A prediction method for shield tunneling through karst caves in karst areas

Abstract

The present invention discloses a method for predicting karst caves in the karst area during shield tunneling, belonging to the technical field of karst cave prediction. In the present invention, reflection waves in the karst area are collected by geophones during the operation of the shield machine, and the instantaneous frequency, instantaneous amplitude, and instantaneous phase of P-waves and S-waves are calculated respectively, reflecting the changes in frequency, amplitude, and phase at each moment. By capturing the phase reversal coefficient, amplitude change coefficient, frequency change coefficient, and energy attenuation coefficient of P-waves, as well as the amplitude change coefficient and energy attenuation coefficient of S-waves, the response characteristics of P-waves and S-waves to karst caves can be effectively reflected. Combining these characteristic coefficients with the processing of the karst cave prediction model can further improve the accuracy of karst cave prediction and provide a strong guarantee for the safety of shield construction.

Description

Technical Field

[0001] The present invention relates to the technical field of karst cave prediction, and particularly relates to a method for predicting karst caves during shield tunneling through a karst area. Background Art

[0002] With the continuous development of urban underground engineering construction, shield tunneling construction has become increasingly common in the construction of infrastructure such as transportation and subways. However, during the shield construction process, encountering complex geological conditions such as karst areas and karst caves often brings potential safety hazards, increased construction difficulty, and economic losses. Therefore, accurately predicting the distribution of karst caves in karst areas is of great significance for ensuring the construction safety and construction efficiency of shield projects.

[0003] Karst areas are special geological structures formed by soluble rocks (such as limestone, dolomite, etc.) under the erosion of water, with obvious spatial variations and inhomogeneities. The existence of karst caves can lead to risks such as weakened soil strength, settlement, and collapse. Traditional geological exploration methods, such as drilling and ground penetrating radar, usually have problems such as high cost, long cycle, and difficulty in comprehensive coverage, and are difficult to meet the real-time and accuracy requirements of shield construction.

[0004] In recent years, geological information acquisition technologies based on seismic reflection method detection have been gradually applied. By analyzing the signal components of the reflected wave signals collected by geophones, geological information is obtained. However, in karst areas with obvious spatial variations and inhomogeneities, it is difficult to fully reflect the location of karst caves from the signal components, resulting in the problem of low prediction accuracy of karst caves. Summary of the Invention

[0005] Aiming at the above deficiencies in the prior art, the present invention provides a method for predicting karst caves during shield tunneling through a karst area, which solves the problem of low prediction accuracy of karst caves existing in the prior art.

[0006] In order to achieve the above invention purpose, the technical solution adopted by the present invention is: a method for predicting karst caves during shield tunneling through a karst area, comprising the following steps:

[0007] S1. Collect the reflected waves in the karst area when the shield machine is working through a geophone;

[0008] S2. Calculate the instantaneous frequency, instantaneous amplitude, and instantaneous phase of the P-wave and S-wave in the reflected wave respectively;

[0009] S3. Extract the phase reversal coefficient, amplitude change coefficient, frequency change coefficient, and energy attenuation coefficient according to the instantaneous phase, instantaneous amplitude, and instantaneous frequency at multiple consecutive moments in the P-wave;

[0010] S4. Extract the amplitude change coefficient and energy attenuation coefficient according to the instantaneous amplitude at multiple connected moments in the S-wave;

[0011] S5. According to the karst cave prediction model, process the phase inversion coefficient, amplitude change coefficient, frequency change coefficient, and energy attenuation coefficient of the P wave, as well as the amplitude change coefficient and energy attenuation coefficient of the S wave, to obtain the predicted distance of the karst cave.

[0012] Further, the extraction of the phase inversion coefficient in S3 includes the following steps:

[0013] A1. Among consecutive multiple moments, subtract the instantaneous phase of the latter moment from the instantaneous phase of the former moment in two adjacent moments, take the absolute value to obtain the phase difference, and add up multiple phase differences to obtain the result of adding up the phase differences;

[0014] A2. Among consecutive multiple moments, subtract the maximum instantaneous phase from the minimum instantaneous phase to obtain the maximum difference;

[0015] A3. Add up the maximum difference and the result of adding up the phase differences, and normalize to obtain the phase inversion coefficient.

