Construction process data retrieval method and system

Abstract

The invention provides a construction process data retrieval method and system, and relates to the field of railway tunnel engineering management.The method comprises the steps that multi-source heterogeneous data related to the construction process is recognized based on a large model, and recognition content is output; performing entity extraction on the construction data, and outputting a space-time dynamic map of the construction process data according to a local relationship and a global relationship between entities; generating fusion data of the text information and the LaTeX formula sequence according to the technical specification knowledge; generating a business scene logic graph according to the business scene document; constructing a multimode information database according to the spatio-temporal dynamic map, the fusion data of the text information and the LaTeX formula sequence and the business scene logic map; and judging a service scene based on the trained expert model, selecting different mixed retrieval modes according to the service scene to carry out data search in the multimode information database, and outputting a retrieval result. According to the method, construction process data retrieval which is high in query speed, comprehensive in query result and high in credibility is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of railway tunnel engineering management, and in particular to a construction process data retrieval method and system. Background Art

[0002] Railway construction process data covers the entire process of railway construction and construction, involving multiple disciplines, dozens of file formats, and hundreds of file contents. It includes documents from multiple sources and complex data types, such as geological forecasts, construction logs, construction progress ledgers, maintenance records, and experimental test reports. In the process of realizing the present invention, the applicant found that the existing information retrieval technology is difficult to meet the efficient retrieval and utilization of railway construction process data. RAG (Retrieval-Augmented Generation) technology is difficult to accurately capture and represent the complex relationships between entities, which easily leads to relationship confusion; the knowledge graph-based method can obtain the relationship between data, provide structured knowledge representation, and is suitable for the management and query of complex relationships. However, when processing large-scale railway construction process data, the knowledge graph-based method will result in a huge graph scale and the query speed cannot meet the demand. Therefore, there is an urgent need for a railway construction process data retrieval method with fast query speed, comprehensive query results, and high credibility. Summary of the invention

[0003] The present invention aims to solve at least one of the technical problems existing in the prior art or related technology, discloses a construction process data retrieval method and system, and realizes construction process data retrieval with fast query speed, comprehensive query results and high credibility.

[0004] The first aspect of the present invention discloses a construction process data retrieval method, including: construction data parsing: based on a large model, calling multiple document parsing agents to identify multi-source heterogeneous data related to the construction process, and outputting the identified content; construction data modeling: based on a regular expression or rule extraction method, processing structured data and unstructured data in the identified content, and outputting construction process data entities; processing construction process data entities based on a graph convolutional network and a graph attention network, and outputting local and global relationships between entities; processing local and global relationships between entities based on a time-series graph neural network to output a spatiotemporal dynamic graph of construction process data; technical specification knowledge modeling: based on a large model, processing technical specification documents in the identified content, outputting a LaTeX formula sequence according to the mathematical formula information in the technical specification documents, outputting a text information summary according to the text information in the technical specification documents, and outputting fusion data of the text information and the LaTeX formula sequence according to the similarity relationship between the vector representation of the text information and the vector representation of the formula sequence; business scenario logic modeling: based on a large model of the vertical domain of construction, processing the identified content, and outputting the fusion data of the text information and the LaTeX formula sequence. The business scenario document description in the identified content is used to output the entity relationship of the business scenario data. The business scenario flowchart of the construction process is processed based on the flowchart extraction algorithm, and the business scenario processing flow sequence is output. The business scenario data entity relationship and the business scenario processing flow sequence are combined to output the business scenario logic map. A multi-modal information database is constructed: According to the spatio-temporal dynamic map, the fusion data of text information and LaTeX formula sequence, and the business scenario logic map, entity disambiguation is performed between different types of data structures, a unified indexing system is established, and different types of data structures are stored in corresponding types of databases respectively. An expert model is trained: the business scenario data is used as the input of the expert model, and the scenario question answer text is used as the output of the expert model. The expert model is trained until convergence to obtain a trained expert model. Hybrid retrieval: The natural language question text input by the user is received, and the natural language question text is processed based on the trained expert model to identify the business scenario. Different hybrid retrieval methods are selected according to the business scenario to search for data in the multi-modal information database, and the retrieval results are output.

