A method and device for sensing the lithology of a shield face

Abstract

The present application belongs to the technical field of tunnel and underground engineering geological detection, and discloses a method and device for sensing the lithology of a shield face, comprising the following steps: (1) selecting samples within a predetermined distance from a borehole in a sample point set to form a training sample set, and the remaining samples constituting a test sample set; (2) mapping the fluctuation scale of the vertical thickness of the lithology to an anisotropy scale factor in spatial interpolation, and then calculating the weight of each sample in the training sample set; (3) constructing an input feature vector; training a machine learning nonlinear model based on the input feature vector and a prior probability field to obtain a lithology real-time sensing model; (4) fusing the prior probability field and the sensing probability of the samples in the test sample set sensed by the lithology real-time sensing model to obtain a final lithology probability, and then reconstructing the shield face according to the final lithology probability. The present application improves the sensing accuracy.

Description

Technical Field

[0001] This invention belongs to the technical field of geological exploration for tunnels and underground engineering, and more specifically, relates to a method and equipment for sensing the lithological composition of a shield tunnel face. Background Technology

[0002] Currently, lithology sensing methods in tunnel boring machine (TBM) construction are mainly divided into two categories: one is based on prior geological knowledge, which uses static information such as borehole data and ordinary interpolation algorithms to construct geological models. However, it does not consider the differences in vertical heterogeneity and spatial continuity of strata, resulting in large prediction errors for lithology far from the borehole and an inability to adapt to dynamic changes in geological conditions during tunneling. The other is based on data analysis of real-time TBM parameters, which uses machine learning models to uncover rock-machine interaction patterns to achieve lithology identification. However, this type of method does not effectively integrate prior geological information, is easily affected by noise fluctuations in TBM parameters, and has insufficient prediction stability in high-confidence geological areas such as near the borehole. At the same time, the model training samples lack a dynamic screening mechanism, making it difficult to adapt to real-time changes in the geological environment, leading to a gradual decrease in prediction accuracy during long-term tunneling. Summary of the Invention

[0003] In view of the above-mentioned defects or improvement needs of the existing technology, the present invention provides a method and equipment for sensing the lithological composition of the tunnel face, which aims to solve the problems that the existing lithological sensing methods do not fully integrate prior geological knowledge with real-time tunneling data and do not consider the impact of the non-uniformity of the bottom layer on the sensing.

[0004] To achieve the above objectives, according to one aspect of the present invention, a method for sensing the lithological composition of a shield tunnel face is provided, comprising the following steps: (1) The heterogeneous geological information is uniformly transformed into a set of indexable sample points in three-dimensional space. Samples within a predetermined distance from the borehole are selected from the sample point set to form a training sample set, and the remaining samples constitute a test sample set. (2) The vertical thickness fluctuation scale of the lithology is mapped to the anisotropic scale factor in spatial interpolation, and then the weight of each sample in the training sample set is calculated by the anisotropic scale factor. (3) Perform sliding window statistical transformation on the original tunnel boring machine parameters to extract past data. The mean, standard deviation, and final state values ​​of the loop parameters are used to construct an input feature vector. Based on the obtained input feature vector and the prior probability field, a machine learning nonlinear model is trained to obtain a real-time lithology sensing model. The prior probability field is calculated from the weights of the samples. The location to be predicted Belongs to the Composed of the prior probabilities of lithological types; (4) The prior probability field and the samples in the test sample set are fused with the perceived probabilities perceived by the real-time lithology perception model to obtain the final lithology probability, and then the shield tunnel face is reconstructed based on the final lithology probability.

[0005] Furthermore, the anisotropic distance is calculated based on the obtained anisotropic scale factor, using the following formula:

[0006] In the formula, For the first The known high-confidence sample points and the... The inner ring Anisotropic distance between the locations to be predicted; and These are represented as anisotropy scaling factors in the horizontal and vertical directions, respectively; For the first The three-dimensional spatial location of a known high-confidence sample point; For the first The inner ring The three-dimensional spatial location of the location to be predicted.

