A Similarity Retrieval and Recommendation Method for Shield Tunneling Construction Cases Based on Multidimensional Feature Vectors
By constructing a multi-dimensional feature vector system and differential similarity calculation, and combining user feedback to optimize the case library, the problems of low efficiency and accuracy in shield tunneling construction scheme formulation were solved, and intelligent construction scheme recommendation was achieved.
Patent Information
- Application Number
- CN202611251399.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-18
- Publication Date
- 2026-11-17
AI Technical Summary
Existing technologies in shield tunneling construction scheme development suffer from single-dimensional feature extraction, lack of multi-dimensional similarity calculation, lack of intelligent recommendation in search results, and static case databases that lack dynamic optimization, resulting in low efficiency in construction scheme development and easy introduction of subjectivity.
A multi-dimensional feature vector system is constructed, a differentiated similarity calculation method is adopted, an intelligent ranking and recommendation mechanism is introduced, and the case library is dynamically optimized through user feedback and construction results.
It has achieved efficient and accurate retrieval and intelligent recommendation of shield tunneling construction cases, improved the retrieval accuracy under complex working conditions, and continuously improved the system performance through dynamic optimization.
Smart Images

Figure CN123115482A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of information technology and intelligent decision-making technology for tunnel boring machine (TBM) construction, specifically a method for similarity retrieval and recommendation of TBM construction cases based on multi-dimensional feature vectors. Background Technology
[0002] Shield tunneling is currently one of the mainstream methods for underground tunnel construction. The success rate of shield tunneling largely depends on whether the construction plan is suitable for the specific engineering geology, hydrogeology, and surrounding environmental conditions. Traditional construction plan formulation mainly relies on the experience and judgment of experts in the field. This involves collecting geological survey reports, hydrogeological data, design documents, and information about the surrounding environment, and then determining the construction plan after expert discussion. This method is labor-intensive, has a long evaluation cycle, and is prone to introducing subjective factors.
[0003] With the accumulation of extensive experience in tunnel boring machine (TBM) engineering, a vast number of historical construction cases contain valuable engineering experience and knowledge. How to quickly and accurately retrieve similar cases from these historical examples to effectively reuse engineering experience has become a crucial issue in the intelligentization of TBM construction.
[0004] Currently, some technologies have attempted to address this problem. For example, one patent proposes a shield tunneling machine (TBM) selection method based on case-based reasoning. This method stores past TBM projects and retrieves similar projects to recommend construction plans for new projects. The method searches according to three categories: soil layers, rock layers, and soil-rock composite layers. Search parameters include particle size distribution, TBM diameter, TBM burial depth, consistency index, and permeability coefficient. Another patent proposes a TBM selection system based on historical cases, which assists in selection by building a historical case database and matching it with new project parameters. Still other technical solutions employ natural language processing and vector space models to calculate the similarity of construction safety cases.
[0005] However, the existing technologies still have significant shortcomings. First, the feature extraction dimensions are relatively limited, mainly focusing on geological classification and a few selection parameters, and a systematic feature vector system covering multiple dimensions such as geological conditions, hydrological conditions, surrounding environment, shield tunneling parameters, construction technology, risk events, and disposal measures has not yet been established. Second, similarity calculations mostly rely on keyword matching or single-category retrieval, lacking a multi-dimensional joint similarity calculation mechanism tailored to the characteristics of shield tunneling construction, making it difficult to meet the accurate retrieval needs under complex working conditions. Third, the search results lack automatic sorting and intelligent recommendation functions based on comprehensive similarity, requiring users to manually verify each result. Fourth, the case database is mostly statically stored, lacking dynamic optimization and weight self-learning mechanisms based on feedback.
[0006] To address the aforementioned issues, this invention proposes a method for similarity retrieval and recommendation of shield tunneling construction cases based on multi-dimensional feature vectors. Summary of the Invention
[0007] To address the aforementioned problems in existing technologies, this invention provides a method for similarity retrieval and recommendation of shield tunneling construction cases based on multidimensional feature vectors. By constructing a systematic multidimensional feature vector system, designing differentiated joint similarity calculation methods, establishing an intelligent ranking and recommendation mechanism, and introducing a feedback-based dynamic optimization strategy, this method achieves efficient and accurate retrieval and intelligent recommendation of historical shield tunneling construction cases, providing scientific decision support for the formulation of shield tunneling construction plans under extreme environments.
