Conductive field source-based seismic-electric combined excavation-following advanced forecasting method in TBM environment

Abstract

The invention provides a conductivity field source-based seismic-electric joint excavation-following advanced forecasting method in a TBM environment, and belongs to the technical field of excavation-following advanced forecasting. Constructing a training set through simulation data generated by transient electromagnetic forward modeling and tunnel earthquake advanced prediction forward modeling, and training the Wnet network based on the training set; the trained Wnet network is used for receiving transient electromagnetic data and tunnel seismic wave data and outputting an advanced geological forecast result; the loss function in the Wnet network training process comprises a model error of predicted resistivity and real resistivity, a model error of predicted wave impedance and real wave impedance, and an error of theoretical response and input response calculated by substituting a Wnet network prediction result into a transient electromagnetic forward equation and a tunnel seismic wave forward equation. According to the method, the limitation of a single geophysical method is overcome through multi-source information complementation, and the recognition capability of a complex geological disaster body is improved by utilizing the sensitivity of the TEM to water and the sensitivity of tunnel seismic exploration to a rock mass structure.

Description

Technical Field

[0001] This invention relates to the field of tunneling advance prediction technology, and in particular to a tunneling advance prediction method based on the combined seismic and electrical characteristics of conductive field sources in a TBM environment. Background Technology

[0002] In recent years, tunnel boring machines (TBMs), as efficient and safe mechanized tunnel construction equipment, have been widely used in transportation, water conservancy, mining, and other fields. During TBM construction, geological hazards such as fault fracture zones and high ground stress often arise in the geological formations ahead, which can easily lead to collapses and other engineering disasters, seriously threatening construction safety and progress. Therefore, achieving accurate advance detection of the geological formations ahead during TBM excavation and adopting different measures for different geological conditions are core technical requirements for ensuring project safety.

[0003] In existing methods, Transient Electromagnetic Transmission (TEM) is sensitive to low-resistivity bodies (such as aquifers), while Tunnel Seismic Wave (TSP) is sensitive to interfaces with impedance differences (such as faults and lithological boundaries), and the two are theoretically complementary. However, traditional single-method inversion suffers from multiple solutions, and the accuracy and reliability of data interpretation are difficult to guarantee in the environment of strong TBM interference. As a huge low-resistivity metallic body, the TBM generates extremely strong early interference signals in the transient electromagnetic response, with amplitudes nearly three orders of magnitude higher than the background response without the TBM. This strong interference mainly manifests as a significant low-resistivity false anomaly, which completely submerges the real, weak, and effective signals generated by aquifers (such as water-bearing faults) ahead of the tunnel face in the early stages. This makes it impossible to directly identify the real geological anomalies from the original attenuation curves, severely reducing the accuracy and reliability of the detection, and constituting a major technical bottleneck in advanced geological prediction for tunnel construction. Currently, there is a lack of a joint inversion and interpretation technique that can effectively integrate TEM and TSP data and adapt to the TBM environment. Traditional advanced geological prediction typically uses seismic (TSP) and electromagnetic (TEM) data separately or in series. For example, the initial resistivity model for TEM inversion is derived using tunnel seismic imaging results. This method relies on empirical rock physics relationships (such as the Faust equation relating velocity and resistivity) to convert seismic impedance into a resistivity profile. While an improvement over single methods, it has limitations: these relationships are regional rather than universal; and the non-uniqueness of TEM inversion remains prominent if the initial model is significantly biased. Overall, previous joint inversion frameworks either couple the data through fixed rock physics equations or impose structural similarity (such as cross-gradient constraints) between the respective inverted models. These methods reduce ambiguity to some extent, but require pre-defined coupling assumptions and struggle to fully exploit the complex and implicit correlations within the data.

[0004] Therefore, a method for advance prediction based on the combined seismic and electrical currents of conductive field sources in the TBM environment is needed. Summary of the Invention

[0005] In view of this, this invention provides a tunnel advance prediction method based on conductive field source seismic-electric joint inversion in the TBM environment, which uses physics-guided deep learning for tunnel advance prediction through TEM-TSP joint inversion. By coupling two heterogeneous geophysical exploration methods in a unified deep network, higher accuracy, less ambiguity, stronger noise resistance, and faster inversion output are achieved.

