Shield adaptability adjusting method based on stratum structure

Abstract

The invention provides a shield adaptability adjusting method based on a stratum structure, which comprises the following steps: acquiring geological parameters of a construction area of a shield tunneling machine through drilling and geophysical prospecting, and arranging sensors at key positions of the shield tunneling machine to acquire construction parameters of the shield tunneling machine in real time; a convolutional neural network CNN is adopted to analyze the geological parameters to identify stratum features, and a geological model is constructed based on the identified stratum features; constructing a disaster risk index system, and predicting a disaster risk based on the geological model and the construction parameters by adopting Logistic regression and a random forest algorithm; according to the predicted disaster risk, the cutterhead configuration of the shield tunneling machine is dynamically adjusted, and tunneling parameters are optimized or a muck improvement scheme is adopted; by monitoring geological parameters, construction parameters and environmental changes in real time, a multistage early warning mechanism is adopted to trigger emergency response measures, and construction safety is guaranteed. According to the method, the safety and efficiency of shield construction can be improved, the geological disaster risk is reduced, and reliable technical guarantee is provided for underground space development.

Description

Technical Field

[0001] The invention relates to the technical field of shield construction, in particular to a shield adaptability adjustment method based on stratum structure. Background Art

[0002] With the acceleration of urbanization and the growing demand for underground space development, shield construction technology has been widely used in underground projects such as tunnels and subways. Currently, shield construction technology has made significant progress, with a relatively mature technical system in areas such as shield machine design and manufacturing, tunneling control, and geological survey. However, the shield construction environment is complex and changeable, and factors such as stratum structure and geological conditions have a significant impact on shield construction. How to adapt the shield machine to the actual stratum structure has become a key and difficult issue in current shield construction technology research.

[0003] Existing shield construction technology often suffers from insufficient adaptability when faced with complex strata. For one thing, the relatively limited methods used to acquire and analyze geological survey data make it difficult to fully and accurately reflect the actual stratum structure, leading to the risk of unknown geological hazards encountered by the shield machine during tunneling. Furthermore, the shield machine's tunneling parameters and adjustment measures are often based on empirical settings, lacking scientific evidence and real-time feedback mechanisms. This makes it difficult to dynamically match the shield machine with the stratum structure, impacting construction efficiency and safety.

[0004] The proposed shield tunneling adaptive adjustment method based on stratum structure utilizes a combination of drilling, geophysical exploration, and sensor monitoring to obtain geological and construction parameters for the shield tunneling area. It then employs advanced convolutional neural network (CNN) algorithms and machine learning methods such as logistic regression and random forests to accurately predict stratum type, characteristics, and disaster risks. Based on these predictions, the shield machine is then adjusted and optimized, achieving intelligent matching and dynamic adaptation between the shield machine and the stratum structure, effectively improving the safety and efficiency of shield tunneling operations.

[0005] To address these research deficiencies, this paper proposes a method that integrates drilling, geophysical exploration, and sensor monitoring to obtain geological and construction parameters within the shield construction area. This method achieves fusion analysis of multi-source data, improving the accuracy and comprehensiveness of geological survey data. A convolutional neural network (CNN) algorithm is used to perform deep learning on geological parameters to identify stratigraphic characteristics. Furthermore, a disaster risk indicator system is constructed, using logistic regression and random forest methods to accurately predict disaster risks based on geological models and construction parameters. Based on these predictions, the paper proposes a dynamic adjustment and optimization mechanism for shield machines. This mechanism uses methods such as cutterhead adjustment, tunneling parameter optimization, and soil improvement schemes to adjust and optimize the shield machine, achieving intelligent matching and dynamic adaptation between the shield machine and the stratigraphic structure. The paper also establishes a comprehensive early warning and emergency response system that provides real-time monitoring and early warning of potential disaster risks during shield construction, and formulates appropriate emergency response measures to ensure the safe and stable operation of the shield machine. Summary of the Invention

[0006] The main purpose of the present invention is to provide a shield adaptive adjustment method based on the stratum structure, aiming to solve the problems of insufficient adaptability and weak disaster risk prevention and control capabilities of traditional shield construction under complex geological conditions. By comprehensively using multi-source geological parameter acquisition technology, intelligent data analysis algorithms and dynamic adjustment mechanisms, the shield machine excavation parameters are accurately matched with the stratum structure, thereby improving the safety and efficiency of shield construction, reducing the risk of geological disasters, and providing reliable technical guarantees for underground space development.

[0007] The technical solution adopted in the present invention is:

[0008] A shield adaptability adjustment method based on stratum structure includes the following steps:

[0009] S101: Obtain geological parameters of the shield machine construction area through drilling and geophysical exploration, and place sensors at key locations of the shield machine to collect the shield machine construction parameters in real time;

[0010] S102: Analyzing the geological parameters using a convolutional neural network (CNN) to identify stratum characteristics, and constructing a geological model based on the identified stratum characteristics; constructing a disaster risk indicator system, using logistic regression and random forest algorithms to predict disaster risks based on the geological model and construction parameters, wherein the disaster risks include landslides, mud bursts, and ground subsidence;

[0011] S103: Dynamically adjust the cutterhead configuration of the shield machine, optimize excavation parameters, or adopt a slag improvement plan based on the predicted disaster risk. The excavation parameters include soil bin pressure and propulsion speed.

[0012] S104: Through real-time monitoring of geological parameters, construction parameters and environmental changes, a multi-level early warning mechanism is used to trigger emergency response measures to ensure construction safety.

[0013] Furthermore, the geological parameters include rock and soil type, structure, construction, particle composition, density, humidity, groundwater type and permeability coefficient; the construction parameters include soil bin pressure, propulsion cylinder pressure, daily feed rate, screw conveyor torque and speed, roll angle and pitch angle of the shield center, shield tail gap, and tool temperature.

[0014] Furthermore, in step S101:

[0015] During drilling, exploration holes are arranged along the tunnel axis, with the drilling spacing adjusted to 10-50 meters according to the complexity of the strata, and the drilling depth exceeds 1.5 times the tunnel burial depth;

[0016] Geophysical exploration uses geological radar, high-density electrical method and seismic wave method to detect the distribution of caves and faults, and constructs a three-dimensional geological model through multi-source data fusion; including

[0017] Sensors placed at key locations on the shield machine include a soil bin pressure sensor placed on the inner wall of the shield bin, a cutter head torque sensor integrated into the drive shaft, a temperature sensor embedded in the cutter body, a posture sensor installed in the center of the shield body, and a screw conveyor sensor installed on the screw conveyor.

