Targeted positioning grouting efficient plugging and reinforcing method and system based on deep learning

Abstract

The application discloses a deep learning-based targeted positioning grouting efficient plugging and reinforcing method and system, geological data is input into a three-dimensional geological weak point prediction model for prediction, three-dimensional geological weak point distribution is obtained and input into a targeted grouting prediction model for prediction, the best grouting point position and grouting parameters of the current engineering area are obtained; based on the setting threshold of the reconstruction error of the targeted grouting prediction model, when the error is within the setting threshold, the best grouting point position and grouting parameters are used for targeted grouting; when the error exceeds the setting threshold, the geological environment is updated through the three-dimensional geological weak point prediction model, and the weak point distribution of the targeted grouting prediction model is updated again; and based on the grouting parameters, the operation parameters of the grouting equipment are set to realize the target grouting effect. The application periodically determines whether there is an abnormal grouting condition on the basis of the established grouting strategy through error threshold analysis, and can adjust the original predicted grouting best point position and grouting parameters in a timely manner according to the abnormal condition.

Description

Technical Field

[0001] This invention relates to the field of grouting and sealing reinforcement technology, specifically to a method and system for efficient grouting and sealing reinforcement based on deep learning and targeted positioning. Background Technology

[0002] The statements in this section are merely background information relating to this disclosure and do not necessarily constitute prior art.

[0003] With the rapid and vigorous development of the economy, underground engineering construction, including tunnel engineering, has entered a new stage. As the scale of construction projects increases, the depth of penetration deepens, and the geological conditions traversed become more complex, the geological disasters encountered during construction are becoming increasingly sudden and destructive, making effective prevention and control of complex geological disasters increasingly important. Grouting engineering can effectively treat sudden disasters in underground engineering projects. Through the solidification and bonding of grout with the surrounding rock structure, it can effectively block the recharge and runoff of aquifers, modify the water-bearing capacity of aquifers, and reinforce weak rock masses. However, conventional grouting techniques suffer from severe ineffective diffusion of grout under different geological conditions, and are not ideal for reinforcing soil and rock masses or effectively sealing sudden water inrushes in strata with complex dynamic water sealing and fracture connectivity.

[0004] Current targeted construction research on grouting projects mainly focuses on the parameter control of grouting pressure, flow rate, and material ratio. This involves global adjustments to the grouting process, making it difficult to address problems in real time. This leads to grouting blind spots and difficulty in controlling the grout's penetration path and diffusion depth, resulting in poor regional grouting effects. Furthermore, adjustments to grouting schemes do not comprehensively consider real-time issues arising during the grouting process, or timely feedback on grouting equipment operation and grouting effects. Summary of the Invention

[0005] To overcome the shortcomings of the prior art, this invention provides a method and system for efficient sealing and reinforcement through targeted grouting based on deep learning. By periodically determining whether there are any abnormal grouting situations based on the established grouting strategy through error threshold analysis, the method can promptly adjust the original predicted optimal grouting points and grouting parameters according to the abnormal situations. At the same time, by combining the operating status of the grouting equipment parameters, the real-time status of the equipment is compared with the target grouting effect, and the output power and pressure of the grouting equipment are adjusted to ensure that the grout can accurately reach the target area.

[0006] To achieve the above objectives, one or more embodiments of the present invention provide the following technical solutions:

[0007] In a first aspect, the present invention provides a targeted grouting method for efficient sealing and reinforcement based on deep learning, comprising:

[0008] Obtain geological data of the current engineering area and preprocess it to obtain preprocessed geological data;

[0009] The preprocessed geological data is input into a pre-trained three-dimensional geological weakness prediction model to predict the distribution of three-dimensional geological weaknesses.

[0010] The three-dimensional geological weakness distribution is input into a pre-trained targeted grouting prediction model for prediction, to obtain the optimal grouting points and grouting parameters for the current engineering area. Based on a set threshold for the reconstruction error of the targeted grouting prediction model, when the error is within the set threshold, targeted grouting is performed using the optimal grouting points and grouting parameters. When the error exceeds the set threshold, the geological environment is updated through the three-dimensional geological weakness prediction model, and the weakness distribution of the targeted grouting prediction model is updated again.

[0011] Based on the grouting parameters, the operating parameters of the grouting equipment are set. Based on the basic principle of multi-agent deep reinforcement learning, each grouting device is treated as an individual agent. The operating parameters of each agent are optimized through reinforcement learning strategies to achieve the target grouting effect.