[0016] Further, the specific process of extracting the amplitude change coefficient in S3 and S4 is as follows: Among consecutive multiple moments, subtract the instantaneous amplitude of the latter moment from the instantaneous amplitude of the former moment in two adjacent moments, take the absolute value to obtain the amplitude difference, add up multiple amplitude differences to obtain the result of adding up the amplitude differences, and perform normalization processing on the result of adding up the amplitude differences to obtain the amplitude change coefficient;

[0017] The specific process of extracting the frequency change coefficient in S3 is as follows: Among consecutive multiple moments, subtract the instantaneous frequency of the latter moment from the instantaneous frequency of the former moment in two adjacent moments, take the absolute value to obtain the frequency difference, add up multiple frequency differences to obtain the result of adding up the frequency differences, and perform normalization processing on the result of adding up the frequency differences to obtain the frequency change coefficient.

[0018] Further, the specific process of extracting the energy attenuation coefficient in S3 and S4 is as follows: Among consecutive multiple moments, square the instantaneous amplitude, accumulate the squared instantaneous amplitudes to obtain the energy, and in the time length N, take the normalized value of the ratio of the maximum energy to the minimum energy as the energy attenuation coefficient, where N is a positive integer.

[0019] Further, the karst cave prediction model in S5 includes: an S-wave feature extraction unit, a P-wave feature extraction unit, a first Concat layer, a transpose operation layer, a multiplier M1, a Conv layer, a dual-channel feature enhancement unit, and a fully connected layer;

[0020] The input end of the S-wave feature extraction unit is used to input the amplitude change coefficient and energy attenuation coefficient of the S-wave; the input end of the P-wave feature extraction unit is used to input the phase inversion coefficient, amplitude change coefficient, frequency change coefficient and energy attenuation coefficient of the P-wave; the input end of the first Concat layer is respectively connected to the output ends of the S-wave feature extraction unit and the P-wave feature extraction unit, and its output end is respectively connected to the input end of the transpose operation layer and the first input end of the multiplier M1; the second input end of the multiplier M1 is connected to the output end of the transpose operation layer, and its output end is connected to the input end of the Conv layer; the input end of the dual-channel feature enhancement unit is connected to the output end of the Conv layer, and its output end is connected to the input end of the fully connected layer; the output end of the fully connected layer is used as the output end of the karst cave prediction model.

[0021] Further, the S-wave feature extraction unit includes: an amplitude processing channel and an energy processing channel;

[0022] In the S-wave feature extraction unit, the amplitude processing channel is used to process the amplitude change coefficient of the S-wave, and the energy processing channel is used to process the energy attenuation coefficient of the S-wave;

[0023] The P-wave feature extraction unit includes: a phase processing channel, an amplitude processing channel, a frequency processing channel and an energy processing channel;

[0024] In the P-wave feature extraction unit, the phase processing channel is used to process the phase inversion coefficient of the P-wave, the amplitude processing channel is used to process the amplitude change coefficient of the P-wave, the frequency processing channel is used to process the frequency change coefficient of the P-wave, and the energy processing channel is used to process the energy attenuation coefficient of the P-wave.

[0025] Further, the phase processing channel, the amplitude processing channel, the frequency processing channel and the energy processing channel all include: a Tanh activation layer, a Sigmoid activation layer and a multiplier M2;

[0026] The Tanh activation layer is used to assign weights and biases to the coefficients input to the phase processing channel, amplitude processing channel, frequency processing channel or energy processing channel to obtain a first eigenvalue; the Sigmoid activation layer is used to assign weights and biases to the coefficients input to the phase processing channel, amplitude processing channel, frequency processing channel or energy processing channel to obtain a second eigenvalue; the multiplier M2 is used to multiply the first eigenvalue and the second eigenvalue to obtain the channel output.

