A method for targeted treatment of subterranean cavities in water-rich formations

Abstract

The application discloses a method for targeted treatment of underground cavities in water-rich strata, and relates to the field of targeted treatment of underground cavities, which comprises the following steps: constructing an underground cavity identification model by using an improved YOLOv8 to identify underground cavities in water-rich strata; constructing a three-dimensional geological model, designing a drilling trajectory from the ground or an existing structure to the cavity, and constructing a targeted drilling hole to form a treatment channel; arranging curtain grouting holes around the cavity and injecting rapid-setting double-liquid slurry to form a three-dimensional water stop curtain; guiding and draining underground water in the three-dimensional water stop curtain through part of the targeted drilling holes and dredging; pumping self-compacting filling materials into the cavity through the remaining targeted drilling holes for main backfilling, and reinforcing the cavity fissures by using sleeve valve pipe grouting technology. The filling and reinforcement effects are verified, and a monitoring system is arranged to monitor the stability of the strata. Through the technical path combining artificial intelligence identification, three-dimensional accurate positioning and active three-dimensional water stop, safe and efficient targeted treatment of underground cavities in water-rich strata is realized.

Description

Technical Field

[0001] This invention relates to the field of targeted treatment technology for underground cavities, and in particular to a method for targeted treatment of underground cavities in water-rich strata. Background Technology

[0002] Underground cavities in water-rich strata pose a significant geological hazard during the construction and operation of infrastructure projects such as tunnels, subways, and municipal roads. These cavities may be formed by karst development, soil erosion, mining subsidence, or historical human activities, and often contain high-pressure groundwater, posing a serious threat to project safety. Once exposed or leaking or collapsing during construction or operation, they can easily trigger catastrophic accidents such as sudden water inrush, mudslides, surface subsidence, or even complete collapse.

[0003] Current technologies for treating underground cavities in water-rich strata generally suffer from the following problems: 1. Inaccurate identification: Relying primarily on geophysical exploration methods such as high-resolution seismic reflection and ground-penetrating radar, combined with a limited number of boreholes for verification, the results depend on manual interpretation, making it difficult to accurately determine the true shape, size, and location of the cavities, and thus impossible to form a reliable three-dimensional model. 2. Poor treatment effect: Especially in water-rich strata, conventional grouting materials are easily washed away by water flow, and the borehole layout lacks specificity, resulting in numerous treatment blind spots, unstable effects, and high construction risks. 3. Disjointed process: From detection and design to construction, information transmission between stages is inefficient, relying on experience-based decision-making and lacking systematic intelligent control, making precise targeted treatment impossible.

[0004] Therefore, a targeted treatment method for underground cavities in water-rich strata is provided to solve the above problems. Summary of the Invention

[0005] To address the aforementioned challenges, this invention provides a targeted treatment method for underground cavities in water-rich strata. By combining artificial intelligence identification, three-dimensional precise positioning, and active three-dimensional water sealing, this method achieves safe and efficient targeted treatment of underground cavities in water-rich strata.

[0006] To achieve the above objectives, the present invention provides a targeted treatment method for underground cavities in water-rich strata, comprising the following steps:

[0007] S1: Obtain geological survey data;

[0008] S2: Based on the underground cavity identification model, geological exploration data is analyzed to identify the spatial location, geometric shape, and scale of underground cavities in water-rich strata; the underground cavity identification model is constructed based on the improved YOLOv8 algorithm;

[0009] S3: Construct a three-dimensional geological model containing cavities. Based on the spatial location and shape of the cavities identified in S2, design a drilling trajectory from the surface or existing structure to the cavities, and construct targeted boreholes along the trajectory to form treatment channels.

[0010] S4: Centered on the identified void, curtain grouting holes are arranged in a ring around it and quick-setting double-liquid grout is injected to form a closed three-dimensional water-stop curtain.

[0011] S5: Through partial targeted drilling, groundwater within the three-dimensional water-stop curtain is drained and some silt or loose filling material in the cavity is removed;

[0012] S6: Through the remaining targeted boreholes, self-compacting filling material is pumped into the cavity for main backfilling, and sleeve valve grouting technology is used to supplement and reinforce the cracks in the surrounding rock and soil.

[0013] S7: Construct inspection holes in the treatment area to verify the filling and reinforcement effects, and deploy a monitoring system to monitor the stability of the formation.