[0006] Therefore, the present invention provides the following technical solution: A method for advance prediction in a TBM environment based on combined seismic and electrical currents from conductive field sources includes: A training set was constructed using simulation data generated from transient electromagnetic forward modeling and tunnel seismic advance prediction forward modeling, and the Wnet network was trained based on the training set. The trained Wnet network is used to receive transient electromagnetic data and tunnel seismic wave data to output advanced geological prediction results. The loss function of the Wnet network training process includes: the model error between predicted resistivity and actual resistivity, the model error between predicted wave impedance and actual wave impedance, and the error between the theoretical response and the input response obtained by substituting the Wnet network prediction results into the transient electromagnetic forward modeling equation and the tunnel seismic wave forward modeling equation.

[0007] Furthermore, the transient electromagnetic transmitter is arranged on the ground directly above the tunnel axis, with two power supply electrodes arranged along the excavation direction as the transmission source; Multiple sets of measuring electrodes are arranged in a grid pattern around the transmitter to form a receiving array.

[0008] Furthermore, the tunnel seismic wave system utilizes a seismic wave exciter and receiver array integrated into the TBM itself or on the shield behind the cutterhead; The seismic source uses the vibration mode of the tunnel boring machine cutterhead to generate seismic waves; The detector array is installed at a preset interval inside the shield or on the inner wall of the tunnel segment to receive reflected seismic wave signals.

[0009] Furthermore, the training set constructed from the simulation data generated through transient electromagnetic forward modeling and tunnel seismic advance prediction forward modeling includes: Preprocessing of forward simulation data includes: The voltage data obtained from transient electromagnetic forward modeling is filtered and averaged, and then converted into an apparent resistivity-time curve. Bandpass filtering, first arrival picking, and wavefield separation were performed on the seismic wave data obtained from TSP forward modeling. Depth profiles reflecting wave impedance differences were obtained through velocity analysis and migration imaging. A training set is constructed based on the preprocessed data.

[0010] Furthermore, the prediction result corresponding to the response includes: The region with low resistivity and a significant decrease in wave impedance is a water-rich fracture zone. Regions with low resistivity but little change in wave impedance are relatively intact aquifers. The high impedance abrupt change interface is a dry fault or lithological interface.

[0011] Furthermore, the construction of the training set based on the preprocessed data includes: The TEM apparent resistivity curve and the corresponding TSP wave impedance profile at each measurement point are spatially aligned and gridded according to their geographical coordinates and detection depth to form a TEM and TSP data volume with spatial location information.

[0012] Advantages and positive effects of the present invention: (1) This method overcomes the limitations of a single geophysical method by complementing multi-source information. By utilizing the sensitivity of TEM to water and the sensitivity of TSP to rock mass structure, it significantly reduces the ambiguity of interpretation and improves the ability to identify complex geological hazards.

[0013] (2) This method effectively avoids strong electromagnetic interference and mechanical vibration interference inside the TBM cave by transmitting and receiving TEM on the ground; joint inversion can suppress some random noise and improve the signal-to-noise ratio through information fusion.

[0014] (3) This method adopts a physical-guided deep learning model, embedding physical laws as prior knowledge into the network to make the inversion results more reliable. Once the network is trained, the inversion accuracy is high and the efficiency is high, which can meet the timeliness requirements of TBM construction.

[0015] (4) This method makes full use of the TBM’s built-in TSP system. It can be implemented simply by adding a ground TEM system and developing joint inversion software. It is convenient to implement and easy to promote. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart of the advanced prediction method based on the combined seismic and electrical currents of conductive field sources in the TBM environment in an embodiment of the present invention.

[0018] Figure 2 This is a schematic diagram of the on-site monitoring setup in an embodiment of the present invention. Detailed Implementation

[0019] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0020] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0021] This invention provides a method for advance prediction based on combined seismic and electromagnetic data from conductive field sources in a TBM environment. A training set is constructed using simulation data generated by transient electromagnetic forward modeling and tunnel seismic advance prediction forward modeling, and the Wnet network is trained based on the training set. The trained Wnet network receives transient electromagnetic data and tunnel seismic wave data and outputs advance geological prediction results.

[0022] Example 1 The hardware implementation of a TBM-based method for advance prediction based on conductive field source seismic-electrical co-current method includes: 1. TEM system: It consists of a transmitting system (dual electrode configuration), a receiving system (surface electrode network) and a data acquisition system (high-speed ADC), used to inject a primary field into the ground and receive the secondary field attenuation signal.

[0023] 2. TSP System: The TBM (tunnel boring machine) uses a seismic wave excitation and reception device, either built into the machine or integrated into it, to collect seismic wave reflection signals.