[0027] Further, the first Concat layer is used to splice the output features of the S-wave feature extraction unit and the output features of the P-wave feature extraction unit to obtain a spliced vector a, and the transpose operation layer is used to perform a transpose operation on the spliced vector a to obtain an operation result a T, the multiplier M1 is used to multiply the operation result a T by the splicing vector a to obtain the fusion matrix A, A = a T a, where T is the transpose operation.

[0028] Further, the dual-channel feature enhancement unit includes: a GAP layer, a GMP layer, a mean feature enhancement layer, a significant feature enhancement layer, and a second Concat layer;

[0029] The input ends of the GAP layer and the GMP layer are both used as the input end of the dual-channel feature enhancement unit;

[0030] The input end of the mean feature enhancement layer is connected to the output end of the GAP layer;

[0031] The input end of the significant feature enhancement layer is connected to the output end of the GMP layer;

[0032] The input ends of the second Concat layer are respectively connected to the output ends of the mean feature enhancement layer and the significant feature enhancement layer, and its output end is used as the output end of the dual-channel feature enhancement unit.

[0033] Further, the mean feature enhancement layer is used to enhance the mean feature output by the GAP layer;

[0034] The significant feature enhancement layer is used to enhance the maximum feature output by the GMP layer.

[0035] The beneficial effects of the present invention are as follows: By using a geophone to collect the reflected waves in the karst area during the operation of the shield machine, the instantaneous frequency, instantaneous amplitude, and instantaneous phase of the P-wave and S-wave are calculated respectively, reflecting the changes in frequency, amplitude, and phase at each moment. In a karst cave, the P-wave shows an obvious frequency decrease and phase inversion. The reason is that the karst cave is usually filled with air or water, forming a significant acoustic impedance difference with the surrounding rocks. This difference leads to a high reflection coefficient, resulting in a large signal intensity returned from the karst cave boundary. At the same time, the S-wave can only propagate in solids and will undergo strong energy conversion and attenuation when encountering a fluid-filled karst cave. Therefore, capturing the phase inversion coefficient, amplitude change coefficient, frequency change coefficient, and energy attenuation coefficient of the P-wave, as well as the amplitude change coefficient and energy attenuation coefficient of the S-wave, can effectively reflect the response characteristics of the P-wave and S-wave to the karst cave. Combining these characteristic coefficients with the processing of the karst cave prediction model can further improve the accuracy of karst cave prediction and provide a strong guarantee for the safety of shield construction. Description of the Drawings

[0036] Figure 1 is a flowchart of a method for predicting karst caves during shield tunneling through a karst area;

[0037] Figure 2Schematic structural diagram of the karst cave prediction model;

[0038] Figure 3 Schematic structural diagram of the S-wave feature extraction unit;

[0039] Figure 4 Schematic structural diagram of the P-wave feature extraction unit;

[0040] Figure 5 Schematic structural diagram of the phase processing channel, amplitude processing channel, frequency processing channel or energy processing channel. Detailed implementation manners

[0041] The following describes the detailed implementation manners of the present invention to facilitate those skilled in the art of the present technology to understand the present invention. However, it should be clear that the present invention is not limited to the scope of the detailed implementation manners. For those of ordinary skill in the art of the present technology, as long as various changes are within the spirit and scope of the present invention defined and determined by the appended claims, these changes are obvious, and all inventions and creations using the concept of the present invention are within the scope of protection.

[0042] As Figure 1 shown, a method for predicting karst caves during shield tunneling through a karst area includes the following steps:

[0043] S1. Collect the reflected waves in the karst area when the shield machine is working through a geophone;

[0044] S2. Calculate the instantaneous frequency, instantaneous amplitude and instantaneous phase of the P-wave and S-wave in the reflected waves respectively;

[0045] S3. Extract the phase reversal coefficient, amplitude change coefficient, frequency change coefficient and energy attenuation coefficient according to the instantaneous phase, instantaneous amplitude and instantaneous frequency at multiple consecutive moments in the P-wave;

[0046] S4. Extract the amplitude change coefficient and energy attenuation coefficient according to the instantaneous amplitude at multiple consecutive moments in the S-wave;

[0047] S5. Process the phase reversal coefficient, amplitude change coefficient, frequency change coefficient and energy attenuation coefficient of the P-wave, and the amplitude change coefficient and energy attenuation coefficient of the S-wave according to the karst cave prediction model to obtain the predicted distance of the karst cave.

