Karst stratum shield tunneling parameter prediction method and system based on federated learning

Abstract

The application provides a karst stratum shield tunneling parameter prediction method and system based on federated learning, relates to the shield tunneling technical field, and comprises the following steps: collecting multi-modal data and performing pretreatment; constructing a deep learning classification model, combining multi-modal data and federated learning technology, constructing a karst cave three-dimensional distribution model and extracting micro feature indexes; analyzing the extracted micro feature indexes through the deep learning classification model to obtain a karst cave type; inputting the quantized micro feature indexes and the karst cave type into a parameter mapping model to establish the internal relationship between the micro features and the tunneling parameters; obtaining a tunneling parameter adjustment instruction according to the internal relationship, the karst cave type and the distribution complexity; and realizing real-time dynamic accurate adjustment of the shield tunneling parameters through the tunneling parameter adjustment instruction. The multi-modal sensor, the edge computing device and the federated learning framework are combined to realize accurate prediction and real-time adjustment of the karst stratum shield tunneling parameters.

Description

Technical Field

[0001] This application relates to the field of tunnel boring machine (TBM) technology, and in particular to a method and system for predicting TBM tunneling parameters in karst formations based on federated learning. Background Technology

[0002] Currently, tunnel boring machines (TBMs) face significant technical challenges during construction in karst formations. Karst formations are widely distributed in tunnel construction areas, and their karst caves are characterized by uneven distribution and high concealment, posing a major threat to the TBM excavation process. The distribution of karst caves and the resulting mudslides and water inrushes are particularly prominent, not only affecting construction efficiency but also posing serious safety hazards.

[0003] One existing approach is to utilize finite element numerical simulation (FEM) technology to establish a simulation model of tunnel boring machine (TBM) excavation based on geological exploration data. For example, patent CN202211102890.8 proposes a technical solution to predict the range of tunneling parameters through orthogonal experimental design and discrete element coupled models. This method obtains the optimal range of tunneling speed, cutterhead rotation speed, and auger rotation speed by simulating adjustments to tunneling parameters under different excavation conditions. However, this method mainly relies on offline analysis and static data, and cannot dynamically respond to sudden changes in geological conditions during tunneling, resulting in poor adaptability.

[0004] Another approach combines slag removal monitoring and image recognition technology to detect the occurrence of karst caves in real time during tunnel boring machine (TBM) excavation. For example, patent CN202410288627.5 uses a YOLO model to train images of karst cave filling materials, and monitors the torque changes of the cutterhead and the distribution of karst cave filling materials through a slag removal detection camera to achieve real-time identification and area calculation of karst caves. This method has certain advantages in identifying karst caves, but it mainly targets post-excavation monitoring and cannot optimize excavation parameters in advance, thus lacking sufficient responsiveness to dynamic changes during the excavation process.

[0005] While the aforementioned methods each have their own focus, they still have the following shortcomings in predicting tunneling parameters in karst formations: First, data acquisition is limited. Existing ground-penetrating radars and sensors are mainly used for detecting macroscopic geological information, while the monitoring accuracy for microscopic features such as karst cave distribution and filling material composition is low, leading to a lag in tunneling parameter adjustment. Second, traditional models have low prediction accuracy and are unable to effectively cope with the highly heterogeneous geological environment in karst formations, especially the complex distribution of filled and cavitary karst caves. During shield tunneling, sudden karst problems (such as water inrush and mudslides) can easily cause significant safety hazards. Existing technologies are mostly based on offline analysis, resulting in weak real-time response capabilities and an inability to adjust parameters in real time during dynamic tunneling, lacking the ability to dynamically respond to abrupt geological changes. Finally, existing technologies typically focus on a single project, resulting in weak inter-project collaboration capabilities and an inability to improve the model's generalization performance through cross-project data sharing. Summary of the Invention

[0006] The purpose of this invention is to address the problem that existing methods for predicting tunneling parameters in karst formations cannot detect microscopic features in real time during dynamic tunneling, resulting in poor prediction accuracy. This invention provides a method and system for predicting shield tunneling parameters in karst formations based on federated learning.