[0008] The technical solution to achieve the above objectives is: A method for similarity retrieval and recommendation of shield tunneling construction cases based on multidimensional feature vectors includes: Step S1: Construct a multi-dimensional structured feature vector system for shield tunneling construction cases, wherein the multi-dimensional structured feature vector... This includes dimensions of geological conditions, hydrological conditions, surrounding environment, tunnel boring equipment, construction technology, risk events, mitigation measures, and construction effectiveness. Step S2: Obtain the multidimensional feature information of the current project to be constructed, and generate the feature vector of the current project according to the same dimension division and quantization rules as in step S1. ; Step S3: Calculate the multi-dimensional joint similarity between the current project feature vector and the feature vectors of each historical case; Step S4: Sort the historical cases from highest to lowest based on their overall similarity, and select the case with the highest similarity. One case is output as the recommendation result; Step S5: Dynamically optimize the case library and perform weight self-learning based on user feedback and construction results.
[0009] Preferably, in step S1, existing historical case data of shield tunneling construction are collected, including but not limited to basic project information, geological survey reports, hydrogeological data, design documents, construction logs, monitoring data, risk event records and handling reports; The following eight feature parameters are extracted from the above data to construct the multidimensional structured feature vector of the case. .
[0010] Preferably, in step S1, each dimension specifically includes the following sub-features: Geological conditions dimension This includes soil layer type and distribution, rock layer type and distribution, soil-rock composite ratio, surrounding rock classification, particle size distribution characteristics, consistency index, permeability coefficient, equivalent quartz content, limiting particle size, and maximum particle size. Hydrological conditions dimension This includes groundwater level, water head at the excavation face, aquifer characteristics, and water pressure; Surrounding environment dimension This includes the distribution of above-ground buildings, the status of underground pipelines, and the distances to existing tunnels and structures. Dimensions of tunnel boring machines This includes the type of tunnel boring machine (TBM), its diameter, cutterhead type, and drive power. Construction process dimension This includes tunneling parameters, grouting parameters, soil removal volume control, and attitude control; Risk event dimension This includes the type of risk, the risk level, the time and location of occurrence; Dimensions of Disposal Measures This includes control measures, emergency plans, and their effectiveness; Construction effect dimension This includes the effectiveness of surface settlement control, project completion status, and quality acceptance results.
[0011] Preferably, in step S1, the sub-features under each dimension are quantized to achieve structured representation: Numerical features are processed using either raw numerical values or normalized values; Categorical features employ encoding mapping; Textual features extract key semantic information through domain dictionaries and natural language processing techniques.
[0012] Preferably, in step S3, for each historical case in the case library... Calculate its eigenvectors relative to the current project. Overall similarity The overall similarity is composed of a weighted sum of the similarities across all dimensions: ; In the formula, For the current project With the Historical Cases The overall similarity between them For the first Weight coefficients for each dimension For the current project With the Historical Cases In the Dimensional similarity across each feature dimension For the current project In the Feature vector values in each feature dimension For the first Historical Cases In the Feature vector values in each feature dimension; A differentiated strategy is used to calculate the similarity across all dimensions: For numerical features, the Gaussian similarity function is used for calculation, and the formula is as follows: ; ; In the formula, This is the normalized Euclidean distance. These are the numerical characteristic values of the project currently under construction. This refers to the numerical feature value corresponding to a specific case in the historical case database. The Gaussian kernel width parameter, This represents the maximum value of the numerical feature across all historical cases. This is the minimum value of the numerical feature across all historical cases; For categorical features, a hierarchical matching degree is used for calculation: similarity is 1.0 when the types are completely identical; similarity is 0.6-0.8 when they belong to the same major category but different subcategories; similarity is 0.3-0.5 when they belong to the same coarser-grained category; and similarity is 0 when they are irrelevant. For text-based features, cosine similarity based on domain word vectors is used for calculation, with the following formula: ; In the formula, The semantic similarity between the current engineering text features and the historical case text features. This is the text semantic vector of the current project. This is the textual semantic vector of historical cases.
[0013] Preferably, in step S3, the weight coefficients of each dimension in the weighted sum of similarities across dimensions are initially determined using the analytic hierarchy process (AHP) and dynamically updated using gradient descent, and each weight coefficient satisfies the following: .