[0048] The steps for extracting the phase reversal coefficient in step S3 include the following:

[0049] A1. In multiple consecutive moments, subtract the instantaneous phase of the latter moment from the instantaneous phase of the former moment in two adjacent moments, take the absolute value to obtain the phase difference, and add up the multiple phase differences to obtain the result of adding up the phase differences;

[0050] A2. Subtract the minimum instantaneous phase from the maximum instantaneous phase at consecutive moments to obtain the maximum difference.

[0051] A3. Add the maximum difference and the result of adding the phase differences, and then normalize to obtain the phase inversion coefficient.

[0052] In this embodiment, the calculation formula for the phase inversion coefficient is: where α t is the phase inversion coefficient at the t-th moment in the P-wave, arctan is the arctangent function, P τ is the instantaneous phase of the P-wave at the τ-th moment, P τ-1 is the instantaneous phase of the P-wave at the (τ - 1)-th moment, max is to take the maximum value, min is to take the minimum value, || is the absolute value, t s is the starting time, t is the current moment, and τ is the superimposed variable.

[0053] In the presence of a karst cave, when the P-wave passes through the boundary of the karst cave, an obvious phase inversion phenomenon will occur. By comparing the phases at adjacent moments, the change in phase can be accurately identified, thus reflecting the presence of the karst cave.

[0054] The specific process of extracting the amplitude change coefficient in S3 and S4 is as follows: At consecutive moments, subtract the instantaneous amplitude of the latter moment from the instantaneous amplitude of the former moment among two adjacent moments, and take the absolute value to obtain the amplitude difference. Add multiple amplitude differences to obtain the result of adding the amplitude differences, and perform normalization processing on the result of adding the amplitude differences to obtain the amplitude change coefficient.

[0055] In this embodiment, the calculation formula for the amplitude change coefficient is: where β t is the amplitude change coefficient at the t-th moment, F τ is the instantaneous amplitude at the τ-th moment, F τ-1 is the instantaneous amplitude at the (τ - 1)-th moment, || is the absolute value operation, t s is the starting time, t is the current moment, τ is the superimposed variable, and arctan is the arctangent function.

[0056] When the P-wave and S-wave encounter a karst cave, due to the sudden change in acoustic impedance, the amplitude of the wave will change significantly. The P-wave will generate a strong reflection at the boundary of the karst cave, resulting in a sharp increase in amplitude; while the S-wave will undergo strong energy conversion and attenuation when encountering a fluid-filled karst cave, resulting in a sharp decrease in amplitude. The present invention calculates the amplitude differences at consecutive moments, accumulates and normalizes these differences, and obtains an index that can quantitatively describe the degree of amplitude change.

[0057] The specific process of extracting the frequency change coefficient in S3 is as follows: In a series of consecutive moments, subtract the instantaneous frequency at the latter moment from the instantaneous frequency at the former moment among two adjacent moments, and take the absolute value to obtain the frequency difference. Add up multiple frequency differences to get the sum of frequency differences. Normalize the sum of frequency differences to obtain the frequency change coefficient.

[0058] In this embodiment, the calculation formula for the frequency change coefficient is: where γ t is the frequency change coefficient at the t-th moment, f τ is the instantaneous frequency at the τ-th moment, f τ-1 is the instantaneous frequency at the (τ - 1)-th moment, || represents the absolute value operation, t s is the starting time, t is the current moment, τ is the superimposed variable, and arctan is the arctangent function.

[0059] When the P-wave passes through the karst cave, since the karst cave is usually filled with air or water, forming a significant acoustic impedance difference with the surrounding rocks, this difference causes a significant decrease in the P-wave frequency. By calculating the instantaneous frequency change at consecutive moments, this frequency decrease phenomenon can be effectively captured, and the frequency change coefficient is extracted as a key index.