[0007] Furthermore, the weight of each sample in the training sample set is calculated using an inverse distance weighting function, which is:

[0008] In the formula, For the first The nth known high-confidence sample point pair The weights of the positions to be predicted; For the first The known high-confidence sample points and the... The inner ring Anisotropic distance between the locations to be predicted; The distance decay exponent, To prevent singular positive numbers at zero distance.

[0009] Furthermore, the weights of the lithological indicator function and the samples are used to calculate the first... The inner ring One location to be predicted Belongs to the The prior probabilities of lithology for each class are used to form a prior probability field; the lithology indicator function is used to identify the first class. 1 known high-confidence sample point Does it belong to the first Lithology, expressed as:

[0010] In the formula, For the first 1 known high-confidence sample point Regarding the first Lithological indicator functions; For the first The lithological category number corresponding to each known high-confidence sample point; This represents the total number of lithological categories within the study area.

[0011] Furthermore, the first The inner ring One location to be predicted Belongs to the Prior probability of lithology The calculation formula is:

[0012] In the formula, This represents the number of known high-confidence sample points used in the prior probability calculation.

[0013] Furthermore, a dynamic fusion weight based on geological determinism is constructed. The dynamic fusion weight is related to the weight within the i-th ring. Distance from the predicted location to the nearest borehole It shows a negative correlation.

[0014] Furthermore, based on distance Different fusion strategies are adopted for different ranges to adapt to scenarios with different geological information credibility: when When located in the near field area, Approaching 0.8; when When in the transition zone, Follow The increase of decreases linearly; when When located in the far field region, It approaches 0.1.

[0015] Furthermore, when training the nonlinear model of machine learning, a model retraining interval is set. After each preset number of tunneling passes, the linear model of machine learning is fully fine-tuned using samples from the accumulated training sample set to ensure that the model can adapt to the latest geological environment changes in real time.

[0016] The present invention also provides a sensing system for the lithological composition of a shield tunnel face. The system includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to perform the sensing method for the lithological composition of a shield tunnel face as described above.

[0017] The present invention also provides a computer-readable storage medium storing machine-executable instructions, which, when invoked and executed by a processor, cause the processor to implement the method for sensing the lithological composition of the tunnel face as described above.

[0018] In summary, compared with the prior art, the method and equipment for sensing the lithological composition of the tunnel face provided by this invention have the following advantages: 1. This invention introduces... This method quantitatively characterizes the vertical heterogeneity of different lithologies and incorporates it into the calculation of anisotropic scale factors. The weight of each sample in the training sample set is then calculated using these anisotropic scale factors, making the use of prior geological knowledge more closely aligned with actual stratigraphic structures. Compared to traditional interpolation methods, the accuracy of lithology prediction in the near-borehole area is improved by 15%-20%. Simultaneously, the prior probability field and the perceived probabilities of samples in the test sample set, obtained through the real-time lithology perception model, are fused to obtain the final lithology probability. The shield tunnel face is then reconstructed based on this final lithology probability, effectively coupling prior experience with real-time tunneling data and improving perception accuracy.

[0019] 2. The weights of prior probability and real-time perception probability are adaptively allocated based on the distance between the location to be predicted and the borehole, which effectively balances the prediction accuracy of different geological regions. The global accuracy of lithology prediction in the entire tunneling section reaches 82.61%, and the accuracy in unknown areas (≥5m from the borehole) reaches 82.90%. The prediction accuracy of the six major lithologies ranges from 50.52% to 95.77%.

[0020] 3. Only samples closely related to the current tunneling ring and with high geological label reliability are included in the training sample set. This provides high-quality data support for the machine learning model, effectively reducing noise interference caused by shield parameter fluctuations. Compared to a single data-driven method without sample selection, the stability of the prediction results is improved by more than 40%. Simultaneously, a model retraining interval is set. After each preset number of tunneling rings, the accumulated high-confidence samples are used to perform full fine-tuning of the machine learning model. This allows the model to adapt to changes in the geological environment in real time, avoiding prediction accuracy decay during long-term tunneling and solving the problem that traditional fixed models are difficult to adapt to dynamic geological conditions.

