Shield tunneling noise source surface wave velocity intelligent inversion method and system

Abstract

The application discloses a shield tunneling noise source surface wave velocity intelligent inversion method and system, comprising: taking each modal dispersion curve as input and corresponding wave velocity curve as output, constructing an inversion network unit corresponding to each modal dispersion curve, weighting and connecting the inversion network units in parallel through setting a weighted mask of each modal dispersion curve to obtain a multi-modal dispersion curve weighted inversion network; after discarding random dispersion points of the input multi-modal dispersion curve, training the multi-modal dispersion curve weighted inversion network; acquiring tunneling noise source surface wave data corresponding to measuring points in a detection area, extracting a multi-modal dispersion curve, and predicting wave velocity distribution of the multi-modal dispersion curve by using the trained multi-modal dispersion curve weighted inversion network. High-robustness dispersion curve nonlinear inversion is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of geophysical exploration, in particular to a shield tunneling noise source surface wave velocity intelligent inversion method and system. BACKGROUND

[0002] The statements in this section merely provide background information related to the present application and do not necessarily constitute the prior art.

[0003] For the problem of advanced prediction of adverse geology in front of the shield tunnel, the shield tunneling noise can be used as a seismic source, long measuring line geophones are laid on the ground to receive surface wave signals, and an advanced prediction method based on noise source surface wave data is used to realize disaster source imaging.

[0004] The tunneling noise source surface wave data contains rich multi-modal dispersion information, and the high-order modal surface wave is more sensitive to the change of the stratum wave velocity than the surface wave, so that the joint inversion is carried out through the multi-modal dispersion curve, and the accuracy of the inversion method is improved.

[0005] The existing surface wave inversion method mainly uses the base order mode for inversion, or uses the base order and part of the high-order mode dispersion points for inversion, so that the characteristics contained in the multi-modal dispersion curve are not fully mined and utilized, resulting in strong multi-solution of the inversion and affecting the accuracy of the inversion. SUMMARY

[0006] In order to solve the above problems, the present application provides a shield tunneling noise source surface wave velocity intelligent inversion method and system, proposes an inversion network construction method based on weighted mask, constructs an inversion network structure in which the multi-modal inversion modules are parallel to each other, provides weighted constraints for the inversion process, and uses a training method of randomly discarding dispersion points to accurately predict the corresponding wave velocity distribution of the multi-modal dispersion curve without the need to pre-set the initial model and layer thickness.

[0007] In order to achieve the above purpose, the present application adopts the following technical solutions:

[0008] In the first aspect, the present application provides a shield tunneling noise source surface wave velocity intelligent inversion method, comprising:

[0009] The theoretical dispersion curve of each order mode is taken as the input, and the corresponding wave velocity curve is taken as the output, the inversion network unit corresponding to each order mode dispersion curve is constructed, the weighted mask of each order mode dispersion curve is set, the weighted parallel connection of the inversion network unit is obtained, and the multi-modal dispersion curve weighted inversion network is obtained;

[0010] After the input multi-modal dispersion curve is discarded randomly, the multi-modal dispersion curve weighted inversion network is trained, and in this way, the incomplete condition of the dispersion curve extraction in the actual detection is simulated;

[0011] The tunnel noise source surface wave data corresponding to the measuring points in the detection area is acquired, and multi-modal dispersion curves are extracted, and the multi-modal dispersion curve is predicted by using the trained multi-modal dispersion curve weighted inversion network to obtain the wave velocity distribution.

[0012] As an optional implementation, the multi-modal dispersion curve is a dispersion curve of five modes of a base mode and first-order to fourth-order high-order modes.

[0013] As an optional implementation, the weighting mask of each mode dispersion curve is:

[0014]

[0015] In the formula, M m1 -M m5 is the weighting mask of the base mode to the fourth-order high-order mode dispersion curve; z is the depth axis coordinate, and h is the total depth of the inversion area.