[0018] Data transmission adopts a hybrid networking strategy, combining LoRa and 5G / 4G dual-channel redundant transmission. The data is encrypted with AES-256 and synchronized with the BIM model in real time.

[0019] Furthermore, in the geophysical exploration:

[0020] Geological radar detects shallow strata, high-density electrical methods generate two-dimensional / three-dimensional resistivity distribution maps to identify caves and faults, and seismic wave methods use dispersion curves to assess shear wave velocity in strata.

[0021] The geophysical data are jointly corrected with the drilling data through the Bayesian inversion algorithm to generate a lithologic-physical property joint model.

[0022] Furthermore, in step S102:

[0023] The convolutional neural network (CNN) uses an improved ResNet-50 network structure and combines the attention mechanism to fuse core images, geophysical profiles, and geotechnical parameter features.

[0024] The geological model uses the output of weighted average logistic regression and random forest, with input features including lithology, cutterhead torque, and groundwater level, to identify key risk factors by calculating the Gini impurity.

[0025] Furthermore, in the random forest model:

[0026] Analyze key risk factors through Gini impurity;

[0027] The geological model is updated every 10 meters, and the fully connected layer is fine-tuned using online learning, and the data confidence is evaluated through a Bayesian neural network.

[0028] Furthermore, in step S103:

[0029] For soft soil, use a cutterhead with an opening ratio of 45%-55%, and use foam injection to reduce cutterhead torque. The foam concentration is 3%-5%.

[0030] For hard rock formations, a mixed configuration of roller cutters and tooth cutters is used, with roller cutters accounting for 60%, and a lubrication and cooling system is installed;

[0031] The soil bin pressure is regulated in a closed loop by a PID controller, and the screw conveyor speed is automatically adjusted when the deviation exceeds ±5%.

[0032] Furthermore, the slag soil improvement plan includes:

[0033] Inject foam into soft soil formations at 0.5-1.0 L / m³, and into hard rock formations at 0.2-0.5 L / m³;

[0034] Bentonite is prepared at a mass ratio of 1:10, and the injection amount is dynamically adjusted to increase the plasticity index of the slag to 12-18.

[0035] Furthermore, in step S104:

[0036] The yellow warning trigger conditions are soil bin pressure fluctuations of ±5%-10% and cutterhead torque fluctuations of ±15%-25%. Countermeasures include intensified monitoring and manual inspections.

[0037] The triggering conditions for the orange warning are soil bin pressure fluctuations of more than ±10% or shield tail gaps greater than 30 mm. Countermeasures include reducing the speed to 60%-80% of the original value and grouting reinforcement.

[0038] The red alert triggering conditions are cutter head torque exceeding the limit or ground settlement >10 mm / d. Response measures include shutting down the machine, freezing the ground, and evacuating personnel.

[0039] Furthermore, in the emergency measures:

[0040] Grouting reinforcement uses cement slurry with a grouting pressure of 0.5-1.0 MPa;

[0041] The freezing method reduces the formation temperature to -20℃ to -30℃ through the circulation of low-temperature brine, and the freezing time is 2-4 weeks.

[0042] Compared with existing technologies, the technical solution provided by the present invention has the following beneficial effects: by comprehensively utilizing multiple technologies such as drilling, geophysical exploration, and sensor monitoring to obtain geological and construction parameters of the shield construction area, and realizing multi-source data fusion analysis, the accuracy and comprehensiveness of geological survey data are greatly improved, laying a solid foundation for subsequent research; a convolutional neural network (CNN) algorithm is used to deeply learn geological parameters, efficiently and accurately identifying stratum characteristics, overcoming the efficiency and accuracy limitations of traditional identification methods; a disaster risk indicator system is constructed, combining geological models with construction parameters, and accurately predicting disaster risks using the logistic regression algorithm and random forest method, greatly enhancing risk prediction capabilities; based on the accurate prediction results, a dynamic adjustment and optimization mechanism for the shield machine is proposed. From cutterhead control and tunneling parameter optimization to the formulation of slag improvement plans, the shield machine and the stratum structure are intelligently matched and dynamically adapted in multiple dimensions, improving construction efficiency and quality; a comprehensive early warning and emergency response system is also established to monitor and warn of construction risks in real time and formulate targeted emergency measures, comprehensively ensuring the safe and stable operation of the shield machine, and providing more intelligent, reliable, and comprehensive technical support for shield construction. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following is a brief introduction to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0044] Figure 1 This is a flow chart of a shield adaptive adjustment method based on stratum structure according to an embodiment of the present invention;

[0045] Figure 2 This is a flow chart for obtaining regional geological parameters and construction parameters according to an embodiment of the present invention;

[0046] Figure 3 This is a diagram showing the working principle of a geological radar according to an embodiment of the present invention;

[0047] Figure 4 This is a schematic diagram of the principle of measuring stratum structure using a seismic wave method according to an embodiment of the present invention;

[0048] Figure 5 is a flow chart of geological data analysis and disaster risk assessment according to an embodiment of the present invention;

[0049] Figure 6 This is a flow chart of geological parameter analysis using a convolutional neural network according to an embodiment of the present invention. DETAILED DESCRIPTION

[0050] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0051] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0052] It should be noted that the illustrations provided in the embodiments of the present invention are only schematic illustrations of the basic concept of the present invention. Therefore, the drawings only show components related to the present invention and are not drawn according to the number, shape and size of components in actual implementation. In actual implementation, the type, quantity and proportion of each component can be changed at will, and the component layout type may also be more complicated.

[0053] It should also be noted that in the present invention, the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate positions or locations based on those shown in the accompanying drawings. These terms are intended solely to facilitate and simplify the description of the present invention and are not intended to indicate or imply that the devices or components referred to must have, be constructed, or operate in a specific orientation. Therefore, they should not be construed as limitations on the present invention. Furthermore, the terms "first" and "second" are used solely for descriptive and distinguishing purposes and should not be construed as indicating or implying relative importance.

[0054] Please refer to Figure 1 The embodiment of the present invention provides a shield adaptive adjustment method based on stratum structure, comprising the following steps:

[0055] The S101 acquires geological parameters of the shield machine's construction area through drilling and geophysical exploration. Sensors are placed in the shield machine's soil bin, screw conveyor, cutterhead, and other locations to collect real-time shield machine construction parameters. These geological and construction parameters are then transmitted to a ground monitoring center via wireless transmission technology.

[0056] Specific implementation method Figure 2 As shown, the following steps are included:

[0057] S201 arranges exploration holes along the tunnel axis through drilling to obtain information such as rock and soil type and groundwater status.