[0005] According to the construction process data retrieval method disclosed in the present invention, preferably, it also includes: training the construction vertical domain big model: using the multi-source heterogeneous data of railway construction as the input of the construction vertical domain big model, using text generation, category labels, structured information extraction and prediction results as the output of the construction vertical domain big model, and training the construction vertical domain big model until convergence.

[0006] According to the construction process data retrieval method disclosed in the present invention, preferably, the steps of processing construction process data entities based on a graph convolutional network and a graph attention network, and outputting local and global relationships between entities, specifically include: during the training process, firstly using a graph convolutional network to capture the local connection features between entities: the graph convolutional network integrates the information of adjacent nodes into the current node through a neighborhood aggregation operation, so that the features of each node not only contain its own information, but also incorporate the contextual information of the neighborhood structure; then the graph attention network is introduced to dynamically calculate the weights of the edges: the graph attention network assigns adaptive weights between neighborhood nodes through an attention mechanism, thereby enhancing the ability to express key relationships.

[0007] According to the construction process data retrieval method disclosed in the present invention, preferably, the business scenarios specifically include: construction progress analysis, concrete usage analysis, second lining construction timing judgment and second lining concrete quality data tracking.

[0008] According to the construction process data retrieval method disclosed in the present invention, preferably, the index system contains all entity information of construction data, technical specifications and business scenarios.

[0009] According to the construction process data retrieval method disclosed in the present invention, preferably, the multi-source heterogeneous data comes from actual construction documents, technical specification documents and business scenario documents.

[0010] According to the construction process data retrieval method disclosed in the present invention, preferably, the database includes: a relational database, a graph database and a vector database.

[0011] The second aspect of the present invention discloses a construction process data retrieval system, comprising: a memory for storing program instructions; a processor for calling the program instructions stored in the memory to implement a construction process data retrieval method such as any of the above technical solutions.

[0012] The beneficial effects of the present invention include at least: the present invention constructs a multi-modal information structure of railway construction process data by modeling construction data, technical specification knowledge, and business scenario logic, and uses multi-modal information training to obtain an expert model in the railway engineering field, guiding the intelligent agent cluster to perform hybrid retrieval on input questions, thereby realizing efficient and intelligent retrieval and utilization of railway construction process data. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 A schematic diagram of the overall architecture of a construction process data retrieval method according to an embodiment of the present invention is shown.

[0014] Figure 2 A schematic block diagram of a construction process data retrieval system according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0015] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention can also be implemented in other ways different from those described herein, and therefore, the present invention is not limited to the limitations of the specific embodiments disclosed below.

[0016] According to an embodiment of the present invention, a construction process data retrieval method is disclosed, comprising: Construction data analysis: Based on the large model, multiple document analysis agents are called to identify multi-source heterogeneous data related to the construction process and output the identified content; Construction data modeling: Based on regular expressions or rule extraction methods, structured data and unstructured data in the recognition content are processed and output as construction process data entities; based on graph convolutional networks and graph attention networks, construction process data entities are processed and local and global relationships between entities are output; based on time-series graph neural networks, local and global relationships between entities are processed and spatiotemporal dynamic graphs of construction process data are output; Technical specification knowledge modeling: Based on the large model, the technical specification documents in the content are processed and identified, and LaTeX formula sequences are output according to the mathematical formula information in the technical specification documents. The text information summary is output according to the text information in the technical specification documents. Based on the similarity relationship between the vector representation of the text information and the vector representation of the formula sequence, the fusion data of the text information and the LaTeX formula sequence is output; Business scenario logic modeling: Based on the construction vertical domain big model, the business scenario document description in the identification content is processed, and the business scenario data entity relationship is output. The business scenario flowchart of the construction process is processed based on the flowchart extraction algorithm, and the business scenario processing process sequence is output. The business scenario data entity relationship and the business scenario processing process sequence are combined to output the business scenario logic map; Construct a multimodal information database: Based on the spatiotemporal dynamic graph, the fusion data of text information and LaTeX formula sequences, and the business scenario logic graph, perform entity disambiguation between different types of data structures, establish a unified indexing system, and store different types of data structures in corresponding types of databases; Training the expert model: Use the business scenario data as the input of the expert model, and the scenario question answer text as the output of the expert model. Train the expert model until convergence to obtain a trained expert model. Hybrid retrieval: Receive natural language question text input by users, process the natural language question text based on the trained expert model to identify the business scenario, and select different hybrid retrieval methods according to the business scenario to search for data in the multi-modal information database and output the retrieval results.