[0021] 4. Based on distance Different fusion strategies are adopted for different ranges to adapt to scenarios with different geological information credibility: when (Near field area) Approaching 0.8, the prior probability obtained by strong trust space interpolation is... By increasing the weighting of the perception probability, noise interference caused by shield tunneling parameter fluctuations can be effectively suppressed, ensuring the stability of lithology prediction in the near-borehole area; when (Transition zone) Follow The increase in probability decreases linearly, gradually increasing the perceived probability. The weighting ratio is used to achieve a smooth transition between prior geological knowledge and real-time tunneling data, balancing the continuity and timeliness of predictions; when (in the far field region) When the value approaches 0.1, the prior information of the borehole is only used as a background reference. The main reliance is on the real-time parameters of the shield to perceive geological changes and ensure a rapid response to geological anomalies in the far borehole area.

[0022] 5. The anisotropic inverse distance weighting method adopted in this invention does not require the introduction of stationarity assumptions or complex statistical distribution fitting. Combined with a dynamic sample screening mechanism, it reduces the amount of invalid computation, enabling rapid calculation of the entire process of single-ring lithology perception. For example, lithology prediction and voxel reconstruction of 960 rings and 23,040 voxels for the entire tunnel can be completed efficiently in 1 minute. The generated shield tunnel face can directly provide a reference for construction decisions without additional data post-processing steps. This not only meets the time requirements for real-time decision-making in shield tunneling but also lowers the threshold for engineering applications. Attached Figure Description

[0023] Figure 1 This is a flowchart of a method for sensing the lithological composition of a shield tunnel face provided in an embodiment of the present invention; Figure 2 It is a roadmap for multi-scale anisotropic lithology prediction based on borehole local statistics and spacing constraints; Figure 3 This is a schematic diagram showing borehole and lithological composition information; Figure 4 This is a schematic diagram of the real-time lithological sensing results at the working face using a dual-driven fusion method of knowledge and data. Figure 5 This is a schematic diagram of the lithology discrimination confusion matrix within a 5 m radius around the borehole. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0025] This invention provides a method for perceiving the lithological composition of a tunnel boring machine (TBM) face. This method is a dynamic lithological composition perception method that integrates prior knowledge and real-time tunneling data, enabling accurate and real-time perception of lithology throughout the entire TBM tunneling process. It improves the accuracy and stability of lithological prediction in different geological regions (near-drilling area, transition zone, and far-drilling area). The method constitutes a complete simulation of the entire TBM tunneling process from the initial ring to the final ring. Through seven core steps—multi-source information preprocessing, scale modeling, interpolation calculation, probability field construction, real-time perception, dynamic fusion, and voxel reconstruction—it achieves deep integration of prior geological knowledge and real-time tunneling data.

[0026] Please see Figure 1 The method mainly includes the following steps: Step 1: The heterogeneous geological information is uniformly transformed into a set of indexable sample points in three-dimensional space. Samples within a predetermined distance from the borehole are selected from the sample point set to form a training sample set, and the remaining samples constitute a test sample set.

[0027] Heterogeneous geological information such as boreholes and revealed rings is standardized and transformed into a set of indexable sample points in three-dimensional space. Samples within a predetermined distance from the borehole (i.e., samples with high confidence in lithology prediction at the current time point) are selected from the sample point set to form a training sample set, and the remaining samples constitute the test sample set.

[0028] Each discrete lithological volume element of a borehole or revealed ring is represented as a correspondence between location and label, as shown in the following mathematical expression: (1) In the formula, Indicates the first Three-dimensional spatial coordinates of a geological sample point; The lithology category number corresponding to this sample point; This indicates the total number of lithological categories within the study area; This represents the total number of usable samples after preprocessing.

[0029] During the excavation of each ring of the tunnel boring machine, the Euclidean distance between the current excavation ring position and the known borehole is calculated in real time, and a confidence region threshold is predefined. Then, based on the comparison between the Euclidean distance and the confidence threshold, samples with high confidence are selected, and samples with an Euclidean distance greater than the confidence threshold are high-confidence samples.

[0030] After digging to the... During the loop, the selection rules for high-confidence samples are as follows: (2) In the formula, For the first An index set of high-confidence samples in the ring tunneling scenario; For the first The known sample (drill hole or revealed ring) and the first The distance between the rings.