[0058] In conventional strata, borehole spacing is typically 30-50 meters, but this needs to be adjusted dynamically based on geological complexity. For example, when crossing fault fracture zones or areas with developed karst caves, spacing needs to be increased to 10-20 meters to capture sudden changes in the strata. The borehole depth must penetrate the target stratum and extend to the stable layer, typically exceeding 1.5 times the tunnel burial depth to ensure coverage of the entire tunnel construction area.

[0059] After the boreholes are laid out, drilling equipment and coring techniques must be combined to obtain raw data on the formation's geotechnical parameters. Use a double-barrel drill or rotary drilling process to ensure a core recovery rate of ≥85% (this limit can be relaxed to 60% for soft soil layers). Cores should be segmented and numbered by depth, with visual characteristics such as color, structure, and degree of fracture development recorded. For cobblestone layers or fractured zones, use a diamond drill bit and control the drilling pressure to avoid core disturbance.

[0060] Core sampling and description are the basis for obtaining rock and soil types, while further physical and mechanical tests provide a quantitative basis for shield adaptability adjustments. Rock and soil samples must be delivered to the laboratory within 48 hours for the following tests:

[0061] (1) Rock strength test: The rock compressive strength test (ASTM D4964) is used to evaluate the excavation resistance of hard rock formations. For formations with a compressive strength greater than 100 MPa, it is necessary to be vigilant about the risk of tool wear.

[0062] (2) Soil classification and parameter determination: The plasticity index of clayey soil was determined using the combined liquid-plastic limit tester (ASTM D4318), and the internal friction angle and cohesion were obtained through the direct shear test (GB / T 50123).

[0063] (3) Treatment of special strata: For expansive soil or saline soil, a free expansion rate test (GB / T 50125) and a salt dissolution collapse test are required to assess the stability risk during shield advancement.

[0064] Laboratory analysis of geotechnical parameters needs to be combined with in-situ testing (such as standard penetration test (SPT)) to verify the reliability of the data.

[0065] During drilling, groundwater levels must be monitored simultaneously. A water level gauge should be installed through the borehole pipe to record the initial water level elevation, and the permeability coefficient (K value) should be measured during pumping tests. For confined water formations, a pressure gauge should be used to monitor the borehole water pressure in real time to prevent sudden surges. Accurate groundwater parameters are crucial for setting the shield chamber pressure and must form a closed-loop feedback loop with subsequent construction parameters (such as the earth pressure setpoint).

[0066] Quality control measures should be implemented during the drilling process to reduce errors. Process parameters such as drilling speed, pump volume, and torque should be recorded during the drilling process. Any abnormal fluctuations should be verified by re-drilling. Core descriptions should use a unified coding system (such as the USGS Core Description Manual) to avoid subjective bias.

[0067] The obtained drilling data should be integrated to form a drilling histogram and a hydrogeological profile, and the stratigraphic mutation points (such as the burial depth of the cave roof) should be marked.

[0068] S202 uses geophysical methods such as geological radar, high-density electrical method, seismic waves, etc. to detect the distribution of adverse geological bodies (such as caves and faults).

[0069] During shield tunneling, continuous strata exploration using non-destructive geophysical techniques compensates for the discrete nature of drilling data and provides critical support for geological model construction. Multi-technique integration and data fusion identify the distribution and scale of unfavorable geological structures (such as karst caves and fault fracture zones), complementing drilling data and ultimately providing a basis for shield tunneling adaptability adjustments.

[0070] Geological radar (GPR) is used to conduct high-resolution detection of shallow layers. It mainly consists of five parts: a host (main control unit), a transmitter, a transmitting antenna, a receiver, and a receiving antenna. The host is the acquisition system, which is used to send a series of control commands to the transmitter and receiver. The transmitter transmits electromagnetic waves into the ground according to the commands, and the receiver collects data according to the commands. The transmitting and receiving antennas are used to transmit and receive electromagnetic waves. After sampling and A / D conversion, the received reflected signal is converted into a digital signal and displayed and saved ( Figure 3 ).

[0071] The transmitting antenna continuously emits electromagnetic waves, which penetrate the underground medium. Different media have different dielectric constants. As the electromagnetic waves propagate underground, they are reflected and refracted at the interfaces between the dielectric layers. The receiving antenna picks up multiple reflected echoes (A-Scan signals), which are processed to form a B-Scan image, enabling detection, identification, and location of various targets.

[0072] By emitting high-frequency electromagnetic waves (10-1000 MHz), geological radar can rapidly image shallow (<30 meters) stratigraphic structures with a vertical resolution of up to 0.1 meter. This allows for the detection of cave roof depths, fracture zones, and interfaces between soft and hard rock layers. In shallow urban strata, a 400 MHz antenna deployed along the tunnel axis and along its sides (with spacing of 2-5 meters) can accurately identify interfaces between clay and sand layers, or cavities above tunnels. However, the penetration depth of geological radar is limited, necessitating the integration of high-density electrical methods to provide additional information from deeper layers.

[0073] High-density resistivity (HDR) deep electrical imaging is used to detect deep, unfavorable geological structures, such as caves and fault fracture zones. The High-Density Resistivity Method (HDR) is a geophysical exploration method that utilizes a multi-electrode system, densely arranged electrodes, and multi-channel data acquisition equipment to measure the subsurface resistivity distribution at high resolution. This improved resistivity method provides faster and more accurate information on the subsurface electrical structure.

[0074] The high-density electrical method is based on the principle of resistivity method. By injecting current into the ground and measuring the potential difference between electrodes, the apparent resistivity distribution of the underground medium is calculated, thereby inferring the spatial distribution characteristics of different underground geological bodies. Resistivity formula:

[0075]

[0076] Where ρ is the apparent resistivity (Ω∙m); K is the geometric coefficient, which is determined by the electrode arrangement; ΔV is the measured potential difference (V); and I is the injected current (A).

[0077] By rapidly switching positions of a multi-electrode array, two-dimensional / three-dimensional resistivity distribution maps are generated. Caves typically appear as low-resistivity (water-filled) or high-resistivity (gas-filled) anomalies in electrical methods, while fault fracture zones exhibit linear low-resistivity characteristics. Data processing requires the use of resistivity inversion algorithms (such as Occam inversion) to construct models, and electrical parameters are corrected using borehole data. For example, HDR identified a fault fracture zone at a depth of 50 meters, whose resistivity was only one-third of the surrounding rock mass. Combined with borehole verification, this provided key information for shield cutterhead selection.