[0017] like Figure 1 As shown, according to the overall architecture diagram of the construction process data retrieval method disclosed in this embodiment, it can be seen that multi-mode information represents different data structures of railway construction process documents stored in the system, including graph-type data and structured data composed of actual construction documents, unstructured data composed of technical specification documents, and relational data composed of business scenario documents. By modeling the construction data, technical specification knowledge, and business scenario logic, a multi-mode information structure of railway construction process data is constructed, and an expert model in the railway engineering field is obtained by multi-mode information training to guide the intelligent agent cluster to perform mixed retrieval on the input questions, thereby realizing efficient and intelligent retrieval and utilization of railway construction process data.

[0018] According to the above embodiment, preferably, it also includes: training the vertical domain big model for construction: using the multi-source heterogeneous data of railway construction as the input of the vertical domain big model for construction, using the text generation, category labels, structured information extraction and prediction results as the output of the vertical domain big model for construction, and training the vertical domain big model for construction until convergence.

[0019] According to the above embodiment, preferably, the steps of processing construction process data entities based on graph convolutional networks and graph attention networks and outputting local and global relationships between entities specifically include: during the training process, firstly using the graph convolutional network to capture the local connection features between entities: the graph convolutional network integrates the information of adjacent nodes into the current node through the neighborhood aggregation operation, so that the features of each node not only contain its own information, but also incorporate the contextual information of the neighborhood structure; then the graph attention network is introduced to dynamically calculate the weights of the edges: the graph attention network assigns adaptive weights between neighborhood nodes through the attention mechanism, thereby enhancing the ability to express key relationships.

[0020] According to the above embodiment, preferably, the business scenarios specifically include: construction progress analysis, concrete usage analysis, secondary lining construction timing judgment and secondary lining concrete quality data tracking.

[0021] According to the above embodiment, preferably, the indexing system contains all entity information of construction data, technical specifications and business scenarios.

[0022] According to the above embodiment, preferably, the multi-source heterogeneous data comes from actual construction documents, technical specification documents and business scenario documents.

[0023] According to the above embodiment, preferably, the database includes: a relational database, a graph database and a vector database.

[0024] According to another embodiment of the present invention, a specific business scenario related to railway tunnel construction is also disclosed, which mainly includes the following four aspects: construction progress analysis, concrete usage analysis, secondary lining construction timing judgment and secondary lining concrete quality data tracking.

[0025] Scenario description of construction progress analysis: In the construction of railway tunnels using the drilling and blasting method, construction progress analysis is the process of comparing, counting and analyzing the actual construction progress of the tunnel face excavation, invert and secondary lining with the planned targets within a specific time period, a specific mileage range and a specific surrounding rock grade. It provides data support for the subsequent analysis of factors affecting the construction progress and the evaluation of the impact of influencing factors on the construction progress, cost, safety and quality. It includes four sub-scenarios: face excavation progress analysis, invert construction progress analysis, secondary lining construction progress analysis and cycle process time analysis. Through the comprehensive analysis of the construction progress data, the rationality of the construction group plan can be evaluated, optimization measures can be proposed, the construction plan can be adjusted, construction management can be strengthened, and the project can be completed smoothly as planned.

[0026] Entity objects of construction progress analysis: Based on the application scenario of construction progress analysis, the key fields of the original data are analyzed to obtain the entity object set of construction progress analysis, including: construction site, surrounding rock, progress index, face excavation, initial support, initial support of invert, invert filling, invert lining, lining arch wall, forecast conclusion, drilling, charging and blasting, slag discharge, frame erection and mesh hanging, locking feet and anchor rods, inspection, shotcrete spraying, and cross section excavation.