[0007] The above-mentioned objective of this application is achieved through the following technical solution:

[0008] S1: Collect and preprocess multimodal data of karst formations; multimodal data includes: high-resolution point cloud data of karst caves, pressure change information, temperature information, humidity information, and hydrological data;

[0009] S2: Construct a deep learning classification model using Convolutional Neural Networks (CNN) and Transformer algorithms;

[0010] S3: By using a deep learning classification model, combined with multimodal data and federated learning techniques, a three-dimensional distribution model of karst caves is constructed and microscopic feature indicators are extracted.

[0011] S4: The extracted microscopic feature indicators are analyzed using a deep learning classification model to determine the type of karst cave;

[0012] S5: Construct a parameter mapping model; input the quantified microscopic feature indicators and karst cave types into the parameter mapping model, and establish the intrinsic relationship between microscopic features and tunnel boring parameters through multivariate regression analysis;

[0013] S6: Divide the karst cave area in the 3D distribution model of the karst cave into different geological zones, and obtain the tunneling parameter adjustment instructions based on the internal relationships, karst cave types and distribution complexity;

[0014] S7: By adjusting the tunneling parameters through the tunneling parameter adjustment command, the tunneling parameters of each geological zone are adjusted, and the hydraulic pumps, cutterhead drive motors, water injection pumps and valves of each project are controlled to complete the real-time dynamic and precise adjustment of the tunneling parameters of the shield machine.

[0015] Optionally, step S1 includes: the preprocessing includes: data cleaning, feature extraction and standardization operations.

[0016] Optionally, step S3 includes:

[0017] Mineral spectral data obtained through spectral analysis instruments;

[0018] By combining mineral spectral data and multimodal data with a deep learning classification model, a three-dimensional distribution model of karst caves is constructed and microscopic feature indicators are extracted.

[0019] Microscopic characteristic indicators include: wall roughness Crack density Mineral composition index and moisture content index .

[0020] Optionally, step S3 may also include:

[0021] The surface roughness was calculated by fitting the high-resolution point cloud data of the cave. ;

[0022] The crack density is obtained by identifying and counting the number of cracks per unit area using a deep learning model. ;

[0023] By using mineral spectral data and combining it with deep learning algorithms, the proportions of each mineral component are identified, and a mineral composition index is derived. ;

[0024] The water content index is calculated using temperature, humidity, and hydrological data. The formula is as follows:

[0025]

[0026] in, The water content measured by the moisture content sensor, i.e., hydrological data; Relative humidity is calculated from temperature and humidity information. and Let be the weighting coefficient, satisfying .

[0027] Optionally, step S4 may also include:

[0028] The extracted microscopic feature indicators are analyzed using a deep learning classification model to determine the types of karst caves; the types of karst caves include: filled karst caves and hollow karst caves.

[0029] The output of the deep learning classification model is a probability distribution of the cave type. Defined as:

[0030]

[0031] in, This represents the input multimodal sensor data. This is a feature extraction and modeling function that combines CNN and Transformer.

[0032] Optionally, step S6 includes:

[0033] The specific steps for calculating the distribution complexity of karst caves are as follows:

[0034] S61: Extract the geometric features of the cave using high-resolution point cloud data; geometric features include: shape, size, and volume;

[0035] S62: Based on the high-resolution point cloud data of the cave, perform connectivity analysis and calculate connectivity. Specifically, it includes:

[0036] Set a distance threshold;

[0037] If the distance between two caves is less than a distance threshold, they are considered to be connected and have a connection relationship.

[0038] Construct an undirected graph with each cave as a vertex and connectivity as edges, and calculate connectivity. ;

[0039] S63: Based on the geometric features and connectivity of the caves The distributed complexity is calculated as follows:

[0040]

[0041] in, For distributed complexity, The volume of the cave. The length of the cave. This refers to the connectivity of the karst caves.

[0042] Optionally, step S7 includes: the tunneling parameter adjustment command includes: thrust increase / decrease, cutterhead speed adjustment, and water injection volume control; tunneling parameters include thrust. Cutter head speed and water injection volume .

[0043] A parameter prediction system for shield tunneling in karst formations based on federated learning, the system comprising: a multimodal sensor module, an edge computing device, a communication module, a shield machine control unit, and a central server;

[0044] The multimodal sensor module is connected to the edge computing device; the edge computing device is connected to the communication module; the communication module is connected to the central server; the tunnel boring machine control unit is connected to the edge computing device.

