Electromagnetic acoustic combined early warning system for permeability characteristics of deep-buried tunnel

Abstract

The invention provides an electromagnetic acoustic combined early warning system for permeability characteristics of a deep-buried tunnel, and relates to the technical field of data analysis. The method comprises the following steps: firstly, acquiring electromagnetic wave data and ultrasonic wave data of the deep-buried tunnel; secondly, taking harmonic semantic information as auxiliary mining information in a frequency domain, and performing permeability characteristic semantic mining on the electromagnetic wave data to form an electromagnetic wave permeability characteristic vector; secondly, the harmonic semantic information serves as auxiliary mining information in a frequency domain, permeability characteristic semantic mining is conducted on the ultrasonic data, and an ultrasonic permeability characteristic vector is formed; further, carrying out association semantic extraction on the electromagnetic wave permeability characteristic vector and the ultrasonic wave permeability characteristic vector to form a multi-dimensional permeability characteristic vector; and finally, performing semantic reduction based on the multi-dimensional permeability characteristic vector to form target permeability characteristic data. On the basis of the content, the problem that in the prior art, the reliability of deep-buried tunnel permeability characteristic early warning is relatively low can be solved.

Description

Technical Field

[0001] This application relates to the field of data analysis technology, and more specifically, to an electromagnetic acoustic combined early warning system for the permeability characteristics of deeply buried tunnels. Background Technology

[0002] The permeability characteristics of deeply buried tunnels in complex underground environments are a crucial issue in underground engineering and geological exploration. To monitor and provide early warning of changes in tunnel permeability, a method combining electromagnetic and acoustic technologies promises to offer more accurate and real-time monitoring. Electromagnetic waves can penetrate soil and respond differently to changes in the medium (such as changes in permeability and groundwater flow). Information such as reflection, refraction, or attenuation of electromagnetic waves can reveal anomalies like groundwater flow, soil permeability, tunnel rupture, or seepage. The propagation speed and attenuation characteristics of sound waves in different media vary depending on the physical properties of the soil (such as porosity and permeability). Acoustic wave detection can reveal the health status of the tunnel structure and the impact of seepage (such as crack propagation or changes caused by water flow). However, current technologies have not fully explored and integrated the information from electromagnetic and acoustic waves, resulting in relatively low reliability for permeability characteristic early warning systems. Summary of the Invention

[0003] In view of this, the purpose of this application is to provide an electromagnetic acoustic combined early warning system for the seepage characteristics of deeply buried tunnels, so as to improve the problem of relatively low reliability of early warning of seepage characteristics of deeply buried tunnels in the prior art.

[0004] To achieve the above objectives, this application adopts the following technical solution: An electromagnetic acoustic joint early warning system for the permeability characteristics of deeply buried tunnels includes a processor and a memory. The processor executes a computer program stored in the memory to implement an electromagnetic acoustic joint early warning method for the permeability characteristics of deeply buried tunnels. The electromagnetic acoustic joint early warning method for the permeability characteristics of deeply buried tunnels includes: Acquire electromagnetic wave data and ultrasonic data generated from data collection of deeply buried tunnels; In the frequency domain, harmonic semantic information is used as auxiliary mining information to perform penetration characteristic semantic mining on the electromagnetic wave data, forming an electromagnetic wave penetration characteristic vector. In the process of electromagnetic wave propagation, the nonlinear effect caused by the change in permeability leads to the appearance of harmonics. In the frequency domain, harmonic semantic information is used as auxiliary mining information to perform penetration characteristic semantic mining on the ultrasonic data, forming an ultrasonic penetration characteristic vector. In the process of ultrasonic propagation, the nonlinear effect caused by the change in permeability leads to the appearance of harmonics. The electromagnetic wave penetration characteristic vector and the ultrasonic wave penetration characteristic vector are correlated and semantically extracted to form a multidimensional penetration characteristic vector. Semantic reconstruction is performed based on the multidimensional permeability characteristic vector to form target permeability characteristic data. The target permeability characteristic data is used to reflect the permeability of the deep-buried tunnel and serves as the basis for corresponding early warning operations.

