strongly weathered to moderately weathered rock strata cannot be directly considered identical to moderately weathered rock strata; the transition conditions must be preserved.

[0041] For parameters with different units of measurement but the same technical meaning, the system converts them to a unified unit while saving both the original value and the original unit. This conversion process is only used for subsequent comparisons and does not overwrite the original report data. For depth and elevation, the system first identifies whether the data uses the depth below the borehole opening, the site design elevation, or the national elevation datum, and then converts it into a comparable spatial range. If the report does not specify an elevation datum, the data is marked as a datum to be verified and is not directly used for conflict determination with other elevation data.

[0042] For engineering descriptions with defined conditions, the system converts them into report evidence that retains the spatial scope, temporal scope, and observation conditions. For example, the absence of stable groundwater during the exploration period is resolved as groundwater observation evidence. Here, the engineering object is groundwater, the object attribute is stable water level observation results, the attribute value is "not observed," the spatial scope is the completed drilling points and drilling depths, the temporal scope is the exploration period, and the applicable conditions are the corresponding observation method and observation duration. This evidence cannot be directly equated to the long-term absence of groundwater at the site.

[0043] For example, the statement that a stratum is generally in a plastic state and locally in a stiff plastic state is interpreted as two reports with an inclusion relationship. One report corresponds to the overall stratum extent and the plastic state, while the other corresponds to the local extent and the stiff plastic state. If subsequent engineering recommendations suggest using the parameters corresponding to the stiff plastic state to represent the entire stratum, the system can generate conflict candidate relationships based on the differences between the local and overall extents.

[0044] Each report's evidence includes the project object, object attributes, attribute values ​​or evaluation conclusions, spatial scope, temporal scope, applicable conditions, source location, data source type, and evidence reliability indicator. Data source types are categorized according to original survey data, parameter statistical results, project evaluation conclusions, and project recommendations. Original drilling records, in-situ test records, laboratory test results, and formal design conditions are considered basic data sources; parameter statistical tables are considered processing result sources; site evaluations are considered analytical conclusion sources; and foundation and construction recommendations are considered project recommendation sources.

[0045] While forming the report evidence set, the system generates a project condition profile based on the basic engineering data. The project condition profile includes the project location, exploration stage, engineering type, site elevation, geomorphic units, stratigraphic composition, groundwater conditions, earthquake-related conditions, seasonal permafrost conditions, and special geotechnical conditions.

[0046] This embodiment corresponds to a project in the high-altitude region of Qinghai. Therefore, site elevation, seasonal permafrost, seasonal variations in groundwater, elevation datum, and special soil and rock conditions such as saline soil are all recorded as independent condition items. If the engineering foundation data indicates that a certain special soil or rock is only locally distributed within the site, the project condition profile also records the corresponding area and depth range of that special soil or rock, avoiding the extension of local applicable conditions to the entire site.

[0047] The project profile also records the type of proposed building structures, the stage of the survey, and the required foundation type. This information is used to differentiate the regulatory requirements for different project categories and different stages of the survey, ensuring that subsequent selection of regulatory clauses does not solely rely on textual similarity.

[0048] After the above processing, S2 outputs a report evidence set and a project condition profile. The report evidence set provides standardized input for subsequent semantic vector processing, while the project condition profile provides a conditional basis for judging the applicability of regulatory clauses.

[0049] S3 divides the normative documents on which the report is prepared and reviewed into normative clause units. Based on the matching relationship between the applicable conditions of the normative clause units and the project condition profile, a set of normative clauses applicable to the project is obtained. Using a large language model, engineering semantic parsing and vectorization processing are performed on the report evidence set and the set of normative clauses applicable to the project, respectively. The information of engineering objects, object attributes, project conditions, numerical and measurement units, spatiotemporal range, and semantic polarity are integrated to obtain the report enhanced semantic vector set and the normative clause enhanced semantic vector set.

[0050] In this step, the system retrieves the regulatory documents upon which the report preparation and review are based. These documents include current technical standards, local technical regulations, enterprise technical regulations, and internal review rules confirmed by the enterprise, all relevant to geotechnical engineering investigation. To avoid using outdated or inapplicable regulatory content, the document name, version information, implementation status, applicable region, and applicable project type are recorded during import.

[0051] The system divides the specification document into specification clause units according to the requirements of articles, clauses, items, and appendices. Each specification clause unit maintains complete conditional relationships as much as possible. When a clause simultaneously includes applicable conditions, technical requirements, and exceptions, these three are not separated into unrelated contents, but rather their conditional chain is preserved.

[0052] For each clause unit, the system extracts the applicable project type, applicable survey stage, applicable site conditions, prerequisites, constraints, attributes to be verified, data or evaluation conclusions to be generated, and the logical relationships between clauses. The logical relationships between clauses include necessary relationships, prohibitive relationships, and conditional relationships.