[0045] The multimodal sensor module includes: ground-penetrating radar, pressure sensor, temperature and humidity sensor, and moisture content detector;

[0046] The ground-penetrating radar is installed at the front end of the tunnel boring machine to scan the distribution and geometric features of karst caves in the karst strata in real time, obtain high-resolution point cloud data of the caves, and generate spatial location and structural information of the caves.

[0047] The pressure sensors are installed on the cutterhead and screw conveyor of the tunnel boring machine to monitor the dynamic changes in the surrounding rock pressure during the tunneling process and obtain pressure change information.

[0048] The temperature and humidity sensors are installed around the cutterhead of the tunnel boring machine to monitor the temperature and humidity information in the karst cave environment;

[0049] The moisture content detector is installed inside the tunnel boring machine's excavation chamber and is used to monitor hydrological data in the karst cave environment;

[0050] Edge computing devices include: high-performance computing units (GPUs), data storage units, and communication interface modules;

[0051] The multimodal sensor module is used to collect multimodal data, including: the spatial location of the cave, the structural information of the cave, pressure change information, temperature information, humidity information, and hydrological data.

[0052] The edge computing device is used to preprocess the received multimodal data and perform edge computing to obtain local model parameters; the preprocessing includes: data cleaning, feature extraction and standardization operations;

[0053] The communication module includes: a 5G module and a cache memory;

[0054] The communication module is used to transmit local model parameters to the central server;

[0055] The central server is used to build deep learning classification models based on convolutional neural networks (CNN) and Transformer algorithms;

[0056] The central server is used to obtain globally optimized model parameters based on local model parameters, deep learning classification models, and federated learning techniques.

[0057] The tunnel boring machine control unit includes: a PLC controller and an actuator drive module;

[0058] The edge computing device is also used to receive and process the globally optimized model parameters, obtain tunneling parameter adjustment instructions, and send them to the shield machine control unit.

[0059] The tunnel boring machine control unit includes: a hydraulic pump, a cutterhead drive motor, a water injection pump, and valves;

[0060] The tunnel boring machine control unit is used to control the hydraulic pump, cutterhead drive motor, water injection pump and valves through tunneling parameter adjustment commands, so as to complete the real-time dynamic and precise adjustment of the tunneling parameters of the tunnel boring machine.

[0061] A computer-readable storage medium storing instructions that, when executed, perform a method for predicting shield tunneling parameters in karst formations based on federated learning.

[0062] The beneficial effects of the technical solution provided in this application are:

[0063] 1. Application of Multimodal Data Fusion Technology: The system integrates multiple sensors, such as ground-penetrating radar, pressure sensors, temperature and humidity sensors, and water content detectors, to collect environmental data of karst formations in real time. After preprocessing by edge computing devices, this data is cleaned and standardized. Then, through convolutional neural networks (CNN) and Transformer algorithms, the multi-source data is fused to achieve accurate modeling of karst cave distribution and extraction of key features, generating a real-time updated 3D distribution model of karst caves.

[0064] 2. Application of the Federated Learning Framework: The system relies on the federated learning framework and adopts homomorphic encryption technology and differential privacy protection mechanism to upload the local model update parameters of different projects to the central server. Data collaborative optimization is achieved through global model training, while ensuring that data privacy between projects is not leaked.

[0065] 3. Dynamic tunneling parameter prediction and real-time adjustment: Based on the 3D model of the karst cave and real-time sensor data, and according to the classification model of filled karst caves and hollow karst caves, the system dynamically predicts key tunneling parameters, such as thrust, cutterhead speed and water injection volume, and achieves automatic optimization of parameters by combining a real-time feedback mechanism.

[0066] 4. Collaborative operation of edge computing and communication modules: By deploying high-performance computing units at the edge of the tunnel boring machine, the system achieves real-time processing of sensor data, model inference, and command generation, avoiding the latency and uncertainty caused by transmitting large amounts of data to the central server. The communication module plays a relay and storage role in the process of uploading data and distributing models.