[0014] Preferably, the geological exploration data acquired in S1 includes one or more of the following: high-resolution seismic reflection wave data, ground-penetrating radar data, and cross-hole CT data.

[0015] Preferably, the underground cavity identification model specifically includes:

[0016] A 3D deformable convolutional layer with a kernel size of 3×3×3 was inserted into the 4th layer of the original YOLOv8 backbone to adaptively capture the irregular geometry of underground cavities.

[0017] A branch for detecting small-scale holes is added to the detection head. The kernel size of this branch is 3×3×3, which is used to improve the ability to identify small-scale holes.

[0018] A physical consistency term is introduced into the loss function to ensure that the cavity depth predicted by the model is physically and logically consistent with the wave velocity information in the geological exploration data.

[0019] Preferably, the 3D deformable convolutional layer specifically includes:

[0020] Parallel configuration of the main convolutional path and offset prediction branch;

[0021] The offset prediction branch consists of two cascaded 3D standard convolutional layers. The kernel size of the second convolutional layer is 3×3×3, and its output channel number is 81. It is used to generate an offset field that has the same spatial dimension as the input feature map and contains 27 sampling points in 3D space with scalar offsets.

[0022] The main convolutional path uses a 3×3×3 deformable convolutional kernel. The kernel predicts the offset field of the branch output based on the offset, samples the input feature map, and calculates the feature value of the sampled point through bilinear interpolation.

[0023] The preferred method for constructing and calculating the physical consistency term is as follows:

[0024] Based on the wave velocity-depth relationship of seismic waves propagating in different media, a physical mapping function d=f(v) between cavity depth d and seismic wave velocity v is established.

[0025] The physical consistency term L_phy is composed of the difference between the cavity depth d_pred predicted by the model and the theoretical depth d_phy=f(v_obs) calculated by the measured wave velocity v_obs through the physical mapping function;

[0026] The physical consistency term L_phy and the model's object detection loss function L_det both participate in model training, and the total loss function is... , where λ is a hyperparameter used to balance the weights of the two items.

[0027] Preferably, the design of the precise borehole trajectory in S3 includes simulating the borehole path in a three-dimensional geological model, calculating the trajectory length, assessing the spatial relationship with underground obstacles, and optimizing the borehole incident angle.

[0028] Preferably, the S4 medium-speed setting type two-component grout is a cement-water glass two-component grout, and its setting time is adjusted between 30 seconds and 3 minutes according to the formation water flow and pressure.

[0029] The construction of the three-dimensional water-stop curtain adopts a segmented, intermittent grouting process, and the grouting pressure is dynamically adjusted according to real-time monitoring data.

[0030] Preferably, the self-compacting filling material in S6 is self-compacting concrete;

[0031] The surrounding rock and soil fissures of the cavity are reinforced by: embedding a sleeve valve pipe in the targeted borehole, and repeatedly, quantitatively and controllably grouting the fissures at different depths and locations through the one-way valve on the sleeve valve pipe. The grouting material is cement-based grout.

[0032] Preferably, in step S7, a verification hole is drilled, and core sampling, optical imaging, and / or permeation tests are performed inside the hole to quantitatively obtain parameters of the integrity, bonding degree, and impermeability of the filler. The obtained parameters are then compared with preset thresholds to determine whether the filling and reinforcement effects are qualified.

[0033] Preferably, the monitoring system deployed in S7 includes stress sensors embedded in the filling material, hydraulic gauges deployed inside and outside the curtain, and settlement monitoring points set on the ground surface. The monitoring system transmits data in real time and performs intelligent analysis and early warning through an Internet of Things platform.

[0034] Therefore, the above-mentioned targeted treatment method for underground cavities in water-rich strata, as described in this invention, has the following beneficial effects:

[0035] (1) The present invention uses an underground cavity identification model based on the improved YOLOv8 algorithm to automatically and accurately identify the spatial location, geometric shape and scale of cavities in water-rich strata, effectively overcoming the errors and limitations of traditional manual interpretation, and providing a reliable basis for subsequent targeted treatment.

[0036] (2) Based on the identification results, the present invention constructs a three-dimensional geological model and designs a precise drilling trajectory, so that the targeted drilling can reach the core area of ​​the cavity, significantly improving the accuracy of treatment and avoiding ineffective drilling and waste of resources caused by blind construction.

[0037] (3) The present invention constructs a three-dimensional water-stop curtain around the cavity, which isolates the interference of external water flow on the treatment process, greatly reduces the risk of water inrush, and provides safe and stable construction conditions for subsequent operations inside the cavity.

