Tunnel dynamic water grouting plugging optimization design method and system based on multi-source information fusion
Through multi-source information fusion and reinforcement learning, real-time optimization decisions for tunnel grouting are realized, and the problem that grouting design is difficult to meet safety and economic requirements is solved, and the intelligence and safety of construction are improved.
Patent Information
- Application Number
- CN202410221558.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-02-28
- Publication Date
- 2024-11-22
- Estimated Expiration
- 2044-02-28
AI Technical Summary
In tunnel projects, grouting sealing design is difficult to meet safety and economic requirements at the same time, and the existing methods cannot realize real-time monitoring and optimization of water influx and grouting parameters, resulting in a lack of intelligence and real-time grouting construction.
Using multi-source information fusion method, through binocular stereoscopic visual recognition and 3D scene reconstruction, combined with geophysical exploration information, a mapping data set of grouting parameters and water inflow volume is established, and real-time intelligent decision-making is made using reinforcement learning to optimize grouting and blocking strategies. .
Real-time optimization decisions for grouting sealing are realized, the intelligence and safety of grouting construction are improved, the water influx is reduced, the dependence on construction personnel's experience is reduced, and the construction efficiency is improved.
Smart Images

Figure CN118296929B_ABST
Abstract
Description
Technical Field
[0001] The present disclosure relates to the technical field of tunnel engineering, and in particular to a tunnel dynamic water grouting plugging optimization design method and system based on multi-source information fusion. Background Art
[0002] The statements in this section merely provide background information related to the present disclosure and do not necessarily constitute prior art.
[0003] The geological conditions of the construction of extra-long, deep-buried, water-rich tunnels are complex, and water gushing is often encountered. Grouting is a key measure to solve the problem. However, it is difficult for the grouting construction design to meet the requirements of safety and economy at the same time, which brings challenges to the dynamic design of grouting plugging. The water gushing at the tunnel face in front will change with time and tunnel construction. It is an important real-time parameter that needs to be measured in the project, which will provide a key basis for predicting the amount of water gushing ahead and grouting construction. However, due to the current limitations of the project organization form, the water gushing detection lacks real-time perception of the water gushing, and when encountering large sudden and gushing water ahead, the layout of sensors is inconvenient. In addition, there are similar problems in the measurement of drilling angles, lengths and grouting positions at the tunnel construction site. With the development of computer vision technology and the popularization of image acquisition equipment, it has become a hot issue to quickly grasp the on-site conditions through digital images of the tunnel face.
[0004] In addition, the control of grouting parameters such as the amount of grouting on site in the tunnel still depends on the experience of the construction personnel to a certain extent. The proposed intelligent grouting decision-making methods mainly focus on expert system reasoning, water inrush risk assessment, grouting alarm information decision-making, and direct simulation of the grouting water blocking process through numerical simulation algorithms. It usually involves the calculation of some intermediate physical quantities, which is time-consuming and greatly affected by the selection of parameters and boundary conditions. From the perspective of available information on site and the problems to be solved, combined with the powerful ability and generalization of reinforcement learning to deal with dynamic sequence optimization problems, the existing methods are subject to certain limitations in the dynamic tunnel excavation and grouting scenes, in which real-time information is obtained on site to guide the selection of grouting construction parameters and grouting sequence, and cannot effectively achieve unmanned and less-manned intelligent real-time monitoring of water inrush and some grouting parameters during construction, as well as real-time optimization of grouting blocking strategies. Summary of the invention
[0005] In order to solve the above problems, the present invention proposes a tunnel dynamic water grouting plugging optimization design method and system with multi-source information fusion, which is based on binocular stereo vision recognition, 3D scene reconstruction, and multi-source tunnel information assisted by other geophysical information, and combines on-site grouting parameters to establish a mapping data set of grouting parameters, water inflow and grouting plugging effect, and finally realizes real-time optimization decision of grouting plugging.