[0005] The electromagnetic-acoustic joint early warning system for the permeability characteristics of deeply buried tunnels provided in this application first acquires electromagnetic wave data and ultrasonic wave data of the deeply buried tunnel; second, in the frequency domain, harmonic semantic information is used as auxiliary mining information to perform permeability characteristic semantic mining on the electromagnetic wave data, forming an electromagnetic wave permeability characteristic vector; then, in the frequency domain, harmonic semantic information is used as auxiliary mining information to perform permeability characteristic semantic mining on the ultrasonic wave data, forming an ultrasonic wave permeability characteristic vector; further, the electromagnetic wave permeability characteristic vector and the ultrasonic wave permeability characteristic vector are correlated with semantic extraction to form a multidimensional permeability characteristic vector; finally, semantic reconstruction is performed based on the multidimensional permeability characteristic vector to form target permeability characteristic data. Based on the above, on the one hand, because harmonic semantic information is used as auxiliary mining information in the frequency domain to perform permeability characteristic semantic mining on the electromagnetic wave data and ultrasonic wave data respectively, semantic information related to permeability characteristics can be fully mined, that is, high-precision semantic mining and characterization of permeability characteristics can be achieved, thereby improving the reliability of the corresponding permeability characteristic early warning. On the other hand, since the semantic information of permeability characteristics mined from both dimensions is further correlated and extracted, rather than simply spliced ​​or superimposed, the accuracy of semantic information representation can be further improved. This, in turn, can further enhance the reliability of subsequent permeability characteristic early warnings. Therefore, it can address the problem of relatively low reliability in existing technologies for permeability characteristic early warnings of deeply buried tunnels. Attached Figure Description

[0006] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings.

[0007] Figure 1 This is a structural block diagram of an electromagnetic acoustic combined early warning system for the permeability characteristics of deeply buried tunnels provided in an embodiment of this application.

[0008] Figure 2 This is a flowchart illustrating the electromagnetic-acoustic combined early warning method for the permeability characteristics of deeply buried tunnels provided in this application embodiment.

[0009] Figure 3 This is a first schematic diagram of penetration feature semantic mining provided in an embodiment of this application.

[0010] Figure 4 This is a schematic diagram illustrating semantic mining of electromagnetic wave frequency domain graphs provided in an embodiment of this application.

[0011] Figure 5 This is a second schematic diagram of penetration feature semantic mining provided in an embodiment of this application.

[0012] Figure 6 This is a schematic diagram illustrating semantic mining of an ultrasonic frequency domain map as provided in an embodiment of this application.

[0013] Figure 7 This is a schematic diagram of semantic extraction provided in an embodiment of this application. Detailed Implementation

[0014] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0015] Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0016] like Figure 1 As shown in the figure, this application provides an electromagnetic acoustic combined early warning system for the seepage characteristics of deeply buried tunnels, which may include a memory and a processor.

[0017] In detail, the memory and the processor are electrically connected directly or indirectly to enable data transmission or interaction. For example, the memory and the processor can be electrically connected via one or more communication buses or signal lines. The processor is used to execute executable computer programs stored in the memory to implement the electromagnetic acoustic joint early warning method for the seepage characteristics of deeply buried tunnels provided in the embodiments of this application.

[0018] Optionally, the memory may be, but is not limited to, random access memory (RAM), read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), etc.

[0019] Optionally, the processor may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), a system on chip (SoC), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0020] Understandable Figure 1 The structure shown is for illustrative purposes only. The electromagnetic acoustic combined early warning system for the permeability characteristics of deeply buried tunnels may also include components that are more advanced than those shown in the diagram. Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown may include, for example, a communication unit for exchanging information with other devices, such as electromagnetic wave sensors and ultrasonic sensors.

[0021] Combination Figure 2 This application also provides an electromagnetic acoustic joint early warning method for the permeability characteristics of deeply buried tunnels, applicable to the aforementioned electromagnetic acoustic joint early warning system for the permeability characteristics of deeply buried tunnels. The method steps defined in the relevant process of the electromagnetic acoustic joint early warning method for the permeability characteristics of deeply buried tunnels can be implemented by the electromagnetic acoustic joint early warning system for the permeability characteristics of deeply buried tunnels (hereinafter referred to as the early warning system). The following will describe... Figure 2 The specific process shown will be explained in detail.