[0038] (4) The present invention utilizes targeted drilling to implement drainage and dredging, self-compacting material backfilling and crack grouting reinforcement in stages, forming a multi-level treatment system from the inside of the cavity to the surrounding rock and soil, comprehensively improving the compactness and long-term stability of the treated area.

[0039] (5) By deploying construction inspection holes and a long-term monitoring system, this invention enables real-time verification and continuous tracking of the treatment effect, ensuring that the quality of the treatment project is controllable and safe and reliable.

[0040] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0041] Figure 1 This is a flowchart illustrating a targeted treatment method for underground cavities in water-rich strata according to the present invention.

[0042] Figure 2 This is a diagram of underground cavities in a water-rich stratum, as shown in an embodiment of the present invention.

[0043] Figure 3 This is a diagram of the 3D deformable convolutional layer structure in an embodiment of the present invention;

[0044] Figure 4 This is a branch structure diagram of small-scale cavity detection in an embodiment of the present invention;

[0045] Figure 5 This is a schematic diagram of the sleeve valve tube structure in an embodiment of the present invention;

[0046] Figure 6 The diagram below shows the stress sensor installation process in an embodiment of the present invention. (a) is a positioning diagram during the stress sensor installation process, (b) is a slotting diagram during the stress sensor installation process, (c) is an installation diagram during the stress sensor installation process, (d) is a lead wire diagram during the stress sensor installation process, (e) is a backfilling diagram during the stress sensor installation process, and (f) is a compaction diagram during the stress sensor installation process. Detailed Implementation

[0047] The following detailed description of embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0048] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning as understood by one of ordinary skill in the art to which this invention pertains.

[0049] The terms "comprising" or "including" as used in this invention mean that the element preceding the term encompasses the element listed after the term, and do not exclude the possibility of encompassing other elements. Terms such as "inner," "outer," "upper," and "lower" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. When the absolute position of the described object changes, the relative positional relationship may also change accordingly. In this invention, unless otherwise explicitly specified and limited, the term "attached" and similar terms should be interpreted broadly. For example, it can refer to a fixed connection, a detachable connection, or an integral part; it can refer to a direct connection or an indirect connection through an intermediate medium; it can refer to the internal communication of two elements or the interaction relationship between two elements. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0050] Example

[0051] A targeted treatment method for underground cavities in water-rich strata, such as Figure 1 As shown, it includes the following steps:

[0052] S1: Acquire geological exploration data; geological exploration data includes one or more of the following: high-resolution seismic reflection wave data, ground-penetrating radar data, and cross-hole CT data, as well as underground cavities such as... Figure 2 As shown.

[0053] S2: Based on the underground cavity identification model, geological exploration data is analyzed to identify the spatial location, geometric shape, and scale of underground cavities in water-rich strata; the underground cavity identification model is constructed based on the improved YOLOv8 algorithm;

[0054] The underground cavity identification model specifically includes:

[0055] Insert a 3D deformable convolutional layer into the 4th layer of the original YOLOv8 backbone, such as... Figure 3 As shown, its convolution kernel size is 3×3×3, which is used to adaptively capture the irregular geometry of underground cavities; the offset learning rate of the deformable convolution layer is set to 0.01.

[0056] 3D deformable convolutional layers specifically include:

[0057] Parallel configuration of the main convolutional path and offset prediction branch;

[0058] The offset prediction branch consists of two cascaded 3D standard convolutional layers. The kernel size of the second convolutional layer is 3×3×3, and its output channel number is 81. It is used to generate an offset field that has the same spatial dimension as the input feature map and contains 27 sampling points in 3D space with scalar offsets.

[0059] The main convolutional path uses a 3×3×3 deformable convolutional kernel. The kernel predicts the offset field of the branch output based on the offset, samples the input feature map, and calculates the feature value of the sampled point through bilinear interpolation.

[0060] To address the issues of missed detection and inaccurate localization caused by the small size and weak features of cavities, a branch for small-scale cavity detection is added to the detection head, such as... Figure 4 As shown, this branch has a 3×3×3 kernel size to improve the ability to identify small-scale holes. Specifically, this branch is not connected to the end of the network, but directly to the shallow output of the feature pyramid. On this high-resolution feature map, this branch uses a standard convolutional layer with a 3×3×3 kernel size for processing. This convolutional layer is followed by a lightweight prediction head, which independently outputs the detection results for small-scale holes (including bounding box coordinates, confidence score, and category). This achieves a "divide and conquer" strategy for the detection task, with a dedicated branch responsible for small-scale targets, avoiding the performance loss caused by different scale targets competing for the same set of network weights. Together with the original detection head, it forms a multi-scale collaborative detection system, enabling the model to have full-scale hole recognition capabilities.