[0031] Only when historical tunneling rings or borehole samples fall within the confidence domain threshold are their corresponding geological labels recognized as high-confidence ground truth and included in the training sample set; regional data outside the confidence domain are used only as test samples. These high-confidence samples serve as both the core input for constructing spatial prior knowledge and the source of training samples for the shield tunneling parameter-driven model.

[0032] Step 2: Map the vertical thickness fluctuation scale of the lithology to an anisotropic scale factor in spatial interpolation, and then calculate the weight of each sample in the training sample set through the anisotropic scale factor.

[0033] Specifically, targeting the first within the study area Lithology (or stratigraphic unit), defining the first The scales of vertical thickness corresponding to different lithologies (abbreviated as Scales of Fluctuation) This is used to quantitatively describe the vertical heterogeneity of formations, and its expression is: (3) In the formula, , The maximum and minimum layer thicknesses of lithological types observed in all borehole samples; This represents the vertical thickness variation scale of the lithology. The magnitude of the value directly reflects the degree of vertical heterogeneity of the strata: The larger the value, the more drastic the variation in the thickness of the lithological layer and the weaker its spatial continuity; The smaller the value, the more stable the stratum thickness and the stronger the spatial continuity.

[0034] The above This is further mapped to an anisotropic scale factor in spatial interpolation, used to measure the stratigraphic correlation length in different spatial directions. The mapping formula is as follows: (4) In the formula, , and The first Lithology in , , Anisotropy scaling factor in three coordinate directions; , The equivalent length scale of the study area in the horizontal direction; For vertical adjustment scale (usually) ); This refers to the maximum depth revealed by the borehole or the characteristic scale of the study area.

[0035] Through the above mapping, the greater the fluctuation in formation thickness ( The larger the value, the smaller the corresponding anisotropic scale factor, enabling the spatial interpolation weight to automatically reflect the uncertainty of the stratigraphic structure.

[0036] based on Constrained Inverse Distance Weighting (IDW): Based on the obtained anisotropic scaling factor, the anisotropic distance is calculated, and the weight of each sample in the training sample set is calculated using the inverse distance weighting function.

[0037] Specifically, considering the bedding characteristics of sedimentary strata, the rate of lithological variation in the vertical direction is much higher than that in the horizontal direction. Therefore, based on the anisotropic scale factor obtained in step two, the anisotropic distance is first calculated, and the corresponding formula is: (5) In the formula, For the first The known high-confidence sample points and the... The inner ring Anisotropic distance between the locations to be predicted; and These are represented as anisotropy scaling factors in the horizontal and vertical directions, respectively; For the first The three-dimensional spatial location of a known high-confidence sample point; For the first The inner ring The three-dimensional spatial location of the location to be predicted.

[0038] Based on this, the inverse distance weighting function (IDW weight) is calculated. For each known high-confidence sample point, its weight is defined as: (6) In the formula, For the first The nth known high-confidence sample point pair The weights of the positions to be predicted; For the first The known high-confidence sample points and the... The inner ring Anisotropic distance between the locations to be predicted; The distance decay exponent, To prevent singular positive numbers at zero distance.

[0039] This step achieves an interpolation effect of "local samples dominating and far-field samples weakening" by introducing anisotropic distance and inverse distance weighting functions, without the need to introduce stationarity or statistical distribution assumptions, and is suitable for real-time update calculations during the tunneling process.

[0040] Step 3: Perform a sliding window statistical transformation on the original tunnel boring machine parameters to extract past data. The mean, standard deviation, and final state values ​​of the loop parameters are used to construct an input feature vector. Based on the obtained input feature vector and the prior probability field, a machine learning nonlinear model is trained to obtain a real-time lithology sensing model. The prior probability field is calculated from the weights of the samples. The location to be predicted Belongs to the Composed of the prior probabilities of lithological types.

[0041] Among them, the weights of the lithology indicator function and the sample are used to calculate the first... The inner ring One location to be predicted Belongs to the The prior probabilities of lithology for each class are used to form a prior probability field; the lithology indicator function is used to identify the first class. 1 known high-confidence sample point Does it belong to the first Lithology, expressed as: (7) In the formula, For the first 1 known high-confidence sample point Regarding the first Lithological indicator functions; For the first The lithological category number corresponding to each known high-confidence sample point; This represents the total number of lithological categories within the study area.