[0022] Step S110: Acquire electromagnetic wave data and ultrasonic data generated from data collection for the deep-buried tunnel.

[0023] In this embodiment of the application, the early warning system can acquire electromagnetic wave data and ultrasonic data generated by data collection on deeply buried tunnels. For example, it can acquire data collected by electromagnetic wave sensors deployed in deeply buried tunnels to obtain electromagnetic wave data. Or, for example, it can acquire data collected by ultrasonic sensors deployed in deeply buried tunnels to obtain ultrasonic data.

[0024] Step S120: In the frequency domain, harmonic semantic information is used as auxiliary mining information to perform penetration characteristic semantic mining on the electromagnetic wave data, forming an electromagnetic wave penetration characteristic vector.

[0025] In this embodiment, after acquiring the electromagnetic wave data, the early warning system can use harmonic semantic information as auxiliary mining information in the frequency domain to perform permeability characteristic semantic mining on the electromagnetic wave data, forming an electromagnetic wave permeability characteristic vector. During electromagnetic wave propagation, the nonlinear effect caused by changes in permeability leads to the appearance of harmonics. Therefore, harmonic semantic information plays an important role in characterizing the permeability characteristics of deeply buried tunnels, enabling reliable permeability characteristic semantic mining based on harmonic semantic information, thereby forming a reliable electromagnetic wave permeability characteristic vector.

[0026] Step S130: In the frequency domain, harmonic semantic information is used as auxiliary mining information to perform penetration characteristic semantic mining on the ultrasonic data, forming an ultrasonic penetration characteristic vector.

[0027] In this embodiment, after acquiring the ultrasonic data, the early warning system can use harmonic semantic information as auxiliary mining information in the frequency domain to perform permeability characteristic semantic mining on the ultrasonic data, forming an ultrasonic permeability characteristic vector. During ultrasonic propagation, the nonlinear effect caused by changes in permeability leads to the appearance of harmonics. Therefore, harmonic semantic information plays an important role in characterizing the permeability characteristics of deeply buried tunnels, enabling reliable permeability characteristic semantic mining based on harmonic semantic information, thereby forming a reliable ultrasonic permeability characteristic vector.

[0028] Step S140: Perform semantic extraction on the electromagnetic wave penetration characteristic vector and the ultrasonic wave penetration characteristic vector to form a multidimensional penetration characteristic vector.

[0029] In this embodiment, after forming the electromagnetic wave penetration characteristic vector and the ultrasonic wave penetration characteristic vector, the early warning system can perform semantic association extraction on the electromagnetic wave penetration characteristic vector and the ultrasonic wave penetration characteristic vector to form a multidimensional penetration characteristic vector. That is, by performing association analysis, associated semantic information can be extracted from the electromagnetic wave penetration characteristic vector and the ultrasonic wave penetration characteristic vector, thus achieving mutual constraint of the semantic information in the two dimensions of the electromagnetic wave penetration characteristic vector and the ultrasonic wave penetration characteristic vector, thereby obtaining a multidimensional penetration characteristic vector with higher semantic representation accuracy.

[0030] Step S150: Semantic restoration is performed based on the multidimensional permeability characteristic vector to form target permeability characteristic data.

[0031] In this embodiment, after forming the multidimensional permeability characteristic vector, the early warning system can perform semantic reconstruction based on the multidimensional permeability characteristic vector to form target permeability characteristic data. The target permeability characteristic data reflects the permeability of the deep-buried tunnel and serves as the basis for corresponding early warning operations. For example, the target permeability characteristic data can be a probability prediction or classification result of permeability risk; for instance, it can be a continuous value (representing the degree of permeability change) or a discrete category (such as normal, slight permeability, severe permeability, etc.). Thus, corresponding early warning notifications can be issued in cases of severe permeability.

