CN-121979854-A - TBM construction whole process intelligent decision expert knowledge base construction method

CN121979854ACN 121979854 ACN121979854 ACN 121979854ACN-121979854-A

Abstract

The invention relates to the technical field of geospatial information systems, and particularly discloses a method for constructing a TBM construction whole-process intelligent decision expert knowledge base, which comprises the steps of realizing full-automatic structuring of multi-source heterogeneous data aiming at the characteristics of scattered construction information, platform division and unstructured construction information of TBM clusters of tunnels along ultra-long and ultra-large railways; establishing a TBM construction whole process intelligent decision knowledge pattern layer, establishing a knowledge pattern with space-time correlation, realizing knowledge pattern inquiry and knowledge recommendation based on semantic similarity according to user inquiry content, and realizing knowledge base dynamic update. According to the invention, the knowledge graph is utilized to carry out association expression on TBM security risk information which is subjected to coverage investigation and design to a construction stage, a construction risk disposal expert knowledge base is constructed, and the expert knowledge base is mined by combining semantic matching and a graph algorithm, so that intelligent recommendation of risk prevention and control information facing different construction scenes is realized, and intelligent decision of the whole TBM construction process is effectively assisted.

Inventors

ZHU QING
HUANG QIYU
JIANG QIAN
DING YULIN
Wu Tingchen
LUO XUN

Assignees

西南交通大学
川藏铁路有限公司
中国国家铁路集团有限公司

Dates

Publication Date: 20260505
Application Date: 20251217

Claims (10)

1. The construction method of the intelligent decision expert knowledge base of the whole TBM construction process is characterized by comprising the following steps: s10, realizing full-automatic structuring of multi-source heterogeneous data according to the characteristics of construction information dispersion, platform division and unstructured of TBM clusters of tunnels along ultra-long and ultra-large railways; S20, aiming at the structural data with staged, multidimensional attribute and multimodal characteristics, constructing a knowledge graph mode layer for intelligent decision in the whole TBM construction process, and automatically constructing a knowledge graph with space-time correlation based on the knowledge graph mode layer to realize knowledge graph embedding; s30, according to the query content of the user, realizing knowledge graph query and knowledge recommendation based on semantic similarity; S40, based on the data source updating information and expert schemes obtained after user inquiry, the knowledge base is dynamically updated.
2. The method for constructing a knowledge base of intelligent decision expert in the whole process of TBM construction according to claim 1, wherein the method comprises the following steps: S101, aiming at different construction data types and different construction scenes, including advanced horizontal drilling, seismic wave reflection, electromagnetic wave reflection, sketch of a tunnel body, deepened blastholes, horizontal acoustic wave sections, transient electromagnetic waves, microseismic detection, sketch of a face and risk treatment information, realizing automatic acquisition and structuring of construction information; S102, aiming at multi-type and multi-source document data, including investigation design evaluation reports and change design reports, automatic extraction of information contained in the multi-type and multi-source document data is realized by a natural language processing mode, and the association index of the documents is established by storing the names, types, storage positions and relations among the associated documents in a graph database.
3. The method for constructing a knowledge base of intelligent decision expert in the whole process of TBM construction according to claim 1, wherein the method comprises the steps of: S201, aiming at TBM construction characteristics, building a TBM construction whole-process intelligent decision knowledge graph mode layer according to construction whole-process intelligent decision demand; S202, according to the constructed knowledge graph mode layer, based on structural data, automatically establishing a knowledge graph by using Neo4j, and realizing space-time correlation of risk investigation design, construction early warning and actually revealed information of each stage by taking a construction mileage section as a constraint; s203, embedding the knowledge graph according to the types of all nodes of the knowledge graph and the attribute information contained by the knowledge graph by using a BERT encoder.
4. The method for constructing a knowledge base of intelligent decision expert in the whole TBM construction process according to claim 3, wherein in step S202, 1) knowledge maps of design sections, construction sections and change sections of each TBM are respectively created according to entity types and relation types, 2) knowledge map nodes of each construction section and change section are traversed, segmented positions of each mileage section entity are obtained, and the segmented positions of the mileage section entity are judged to be located in the interval of the mileage position of the design section, and space-time association relation is established between the segmented positions and the knowledge map nodes of each mileage section.
5. The TBM construction whole process intelligent decision expert knowledge base construction method according to claim 4, wherein the entity type comprises a TBM entity, a mileage section entity, a surrounding rock grade entity, a detection method entity, a surrounding rock risk entity, a risk grade entity and a disposal scheme entity, and the relationship type comprises a mileage section relationship, a surrounding rock grade relationship, a surrounding rock risk relationship, a risk grade relationship, a disposal result relationship and a time-space association relationship.
6. The method for constructing a knowledge base of intelligent decision expert in the whole process of TBM construction according to claim 1, wherein the step S30 comprises the steps of: S301, cleaning the user query information, and embedding the cleaned query information by using the same BERT encoder; S302, obtaining the embedding of each node of the constructed knowledge graph, carrying out vector similarity comparison with the embedding of the query information of the user, and retrieving the knowledge graph node which is most matched with the query information; s303, recommending the knowledge contained in the retrieved knowledge graph to the user.
7. The method for constructing a TBM construction whole process intelligent decision expert knowledge base according to claim 6, wherein in step S301, the method specifically comprises: S3011, constructing a professional private dictionary, and extracting entities from the user query information based on the dictionary to obtain key information of the user query; S3012, extracting the obtained key information from the entity, carrying out syntactic reconstruction, and embedding all the reconstructed query sentences based on BERT to obtain semantic vectors.
8. The method for constructing a knowledge base of intelligent decision expert in the whole process of TBM construction according to claim 6, wherein in step S302, the method specifically comprises: S3021, obtaining the embedding of each node of the constructed knowledge graph and the unique identifier ID of each graph node; s3022, embedding the grammar reconstructed multiple user query embedded vectors into each node of each knowledge graph to search for vector similarity taking cosine similarity as a reference, selecting the node with the highest vector similarity in each search result, and obtaining the ID number of the node to further obtain the node related information.
9. The method for constructing a knowledge base of intelligent decision expert in the whole process of TBM construction according to claim 6, wherein in step S303, the method specifically comprises: s3031, a proprietary large model is deployed locally and fine-tuned to realize screening of questions raised by a user, and search query is not performed on questions related to a non-knowledge base; S3032, prompt word engineering is built based on LANGCHAIN, and node information obtained in the step S3022 is transmitted into a language model to return a user query result, so that knowledge recommendation is completed.
10. The method for constructing a knowledge base of intelligent decision expert in the whole process of TBM construction according to claim 1, wherein in step S40, the method specifically comprises: s401, establishing nodes based on data source updating information, and establishing connection between the newly established nodes and associated mileage segments in a knowledge graph according to the mileage segments of the updating information; s402, extracting information of a user query problem, creating nodes, and associating a solution proposed by an expert with the problem nodes to realize dynamic update of a knowledge base.