[0016] As an optional implementation, the random dispersion point discarding mode is that the base mode and the first-order high-order mode dispersion points in the 0-10Hz frequency range are randomly discarded at a set probability, and the dispersion points of all five modes in the 11-55Hz frequency range are randomly discarded at a set probability.

[0017] As an optional implementation, the multi-modal dispersion curve weighted inversion network is:

[0018]

[0019] In the formula, U mn represents the inversion network unit corresponding to each mode dispersion curve, l mn represents each mode dispersion curve, c represents the wave velocity curve output by the multi-modal dispersion curve weighted inversion network, c1-c5 are the wave velocity curves output by the inversion network unit corresponding to the base mode to the fourth-order high-order mode dispersion curve, M m1 -M m5 is the weighting mask of the base mode to the fourth-order high-order mode dispersion curve, and w1-w5 are network parameters of each inversion network unit.

[0020] As an optional implementation, the target function set in the training process is:

[0021]

[0022] In the formula, U m1-5 represents the inversion network unit corresponding to each mode dispersion curve, and c is the wave velocity curve; l m1-5 is the theoretical dispersion curve of the five modes; L1 represents a norm, L2 represents a two norm, and b is the number of samples.

[0023] As an alternative embodiment, the extracted dispersion curve is subjected to one-dimensional wave velocity curve prediction by using the trained multi-modal dispersion curve weighted inversion network, and after all the measuring points are processed, the two-dimensional wave velocity profile of the rock-soil medium distribution in the detection area is formed by interpolating and fitting all the one-dimensional wave velocity curves.

[0024] In a second aspect, the present application provides a shield tunneling noise source wave velocity intelligent inversion system, comprising:

[0025] The network construction module is configured to construct an inversion network unit corresponding to each order of modal dispersion curve with the dispersion curve as input and the corresponding wave velocity curve as output, and obtain the multi-modal dispersion curve weighted inversion network by connecting the inversion network units in parallel after weighting through setting the weighting mask of each order of modal dispersion curve.

[0026] The network training module is configured to train the multi-modal dispersion curve weighted inversion network after discarding the random dispersion points of the input multi-modal dispersion curve.

[0027] The wave velocity inversion module is configured to obtain the tunneling noise source surface wave data corresponding to the measuring points in the detection area, extract the multi-modal dispersion curve, and predict the wave velocity distribution by using the trained multi-modal dispersion curve weighted inversion network.

[0028] In a third aspect, the present application provides an electronic device comprising a memory and a processor, and computer instructions stored in the memory and running on the processor, wherein when the computer instructions are run by the processor, the method of the first aspect is completed.

[0029] In a fourth aspect, the present application provides a computer readable storage medium for storing computer instructions, wherein when the computer instructions are executed by a processor, the method of the first aspect is completed.

[0030] Compared with the prior art, the present application has the following beneficial effects:

[0031] The multi-modal dispersion curve of the tunneling noise source surface wave data contains the response to the wave velocity change of the underground medium, and in order to fully utilize the multi-modal dispersion curve information, the present application analyzes the stratum wave velocity change law reflected in the multi-modal dispersion curve shape characteristics, and proposes a corresponding multi-modal inversion network weighting mask construction method to provide a weighted constraint for the subsequent inversion process.

[0032] In view of the strong nonlinearity of the inversion of the multimodal dispersion curve of the surface wave and the robustness of the practical application, an inversion network structure in which high-order mode inversion modules are parallel to each other is constructed, and the inversion effect is better than that of a serial structure; in the training mode, a training method of randomly discarding the dispersion points of the input curve is adopted to realize the high-robustness nonlinear inversion of the dispersion curve.

[0033] Advantages of the additional aspects of the application will be partially given in the following description, partially become obvious from the following description, or be known by the practice of the application. BRIEF DESCRIPTION OF DRAWINGS

[0034] The drawings accompanying the specification of this application form a part thereof, serve to provide further understanding of the application, and together with the description of the exemplary embodiments of the application and the explanation thereof serve to explain the application, and do not constitute an improper limitation of the application.