[0012] A further technical solution is that the three-dimensional geological weakness prediction model is constructed using a generative adversarial network, which includes a generator and a discriminator. A variational autoencoder is used as the backbone structure of the network generator, and a Transformer is used as the backbone structure of the discriminator.

[0013] A further technical solution is that the data processing process in the three-dimensional geological weakness prediction model is as follows:

[0014] The preprocessed geological data is used as input data and passed through a generator with a variational autoencoder as the backbone to output a three-dimensional geological weakness distribution. At the same time, the three-dimensional geological weakness distribution is correlated with the geological data to obtain geological data carrying parameter information.

[0015] Then, a discriminator with Transformer as its backbone structure compares the distribution of unfavorable geological features in the actual strata with the generated distribution of three-dimensional geological weaknesses, while using mutual information as a loss term.

[0016] The model training is completed when the model generated by the generator is close to the distribution of the real model and the loss term meets the requirements, based on the backpropagation update optimization solution.

[0017] A further technical solution is that the targeted grouting prediction model is constructed based on a convolutional neural network, and outputs different three-dimensional geological weakness distributions through a three-dimensional geological weakness prediction model, including single weaknesses and combinations of multiple weaknesses.

[0018] A further technical solution is that the single weakness is targeted grouting based on the characteristics of different weaknesses, and the combination of multiple weaknesses is predicted based on a pre-trained targeted grouting prediction model. The neural network learns from historical grouting project data to learn the relationship between the distribution of different three-dimensional geological weaknesses and the optimal grouting point and grouting parameters when multiple weaknesses exist.

[0019] The formula is expressed as:

[0020] S(z i ) t =v(x i ,S(z i ) t-1 ) t +b(t)

[0021] Where S represents the different distributions of weaknesses at the current time step, z i This represents the different distributions of weaknesses, where t and t-1 represent time steps, i represents different weakness categories, and x represents the different distributions of weaknesses. i denoted by , v represents the actual grouting parameters corresponding to the weakness, v represents the weakness parameters at the current time step resulting from the combined effect of the grouting parameter selection at the current time step and the weakness parameter situation at the previous time step, and b(t) represents the influence of the surrounding environment on grouting at each time step.

[0022] A further technical solution involves optimizing the parameters of each agent using a reinforcement learning strategy, where the reward function is expressed as:

[0023]

[0024] Where R represents the actual grouting progress corresponding to the state quantity of different data quantization for each action, m represents the total number of equipment, n represents the total number of actions of equipment j, j represents different equipment, k represents a certain action of equipment j, t is the time step, α(t) and β(t) represent the adjustment coefficients at different times, and w j (t) represents the loss of device j at a certain time, λ j (t) represents the reward coefficient for the actual grouting progress corresponding to the different data quantification state quantities of each action of each device at different times, Y. j y represents the state quantity of device j at a certain time during a certain action. k Y represents the state quantity of device j at a certain time step under action k. j (y k ,t) represents the state quantity of each action of each device at different times.

[0025] A further technical solution involves determining the action combination values ​​required for the grouting parameters corresponding to each intelligent agent. The calculation of these combination values ​​is expressed as follows:

[0026]

[0027] Where, x i This represents the actual grouting parameters corresponding to the weakness, j represents different equipment, k represents a certain action of equipment j, n represents the total number of actions of equipment j, and γ represents the total number of actions of equipment j. j,k y represents the weight of a certain action k of device j on the contribution value of parameter j. k Z represents the state quantity of device j at a certain time step under action k. j (y k ) represents the contribution value corresponding to the state variable.

[0028] Secondly, this invention provides a deep learning-based targeted grouting high-efficiency sealing and reinforcement system, comprising:

[0029] The data acquisition module is configured to acquire geological data of the current engineering area and preprocess it to obtain preprocessed geological data.

[0030] The weakness distribution prediction module is configured to input preprocessed geological data into a pre-trained three-dimensional geological weakness prediction model to predict the three-dimensional geological weakness distribution.

[0031] The grouting prediction module is configured to: input the three-dimensional geological weakness distribution into a pre-trained targeted grouting prediction model for prediction, and obtain the optimal grouting point location and grouting parameters for the current engineering area; based on a set threshold for the reconstruction error of the targeted grouting prediction model, when the error is within the set threshold, perform targeted grouting using the optimal grouting point location and grouting parameters; when the error exceeds the set threshold, update the geological environment through the three-dimensional geological weakness prediction model, and then update the weakness distribution of the targeted grouting prediction model again.