[0027] The relationship between entities is constructed as follows:

[0028] Logical relationship: Taking a certain period of time, a certain work point, and the excavation of the tunnel face as an example, the construction progress analysis logic of the railway tunnel drilling and blasting method is designed, including the following steps: (1) Determine the mileage range: Based on the daily advance of the tunnel face and the front mileage of the tunnel face, the excavation mileage range of the tunnel face is obtained; (2) Determine the surrounding rock conditions: Compare the forecast conclusion within the mileage range with the planned surrounding rock grade within the mileage range to determine whether the forecast is consistent with the design. If they are consistent, conduct subsequent analysis according to the original planned progress indicators; if they are inconsistent, conduct subsequent analysis according to the progress indicators corresponding to the construction method recommended in the forecast conclusion; (3) Calculate the actual footage data: Accumulate the daily footage of the tunnel face to obtain the total actual footage of the tunnel face, and calculate the actual average daily footage of the tunnel face.

[0029] (4) Calculation of planned advance index: Based on the excavation progress index of the tunnel face, the planned daily advance is accumulated and the planned average daily advance of the tunnel face is calculated.

[0030] (5) Data comparison and judgment: The actual average daily advance of the tunnel face is compared with the planned average daily advance of the tunnel face. If the actual average daily advance of the tunnel face is greater than the planned average daily advance of the tunnel face, the analysis ends. If the actual average daily advance of the tunnel face is less than the planned average daily advance of the tunnel face, the subsequent "cycle process time consumption analysis" is carried out.

[0031] (6) Cycle process time consumption analysis: Statistics are collected on the time consumption of the face excavation process within the mileage range. The cycle process time consumption within the mileage range is compared with the process assessment indicators. Based on the data on the factors affecting the cycle process, further reasons are analyzed.

[0032] Noun constraints: (1) The surrounding rock grade in the visualized surrounding rock list of the tunnel must be consistent with the surrounding rock grade recommended by the advanced geological forecast; (2) The actual average daily advance of the tunnel face must be greater than the planned average daily advance of the tunnel face.

[0033] Scenario description of concrete consumption analysis: In the construction of railway tunnels using the drilling and blasting method, concrete consumption analysis refers to the process of comparing, counting and analyzing the actual consumption of concrete with the planned consumption at a specific construction stage and under a specific surrounding rock grade. Through a comprehensive analysis of concrete consumption data, the difference between the actual concrete consumption and the designed amount in construction is revealed, and the specific reasons for overconsumption or saving are analyzed. It includes four sub-scenarios: initial support concrete consumption analysis, secondary lining concrete consumption analysis, invert construction concrete consumption analysis, and concrete overconsumption analysis. Through concrete consumption analysis, the rationality of the construction plan is evaluated, optimization measures are proposed, the construction plan is adjusted, and construction management is strengthened to ensure that the project is completed smoothly as planned.

[0034] Entity objects for concrete usage analysis: Based on the application scenario of concrete usage analysis, the key fields of the original data are analyzed to obtain a set of entity objects for concrete usage analysis, including: construction site, concrete, over-excavation volume, over-limit volume, face excavation, initial support, initial support of invert, invert filling, invert lining, lining arch wall, actual usage, disc volume, and excess consumption.

[0035] The relationship between entities is constructed as follows:

[0036] Logical relationship: Taking the initial support of a certain work point in a certain period of time as an example, the analysis logic of concrete consumption in railway tunnel drilling and blasting construction is designed, including the following steps: (1) Planned volume of concrete used: Based on the construction site, the planned volume of concrete used for that construction site is obtained through the concrete production notice; (2) Actual production volume: Based on the concrete production notice number, the volume of concrete produced at the construction site is obtained through the concrete production ledger, which is the actual production volume; (3) Actual completion volume: The completion volume for the day is accumulated to obtain the actual completion volume of concrete during the time period.

[0037] (4) Determination of concrete overconsumption: Compare the actual volume of concrete completed during the time period with the actual volume of concrete produced. If the actual volume of concrete produced is equal to the actual volume completed, there is no overconsumption of concrete and the analysis ends. If the actual volume of concrete produced is greater than the actual volume completed, there is overconsumption of concrete and subsequent analysis is performed.