[0067] 5. Closed-Loop Feedback Mechanism for the Entire System: The system achieves a complete workflow of data acquisition, model optimization, parameter adjustment, and feedback through a closed-loop feedback mechanism. The sensor module continuously collects real-time data during construction and transmits feedback information back to the edge computing device for further optimization of the tunneling parameter prediction model. This closed-loop design ensures the system's prediction accuracy and response speed, enabling it to better cope with the complex and variable geological conditions of karst formations. Attached Figure Description

[0068] The present application will be further described below with reference to the accompanying drawings and embodiments. In the accompanying drawings:

[0069] Figure 1 This is a system structure diagram in an embodiment of this application;

[0070] Figure 2 This is a flowchart of the three-dimensional modeling process of the karst cave in the embodiments of this application;

[0071] Figure 3 This is a flowchart of the dynamic tunneling parameter prediction process in the embodiments of this application;

[0072] Figure 4 This is a diagram showing the setup of the cutter head sensor in an embodiment of this application. Detailed Implementation

[0073] To provide a clearer understanding of the technical features, objectives, and effects of this application, the specific embodiments of this application will now be described in detail with reference to the accompanying drawings.

[0074] The embodiments of this application provide a method for predicting shield tunneling parameters in karst formations based on federated learning.

[0075] Please refer to Figure 1 , Figure 1 This is a system structure diagram of a method for predicting shield tunneling parameters in karst formations based on federated learning, as described in an embodiment of this application, including:

[0076] S1: Collect and preprocess multimodal data of karst formations; multimodal data includes: high-resolution point cloud data of karst caves, pressure change information, temperature information, humidity information, and hydrological data;

[0077] S2: Construct a deep learning classification model using Convolutional Neural Networks (CNN) and Transformer algorithms;

[0078] S3: By using a deep learning classification model, combined with multimodal data and federated learning techniques, a three-dimensional distribution model of karst caves is constructed and microscopic feature indicators are extracted.

[0079] S4: The extracted microscopic feature indicators are analyzed using a deep learning classification model to determine the type of karst cave;

[0080] S5: Construct a parameter mapping model; input the quantified microscopic feature indicators and karst cave types into the parameter mapping model, and establish the intrinsic relationship between microscopic features and tunnel boring parameters through multivariate regression analysis;

[0081] The intrinsic relationship refers to the correlation between the microscopic geometric characteristics of a karst cave (such as diameter, depth, volume, and shape parameters) and the tunnel boring machine (TBM) parameters (cutterhead rotation speed, thrust, and water injection volume). This relationship is constructed using the following model:

[0082]

[0083] Wherein, P represents the preliminary tunneling parameter adjustment suggestions, including thrust F and cutterhead rotation speed. And the water injection volume Q. D is the depth of the cave, R is the diameter of the cave, V is the volume of the cave, and the shape characteristics include the aspect ratio, irregularity coefficient, etc.

[0084] Using the above model, the system takes microscopic features as input and outputs preliminary suggestions for adjusting tunneling parameters. These preliminary suggestions will be further integrated with the type and distribution complexity of the karst caves to ultimately generate complete results for adjusting tunneling parameters.

[0085] S6: Divide the karst cave area in the 3D distribution model of the karst cave into different geological zones, and obtain the tunneling parameter adjustment instructions based on the internal relationships, karst cave types and distribution complexity;

[0086] S7: By adjusting the tunneling parameters through the tunneling parameter adjustment command, the tunneling parameters of each geological zone are adjusted, and the hydraulic pumps, cutterhead drive motors, water injection pumps and valves of each project are controlled to complete the real-time dynamic and precise adjustment of the tunneling parameters of the shield machine.

[0087] As one example, the system employs a zoning and adaptive adjustment strategy to address the complex distribution of karst caves.

[0088] Step S1 includes: The preprocessing includes: data cleaning, feature extraction and standardization operations.

[0089] Step S3 includes:

[0090] Mineral spectral data obtained through spectral analysis instruments;

[0091] By combining mineral spectral data and multimodal data with a deep learning classification model, a three-dimensional distribution model of karst caves is constructed and microscopic feature indicators are extracted.