[0032] The grouting equipment operation module is configured to: set the operation parameters of the grouting equipment based on the grouting parameters; and, based on the basic principle of multi-agent deep reinforcement learning, treat each grouting equipment as an individual agent and optimize the operation parameters of each agent through reinforcement learning strategies to achieve the target grouting effect.

[0033] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the deep learning-based targeted grouting efficient sealing and reinforcement method described in the first aspect.

[0034] Fourthly, the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the deep learning-based targeted grouting efficient sealing and reinforcement method described in the first aspect.

[0035] The above one or more technical solutions have the following beneficial effects:

[0036] This invention acquires geological data of the engineering area, analyzes the adverse geological attributes (weaknesses) of the construction area, preprocesses the data, establishes a geological database containing data acquired through different monitoring methods, and performs three-dimensional tagging of the data. A three-dimensional geological weakness prediction model is established, using a generative adversarial network to perform localized estimation of existing voids, loose strata, and water-rich areas, providing precise location information for targeted reinforcement through grouting. A targeted grouting prediction model is established, predicting the optimal grouting points and parameters for the area based on the determined distribution of three-dimensional geological weaknesses. A threshold for reconstruction error in the targeted grouting prediction model is set to allow for adjustments to the grouting strategy under abnormal conditions. This fully considers anomalies that may occur during the grouting process and enables timely adjustments to the originally predicted optimal grouting points and parameters based on these anomalies.

[0037] This invention also leverages the fundamental principles of multi-agent deep reinforcement learning to ensure consistency in operating and grouting parameters across multiple devices. By comparing the real-time status of the devices with the target grouting effect, the output power and pressure of the grouting equipment are adjusted to ensure the grout accurately reaches the target area. After each grouting cycle, the grouting effect is evaluated and feedback is provided. By comparing the parameter characteristics of adjacent grouting cycles, the success rate of grouting and the long-term stability of the soil and rock reinforcement are verified.

[0038] This invention establishes a three-dimensional geological weakness prediction model to identify unfavorable geological areas in the strata, constructs a targeted grouting prediction model to obtain the optimal grouting points and parameters, and sets an error threshold based on the targeted grouting prediction model (different from the convergence loss during the original model training). Through error threshold analysis, it periodically determines whether there are any grouting anomalies based on the established grouting strategy and makes adjustments accordingly. At the same time, it compares the real-time status of the grouting equipment with the target grouting effect by combining the parameter status and operation status of the grouting equipment, and adjusts the output power and pressure of the grouting equipment to ensure that the grout can accurately reach the target area. Attached Figure Description

[0039] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0040] Figure 1 This is a flowchart of the sealing and reinforcement method according to an embodiment of the present invention. Detailed Implementation

[0041] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0042] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0043] Where there is no conflict, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0044] Example 1

[0045] like Figure 1 As shown in the figure, this embodiment discloses a high-efficiency sealing and reinforcement method for targeted grouting based on deep learning. The method includes the following steps:

[0046] S1: Obtain geological data of the current engineering area and preprocess it to obtain preprocessed geological data;

[0047] In this embodiment, the propagation characteristics of seismic waves and sound waves in different geological formations are utilized to collect reflection and refraction data, revealing information such as density and porosity within the strata. Ground-penetrating radar is used to image the internal structure of the strata, obtaining cross-sectional views of the geological structure and providing stratigraphic structural information in image form, identifying fractures and cavities within the strata. Borehole monitoring is used to obtain information such as pressure and temperature within the strata, determining the density and structural homogeneity of the strata. Physical tests and elemental information identification are performed on borehole core samples to analyze adverse geological properties of the construction area, providing fundamental data for localized identification.

[0048] Missing values ​​in the acquired geological data were filled using mean imputation, and outlier detection methods were used to identify and remove outliers to ensure data quality. Different types of acquired data were standardized and normalized to adapt them to the model input format, ensuring rapid model convergence and improving prediction accuracy. A geological database containing data acquired through various monitoring methods was established.

[0049] The actual grouting area was divided into three dimensions. The three-dimensional data in the database, i.e., the acquired geological data, was tagged to indicate the stratigraphic properties of each area, such as density, porosity, and elemental properties. Specifically, the geological data was divided into zones according to different scales. Three-dimensional meshes of varying coarseness and structure were created in the different scale areas. The centroid of each three-dimensional mesh point was used as the data storage point. By tagging the three-dimensional data in the database and indicating the stratigraphic properties of each area, such as density, porosity, and elemental properties, the data storage points of each mesh were assigned. All data was aligned to a unified three-dimensional coordinate system, and data noise was reduced through convolutional filtering. GAN was used to complete the data and then the data was tagged.