[0060] The specific process of extracting the energy attenuation coefficient in S3 and S4 is as follows: In a series of consecutive moments, square the instantaneous amplitude, accumulate the squared instantaneous amplitudes to obtain the energy. In the time length N, take the normalized value of the ratio of the maximum energy to the minimum energy as the energy attenuation coefficient, where N is a positive integer.

[0061] In this embodiment, the calculation formula for the energy attenuation coefficient is: where θ t is the energy attenuation coefficient at the t-th moment, max is to take the maximum value, min is to take the maximum value, F τ is the instantaneous amplitude at the τ-th moment, t s is the starting time, t is the current moment, τ is the superimposed variable, E t is the energy at the t-th moment, E t-i is the energy at the (t - i)-th moment, E t-N is the energy at the (t - N)-th moment, i is a positive integer, N is the time length, N is a positive integer, and arctan is the arctangent function.

[0062] When P-waves and S-waves propagate to the karst cave area, due to the significant difference in acoustic impedance, the energy of the waves will undergo obvious attenuation. Specifically, the P-wave will maintain a relatively high energy due to strong reflection at the karst cave boundary, while the S-wave will experience significant energy loss when encountering a fluid-filled karst cave. The characteristics of this energy change are quantified by calculating the energy attenuation coefficient at consecutive moments, which can effectively reflect the existence and characteristics of the karst cave.

[0063] As Figure 2 shown, the karst cave prediction model in S5 includes: an S-wave feature extraction unit, a P-wave feature extraction unit, a first Concat layer, a transpose operation layer, a multiplier M1, a Conv layer, a dual-channel feature enhancement unit, and a fully connected layer;

[0064] The input end of the S-wave feature extraction unit is used to input the amplitude change coefficient and energy attenuation coefficient of the S-wave; the input end of the P-wave feature extraction unit is used to input the phase inversion coefficient, amplitude change coefficient, frequency change coefficient, and energy attenuation coefficient of the P-wave; the input ends of the first Concat layer are respectively connected to the output ends of the S-wave feature extraction unit and the P-wave feature extraction unit, and its output end is respectively connected to the input end of the transpose operation layer and the first input end of the multiplier M1; the second input end of the multiplier M1 is connected to the output end of the transpose operation layer, and its output end is connected to the input end of the Conv layer; the input end of the dual-channel feature enhancement unit is connected to the output end of the Conv layer, and its output end is connected to the input end of the fully connected layer; the output end of the fully connected layer serves as the output end of the karst cave prediction model.

[0065] The present invention sets an S-wave feature extraction unit to process the amplitude change coefficient and energy attenuation coefficient of the S-wave, sets a P-wave feature extraction unit to process the phase inversion coefficient, amplitude change coefficient, frequency change coefficient, and energy attenuation coefficient of the P-wave, uses the first Concat layer to splice the P-wave and S-wave features, then uses the transpose operation layer for transpose operation, at the multiplier M1, constructs a fusion matrix A to realize the conversion from one-dimensional data to two-dimensional data, uses the Conv layer for convolution processing, uses the dual-channel feature enhancement unit to enhance different features, and the fully connected layer calculates the predicted distance of the karst cave according to the features output by the dual-channel feature enhancement unit.

[0066] As Figure 3 shown, the S-wave feature extraction unit includes: an amplitude processing channel and an energy processing channel;

[0067] In the S-wave feature extraction unit, the amplitude processing channel is used to process the amplitude change coefficient of the S-wave, and the energy processing channel is used to process the energy attenuation coefficient of the S-wave;

[0068] As Figure 4As shown, the P-wave feature extraction unit includes: a phase processing channel, an amplitude processing channel, a frequency processing channel, and an energy processing channel;

[0069] In the P-wave feature extraction unit, the phase processing channel is used to process the phase inversion coefficient of the P-wave, the amplitude processing channel is used to process the amplitude change coefficient of the P-wave, the frequency processing channel is used to process the frequency change coefficient of the P-wave, and the energy processing channel is used to process the energy attenuation coefficient of the P-wave.