[0092] In one embodiment of this application, such as Figure 2 As shown, based on multimodal data collected by ground-penetrating radar and sensors, a convolutional neural network (CNN) is used to extract karst cave features, and the Transformer algorithm is used to capture global dependencies in the stratigraphic data to generate a real-time updated 3D distribution model of karst caves. This model not only includes the spatial location and geometry of the karst caves, but also refines microscopic features such as wall roughness, fracture density, mineral composition ratio, and water content, providing comprehensive stratigraphic information to support the precise adjustment of tunnel boring machine parameters.

[0093] In one embodiment of this application, such as Figure 3 As shown, using a three-dimensional distribution model of karst caves and real-time multimodal data as input, multimodal data fusion technology is employed to predict key parameters such as shield machine thrust, cutterhead rotation speed, and water injection volume in real time, and to dynamically adjust the tunneling strategy based on abnormal data feedback.

[0094] Microscopic characteristic indicators include: wall roughness Crack density Mineral composition index and moisture content index .

[0095] Step S3 also includes:

[0096] The surface roughness was calculated by fitting the high-resolution point cloud data of the cave. ;

[0097] The crack density is obtained by identifying and counting the number of cracks per unit area using a deep learning model. ;

[0098] By using mineral spectral data and combining it with deep learning algorithms, the proportions of each mineral component are identified, and a mineral composition index is derived. ;

[0099] As an example, deep learning algorithms include: Convolutional Neural Networks (CNN), Transformer, Random Forest, Support Vector Machine (SVM), etc.

[0100] As one example, the real-time acquisition of mineral composition indices relies on a combination of a spectral analyzer and a deep learning model: a mineral spectral analyzer is used to acquire mineral composition spectral data in real time, and then a pre-trained CNN model is used to classify and perform proportion analysis on the spectral data. The analysis of the spectral data is typically completed within seconds, thus allowing for real-time acquisition.

[0101] The water content index is calculated using temperature, humidity, and hydrological data. The formula is as follows:

[0102]

[0103] in, The water content measured by the moisture content sensor, i.e., hydrological data; Relative humidity is calculated from temperature and humidity information. and Let be the weighting coefficient, satisfying ;

[0104] Step S4 also includes:

[0105] The extracted microscopic feature indicators are analyzed using a deep learning classification model to determine the types of karst caves; the types of karst caves include: filled karst caves and hollow karst caves.

[0106] The output of the deep learning classification model is a probability distribution of the cave type. Defined as:

[0107]

[0108] in, This represents the input multimodal sensor data. This is a feature extraction and modeling function that combines CNN and Transformer.

[0109] As one example, based on the output of the deep learning classification model, caves are classified into two types: filled caves and empty caves. Filled caves typically have higher density and a diverse distribution of mineral components, while empty caves exhibit a hollow structure and may contain water or other fluids.

[0110] Step S6 includes:

[0111] The specific steps for calculating the distribution complexity of karst caves are as follows:

[0112] S61: Extract the geometric features of the cave using high-resolution point cloud data; geometric features include: shape, size, and volume;

[0113] S62: Based on the high-resolution point cloud data of the cave, perform connectivity analysis and calculate connectivity. Specifically, it includes:

[0114] Set a distance threshold;

[0115] If the distance between two caves is less than a distance threshold, they are considered to be connected and have a connection relationship.

[0116] Construct an undirected graph with each cave as a vertex and connectivity as edges, and calculate connectivity. The specific calculation formula is as follows:

[0117]

[0118] S63: Based on the geometric features and connectivity of the caves The distributed complexity is calculated as follows:

[0119]

[0120] in, For distributed complexity, The volume of the cave. The length of the cave. This refers to the connectivity of the karst caves.

[0121] In one embodiment of this application, the karst cave area is divided into three categories—simple, complex, and highly complex—based on the degree of distribution complexity to guide the optimization of tunneling parameters, and finally to obtain tunneling parameter adjustment instructions.

[0122] Step S7 includes:

[0123] The commands for adjusting tunneling parameters include: thrust increase / decrease, cutterhead speed adjustment, and water injection volume control; tunneling parameters include thrust. Cutter head speed and water injection volume .

[0124] As one example, filling the karst cave area: because the filling material may cause local instability, the system will increase the water injection volume. To improve soil stability or adjust thrust To accommodate different material densities. Simultaneously, the cutter head rotation speed... It is also necessary to make appropriate adjustments based on the properties of the filler material in order to optimize cutting efficiency and reduce mechanical wear.