[0061] A physical consistency term is introduced into the loss function to ensure that the cavity depth predicted by the model is physically and logically consistent with the wave velocity information in the geological exploration data.

[0062] The physical consistency term is constructed and calculated as follows:

[0063] Based on the wave velocity-depth relationship of seismic waves propagating in different media, a physical mapping function d=f(v) between cavity depth d and seismic wave velocity v is established.

[0064] The physical consistency term L_phy is composed of the difference between the cavity depth d_pred predicted by the model and the theoretical depth d_phy=f(v_obs) calculated by the measured wave velocity v_obs through the physical mapping function;

[0065] The physical consistency term L_phy and the model's object detection loss function L_det both participate in model training, and the total loss function is... , where λ is a hyperparameter used to balance the weights of the two items.

[0066] The underground cavity identification model was trained using original geological survey data from multiple actual engineering projects, specifically including:

[0067] The original data is standardized and preprocessed to form a unified and standardized training sample. This mainly includes: (1) Format unification: Converting the original data from different sources and in different formats (such as SEG-Y, DZT, etc.) into a general tensor format that the model can process (such as NumPy arrays or PyTorch Tensors). (2) Normalization: Normalizing the gray value or amplitude value of each data volume to unify its distribution range to the range of [0, 1] or [-1,1], so as to accelerate model convergence and improve training stability. (3) Size regularization: Trimming or padding all input data to a uniform size (e.g., 128x128x128 voxels) to meet the model input requirements. For two-dimensional data such as GPR profiles, pseudo-three-dimensional data volumes are constructed by stacking or interpolation.

[0068] Using preliminary drilling verification, in-hole television, and expert interpretation results as standards, subsurface cavities were accurately labeled in the preprocessed 3D data volume. The labeling results were converted into label files matching the model's output format, such as YOLO or COCO, to provide ground truth values ​​for subsequent loss calculations. To improve the model's generalization ability and robustness and prevent overfitting, strict data augmentation strategies were applied to the training set, including geometric transformations and numerical perturbations. The dataset was divided into training and validation sets in a 7:3 ratio to ensure fairness in the evaluation.

[0069] S3: Construct a three-dimensional geological model containing cavities. Based on the spatial location and shape of the cavities identified in S2, design a drilling trajectory from the surface or existing structure to the cavities, and construct targeted boreholes along the trajectory to form treatment channels. The design of precise drilling trajectories includes simulating the drilling path in the three-dimensional geological model, calculating the trajectory length, evaluating the spatial relationship with underground obstacles, and optimizing the borehole incident angle.

[0070] Specifically, key parameters such as the spatial location, geometric shape, and scale of the underground cavity identified by the S2 cavity identification model are used as core constraints and integrated into 3D geological modeling software (such as GOCAD, GMS, FLAC3D, etc.). Based on this, original exploration data from boreholes and geophysical surveys are integrated to construct a high-precision, integrated 3D geological model that includes the target cavity, surrounding strata, groundwater information, and existing underground structures. This model serves as the digital foundation for all subsequent design and analysis. In the 3D geological model, intelligent planning of the borehole trajectory is performed according to the following design requirements: calculating the optimal 3D path from the surface or existing structures (such as tunnel sidewalls, construction shafts) to the target cavity, striving for the shortest path to save on engineering workload and costs; the trajectory must accurately avoid all known underground obstacles, such as municipal pipelines, existing pile foundations, pressurized water sacs, and large fracture zones. Based on the main functions of the borehole in subsequent treatment (such as drainage, dredging, and grouting), its incident angle and terminal position are optimized. For example, the endpoint of the grouting hole should be located at the center of the bottom of the cavity, with an angle conducive to grout diffusion; the drainage hole should facilitate water collection and discharge. The trajectory design must comply with the technological limitations of the drilling equipment, such as the maximum drilling depth and the allowable trajectory curvature. A measurement-while-drilling (MWD) and directional drilling system is used for construction. During drilling, the bottom-of-hole measuring tool transmits real-time attitude data such as the borehole's inclination, azimuth, and depth back to the ground control system. The real-time collected borehole trajectory data is compared with the designed theoretical trajectory. If a deviation occurs, the system immediately calculates a correction plan and dynamically controls the drilling direction by adjusting the drill bit tool face angle, forming a real-time closed-loop control of measurement-comparison-correction to ensure the borehole strictly follows the designed trajectory. After drilling is completed, in-hole television and other equipment can be used to finally confirm the channel quality and the condition inside the cavity. At this point, one or more efficient and precise targeted treatment channels for subsequent drainage, dredging, grouting, and reinforcement are formally formed.