[0070] As Figure 5 shown, the phase processing channel, the amplitude processing channel, the frequency processing channel, and the energy processing channel all include: a Tanh activation layer, a Sigmoid activation layer, and a multiplier M2;

[0071] The Tanh activation layer is used to assign weights and biases to the coefficients input to the phase processing channel, the amplitude processing channel, the frequency processing channel, or the energy processing channel to obtain a first eigenvalue; the Sigmoid activation layer is used to assign weights and biases to the coefficients input to the phase processing channel, the amplitude processing channel, the frequency processing channel, or the energy processing channel to obtain a second eigenvalue; the multiplier M2 is used to multiply the first eigenvalue and the second eigenvalue to obtain the channel output.

[0072] In the S-wave feature extraction unit, two channels are used to process two coefficients respectively. In the P-wave feature extraction unit, four channels are used to process four coefficients respectively. In each channel, two different activation functions, Tanh and Sigmoid, are used to enhance the nonlinear expression ability of the model, enabling it to better fit complex data distributions. By multiplying the eigenvalues processed by different activation functions using the multiplier M2, effective feature fusion is achieved, and higher-order feature combinations can be captured.

[0073] The first Concat layer is used to splice the output features of the S-wave feature extraction unit and the output features of the P-wave feature extraction unit to obtain a spliced vector a. The transpose operation layer is used to perform a transpose operation on the spliced vector a to obtain an operation result a T and the multiplier M1 is used to multiply the operation result a T by the spliced vector a to obtain a fusion matrix A, A = a T a, where T is the transpose operation.

[0074] The dual-channel feature enhancement unit includes: a GAP layer, a GMP layer, a mean feature enhancement layer, a significant feature enhancement layer, and a second Concat layer;

[0075] The input ends of the GAP layer and the GMP layer both serve as the input end of the dual-channel feature enhancement unit;

[0076] The input end of the mean feature enhancement layer is connected to the output end of the GAP layer;

[0077] The input end of the significant feature enhancement layer is connected to the output end of the GMP layer;

[0078] The input end of the second Concat layer is respectively connected to the output ends of the mean feature enhancement layer and the significant feature enhancement layer, and its output end serves as the output end of the dual-channel feature enhancement unit.

[0079] The GAP layer is a global average pooling layer, and the GMP layer is a global maximum pooling layer.

[0080] The mean feature enhancement layer is used to enhance the mean features output by the GAP layer;

[0081] The significant feature enhancement layer is used to enhance the maximum features output by the GMP layer.

[0082] Through the parallel GAP and GMP layers, the present invention can capture global average and maximum features simultaneously, providing a more comprehensive feature representation. Secondly, the mean and significant feature enhancement layers are respectively connected to the outputs of the GAP and GMP layers, which can further strengthen and refine different types of features. The introduction of the second Concat layer realizes the effective fusion of features, combines the enhanced mean and significant features, and generates a richer feature representation.

[0083] In this embodiment, the expressions of the mean feature enhancement layer and the significant feature enhancement layer are both: X = BN(x)·(1 + BN(x)), where X is the enhanced feature, BN is the BN layer, and x is the feature to be enhanced.

[0084] The present invention processes the feature to be enhanced using the BN layer, and according to the result of the BN layer, applies attention (1 + BN(x)) to achieve adaptive adjustment of the attention of important features.

[0085] In this embodiment, the karst cave prediction model is trained using the gradient descent method. When constructing the training samples, the phase inversion coefficient, amplitude change coefficient, frequency change coefficient, and energy attenuation coefficient of the P wave, and the amplitude change coefficient and energy attenuation coefficient of the S wave are used as samples, and the karst cave distance is used as the label.