[0014] Preferably, in step S4, the recommendation results include the comprehensive similarity score of each recommended case, the similarity details of each dimension, the key information summary of the recommended case, and the explanation of the recommendation reasons; The recommendation results are presented in a visual format, including a similarity radar chart, a dimensional contribution bar chart, and a case information comparison table.
[0015] Preferably, in step S5, the dynamic optimization of the case library and weight self-learning based on user feedback and construction results includes: The gradient descent method is used to automatically adjust the weight coefficients of each dimension. The formula is as follows: ; In the formula, For the updated number Weight coefficients for each feature dimension For the previous version The weight coefficients of each feature dimension, i.e., the current weights. For learning rate, The overall score is based on the actual construction effect of the current project. The expected construction effect score for recommended cases is directly taken from the feature values of the construction effect dimension of historical cases. For comprehensive similarity For the Weights of each dimension The partial derivatives; Once the current project is completed, it will be stored in the case library as a new historical case. Cases are marked with quality labels based on their construction results, and high-quality cases are given higher priority in the search.
[0016] Compared with the prior art, the beneficial effects of the present invention are: This invention breaks through the limitations of existing technologies that only focus on geological classification and a few selection parameters. It constructs a systematic feature vector system covering eight dimensions, including geological conditions, hydrological conditions, surrounding environment, shield tunneling equipment, construction technology, risk events, disposal measures, and construction effects, thereby realizing the digital and structured comprehensive representation of shield tunneling construction cases. This invention employs differentiated similarity calculation strategies for features of different dimensions—Gaussian similarity, hierarchical matching degree, and semantic similarity algorithms are used for numerical, categorical, and textual features, respectively, and a comprehensive similarity is obtained through weighted fusion. This overcomes the limitations of traditional single similarity calculation methods and significantly improves retrieval accuracy under complex and extreme conditions. This invention not only outputs a sorted list of recommended cases, but also provides multi-level information such as comprehensive similarity scores, details of similarity in each dimension, and explanations of the reasons for recommendation, to help users quickly understand the basis for recommendation and make decisions. This invention introduces a weight adaptive adjustment and case library incremental update mechanism based on user feedback and construction results, enabling the systematic method framework for shield tunneling construction case similarity retrieval and recommendation to be continuously optimized as the case library expands and feedback accumulates, overcoming the limitations of static storage of case libraries in existing technologies. Attached Figure Description
[0017] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1This is a flowchart of the shield tunneling construction case similarity retrieval and recommendation method based on multi-dimensional feature vectors according to the present invention; Figure 2 This is a structural block diagram of the multi-dimensional structured feature vector system in the shield tunneling construction case of this invention; Figure 3 This is a flowchart illustrating the multi-dimensional joint similarity calculation method in this invention; Figure 4 This is a schematic diagram of the feedback-based adaptive weight adjustment and dynamic optimization process of the case library in this invention. Detailed Implementation
[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0019] like Figure 1 As shown, the shield tunneling construction case similarity retrieval and recommendation method based on multi-dimensional feature vectors includes: Step S1: Construct a multi-dimensional structured feature vector system for shield tunneling construction cases, wherein the multi-dimensional structured feature vector... It includes dimensions of geological conditions, hydrological conditions, surrounding environment, tunnel boring equipment, construction technology, risk events, response measures, and construction results.
[0020] In this embodiment, existing historical case data on tunnel boring machine (TBM) construction are collected, including but not limited to basic project information, geological survey reports, hydrogeological data, design documents, construction logs, monitoring data, risk event records and handling reports; The following eight feature parameters are extracted from the above data to construct the multidimensional structured feature vector of the case. ; like Figure 2 As shown, each dimension specifically includes the following sub-features: Geological conditions dimension This includes soil layer type and distribution, rock layer type and distribution, soil-rock composite ratio, surrounding rock classification, particle size distribution characteristics, consistency index, permeability coefficient, equivalent quartz content, limiting particle size, and maximum particle size. Hydrological conditions dimension This includes groundwater level, water head at the excavation face, aquifer characteristics, and water pressure; Surrounding environment dimension This includes the distribution of above-ground buildings, the status of underground pipelines, and the distances to existing tunnels and structures. Dimensions of tunnel boring machines This includes the shield type (earth pressure balance / slurry balance), shield diameter, cutterhead type, and drive power; Construction process dimension This includes tunneling parameters (thrust, torque, tunneling speed), grouting parameters, soil removal control, and attitude control; Risk event dimension This includes the type of risk (collapse, water inrush, machine jamming, mud cake formation, excessive surface subsidence, etc.), risk level, time and location of occurrence; Dimensions of Disposal Measures This includes control measures, emergency plans, and their effectiveness; Construction effect dimension This includes the effectiveness of surface settlement control, project completion status, and quality acceptance results.