[0035] Figure 1 A shield tunneling noise source surface wave velocity intelligent inversion method flowchart is provided for embodiment 1 of the application;

[0036] Figure 2 An InceptionV2 network model schematic diagram is provided for embodiment 1 of the application;

[0037] Figures 3(a)-3(b) A multimodal dispersion curve weighted inversion network schematic diagram is provided for embodiment 1 of the application;

[0038] Figures 4(a)-4(b) A network training stage loss function curve comparison diagram with and without adding a multimodal weighted mask is provided for embodiment 1 of the application;

[0039] Figures 5(a)-5(b) A network loss function curve comparison diagram with and without using a random discarding training method is provided for embodiment 1 of the application;

[0040] Figures 6(a)-6(b) A multimodal dispersion curve weighted inversion network schematic diagram of serial and parallel structures is provided for embodiment 1 of the application;

[0041] Figures 7(a)-7(b) A network training stage loss function curve comparison diagram of serial and parallel structures is provided for embodiment 1 of the application;

[0042] Figure 8 A multimodal dispersion curve weighted inversion network multi-layer model prediction result diagram is provided for embodiment 1 of the application. DETAILED DESCRIPTION

[0043] The application will be further described below in combination with the drawings and embodiments.

[0044] It should be noted that the following detailed description is exemplary in nature and is intended to provide further description of the application. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.

[0045] It is to be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of example embodiments in accordance with the present application. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "comprises" and / or "comprising," when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0046] The embodiments in the present application and the features in the embodiments can be combined with each other without conflict.

[0047] Embodiment 1

[0048] The embodiment provides a shield tunneling noise source surface wave velocity intelligent inversion method, as shown in the method comprises the following steps: Figure 1

[0049] Taking each order modal dispersion curve as input and corresponding wave velocity curve as output, an inversion network unit corresponding to each order modal dispersion curve is constructed, and after weighting and parallel connection of the inversion network units by setting a weighting mask of each order modal dispersion curve, a multi-modal dispersion curve weighting inversion network is obtained.

[0050] After discarding random dispersion points of the input multi-modal dispersion curve, the multi-modal dispersion curve weighting inversion network is trained.

[0051] Obtaining tunneling noise source surface wave data corresponding to measuring points in a detection area, and extracting a multi-modal dispersion curve, the multi-modal dispersion curve is predicted by using the trained multi-modal dispersion curve weighting inversion network.

[0052] In the embodiment, the current shield tunneling mileage and the current buried depth are determined, the position of the ground surface corresponding to the tunneling line is determined, and the measuring line is laid according to the ground surface environment condition, the direction of the measuring line, the offset distance, the trace interval and the length of the measuring line are determined; by setting the sampling parameters of the geophone, the field data collection is started, the geophone is checked regularly during sampling, and after ensuring that the sampling time is sufficient, the geophone is recovered and the data is read in order.

[0053] ​According to the detection requirements, the multi-channel data is grouped, the number of seismic data channels corresponding to each measuring point is determined, the tunneling noise source surface wave data corresponding to each measuring point is obtained, the tunneling noise section is selected, and the selected data is sequentially subjected to the conventional noise source detection data processing steps of band-pass filtering, linear trend removal, spectral whitening, cross-correlation, etc.; the tunneling noise source surface wave real-time measurement data quality optimization method is selectively used according to the processed data waveform and dispersion spectrum quality to suppress multiple types of interference; then the dispersion curve is extracted from the processed data, dispersion imaging is performed, and according to the correspondence between the dispersion spectrum and the extracted dispersion curve, manual curve correction can be performed, and unreasonable dispersion points can be removed and reasonable dispersion points can be supplemented.