[0006] According to some embodiments, the present disclosure adopts the following technical solutions:
[0007] The optimization design method of tunnel dynamic water grouting plugging based on multi-source information fusion includes:
[0008] Obtain optical imaging and thermal imaging video of the tunnel face and segment the video into time-series images;
[0009] Use time-series images to achieve stereoscopic visual recognition of the tunnel, reconstruct the 3D scene of the tunnel, and construct a depth image;
[0010] Based on the depth image, the ViT network is used to identify the water gushing behavior in the tunnel face, collect the geophysical information in the tunnel and the on-site grouting parameters, and obtain the time series data of grouting plugging.
[0011] According to the grouting plugging time series data and water gushing volume, a mapping database between them and the on-site grouting plugging effect is established. Policy-based reinforcement learning is used to make real-time intelligent decisions on the grouting plugging action. The reward is set as the reduction rate of water gushing volume. The Monte Carlo policy gradient method is used to find the optimal solution, randomly initialize the policy parameters, generate the trajectory on the strategy, estimate the return value and update the policy parameter value in each time step, and output the grouting parameters and the choice of grouting timing under the situation of water gushing in front of the tunnel.
[0012] According to some embodiments, the present disclosure adopts the following technical solutions:
[0013] The tunnel dynamic water grouting plugging optimization design system based on multi-source information fusion includes:
[0014] Data acquisition-storage-transmission module, used to acquire optical imaging and thermal imaging video of the tunnel face and segment the video into time-series images;
[0015] Stereoscopic vision recognition-3D reconstruction module is used to realize stereoscopic vision recognition of tunnels using time-series images, reconstruct 3D scenes of tunnels, and construct depth images; based on the depth images, the ViT network is used to identify the water gushing behavior in the tunnel face, collect geophysical information in the tunnel and on-site grouting parameters, and obtain the time-series data of grouting plugging;
[0016] The real-time grouting optimization module is used to establish a mapping database between the grouting and plugging time series data and the water inflow and the on-site grouting and plugging effects. It uses policy-based reinforcement learning to make real-time intelligent decisions on grouting and plugging actions, sets the reward as the rate of reduction of water inflow, uses the Monte Carlo policy gradient method to find the optimal solution, randomly initializes the policy parameters, generates the trajectory on the strategy, estimates the return value and updates the policy parameter value in each time step, and outputs the grouting parameters and the choice of grouting timing under the situation of water inflow in front of the tunnel.
[0017] According to some embodiments, the present disclosure adopts the following technical solutions:
[0018] A non-transitory computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by a processor, the tunnel dynamic water grouting plugging optimization design method based on multi-source information fusion is implemented.
[0019] According to some embodiments, the present disclosure adopts the following technical solutions:
[0020] An electronic device comprises: a processor, a memory and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory, so that the electronic device executes the tunnel dynamic water grouting plugging optimization design method that realizes the multi-source information fusion.
[0021] Compared with the prior art, the present invention has the following beneficial effects:
[0022] The multi-source information fusion optimized design method for tunnel dynamic water grouting plugging disclosed in the present invention collects the image time series data of the tunnel face site, including the identified water gushing, grouting behavior, and induced polarization auxiliary geophysical exploration information, and can directly establish the feedback and mapping time series relationship between the on-site water gushing and grouting behavior without numerical simulation. At the same time, using the principle of binocular stereo vision, it is proposed to use two conventional-thermal imaging cameras to jointly participate in the 3D modeling of the tunnel grouting behavior, and it is proposed to use a 4DOF robotic arm to complete the automatic correction and adjustment of the posture of the two cameras. Then, through the method of reinforcement learning, the selection of grouting parameters and grouting timing under the specific water gushing situation ahead is completed, such as the arrangement of grouting holes, the selection of grouting timing, the determination of grouting amount, the selection of drilling angle and drilling depth, etc.