[0053] The "necessary" relationship indicates that the report must include specific data, analytical processes, or evaluation conclusions when the corresponding project conditions are met. The "prohibited" relationship indicates that a certain conclusion or treatment method must not be adopted under specific conditions. The "conditional" relationship indicates that subsequent technical requirements apply only if the preconditions are met.

[0054] The system matches the applicable engineering type, applicable survey stage, applicable site conditions, and prerequisites of each specification clause unit with the project condition profile item by item. Specification clause units that meet the matching conditions are identified as applicable specification clauses for the project, while specification clause units that do not meet the matching conditions are excluded from the conflict detection scope of the current project.

[0055] For example, clauses related to seasonally frozen soil only enter the set of applicable code clauses when the project condition profile indicates a need for seasonally frozen soil assessment at the site. Clauses applicable only to a certain type of project or a specific investigation stage are not used as a basis for conflict determination if they do not match the current project type or investigation stage. Clauses involving special soil and rock types also need to be matched with the distribution area and depth range of the special soil and rock types to avoid applying code requirements corresponding to local conditions to the entire site.

[0056] After screening the regulatory clauses, the system uses a locally deployed large language model to process the report evidence and the applicable regulatory clauses for the project. The large language model runs in the enterprise's internal computing environment, and the report data and basic engineering data are processed within the internal environment without sending project files to external services.

[0057] The large language model does not directly determine whether a report is compliant. Instead, it performs engineering semantic analysis on each piece of report evidence and applicable regulatory clauses based on pre-defined engineering semantic fields. The analysis includes the subject, evaluation attributes, status or value, conditions for validity, spatial scope, temporal scope, affirmative relations, negative relations, and conditional relations.

[0058] For example, if no stable groundwater was observed within the drilling depth during the exploration period, the large language model extracts groundwater as the subject, stable water level observation results as the evaluation attribute, "not observed" as a negative relation, the exploration period as the time range, and "within the drilling depth" as the spatial range. If there is no groundwater influence at the site, the large language model extracts the groundwater influence at the site as the subject, and "no influence" as the conclusion relation, but this conclusion may not have a clear time range. By comparing the two standardized semantic descriptions, it can be found that there may be a range expansion between observational evidence and engineering conclusions.

[0059] The large language model generates standardized semantic descriptions based on the extracted results. The system then vectorizes these standardized semantic descriptions to obtain basic semantic vectors. These basic semantic vectors are used to represent the semantic similarity between different sentence structures and specialized expressions.

[0060] To avoid the basic semantic vector ignoring project objects, numerical ranges, and applicable conditions, the system further integrates project object categories, object attribute categories, project condition matching results, numerical values ​​and units of measurement, spatial range, time range, semantic polarity, data source type, and chapter function information.

[0061] The engineering object category is used to distinguish between strata, groundwater, test samples, engineering parameters, foundation bearing strata, and proposed buildings and structures. The object attribute category is used to distinguish between soil and rock conditions, density, weathering degree, bearing capacity, compressibility, and engineering suitability. The project condition matching result is used to indicate the degree of correspondence between current evidence or specification clauses and the project location, engineering type, investigation stage, and site conditions.

[0062] Numerical and unit information is used to retain numerical values, data types, and unit categories. Spatial extent information includes site area, exploration point, stratigraphic extent, depth extent, and elevation extent. Temporal extent information includes drilling date, observation date, exploration period, and long-term evaluation conditions. Semantic polarity is used to distinguish between affirmative, negative, restrictive, exceptional, and conditional relationships. Data source type is used to distinguish between basic data, parametric statistics, engineering evaluation, and engineering recommendations. Chapter function information indicates whether the report content is located in the basic data chapter, parametric statistics chapter, engineering evaluation chapter, or design recommendation chapter.

[0063] The system fuses the above information with the basic semantic vector to obtain the report-enhanced semantic vector and the specification clause-enhanced semantic vector. The fused vectors not only reflect the semantic similarity of the text, but also whether the two contents refer to the same project object, the same attribute, and the same scope of application.

[0064] For example, the statement that the overall formation is in a plastic state, the local formation is in a stiff plastic state, and the recommendation of parameters for the entire formation based on the stiff plastic state are highly correlated at the textual level. After enhancement, the vector also carries information such as the overall range, the local range, the parameter statistics section, and the engineering recommendation section, enabling subsequent steps to determine whether the local state has been improperly extended to the basis for overall value selection.

[0065] After the above processing, we obtain the report-enhanced semantic vector set and the specification clause-enhanced semantic vector set.

[0066] S4 groups the report evidence set according to the project object, object attributes, spatiotemporal scope and reference relationship, and generates a set of conflict candidate relationships based on the semantic proximity between the report enhanced semantic vectors and between the report enhanced semantic vectors and the enhanced semantic vectors of the normative clauses.