[0038] (5) Calculation of excess consumption and excess consumption coefficient: The difference between the actual production volume and the actual completion volume is recorded as the concrete excess consumption, and the ratio of the actual production volume to the actual completion volume is recorded as the excess consumption coefficient.

[0039] (6) Overexcavation analysis: Based on the 3D cross-sectional scanning data during the face excavation stage, the overexcavation volume within the mileage range is calculated. The overexcavation volume is compared with the excess concrete consumption, and the proportion of the overexcavation volume in the excess consumption is calculated.

[0040] (7) Mix ratio analysis: Based on the initial support concrete mix ratio report data within the mileage range, compare the theoretical mix ratio with the construction mix ratio to analyze whether the concrete mix ratio meets the requirements of the specifications.

[0041] (9) Rebound rate analysis: Calculate the sum of the over-excavation volume and the planned concrete volume for the mileage. Based on this value, calculate the concrete rebound rate. Compare it with the relevant technical specifications to determine whether the rebound rate meets the requirements of the specifications.

[0042] Noun constraints: (1) The actual volume of concrete produced should be equal to the actual volume completed and the planned volume used; (2) The concrete mix ratio should meet the requirements of relevant technical specifications; (3) The rebound rate of initial support concrete should meet the requirements of relevant technical specifications.

[0043] Scenario description of the timing of the second lining: In the railway tunnel drilling and blasting construction, the compliance cross-verification of the secondary lining construction is to conduct a comprehensive analysis and cross-verification of the compliance of the secondary lining construction through surrounding rock monitoring data and construction logs, ensuring structural safety and stability, improving construction quality, optimizing project progress, reducing project costs, and complying with specifications and regulatory requirements.

[0044] Entity objects for judging the timing of secondary lining construction: Based on the application scenario of judging the timing of secondary lining construction, the key fields of the original data are analyzed to obtain the entity object set for judging the timing of secondary lining construction, including: construction site, surrounding rock, progress index, face excavation, initial support, initial support of invert, invert filling, invert lining, lining arch wall, forecast conclusion, drilling, charging and blasting, slag discharge, erection and mesh hanging, locking feet and anchor rods, inspection, shotcrete, and section digging.

[0045]

[0046] Logical relationship: Taking the judgment of the timing of secondary lining construction within a certain mileage range of a certain work point as an example, the analysis logic of the judgment of the timing of secondary lining construction of railway tunnels is designed, including the following steps: (1) According to the mileage range, the monitoring data of the surrounding rock monitoring points within the mileage range / closest to the mileage range are searched, the monitoring frequency is calculated, and compared with the value required by the technical specifications. If the technical specifications are met, the analysis is continued; if not, the output is "the monitoring data does not meet the requirements of the specifications, and it is impossible to correctly judge the timing of the secondary lining construction."

[0047] (2) Based on the surrounding rock monitoring data, the deformation rate of the surrounding rock periphery convergence and the vault sinking is calculated, and the measurement time when the surrounding rock deformation reaches stability is determined according to the technical specifications; if the surrounding rock has not reached stability as of the current date, the output is "the surrounding rock deformation is not stable and the secondary lining should not be constructed."

[0048] (3) Search the construction log according to the mileage range to determine the construction date of the secondary lining; and search the concrete production notice and production ledger to determine the use time of the concrete required for the secondary lining within the mileage range and compare them. If the dates are consistent, continue the analysis; if not, continue the analysis based on the use date of the secondary lining concrete in the concrete production notice and production ledger.

[0049] (4) According to the construction log, calculate the distance from the tunnel face to the secondary lining and the distance from the tunnel face to the invert, and compare and analyze them with the technical specifications. If both distances meet the safety step distance requirements in the technical specifications, continue the analysis; if not, output "the safety step distance does not meet the technical specifications and is not suitable for construction."