[0023] The use of the comprehensive grouting and plugging intelligent decision-making system disclosed in the present invention will facilitate the intelligent and safe construction of tunnel grouting. BRIEF DESCRIPTION OF THE DRAWINGS
[0024] The accompanying drawings constituting a part of the present disclosure are used to provide a further understanding of the present disclosure. The illustrative embodiments of the present disclosure and their descriptions are used to explain the present disclosure and do not constitute an improper limitation on the present disclosure.
[0025] Figure 1 It is a method flow architecture diagram of an embodiment of the present disclosure;
[0026] Figure 2 It is a module architecture diagram of an embodiment of the present disclosure;
[0027] Figure 3 This is a diagram showing the arrangement of binocular depth cameras according to an embodiment of the present disclosure;
[0028] Figure 4A camera 4DOF robot and layout diagram of an embodiment of the present disclosure;
[0029] Figure 5 This is a schematic diagram of binocular stereoscopic vision perception principle of an embodiment of the present disclosure;
[0030] Figure 6 A diagram showing a baseline length calculation principle according to an embodiment of the present disclosure;
[0031] Figure 7 It is a structural diagram of the ViT model of an embodiment of the present disclosure;
[0032] Figure 8 An example of a ViT model classification object training diagram for an embodiment of the present disclosure;
[0033] Fig. 9 It is a fitting diagram of a single water flow axis of an embodiment of the present disclosure;
[0034] Fig.10 This is a reinforcement learning framework diagram of an embodiment of the present disclosure. DETAILED DESCRIPTION
[0035] The present disclosure is further described below in conjunction with the accompanying drawings and embodiments.
[0036] It should be noted that the following detailed descriptions are all illustrative and are intended to provide further explanation of the present disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present disclosure belongs.
[0037] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present disclosure. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.
[0038] Example 1
[0039] In one embodiment of the present disclosure, a tunnel dynamic water grouting plugging optimization design method based on multi-source information fusion is provided, comprising:
[0040] Step 1: Obtain optical imaging and thermal imaging video of the tunnel face, and segment the video into time-series images;
[0041] Step 2: Use time-series images to realize stereoscopic visual recognition of the tunnel, reconstruct the 3D scene of the tunnel, and construct a depth image;
[0042] Step 3: Based on the depth image, the ViT network is used to identify the water gushing behavior in the tunnel face, collect the geophysical information in the tunnel and the on-site grouting parameters, and obtain the time series data of grouting plugging;
[0043] Step 4: According to the grouting and plugging time series data and water gushing volume, a mapping database is established between them and the on-site grouting and plugging effect. Policy-based reinforcement learning is used to make real-time intelligent decisions on the grouting and plugging actions. The reward is set as the reduction rate of water gushing volume. The Monte Carlo policy gradient method is used to find the optimal solution. The policy parameters are randomly initialized and the trajectory on the strategy is generated. The return value is estimated and the policy parameter value is updated in each time step. The grouting parameters and the choice of grouting timing are output when there is water gushing in front of the tunnel.
[0044] As an embodiment, the specific implementation process of the tunnel dynamic water grouting plugging optimization design method of the multi-source information fusion disclosed in the present invention is as follows:
[0045] Step 1: Obtain optical imaging and thermal imaging video of the tunnel face, and segment the video into time-series images;
[0046] Specifically, first, two conventional thermal imaging cameras are placed at different positions in front of the palm face, such as Figure 3 As shown. Two of the cameras have both conventional optical and thermal imaging functions, and the cameras have their own light sources. Placing two cameras can achieve the following functions:
[0047] (1) Realize passive ranging sensing. Based on the principle of binocular stereo vision, two cameras are used to obtain two sets of images of the same tunnel water inrush scene.
[0048] (2) It can prevent water flows from blocking each other. When water gushes from the face, multiple water flows are usually generated. The blocking between water flows will affect the recognition accuracy.
[0049] (3) Depth information can be collected without the device itself emitting energy. Unlike active depth sensing methods such as time-of-flight (TOF), this method uses a stereo matching algorithm to match corresponding pixels and completes depth estimation based on the parallax principle. It does not require a ranging sensor and can greatly reduce costs.