[0032] Based on the above, on the one hand, since harmonic semantic information is used as auxiliary information in the frequency domain to perform semantic mining of permeability characteristics on electromagnetic wave data and ultrasonic data respectively, it is possible to fully extract semantic information related to permeability characteristics, that is, to achieve high-precision semantic mining and characterization of permeability characteristics, thereby improving the reliability of the corresponding permeability characteristic early warning. On the other hand, since the permeability characteristic semantic information mined from the two dimensions is further correlated with semantic extraction, rather than simply splicing or superimposing, the accuracy of semantic information characterization can be further improved, thus further improving the reliability of subsequent permeability characteristic early warning. Therefore, it can improve the problem of relatively low reliability of permeability characteristic early warning for deep buried tunnels in existing technologies.

[0033] Firstly, regarding step S110, it should be noted that the specific methods for acquiring electromagnetic wave data and ultrasonic data are not limited and can be selected according to actual needs.

[0034] For example, in one alternative implementation, electromagnetic wave data and ultrasonic data collected by electromagnetic wave sensors and ultrasonic sensors can be acquired in real time. In another alternative implementation, stored electromagnetic wave data and ultrasonic data can be retrieved from a connected database. That is, the electromagnetic wave data and ultrasonic data collected by the electromagnetic wave sensors and ultrasonic sensors can be stored in the database first, and then retrieved from the database when early warning analysis is needed to obtain the corresponding electromagnetic wave data and ultrasonic data.

[0035] Secondly, regarding step S120, it should be noted that the specific method for performing penetration characteristic semantic mining on the electromagnetic wave data is not limited and can be selected according to actual needs.

[0036] For example, in an alternative implementation, in order to capture more detailed semantic information during the process of semantic mining of penetration characteristics, thereby improving the semantic representation capability of the formed electromagnetic wave penetration characteristic vector, the above step S120 may further include steps S121, S122, S123, S124 and S125, the specific contents of which are as follows.

[0037] Step S121: Perform Fourier processing on the electromagnetic wave data in the time domain to form an electromagnetic wave frequency domain diagram, identify the fundamental frequency of the electromagnetic wave from the electromagnetic wave frequency domain diagram, and determine at least one harmonic frequency of the electromagnetic wave frequency domain diagram based on the fundamental frequency of the electromagnetic wave.

[0038] In the embodiments of this application, combined with Figure 3 The electromagnetic wave data in the time domain can be subjected to Fourier processing to form an electromagnetic wave frequency domain diagram. The fundamental frequency of the electromagnetic wave can be identified from the electromagnetic wave frequency domain diagram. Based on the fundamental frequency of the electromagnetic wave, at least one harmonic frequency of the electromagnetic wave frequency domain diagram can be determined. For example, the fundamental frequency of the electromagnetic wave can be f0, and correspondingly, the harmonic frequencies can be 2f0, 3f0, 4f0, 5f0, 6f0, 7f0, etc. The specific number can be selected according to actual needs.

[0039] Step S122: Based on the at least one harmonic frequency, the electromagnetic wave frequency domain diagram is masked to form an electromagnetic harmonic frequency domain diagram.

[0040] In this embodiment, the electromagnetic wave frequency domain map can be masked based on the at least one harmonic frequency to form an electromagnetic harmonic frequency domain map. This electromagnetic harmonic frequency domain map contains information about the at least one harmonic frequency but not information about other frequencies. In other words, information about all frequencies other than the at least one harmonic frequency in the electromagnetic wave frequency domain map can be hidden, thus obtaining the electromagnetic harmonic frequency domain map.

[0041] Step S123: Perform semantic mining on the electromagnetic wave frequency domain map to form an electromagnetic wave frequency domain vector.

[0042] In this embodiment of the application, after obtaining the electromagnetic wave frequency domain map, semantic mining can be performed on the electromagnetic wave frequency domain map to form an electromagnetic wave frequency domain vector. That is, by mining the potential semantic information in the electromagnetic wave frequency domain map and representing it in the form of a vector, the electromagnetic wave frequency domain vector is obtained.