[0050] S2: Input the preprocessed geological data into the pre-trained three-dimensional geological weakness prediction model to predict the distribution of three-dimensional geological weaknesses;

[0051] In this embodiment, the three-dimensional geological weakness prediction model is constructed using a generative adversarial network, which includes a generator and a discriminator. The variational autoencoder (VAE) is used as the backbone structure of the network generator, and the Transformer is used as the backbone structure of the discriminator.

[0052] The data processing procedure in the three-dimensional geological weakness prediction model is as follows:

[0053] (1) The preprocessed geological data (3D data) is input into the 3D geological weakness prediction model. After passing through the generator with variational autoencoder as the backbone, the 3D geological weakness distribution (such as loose strata, water-rich areas, karst conduits, etc.) is output. At the same time, the 3D geological weakness distribution is associated with the data in the database to obtain geological data carrying parameter information. That is, each generated 3D geological weakness distribution carries corresponding geological parameter information. For example, high porosity is associated with loose strata, while low density and resistivity are associated with voids, fractures and water-rich areas. The parameter information is normalized and data noise is reduced.

[0054] (2) Then, a discriminator with Transformer as its backbone compares the actual distribution of unfavorable geological features in the strata with the generated distribution of three-dimensional geological weaknesses, while using mutual information as a loss term, i.e.:

[0055] L a =I(M;N) (1)

[0056] Here, M and N are the parametric features carried by different three-dimensional geological weaknesses, and I(M;N) represents the mutual information between them. By minimizing the information between different parameters, the accurate determination of the distribution of different three-dimensional geological weaknesses is ensured.

[0057] (3) The model training is completed when the model generated by the generator is close to the distribution of the real model and the loss term meets the requirements.

[0058] Geological weaknesses refer to unfavorable geological areas in strata. Based on the three-dimensional distribution of geological weaknesses, combined with the area, depth and connectivity of the regional distribution, local estimation is performed on the actual existing void channels, loose strata and water-rich areas, providing accurate positioning information for targeted reinforcement of grouting treatment.

[0059] S3: Input the three-dimensional geological weakness distribution into the pre-trained targeted grouting prediction model for prediction to obtain the optimal grouting point and grouting parameters for the current engineering area; based on the set threshold of the reconstruction error of the targeted grouting prediction model, when the error is within the set threshold, targeted grouting is performed using the optimal grouting point and grouting parameters; when the error exceeds the set threshold, the geological environment is updated through the three-dimensional geological weakness prediction model, and the weakness distribution of the targeted grouting prediction model is updated again.

[0060] In this embodiment, the targeted grouting prediction model is constructed based on a convolutional neural network. It outputs different 3D geological weakness distributions through a 3D geological weakness prediction model, including single and multiple weakness combinations, based on historical data from engineering experience and actual engineering grouting effects. The input is 3D geological weakness distribution information. The model performs a convolution operation on this input data to represent it in 3D data space, and the output data includes the optimal grouting point location and grouting parameters.

[0061] Based on the obtained three-dimensional geological weakness distribution information, the optimal grouting points and grouting parameters for the current engineering area are predicted. Specific classifications of weaknesses are considered, including loose strata, fissures, karst conduits, water-rich areas, and high temperatures.

[0062] For single weaknesses, such as loose strata, different grout types are selected to achieve better penetration and form stable grout veins and stone bodies. For water-rich fissures or karst conduits, targeted grouting is carried out considering the fissure opening and extension, as well as the diameter and branching of the karst conduit.

[0063] For cases with multiple weaknesses, predictions are made based on a pre-trained targeted grouting prediction model. The model learns from historical grouting project data through a neural network, learning the relationship between the distribution of different three-dimensional geological weaknesses carrying different geological parameter information and the optimal grouting point and grouting parameters in cases with multiple weaknesses. The model is trained to quickly output the optimal grouting point and grouting parameters when different three-dimensional geological weakness distributions are input.