[0050] To correct the relative position and angle of the two cameras, the specific structure and layout are as follows: Figure 4 As shown. The bracket under the two cameras uses a 4DOF mechanical arm (4DOF: 4 degrees of freedom), which can realize the setting of the camera in any position. At the same time, in order to measure the baseline angle, a circle of LED infrared identification lights are installed around the top of the bracket.
[0051] Secondly, an induced polarization geophysical exploration device is placed on the tunnel face to detect the water-bearing body in front of the tunnel face. The relationship between the water volume V of the water-bearing body and the half-time difference envelope area A (where the half-time difference envelope area A is the value obtained by induced polarization detection) is used to perform auxiliary geophysical detection:
[0052] V=aA
[0053] Among them, the process of the geophysical inversion module to detect the water content of the water body in the front tunnel face is as follows:
[0054] A mapping database of the half-decay difference envelope area and water volume is established through experimental data to obtain the above-mentioned values. After obtaining the half-decay difference of induced polarization detection, a prediction is made and the values are compared and corrected with the actual identified water inflow.
[0055] Secondly, pressure sensors, flow sensors and other monitoring components are placed on the grouting equipment to measure the grouting rate and pressure in real time.
[0056] Finally, a large-capacity storage device is used to store the videos, baseline length and other parameters taken by the two cameras. The stored videos are divided into thermal images and optical images according to time; baseline length (initially, the two cameras automatically find and align the infrared identification light ring on the bracket through the rotation of the 4DOF mechanical arm, and the baseline length between the two cameras is calculated by the length of the infrared identification light ring in the camera); other parameters include: focal length, relative position of camera-bracket (obtained from the sensor of the 4DOF mechanical arm), etc., and the difference envelope area of the half-decay of the excited polarization. Data transmission adopts a combination of cable and wireless transmission to transmit to the ground data center, which can prevent data loss and monitor data in real time.
[0057] Step 2: Use the time-series images to realize stereoscopic visual recognition of the tunnel, reconstruct the 3D scene of the tunnel, and construct a depth image;
[0058] First, the tunnel is reconstructed in 3D. The left and right cameras C L , C R The principle of achieving binocular stereoscopic vision perception, that is, 3D reconstruction, is as follows Figure 5 The baseline length of the camera center line is b, and point P is on the left camera C. L The imaging point is P L , in the right camera C R The imaging point is P R Point P is the line connecting the camera's projection center and the imaging point C. L P L , C R P R The intersection of the line segment x L and x R Camera CL , C R The imaging point P L , P R location.
[0059] According to the disparity x of point P L -x R The triangular approximate relationship between , baseline length, focal length and depth value Z is:
[0060]
[0061] The baseline length b is obtained by aligning the camera to another camera through the 4DOF robot arm. The calculation formula is:
[0062]
[0063] Where f is the focal length, D marker is the diameter of the infrared identification lamp ring, both of which are known and fixed, l L and l R are the imaging lengths of the infrared marker light rings captured by the left and right cameras. The calculation principle of the baseline length is as follows: Figure 6 As shown, where P U and P D They are the upper and lower points of the infrared marker light ring in the camera’s image edge, C L and C R Represents the left and right cameras respectively.
[0064] Furthermore, the specific steps of tunnel 3D reconstruction are:
[0065] (1) Distortion correction: Remove the radial and tangential distortion of the lens and convert the original image into a non-distorted image.
[0066] (2) Correct and adjust the angle and distance between cameras. The purpose is to align and refine the image. This process is automatically completed by the robot joint driver and sensor at the joint of the 4DOF robot fixed on the bracket. The robot joint sensor can estimate the pose parameters of the camera at the end of the robot, which is solved by the DH parameter model of the serial robot forward kinematics:
[0067]
[0068] where a i is the length of the robot arm connecting rod, α i is the torsion angle of the robot arm link, d i is the displacement of the robot link, θ i is the joint angle of the robot arm. The camera pose at the end of the robot arm is:
[0069]
[0070] Among them A i is the i-th DH transformation matrix.