[0054] In the embodiment, the multi-modal dispersion curve is the dispersion curve of the base mode and the first-order high-order to the fourth-order high-order five modes, and the weighting mask of each mode dispersion curve is set as:

[0055]

[0056] In the formula, M m 1-M m5 is the weighting mask of the inversion network unit output wave velocity characteristics corresponding to the five modes, and the subscripts m1-m5 represent the base mode to the fourth-order high-order mode; z represents the depth axis coordinate, and h represents the total depth of the inversion region.

[0057] The weighting mask of the multi-modal dispersion curve is equivalent to an artificial network attention mechanism, which guides the network parameters associated with each mode dispersion curve to pay attention to the wave velocity information at the most relevant depth in the prediction model; the design idea is:

[0058] (1) Considering that the base mode and the first-order high-order mode dispersion curve often have a "mode kiss" phenomenon at the low-frequency end, if the dispersion curve extraction is not accurate, even if the mode is misjudged, it will directly lead to distortion of the inversion result; and the information contained in the base mode and the first-order high-order mode dispersion curve at the low-frequency end corresponds to the wave velocity of the deep stratum, so the influence of the base mode and the first-order high-order mode on the deep wave velocity in the inversion result should be weakened; and the phase velocity corresponding to the high-frequency end of the above two modes directly reflects the wave velocity of the model surface layer, so the wave velocity of the shallow layer should be enhanced, so the weight corresponding to the shallow layer wave velocity is increased in M m1 and M m2 .

[0059] (2) Considering that the second-order, third-order and fourth-order high-order mode dispersion curve can better reflect the wave velocity variation information of the middle and deep stratum compared with the base mode and the first-order high-order mode dispersion curve, and the highest phase velocity at the low-frequency end can directly reflect the shear wave velocity of the deep bedrock, therefore, the influence of the above modes on the deep wave velocity in the inversion result is enhanced by M m3 , M m4 and M m5 .

[0060] (3) Considering that the main frequency of the surface wave data of tunneling noise source is roughly distributed between 15 Hz and 35 Hz, and that the formation attenuation and absorption in the actual data often lead to the loss of high-frequency information, the high-order mode dispersion curves distributed in the higher frequency band often face the situation of poor extraction effect. Therefore, this embodiment further weakens the influence of the fourth-order high-order mode information on the inversion model.

[0061] In this embodiment, the vertical one-dimensional shear wave velocity curve within a 30m underground range is used as the output, with a grid spacing of 1m. The theoretical dispersion curves of the fundamental and first- to fourth-order higher modes, with a frequency range of 0Hz-55Hz, corresponding to the model, are used as input to construct a multi-mode dispersion curve weighted inversion network U based on the fundamental mode. m1-5 It contains 5 inversion network units U mn , represented as:

[0062]

[0063] In the formula, U mn The l represents the inversion network unit corresponding to each modal dispersion curve. mn The vector represents the dispersion curves of each mode, c represents the wave velocity curve output by the weighted inversion network of the multimode dispersion curve, c1-c5 are the wave velocity curves output by the corresponding inversion network units of the dispersion curves from the fundamental mode to the fourth higher-order mode, and M m1 -M m5 w1-w5 are the weighted masks for the dispersion curves from the basic mode to the fourth higher-order modes, and w1-w5 are the network parameters of each inversion network unit.

[0064] U m 1-5 Its structure requires consideration of fundamental mode information when establishing the mapping relationship between higher-order mode dispersion curves and wave velocities. Furthermore, the corresponding inversion network unit U for each mode... mn The identical structures and parallel connection of weighted masks help to reduce the mutual influence of higher-order modal features during inversion, preventing the error from being propagated to higher-order modal inversion processes after the inversion effect of poor curve extraction in intermediate-order modes is affected. Simultaneously, the above equation demonstrates the weighted mask M. mn Its specific working method is similar to a manually preset network attention mechanism. It constrains the wave velocity characteristics mapped from each modal dispersion curve by weighting them. mn Sensitivity to wave velocity inversion results.