[0024] 3. Synchronization control unit: GPS or high-precision crystal oscillator is used to achieve high-precision time synchronization of TEM and TSP data acquisition.

[0025] The process of using a combined conductive field-source seismic-electric prediction method for tunnel boring machine (TBM) environments includes: 1. Data preprocessing and feature extraction: Denoise and filter the raw TEM data to extract multi-time channel voltage or apparent resistivity curves; Denoise, first arrival picking, and velocity analysis are performed on the TSP data to extract features such as reflected wave amplitude and travel time.

[0026] 2. Multimodal data fusion: Spatially register TEM response data with TSP feature data to construct a time-space corresponding multi-physical quantity dataset, which serves as the input to the WNet network.

[0027] 3. Physics-Guided Deep Learning Joint Inversion: A deep learning network based on the WNet architecture is used as the inversion engine. The fused multi-physical quantity data is input into the network, which simultaneously outputs a subsurface resistivity model and a wave impedance model. The network training employs a physics-constrained loss function, embedding constraints from the electromagnetic diffusion equation and the seismic wave equation as regularization terms into the loss function to ensure that the inversion results conform to physical laws.

[0028] 4. Geological interpretation and 3D visualization: Based on the resistivity and wave impedance models obtained from the inversion, the nature, location and scale of the adverse geological bodies are comprehensively inferred, and a 3D visualization result map is generated.

[0029] Specifically, the data processing flow is as follows: Step 1: During TBM tunneling, two data acquisition tasks are carried out simultaneously: Step 1-1, TEM data acquisition: A conductive field source, namely two transmitting electrodes and a two-dimensional array receiving electrode, is arranged on the ground directly above the tunnel axis. A primary pulse electromagnetic field is injected into the ground, and a secondary field attenuation signal is received.

[0030] Steps 1-2: TSP Data Acquisition: Using the TSP system built into the TBM shield machine, seismic waves are generated and reflected wave signals from the front of the tunnel are received.

[0031] Steps 1-3: Employ high-precision time synchronization technology to control the TEM and TSP systems to start data acquisition synchronously, ensuring the comparability of data in time and space.

[0032] Step 2: Data Preprocessing and Fusion Step 2-1: Perform outlier removal, noise suppression (such as power frequency filtering), and signal enhancement processing on the acquired raw TEM data, and calculate the apparent resistivity-time curve for each measurement point.

[0033] Step 2-2: Preprocess the acquired raw TSP data, including bandpass filtering, first arrival picking, velocity analysis, extraction of reflected wave amplitude, travel time and other features, and pre-stack migration imaging to obtain a preliminary wave impedance difference profile.

[0034] Steps 2-3: Based on the location and depth range of the measuring points, spatially register the apparent resistivity-time curves to construct a paired dataset of "TEM-TSP response-subsurface model" for joint inversion.

[0035] Step 3: Physics-guided deep learning joint inversion, the process is as follows: Figure 1 As shown: Step 3-1: Construct the training dataset. By establishing a large number of forward models covering different geological conditions (such as layered models with different thicknesses, resistivity, and wave impedance, including faults, aquifers, and other anomalous bodies), calculate their TEM response and TSP response respectively, and generate a large-scale sample library of "subsurface model - TEM response - TSP response" for network training.

[0036] Step 3-2: Train the WNet network.

[0037] A W-shaped network, denoted as WNet, is constructed by cascading two UResNets, representing the existing network structure.

[0038] The WNet network employs a dual encoder-dual decoder structure (W-type), with one encoding branch inputting TEM response data and the other encoding branch inputting TSP response data; feature fusion is performed in the intermediate layers of the network; the decoder branch one outputs the predicted resistivity model, and the decoder branch two outputs the predicted wave impedance model.

[0039] Step 3-3: Embedding Physical Constraints. During the network training phase, the loss function consists of three parts: (1) Model error between predicted resistivity and actual resistivity; (2) Model error between predicted wave impedance and actual wave impedance; (3) Physical equation constraint error (the error between the theoretical response and the input response obtained by substituting the network prediction model into the TEM forward equation and the TSP forward equation).

[0040] The loss function guides the network to learn a mapping relationship that simultaneously satisfies data fitting and the inversion of physical laws.

[0041] Steps 3-4: Perform the inversion. Input the field fusion data prepared in Steps 2-3 into the trained WNet network. The network quickly outputs the resistivity distribution model and wave impedance distribution model in front of the tunnel.