[0071] (3) Match the points of the left and right cameras to obtain the disparity map of the left and right cameras.
[0072] (4) Triangulate the disparity map, construct the depth map, and achieve 3D reconstruction.
[0073] Then, the sudden water inrush behavior of the tunnel face is identified and the water inrush volume is calculated in real time. After obtaining the depth map of the tunnel 3D reconstruction, a group of continuous frames of video are accumulated to obtain a 3D scene map in the average sense, and then the water inrush is identified by classification, and then the connected domain of each water flow is obtained. The pixel points on the connected domain are used for edge detection to achieve the segmentation of the tunnel face contour and water flow. After obtaining the outer contour of each water flow, combined with its morphology, the flow rate of each water flow is further calculated, and the real-time water inrush volume is calculated.
[0074] Specifically, the following steps are implemented:
[0075] (1) Accumulate a set of consecutive frames and average the tunnel 3D scene.
[0076] (2) ViT (Vision Transformer) is used to classify and identify the averaged tunnel 3D scene. When there is enough data for pre-training, ViT is better than CNN (Convolutional Neural Networks) and can also break through the limitation of Transformer's lack of inductive bias. The objects recognized and classified by the present invention are fixed objects, which can be pre-trained in large quantities. Therefore, the ViT model is used. The ViT model structure is as follows: Figure 7 As shown in the figure. Unlike traditional convolutional neural networks (CNNs), ViT directly treats images as serialized inputs and uses a self-attention mechanism to process pixel relationships in images. ViT divides each input frame into multiple tiles, adds a linear mapping layer, maps the dimensions of these tiles to the specified embedding dimensions, and then flattens them before inputting them into a multi-layer Transformer Encoder for feature extraction.
[0077] The Transformer encoder contains a self-attention mechanism and a feedforward neural network layer, which can capture the contextual dependencies of different positions in the image, and finally classify the image output by the Transformer encoder. Specifically, the input patch (Embedded Patches) is divided into three parts by the Norm layer, and then simultaneously input into the Multi-Head Attention (the multi-head attention mechanism is an important component of the Transformer Encoder feature extraction), and then added to the residual input, and then output through Norm and MLP (multi-layer perceptron) to complete the image classification.
[0078] The objects of image classification are mainly water flow and machinery such as on-site grouting equipment, so the ViT model can also identify the location of grouting holes, drilling angles, and drilling length information. A large amount of pre-training in the early stage can achieve better migration effects, such as Figure 8 shown.
[0079] (3) Through fitting technology, the trajectory axis of a single water flow is calculated based on the distribution of the actual water flow cross-sectional area, such as Fig. 9 shown.
[0080] First, the centroid and area of the cross section are calculated through the distribution of the water flow cross section, and then the parabolic axis equation of each water flow is obtained according to the fitting calculation formula of the parabolic axis:
[0081]
[0082] in, Obtained by the least square fitting formula of the centroid set of the water flow calculation section:
[0083]
[0084]
[0085]
[0086] where x i ,y i are respectively the x and y coordinates of the centroid of the water flow cross section in the plane of the water flow axis, and the y-axis is vertically upward.
[0087] (4) By fitting the water flow trajectory axis equation, based on the principle of projectile motion, the initial flow velocity and flow size of a single water flow i are calculated. The derivation process is omitted. The water flow velocity v at any point i The calculation formula for (x) is:
[0088]
[0089] Where g is the value of gravity acceleration, which is obtained through field measurement. Next, for the entire water flow i, take the maximum value of the observed flow:
[0090] Q i =MAX{Q i (x)}
[0091] Q i (x) = S i (x)v i (x)
[0092] Where S i (x) is the cross-sectional area at any point of the i-th stream, obtained from the 3D model, Q i (x) is the calculated flow rate at any point on the axis, Q i is the estimated flow rate of the entire stream.