[0078] Different electrode arrangements can be used for different geological conditions: the Wenner arrangement is suitable for uniform formations and has strong noise immunity; the Schlumberger arrangement is suitable for detecting deep structures; the dipole-dipole arrangement has high resolution and is suitable for complex terrain; and the symmetrical quadrupole arrangement is often used for detecting shallow structures.

[0079] Long-distance continuous exploration using seismic wave methods (such as Rayleigh waves and refraction waves) assesses stratum uniformity and the interface between soft and hard rock layers. Seismic wave methods are a geophysical exploration method that uses the differences in elastic wave velocity and wave impedance of various underground media to study the propagation patterns of seismic waves generated by artificial seismic sources (such as hammer strikes) and to determine the distribution of underground media.

[0080] Elastic waves propagate from the earthquake source to the strata. If they encounter a stratum interface with different properties, they will refract according to the law of refraction. When Rayleigh waves of different wavelengths are released, the Rayleigh wave velocity values at different penetration depths can be obtained. Based on the wave velocity values, geological bodies can be evaluated or geological stratification can be performed, thereby achieving the purpose of detection ( Figure 4 ).

[0081] The Rayleigh wave method, by analyzing dispersion curves, can determine shear wave velocities in shallow layers (<50 meters) for assessing liquefaction risk. The refraction wave method detects deep, low-velocity anomalies (such as fracture zones) and inverts the velocity structure using first-wave time-distance curves. For example, in a hard rock tunnel, the seismic wave method identified a low-velocity layer (wave velocity <2000 m / s) at a depth of 80 meters, corresponding to a fault fracture zone revealed by drilling, providing an early warning for adjusting thrust parameters.

[0082] Geophysical data requires precise interpretation through multi-source joint inversion and 3D modeling. First, geological radar, high-density electrical method, and seismic wave data are unified into a tunnel axis coordinate system to eliminate spatial deviations. Subsequently, combined with borehole lithologic data, a Bayesian inversion algorithm is used to generate a joint lithologic-physical property model. For example, the presence of a cave can be determined by combining its low resistivity (electrical method), high electromagnetic wave reflectivity (radar), and low wave velocity (seismic). Finally, a 3D geological model is constructed using GMS or GOCAD software, annotating parameters such as the cave roof depth and fault orientation.

[0083] In complex strata, geophysical exploration techniques require targeted optimization. For example, cave exploration employs a combined "point-surface" strategy: radar accurately locates the roof, and electrical methods determine the type of filling. Fault fracture zones are identified by seismic wave detection of large-scale low-velocity zones, combined with drill hole fragmentation classification. In shallow urban areas, interference from underground pipelines must be avoided, and low-frequency radar (25 MHz) is used, supplemented by offset imaging technology, to eliminate multipath effects.

[0084] S203 arranges sensors at key positions of the shield machine to collect and transmit the shield machine construction parameters.

[0085] Sensors are placed in the shield machine's soil bin, screw conveyor, cutterhead, and other locations to monitor parameters such as soil bin pressure, propulsion speed, and cutterhead torque in real time. The selection and application range of different sensors are shown in Table 1.

[0086] Table 1

[0087]

[0088] The physical layout of sensors must be tailored to the shield structure and the characteristics of the ground. For example, pressure sensors inside the soil bin are flange-mounted to prevent soil clogging and are deployed at multiple locations (at least three) to eliminate local pressure deviations. The cutterhead torque sensor is integrated into the drive shaft, and the temperature sensor is embedded in the cutter body, using wireless transmission to minimize cable interference. In soft soils, the soil bin pressure sensor requires increased redundancy to account for the viscous effects of clayey soils. In hard rock formations, the cutterhead torque sensor's range needs to be expanded to 1500 kN·m, and an additional temperature sensor should be added to monitor tool wear. Furthermore, the attitude sensor is mounted at the center of the shield and calibrated with ground control points using a laser target to ensure data consistency with the BIM model. To ensure reliability, key parameters (such as soil bin pressure and cutterhead torque) are measured using two sensors connected in parallel. Abnormal data automatically switches to the backup sensor, triggering an alarm.

[0089] Sensor protection and anti-interference design are also crucial. All sensors must meet the IP68 protection level, withstand temperatures of -20°C to 80°C and 10G vibration, and be equipped with an electromagnetic shielding layer to reduce interference with the shield motor's signal. At the same time, the sensor has a built-in temperature compensation module and automatically calibrates the zero point before starting each shift to ensure long-term stability. For example, during the construction of a subway tunnel, the cutterhead torque suddenly increased to 1200 kN·m. The system quickly identified it as a hard rock section through redundant sensors, and then automatically reduced the propulsion speed to 20 mm / rev and increased the foam injection volume to bring the torque back to a safe range, ensuring the reliability of the sensor system under complex working conditions.

[0090] Construction parameters are transmitted to the ground monitoring center in real time through wireless transmission technology. Real-time data collected by shield machine sensors is the core basis for construction decision-making. The data transmission system needs to solve the communication difficulties in the complex underground environment, while forming a dynamic association with geological parameters and construction parameters to provide reliable support for shield adaptive adjustment. In engineering practice, a hybrid networking strategy is adopted, combining the advantages of multiple technologies to achieve fast and accurate transmission of construction parameters:

[0091] (1) LoRa technology is used for long-distance underground communications due to its strong penetration and low power consumption. For example, in a subway project, the shield machine is 1 km away from the ground relay station. The LoRa network can achieve stable data transmission, but the transmission rate is low (about 10-50 kbps).

[0092] (2) 5G / 4G networks are deployed through underground waveguides or distributed base stations to address signal attenuation issues. For example, in surface coverage areas, a 5G base station is set up every 200 meters, combined with fiber optic relays to achieve high bandwidth (>100 Mbps) and low latency (<20 ms) transmission, suitable for key parameters such as cutterhead torque and soil pressure.

[0093] (3) The WiFi Mesh network is deployed inside the shield machine to form a local self-organizing network, supporting short-distance high-speed transmission from sensors to the control cabinet, ensuring real-time data processing.

[0094] This layered hybrid networking strategy ensures both long-distance coverage and high-speed transmission of critical data. For example, in hard rock tunneling projects, cutterhead torque data is rapidly transmitted to edge servers via WiFi Mesh. After initial processing, it is then transmitted back to the ground center in real time via the 5G network, enabling dynamic adjustment of tunneling parameters.