[0050] (5) Compare the sequence of the measurement time when the surrounding rock deformation reaches stability in step (2) and the construction date of the second lining in step (3). If the construction date of the second lining is later than the measurement time when the surrounding rock deformation reaches stability, output “the construction timing of the second lining is reasonable and meets the requirements of the technical specifications”; if it does not meet the requirements, output “the construction timing of the second lining does not meet the requirements of the technical specifications”.

[0051] Noun constraints: (1) The distance from the tunnel face to the secondary lining and the distance from the tunnel face to the invert are less than the safe step distance specified in the technical specifications; (2) The time for the surrounding rock deformation to stabilize is determined according to the requirements for the surrounding rock deformation rate in the technical specifications; (3) The construction time of the secondary lining in the construction log is consistent with the concrete production time required for the secondary lining construction in the concrete production notice and production ledger; (4) The frequency of surrounding rock deformation monitoring is greater than the surrounding rock deformation monitoring frequency requirements in the technical specifications; (5) The construction time of the secondary lining must be later than the measurement time of the surrounding rock deformation stabilization.

[0052] Scenario description of secondary lining concrete quality data tracking: In the construction of railway tunnels using the drilling and blasting method, secondary lining concrete quality data tracking is to compare and analyze the strength of the secondary lining concrete based on the compressive strength and rebound strength of the specimens, and accurately evaluate it. Summarize the problems existing in the design and construction process of the secondary lining concrete, ensure structural safety, improve construction quality, reduce economic costs, and optimize construction technology.

[0053] Secondary lining concrete quality data tracking entity object:

[0054] Logical relationship: Taking the secondary lining concrete quality data tracking within a certain mileage range of a certain work point as an example, the secondary lining concrete quality data tracking and analysis logic of the railway tunnel is designed, including the following steps: (1) Find the mix ratio report of the concrete used in the secondary lining according to the mileage range, and compare the raw material inspection results in the report with the raw material quality requirements in the technical specifications. If the requirements are met, continue the analysis; otherwise, output "the quality of the concrete raw materials does not meet the requirements."

[0055] (2) Search for the compressive strength test report of the concrete inspection specimen according to the mileage range, and make a consistency judgment on the concrete mix ratio in the report with the concrete mix ratio report in step (1), and compare it with the technical specifications. If the requirements are met, continue the analysis; otherwise, output "the concrete mix ratio does not meet the requirements".

[0056] (3) Search for the construction log of the secondary lining according to the mileage range, and compare the concrete pouring process in the construction instructions with the concrete pouring requirements in the technical specifications. If the requirements are met, continue the analysis; otherwise, output "concrete pouring does not meet the requirements."

[0057] (4) Search for the concrete maintenance record sheet based on the mileage range, and compare the maintenance methods (maintenance methods, temperature, and humidity) in the maintenance record with the technical specifications. If the requirements are met, continue the analysis; otherwise, output "concrete maintenance does not meet the requirements."

[0058] (5) Search for the concrete rebound strength test report based on the mileage range, calculate the curing period based on the pouring date and rebound date, and compare and analyze the curing period and technical specifications in the concrete curing record sheet in step (4). If the requirements are met, continue the analysis; if not, output "concrete does not meet the curing period requirements."

[0059] (6) Compare and analyze the compressive strength of the concrete specimen in the compressive strength test report of step (2) and the concrete rebound strength in the concrete rebound strength test report of step (5) with the concrete design strength grade. If the compressive strength and rebound strength of the concrete specimen are both greater than the design strength, the requirements are met and the analysis continues; otherwise, the output is "the concrete strength does not meet the requirements".

[0060] (7) Search for the concrete apparent quality acceptance form according to the mileage range, compare the acceptance results with the technical specifications, and if the requirements are met, output "the quality of the secondary lining concrete within a certain mileage range of a certain work site meets the requirements"; if the requirements are not met, output "the apparent quality of the concrete does not meet the acceptance requirements".

[0061] Noun constraints: (1) The mix ratio in the concrete mix report must be consistent with that in the compressive strength test report of the concrete inspection specimen; (2) The raw material test results in the concrete mix report must meet the raw material quality requirements in the technical specifications; (3) The compressive strength in the compressive strength test report and the rebound strength report of the concrete inspection specimen must be greater than the design strength grade; (4) The concrete mix ratio, casting requirements, thickness, curing conditions, and apparent quality must all meet the corresponding requirements in the technical specifications.