[0093] (5) Finally, the sum of the flow rates of all water flows is the total water inflow at the face.
[0094] Q=∑Q i
[0095] Collect geophysical information in the tunnel and grouting volume parameters of on-site grouting, and form time series data with the grouting plugging data (grouting hole location, drilling angle and drilling length information) obtained through identification;
[0096] Step 3: According to the grouting and plugging time series data and water inflow, a mapping database is established between them and the on-site grouting and plugging effects. Policy-based reinforcement learning is used to make real-time intelligent decisions on the grouting and plugging actions. The reward is set as the reduction rate of water inflow. The Monte Carlo policy gradient method is used to find the optimal solution. The policy parameters are randomly initialized and the trajectory on the strategy is generated. The return value is estimated and the policy parameter value is updated in each time step. The grouting parameters and the choice of grouting timing under the situation of water inflow in front of the tunnel are output.
[0097] Specifically, after obtaining the time series data of water inflow, grouting hole layout, grouting amount, and grouting drilling angle, the time series data (including unstructured and structured data) is first stored in the database. Then, a policy-based reinforcement learning system (RL) is used to continuously optimize and improve the grouting construction plan through feedback loops. Fig.10 shown.
[0098] In reinforcement learning, by learning in interaction, the agent can optimize the rewards obtained in a complex and uncertain environment, and guide better actions by sensing the reaction of the state of the environment to the action to obtain the maximum benefit. Corresponding to the real-time intelligent decision-making of grouting and plugging, the reward to be obtained is the rate of reduction of water inflow, that is, the efficiency of plugging water. The specific actions include: the layout of the grouting hole location, the grouting amount, the drilling direction and length, etc., as well as their respective time series.
[0099] The present disclosure adopts the Monte Carlo policy gradient method to find the optimal policy (also known as the REINFORCE algorithm, which is a Policy Gradient algorithm based on the Monte Carlo update method). It is more efficient in continuous action space, directly models and optimizes the policy, and can implement a randomized policy. The steps of the Monte Carlo policy gradient method are: randomly initialize the policy parameter θ t , generating sampling trajectories on the strategy:
[0100] π(θ)={s 1 ,a 1 ,r 1 ,s 2 ,a 2 ,r 2 ,...,s T ,a T ,r T}
[0101] where s n ,a n ,r n The state value, action value and reward value are then estimated at each time step and the policy parameter value θ is updated. t+1 .
[0102] Specifically for the real-time optimization of grouting parameters, first let the strategy parameter be θ i , (i=1,2,...,k), the evaluation index is J(θ), where the strategy parameter θ i Including grouting rate, total number of holes, hole location, hole length, hole angle and grouting material, etc., some of these parameters are obtained by on-site sensors, such as grouting rate and grouting material, and the rest are all obtained by the visual recognition system described in the present invention. The evaluation index J(θ) is the reduction of water inflow / grouting volume in a certain period of time, which can be a cycle of tunnel excavation-drilling-grouting. Then the stochastic gradient ascent method is used to update the strategy parameters, that is:
[0103]
[0104] Among them, α is the learning rate, is the gradient of the evaluation index, and its value is:
[0105]
[0106] Among them, s,a~ π(·) represents the expected value. Finally, the iteration formula of the REINFORCE algorithm is:
[0107]
[0108] in, R k is the reward value, γ is the decay factor, A t |S t For state S t Action A t ,π(A t |S t ,θ t ) is the strategy under given parameters.