[0071] S4: Centered on the identified void, curtain grouting holes are arranged in a ring around it and quick-setting double-liquid grout is injected to form a closed three-dimensional water-stop curtain.

[0072] The quick-setting double-liquid grout is a cement-water glass double-liquid grout, and its setting time is controlled between 30 seconds and 3 minutes according to the formation water flow and pressure. The construction of the three-dimensional water-stop curtain adopts a segmented and intermittent grouting process, and the grouting pressure is dynamically adjusted according to real-time monitoring data.

[0073] Specifically, centered on the three-dimensional morphology of the cavity identified by S2, one or more rows of curtain grouting holes are scientifically arranged in a ring or spherical shell pattern around its perimeter. The hole layout design considers not only planar position but also strictly controls its vertical depth and spacing to ensure that the final curtain completely encloses the target cavity in three-dimensional space, forming a complete waterproof enclosure. Cement-water glass dual-liquid grout is used as the curtain material, and a segmented grouting method is employed for each curtain grouting hole, either from top to bottom or bottom to top. Grouting plugs are used to divide the borehole into several independent sections, which are then grouted segment by segment. This ensures that strata at different depths receive sufficient and uniform grouting, preventing grout from flowing only along the weakest channels. During the grouting process, an intermittent operation mode of grouting-pause-re-grouting is adopted. During the pause, the injected grout is allowed to fully solidify, increase in strength, and consolidate the strata, thus providing higher injection pressure and better constraint conditions for the next round of grouting. Real-time pressure and flow monitoring is conducted throughout the entire grouting process. The grouting pressure is not a fixed value, but is dynamically adjusted based on the formation's grout absorption, previous grouting records, and real-time monitoring data (such as surface uplift and grout leakage from adjacent boreholes). Following the principle of low-pressure slow seepage and high-pressure fracturing, a lower pressure is used initially to ensure seepage, and the pressure is appropriately increased within a controllable range in the later stages to ensure the density and continuity of the grout curtain.

[0074] S5: Through partial targeted drilling, groundwater within the three-dimensional water-stop curtain is drained and some silt or loose filling material in the cavity is removed;

[0075] Specifically, some of the targeted boreholes formed in step S3 are selected as dedicated drainage channels. A deep well pump or submersible pump is lowered to the bottom of the cavity through these selected boreholes. The pumps are then started to continuously discharge the groundwater within the sealed space of the three-dimensional water-stop curtain, actively lowering the water level inside the cavity. During the drainage process, the outflow rate and water quality are monitored in real time until the outflow changes from turbid to clear, the flow rate significantly decreases, and it stabilizes, indicating that the free water inside the cavity has been largely drained.

[0076] After drainage is completed, this section of the targeted borehole will serve as a dredging channel. High-pressure jet pipes will be lowered into the boreholes, using high-pressure water jets to flush, agitate, and liquefy the silt and soft plastic filling material within the cavities, transforming it into a suspended slurry. Simultaneously, a mud pump or air lift pump will be used to establish a negative pressure suction system through the boreholes, continuously pumping the liquefied slurry mixture to the surface. Following a cyclical operation of "flushing-liquefaction-suction," cleaning will be carried out in sections and layers until the pumped material is primarily water with no obvious particulate matter carried out, indicating that most of the loose material within the cavities has been removed.

[0077] After the dredging operation is completed, the thoroughness of the dredging can be indirectly assessed by measuring the volume of water and mud discharged and comparing it with the volume of cavities identified by S2.

[0078] S6: Through the remaining targeted boreholes, self-compacting filling material is pumped into the cavity for main backfilling, and sleeve valve grouting technology is used to supplement and reinforce the cracks in the surrounding rock and soil.