[0095] Data transmission systems typically have a three-tiered architecture, optimizing data quality and transmission efficiency layer by layer. 1. Front-end acquisition and edge computing: After analog-to-digital conversion, sensor data is first preprocessed in the edge server in the shield machine control cabinet. For example, cutterhead torque data (sampling rate 10 Hz) is subjected to Kalman filtering to eliminate noise, and the second-by-second mean and fluctuation are calculated. Only key characteristic values (such as peak torque and abnormal fluctuations) are compressed and transmitted, significantly reducing the data volume. This process, similar to "data slimming," ensures transmission efficiency while preserving core information. 2. Relay transmission and protocol adaptation: Processed data is transmitted using a transmission protocol based on priority. For example, emergency warning data (such as sudden drop in earth pressure) uses the MQTT protocol's "QoS 2" mode to ensure delivery, while routine data (such as daily progress) uses "QoS 0" mode to improve transmission efficiency. In one project, when the shield machine passes through a karst cave, the system automatically increases the sampling frequency of earth pressure data and prioritizes transmission over the 5G network, ensuring that the ground control center can adjust the earth bin pressure setpoint in a timely manner. 3. Integration of the ground center and the cloud: Once data reaches the ground, it connects to BIM models and geological databases via the OPC UA protocol. For example, the ground center dynamically calculates the cutterhead speed threshold by combining core compressive strength revealed by drilling with real-time torque data. This command is then transmitted back to the shield machine via the 5G network, forming a closed loop of "perception-analysis-decision-execution."

[0096] As an important guarantee for construction safety, various methods are used to improve the security and reliability of data transmission. 1. Encryption and Authentication: The AES-256 encryption algorithm is used to protect data privacy. Sensors and the ground center use a pre-shared key (PSK) for two-way authentication to prevent data tampering. For example, during construction in sensitive areas, all tool temperature data is transmitted encrypted to avoid leaking critical engineering information. 2. Anti-interference Technology: In areas with severe electromagnetic interference (such as near the cutterhead), hybrid fiber and wireless transmission is used, or frequency hopping technology (FHSS) is added to the wireless signal to avoid signal congestion. 3. Redundancy and Self-Healing Mechanism: Critical data is transmitted simultaneously through dual channels (such as LoRa + 5G). If a link is interrupted (such as a LoRa gateway failure), the system automatically switches to the backup channel. For example, the ground center continuously receives data through the 5G backup channel to ensure zero loss of excavation parameters.

[0097] S102 uses a convolutional neural network (CNN) algorithm to analyze geological parameters and identify stratigraphic features. It also constructs a disaster risk indicator system, using logistic regression and random forest methods to predict disaster risks based on geological models and construction parameters.

[0098] Specific implementation methods ( Figure 5 ), as shown below:

[0099] S301 uses the convolutional neural network (CNN) algorithm to analyze geological parameters and identify formation characteristics.

[0100] The application of CNN in geological parameter analysis is the core technology of stratum identification. By converting multi-source geological data (such as drilling core images, geophysical radar profiles, and geotechnical test parameters) into structured input, CNN can automatically extract stratum features and classify them ( Figure 6 ). The specific steps are as follows:

[0101] 1. Data preprocessing

[0102] Before using a CNN algorithm to process data, it needs to be standardized or normalized to improve model accuracy and accelerate convergence. Image data (such as rock core photos and geological radar scans) needs to be standardized to a uniform size (such as 256×256 pixels) and contrast enhanced to highlight texture features.

[0103]

[0104] Where μ is the mean of the image, X represents the image matrix, where

[0105]

[0106] Here, σ represents the standard deviation and N represents the number of pixels in image X.

[0107] For non-image data (such as core compressive strength and permeability coefficient), they are converted into the 0-1 interval through normalization processing (Min-Max or Z-score) to form a feature vector.

[0108] Min-Max normalization, also known as linear normalization or maximum-min normalization, scales the original data proportionally. By using the maximum and minimum values (or the maximum value) of a variable, the original data is converted to data within a specific range, eliminating the effects of dimension and magnitude and changing the weight of the variables in the analysis to address different metrics.

[0109]

[0110] Among them, X is the original data, X min and X max are the minimum and maximum values in the data set, respectively, and X norm is the normalized data.

[0111] 2. Geological model construction and training

[0112] An improved ResNet-50 network structure is used to analyze the strata, which includes convolutional layers, pooling layers, and fully connected layers. The first convolution kernel size is 3×3 and the stride is 1 to capture fine stratum textures.

[0113] After processing image and non-image data separately, features are fused using an attention mechanism (such as the SE module). The attention mechanism is a resource allocation mechanism. Neural networks have limited computational power. To improve accuracy within these constraints, more resources (called weights in neural networks) are added to identify important features of an object to improve recognition accuracy. SENet is a newly proposed network architecture that uses the Squeeze-and-Excitation module to model the interdependencies between convolutional feature channels and enhance network representation capabilities. This module includes global information embedding, adaptive recalibration, and feature map reweighting, enabling the network to dynamically adjust channel weights. Experiments have demonstrated that SENet achieves state-of-the-art performance on multiple datasets, demonstrating its effectiveness.

[0114] Using the cross-entropy loss function and the Adam optimizer, the model was iteratively trained on labeled datasets. For example, through joint training on 2,000 sets of rock core images and geophysical data, the model achieved a 92% accuracy rate in identifying sandstone, clay, and fractured zones.

[0115] Adam (Adaptive Moment Estimation) is a deep learning optimization algorithm that combines the ideas of momentum method and RMSProp. It adjusts the learning rate of each parameter by calculating the first-order moment estimate and second-order moment estimate of the gradient, thereby achieving more efficient network training.

[0116] The key to the Adam algorithm is to simultaneously calculate the exponential moving average of the first-order moment (mean) and the second-order moment (uncentered variance) of the gradient and perform bias correction on them to ensure that the gradient estimate does not deviate towards 0 in the early stages of training. Its update rule is as follows.

[0117] Compute the gradient at time step t:

[0118]

[0119] Among them, θ is the parameter to be optimized.

[0120] For time step t, given the gradient g t , the corresponding first-order moment m t and the second moment v t They are:

[0121]

[0122]

[0123] Where β1 and β2 are exponential decay rates that control the weight distribution (momentum and current gradient), and are usually close to 1.

[0124] To m t 、v t Make corrections to remove bias from the initial estimate:

[0125]

[0126]

[0127] Finally, the parameters are updated using the corrected moment values:

[0128]

[0129] In the formula, θ is the parameter to be optimized, α is the learning rate, and ε is a small value (usually 10) added to prevent division by zero. −8 ).

[0130] S302 updates the geological model in real time as the shield tunneling progresses, improving the accuracy of stratum identification.