[0067] In this step, the system groups the report evidence set according to the project object, object attributes, stratigraphic identifier, exploration point identifier, spatial range, and reference relationship.

[0068] Evidence from reports belonging to the same project or having data reference relationships is grouped into the same related evidence group. For example, borehole records, laboratory test results, parameter statistics, site evaluation conclusions, and foundation recommendations for the same stratum are grouped into the same stratum-related evidence group. Groundwater observation records, groundwater descriptions in the report text, and construction dewatering recommendations for the same exploration point are grouped into the same groundwater-related evidence group.

[0069] When an engineering proposal cites parameter statistics, and those statistics originate from laboratory test results, the system connects the three types of evidence into a continuous data chain based on the citation relationship. Even if the engineering proposal does not directly mention the specimen number, as long as the corresponding parameter statistics can be traced back through the stratigraphic number and parameter name, it is also classified into the same related evidence group.

[0070] Within each group of related evidence, the system retrieves pairs of report evidence that may describe the same object attribute based on the semantic proximity between the report-enhanced semantic vectors. Semantic proximity is used only to filter evidence combinations that require further review and is not directly used as a conclusion of conflict.

[0071] The system-generated report contains candidate relationships for internal conflicts, including those related to numerical inconsistencies, status description inconsistencies, inconsistencies between basic data and parameter statistical results, inconsistencies between parameter statistical results and engineering evaluation conclusions, and inconsistencies between engineering evaluation conclusions and engineering recommendations.

[0072] Numerical inconsistency candidate relationships are used to connect the numerical representations of the same engineering object and the same parameter in different chapters. For example, when the same soil bearing capacity parameter shows different values ​​in the parameter statistics table and the foundation recommendations, a numerical inconsistency candidate relationship is formed.

[0073] Inconsistent state description candidate relationships are used to connect descriptions of the state, density, weathering degree, and water-bearing state of the same stratum or rock mass at different locations. For example, if the stratigraphic description section describes a sand layer as slightly dense to medium dense, while the engineering evaluation section directly evaluates it as dense, an inconsistent state description candidate relationship is formed.

[0074] Candidate relationships for inconsistencies between basic data and parameter statistical results are used to determine whether the statistical results can be supported by the corresponding basic data. Candidate relationships for inconsistencies between parameter statistical results and engineering evaluation conclusions are used to determine whether the engineering evaluation exceeds the range that the parameter statistical results can support. Candidate relationships for inconsistencies between engineering evaluation conclusions and engineering recommendations are used to determine whether the foundation recommendations, excavation recommendations, or construction recommendations are consistent with the previous evaluation conclusions.

[0075] The system also generates candidate relationships for inconsistencies between project conditions and applicable conditions of normative clauses, as well as candidate relationships for inconsistencies between report conclusions and preconditions of normative clauses, based on the semantic similarity between the enhanced semantic vector of the report and the enhanced semantic vector of the normative clauses.

[0076] The candidate relationship of inconsistency between project conditions and applicable standard clauses is used to identify situations where the standard requirements cited or adopted in the report do not match the current project conditions. The candidate relationship of inconsistency between report conclusions and prerequisites of standard clauses is used to identify situations where the report has reached a certain evaluation conclusion, but lacks the basic data, experimental data, or condition judgment process required by the standard.

[0077] When generating conflict candidate relationships, the system employs filtering mechanisms based on object, space, time, and applicable conditions. For report evidence with different engineering objects, no direct conflict candidate relationships are generated. For content with no spatial overlap, even if the wording is similar, no conflict candidate relationships are generated. For content with inconsistent observation times where the time difference can reasonably explain the state change, no direct conflict candidate relationships are generated. For evaluation conclusions with different applicable conditions, only when the report does not explicitly distinguish the corresponding conditions are they retained as candidate relationships awaiting review.

[0078] For example, the same strata at different exploration sites may have different states. If two exploration sites are located in different geomorphic units and their spatial ranges do not overlap, then the difference in state will not be considered a conflict. As another example, when there are discrepancies between groundwater observation results during the dry season and survey data during the wet season, the system retains the observation time condition and does not directly classify the water level difference as a conflict.

[0079] By performing semantic retrieval first and then filtering based on engineering conditions, the system can reduce invalid candidate relationships formed solely by textual similarity and concentrate evidence requiring further verification into the conflict candidate relationship set.

[0080] S5 reviews the set of candidate conflict relationships for consistency in engineering objects, numerical values ​​and units of measurement, spatiotemporal scope, applicable conditions, and logical relationships. It determines whether there are internal conflicts in the report or conflicts between the report and the standard clauses, and outputs the conflict type, related evidence, standard basis, source location, and review priority.