[0065] In this embodiment, in the inversion network unit U mn Several InceptionV2 network models containing pooling layers are introduced, such as Figure 2As shown, the increase of the depth of the convolution layer will facilitate the capture of small-scale data pattern information in the input, so that Inception V2 can extract multi-scale features from the input data. Inception V2 can be used for curve pattern feature extraction and inversion of measured tunneling noise source surface wave data dispersion curves of different completeness. For the case where the input is an incomplete dispersion curve, the channel related to the maximum pooling will assist the network to improve the mining of large-scale features in the dispersion curve, and ignore the influence of small-scale features caused by the loss of dispersion points; when the extracted dispersion curve is relatively complete, the network will be able to improve the inversion accuracy by extracting small-scale pattern information contained in the dispersion curve through multi-layer convolution.

[0066] To improve the U m1-5 Robustness of inversion of incomplete input data, U m1-5 In the training phase on the simulation data set, in addition to using the theoretical dispersion curve as the input of U m1-5 The prediction result and the label wave velocity calculation target function, but also to the modal theoretical dispersion curve on the dispersion point to a certain proportion of random loss, also input U m1-5 Generate predicted wave velocity and calculate the target function.

[0067] The random dispersion point loss method is:

[0068] (1) For the base and first-order high-order modal dispersion points in the 0-10Hz frequency range, randomly discard with a probability of 80%.

[0069] (2) For the dispersion points of all five modes in the 11-55Hz frequency range, randomly discard with a probability of 20% to simulate the incomplete dispersion curve extraction in actual detection.

[0070] The vertical one-dimensional shear wave velocity curve is used as the label, and the multi-modal theoretical dispersion curve corresponding to the wave velocity curve and the theoretical dispersion curve after random loss of the dispersion points are used as the input to train U m1-5 , U m1-5 The target function of the training process can be represented as:

[0071]

[0072] In the formula, U mn represents the inversion network unit corresponding to each order of modal dispersion curve, l mn represents the modal dispersion curve, c represents the vertical one-dimensional shear wave velocity curve, i.e. the training label, b represents the number of single training samples, L1 represents the one norm, L2 represents the two norm, is the theoretical dispersion curve after random loss of the dispersion points; the meaning of the subscript m 1-5 is the base to the fourth-order high-order five modes.

[0073] It is worth noting that, unlike conventional dispersion curve inversion methods, U m1-5 It can directly predict the one-dimensional vertical wave velocity distribution curve within a 30m underground range, and does not output the wave velocity magnitude of each layer under the condition that the stratification is known. This process does not require pre-setting the number of layers and layer thickness, etc., of the geological stratification. On the other hand, U m1-5 No initial wave velocity model needs to be set beforehand when making predictions. These two characteristics make this method more practical than conventional dispersion curve inversion methods.

[0074] In this embodiment, surface wave data of tunneling noise sources corresponding to measuring points within the actual detection area are measured, and multi-mode dispersion curves are extracted. A multi-mode dispersion curve weighted inversion network is used to predict one-dimensional wave velocity curves. The network can be selectively trained using physical-driven transfer learning based on the wave velocity prediction effect. After all measuring points have been processed, all one-dimensional wave velocity curves are interpolated and fitted to form a two-dimensional wave velocity profile of the rock and soil medium distribution in the corresponding detection area.

[0075] like Figures 3(a)-3(b) The image shows the weighted inversion network U for multimodal dispersion curves. m1-5 The dispersion curves were adopted from the basic order and the first to fourth order higher-order modes of the theoretical dispersion curves. The network hyperparameters were set as follows: 100 rounds, 24 training samples per session, learning rate of 1e-3, inactivation rate of 0.2, and the optimizer was set to Adam.