[0086] In the present invention, during the operation of the shield machine, a geophone is used to collect the reflected waves in the karst area, and the instantaneous frequency, instantaneous amplitude, and instantaneous phase of the P-wave and S-wave are calculated respectively to reflect the changes in frequency, amplitude, and phase at each moment. In a karst cave, the P-wave shows an obvious frequency decrease and phase inversion. The reason is that the karst cave is usually filled with air or water, forming a significant acoustic impedance difference with the surrounding rocks. This difference leads to a high reflection coefficient, resulting in a large signal intensity returned from the boundary of the karst cave. At the same time, the S-wave can only propagate in solids and will undergo strong energy conversion and attenuation when encountering a karst cave filled with fluid. Therefore, capturing the phase inversion coefficient, amplitude change coefficient, frequency change coefficient, and energy attenuation coefficient of the P-wave, as well as the amplitude change coefficient and energy attenuation coefficient of the S-wave, can effectively reflect the response characteristics of the P-wave and S-wave to the karst cave. Combining these characteristic coefficients with the processing of the karst cave prediction model can further improve the accuracy of karst cave prediction and provide a strong guarantee for the safety of shield construction.

[0087] When predicting karst caves, the present invention abandons the analysis of signal components, focuses on the response characteristics of the P-wave and S-wave to karst caves, and then uses the karst cave prediction model to predict the prediction distance of karst caves, improving the accuracy of karst cave prediction.

[0088] The above are only the preferred embodiments of the present invention and are not used to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for predicting caves in a karst area when a shield machine passes through it, characterized in that: The following steps are involved: S1. Collect the reflected waves from the karst area when the shield machine is working through the detector; S2, respectively calculate the instantaneous frequency, instantaneous amplitude and instantaneous phase of the P wave and S wave in the reflected wave; S3, extracting the phase reversal coefficient, amplitude variation coefficient, frequency variation coefficient and energy attenuation coefficient according to the instantaneous phase, instantaneous amplitude and instantaneous frequency at multiple consecutive moments in the P wave; S4, extracting the amplitude variation coefficient and energy attenuation coefficient according to the instantaneous amplitudes at multiple moments in the S wave; S5. According to the cave prediction model, the phase reversal coefficient, amplitude variation coefficient, frequency variation coefficient and energy attenuation coefficient of the P wave, as well as the amplitude variation coefficient and energy attenuation coefficient of the S wave are processed to obtain the cave prediction distance; The cave prediction model in S5 includes: an S-wave feature extraction unit, a P-wave feature extraction unit, a first Concat layer, a transposition operation layer, a multiplier M1, a Conv layer, a dual-channel feature enhancement unit and a fully connected layer; The input end of the S-wave feature extraction unit is used to input the amplitude variation coefficient and energy attenuation coefficient of the S-wave; the input end of the P-wave feature extraction unit is used to input the phase reversal coefficient, amplitude variation coefficient, frequency variation coefficient and energy attenuation coefficient of the P-wave; the input end of the first Concat layer is respectively connected to the output end of the S-wave feature extraction unit and the output end of the P-wave feature extraction unit, and its output end is respectively connected to the input end of the transposition operation layer and the first input end of the multiplier M1; the second input end of the multiplier M1 is connected to the output end of the transposition operation layer, and its output end is connected to the input end of the Conv layer; the input end of the dual-channel feature enhancement unit is connected to the output end of the Conv layer, and its output end is connected to the input end of the fully connected layer; the output end of the fully connected layer serves as the output end of the cave prediction model.

2. The method for predicting caves in a karst area when a shield machine passes through according to claim 1 is characterized in that: Extracting the phase reversal coefficient in S3 comprises the following steps: A1. In a plurality of consecutive moments, the instantaneous phase of the latter moment is subtracted from the instantaneous phase of the previous moment in two adjacent moments, and the absolute value is taken to obtain a phase difference, and the plurality of phase differences are added to obtain a phase difference addition result; A2. In a plurality of consecutive moments, subtract the maximum instantaneous phase from the minimum instantaneous phase to obtain the maximum difference; A3. Add the maximum difference and the phase difference addition result, and normalize them to obtain a phase reversal coefficient.