[0081] After this deficiency, the system sequentially checks the consistency of engineering objects, numerical values ​​and units of measurement, spatiotemporal range, applicable conditions, and logical relationships of the conflict candidate relationship set.

[0082] For numerical conflict candidate relationships, the system first standardizes the units of measurement and then distinguishes between original measured values, statistical values, recommended values, and design values ​​based on the data source type. Original measured values ​​are data directly generated from drilling, in-situ testing, laboratory experiments, or groundwater observation; statistical values ​​are data generated from multiple original measured values ​​through statistical processing; recommended values ​​are engineering parameters based on statistical results and engineering experience; and design values ​​are the parameters ultimately adopted in the report recommendations or design conditions.

[0083] The system determines whether the numerical discrepancies stem from unit conversion, statistical processing, value range, significant figures, or changes in data version. If the two values ​​belong to different data levels and there is a clear statistical or reduction relationship in the report, they are not considered a numerical conflict. If the two values ​​refer to the same project object, the same parameter, and the same applicable conditions, and are still incompatible after excluding unit conversion and statistical processing factors, then they are determined to be a numerical conflict.

[0084] For state-related conflict candidates, the system determines whether different state descriptions can be simultaneously valid based on the engineering object, stratigraphic scope, depth range, and applicable conditions. If one description applies to the entire stratigraphic system while the other applies only to a local area, the system determines whether the report retains local limitations. A state description conflict is only identified when two state descriptions apply to the same spatial scope and conditions, and cannot be simultaneously valid.

[0085] For example, the site engineering geological conditions section describes a stratum as being in a plastic state overall and a stiff plastic state locally, while the parameter statistics section directly determines the parameters of the entire stratum as being in a stiff plastic state. If the report does not provide evidence that the stiff plastic sample can represent the entire stratum, the system will determine this relationship as an inconsistency between the state range and the basis for parameter values, and include it in the state description conflict or data conclusion conflict.

[0086] For conflicting candidate relationships in the conclusion category, the system establishes an evidence chain from basic engineering data, parameter statistical results, engineering evaluation conclusions to engineering recommendations. Basic engineering data constitutes the starting node of the evidence chain, parameter statistical results constitute the data processing node, engineering evaluation conclusions constitute the analysis node, and engineering recommendations constitute the output node.

[0087] The system verifies the consistency of engineering objects, parameters, and applicable conditions cited in subsequent evidence with those in preceding evidence, step by step, along the chain of evidence. When the geological strata, parameters, or site conditions recommended for the engineering project cannot be traced back to the preceding evidence, or when subsequent conclusions expand the spatial, temporal, or applicable scope of the preceding evidence, a data-conclusion conflict is identified.

[0088] For example, groundwater observation records may only prove that a stable water level was not observed at certain borehole depths during the exploration period, while engineering recommendations may directly conclude that the site is unaffected by groundwater during construction and use. In this case, the subsequent conclusion exceeds the temporal and spatial scope of the preceding observational evidence. The system identifies this as a data conclusion conflict and prompts a review of groundwater seasonal variation data and hydrological conditions during the construction period.

[0089] For candidate conflicts related to regulations, the system reviews the applicable conditions and prerequisites of the regulatory clauses based on the project's condition profile. If the regulatory clauses used in the report do not match the project type, investigation stage, site conditions, or special geotechnical conditions, a regulatory conflict is identified.

[0090] When the project conditions meet the scope of application of the standard clauses, the system further determines whether the report has the basic data, analysis process, and evaluation conclusions required by the clauses. If the standard clauses require a frozen soil evaluation under specific site conditions, and the report has reached a conclusion on frost heave, but the basic engineering data lacks site surveys, soil condition data, or meteorological data corresponding to this conclusion, then the report's conclusions are determined to be inconsistent with the preconditions of the standard.

[0091] If the report's conclusions violate the necessary, prohibited, or conditional relationships in the applicable regulations for the project, it is also considered a conflict of regulations. A lack of necessary relationships manifests as the report failing to produce the required data or evaluation conclusions; a conflict of prohibited relationships manifests as the report using restricted processing methods under specific conditions; a conflict of conditional relationships manifests as the report drawing subsequent conclusions before the preconditions are met.

[0092] The large language model is used in the review process to identify complex sentence structures, elliptical expressions, referential relationships, negation relationships, and conditional relationships. For example, expressions such as "not seen," "inappropriate," "except in specific cases," "under certain conditions," and "within a local scope" are converted into standardized semantic relationships by the large language model. The final conflict type is still determined by the engineering object, numerical value, unit of measurement, spatial scope, temporal scope, applicable conditions, and normative logical rules.

[0093] After determining the conflict outcome, the system determines the review priority based on the position of the conflict outcome in the chain of evidence and the scope of its reference in subsequent reports.