[0042] Step 4: Result Verification and Geological Interpretation: The resistivity anomaly areas obtained from the inversion are superimposed and compared with the acoustic impedance anomaly areas for mutual verification. Low resistivity anomalies usually indicate water content, while abrupt changes in acoustic impedance indicate lithological boundaries or fractures. Combining the geological data of the region, the geological attributes of the anomalies (such as water-rich fault fracture zones, dry fracture zones, etc.) are comprehensively determined, and finally, an advanced prediction map is drawn and its risk level is assessed.

[0043] Example 2 Based on the aforementioned hardware and data processing flow, a method for advance prediction under a TBM environment based on conductive field-source seismic-electrical co-intrusion is proposed. This method employs an electrode-type device to inject a primary electromagnetic field into the subsurface medium using the ground transient electromagnetic method, simultaneously acquiring TSP seismic data. A physics-guided deep learning network is used for joint seismic-electrical inversion to achieve accurate extraction of geological information. Specifically, the method includes the following steps: Figure 1 As shown: Step 1: Hardware Deployment and Data Acquisition During TBM tunneling, hardware deployment and data acquisition are carried out simultaneously: 1-1. TEM (Transient Electromagnetic Transmitter) System Layout: (e.g., ...) Figure 2 As shown, two power supply electrodes (A and B poles) are deployed on the ground directly above the tunnel axis along the excavation direction as transmitters, with the electrode spacing determined to be 20-50 meters based on the required detection depth. Around the transmitters, multiple sets of measuring electrodes (M and N poles) are arranged in a grid pattern to form a receiving array. A high-power transmitter injects a bipolar square wave current greater than 10A into the AB poles. A high-speed acquisition station synchronously records the transient electromagnetic response voltage decay curves of all MN pole pairs.

[0044] 1-2. TSP System Layout: Utilizes an array of seismic wave generators (sources) and receivers (detectors) integrated into the TBM itself or temporarily installed on the shield behind the cutterhead. The source generates seismic waves through the vibration of the TBM cutterhead, and the detector array is installed at preset intervals inside the shield or on the inner wall of the tunnel segments to receive reflected seismic wave signals.

[0045] 1-3. Synchronous acquisition: The trigger signals of the TEM transmitter and the TSP source are controlled by the same high-precision GPS synchronization host to ensure that the clock references of the two data acquisitions are consistent, thereby ensuring that the data can be correlated in the time dimension.

[0046] Step 2: Data Preprocessing and Fusion The collected raw data is transmitted to the data processing center for the following processing: 2-1. TEM data processing: The raw voltage data is filtered (to remove power frequency interference, etc.), averaged (to improve the signal-to-noise ratio), and then converted into an apparent resistivity-time curve.

[0047] 2-2. TSP Data Processing: Bandpass filtering, first arrival picking, and wavefield separation are performed on the seismic records. Through velocity analysis and migration imaging, depth profiles reflecting wave impedance differences are obtained.

[0048] 2-3. Data Fusion: The TEM apparent resistivity curve and the corresponding TSP wave impedance profile at each measurement point are spatially aligned and gridded according to their geographical coordinates and detection depth to form a TEM and TSP data volume with spatial location information, which serves as the input to the joint inversion model.

[0049] Step 3: Physics-guided deep learning joint inversion 3-1 Training Data Preparation: Based on the geological background of the work area, a large number of geoelectric models (forward models) containing different layer thicknesses, typical geological hazard bodies (such as faults and water-bearing chambers), and resistivity are constructed, as well as seismic forward models containing different layer thicknesses, typical geological hazard bodies (such as faults and water-bearing chambers), and wave impedance.

[0050] The TEM response of each model is calculated using the finite element method forward modeling, and its TSP response is calculated using the finite difference method forward modeling, generating a large-scale sample library of "model-response" pairs.

[0051] 3-2. Network training: using methods such as... Figure 1 The WNet network structure shown is used for training. TEM and TSP response data from the paired sample library are simultaneously input into the network, and the network outputs predicted resistivity and wave impedance models. The loss function for the training process is defined as:

[0052] in, , and These are preset parameters; The mean squared error between the TEM prediction and the true model; Loss_physics is the mean squared error between the TSP prediction and the true model. To address the error between the theoretical response calculated by substituting the network output model into the forward modeling equations and the input response, the physical laws governing the propagation of electromagnetic and seismic waves are embedded into the network training. The network parameters are optimized using a backpropagation algorithm until the loss function converges.