[0079] The self-compacting filler material is self-compacting concrete;

[0080] The surrounding rock and soil fissures of the cavity are reinforced by: embedding a sleeve valve pipe in the targeted borehole, and repeatedly, quantitatively and controllably grouting the fissures at different depths and locations through the one-way valve on the sleeve valve pipe. The grouting material is cement-based grout.

[0081] Specifically, self-compacting concrete is used as the main backfill material. Its mix design is optimized to exhibit high fluidity, anti-segregation properties, and excellent self-filling characteristics, allowing it to fill the entire irregular cavity without vibration under its own weight. Through the remaining targeted boreholes not used for dredging (pre-drilled in step S3), the mixed self-compacting concrete is continuously pumped from the surface mixing plant into the drained and emptied cavities using a high-pressure concrete pump. The pumping process continues until a full volume of slurry flows smoothly from the pre-drilled vents or observation holes, indicating that the cavity has been completely filled and compacted. This process forms the core structure that resists ground deformation.

[0082] After the main backfill body is formed, in order to seal the cracks in the surrounding soil and rock mass and form an integrated anti-seepage reinforcement ring, sleeve valve pipe grouting technology is used for supplementary reinforcement.

[0083] Specifically, the sleeve valve system is installed by inserting sleeve valves into pre-designed targeted boreholes. A sleeve valve is a specialized grouting pipe, such as... Figure 5As shown, the pipe wall is equipped with one-way grouting valves at certain intervals, and is protected by a double-layer rubber sleeve. A sleeve material is injected between the sleeve valve pipe and the borehole wall to firmly fix it inside the hole. The function of the sleeve material is to seal the annular gap, forcing subsequent grouting pressure to only open the designated sleeve valve and preventing it from leaking out of the pipe. Segmented, controllable grouting operation: Using a double-plug grouting core pipe, grouting is carried out in segments from the bottom of the sleeve valve pipe upwards. By sealing both ends of a specific segment, independent and precise grouting of the corresponding depth of the formation fractures in that segment is achieved. For each grouting segment, the grouting pressure and grouting volume are strictly controlled. Adjustments are made in real time according to geological conditions to ensure that the grout can fully penetrate into the fractures while avoiding excessive pressure that could cause the formation to split. Repeated grouting capability: Thanks to the one-way valve design on the sleeve valve pipe, if the initial grouting effect of a segment does not meet expectations, the segment can be repeatedly reinforced with secondary or multiple reinforcement grouting, ensuring the reliability and flexibility of the reinforcement.

[0084] The grouting material used is cement-based grout. This material has the advantages of wide availability, high strength, good durability, and low cost. Its water-cement ratio and admixtures can be dynamically adjusted according to the formation water absorption rate and reinforcement requirements.

[0085] S7: Construct inspection holes in the treatment area to verify the filling and reinforcement effects, and deploy a monitoring system to monitor the stability of the formation.

[0086] By drilling verification holes, core sampling, optical imaging, and / or permeability tests are performed inside the holes to quantitatively obtain parameters of the filler integrity, bonding degree, and impermeability. The obtained parameters are then compared with preset thresholds to determine whether the filling and reinforcement effects are qualified.

[0087] The deployed monitoring system includes stress sensors embedded in the filling material, hydraulic gauges installed inside and outside the curtain wall, and settlement monitoring points set on the ground surface. The system transmits data in real time and provides intelligent analysis and early warning through an Internet of Things (IoT) platform. For example... Figure 6 The diagram illustrates the specific process for burying the stress sensor.

[0088] Therefore, this invention adopts the above-mentioned targeted treatment method for underground cavities in water-rich strata. Through a systematic technical path of precise identification by artificial intelligence, intelligent decision-making by three-dimensional model, active prevention and control of three-dimensional water stopping, precise targeted drilling and step-by-step treatment and reinforcement, a complete treatment closed loop is formed. This realizes the technical leap from traditional experience-based treatment to modern precision treatment, significantly improving the safety, reliability and economy of treating underground cavities in water-rich strata, and providing effective technical protection for the safe development of underground space.