[0094] If a conflict occurs in the basic engineering data or parameter statistics stage and is cited multiple times in site evaluation, foundation design, and construction recommendations, this conflict may affect multiple subsequent conclusions, and the system will set it to a higher review priority. If the conflict only involves differences in wording that do not affect the technical conclusions, it will be set to a lower review priority.

[0095] Conflicts that may affect stratigraphic division, geotechnical parameter values, foundation selection, groundwater assessment, seismic assessment, permafrost assessment, or construction recommendations are all subject to priority review. In cases of conflicting basic data, the system prioritizes prompting reviewers to verify the exploration point number, sample number, test depth, observation time, and data version. For conflicts between engineering recommendations and basic data, the system traces back along the evidence chain to pinpoint the first instance of inconsistency.

[0096] The system's final output report may indicate internal conflicts or conflicts with standard clauses. Each conflict detection result includes the conflict type, related evidence, standard basis, source location, and review priority.

[0097] Related evidence includes evidence from the first and second reports, or evidence from reports and applicable standard clauses for the project. Source location includes report chapters, page numbers, paragraph positions, table names, table row and column positions, or drawing locations. Standard basis includes the standard document name, clause location, applicable conditions, and prerequisites. The test results also show the engineering objects involved in the conflict, object attributes, spatial scope, temporal scope, judgment criteria, potential impact on report conclusions, and recommended engineering baseline data for review.

[0098] The conflict detection results are output to the report review interface. After the reviewer selects a conflict result, the review interface simultaneously locates the relevant paragraphs, table units, or figure descriptions in the geotechnical engineering investigation report to be tested, and displays the associated drilling records, test results, groundwater observation records, or code clauses. This allows the reviewer to directly trace back to the original evidence from the conflict result without having to search repeatedly throughout the entire report and multiple appendices.

[0099] Reviewers can mark test results as confirmed conflicts, reasonable discrepancies, or requiring supplementary information. For content confirmed by manual review to be synonyms, the system updates the mapping relationship between the project object name and its synonym. For reasonable discrepancies arising from differences in spatial scope, observation time, or survey stage, the system records the corresponding conditions and forms conditional exclusion rules. For content confirmed to be conflicting, the system supplements the corresponding object relationships, attribute relationships, and conditional relationships into the review rules for similar reports.

[0100] The results of manual review are only used to adjust terminology mapping relationships, conditional relationships, and review rules; they do not replace basic engineering data, nor do they directly modify the content of the original report. Through the above processing, this embodiment can detect cross-chapter data conflicts, state description conflicts, data conclusion conflicts, and standard application conflicts in geotechnical engineering investigation reports in high-altitude areas of Qinghai, and generate traceable results with clear source location, engineering basis, and review priority.

[0101] In summary, this embodiment transforms the technical content scattered in the main text, tables, and appendices into report evidence with engineering objects, applicable conditions, and source locations by performing layered analysis and source association on the survey report and basic engineering data. Then, it combines project conditions to screen applicable standard clauses and uses large model semantic vectors to establish cross-chapter semantic relationships to form conflict candidate relationships. Finally, through review of objects, values, spatiotemporal range, and logical relationships, it can accurately identify internal conflicts in the report and conflicts in the application of standards, and realize the location, traceability, and hierarchical review of conflict results.

[0102] The following describes a geotechnical engineering investigation report compliance conflict detection device combining large model semantic vector enhancement, provided by an embodiment of this application. The geotechnical engineering investigation report compliance conflict detection device and the geotechnical engineering investigation report compliance conflict detection method combining large model semantic vector enhancement described below can be referred to in correspondence with each other.

[0103] Please refer to Figure 2 , Figure 2 This is a schematic diagram of a geotechnical engineering investigation report compliance conflict detection device that combines large model semantic vector enhancement, as provided in an embodiment of this application.

[0104] In this embodiment, the device may include: The data parsing module 100 is used to obtain the geotechnical engineering investigation report to be tested and the corresponding basic engineering data, perform layered parsing of the geotechnical engineering investigation report to be tested, and establish the source association relationship between the basic engineering data and the content of the parsed report to obtain a set of original content units with source location information. The report standardization processing module 200 is used to identify geotechnical engineering objects and standardize engineering semantics on the original set of content units, extract engineering objects, object attributes, attribute values ​​or evaluation conclusions, applicable conditions and source information, form a set of report evidence, and form a project condition profile based on basic engineering data; The report vectorization module 300 is used to divide the normative documents on which the report is prepared and reviewed into normative clause units. Based on the matching relationship between the applicable conditions of the normative clause units and the project condition profile, a set of applicable normative clauses for the project is obtained. The large language model is used to perform engineering semantic parsing and vectorization processing on the report evidence set and the set of applicable normative clauses for the project, and integrates engineering object information, object attribute information, project condition information, numerical and measurement unit information, spatiotemporal range information, and semantic polarity information to obtain the report enhanced semantic vector set and the normative clause enhanced semantic vector set. The conflict candidate module 400 is used to group the report evidence set according to the project object, object attribute, spatiotemporal range and reference relationship, and generate a conflict candidate relationship set according to the semantic proximity between the report enhanced semantic vectors and between the report enhanced semantic vectors and the enhanced semantic vectors of the specification clauses. The conflict result output module 500 is used to review the consistency of engineering objects, numerical values ​​and units of measurement, spatiotemporal range, applicable conditions and logical relationships of the conflict candidate relationship set, determine the internal conflict of the report or the conflict between the report and the standard clauses, and output the conflict type, related evidence, standard basis, source location and review priority.