[0125] The specific adjustment formula is as follows:

[0126]

[0127]

[0128]

[0129] in, Represents filling caves; This is an adjustment coefficient for different types of caves, with different adjustment strategies set according to different cave types. This indicates the thrust, cutterhead speed, and increased water injection volume before adjustment; This indicates the adjusted thrust, cutterhead speed, and increased water injection volume; Indicates adjustment factor; wall roughness Crack density Mineral composition index and moisture content index ;

[0130] As one example, in the cavity / cavity area: special attention needs to be paid to strengthening the support structure, and the system will appropriately reduce the thrust. To avoid putting excessive pressure on the cave walls, we will also strengthen the monitoring of water pressure changes and adjust the water injection volume in a timely manner. At the same time, the cutter head rotation speed It is also necessary to make appropriate adjustments based on the properties of the filler material in order to optimize cutting efficiency and reduce mechanical wear.

[0131] The specific adjustment formula is as follows:

[0132]

[0133]

[0134]

[0135] in, It represents a hollow cavern.

[0136] A parameter prediction system for shield tunneling in karst formations based on federated learning, the system comprising: a multimodal sensor module, an edge computing device, a communication module, a shield machine control unit, and a central server;

[0137] The multimodal sensor module is connected to the edge computing device; the edge computing device is connected to the communication module; the communication module is connected to the central server; the tunnel boring machine control unit is connected to the edge computing device.

[0138] The multimodal sensor module includes: ground-penetrating radar, pressure sensor, temperature and humidity sensor, and moisture content detector;

[0139] The ground-penetrating radar is installed at the front end of the tunnel boring machine to scan the distribution and geometric features of karst caves in the karst strata in real time, obtain high-resolution point cloud data of the caves, and generate spatial location and structural information of the caves.

[0140] The pressure sensors are installed on the cutterhead and screw conveyor of the tunnel boring machine to monitor the dynamic changes in the surrounding rock pressure during the tunneling process and obtain pressure change information.

[0141] The temperature and humidity sensors are installed around the cutterhead of the tunnel boring machine to monitor the temperature and humidity information in the karst cave environment;

[0142] The moisture content detector is installed inside the tunnel boring machine's excavation chamber and is used to monitor hydrological data in the karst cave environment;

[0143] Edge computing devices include: high-performance computing units (GPUs), data storage units, and communication interface modules;

[0144] The multimodal sensor module is used to collect multimodal data, including: the spatial location of the cave, the structural information of the cave, pressure change information, temperature information, humidity information, and hydrological data.

[0145] The edge computing device is used to preprocess the received multimodal data and perform edge computing to obtain local model parameters; the preprocessing includes: data cleaning, feature extraction and standardization operations;

[0146] In one embodiment, all sensors aggregate data to an edge computing device via an interface to complete real-time data acquisition. The edge computing device is the core hardware module of the system, with built-in high-performance computing units (such as GPUs), data storage units, and communication interface modules, enabling rapid processing of sensor data and model inference. The device receives multimodal data from the sensor modules through a standardized interface, performs data cleaning, feature extraction, and standardization operations, and transmits the processing results to the central server via the communication module.

[0147] The communication module includes: a 5G module and a cache memory;

[0148] The communication module is used to transmit local model parameters to the central server;

[0149] In one embodiment, the communication module consists of a wireless communication device (such as a 5G module) and a cache memory, acting as a bridge in the system. On one hand, the communication module receives local model parameters uploaded by the edge computing device and transmits them to the central server; on the other hand, the central server sends globally optimized model parameters to the edge computing device through the communication module.

[0150] The central server is used to build deep learning classification models based on convolutional neural networks (CNN) and Transformer algorithms;

[0151] The central server is used to obtain globally optimized model parameters based on local model parameters, deep learning classification models, and federated learning techniques.

[0152] The tunnel boring machine control unit includes: a PLC controller and an actuator drive module;

[0153] The edge computing device is also used to receive and process the globally optimized model parameters, obtain tunneling parameter adjustment instructions, and send them to the shield machine control unit.