[0021] In this embodiment, the sub-features under each dimension are quantized to achieve structured representation: Numerical features are processed using either raw numerical values or normalized values; Categorical features employ encoding mapping; Textual features extract key semantic information through domain dictionaries and natural language processing techniques.
[0022] Step S2: Obtain the multidimensional feature information of the current project to be constructed, and generate the feature vector of the current project according to the same dimension division and quantization rules as in Step S1. .
[0023] Step S3: Calculate the multi-dimensional joint similarity between the current project feature vector and the feature vectors of each historical case.
[0024] like Figure 3 As shown, for each historical case in the case library Calculate its eigenvectors relative to the current project. Overall similarity The overall similarity is composed of a weighted sum of the similarities across all dimensions: ; In the formula, For the current project With the Historical Cases The overall similarity between them For the first Weight coefficients for each dimension For the current project With the Historical Cases In the Dimensional similarity across each feature dimension For the current project In the Feature vector values in each feature dimension For the first Historical Cases In the Feature vector values in each feature dimension; A differentiated strategy is used to calculate the similarity across all dimensions: For numerical features (such as shield diameter, burial depth, permeability coefficient, etc.), the Gaussian similarity function is used for calculation, and the formula is: ; ; In the formula, This is the normalized Euclidean distance. These are the numerical characteristic values of the project currently under construction. This refers to the numerical feature value corresponding to a specific case in the historical case database. The Gaussian kernel width parameter is set to 0.15–0.25. This represents the maximum value of the numerical feature across all historical cases. This is the minimum value of the numerical feature across all historical cases; For categorized features (such as soil type, shield tunnel type, risk type, etc.), a hierarchical matching degree is used for calculation: the similarity is 1.0 when the types are completely identical; the similarity is 0.6-0.8 when they belong to the same major category but different subcategories; the similarity is 0.3-0.5 when they belong to the same coarser-grained category; and 0 when they are irrelevant. For text-based features (such as engineering descriptions, risk descriptions, etc.), cosine similarity based on domain word vectors is used for calculation, with the following formula: ; In the formula, The semantic similarity between the current engineering text features and the historical case text features. This is the text semantic vector of the current project. This is the textual semantic vector of historical cases.
[0025] In this embodiment, in the weighted sum of similarities across dimensions, the weight coefficients of each dimension are initially determined using the analytic hierarchy process (AHP) and dynamically updated using gradient descent, and each weight coefficient satisfies the following: .
[0026] Step S4: Sort the historical cases from highest to lowest based on their overall similarity, and select the case with the highest similarity. Each case is output as a recommendation result.
[0027] In this example, the recommendation results include the overall similarity score of each recommended case, the details of similarity in each dimension, the key information summary of the recommended case, and the explanation of the recommendation reasons; The recommendation results are presented in a visual format, including a similarity radar chart, a dimensional contribution bar chart, and a case information comparison table.
[0028] Step S5: Dynamically optimize the case library and perform weight self-learning based on user feedback and construction results.
[0029] like Figure 4 As shown, the dynamic optimization of the case library and weight self-learning based on user feedback and construction results include: The gradient descent method is used to automatically adjust the weight coefficients of each dimension. The formula is as follows: ; In the formula, For the updated number Weight coefficients for each feature dimension For the previous version The weight coefficients of each feature dimension, i.e., the current weights. For learning rate, The overall score is based on the actual construction effect of the current project. The expected construction effect score for recommended cases is directly taken from the feature values of the construction effect dimension of historical cases. For comprehensive similarity For the Weights of each dimension The partial derivatives; Once the current project is completed, it will be stored in the case library as a new historical case. Cases are marked with quality labels based on their construction results, and high-quality cases are given higher priority in the search.
[0030] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. The following embodiments are only used to illustrate the present invention and should not be regarded as limiting the scope of the present invention.