[0105] This application also provides a testing device, please refer to... Figure 3 , Figure 3 This is a schematic diagram of the structure of a testing device provided in an embodiment of this application. The testing device may include: Memory, used to store computer programs; A processor, used to execute computer programs, can implement steps such as any of the above-described methods for detecting compliance conflicts in geotechnical engineering investigation reports by combining large model semantic vector enhancement.

[0106] like Figure 3 The diagram shows the structural composition of a detection device, which may include a processor 10, a memory 11, a communication interface 12, and a communication bus 13. The processor 10, memory 11, and communication interface 12 communicate with each other through the communication bus 13.

[0107] In this embodiment, the processor 10 may be a central processing unit (CPU), an application-specific integrated circuit, a digital signal processor, a field-programmable gate array, or other programmable logic devices.

[0108] The processor 10 can call the program stored in the memory 11. Specifically, the processor 10 can execute the operations in the embodiment of the abnormal IP identification method.

[0109] The memory 11 is used to store one or more programs. The programs may include program code, which includes computer operation instructions. In this embodiment, the memory 11 stores at least a program for implementing the following functions: S1. Obtain the geotechnical engineering investigation report to be tested and the corresponding basic engineering data. Perform layered analysis on the geotechnical engineering investigation report to be tested and establish the source association relationship between the basic engineering data and the content of the analyzed report to obtain a set of original content units with source location information. S2, performs geotechnical engineering object identification and engineering semantic standardization processing on the original content unit set, extracts engineering objects, object attributes, attribute values ​​or evaluation conclusions, applicable conditions and source information, forms a report evidence set, and forms a project condition profile based on basic engineering data; S3 divides the normative documents on which the report is based into normative clause units. Based on the matching relationship between the applicable conditions of the normative clause units and the project condition profile, a set of normative clauses applicable to the project is obtained. Using a large language model, engineering semantic parsing and vectorization processing are performed on the report evidence set and the set of normative clauses applicable to the project, respectively. The information of engineering objects, object attributes, project conditions, numerical and measurement units, spatiotemporal range, and semantic polarity are integrated to obtain the report enhanced semantic vector set and the normative clause enhanced semantic vector set. S4. Group the report evidence set according to the project object, object attributes, spatiotemporal scope and reference relationship, and generate a set of conflict candidate relationships based on the semantic proximity between the report enhanced semantic vectors and between the report enhanced semantic vectors and the enhanced semantic vectors of the normative clauses. S5 reviews the set of candidate conflict relationships for consistency in engineering objects, numerical values ​​and units of measurement, spatiotemporal scope, applicable conditions, and logical relationships. It determines whether there are internal conflicts in the report or conflicts between the report and the standard clauses, and outputs the conflict type, related evidence, standard basis, source location, and review priority.

[0110] In one possible implementation, the memory 11 may include a program storage area and a data storage area, wherein the program storage area may store the operating system and applications required for at least one function; and the data storage area may store data created during use.

[0111] In addition, memory 11 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device or other volatile solid-state storage device.

[0112] Communication interface 12 can be an interface for the communication module, used to connect with other devices or systems.

[0113] Of course, it should be noted that, Figure 3 The structure shown does not constitute a limitation on the detection device in the embodiments of this application. In practical applications, a detection device may include more than Figure 3 More or fewer components as shown, or combinations of certain components.

[0114] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, can implement the steps of any of the above-described methods for detecting compliance conflicts in geotechnical engineering investigation reports by combining large model semantic vector enhancement.

[0115] The computer-readable storage medium may include various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0116] For a description of the computer-readable storage medium provided in this application, please refer to the above method embodiments; further details will not be repeated here.

[0117] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0118] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0119] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0120] The foregoing has provided a detailed description of the method, device, testing equipment, and computer-readable storage medium for detecting compliance conflicts in geotechnical engineering investigation reports, which combines large-model semantic vector enhancement. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the embodiments above are merely for the purpose of helping to understand the method and its core ideas. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from its principles, and these improvements and modifications also fall within the protection scope of the claims of this application.