[0042] Based on the lithology indicator function and the calculated weights, the first The inner ring One location to be predicted Belongs to the Prior probability of lithology The calculation formula is: (8) In the formula, This represents the number of known high-confidence sample points used in the prior probability calculation.

[0043] Based on the above calculations, the prior probability... Constitute a by A constrained three-dimensional spatial lithology prior probability field is constructed, which directly inherits the influence of stratigraphic thickness fluctuations on spatial correlation, fully preserving the core information of prior geological knowledge. The calculation steps of this process are as follows: Figure 2 As shown.

[0044] The input to the real-time lithology sensing model is the input feature vector, and the output is the first... Predicted location within the ring Perception probability of various types of lithology .

[0045] Let the first The original shield tunneling parameter vector for the ring is: (9) In the formula, This refers to the number and types of parameters for the tunnel boring machine. For the first Ring 1 Measured values ​​of shield-like tunneling parameters.

[0046] Constructing input feature vectors based on a sliding window strategy : (10) In the formula, For the first Ring to the first Ring 1 The mean values ​​of shield-like tunneling parameters; For the th interval Standard deviation of shield-type tunneling parameters; For the first Ring 1 Final state values ​​of shield tunneling parameters.

[0047] Using a machine learning (ML) nonlinear model as the base model, The input is the number of elements, and the output is the number of elements. Ring 1 One location to be predicted Real-time sensing probability of various types of lithology The real-time sensing probability It directly reflects the immediate geological perception results under rock-machine interaction.

[0048] At the same time, a model retraining interval is set. After each preset number of tunneling passes, the nonlinear machine learning model is fully fine-tuned using samples from the accumulated training sample set to ensure that the model can adapt to the latest geological environment changes in real time.

[0049] Step four: The prior probability field and the perceived probabilities of the samples in the test sample set are fused with the real-time lithology perception model to obtain the final lithology probability, and then the shield tunnel face is reconstructed based on the final lithology probability.

[0050] Specifically, a dynamic fusion weight based on geological determinism is constructed. The dynamic fusion weight Distance from the current location to be predicted to the nearest borehole There is a negative correlation (i.e., the closer the distance, the higher the credibility of prior geological knowledge). Fusion weights Satisfies the value range constraint: (11) in, Follow It increases while monotonically decreasing.

[0051] Based on distance Different fusion strategies are adopted for different ranges to adapt to scenarios with different geological information credibility: when (Near field area) Approaching 0.8, the prior probability obtained by strong trust space interpolation is... By increasing the weighting of the perception probability, noise interference caused by shield tunneling parameter fluctuations can be effectively suppressed, ensuring the stability of lithology prediction in the near-borehole area; when (Transition zone) Follow The increase in probability decreases linearly, gradually increasing the perceived probability. The weighting ratio is used to achieve a smooth transition between prior geological knowledge and real-time tunneling data, balancing the continuity and timeliness of predictions; when (in the far field region) When the value approaches 0.1, the prior information of the borehole is only used as a background reference. The main reliance is on the real-time parameters of the shield to perceive geological changes and ensure a rapid response to geological anomalies in the far borehole area.

[0052] The final lithology probability fusion formula is: (12) In the formula, For the first The inner ring One location to be predicted Belongs to the The final probability of lithological characteristics.

[0053] The fused probability vector is normalized to ensure... .

[0054] Based on the final lithology probability reconstruction of the shield tunnel face: at the... After the ring excavation is completed, the fusion probability vector corresponding to the ring is obtained: (13) In the formula, For the first The final lithological probability set for all locations to be predicted within the ring face area; For the first The inner ring One location to be predicted; For the first The total number of locations or voxels to be predicted within the annular facet.

[0055] Finally, the predicted lithological labels of all voxels in the current ring are used to construct the tunnel face lithological sequence. Combined with the tunnel face lithological sequences of all tunneling rings, the stratigraphic voxel reconstruction and shield tunnel face generation of the study area are completed.

[0056] The present invention will be further described in detail below with reference to specific embodiments.