[0043] Step S124: Perform semantic mining on the electromagnetic harmonic frequency domain map to form an electromagnetic harmonic frequency domain vector.

[0044] In this embodiment of the application, after obtaining the electromagnetic harmonic frequency domain map, semantic mining can be performed on the electromagnetic harmonic frequency domain map to form an electromagnetic harmonic frequency domain vector. That is, by mining the potential semantic information in the electromagnetic harmonic frequency domain map and representing it in the form of a vector, the electromagnetic harmonic frequency domain vector is obtained.

[0045] Step S125: Using the electromagnetic harmonic frequency domain vector as auxiliary mining information, perform penetration characteristic semantic mining on the electromagnetic frequency domain vector to form an electromagnetic wave penetration characteristic vector.

[0046] In this embodiment, after obtaining the electromagnetic harmonic frequency domain vector and the electromagnetic wave frequency domain vector, the electromagnetic harmonic frequency domain vector can be used as auxiliary mining information to perform penetration characteristic semantic mining on the electromagnetic wave frequency domain vector, forming an electromagnetic wave penetration characteristic vector. That is, since the electromagnetic harmonic frequency domain vector has a better characterization ability for penetration characteristics, using it as auxiliary mining information can improve the accuracy of penetration characteristic semantic mining on the electromagnetic wave frequency domain vector.

[0047] It is understood that the specific method of semantic mining of the electromagnetic wave frequency domain map in step S123 above is not limited. For example, in an alternative implementation, in order to ensure the reliability of semantic mining and make the semantic information represented by the electromagnetic wave frequency domain vector richer, step S123 above may further include steps S123a, S123b, S123c and S123d, the specific contents of which are as follows.

[0048] Step S123a: Perform convolution mining on the electromagnetic wave frequency domain map to form an electromagnetic wave frequency domain convolution vector.

[0049] In the embodiments of this application, combined with Figure 4The electromagnetic wave frequency domain graph can be convolutionally mined to form electromagnetic wave frequency domain convolution vectors. This convolutional mining can be implemented using convolutional network layers.

[0050] Step S123b: The electromagnetic wave frequency domain map is segmented, and each local frequency domain map obtained by the segmentation process is convolutionally mined to form a local frequency domain convolution vector sequence.

[0051] In this embodiment, the electromagnetic wave frequency domain map can be segmented, and each segmented local frequency domain map can be subjected to convolution mining to form a local frequency domain convolution vector sequence. For example, the electromagnetic wave frequency domain map can be segmented to the same size, resulting in local frequency domain maps of the same size. Furthermore, the number of local frequency domain maps formed by the segmentation can be greater than or equal to two, thus the number of local frequency domain convolution vectors included in the local frequency domain convolution vector sequence is also greater than or equal to two. Additionally, in the local frequency domain convolution vector sequence, the local frequency domain maps corresponding to two adjacent local frequency domain convolution vectors are also adjacent in the electromagnetic wave frequency domain map. This facilitates semantic enhancement based on association relationships in subsequent steps.

[0052] Step S123c: For each local frequency domain convolution vector in the local frequency domain convolution vector sequence, based on the spatial enhancement convolution vector corresponding to the local frequency domain convolution vector, perform semantic enhancement processing on the local frequency domain convolution vector to form the local frequency domain enhancement vector corresponding to the local frequency domain convolution vector.

[0053] In this embodiment, after forming the local frequency domain convolution vector sequence, semantic enhancement processing can be performed on each local frequency domain convolution vector in the sequence, based on the spatial enhancement convolution vector corresponding to that local frequency domain convolution vector, to form a local frequency domain enhancement vector corresponding to that local frequency domain convolution vector. Specifically, the spatial enhancement convolution vector corresponding to the first local frequency domain convolution vector is that local frequency domain convolution vector, and the spatial enhancement convolution vectors corresponding to other local frequency domain convolution vectors are the local frequency domain enhancement vectors corresponding to the preceding local frequency domain convolution vector. That is, the semantic enhancement processing can be performed sequentially according to the order in the local frequency domain convolution vector sequence, such as performing semantic enhancement processing on the first local frequency domain convolution vector first, and then performing semantic enhancement processing on the second local frequency domain convolution vector. Furthermore, the semantic enhancement processing can be implemented based on an attention mechanism, that is, through the association mining of the attention mechanism, the semantic information of two adjacent local frequency domain maps can be fused.