Description

TBM construction whole process intelligent decision expert knowledge base construction method Technical Field The invention relates to the technical field of geospatial information systems, in particular to a method for constructing a TBM construction whole-process intelligent decision expert knowledge base. Background The TBM cluster construction risk information of the tunnel along the ultra-long and ultra-large railway has the characteristics of multiple sources, multiple structures and multiple types, is low in information structuring degree, large in data redundancy and difficult to establish a high-efficiency and convenient use method, and the multi-source heterogeneous construction information management has the characteristics of dispersion, phasing, subsystem, and the like, and is insufficient in information space-time correlation of various stages such as risk investigation design, construction early warning and actual disclosure, so that powerful data support is difficult to provide for intelligent decision of the whole construction process. In order to meet the data service requirements of timely and effective construction safety risk prevention and control, automation of building from a data source to a knowledge base is realized, and intelligence of knowledge service is achieved from data retrieval, a TBM construction whole process intelligent decision expert knowledge base is needed to be built, and TBM cluster construction is energized. The existing database mainly uses a relational database, is suitable for application scenes with high requirements on structured data and transaction processing, is commonly used for business systems requiring strict data consistency and complex query, but is difficult to update, and difficult to express the space-time relevance of TBM construction information, and the graph database is suitable for application scenes requiring representing and querying complex relationships, is commonly used for data application requiring dynamic modes and high expandability, has flexible modes, can dynamically update data, and is easy to establish the relevance between the data. Based on the knowledge graph, the graph database is established to deeply integrate the advantages of a plurality of discipline theoretical methods such as semantic technology, data mining, deep learning, graph theory and the like. On the other hand, the tight combination of the current graph database and the large language model is a new paradigm of a retrieval enhancement technology (RAG) and is also a research hotspot in the current knowledge service field. The explicit knowledge in the graph database is combined with the implicit knowledge of the language model in an effort to solve the illusion problem inherent to the language model. The method has the advantages that the traditional numerous database query sentences are abandoned from natural language, semantic similarity is used as constraint, the graph database knowledge is searched, and the structured data contained in the knowledge graph nodes are converted into natural language to be used as output, so that users can understand the structured data conveniently, and end-to-end information service is realized, so that the method is one of the paths of the current efficient knowledge service. Disclosure of Invention Aiming at the defects of acquisition, management, application and the like of TBM cluster construction risk information of a tunnel along a current ultra-long and ultra-large railway, the invention provides a construction method of a TBM construction whole process intelligent decision expert knowledge base, which utilizes a knowledge graph to correlate and express TBM safety risk information which is designed to a construction stage by coverage investigation, constructs a construction risk disposal expert knowledge base, and combines semantic matching and a graph algorithm to mine the expert knowledge base, thereby realizing intelligent recommendation of risk prevention and control information facing different construction scenes, effectively assisting the intelligent decision of the TBM construction whole process, and solving the problems mentioned in the background art. In order to achieve the purpose, the invention provides the following technical scheme that the method for constructing the intelligent decision expert knowledge base in the whole TBM construction process comprises the following steps: s10, realizing full-automatic structuring of multi-source heterogeneous data according to the characteristics of construction information dispersion, platform division and unstructured of TBM clusters of tunnels along ultra-long and ultra-large railways; S20, aiming at the structural data with staged, multidimensional attribute and multimodal characteristics, constructing a knowledge graph mode layer for intelligent decision in the whole TBM construction process, and automatically constructing a knowledge graph with space-time cor