Claims

1. A method for detecting compliance conflicts in geotechnical engineering investigation reports by combining large-model semantic vector enhancement, characterized in that, include: S1, obtain the geotechnical engineering investigation report to be tested and the corresponding basic engineering data, perform layered analysis on the geotechnical engineering investigation report to be tested, and establish the source association relationship between the basic engineering data and the content of the analyzed report to obtain a set of original content units with source location information; S2, perform geotechnical engineering object identification and engineering semantic standardization processing on the original content unit set, extract engineering objects, object attributes, attribute values ​​or evaluation conclusions, applicable conditions and source information to form a report evidence set, and form a project condition profile based on the engineering basic data; S3. The normative documents on which the report is prepared and reviewed are divided into normative clause units. Based on the matching relationship between the applicable conditions of the normative clause units and the project condition profile, a set of normative clauses applicable to the project is obtained. The report evidence set and the set of normative clauses applicable to the project are subjected to engineering semantic parsing and vectorization processing using a large language model. The engineering object information, object attribute information, project condition information, numerical and measurement unit information, spatiotemporal range information, and semantic polarity information are integrated to obtain a set of enhanced semantic vectors for the report and a set of enhanced semantic vectors for the normative clauses. S4. The report evidence set is grouped according to the project object, the object attribute, the spatiotemporal range and the reference relationship, and a conflict candidate relationship set is generated according to the semantic proximity between the report enhanced semantic vectors and between the report enhanced semantic vectors and the normative clause enhanced semantic vectors. S5, perform consistency checks on the set of conflict candidate relationships in terms of engineering object consistency, numerical and measurement unit consistency, spatiotemporal range consistency, applicable conditions consistency, and logical relationship consistency, determine internal conflicts in the report or conflicts between the report and standard clauses, and output the conflict type, related evidence, standard basis, source location, and review priority.

2. The method for detecting compliance conflicts in geotechnical engineering investigation reports according to claim 1, characterized in that, In step S1, the geotechnical engineering investigation report to be tested is analyzed in layers, including: According to the document hierarchy and content carrier, the geotechnical engineering investigation report to be tested is divided into title unit, paragraph unit, item unit, table unit, figure description unit and attachment reference unit; Record the chapter name, page number, paragraph position, table position, or figure position corresponding to each original content unit; Extract exploration point identifiers, stratigraphic identifiers, sample identifiers, test depths, and observation times from the engineering basic data, and establish source association relationships between the engineering basic data and the corresponding original content units based on the exploration point identifiers, stratigraphic identifiers, sample identifiers, test depths, or observation times.

3. The method for detecting compliance conflicts in geotechnical engineering investigation reports according to claim 2, characterized in that, In step S2, geotechnical engineering object identification and engineering semantic standardization processing are performed on the original content unit set, including: Extract the engineering object, object attributes, attribute values ​​or evaluation conclusions, spatial range, time range, applicable conditions, and data source type from each of the original content units; Establish a mapping relationship between the object name, object abbreviation, object number, and text description corresponding to the same project object; Standardize the names of soil and rock conditions, units of measurement, depth benchmarks, elevation benchmarks, and synonyms; Transform engineering descriptions with spatial, temporal, or observational limitations into report evidence that retains the corresponding limitations.

4. The method for detecting compliance conflicts in geotechnical engineering investigation reports according to claim 3, characterized in that, The project condition profile includes the project location, exploration stage, engineering type, site elevation, geomorphic unit, stratigraphic composition, groundwater conditions, earthquake-related conditions, seasonal permafrost conditions, and special soil and rock conditions. Accordingly, in S3, the matching relationship between the applicable conditions of the standard clause unit and the project condition profile includes: matching the applicable engineering type, applicable survey stage, applicable site conditions and prerequisites corresponding to each standard clause unit with the project condition profile, determining the standard clause unit that meets the matching conditions as the applicable standard clause for the project, and excluding the standard clause unit that does not meet the matching conditions from the conflict detection range.

5. The method for detecting compliance conflicts in geotechnical engineering investigation reports according to claim 4, characterized in that, In step S3, the large language model is used for engineering semantic parsing and vectorization processing, including: Using the large language model, the subject, evaluation attributes, status or value, conditions for establishment, spatial range, temporal range, affirmative relationship, negative relationship and conditional relationship are extracted from each of the report evidence and each of the applicable normative clauses of the project to generate a standardized semantic description; The standardized semantic description is vectorized to obtain the basic semantic vector; The report enhanced semantic vector and the specification clause enhanced semantic vector are obtained by fusing the engineering object category, object attribute category, project condition matching result, numerical value and unit of measurement, spatial range, time range, semantic polarity, data source type and chapter function information with the basic semantic vector.