[0064] The formula is expressed as:

[0065] S(z i )t =v(x i ,S(z i ) t-1 ) t +b(t) (2)

[0066] Where S represents the different distributions of weaknesses at the current time step, z i This represents the different distributions of weaknesses, where t and t-1 represent time steps, i represents different weakness categories, and x represents the different distributions of weaknesses. i The parameters represent the actual grouting parameters corresponding to the weaknesses, such as grout type and grouting point location. `v` represents the weakness parameter status at the current time step, resulting from the combined effect of the grouting parameter selection at the current time step and the weakness (adverse geological area) parameter status at the previous time step. `b(t)` represents the influence of the surrounding environment on grouting at each time step. This can be considered as follows: within a set time step cycle, based on the acquired weakness distribution information, the three-dimensional geological weakness distribution is completed using the optimal grouting point location and grouting parameters at each time step. Numerically, this can be represented as follows: within the set time step cycle, assuming the initial three-dimensional geological weakness is assigned a value of 1, the grouting point location and grouting parameters are dynamically adjusted at each time step, ultimately assigning the three-dimensional geological weakness distribution a value of 0. That is, grouting can complete the completion of the three-dimensional geological weakness.

[0067] Three-dimensional pooling is used to preserve important features, improving the model's computational efficiency and spatial awareness. Multi-layer convolutional pooling is performed, and residual blocks and skip connections are added to the deep network to maintain the model's depth, avoid gradient vanishing, and further enhance the model's ability to represent complex geological structures.

[0068] An adaptive optimization strategy was adopted to adjust the parameters of the training data, and the training data was used to obtain a targeted grouting prediction model under normal conditions.

[0069] In this embodiment, the targeted grouting prediction model has corresponding time-step data at each time step. By setting a threshold for reconstruction error, it periodically determines whether geological weaknesses change under the predicted optimal grouting point and grouting parameters. Periodicity refers to setting error thresholds for different time periods. In other words, the original loss was used to train the model, while the current reconstruction error threshold is used to periodically determine whether the original optimal grouting point and parameters can be achieved under existing grouting parameters. The error threshold is set as a percentage error between the actual grouting process and the original model's prediction.

[0070] Based on the established targeted grouting prediction model, a threshold for reconstruction error is set to accommodate adjustments to the grouting strategy under abnormal conditions. The period for each threshold update is also set. Specifically, after the targeted grouting prediction model outputs the optimal grouting points and parameters for the current engineering area, an error threshold determination is performed again (this threshold determination differs from the convergence loss determination during the original targeted grouting prediction model training). Through error threshold analysis, it is periodically determined whether any grouting anomalies exist based on the established grouting strategy, and adjustments are made accordingly.

[0071] (1) When the error is within the set threshold, it meets the target grouting standard, and target grouting is performed based on the output of the target grouting prediction model.

[0072] (2) When the error exceeds the set threshold, that is, an abnormal situation is detected during the grouting process. There are two types of grouting abnormalities: First, as the grouting progresses, the three-dimensional geological weaknesses in the grouting area will be continuously improved, that is, the optimal parameter ratio generated above needs to be updated and adjusted periodically; Second, during the grouting process, the grouting area is disturbed by the grouting and new three-dimensional geological weaknesses are generated, or the grouting effect is not ideal due to the influence of human and environmental factors during the actual grouting process. Adjustments are made to the original predicted optimal grouting points and grouting parameters.

[0073] Within the set error threshold, no further updates are needed. Outside the threshold, the system assesses the extent of the error by monitoring factors such as failure to reduce water flow in the injected area, abnormal borehole collapse, abnormal fluctuations in grout concentration, grout loss and blockage, and sudden temperature changes. This involves determining whether the situation falls under the first or second type of anomaly. The system then updates the targeted grouting prediction model by analyzing the geological environment ahead and considering the adaptability of the grouting effect to environmental influences.

[0074] Simultaneously, considering the periodic updates of the targeted grouting area required ahead under the existing grouting and sealing conditions, i.e., the abnormal situation that occurs is that the targeted grouting and sealing effect is better, resulting in an increase in the grouting and sealing reinforcement efficiency, the time step of the targeted grouting prediction model is adjusted to adapt to the actual grouting effect on site.

[0075] S4: Based on the grouting parameters, set the operating parameters of the grouting equipment. Based on the basic principle of multi-agent deep reinforcement learning, treat each grouting equipment as an individual agent and optimize the operating parameters of each agent through reinforcement learning strategies to achieve the target grouting effect.