[0054] Step S123d: Based on the local frequency domain enhancement vector corresponding to the last local frequency domain convolution vector, perform global semantic enhancement processing on the electromagnetic wave frequency domain convolution vector to form an electromagnetic wave frequency domain vector.

[0055] In this embodiment, after obtaining the local frequency domain enhancement vector corresponding to the last local frequency domain convolution vector, global semantic enhancement processing can be performed on the electromagnetic wave frequency domain convolution vector based on the local frequency domain enhancement vector corresponding to the last local frequency domain convolution vector to form an electromagnetic wave frequency domain vector. That is, based on the attention mechanism, the local frequency domain enhancement vector corresponding to the last local frequency domain convolution vector can be fused into the electromagnetic wave frequency domain convolution vector. In this way, it can be ensured that the formed electromagnetic wave frequency domain vector can take into account both the accuracy and comprehensiveness of semantic information representation.

[0056] It is understood that the specific method of semantic mining of the electromagnetic harmonic frequency domain map in step S124 above is not limited. For example, in an alternative implementation, in order to ensure the reliability of semantic mining and make the semantic information represented by the formed electromagnetic harmonic frequency domain vector richer, step S124 above may further include steps S124a, S124b, S124c and S124d, the specific contents of which are as follows.

[0057] Step S124a: Perform convolution mining on the electromagnetic harmonic frequency domain map to form an electromagnetic harmonic frequency domain convolution vector.

[0058] In this embodiment of the application, convolution mining can be performed on the electromagnetic harmonic frequency domain map to form electromagnetic harmonic frequency domain convolution vectors. Convolution mining can be implemented using convolutional network layers.

[0059] Step S124b: The electromagnetic harmonic frequency domain map is segmented, and each local harmonic frequency domain map obtained by the segmentation process is convolutionally mined to form a local harmonic convolution vector sequence.

[0060] In this embodiment, the electromagnetic harmonic frequency domain map can be segmented, and each segmented local harmonic frequency domain map can be subjected to convolution mining to form a local harmonic convolution vector sequence. For example, the electromagnetic harmonic frequency domain map can be segmented to the same size, ensuring that the resulting local harmonic frequency domain maps are all the same size. Furthermore, the number of local harmonic frequency domain maps formed by the segmentation can be greater than or equal to two. Thus, the number of local harmonic convolution vectors included in the local harmonic convolution vector sequence is also greater than or equal to two. Additionally, in the local harmonic convolution vector sequence, the local harmonic frequency domain maps corresponding to two adjacent local harmonic convolution vectors are also adjacent in the electromagnetic harmonic frequency domain map. This facilitates semantic enhancement based on association relationships in subsequent steps.

[0061] Step S124c: For each local harmonic convolution vector in the local harmonic convolution vector sequence, based on the spatial enhancement convolution vector corresponding to the local harmonic convolution vector, perform semantic enhancement processing on the local harmonic convolution vector to form the local frequency domain enhancement vector corresponding to the local harmonic convolution vector.

[0062] In this embodiment, after forming the local harmonic convolution vector sequence, semantic enhancement processing can be performed on each local harmonic convolution vector in the sequence, based on the spatial enhancement convolution vector corresponding to that local harmonic convolution vector, to form a local frequency domain enhancement vector corresponding to that local harmonic convolution vector. Specifically, the spatial enhancement convolution vector corresponding to the first local harmonic convolution vector is that local harmonic convolution vector, and the spatial enhancement convolution vectors corresponding to other local harmonic convolution vectors are the local frequency domain enhancement vectors corresponding to the preceding local harmonic convolution vector. That is, the semantic enhancement processing can be performed sequentially according to the order in the local harmonic convolution vector sequence, such as performing semantic enhancement processing on the first local harmonic convolution vector first, and then performing semantic enhancement processing on the second local harmonic convolution vector. Furthermore, the semantic enhancement processing can be implemented based on an attention mechanism, that is, through the association mining of the attention mechanism, the semantic information of two adjacent local harmonic frequency domain maps can be fused.