[0053] 3-3. Field data inversion: Input the field fusion data processed in step 2 into the trained WNet network. After the network propagates forward, it can quickly output the resistivity distribution and wave impedance distribution results of the area in front of the tunnel.

[0054] Step 4, Result Interpretation and Output: 4-1. Comprehensive Interpretation: Compare and analyze the resistivity model and wave impedance model obtained from the inversion. Regions with low resistivity and significantly reduced wave impedance can be interpreted as water-rich fracture zones; regions with low resistivity but little change in wave impedance may be relatively intact water-bearing strata; interfaces with abrupt changes in high wave impedance may be dry faults or lithological boundaries.

[0055] 4-2. Results Generation: Using 3D visualization software, the inversion results are used to generate resistivity-wave impedance joint profile map along the tunnel axis, 3D geological model map and other results maps, and the location, scale and nature of adverse geological bodies are marked. Finally, an advanced geological prediction report is generated to guide TBM construction decisions.

[0056] This method combines TEM and TSP data acquisition with a physics-guided deep learning joint inversion model to fuse and interpret multi-source geophysical data, ultimately achieving comprehensive and advanced prediction of the geological structure and hydrogeological conditions ahead of the tunnel. Training the network on large-scale synthetic samples including faults and water-rich bodies enables it to recognize complex patterns. Once training is complete, the inversion of new field data can output resistivity and impedance profiles in near real-time, which is crucial for rapid TBM construction decisions. Traditional (even joint) inversion iterations are time-consuming and do not match the pace of TBM advancement; the deep learning solution fills this gap, achieving a dual improvement in high fidelity and timeliness. Furthermore, the solution fully utilizes the TBM's built-in TSP and easily deployable surface TEM, and uses GPS for high-precision synchronization, enabling true fusion of the two types of data based on spatiotemporal alignment. The surface TEM can also avoid strong electromagnetic noise within the TBM, while feature fusion at the network level helps to suppress and correct noise, resulting in a higher signal-to-noise ratio and more reliable anomaly identification.

[0057] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for advance prediction in a TBM environment based on combined seismic and electrical currents from conductive field sources, characterized in that, include: A training set was constructed using simulation data generated from transient electromagnetic forward modeling and tunnel seismic advance prediction forward modeling, and the Wnet network was trained based on the training set. The trained Wnet network is used to receive transient electromagnetic data and tunnel seismic wave data to output advanced geological prediction results. The loss function of the Wnet network training process includes: the model error between predicted resistivity and actual resistivity, the model error between predicted wave impedance and actual wave impedance, and the error between the theoretical response and the input response obtained by substituting the Wnet network prediction results into the transient electromagnetic forward modeling equation and the tunnel seismic wave forward modeling equation.

2. The method according to claim 1, characterized in that, The transient electromagnetic transmitter is positioned on the ground directly above the tunnel axis, with two power supply electrodes arranged along the excavation direction as the transmission source. Multiple sets of measuring electrodes are arranged in a grid pattern around the transmitter to form a receiving array.

3. The method according to claim 1, characterized in that, The tunnel seismic wave system utilizes a seismic wave exciter and receiver array integrated into the TBM itself or on the shield body behind the cutterhead; The seismic source uses the vibration mode of the tunnel boring machine cutterhead to generate seismic waves; The detector array is installed at a preset interval inside the shield or on the inner wall of the tunnel segment to receive reflected seismic wave signals.

4. The method according to claim 1, characterized in that, The training set, constructed from simulation data generated through transient electromagnetic forward modeling and tunnel seismic advance prediction forward modeling, includes: Preprocessing of forward simulation data includes: The voltage data obtained from transient electromagnetic forward modeling is filtered and averaged, and then converted into an apparent resistivity-time curve. Bandpass filtering, first arrival picking, and wavefield separation were performed on the seismic wave data obtained from TSP forward modeling. Depth profiles reflecting wave impedance differences were obtained through velocity analysis and migration imaging. A training set is constructed based on the preprocessed data.

5. The method according to claim 1, characterized in that, The prediction results corresponding to the response include: The region with low resistivity and a significant decrease in wave impedance is a water-rich fracture zone. Regions with low resistivity but little change in wave impedance are relatively intact aquifers. The high impedance abrupt change interface is a dry fault or lithological interface.

6. The method according to claim 1, characterized in that, The construction of the training set based on the preprocessed data includes: The TEM apparent resistivity curve and the corresponding TSP wave impedance profile at each measurement point are spatially aligned and gridded according to their geographical coordinates and detection depth to form a TEM and TSP data volume with spatial location information.

Images