[0076] In this embodiment, multimodal control of the grouting equipment is also incorporated. Specifically, based on the fundamental principles of multi-agent deep reinforcement learning, each grouting device and monitoring element is treated as a separate agent. The operating parameters of each agent are optimized through reinforcement learning strategies, while simultaneously monitoring and controlling the coordinated operation of multiple grouting devices. Consistency between the operating and grouting parameters of the multiple devices is established. The operating parameters include the pressure, flow rate, and monitoring angle / azimuth of the grouting device.

[0077] By continuously optimizing the changes in parameters such as pressure, flow rate, and monitoring angle and orientation of each agent through reinforcement learning strategies, analyzing the independent actions and mutual influences of multiple agents, handling nonlinear collaborative problems, realizing multi-source information fusion, precisely adjusting every link in the grouting process, taking action based on real-time status data (i.e., data from the grouting equipment and monitoring elements themselves and acquired data), and achieving the overall grouting goal through the collaborative operation of agents.

[0078] The specific steps of reinforcement learning strategies are as follows:

[0079] (1) Establish a continuous motion space, including the start-up of each device, and establish different continuous motion models for different devices (such as the start-up of grouting equipment, grouting flow rate, grouting pressure, grout selection, grouting pipe position and pipe type, etc., the installation of sensing devices and monitoring angle, etc.), and set the continuous movement of different devices in the space.

[0080] (2) Quantify the data of each step of the operation of each device in the continuous space (e.g., set the rotation angle to 360°, the grouting rate to 5m). 3 / h, etc., will display the status variables of each device in a data format.

[0081] (3) Based on the aforementioned distribution of three-dimensional geological weaknesses, optimal grouting locations, and grouting parameters, and combined with historical data analysis, the optimal actions and their state variables are selected for different combinations to assess the actual progress of grouting treatment of three-dimensional geological weaknesses under different data quantification states for each action. The reward function corresponding to the actual progress for different actions and different data quantification states can be set as follows:

[0082]

[0083] Where R represents the actual grouting progress corresponding to the state quantity of different data quantization for each action, m represents the total number of equipment, n represents the total number of actions of equipment j, j represents different equipment, k represents a certain action of equipment j, t is the time step, α(t) and β(t) represent the adjustment coefficients at different times, and w j (t) represents the loss of device j at a certain time, λ j(t) represents the reward coefficient for the actual grouting progress corresponding to the different data quantification state quantities of each action of each device at different times, Y. j y represents the state quantity of device j at a certain time during a certain action. k Y represents the state quantity of device j at a certain time step under action k. j (y k ,t) represents the state quantity of each action of each device at different times.

[0084] By comparing the real-time status of the equipment with the target grouting effect, the output power and pressure of the grouting equipment are adjusted to ensure that the grout can accurately reach the target area.

[0085] Based on the grouting parameters predicted by the targeted grouting prediction model using the quantified state variables of existing equipment data, the overall grouting progress R is optimized. This means that the operation of each grouting device corresponds to achieving the actual grouting parameters. After determining the grouting parameters through the targeted grouting prediction model, the required action combinations for each agent's corresponding grouting parameters are determined. For example, after determining the grout type, the target value is achieved by adjusting the grout-water ratio and conveying speed of the grouting platform. The combination value calculation can be expressed as:

[0086]

[0087] Where, x i This represents the actual grouting parameters corresponding to the weakness, such as grout type and grouting point location. j represents different equipment, k represents a specific action of equipment j, n represents the total number of actions of equipment j, and γ represents... j,k y represents the weight of a certain action k of device j on the contribution value of parameter j. k Z represents the state quantity of device j at a certain time step under action k. j (y k ) represents the contribution value corresponding to the state variable.

[0088] Based on the known actions conforming to the above formula, the maximum value of the total grouting progress R is obtained using the Lagrange multiplier method.

[0089] Grouting advance cycle refers to the distance advanced to complete one grouting cycle during grouting construction. After each grouting advance cycle, geophysical exploration and data acquisition of the preceding strata are conducted again. A three-dimensional geological weakness prediction model is used to predict weaknesses in the preceding strata. Simultaneously, the water flow path and pressure distribution after each grouting cycle are compared with the previous grouting cycle in three-dimensional space to analyze the grouting sealing and reinforcement effect. Geological samples are taken for soil and rock physical property testing to verify the grouting success rate and the long-term stability of the soil and rock reinforcement. The grouting effect is evaluated and feedback is provided.