[0131] As the shield tunneling machine advances, the model needs to be dynamically updated with newly collected geological parameters to avoid historical data bias. Every 10 meters, newly collected rock core images and earth pressure data are fed into the model, and the weights are gradually updated using online learning. To prevent catastrophic forgetting, the bottom convolutional layers (basic feature extraction layers) of the CNN are frozen, and only the top fully connected layers are fine-tuned. For example, during tunneling in hard rock, the model reduced the tool wear prediction error from 8% to 3% using 10 sets of new data.

[0132] A Bayesian neural network (BNN) module was introduced to assess the confidence of new data. When the model's confidence in the formation prediction falls below 70%, manual review and correction of the annotated data are triggered.

[0133] New model versions are generated daily and compared with historical data through cross-validation to ensure consistent performance. For example, in one project, after the model crossed a fault fracture zone, incremental learning improved the fracture zone recognition rate from 85% to 94%.

[0134] S303 constructs a disaster risk indicator system, using the Logistic regression algorithm and random forest method to predict disaster risks based on geological models and construction parameters.

[0135] Combining geological characteristics with construction parameters, a quantitative assessment system for disaster risk indicators such as landslides, mud bursts, and ground subsidence is formed.

[0136] The collapse risk indicator system includes the following: 1. Formation strength index: The ratio of the core compressive strength to the shield cutterhead torque. The lower the ratio, the more likely it is to collapse. 2. Support condition index: The risk increases significantly when the difference between the soil bin pressure and the formation water and soil pressure is less than 0.5 MPa.

[0137] Risk indicators of mud and water inrush include: 1. Permeability coefficient (K): obtained through pumping test, K> 10 -4 m / s is prone to water inrush. 2. Water pressure gradient (i): The ratio of the confined water head (H) to the excavation depth (h) (i = H / h). When i > 1, there is a risk of confined water inrush.

[0138] Ground collapse risk indicators include: 1. Shield tail gap (δ): When δ > 30 mm, the risk of separation between the segments and the formation increases. 2. Ground subsidence rate (S): Measured through subsidence monitoring points, an S > 5 mm / day requires an early warning.

[0139] Logistic regression and random forest algorithms are used to predict disaster risks based on geological models and construction parameters.

[0140] Logistic regression and random forest are two complementary prediction models, each suitable for different scenarios.

[0141] The logistic regression model is based on the logistic distribution. The parameters are directly estimated through the parameter estimation method to obtain P(Y|X).

[0142] Let X be a continuous random variable that follows a Logistic distribution with the following distribution function and density function:

[0143]

[0144]

[0145] Where μ is the location parameter and γ>0 is the shape parameter.

[0146] In order to predict the category based on all inputs, the sigmoid function is introduced:

[0147]

[0148] The input of the sigmoid function is recorded as z, and the linear model result is assigned to z, that is:

[0149]

[0150] If we use vector writing, the above formula can be written as:

[0151]

[0152] The vector X is the input data of the classifier, that is, the eigenvalue; the vector W is the optimal parameter, so that the classifier is as accurate as possible.

[0153] On this basis, a Logistic regression model is established:

[0154]

[0155] Where, is the input, is the output, is a parameter, and ω is called the weight vector.

[0156] Specifically, in disaster risk prediction, continuous variables such as collapse risk indicators, mud burst indicators, and ground subsidence indicators are used as input features. Regularization is used to prevent overfitting, and a 0-1 probability output (0 = low risk, 1 = high risk) is used to construct a disaster warning model. For example, the model's AUC for collapse risk reached 0.89. When the predicted probability exceeds 0.7, the system triggers an alert and automatically reduces the advance speed.

[0157] Random Forest is an ensemble learning method used for classification, regression, and other tasks. It improves prediction accuracy and controls overfitting by building multiple decision trees and combining their results. The basic steps of the Random Forest algorithm are as follows:

[0158] 1. Create multiple decision trees from the sample: Randomly sample with replacement (bootstrap sample) from the original dataset to generate a different training set for each tree.

[0159] 2. Select the best split point: At each node, instead of searching all features to find the best split point, we first randomly select a subset of features and then find the best split from this subset. This increases the diversity of the model and helps prevent overfitting.

[0160] 3. Build a decision tree: Recursively build a decision tree using the optimal split points determined in the previous steps. For classification problems, this is usually done until each leaf node is as "pure" as possible (i.e., contains almost identical class labels). For regression problems, the split may be stopped based on some criteria, such as reaching the minimum squared error or the depth limit of the tree.

[0161] 4. Voting mechanism (classification) or average calculation (regression): For each new data point, each decision tree will produce its own classification result (for classification problems) or numerical prediction (for regression problems). In classification problems, random forests use majority voting to determine the final output category; in regression problems, the final output is the average of all tree predictions.

[0162] In disaster risk prediction, the random forest model is suitable for handling nonlinear relationships and high-dimensional data, and is well-suited for complex strata. For example, input features include over 30 parameters (such as lithology, cutterhead torque, and groundwater level). By calculating the Gini impurity, key risk factors can be identified. For example, the model shows that the shield tail gap (δ) contributes 35% to ground collapse, making it a key monitoring parameter.

[0163] The outputs of Logistic regression and random forest are weighted averaged to comprehensively improve the prediction robustness.

[0164] S103 takes adjustment and optimization measures for the shield machine based on the predicted disaster risks, such as adjusting the cutter head, optimizing tunneling parameters, and implementing slag improvement plans.

[0165] For soft soil formations (which may increase the risk of ground collapse), a large-opening cutterhead design is used. The cutterhead opening ratio is increased to 45%-55%, reducing cutting resistance and preventing soft soil compression and ground uplift. For example, in subway projects, increasing the cutterhead opening ratio from 40% to 50% reduced cutterhead torque by 15%, keeping ground settlement within 10 mm. A foam injection system is also deployed. Foam parameters are set to a foam concentration of 3%-5%, an injection pressure of 0.3-0.5 MPa, and a foam volume ratio of 1:5 to 1:10. This reduces soil viscosity, improves fluidity, and prevents cake formation in the soil bin. For example, in clay formations, foam injection reduces cutterhead torque fluctuation from ±20% to ±5%.

[0166] For hard rock formations (which may increase the risk of ground collapse), a hybrid cutterhead configuration with disc cutters and tooth cutters is used to enhance rock breaking capabilities. Six to eight disc cutters (250-300 mm diameter) are positioned in the center of the cutterhead to break the primary rock formation; 12-16 alloy tooth cutters are positioned around the disc cutters to handle core residue or isolated boulders. For example, in a hard rock tunnel, disc cutters account for 60% and tooth cutters for 40%, improving excavation efficiency by 30%. A wear-resistant and cooling system is implemented, with grease injection ports added to the cutterhead's main bearings, with grease injection (10-15 L) every 50 meters of excavation progress. The pressure in the tool cooling water ring is maintained at 0.6-0.8 MPa to prevent tool overheating and wear.