[0062] like Figure 2 As shown, according to another embodiment of the present invention, a construction process data retrieval system 200 is also disclosed, including: a memory 201, used to store program instructions; a processor 202, used to call the program instructions stored in the memory to implement the construction process data retrieval method as described in the above embodiment.

[0063] In this embodiment, the construction process data retrieval system specifically includes: Multi-source heterogeneous data parsing module: It is used to call multi-category document parsing agents based on large model technology according to the various formats of data in the railway construction process, and automatically identify and parse structured and unstructured data. Taking PDF documents as an example, the document parsing agent will call PyMuPDF or PDFMiner tools to extract PDF content.

[0064] Railway construction process data entity recognition module: used to extract and output construction process data entities based on regular expressions or rules according to the structured and unstructured data in the railway construction process. For example, information such as "construction mileage" and "excavation method" in railway engineering projects usually follows fixed naming rules, and regular expressions can be used to efficiently extract relevant information in batches. Construction mileage recognition: According to specific coding rules (such as the coding format of "DKXXX+YYY-ZZZ"), the system uses regular expressions to match all cells that meet the rules, and identifies and marks them as construction mileage entities. Excavation method classification: According to the predefined types in the engineering material list (such as "step method", "full section method", etc.), the system directly classifies the excavation method through field matching to ensure the consistency and accuracy of the classification results.

[0065] Railway construction process data entity relationship extraction module: It is used to output local and global relationships between entities based on the construction process data entities, based on the graph convolutional network (GCN) and graph attention network (GAT) technology. In the feature training process, GCN is first used to capture the local connection features between entities in the graph. GCN integrates the information of adjacent nodes into the current node through the neighborhood aggregation operation, so that the features of each node not only contain its own information, but also incorporate the contextual information of its neighborhood structure. Subsequently, GAT is introduced to dynamically calculate the weights of the edges. GAT uses the attention mechanism to assign adaptive weights between neighborhood nodes, enhancing the ability to express key relationships.

[0066] Spatiotemporal dynamic graph construction module: This module is based on the temporal graph neural network (Temporal GNN) and uses the local and global relationships between entities to model spatiotemporal dynamic relationships. Temporal GNN captures the dynamic characteristics of nodes and edges changing over time in the time dimension through hierarchical temporal learning, while supporting dynamic updates of the graph, thereby maintaining the consistency of the graph with the actual situation at the construction site.

[0067] Technical specification document information segmentation module: It is used to output text information summaries at all levels based on the text information of technical specification documents in railway construction and hierarchical recursive summary of text through a large model.

[0068] Technical specification mathematical formula extraction module: It is used to output LaTeX formula sequences based on the mathematical formula recognition model of the encoder-decoder according to the mathematical formulas of the technical specification documents in railway construction.

[0069] Technical specification mathematical formula and text fusion module: It is used to output the fusion data of text and LaTeX formula sequence based on the text information and LaTeX formula sequence of the technical specification documents in railway construction, based on the similarity relationship represented by the calculation of text and formula sequence vectors.

[0070] Business scenario element extraction module: used to output business scenario data entity relationships based on the business scenario document description of railway construction and the vertical domain big model of railway construction.

[0071] Business scenario logic flowchart extraction module: used to output the business scenario processing process sequence based on the business scenario flowchart of railway construction and flowchart extraction technology.

[0072] Multimodal information construction module: It is used to perform entity disambiguation between different types of data structures based on railway construction data, technical specification knowledge and business scenario knowledge, and establish a unified index system to index all entity information including railway construction data, technical specification knowledge and business scenario knowledge. Finally, different types of data structures are stored in different types of databases.

[0073] Hybrid retrieval module: used to select different hybrid retrieval methods according to business scenarios. The selection of hybrid retrieval methods is carried out through the instructions of the expert model, and then the retrieval method is determined to perform data search.