6. The method for detecting compliance conflicts in geotechnical engineering investigation reports according to claim 5, characterized in that, In step S4, based on the engineering object, the object attribute, the stratigraphic identifier, the exploration point identifier, the spatial range, and the reference relationship, the report evidence belonging to the same engineering object or having a data reference relationship is divided into the same related evidence group; Within each of the aforementioned related evidence groups, based on the semantic proximity between the report-enhanced semantic vectors, candidate relationships for numerical inconsistency, candidate relationships for state description inconsistency, candidate relationships for inconsistency between basic data and parameter statistical results, candidate relationships for inconsistency between parameter statistical results and engineering evaluation conclusions, and candidate relationships for inconsistency between engineering evaluation conclusions and engineering recommendations are generated. Based on the semantic similarity between the enhanced semantic vector of the report and the enhanced semantic vector of the normative clause, candidate relationships of inconsistency between project conditions and applicable conditions of normative clauses, and candidate relationships of inconsistency between report conclusions and preconditions of normative clauses are generated. For report evidence that differs in the engineering object, has no overlap in the spatial range, does not correspond in the time range, or has inconsistent applicable conditions, no conflict candidate relationship is generated.

7. The method for detecting compliance conflicts in geotechnical engineering investigation reports according to claim 6, characterized in that, In step S5, the set of conflict candidate relationships is reviewed, including: For numerical conflict candidate relationships, the unit of measurement is unified, and the original measured value, statistical value, suggested value and design adopted value are distinguished according to the data source type. After excluding unit conversion, statistical processing, value range and data version differences, numerical conflicts are determined. For state-related conflict candidate relationships, based on the corresponding engineering object, stratum range, depth range and applicable conditions, it is determined whether different state descriptions can be established simultaneously, and when the different state descriptions cannot be established simultaneously under the same conditions, a state description conflict is determined. For candidate relationships of conflicting conclusions, an evidence chain is established from the basic engineering data, the statistical results of the parameters, the engineering evaluation conclusions to the engineering recommendations. When the engineering objects, parameters or applicable conditions cited in the subsequent evidence are inconsistent with the preceding evidence, a data-conclusion conflict is determined. For candidate relationships of normative conflicts, the applicable conditions and prerequisites of the corresponding normative clause units are reviewed according to the project condition profile. When the normative clauses used in the geotechnical engineering investigation report to be tested do not match the project condition profile, or when the report conclusions are inconsistent with the necessary, prohibited, or conditional relationships in the applicable normative clauses of the project, a normative application conflict is determined. The review priority is determined based on the position of each conflict outcome in the chain of evidence and the extent to which it is referenced in subsequent reports.

8. A device for detecting compliance conflicts in geotechnical engineering investigation reports by combining large-model semantic vector enhancement, characterized in that, include: The data parsing module is used to obtain the geotechnical engineering investigation report to be tested and the corresponding basic engineering data, perform layered parsing on the geotechnical engineering investigation report to be tested, and establish the source association relationship between the basic engineering data and the content of the parsed report to obtain a set of original content units with source location information; The report standardization processing module is used to perform geotechnical engineering object identification and engineering semantic standardization processing on the original content unit set, extract engineering objects, object attributes, attribute values ​​or evaluation conclusions, applicable conditions and source information, form a report evidence set, and form a project condition profile based on the engineering basic data; The report vectorization module is used to divide the normative documents on which the report is prepared and reviewed into normative clause units. Based on the matching relationship between the applicable conditions of the normative clause units and the project condition profile, a set of applicable normative clauses for the project is obtained. The report evidence set and the set of applicable normative clauses for the project are subjected to engineering semantic parsing and vectorization processing using a large language model. The module also integrates engineering object information, object attribute information, project condition information, numerical and measurement unit information, spatiotemporal range information, and semantic polarity information to obtain a report enhanced semantic vector set and a normative clause enhanced semantic vector set. The conflict candidate module is used to group the report evidence set according to the project object, the object attribute, the spatiotemporal range and the reference relationship, and generate a conflict candidate relationship set according to the semantic similarity between the report enhanced semantic vectors and between the report enhanced semantic vectors and the specification clause enhanced semantic vectors; The conflict result output module is used to review the consistency of the candidate conflict relationships in terms of engineering objects, numerical values ​​and units of measurement, spatiotemporal range, applicable conditions, and logical relationships, determine the internal conflict of the report or the conflict between the report and the standard clauses, and output the conflict type, related evidence, standard basis, source location, and review priority.

9. A testing device, characterized in that, include: Memory, used to store computer programs; A processor, configured to implement the steps of the geotechnical engineering investigation report compliance conflict detection method as described in any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the geotechnical engineering investigation report compliance conflict detection method as described in any one of claims 1 to 7.

Images