[0167] For complex strata (where geological parameters fluctuate frequently, requiring dynamic adjustment of construction parameters to prevent sudden increases in risk), a modular cutterhead is used, allowing tool changes to adapt to stratum fluctuations. This modular design ensures rapid replacement of cutting tools (such as roller cutters, tooth cutters, and bionic cutters) in the cutterhead, with a switchover time of less than four hours. When the cutterhead torque fluctuation exceeds 20% and propulsion resistance increases suddenly, the system recommends switching cutterhead modules. For example, in interbedded soft and hard strata, the shield machine's modular cutterhead allows for rapid switching between sandstone and clay layers, increasing its average daily advance rate from 15 meters to 22 meters.

[0168] Dynamic soil bin pressure control is employed. High pressure is used for soft formations, set at 1.2 times the static earth pressure. For example, in clay formations with a static earth pressure of 0.8 MPa, the soil pressure is controlled at 0.96 ± 0.1 MPa. Low pressure is used for hard rock formations, reducing the soil pressure to 0.8 times the static earth pressure, thereby reducing friction between the cutterhead and the formation. A closed-loop regulation mechanism is employed, linking the screw conveyor speed to a PID controller. When the soil pressure deviates by ±5% from the set value, the screw conveyor speed is automatically adjusted by ±5 rpm.

[0169] The shield machine uses intelligent propulsion speed adjustment to adapt to uneven ground conditions. The standard speed is ≤30 mm / rev (approximately 20 meters / day). When the cutterhead torque suddenly increases by >30%, the speed automatically decreases to 15 mm / rev, triggering foam injection (increased to 5% concentration). When crossing a fault fracture zone, the shield machine reduces speed to 10 mm / rev and, in conjunction with bentonite injection (5% concentration), successfully prevents segment misalignment.

[0170] By controlling the amount of soil discharged, the soil pressure in the soil bin is maintained. The rotation speed is linked to the soil pressure, that is:

[0171]

[0172] Where n is the rotational speed, P_c is the soil bin pressure, and P_f is the ground pressure. K = 0.5 rpm / MPa. A limiting protection system is also implemented, limiting the screw conveyor speed to a maximum of 80 rpm to prevent soil sluices.

[0173] The system is designed with automatic deviation correction. When the soil pressure is continuously lower than the set value, the system automatically reduces the screw conveyor speed to 80% of the original value. When the soil is blocked, the screw conveyor speed is temporarily increased to 100 rpm (lasting for <30 seconds).

[0174] Foam injection is used to improve soil fluidity and reduce cutter head torque. For soft soil improvement, the foam injection rate is 0.5-1.0 L / m 3 For soil modification, the foam concentration is 3%-5%. For hard rock modification, the foam injection rate is 0.2-0.5 L / m³ at a concentration of 2%-3% to prevent cutterhead mud buildup. Intelligent foam injection is triggered by torque sensor data. When the average cutterhead torque exceeds 800 kN·m, foam injection (concentration increased to 5%) is automatically initiated.

[0175] Bentonite injection enhances the soil's watertightness and plasticity. Bentonite is mixed with water at a mass ratio of 1:10. The injection rate is adjusted dynamically based on the soil's moisture content: when the moisture content is <15%, the injection rate is 20-30 L / m³; when the moisture content is >25%, the injection rate is 10-15 L / m³. This improved soil's plasticity index (IP) increases to 12-18, and its shear strength increases by 20%-30%.

[0176] The injection rate of the modifier is automatically adjusted based on the soil's characteristics, using a dual-system collaborative control system. In alternating soft and hard strata, a "foam + bentonite" mixture is used for modification, with a ratio of 60% foam and 40% bentonite, ensuring both fluidity and water-stopping properties.

[0177] Early warning and emergency response measures are implemented during the construction of the S104 shield machine to ensure its safe and stable operation.

[0178] During shield machine construction, early warning systems and emergency response mechanisms are key to ensuring construction safety. By real-time monitoring of geological parameters, construction parameters, and environmental changes, issuing early warnings and implementing appropriate measures can effectively prevent disasters and ensure the safety of personnel and equipment.

[0179] 1. Set up early warning responses during the operation of the shield machine.

[0180] A yellow alert indicates a low disaster risk and requires enhanced monitoring. Trigger conditions are: 1. The soil silo pressure fluctuation range exceeds the set value by ±5% but is less than ±10%; 2. The cutterhead torque fluctuation rate exceeds ±15% but is less than ±25%; 3. The ground subsidence rate S is greater than 3 mm / day but less than 5 mm / day. Countermeasures: 1. Increase monitoring frequency: Increase the sampling frequency of key parameters such as soil pressure and cutterhead torque from once per hour to every 15 minutes. 2. Increase manual inspections: Add one manual inspection per shift, focusing on the condition of the soil in the silo and cutter wear.

[0181] An orange alert indicates a moderate disaster risk, requiring adjustments to construction parameters. Triggering conditions are: 1. Fluctuation in the earth bunker pressure exceeds ±10%; 2. Fluctuation in the cutterhead torque exceeds ±25%; 3. Ground subsidence rate S > 5 mm / day or cracks appear; 4. Shield tail gap δ > 30 mm and continues to increase. Countermeasures are as follows: 1. Adjust excavation parameters: Reduce advance speed to 60%-80% of the original value, increase earth bunker pressure to 1.2 times the static earth pressure, and increase foam injection to improve soil fluidity. 2. Activate the emergency plan: Notify the ground monitoring center and prepare to initiate ground reinforcement measures (such as grouting).

[0182] A red alert indicates a high risk of disaster and requires the activation of the emergency plan. Triggering conditions include: 1. Sudden drop in soil bin pressure exceeding ±20%; 2. Cutterhead torque exceeding the design limit (e.g., 1200 kN·m); 3. Ground subsidence rate (S) > 10 mm / d or significant collapse; 4. Discovery of a karst cave or cavity, with the roof depth less than 1.5 times the shield diameter. Response measures are as follows: 1. Immediately halt excavation: Shut down all power systems and ensure the cutterhead is no longer rotating to prevent further damage. 2. Activate the emergency plan: Select appropriate reinforcement measures (e.g., grouting, freezing, etc.) based on the specific situation, and evacuate construction personnel.

[0183] 2. Emergency measures during shield machine construction.