[0031] Implementation Cases: Case Studies of Cross-River Shield Tunnel Construction under High Water Pressure Environment 1. Project Background A cross-river tunnel project is planned to be constructed using the shield tunneling method. The tunnel is approximately 6.4 km long, with a maximum underwater depth of 54 m and a maximum water pressure of 0.76 MPa. The strata it traverses are primarily high-water-pressure, highly permeable soft soil, characterized by high water content, high compressibility, high rheological properties, and low strength. The tunnel lies above the main channel of the Yangtze River, presenting complex navigation conditions and posing extremely stringent requirements for construction disturbance control. Engineers need to search through historical case studies of shield tunneling construction in high-water-pressure environments to reference waterproofing and sealing solutions, tunneling parameter control, and risk mitigation measures.
[0032] 2. Implementation of Step S1 – Case Library Construction The system has pre-built a multi-dimensional feature vector library containing over 180 historical shield tunneling cases, including 12 high-water-pressure (≥0.5MPa) river-crossing and sea-crossing shield tunneling cases. Each case is stored in the following format: Example Case —Jiangyin-Jingjiang Yangtze River Tunnel Project Geological conditions dimension: High water pressure and strong permeability soft soil strata (Code: S-3-2, soft soil type - high compressibility - strong permeability) Hydrological conditions: head 52m, water pressure 0.73MPa, water-rich confined aquifer Surrounding environment dimension: Cross-river project, busy navigation (Code: E-2) Tunnel boring machine dimensions: slurry balance shield, diameter 16.1m (code: S-2-1) Construction technology dimension: asymmetric freezing method, visual system for opposing tunneling Risk event dimensions: High water pressure seal failure, shield attitude loss (code: R-1-1, R-2-3) Treatment measures include: watertight welding, triple-ring tightening of segment bolts, and secondary grouting. Construction effect dimension: Connection accuracy 2.3mm, zero leakage (score: 0.95) 3. Implementation of step S2 – Input of current project feature vector Engineers input basic information about the current project, and the system automatically extracts and quantifies features across various dimensions to generate a feature vector for the current project. As shown in Table 1: Table 1 4. Implementation of Step S3 – Multi-dimensional Joint Similarity Calculation The system traverses the case library and calculates each case one by one. Overall similarity with various historical cases. (Based on historical cases) (Taking the Jiangyin Jingjiang Yangtze River Tunnel Project as an example): (1) Geological conditions dimension (weight) ) Both are soft soil strata with high water pressure and strong permeability, and their classification characteristics are completely matched: ; (2) Hydrological conditions dimension (weight) ) The water head heights are 54m and 52m respectively, with a range of 10-80m. ; Pick : ; The water pressure similarity is approximately 0.98, which is consistent with the overall trend. ; (3) Surrounding environment dimension (weight) ) All are river-crossing projects with busy navigation; the codes match perfectly. ; (4) Shield tunneling equipment dimension (weight) ) Both are slurry balance shield tunnels, with diameters of 16.0m and 16.1m respectively, ranging from 3.0 to 20.0m. ; ; (5) Construction technology dimension (weight) ) The current construction process is not yet determined, but recommended process characteristics can be deduced based on the input geological and hydrological conditions. This is based on the semantic similarity of the process description text. (Some processes, such as freezing schemes and tunneling control, have a high degree of compatibility). (6) Risk event dimension (weight) ) Both involve risks such as high water pressure seal failure and shield attitude control. ; (7) Dimension of Disposal Measures (Weight) ) The handling measures of historical cases have high reference value for current projects: ; (8) Construction effect dimension (weight) ) Historical case studies show good construction results (score 0.95): ; Overall similarity calculation: ; 5. Implementation of Step S4 – Sorting and Recommendation The similarity calculation results for all cases were sorted, and the recommendation results are shown in Table 2: Table 2 Simultaneously display The reasons for the recommendation are: "The hydrological conditions are highly compatible (similar water head, with a deviation of only 3.7%), the geological types are completely consistent (both are soft soils with high water pressure and strong permeability), the surrounding environmental characteristics are the same (busy cross-river navigation conditions), and the sealing scheme and attitude control measures have direct reference value." Key information summary of the recommended case: Asymmetric freezing method was used for formation water control; Create a visualization system for opposing tunneling to collect tunneling data and environmental information in real time; The shield tunnel attitude is accurately monitored using the "same-side intersecting double-wire" measurement technology; Watertight welding is used to reinforce the seal under high water pressure. Innovative processes such as "three-ring re-tightening" of segment bolts and secondary grouting were used. Construction results: The connection accuracy is 2.3mm, and the leakage is zero.