[0057] This embodiment takes the longitudinal two-dimensional model lithology identification of the tunnel face of Hangzhou Metro Line 6 as the core background, and combines the actual geological conditions of the project with the characteristics of shield tunneling to elaborate in detail the specific implementation process, parameter configuration, equipment selection and test verification of the invention "a dynamic perception method of lithology composition integrating prior knowledge and data".

[0058] 1. Implementation Scenarios and Basic Conditions (1) Data source and specifications This embodiment is based on the Hangzhou Metro Line 6 Phase I project. Located in Binjiang District of Hangzhou City, the terrain along the line is relatively flat, with ground elevations generally ranging from 5.16 m to 9.70 m. The shield tunnel section is 2258.61 m long, with a burial depth between 9.5 m and 37.5 m. The tunnel structure adopts a circular cross-section with an outer diameter of 11.36 m and an inner diameter of 10.36 m, with segment ring widths of 2 m and a thickness of 500 mm.

[0059] A total of 60 boreholes were drilled along the route (see borehole columnar diagram). Figure 3 (As shown), maximum depth of borehole revealed. The tunnel face traverses strata encompassing six main lithologies (numbered 6-2, 8-2, 8-3, 10-1, 12-1, and 12-4), including silty clay, sandy silt, gravel, and moderately weathered sandstone. The strata exhibit significant vertical heterogeneity, with some lithological layers showing considerable thickness fluctuations, consistent with the complex geological scenarios this invention is designed for. Lithology and thickness information were recorded every 0.5 m in 60 boreholes, covering key geological sections throughout the tunnel's length. During excavation, a total of 960 sets of lithology tags (corresponding to 960 rings) were revealed, and these tags are consistent with the borehole lithology classification standards. The tunnel excavation employed an earth pressure balance shield machine. Real-time data collection parameters included thrust, torque, cutterhead speed, and penetration depth. The shield machine was sampled once every 20mm of excavation. The average value of the data collected for each ring was used in the calculations, resulting in a total of 960 sets of shield machine data.

[0060] The tunnel's three-dimensional coordinate data was used for spatial positioning of sample points; each ring was vertically discretized into 24 voxel units, generating a total of 960 voxel units for the entire tunnel. 24 = 23040 voxels (voxel size 0.5 m) 0.5 m 0.5 m).

[0061] (2) Specific implementation steps Step 1: Spatialization of multi-source geological information and construction of high-confidence samples The lithological data from 60 boreholes (including complete layer thickness information within a depth of 75 m) and 960 sets of revealed ring data were uniformly converted into a three-dimensional sample point set. lithological tags Based on the dense distribution characteristics of 60 boreholes and the engineering geological survey report, the confidence region threshold was determined. Tunneling to the... During looping, the Euclidean distance between the current loop position coordinates and the known sample loop position coordinates is calculated in real time, and high-confidence samples are selected. Only samples within the confidence domain are included in the training sample library, while the rest are used as samples to be predicted. For example, when the tunnel reaches the 500th ring, 63 samples within a 5m radius are selected (including relevant data from the three surrounding boreholes) for subsequent prior knowledge construction and model training.

[0062] Step 2, Introduction Stratigraphic-scale modeling Calculation of vertical thickness fluctuation scale of lithology: Based on the complete layer thickness records of 60 boreholes, the maximum layer thickness of 6 types of lithology was statistically analyzed. With minimum layer thickness (m), through calculate The values ​​quantitatively describe the vertical heterogeneity of each lithology, and the results are as follows: Table 1 Vertical orientation of each lithology value

[0063] Define the equivalent length scale in the horizontal direction Vertical adjustment scale (satisfy ), maximum depth By The mapping is to anisotropic scale factors, so that the interpolation weights can automatically reflect the uncertainty of the stratigraphic structure.

[0064] Step 3, based on Constrained anisotropic inverse distance weighted (IDW) Regarding the first Predicted location within the ring The combined results , , By calculating the anisotropic distance between high-confidence sample points and the location to be predicted. This fully considers the characteristic that the vertical lithological variation rate of sedimentary strata is higher than that in the horizontal direction. A distance attenuation index is set. To prevent singularly small positive numbers at zero distance By calculating the weight of each sample point, it achieves the interpolation effect of "local samples dominating and far-field samples weakening", without requiring additional assumptions, and is suitable for real-time computing needs.