[0089] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for targeted treatment of a subterranean void in a water- enriched formation, comprising: The method comprises the following steps: S1: obtaining geological survey data; the geological survey data comprises one or more of high-resolution seismic reflection wave data, geological radar data, and cross-hole CT data; S2: analyzing the geological survey data based on an underground cavity identification model to identify the spatial position, geometric shape, and scale of the underground cavity in the water-rich stratum; the underground cavity identification model is constructed based on an improved YOLOv8 algorithm; The underground cavity identification model specifically comprises: A 3D deformable convolution layer is inserted at the 4th layer of the original backbone of YOLOv8, and the convolution kernel size thereof is 3×3×3, which is used to adaptively capture the irregular geometric shape of the underground cavity; A branch for small-scale cavity detection is added to the detection head, and the convolution kernel size of the branch is 3×3×3, which is used to improve the recognition ability of small-scale cavities; A physical consistency term is introduced in the loss function, which is used to constrain the consistency of the predicted cavity depth and the wave velocity information in the geological survey data in the physical logic; S3: constructing a three-dimensional geological model containing the cavity, designing a drilling trajectory from the ground or an existing structure to the cavity based on the spatial position and shape of the cavity identified in S2, and constructing a target drilling along the trajectory to form a treatment channel; the design of the drilling trajectory comprises simulating the drilling path in the three-dimensional geological model, calculating the trajectory length, evaluating the spatial relationship with underground obstacles, and optimizing the drilling incidence angle; S4: taking the identified cavity as the center, arranging curtain grouting holes in the periphery of the cavity and injecting rapid-setting double-liquid slurry to form a closed three-dimensional waterproof curtain; S5: through part of the target drilling, guiding and draining the underground water in the three-dimensional waterproof curtain and removing part of the silt or loose filling in the cavity; S6: through the remaining target drilling, pumping self-compacting filling material into the cavity for main backfilling, and using sleeve valve pipe grouting technology to supplement and reinforce the fractures of the rock-soil mass around the cavity; S7: constructing inspection holes in the treatment area to verify the filling and reinforcement effect, and arranging a monitoring system to monitor the stability of the stratum.

2. The method of claim 1, wherein the method further comprises: The 3D deformable convolution layer specifically comprises: A main convolution path and an offset prediction branch arranged in parallel; The offset prediction branch is composed of two cascaded 3D standard convolution layers, wherein the convolution kernel size of the second convolution layer is 3×3×3, and the number of output channels thereof is 81, which is used to generate an offset field consistent with the spatial dimension of the input feature map and containing the scalar offset of 27 sampling points in the three-dimensional space; The main convolution path adopts a 3×3×3 deformable convolution kernel, which samples the input feature map according to the offset field output by the offset prediction branch and calculates the feature values of the sampling points through bilinear interpolation.

3. The method of claim 2, wherein the method further comprises: The construction and calculation method of the physical consistency term is as follows: Based on the wave velocity-depth relationship of seismic waves propagating in different media, a physical mapping function d=f(v) is established between the cavity depth d and the seismic wave velocity v; The physical consistency term L_phy is composed of the difference between the predicted cavity depth d_pred and the theoretical depth d_phy=f(v_obs) calculated by the physical mapping function from the measured wave velocity v_obs. The physical consistency term L phy and the target detection loss function L det of the model jointly participate in the model training, and the total loss function is L total = L det + λL phy L phy, wherein λ is a hyperparameter for balancing the weights of the two terms.

4. The method of claim 3, wherein: The S4 medium setting type double slurry is a cement-sodium silicate double slurry, and the setting time is regulated at 30 seconds to 3 minutes according to the formation water flow and pressure; The construction of the three-dimensional water stop curtain adopts a segmented and intermittent grouting process, and the grouting pressure is dynamically adjusted according to real-time monitoring data.

5. The method of claim 1, wherein: The S6 self-compacting filling material is self-compacting concrete; The fractures around the cavity are supplemented and reinforced, specifically, sleeve valve pipes are buried in the targeted drill holes, and through the one-way valves on the sleeve valve pipes, the fractures at different depths and different positions are repeatedly, quantitatively and controllably grouted, and the grouting material is a cement-based slurry.

6. The method of claim 1, wherein: In step S7, a verification hole is drilled, coring, optical imaging and / or permeation test are performed in the hole, the integrity, cementation and impermeability parameters of the filling body are quantitatively obtained, and the obtained parameters are compared with the preset threshold value to determine whether the filling and reinforcement effect is qualified.

7. The method of claim 1, wherein: The monitoring system arranged in S7 includes stress sensors buried in the filling body, water pressure gauges arranged on the inside and outside of the curtain, and settlement monitoring points set on the ground surface, and the monitoring system transmits and intelligently analyzes and warns the data in real time through an Internet of Things platform.

Images