[0090] After each grouting cycle, data from advanced detection methods such as induced polarization, borehole television, cross-hole CT, and advanced horizontal drilling, along with data obtained from the engineering site, are used to analyze the water flow path and pressure distribution after each grouting cycle. Simultaneously, based on the geological data obtained for the current cycle segment, a three-dimensional geological weakness prediction model is used to obtain the three-dimensional geological weakness distribution of the current grouting cycle segment. The three-dimensional geological weakness distribution, water flow path, and pressure distribution of the current grouting cycle segment are compared with those of the previous grouting cycle to analyze the grouting sealing and reinforcement effect. Geological samples are taken from the geological area for soil and rock physical property testing to verify the grouting success rate and the long-term stability of the soil and rock reinforcement. The grouting effect is evaluated and feedback is provided.

[0091] Example 2

[0092] This embodiment provides a deep learning-based targeted grouting high-efficiency sealing and reinforcement system, including:

[0093] The data acquisition module is configured to acquire geological data of the current engineering area and preprocess it to obtain preprocessed geological data.

[0094] The weakness distribution prediction module is configured to input preprocessed geological data into a pre-trained three-dimensional geological weakness prediction model to predict the three-dimensional geological weakness distribution.

[0095] The grouting prediction module is configured to: input the three-dimensional geological weakness distribution into a pre-trained targeted grouting prediction model for prediction, and obtain the optimal grouting point location and grouting parameters for the current engineering area; based on a set threshold for the reconstruction error of the targeted grouting prediction model, when the error is within the set threshold, perform targeted grouting using the optimal grouting point location and grouting parameters; when the error exceeds the set threshold, update the geological environment through the three-dimensional geological weakness prediction model, and then update the weakness distribution of the targeted grouting prediction model again.

[0096] The grouting equipment operation module is configured to: set the operation parameters of the grouting equipment based on the grouting parameters; and, based on the basic principle of multi-agent deep reinforcement learning, treat each grouting equipment as an individual agent and optimize the operation parameters of each agent through reinforcement learning strategies to achieve the target grouting effect.

[0097] Example 3

[0098] The purpose of this embodiment is to provide a computing device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the method of Embodiment 1.

[0099] Example 4

[0100] The purpose of this embodiment is to provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the steps of the method of Embodiment 1.

[0101] The steps and methods involved in the apparatuses of Embodiments 3 and 4 above correspond to those in Embodiment 1. For specific implementation details, please refer to the relevant description section of Embodiment 1. The term "computer-readable storage medium" should be understood as a single medium or multiple media including one or more instruction sets; it should also be understood as including any medium capable of storing, encoding, or carrying an instruction set for execution by a processor and enabling the processor to perform any of the methods in this invention.

[0102] Those skilled in the art will understand that the modules or steps of the present invention described above can be implemented using general-purpose computer devices. Optionally, they can be implemented using computer-executable program code, thereby allowing them to be stored in a storage device for execution by a computer device, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. The present invention is not limited to any particular combination of hardware and software.

[0103] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0104] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A deep learning-based targeted positioning grouting efficient plugging and reinforcement method, characterized in that, The method comprises the following steps: obtaining geological data of a current engineering area and preprocessing the geological data to obtain preprocessed geological data; inputting the preprocessed geological data into a pre-trained three-dimensional geological weak point prediction model for prediction to obtain a three-dimensional geological weak point distribution; inputting the three-dimensional geological weak point distribution into a pre-trained targeted grouting prediction model for prediction to obtain an optimal grouting point and grouting parameters of the current engineering area; based on the set threshold of the error of the targeted grouting prediction model, when the error is within the set threshold, the optimal grouting point and grouting parameters are used for targeted grouting; when the error exceeds the set threshold, the geological environment is updated through the three-dimensional geological weak point prediction model, and the weak point distribution of the targeted grouting prediction model is updated again; based on the grouting parameters, the operating parameters of the grouting equipment are set, and based on the basic principle of multi-agent deep reinforcement learning, each grouting equipment is regarded as a separate agent, and the operating parameters of each agent are optimized through a reinforcement learning strategy to achieve a target grouting effect; the reward function set in the reinforcement learning strategy is represented as: wherein, the state quantity of different data quantity for each action corresponds to the actual grouting progress, denotes the total number of devices, denotes the device the total number of actions, denotes different devices, denotes the device a certain action, is the time step, and respectively denote the adjustment coefficient at different times, denotes the loss amount of the device at a certain time, is the reward coefficient of the actual grouting progress corresponding to the state quantity of different data quantity for each action of each device at different times, denotes the state quantity of a certain action of the device at a certain time, denotes the state quantity of the device at a certain time step in action , is the state quantity of each action of each device at different times.