[0154] The tunnel boring machine control unit includes: a hydraulic pump, a cutterhead drive motor, a water injection pump, and valves;

[0155] The tunnel boring machine control unit is used to control the hydraulic pump, cutterhead drive motor, water injection pump and valves through tunneling parameter adjustment commands, so as to complete the real-time dynamic and precise adjustment of the tunneling parameters of the tunnel boring machine.

[0156] In one embodiment, the tunnel boring machine (TBM) control unit is a crucial component that executes the adjustment strategies generated by the system. It mainly consists of a PLC controller and an actuator drive module. The TBM control unit receives tunneling parameter adjustment commands from the edge computing device, including thrust increase / decrease, cutterhead speed adjustment, and water injection volume control. It then uses internal program logic to drive the hydraulic system, cutterhead drive motor, water injection pump, and valves to complete the parameter adjustments.

[0157] In one embodiment, the hydraulic system achieves precise control of propulsion by adjusting the output flow of the hydraulic pump; the cutter head drive motor achieves dynamic control of the cutter head rotation speed by adjusting the motor speed through a frequency converter; and the water injection pump and valves achieve dynamic adjustment of the water injection volume by adjusting the pump power or valve opening.

[0158] The above are merely exemplary embodiments of this disclosure and should not be construed as limiting the scope of this disclosure. Any equivalent changes and modifications made in accordance with the teachings of this disclosure shall still fall within the scope of this disclosure.

[0159] This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not described in this disclosure. The specification and embodiments are to be considered exemplary only, and the scope and spirit of this disclosure are defined by the claims.

Claims

1. A karst stratum shield tunneling parameter prediction method based on federated learning, characterized in that, The method comprises the following steps: S1: collecting multi-modal data of karst stratum and preprocessing; the multi-modal data comprises high-resolution point cloud data of karst cave, pressure change information, temperature information, humidity information and hydrological data; S2: constructing a deep learning classification model by using a convolutional neural network (CNN) and a Transformer algorithm; S3: constructing a three-dimensional distribution model of karst caves and extracting micro-feature indexes by using the deep learning classification model, combining multi-modal data and federated learning technology; S4: analyzing the extracted micro-feature indexes by using the deep learning classification model to obtain the type of karst caves; S5: constructing a parameter mapping model; inputting the quantized micro-feature indexes and the type of karst caves into the parameter mapping model to establish the internal relationship between the micro-feature and the shield tunneling parameters by multivariate regression analysis; S6: calculating the distribution complexity of karst caves; dividing the karst cave area of the three-dimensional distribution model of karst caves into different geological partitions, and obtaining the tunneling parameter adjustment instruction according to the internal relationship, the type of karst caves and the distribution complexity; S7: adjusting the tunneling parameters of each geological partition by using the tunneling parameter adjustment instruction, controlling the hydraulic pump, cutter drive motor, water injection pump and valve of each project, and completing the real-time dynamic and accurate adjustment of the tunneling parameters of the shield machine.

2. The karst stratum shield tunneling parameter prediction method based on federated learning according to claim 1, wherein, Step S1 comprises: the preprocessing comprises data cleaning, feature extraction and standardization operation.

3. The karst stratum shield tunneling parameter prediction method based on federal learning according to claim 1, characterized in that, Step S3 comprises: mineral spectrum data obtained by a spectrum analysis instrument; constructing a three-dimensional distribution model of karst caves and extracting micro-feature indexes by using the deep learning classification model, combining the mineral spectrum data and the multi-modal data; Microscopic characteristic indicators include: wall roughness , fracture density , mineral composition index and water content index .

4. The karst stratum shield tunneling parameter prediction method based on federated learning according to claim 3, characterized in that, Step S3 further comprises: Surface fitting by high resolution point cloud data of the cave, wall roughness is calculated ; The number of fissures in a unit area is identified and counted by a deep learning model, and fissure density is obtained ; By mineral spectral data, combined with deep learning algorithm, the proportion of each mineral component is identified, and a mineral component index is obtained ; Through temperature information, humidity information and hydrological data, the water content index is calculated , as follows: wherein, is the water content measured by the water content sensor, i.e. the hydrograph data; is the relative humidity, calculated from the temperature information and the humidity information; and is a weighting factor, satisfying .