[0074] All or part of the steps in the various methods of the above embodiments can be completed by controlling the relevant hardware through a program, and the program can be stored in a readable storage medium, and the storage medium includes a read-only memory (ROM), a random access memory (RAM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), a one-time programmable read-only memory (OTPROM), an electronically erasable programmable read-only memory (EEPROM), a compact disc (CD-ROM) or other optical disc storage, magnetic disk storage, magnetic tape storage, or any other readable medium that can be used to carry or store data.

[0075] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A construction process data retrieval method, characterized in that: include: Construction data analysis: Based on the large model, multiple document analysis agents are called to identify multi-source heterogeneous data related to the construction process and output the identified content; Construction data modeling: Based on regular expressions or rule extraction methods, the structured data and unstructured data in the identified content are processed to output construction process data entities; based on graph convolutional networks and graph attention networks, the construction process data entities are processed to output local and global relationships between entities; based on time-series graph neural networks, local and global relationships between entities are processed to output spatiotemporal dynamic graphs of construction process data; Technical specification knowledge modeling: processing the technical specification documents in the identified content based on the big model, outputting a LaTeX formula sequence according to the mathematical formula information in the technical specification documents, outputting a text information summary according to the text information in the technical specification documents, and outputting fusion data of the text information and the LaTeX formula sequence according to the similarity relationship between the vector representation of the text information and the vector representation of the formula sequence; Business scenario logic modeling: Process the business scenario document description in the identified content based on the construction vertical domain big model, output the business scenario data entity relationship, process the business scenario flowchart of the construction process based on the flowchart extraction algorithm, output the business scenario processing process sequence, combine the business scenario data entity relationship and the business scenario processing process sequence, and output the business scenario logic map; Constructing a multimodal information database: performing entity disambiguation processing between different types of data structures based on the spatiotemporal dynamic graph, the fusion data of the text information and the LaTeX formula sequence, and the business scenario logic graph, establishing a unified indexing system, and storing different types of data structures in corresponding types of databases; Training the expert model: Use the business scenario data as the input of the expert model, and the scenario question answer text as the output of the expert model. Train the expert model until convergence to obtain a trained expert model. Hybrid retrieval: Receive natural language question text input by the user, process the natural language question text based on the trained expert model to determine the business scenario, and select different hybrid retrieval methods according to the business scenario to search for data in the multi-modal information database, and output the retrieval results.

2. The construction process data retrieval method according to claim 1, characterized in that: Also includes: Training the construction vertical domain big model: Use the multi-source heterogeneous data of railway construction as the input of the construction vertical domain big model, use the text generation, category label, structured information extraction and prediction results as the output of the construction vertical domain big model, and train the construction vertical domain big model until convergence.

3. The construction process data retrieval method according to claim 1, characterized in that: The step of processing the construction process data entities based on the graph convolution network and the graph attention network and outputting the local relationship and global relationship between the entities specifically includes: During the training process, the graph convolutional network is first used to capture the local connection features between entities: the graph convolutional network integrates the information of adjacent nodes into the current node through the neighborhood aggregation operation, so that the characteristics of each node not only contain its own information, but also incorporate the contextual information of the neighborhood structure; then the graph attention network is introduced to dynamically calculate the weights of the edges: the graph attention network assigns adaptive weights between neighborhood nodes through the attention mechanism, enhancing the ability to express key relationships.

4. The construction process data retrieval method according to claim 1, characterized in that: The business scenarios specifically include: construction progress analysis, concrete usage analysis, secondary lining construction timing judgment and secondary lining concrete quality data tracking.

5. The construction process data retrieval method according to claim 1, characterized in that: The index system contains all entity information of construction data, technical specifications and business scenarios.

6. The construction process data retrieval method according to any one of claims 1 to 5, characterized in that: The multi-source heterogeneous data comes from actual construction documents, technical specification documents and business scenario documents.

7. The construction process data retrieval method according to any one of claims 1 to 5, characterized in that: The database includes: a relational database, a graph database and a vector database.

8. Construction process data retrieval system, characterized in that: include: A memory for storing program instructions; A processor, configured to call the program instructions stored in the memory to implement the construction process data retrieval method as described in any one of claims 1 to 7.

Images