[0184] Adjust construction parameters, such as reducing the advance speed and adjusting the soil bunker pressure. When an abnormally high cutterhead torque is detected, gradually reduce the advance speed to 50%-70% of the original value to reduce the impact on the formation. For example, when excavating in hard rock, if the cutterhead torque approaches the limit, reduce the advance speed from 30 mm / rev to 15 mm / rev. Dynamically adjust the soil bunker pressure based on the formation characteristics. For soft formations, appropriately increase the soil bunker pressure to 1.2-1.3 times the static earth pressure; for hard rock formations, maintain the soil bunker pressure slightly below the static earth pressure.

[0185] Ground reinforcement: such as grouting and freezing methods. Grouting is suitable for unstable ground such as fractured zones and caves. Cement slurry or chemical slurry is injected through drilling holes to fill the ground voids and enhance ground stability. The grouting pressure is controlled at 0.5-1.0 MPa, and the grouting volume is estimated based on the ground porosity. The freezing method is suitable for ground with high water content and prone to collapse. Freezing pipes are arranged in front of the shield machine, and the ground is frozen into a solid frozen wall by circulating low-temperature brine. The freezing temperature is controlled at -20°C to -30°C, and the freezing time is determined by the thickness of the ground (usually 2-4 weeks).

[0186] Evacuate personnel: If the risk is extremely high, evacuate construction personnel and suspend construction. For extreme evacuation situations, if the system issues a red alert and confirms a major safety hazard, initiate evacuation procedures immediately. Notify all on-site personnel via the PA system to evacuate to a safe area immediately. Shut down all non-essential equipment and ensure a secure power supply. Establish a cordon to prohibit unauthorized personnel from entering the danger zone.

[0187] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A shield adaptive adjustment method based on stratum structure, characterized in that: The steps include: S101: Obtain geological parameters of the shield machine construction area through drilling and geophysical exploration, and place sensors at key locations of the shield machine to collect the shield machine construction parameters in real time; S102: Analyzing the geological parameters using a convolutional neural network (CNN) to identify stratum characteristics, and constructing a geological model based on the identified stratum characteristics; constructing a disaster risk indicator system, using logistic regression and random forest algorithms to predict disaster risks based on the geological model and construction parameters, wherein the disaster risks include landslides, mud bursts, and ground subsidence; S103: Dynamically adjust the cutterhead configuration of the shield machine, optimize excavation parameters, or adopt a slag improvement plan based on the predicted disaster risk. The excavation parameters include soil bin pressure and propulsion speed. S104: Through real-time monitoring of geological parameters, construction parameters and environmental changes, a multi-level early warning mechanism is used to trigger emergency response measures to ensure construction safety.

2. The method according to claim 1, characterized in that The geological parameters include rock and soil type, structure, construction, particle composition, density, humidity, groundwater type and permeability coefficient; the construction parameters include soil bin pressure, propulsion cylinder pressure, daily feed rate, screw conveyor torque and speed, roll angle and pitch angle of the shield center, shield tail gap, and tool temperature.

3. The method according to claim 1, characterized in that In step S101: During drilling, exploration holes are arranged along the tunnel axis, with the drilling spacing adjusted to 10-50 meters according to the complexity of the strata, and the drilling depth exceeds 1.5 times the tunnel burial depth; Geophysical exploration uses geological radar, high-density electrical method and seismic wave method to detect the distribution of caves and faults, and constructs a three-dimensional geological model through multi-source data fusion; including Sensors placed at key locations on the shield machine include a soil bin pressure sensor placed on the inner wall of the shield bin, a cutter head torque sensor integrated into the drive shaft, a temperature sensor embedded in the cutter body, a posture sensor installed in the center of the shield body, and a screw conveyor sensor installed on the screw conveyor. Data transmission adopts a hybrid networking strategy, combining LoRa and 5G / 4G dual-channel redundant transmission. The data is encrypted with AES-256 and synchronized with the BIM model in real time.

4. The method according to claim 2, characterized in that In the geophysical exploration: Geological radar detects shallow strata, high-density electrical methods generate two-dimensional / three-dimensional resistivity distribution maps to identify caves and faults, and seismic wave methods use dispersion curves to assess shear wave velocity in strata. The geophysical data are jointly corrected with the drilling data through the Bayesian inversion algorithm to generate a lithologic-physical property joint model.

5. The method according to claim 1, wherein In step S102: The convolutional neural network (CNN) uses an improved ResNet-50 network structure and combines the attention mechanism to fuse core images, geophysical profiles, and geotechnical parameter features. The geological model uses the output of weighted average logistic regression and random forest, with input features including lithology, cutterhead torque, and groundwater level, to identify key risk factors by calculating the Gini impurity.

6. The method according to claim 4, characterized in that In the random forest model: Analyze key risk factors through Gini impurity; The geological model is updated every 10 meters, and the fully connected layer is fine-tuned using online learning, and the data confidence is evaluated through a Bayesian neural network.

7. The method according to claim 1, characterized in that In step S103: For soft soil, use a cutterhead with an opening ratio of 45%-55%, and use foam injection to reduce cutterhead torque. The foam concentration is 3%-5%. For hard rock formations, a mixed configuration of roller cutters and tooth cutters is used, with roller cutters accounting for 60%, and a lubrication and cooling system is installed; The soil bin pressure is regulated in a closed loop by a PID controller, and the screw conveyor speed is automatically adjusted when the deviation exceeds ±5%.

8. The method according to claim 1, characterized in that The soil improvement plan includes: Inject foam into soft soil formations at 0.5-1.0 L / m³, and into hard rock formations at 0.2-0.5 L / m³; Bentonite is prepared at a mass ratio of 1:10, and the injection amount is dynamically adjusted to increase the plasticity index of the slag to 12-18.

9. The method according to claim 1, characterized in that In step S104: The yellow warning trigger conditions are soil bin pressure fluctuations of ±5%-10% and cutterhead torque fluctuations of ±15%-25%. Countermeasures include intensified monitoring and manual inspections. The triggering conditions for the orange warning are soil bin pressure fluctuations of more than ±10% or shield tail gaps greater than 30 mm. Countermeasures include reducing the speed to 60%-80% of the original value and grouting reinforcement. The red alert triggering conditions are cutter head torque exceeding the limit or ground settlement >10 mm / d. Response measures include shutting down the machine, freezing the ground, and evacuating personnel.

10. The method according to claim 9, characterized in that Among the emergency measures: Grouting reinforcement uses cement slurry with a grouting pressure of 0.5-1.0 MPa; The freezing method reduces the formation temperature to -20℃ to -30℃ through the circulation of low-temperature brine, and the freezing time is 2-4 weeks.

Images