[0065] Step 4: Construction of the prior lithological probability field based on IDW For the first Lithology setting indicator function , mark 1 known high-confidence sample point Whether it belongs to this type of lithology. This is determined by formula (8) combined with the sample weights. With indicator functions Calculate the first The inner ring One location to be predicted Prior probability of belonging to various types of lithology , construct by Constrained three-dimensional spatial lithology probability field It fully preserves the core information of prior geological knowledge. For example, the location of the 300th ring to be predicted. The entire sample calculation process for the study area is as follows: Figure 2 As shown.

[0066] Step 5: Real-time lithology sensing model driven by shield tunneling parameters Set Time Backtracking Window The mean, standard deviation, and final state values ​​of the shield tunneling parameters from the past 5 rings are extracted to construct an input feature vector with a dimension of 3×4=12. To capture the temporal effects of rock-machine interaction, a random forest algorithm was employed, with parameters such as 100 decision trees and a maximum tree depth of 15. Initial training (100 iterations) was completed using high-confidence samples as the training set. A retraining interval of 10 cycles was set, and the model was fully fine-tuned every 10 cycles using newly added high-confidence samples to ensure adaptation to changes in the geological environment. The feature vectors were then... Input model, output the first The inner ring One location to be predicted Real-time sensing probability of various types of lithology For example, the 500th ring. .

[0067] Step 6: Dynamic probabilistic fusion based on spatial credibility Get the The inner ring One location to be predicted Distance to the nearest borehole fusion weight Calculated according to the linear decay rule, the closer the distance, the greater the weight, and the more trust is placed in prior geological knowledge. For example, the 500th ring. (Transition zone), calculated as follows .

[0068] Will , and Complete the weighted fusion and normalize the results to ensure... For example, the final fusion probability of ring 500. .

[0069] Step 7: Probabilistic Voxel Reconstruction and Profile Generation Fusion probability vector for each ring The lithology corresponding to the highest probability for each voxel is taken as the discrete lithology label. Combining the lithology sequence of 960 rings with 23040 voxel data, a two-dimensional voxel model of the tunnel face is generated for the entire tunnel mileage. The profile is labeled with lithology type, layer thickness, predicted probability, and borehole location. The technical route for all the above steps is as follows: Figure 1 As shown, the final real-time lithological assessment results for the tunnel face area are as follows: Figure 4 As shown.

[0070] (3) Implementation effect verification The model's predictive performance is measured using global accuracy ( ) and accuracy in unknown areas ( Two metrics are used. Global accuracy is the overall degree of agreement between the model's prediction of the tunnel face and the actual tunnel face throughout the entire real-time simulation phase. Global accuracy includes both the "known region" near the borehole and the "unknown region" far from the borehole. Unknown region accuracy only evaluates the prediction performance in areas far from the borehole. In this embodiment, rings less than or equal to 5 m from the borehole are considered "training set / known region," and rings greater than 5 m from the borehole are considered "prediction set / unknown region." The formulas for calculating global accuracy and unknown region accuracy are as follows: (14) (15) Calculations showed a global accuracy rate of 82.61% and an accuracy rate of 82.90% for unknown areas; the lithology discrimination confusion matrix within a 5m radius around the borehole was as follows. Figure 5 As shown in Table 2, the accuracy rates for the six lithologies are 95.77%, 70.98%, 52.89%, 50.52%, 85.47%, and 87.81%, respectively.

[0071] Table 2 shows the discrimination accuracy of the method proposed in this invention on different lithologies.

[0072] The present invention also provides a sensing system for the lithological composition of a shield tunnel face. The system includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to perform the sensing method for the lithological composition of a shield tunnel face as described above.

[0073] The present invention also provides a computer-readable storage medium storing machine-executable instructions, which, when invoked and executed by a processor, cause the processor to implement the method for sensing the lithological composition of the tunnel face as described above.