[0076] Figures 4(a)-4(b) Demonstrates the addition of a multimodal weighted mask M mn U before and after m1-5 Loss function curves for the training and validation sets. U is calculated by adding a multimodal weighted mask to the training set. m1-5 The convergence speed of the loss function is slightly improved, and the convergence value is slightly reduced; however, on the validation set, the U function with added multimodal weighted mask shows a slight improvement. m1-5 The loss function converges to 0.0365; without the added weighted mask, it converges to 0.0483. After adding the weighted mask, U... m1-5 It clearly exhibits superior training convergence. Comparing U before and after adding a multimodal mask... m1-5 Performance on the test set was evaluated using Mean Absolute Error (MAE) and Mean Squared Error (MSE) to assess the accuracy of predictions on the test set data. Without a multimodal mask, the MAE and MSE were 0.044679 and 0.004997, respectively; in this embodiment, the MAE and MSE were 0.037206 and 0.004103, respectively. With a multimodal mask added, the accuracy of predictions was... m1-5 It has a clear advantage in both of the above indicators.

[0077] When calculating the objective function, the input dispersion curve is randomly dropped to a certain extent to approximate the dispersion curve extracted from actual data, thereby enhancing the practicality of the method. Without random dropping, the MAE and MSE are 0.042824 and 0.004850, respectively; in this embodiment, the MAE and MSE are 0.037206 and 0.004103, respectively. Figures 5(a)-5(b) The loss function curves for the training and validation sets demonstrate the performance improvement of the inversion network achieved by this training strategy. The input dispersion curves for both the validation and test sets underwent random dispersion point discarding in the same manner. In this embodiment, the validation set loss function curve converges to 0.0365, lower than the convergence value of 0.0376 for the validation set loss function curve without this training method. Furthermore, the difference in metrics is even more pronounced on the test set.

[0078] In addition, U m1-5 Inversion network unit U of intermediate and high-order modes mn The network structures were configured to operate in parallel to reduce the mutual influence of the inversion of the dispersion curves of each mode, as shown in Figure 6(a). Comparing the training effect of the serial network structure for high-order mode inversion shown in Figure 6(b), random dispersion points were also discarded from the input dispersion curves of the network throughout the training process. Figures 7(a)-7(b) As shown, the parallel structure exhibits superior performance on both the training set loss function curve and the validation set loss function curve. These results can simulate the situation where dispersion curve extraction is incomplete in real-world scenarios, and the corresponding results demonstrate the superior performance of the U-structure described in this embodiment. m1-5 The inversion network structure has better robustness.

[0079] Finally, the U obtained by training using the method described in this embodiment m1-5 The prediction results on the test set data are as follows: Figure 8 As shown, the multimodal dispersion curve inversion method based on difference optimization is used as a baseline for comparison. To ensure the stability and convergence of this method, the inversion is performed using 10 thin layers of equal thickness, each 3m thick. The initial model is set with progressively increasing velocities based on the original model's wave velocity range. During testing, the dispersion curves are also processed in the manner described above, with dispersion points randomly discarded.

[0080] like Figure 8 As shown, the first row contains three velocity-increasing models. The first column of models has a relatively simple structure, while the second and third columns contain multiple strata with thin layers and small increases in wave velocity with depth. m1-5 The predictions from the above models are almost identical to the actual models, with mean squared error (MSE) as low as 10. -4 The mean square error is about an order of magnitude lower than that of the benchmark method; the second row shows models with high-speed interlayers, U m1-5The prediction result can accurately reflect the position and wave velocity of the high wave velocity body, and the inversion accuracy is generally higher than that of the benchmark method; the third row is a model containing a low-velocity interlayer, U m1-5 The prediction result is generally accurate, and for the rightmost column of the model with a sharp change in wave velocity, U m1-5 The prediction result is still relatively accurate, and the inversion result of the benchmark method has significantly increased error.

[0081] From the overall inversion result, the lowest wave velocity in these models is lower than 200 m / s, the highest wave velocity can reach about 950 m / s, and the number of layers ranges between 3-6 layers, U m1-5 The wave velocity of the shallowest layer and the deepest bedrock within a range of 30m underground can be accurately predicted, and the wave velocity distribution of each horizontal layer in the middle can be accurately described. Moreover, the prediction process does not require setting an initial model or prior layering information, and has great advantages in practicality compared with the benchmark method, which is a conventional dispersion curve inversion method.