[0033] 6. Implementation of Step S5 – Feedback and Dynamic Optimization The engineer adopted the case study. The existing freezing and water control scheme and sealing enhancement measures were adjusted and implemented based on the actual conditions of this project. No major leakage incidents occurred during construction, and the tunnel boring machine docking achieved millimeter-level precision.
[0034] After construction is completed, collect feedback information: User adoption status: Fully adopted (rating 1.0); Actual construction results: excellent leakage control and good shield posture (overall score 0.96). Compared with the expected performance of the recommended case (0.95): , .
[0035] The system performs adaptive weight adjustment (taking) ): Hydrological condition dimension weights: ; Geological condition dimension weights: ; After normalization, the weight of hydrological conditions increased slightly (reflecting the high predictive power of this dimension). At the same time, this project (high water pressure cross-river shield tunnel) will be stored in the case library as a new high-quality case, marked with a quality level of A, for reference in subsequent projects.
[0036] (III) Method for determining the initial weight assignment The initial weights for each dimension were determined using the analytic hierarchy process (AHP). The specific steps are as follows: Step 1: Invite 5 experts in the field of tunnel boring machine (TBM) construction to conduct pairwise comparisons of the relative importance of the eight dimensions according to the Saaty scale (1-9 scale) to construct a judgment matrix. .
[0037] Step 2: Calculate the largest eigenvalue of the judgment matrix and its corresponding eigenvector, and obtain the initial weight vector after normalization. .
[0038] Step 3: Perform a consistency check and calculate the consistency ratio. ,when It is assumed that the judgment matrix satisfies the consistency requirement.
[0039] Based on the above calculations, the initial weights for each dimension are shown in Table 3: Table 3 (iv) Cold start strategy In the initial stage of case database accumulation (number of cases < 50), the fixed weights determined by the above-mentioned analytic hierarchy process are used for similarity calculation. Once the case database exceeds 50 cases, the weight adaptive adjustment mechanism described in step S5 is activated, and the weight coefficients are continuously optimized using the gradient descent method.
[0040] (v) System Implementation This invention can be deployed as a client-server architecture or a cloud-based SaaS service platform. Users input current project feature information through a web interface or mobile terminal, the server performs similarity calculations and case recommendations, and returns the results to the user in a visual format.
[0041] Finally, it should be noted that the above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A method for similarity retrieval and recommendation of shield tunneling construction cases based on multi-dimensional feature vectors, characterized in that, include: Step S1: Construct a multi-dimensional structured feature vector system for shield tunneling construction cases, wherein the multi-dimensional structured feature vector... This includes dimensions of geological conditions, hydrological conditions, surrounding environment, tunnel boring equipment, construction technology, risk events, mitigation measures, and construction effectiveness. Step S2: Obtain the multidimensional feature information of the current project to be constructed, and generate the feature vector of the current project according to the same dimension division and quantization rules as in step S1. ; Step S3: Calculate the multi-dimensional joint similarity between the current project feature vector and the feature vectors of each historical case; Step S4: Sort the historical cases from highest to lowest based on their overall similarity, and select the case with the highest similarity. One case is output as the recommendation result; Step S5: Dynamically optimize the case library and perform weight self-learning based on user feedback and construction results.
2. The shield tunneling construction case similarity retrieval and recommendation method based on multi-dimensional feature vectors according to claim 1, characterized in that, In step S1, existing historical case data of shield tunneling construction are collected, including but not limited to basic project information, geological survey reports, hydrogeological data, design documents, construction logs, monitoring data, risk event records and handling reports; Extract the following eight feature parameters from the above data to construct the multidimensional structured feature vector of the case. .