[0063] Step S124d: Based on the local frequency domain enhancement vector corresponding to the last local harmonic convolution vector, perform global semantic enhancement processing on the electromagnetic harmonic frequency domain convolution vector to form an electromagnetic harmonic frequency domain vector.

[0064] In this embodiment, global semantic enhancement processing can be performed on the electromagnetic harmonic frequency domain convolution vector based on the local frequency domain enhancement vector corresponding to the last local harmonic convolution vector to form an electromagnetic harmonic frequency domain vector. That is, based on an attention mechanism, the local frequency domain enhancement vector corresponding to the last local harmonic convolution vector can be fused into the electromagnetic harmonic frequency domain convolution vector. This ensures that the resulting electromagnetic harmonic frequency domain vector maintains both the accuracy and comprehensiveness of semantic information representation.

[0065] It is understood that the specific method of performing penetration characteristic semantic mining on the electromagnetic wave frequency domain vector in step S125 above is not limited. For example, in an alternative implementation, in order to ensure the reliability of penetration characteristic semantic mining, step S125 above may further include steps S125a, S125b, S125c and S125d, the specific contents of which are as follows.

[0066] Step S125a: Gated mapping is performed on the electromagnetic harmonic frequency domain vector to form the first gated mapping parameter.

[0067] In this embodiment, the electromagnetic harmonic frequency domain vector can be gated to form a first gated mapping parameter. For example, the electromagnetic harmonic frequency domain vector can first be linearly mapped (e.g., through a fully connected network layer), and then the result of the linear mapping can be nonlinearly activated (e.g., through a sigmoid function) to obtain the first gated mapping parameter. The parameters in the first gated mapping parameter are between 0 and 1, thus representing the importance of corresponding positions. This allows for appropriate gating adjustments in subsequent steps, such as retaining important semantic information while intercepting or suppressing unimportant or low-importance semantic information.

[0068] Step S125b: Perform self-attention processing on the electromagnetic harmonic frequency domain vector to form an electromagnetic wave attention vector, and perform gated mapping on the electromagnetic wave attention vector to form a second gated mapping parameter.

[0069] In this embodiment, considering the high quality requirements of the gating mechanism for the electromagnetic harmonic frequency domain vector, self-attention processing can be applied to the electromagnetic harmonic frequency domain vector to form an electromagnetic wave attention vector. Then, gating mapping can be applied to the electromagnetic wave attention vector to form a second gating mapping parameter. In other words, self-attention processing is first used to achieve preliminary screening of the main semantic information, and then gating mapping is used to achieve high-precision mapping.

[0070] Step S125c: Fuse the first gating mapping parameters and the second gating mapping parameters to form auxiliary gating mapping parameters.

[0071] In this embodiment of the application, after the first gated mapping parameter and the second gated mapping parameter are formed, the first gated mapping parameter and the second gated mapping parameter can be fused to form an auxiliary gated mapping parameter. For example, the first gated mapping parameter and the second gated mapping parameter can be averaged or weighted averaged to obtain the auxiliary gated mapping parameter.

[0072] Step S125d: Based on the auxiliary gating mapping parameters, the electromagnetic wave frequency domain vector is gating and adjusted to realize semantic mining of penetration characteristics and form an electromagnetic wave penetration characteristic vector.

[0073] In this embodiment, after obtaining the auxiliary gating mapping parameters, the electromagnetic wave frequency domain vector can be gated and adjusted based on the auxiliary gating mapping parameters to achieve semantic mining of penetration characteristics and form an electromagnetic wave penetration characteristic vector. For example, the auxiliary gating mapping parameters and the electromagnetic wave frequency domain vector can be multiplied bitwise to retain, suppress, or intercept semantic information of different importance levels, thereby forming an electromagnetic wave penetration characteristic vector.