[0082] Embodiment 2

[0083] The embodiment provides a shield tunneling noise source wave velocity intelligent inversion system, comprising:

[0084] The network construction module is configured to input each modal dispersion curve as input and output the corresponding wave velocity curve, construct the inversion network unit corresponding to each modal dispersion curve, and obtain the multi-modal dispersion curve weighted inversion network by setting the weighted mask of each modal dispersion curve and connecting the inversion network units in parallel.

[0085] The network training module is configured to discard random dispersion points of the input multi-modal dispersion curve, and train the multi-modal dispersion curve weighted inversion network;

[0086] The wave velocity inversion module is configured to obtain the tunneling noise source surface wave data corresponding to the detection points in the detection area, extract the multi-modal dispersion curve, and predict the wave velocity distribution by using the trained multi-modal dispersion curve weighted inversion network on the multi-modal dispersion curve.

[0087] It should be noted that the above modules correspond to the steps described in Embodiment 1, and the above modules and the examples and application scenarios realized by the corresponding steps are the same, but are not limited to the content disclosed in Embodiment 1. It should be noted that the above modules as part of the system can be executed in a computer system such as a group of computer executable instructions.

[0088] In more embodiments, there are also provided:

[0089] An electronic device includes a memory and a processor and computer instructions stored on the memory and run on the processor, when the computer instructions are run by the processor, the method described in embodiment 1 is completed. For the sake of brevity, it will not be repeated here.

[0090] It should be understood that in the embodiments, the processor can be a central processing unit CPU, and the processor can also be other general-purpose processors, digital signal processors DSPs, application-specific integrated circuits ASICs, ready-to-program gate arrays FPGA or other programmable logic devices, discrete gates or transistor logic, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor, etc.

[0091] The memory can include read-only memory and random access memory, and provide instructions and data to the processor, and a part of the memory can also include non-volatile random access memory. For example, the memory can also store device type information.

[0092] A computer readable storage medium for storing computer instructions, when the computer instructions are executed by the processor, the method described in embodiment 1 is completed.

[0093] The method in embodiment 1 can be directly embodied as a hardware processor to complete, or a combination of hardware and software modules in the processor to complete. The software module can be located in a storage medium mature in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory or electrically erasable programmable memory, register, etc. The storage medium is located in the memory, and the processor reads the information in the memory to complete the steps of the above method in combination with its hardware. To avoid repetition, it will not be described in detail here.

[0094] Those of ordinary skill in the art can realize that the units of the examples described in combination with the embodiments, i.e. the algorithm steps, can be realized in electronic hardware or in combination of computer software and electronic hardware. Whether the functions are executed in hardware or software mode depends on the specific application and design constraints of the technical solution. The skilled person can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0095] The above describes the specific embodiments of the application in combination with the drawings, but is not a limitation on the protection scope of the application. Those skilled in the art should understand that various modifications or variations made by those skilled in the art on the basis of the technical solutions of the application without creative labor are still within the protection scope of the application.

Claims

1. A shield tunneling noise source surface wave velocity intelligent inversion method, characterized in that, Comprise: With each modal dispersion curve as input, with corresponding wave velocity curve as output, build each modal dispersion curve corresponding inversion network unit, set the weighted mask of each modal dispersion curve, and get the multi-modal dispersion curve weighted inversion network after the weighted parallel connection of the inversion network unit; The multi-modal dispersion curve weighted inversion network is: wherein, U m n represents the inversion network unit corresponding to each order modal dispersion curve, l m n represents each modal dispersion curve, c represents the wave velocity curve output by the multi-modal dispersion curve weighted inversion network, c1-c5 are the wave velocity curves output by the inversion network unit corresponding to the base order modal to the fourth order high order modal dispersion curve, M m1 -M m5 is the weighted mask of the base order modal to the fourth order high order modal dispersion curve, w1-w5 are the network parameters of each inversion network unit; After discarding the random dispersion points of the input multi-modal dispersion curve, the multi-modal dispersion curve weighted inversion network is trained; Obtain the tunneling noise source surface wave data corresponding to the measuring points in the detection area, extract the multi-modal dispersion curve, and predict the wave velocity distribution of the multi-modal dispersion curve using the trained multi-modal dispersion curve weighted inversion network.