[0109] Example 2
[0110] In one embodiment of the present disclosure, a tunnel dynamic water grouting plugging optimization design system based on multi-source information fusion is provided, comprising:
[0111] A data acquisition module is used to acquire optical imaging and thermal imaging video of the tunnel face and segment the video into time-series images;
[0112] The reconstruction module is used to realize stereoscopic visual recognition of the tunnel using time-series images, reconstruct the 3D scene of the tunnel, and construct a depth image;
[0113] The recognition and classification module is used to identify the water gushing behavior in the tunnel face based on the depth image and the ViT network, collect the geophysical information in the tunnel and the on-site grouting parameters, and obtain the time series data of grouting plugging;
[0114] The optimization decision module is used to establish a mapping database between the grouting and plugging time series data and the water inflow and the on-site grouting and plugging effects, and use policy-based reinforcement learning to make real-time intelligent decisions on the grouting and plugging actions. The reward is set as the reduction rate of the water inflow, and the Monte Carlo policy gradient method is used to find the optimal solution. The policy parameters are randomly initialized, and the trajectory on the strategy is generated. The return value is estimated and the policy parameter value is updated in each time step, and the grouting parameters and the choice of grouting timing are output under the condition of water inflow in front of the tunnel.
[0115] Example 3
[0116] In one embodiment of the present disclosure, a non-transitory computer-readable storage medium is provided, wherein the non-transitory computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by a processor, the tunnel dynamic water grouting plugging optimization design method based on multi-source information fusion is implemented.
[0117] Example 4
[0118] In one embodiment of the present disclosure, an electronic device is provided, including: a processor, a memory, and a computer program; wherein the processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory, so that the electronic device executes the tunnel dynamic water grouting plugging optimization design method that realizes the multi-source information fusion.
[0119] The present disclosure is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present disclosure. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0120] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.
[0121] Although the above describes the specific implementation methods of the present disclosure in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present disclosure. Technical personnel in the relevant field should understand that on the basis of the technical solution of the present disclosure, various modifications or variations that can be made by those skilled in the art without creative work are still within the scope of protection of the present disclosure.
Claims
1. A tunnel dynamic water grouting plugging optimization design method based on multi-source information fusion, characterized in that: include: Obtain optical imaging and thermal imaging video of the tunnel face and segment the video into time-series images; Use time-series images to achieve stereoscopic visual recognition of the tunnel, reconstruct the 3D scene of the tunnel, and construct a depth image; Based on the depth image, the ViT network is used to identify the water gushing behavior in the tunnel face, collect the geophysical information in the tunnel and the on-site grouting parameters, and obtain the time series data of grouting plugging. The specific contents of using the ViT network to identify water inrush behavior in the tunnel face include: The depth image is input into the ViT network to classify and identify the 3D scene in the tunnel, output the water inrush behavior, and then obtain the connected domain of each water flow. The pixel points on the connected domain are used for edge detection to achieve the segmentation of the tunnel face contour and water flow. After obtaining the outer contour of each water flow, the flow velocity of each water flow is calculated based on its morphology, and the real-time water inrush volume is calculated. Through fitting technology, the trajectory axis of a single water flow is calculated through the distribution of the actual water flow cross-sectional area. Through the trajectory axis equation of a single water flow, based on the principle of projectile motion, the initial flow velocity and flow size of a single water flow are calculated. The sum of the flow sizes of all water flows is the total water inflow of the face. According to the grouting plugging time series data and water gushing volume, a mapping database between them and the on-site grouting plugging effect is established. Policy-based reinforcement learning is used to make real-time intelligent decisions on the grouting plugging action. The reward is set as the reduction rate of water gushing volume. The Monte Carlo policy gradient method is used to find the optimal solution, randomly initialize the policy parameters, generate the trajectory on the policy, estimate the return value and update the policy parameter value in each time step, and output the grouting parameters and the choice of grouting timing under the situation of water gushing in front of the tunnel.
2. The tunnel dynamic water grouting plugging optimization design method based on multi-source information fusion according to claim 1 is characterized in that: Using binocular stereo vision perception, the time-series images are processed and converted into non-distorted images. The angle and distance between the cameras are corrected and adjusted. The points of the left and right cameras are matched. The disparity maps of the left and right cameras are obtained. The disparity maps are triangulated and the depth images are constructed to achieve 3D reconstruction of the tunnel.