3. The shield tunneling construction case similarity retrieval and recommendation method based on multi-dimensional feature vectors according to claim 2, characterized in that, In step S1, each dimension specifically includes the following sub-features: Geological conditions dimension This includes soil layer type and distribution, rock layer type and distribution, soil-rock composite ratio, surrounding rock classification, particle size distribution characteristics, consistency index, permeability coefficient, equivalent quartz content, limiting particle size, and maximum particle size. Hydrological conditions dimension This includes groundwater level, water head at the excavation face, aquifer characteristics, and water pressure; Surrounding environment dimension This includes the distribution of above-ground buildings, the status of underground pipelines, and the distances to existing tunnels and structures. Dimensions of tunnel boring machines This includes the type of tunnel boring machine (TBM), its diameter, cutterhead type, and drive power. Construction process dimension This includes tunneling parameters, grouting parameters, soil removal volume control, and attitude control; Risk event dimension This includes the type of risk, the risk level, the time and location of occurrence; Dimensions of Disposal Measures This includes control measures, emergency plans, and their effectiveness; Construction effect dimension This includes the effectiveness of surface settlement control, project completion status, and quality acceptance results.
4. The method for similarity retrieval and recommendation of shield tunneling construction cases based on multi-dimensional feature vectors according to claim 1, characterized in that, In step S1, the sub-features under each dimension are quantized to achieve structured representation: Numerical features are processed using either raw numerical values or normalized values; Categorical features employ encoding mapping; Textual features extract key semantic information through domain dictionaries and natural language processing techniques.
5. The shield tunneling construction case similarity retrieval and recommendation method based on multi-dimensional feature vectors according to claim 1, characterized in that, In step S3, for each historical case in the case library Calculate its feature vector with that of the current project. Overall similarity The overall similarity is composed of a weighted sum of the similarities across all dimensions: ; In the formula, For the current project With the Historical Cases The overall similarity between them For the first Weight coefficients for each dimension For the current project With the Historical Cases In the Dimensional similarity across each feature dimension For the current project In the Feature vector values in each feature dimension For the first Historical Cases In the Feature vector values in each feature dimension; A differentiated strategy is used to calculate the similarity across all dimensions: For numerical features, the Gaussian similarity function is used for calculation, and the formula is as follows: ; ; In the formula, This is the normalized Euclidean distance. These are the numerical characteristic values of the project currently under construction. This refers to the numerical feature value corresponding to a specific case in the historical case database. The Gaussian kernel width parameter, This represents the maximum value of the numerical feature across all historical cases. This is the minimum value of the numerical feature across all historical cases; For categorical features, a hierarchical matching degree is used for calculation: similarity is 1.0 when the types are completely identical; similarity is 0.6-0.8 when they belong to the same major category but different subcategories; similarity is 0.3-0.5 when they belong to the same coarser-grained category; and similarity is 0 when they are irrelevant. For text-based features, cosine similarity based on domain word vectors is used for calculation, with the following formula: ; In the formula, The semantic similarity between the current engineering text features and the historical case text features. This is the text semantic vector of the current project. This is the textual semantic vector of historical cases.
6. The shield tunneling construction case similarity retrieval and recommendation method based on multi-dimensional feature vectors according to claim 5, characterized in that, In step S3, the weight coefficients of each dimension in the weighted sum of similarities across dimensions are initially determined using the analytic hierarchy process (AHP) and dynamically updated using gradient descent. Each weight coefficient satisfies the following conditions: 。 7. The shield tunneling construction case similarity retrieval and recommendation method based on multi-dimensional feature vectors according to claim 1, characterized in that, In step S4, the recommendation results include the comprehensive similarity score of each recommended case, the similarity details of each dimension, the key information summary of the recommended case, and the explanation of the recommendation reasons; The recommendation results are presented in a visual format, including a similarity radar chart, a dimensional contribution bar chart, and a case information comparison table.
8. The method for similarity retrieval and recommendation of shield tunneling construction cases based on multi-dimensional feature vectors according to claim 6, characterized in that, In step S5, the dynamic optimization of the case library and weight self-learning based on user feedback and construction results include: The gradient descent method is used to automatically adjust the weight coefficients of each dimension. The formula is as follows: ; In the formula, For the updated number Weight coefficients for each feature dimension For the previous version The weight coefficients of each feature dimension, i.e., the current weights. For learning rate, The overall score is based on the actual construction effect of the current project. The expected construction effect score for recommended cases is directly taken from the feature values of the construction effect dimension of historical cases. For comprehensive similarity For the Weights of each dimension The partial derivatives; Once the current project is completed, it will be stored in the case library as a new historical case. Cases are marked with quality labels based on their construction results, and high-quality cases are given higher priority in the search.