[0074] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for sensing the lithological composition of a shield tunnel face, characterized in that, The steps are as follows: (1) The heterogeneous geological information is uniformly transformed into a set of indexable sample points in three-dimensional space. Samples within a predetermined distance from the borehole are selected from the sample point set to form a training sample set, and the remaining samples constitute a test sample set. (2) The vertical thickness fluctuation scale of the lithology is mapped to the anisotropic scale factor in spatial interpolation, and then the weight of each sample in the training sample set is calculated by the anisotropic scale factor. (3) Perform sliding window statistical transformation on the original tunnel boring machine parameters to extract past data. The mean, standard deviation, and final state values ​​of the loop parameters are used to construct an input feature vector. Based on the obtained input feature vector and the prior probability field, a machine learning nonlinear model is trained to obtain a real-time lithology sensing model. The prior probability field is calculated from the weights of the samples. The inner ring One location to be predicted Belongs to the Composed of the prior probabilities of lithological types; (4) The prior probability field and the samples in the test sample set are fused with the perceived probabilities perceived by the real-time lithology perception model to obtain the final lithology probability, and then the shield tunnel face is reconstructed based on the final lithology probability.

2. The method for sensing the lithological composition of a shield tunnel face as described in claim 1, characterized in that: The anisotropic distance is calculated based on the obtained anisotropic scale factor, using the following formula: In the formula, For the first The known high-confidence sample points and the... The inner ring Anisotropic distance between the locations to be predicted; and These are represented as anisotropy scaling factors in the horizontal and vertical directions, respectively; For the first The three-dimensional spatial location of a known high-confidence sample point; For the first The inner ring The three-dimensional spatial location of the location to be predicted.

3. The method for sensing the lithological composition of a shield tunnel face as described in claim 2, characterized in that: The weight of each sample in the training sample set is calculated using the inverse distance weighting function, which is: In the formula, For the first The nth known high-confidence sample point pair The weights of the positions to be predicted; For the first The known high-confidence sample points and the... The inner ring Anisotropic distance between the locations to be predicted; The distance decay exponent, To prevent singular positive numbers at zero distance.

4. The method for sensing the lithological composition of a shield tunnel face as described in claim 1, characterized in that: The weighted calculation of the first sample based on the lithology indicator function and the sample The inner ring One location to be predicted Belongs to the The prior probabilities of lithology for each class are used to form a prior probability field; the lithology indicator function is used to identify the first class. Known high-confidence sample points Does it belong to the first Lithology, expressed as: In the formula, For the first Known high-confidence sample points Regarding the first Lithological indicator functions; For the first The lithological category number corresponding to each known high-confidence sample point; This represents the total number of lithological categories within the study area.

5. The method for sensing the lithological composition of a shield tunnel face as described in claim 4, characterized in that: No. The inner ring One location to be predicted Belongs to the Prior probability of lithology The calculation formula is: In the formula, N is the number of known high-confidence sample points involved in the prior probability calculation.

6. The method for sensing the lithological composition of a shield tunnel face as described in any one of claims 1-5, characterized in that: Constructing dynamic fusion weights based on geological determinism The dynamic fusion weight is related to the weight within the i-th ring. Distance from the predicted location to the nearest borehole It shows a negative correlation.

7. The method for sensing the lithological composition of a shield tunnel face as described in claim 6, characterized in that: Based on distance Different fusion strategies are adopted for different ranges to adapt to scenarios with different geological information credibility: when hour, Approaching 0.8; when hour, Follow The increase of decreases linearly; when hour, It approaches 0.

1.

8. The method for sensing the lithological composition of a shield tunnel face as described in any one of claims 1-5, characterized in that: When training a nonlinear machine learning model, a retraining interval is set. After each preset number of tunneling passes, the machine learning linear model is fully fine-tuned using samples from the accumulated training sample set to ensure that the model can adapt to the latest geological environment changes in real time.

9. A sensing system for the lithological composition of a shield tunnel face, characterized in that: The system includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it performs the method for sensing the lithological composition of the shield tunnel face as described in any one of claims 1-8.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores machine-executable instructions, which, when invoked and executed by a processor, cause the processor to implement the method for sensing the lithological composition of the shield tunnel face as described in any one of claims 1-8.

Images