5. The karst stratum shield tunneling parameter prediction method based on federal learning according to claim 1, characterized in that, Step S4 further comprises: analyzing the extracted micro-feature indexes by using the deep learning classification model to obtain the type of karst caves; the type of karst caves comprises filled karst caves and cavity karst caves; The output of the deep learning classification model is a probability distribution of the cave type is defined as: wherein, represents inputted multimodal sensor data, is a feature extraction and modeling function combining CNN and Transformer.

6. The karst stratum shield tunneling parameter prediction method based on federal learning according to claim 1, characterized in that, Step S6 comprises: The specific steps of calculating the distribution complexity of karst caves are as follows: S61: extracting the geometric features of karst caves by using the high-resolution point cloud data of karst caves; the geometric features comprise shape, size and volume; S62: According to the high-resolution point cloud data of the cave, connectivity analysis is performed, and connectivity is calculated , specifically comprising: setting a distance threshold; if the distance between two karst caves is less than the distance threshold, it is considered that the two karst caves are connected and have a connection relationship; With each cave as the vertex, the edge of the connected relationship, the construction of undirected graph, the connectivity is calculated ; S63: Based on the geometry of the cave and its connectivity , a distribution complexity calculation is performed as follows: wherein, is the distribution complexity, is the cave volume, is the cave length, is the cave connectivity.

7. The karst stratum shield tunneling parameter prediction method based on federal learning according to claim 1, characterized in that, Step S7 comprises: The tunneling parameter adjustment instruction includes: thrust increase or decrease, cutter head rotating speed adjustment and water injection amount control; the tunneling parameters include thrust , cutter head rotating speed and water injection amount .

8. A karst stratum shield tunneling parameter prediction system based on federated learning, used to implement the karst stratum shield tunneling parameter prediction method based on federated learning in any one of claims 1-7. The system comprises a multi-modal sensor module, an edge computing device, a communication module, a shield machine control unit and a central server; the multi-modal sensor module is connected to the edge computing device; the edge computing device is connected to the communication module; the communication module is connected to the central server; the shield machine control unit is connected to the edge computing device; the multi-modal sensor module comprises a geological radar, a pressure sensor, a temperature and humidity sensor and a water content detector; the geological radar is arranged at the front end of the shield machine, and is used for real-time scanning of the distribution and geometric features of karst caves in karst stratum, obtaining high-resolution point cloud data of karst caves, and generating spatial position and structure information of karst caves; the pressure sensor is arranged on the cutter head and the screw conveyor of the shield machine, and is used for monitoring the dynamic change of the surrounding rock pressure in the tunneling process to obtain pressure change information; the temperature and humidity sensor is arranged around the cutter head of the shield machine, and is used for monitoring the temperature information and humidity information in the karst cave environment; The water content detector is arranged in a tunneling cabin of the shield machine, and is used for monitoring hydrological data in a karst cave environment; The edge computing device comprises a high-performance computing unit GPU, a data storage unit and a communication interface module; The multi-modal sensor module is used for collecting multi-modal data; the multi-modal data comprises spatial position of the karst cave, structural information of the karst cave, pressure change information, temperature information, humidity information and hydrological data; The edge computing device is used for preprocessing and edge computing of the received multi-modal data to obtain local model parameters; the preprocessing comprises data cleaning, feature extraction and standardization operation; The communication module comprises a 5G module and a cache memory; The communication module is used for transmitting the local model parameters to a central server; The central server is used for constructing a deep learning classification model based on a convolutional neural network CNN and a Transformer algorithm; The central server is used for obtaining globally optimized model parameters according to the local model parameters, the deep learning classification model and a federated learning technology; The shield machine control unit comprises a PLC controller and an actuator driving module; The edge computing device is further used for receiving the globally optimized model parameters and processing to obtain tunneling parameter adjustment instructions and transmitting to the shield machine control unit; The shield machine control unit comprises a hydraulic pump, a cutter head driving motor, a water injection pump and a valve; The shield machine control unit is used for controlling the hydraulic pump, the cutter head driving motor, the water injection pump and the valve through the tunneling parameter adjustment instructions to complete real-time dynamic accurate adjustment of the tunneling parameters of the shield machine.

9. A computer-readable storage medium, characterized in that, The computer readable storage medium stores instructions, when the instructions are executed by a computer, the method of any one of claims 1-7 is executed.

Images