3. The method for predicting a karst cave when a shield machine passes through a karst area according to claim 1, characterized in that: The specific process of extracting the amplitude variation coefficient in S3 and S4 is as follows: in a plurality of consecutive moments, the instantaneous amplitude at the latter moment is subtracted from the instantaneous amplitude at the previous moment between two adjacent moments, and the absolute value is taken to obtain an amplitude difference, multiple amplitude differences are added to obtain an amplitude difference addition result, and the amplitude difference addition result is normalized to obtain an amplitude variation coefficient; The specific process of extracting the frequency change coefficient in S3 is: in a plurality of consecutive moments, subtract the instantaneous frequency of the latter moment from the instantaneous frequency of the previous moment in two adjacent moments, and take the absolute value to obtain a frequency difference, add a plurality of frequency differences to obtain a frequency difference addition result, and normalize the frequency difference addition result to obtain the frequency change coefficient.

4. The method for predicting a karst cave when a shield machine passes through a karst area according to claim 1, characterized in that: The specific process of extracting the energy attenuation coefficient in S3 and S4 is as follows: in a plurality of consecutive moments, square the instantaneous amplitude, accumulate the squared instantaneous amplitude to obtain energy, and in a time length N, take the normalized value of the ratio of the maximum energy to the minimum energy as the energy attenuation coefficient, where N is a positive integer.

5. The method for predicting a karst cave when a shield machine passes through a karst area according to claim 1, characterized in that: The S wave feature extraction unit includes: an amplitude processing channel and an energy processing channel; In the S-wave feature extraction unit, the amplitude processing channel is used to process the amplitude variation coefficient of the S-wave, and the energy processing channel is used to process the energy attenuation coefficient of the S-wave; The P wave feature extraction unit includes: a phase processing channel, an amplitude processing channel, a frequency processing channel and an energy processing channel; In the P wave feature extraction unit, the phase processing channel is used to process the phase reversal coefficient of the P wave, the amplitude processing channel is used to process the amplitude variation coefficient of the P wave, the frequency processing channel is used to process the frequency variation coefficient of the P wave, and the energy processing channel is used to process the energy attenuation coefficient of the P wave.

6. The method for predicting a karst cave when a shield machine passes through a karst area according to claim 5, characterized in that: The phase processing channel, amplitude processing channel, frequency processing channel and energy processing channel all include: a Tanh activation layer, a Sigmoid activation layer and a multiplier M2; The Tanh activation layer is used to assign weights and biases to the coefficients of the input phase processing channel, amplitude processing channel, frequency processing channel or energy processing channel to obtain a first eigenvalue; the Sigmoid activation layer is used to assign weights and biases to the coefficients of the input phase processing channel, amplitude processing channel, frequency processing channel or energy processing channel to obtain a second eigenvalue; the multiplier M2 is used to multiply the first eigenvalue and the second eigenvalue to obtain a channel output.

7. The method for predicting caves in a karst area when a shield machine passes through according to claim 1 is characterized in that: The first Concat layer is used to concatenate the output features of the S wave feature extraction unit and the output features of the P wave feature extraction unit to obtain a concatenated vector a, and the transposition operation layer is used to perform a transposition operation on the concatenated vector a to obtain an operation result a T The multiplier M1 is used to convert the operation result a T Multiply it with the concatenated vector a to get the fusion matrix A, A = a T a, where T is the transpose operation.

8. The method for predicting caves in a karst area when a shield machine passes through according to claim 1 is characterized in that: The dual-channel feature enhancement unit includes: a GAP layer, a GMP layer, a mean feature enhancement layer, a significant feature enhancement layer and a second Concat layer; The input ends of the GAP layer and the GMP layer are both used as input ends of the dual-channel feature enhancement unit; The input end of the mean feature enhancement layer is connected to the output end of the GAP layer; The input end of the salient feature enhancement layer is connected to the output end of the GMP layer; The input end of the second Concat layer is connected to the output end of the mean feature enhancement layer and the output end of the significant feature enhancement layer respectively, and the output end thereof serves as the output end of the dual-channel feature enhancement unit.

9. The method for predicting a karst cave when a shield machine passes through a karst area according to claim 8, characterized in that: The mean feature enhancement layer is used to enhance the mean feature output by the GAP layer; The significant feature enhancement layer is used to enhance the maximum value feature output by the GMP layer.

Images