3. The tunnel dynamic water grouting plugging optimization design method based on multi-source information fusion as claimed in claim 2 is characterized in that: Correcting and adjusting the angle and distance between cameras includes: estimating the pose parameters of the camera at the end of the robotic arm through the robotic arm joint sensor based on the robotic arm joint driver and sensor at the joint of the 4DOF robotic arm fixed on the bracket, and solving it by the DH parameter model of the forward kinematics of the serial robot.
4. The tunnel dynamic water grouting plugging optimization design method based on multi-source information fusion as claimed in claim 1 is characterized in that: Based on the grouting and plugging time series data and water gushing volume, a mapping database between them and the on-site grouting and plugging effects is established. The time series data includes unstructured and structured data. Then, a strategy-based reinforcement learning system is used to continuously optimize and improve grouting construction decisions through feedback loops.
5. The tunnel dynamic water grouting plugging optimization design method based on multi-source information fusion as claimed in claim 4 is characterized in that: In reinforcement learning, by learning in interaction, the intelligent agent can optimize the rewards obtained in a complex and uncertain environment, and guide better actions by sensing the reaction of the state of the environment to the actions to obtain the maximum benefit. Corresponding to the real-time intelligent decision-making of grouting and plugging, the reward obtained is the rate of reduction of water inflow, that is, the efficiency of plugging water. The specific actions include: the layout of the grouting hole location, the grouting amount, the drilling direction and length, and their respective time series.
6. The tunnel dynamic water grouting plugging optimization design system based on multi-source information fusion is characterized by: include: Data acquisition-storage-transmission module, used to acquire optical imaging and thermal imaging video of the tunnel face and segment the video into time-series images; Stereoscopic vision recognition-3D reconstruction module, which is used to realize stereoscopic vision recognition of tunnels using time-series images, reconstruct 3D scenes of tunnels, and construct depth images; Based on the depth image, the ViT network is used to identify the water gushing behavior in the tunnel face, collect the geophysical information in the tunnel and the on-site grouting parameters, and obtain the time series data of grouting plugging. The specific contents of using the ViT network to identify water inrush behavior in the tunnel face include: The depth image is input into the ViT network to classify and identify the 3D scene in the tunnel, output the water inrush behavior, and then obtain the connected domain of each water flow. The pixel points on the connected domain are used for edge detection to achieve the segmentation of the tunnel face contour and water flow. After obtaining the outer contour of each water flow, the flow velocity of each water flow is calculated based on its morphology, and the real-time water inrush volume is calculated. Through fitting technology, the trajectory axis of a single water flow is calculated through the distribution of the actual water flow cross-sectional area. Through the trajectory axis equation of a single water flow, based on the principle of projectile motion, the initial flow velocity and flow size of a single water flow are calculated. The sum of the flow sizes of all water flows is the total water inflow of the face. The real-time grouting optimization module is used to establish a mapping database between the grouting and plugging time series data and the water inflow and the on-site grouting and plugging effects. It uses policy-based reinforcement learning to make real-time intelligent decisions on grouting and plugging actions, sets the reward as the rate of reduction of water inflow, uses the Monte Carlo policy gradient method to find the optimal solution, randomly initializes the policy parameters, generates the trajectory on the strategy, estimates the return value and updates the policy parameter value in each time step, and outputs the grouting parameters and the choice of grouting timing under the situation of water inflow in front of the tunnel.
7. A non-transitory computer-readable storage medium, characterized in that: The non-transitory computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by the processor, the tunnel dynamic water grouting plugging optimization design method based on multi-source information fusion as described in any one of claims 1 to 5 is implemented.
8. An electronic device, characterized in that: include: A processor, a memory and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory so that the electronic device executes the tunnel dynamic water grouting plugging optimization design method based on multi-source information fusion as described in any one of claims 1-5.