2. The deep learning-based targeted positioning grouting efficient plugging and reinforcing method of claim 1, wherein, the three-dimensional geological weak point prediction model is constructed by using a generative adversarial network, including a generator and a discriminator, a variational autoencoder is used as the backbone structure of the network generator, and a Transformer is used as the backbone structure of the discriminator.

3. The deep learning-based targeted positioning grouting efficient plugging and reinforcing method of claim 2, wherein, the data processing process in the three-dimensional geological weak point prediction model is specifically as follows: the preprocessed geological data is input as input data, and is output through the generator with the variational autoencoder as the backbone structure to obtain a three-dimensional geological weak point distribution; meanwhile, the three-dimensional geological weak point distribution is associated with the geological data to obtain geological data carrying parameter information; then, the actual poor geological distribution in the stratum is compared with the generated three-dimensional geological weak point distribution through the discriminator with the Transformer as the backbone structure, and mutual information is used as a loss term; based on the back propagation update optimization solution, when the model generated by the generator is close to the real model distribution and the loss term meets the requirements, the model training is completed.

4. The deep learning-based targeted positioning grouting efficient plugging and reinforcing method of claim 1, wherein, the targeted grouting prediction model is constructed based on a convolutional neural network, different three-dimensional geological weak point distributions are output through the three-dimensional geological weak point prediction model, including a single weak point and a multi-weak point combination.

5. The deep learning-based targeted positioning grouting high-efficiency plugging and reinforcing method according to claim 4, characterized in that, the single weak point is targeted grouting according to the characteristics of different weak points, and the multi-weak point combination is predicted according to the pre-trained targeted grouting prediction model, and the relationship between different three-dimensional geological weak point distributions and the optimal grouting point and grouting parameters under the condition of the existence of multiple weak points is learned through neural network learning on historical grouting engineering data; the formula is represented as: wherein, represents different distribution of vulnerabilities at the current time step, represents different distribution of vulnerabilities, and represents time step, represents different vulnerability classification, represents actual grouting parameters corresponding to the vulnerability, represents the vulnerability parameter situation at the current time step under the joint action of the grouting parameter selection at the current time step and the vulnerability parameter situation at the previous time step, represents that grouting is affected by the surrounding environment at each time step.

6. The deep learning-based targeted positioning grouting high-efficiency plugging and reinforcing method according to claim 1, characterized in that, the action combination value required for determining the grouting parameters corresponding to each agent is calculated as: wherein, denotes the actual grouting parameter corresponding to the weakness, denotes a different device, denotes a device of a certain action, denotes the total number of actions of a device , denotes a certain action of a device , denotes the weight of the parameter contribution value, denotes a device in an action at a certain time step, denotes the contribution value corresponding to the state quantity.

7. The deep learning-based targeted positioning grouting efficient plugging and reinforcing system, which realizes the deep learning-based targeted positioning grouting efficient plugging and reinforcing method according to any one of claims 1-6, characterized in that, The method comprises the following steps: a data acquisition module configured to obtain geological data of a current engineering area and preprocess the geological data to obtain preprocessed geological data; a weak point distribution prediction module configured to input the preprocessed geological data into a pre-trained three-dimensional geological weak point prediction model for prediction to obtain a three-dimensional geological weak point distribution; The grouting prediction module is configured to input the three-dimensional geological weak point distribution into a pre-trained targeted grouting prediction model for prediction to obtain optimal grouting points and grouting parameters of the current engineering area; based on a set threshold of error of the targeted grouting prediction model, when the error is within the set threshold, the optimal grouting points and grouting parameters are used for targeted grouting; when the error exceeds the set threshold, the geological environment is updated through the three-dimensional geological weak point prediction model, and the weak point distribution of the targeted grouting prediction model is updated again; The grouting equipment operation module is configured to set operation parameters of grouting equipment based on the grouting parameters, and based on the basic principle of multi-agent deep reinforcement learning, each grouting equipment is regarded as a separate agent, and the operation parameters of each agent are optimized through a reinforcement learning strategy to achieve a target grouting effect.

8. A computer-readable storage medium having stored thereon a computer program, characterized in that, The program, when executed by a processor, implements the steps in the deep learning-based targeted positioning grouting efficient plugging and reinforcement method of any one of claims 1-6.

9. A computer device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor executes the program to implement the steps in the deep learning-based targeted positioning grouting efficient plugging and reinforcement method of any one of claims 1-6.

Images