2. The shield tunneling noise source surface wave velocity intelligent inversion method of claim 1, wherein, The multi-modal dispersion curve is the dispersion curve of the base mode and the first to fourth high-order modes.

3. The shield tunneling noise source surface wave velocity intelligent inversion method of claim 2, wherein, The set weighted mask of each modal dispersion curve is: wherein M m1 -M m5 is a weighted mask of the first to fourth higher order mode dispersion curves; z is the depth axis coordinate and h is the total depth of the inversion region.

4. The shield tunneling noise source surface wave velocity intelligent inversion method of claim 2, wherein, The random dispersion point discarding mode is: for the base mode and the first high-order mode dispersion points in the 0-10Hz frequency range, randomly discard with a set probability; for all five modes of dispersion points in the 11-55Hz frequency range, randomly discard with a set probability.

5. The shield tunneling noise source surface wave velocity intelligent inversion method of claim 1, wherein, The target function set in the training process is: In the formula, U m 1-5 The corresponding inversion network unit of each order modal dispersion curve is represented, and c is the wave velocity curve; l m 1-5 The theoretical dispersion curves of the five modes are shown in the following table: The theoretical dispersion curve after random discarding of the dispersion points is shown in the following table; L1 represents a norm, L2 represents a two-norm, and b is the sample number.

6. The shield tunneling noise source surface wave velocity intelligent inversion method of claim 1, wherein, For the extracted dispersion curve, the trained multi-modal dispersion curve weighted inversion network is used to predict the one-dimensional wave velocity curve, and after all the measuring points are processed, the one-dimensional wave velocity curves are interpolated and fitted to form the two-dimensional wave velocity profile of the rock-soil medium distribution in the detection area.

7. A shield tunneling noise source surface wave velocity intelligent inversion system, characterized in that, Comprise: The network construction module is configured to build each modal dispersion curve corresponding inversion network unit with each modal dispersion curve as input and corresponding wave velocity curve as output, set the weighted mask of each modal dispersion curve, and get the multi-modal dispersion curve weighted inversion network after the weighted parallel connection of the inversion network unit; The multi-modal dispersion curve weighted inversion network is: wherein, U m n represents the inversion network unit corresponding to each order modal dispersion curve, l m n represents each modal dispersion curve, c represents the wave velocity curve output by the multi-modal dispersion curve weighted inversion network, c1-c5 are the wave velocity curves output by the inversion network unit corresponding to the base order modal to the fourth order high order modal dispersion curve, M m1 -M m5 is the weighted mask of the base order modal to the fourth order high order modal dispersion curve, w1-w5 are the network parameters of each inversion network unit; The network training module is configured to discard the random dispersion points of the input multi-modal dispersion curve, and train the multi-modal dispersion curve weighted inversion network. The wave velocity inversion module is configured to obtain the tunneling noise source surface wave data corresponding to the measuring points in the detection area, extract the multi-modal dispersion curve, and predict the wave velocity distribution of the multi-modal dispersion curve using the trained multi-modal dispersion curve weighted inversion network.

8. An electronic device, comprising: A computer program product comprising a memory and a processor and computer instructions stored on the memory and running on the processor, when the computer instructions are run by the processor, the method of any one of claims 1-6 is completed.

9. A computer-readable storage medium, characterized in that, A computer program product for storing computer instructions, when the computer instructions are executed by the processor, the method of any one of claims 1-6 is completed.

Images