[0074] Thirdly, regarding step S130, it should be noted that the specific method for performing penetration characteristic semantic mining on the ultrasonic data is not limited and can be selected according to actual needs.

[0075] For example, in an alternative implementation, in order to capture more detailed semantic information during the semantic mining of permeability characteristics, thereby improving the semantic representation capability of the formed ultrasonic permeability characteristic vector, the above step S130 may further include steps S131, S132, S133, S134 and S135, as detailed below.

[0076] Step S131: Perform Fourier processing on the time-domain ultrasonic data to form an ultrasonic frequency domain graph, identify the fundamental frequency of the ultrasonic wave from the ultrasonic frequency domain graph, and determine at least one harmonic frequency of the ultrasonic frequency domain graph based on the fundamental frequency of the ultrasonic wave.

[0077] In the embodiments of this application, combined with Figure 5 The ultrasonic data in the time domain can be subjected to Fourier processing to form an ultrasonic frequency domain graph. The fundamental frequency of the ultrasonic wave can be identified from the ultrasonic frequency domain graph. Based on the fundamental frequency of the ultrasonic wave, at least one harmonic frequency of the ultrasonic frequency domain graph can be determined. For example, the fundamental frequency of the electromagnetic wave can be f1, and correspondingly, the harmonic frequencies can be 2f1, 3f1, 4f1, 5f1, 6f1, 7f1, etc. The specific number can be selected according to actual needs.

[0078] Step S132: Based on the at least one harmonic frequency, the ultrasonic frequency domain diagram is masked to form an ultrasonic harmonic frequency domain diagram.

[0079] In this embodiment, after forming the at least one harmonic frequency, the ultrasonic frequency domain map can be masked based on the at least one harmonic frequency to form an ultrasonic harmonic frequency domain map. The ultrasonic harmonic frequency domain map contains information about the at least one harmonic frequency but not information about other frequencies. In other words, information about all frequencies other than the at least one harmonic frequency in the ultrasonic frequency domain map can be hidden, thus obtaining the ultrasonic harmonic frequency domain map.

[0080] Step S133: Perform semantic mining on the ultrasonic frequency domain map to form an ultrasonic frequency domain vector.

[0081] In this embodiment of the application, after obtaining the ultrasonic frequency domain map, semantic mining can be performed on the ultrasonic frequency domain map to form an ultrasonic frequency domain vector. That is, by mining the potential semantic information in the ultrasonic harmonic frequency domain map and representing it in the form of a vector, the ultrasonic harmonic frequency domain vector is obtained.

[0082] Step S134: Perform semantic mining on the ultrasonic harmonic frequency domain map to form ultrasonic harmonic frequency domain vectors.

[0083] In this embodiment of the application, after obtaining the ultrasonic harmonic frequency domain map, semantic mining can be performed on the ultrasonic harmonic frequency domain map to form an ultrasonic harmonic frequency domain vector. That is, by mining the potential semantic information in the ultrasonic harmonic frequency domain map and representing it in the form of a vector, the ultrasonic harmonic frequency domain vector is obtained.

[0084] Step S135: Using the ultrasonic harmonic frequency domain vector as auxiliary mining information, perform semantic mining of the ultrasonic frequency domain vector to form an ultrasonic penetration characteristic vector.

[0085] In this embodiment, after obtaining the ultrasonic harmonic frequency domain vector and the ultrasonic frequency domain vector, the ultrasonic harmonic frequency domain vector can be used as auxiliary mining information to perform semantic mining of penetration characteristics on the ultrasonic frequency domain vector, forming an ultrasonic penetration characteristic vector. That is, since the ultrasonic harmonic frequency domain vector has a better characterization ability for penetration characteristics, using it as auxiliary mining information can improve the accuracy of semantic mining of penetration characteristics from the ultrasonic frequency domain vector.

[0086] It is understood that the specific method of semantic mining of the ultrasonic frequency domain map in step S133 above is not limited. For example, in an alternative implementation, in order to ensure the reliability of semantic mining and make the semantic information represented by the formed ultrasonic frequency domain vector richer, step S133 above may further include steps S133a, S133b, S133c and S133d, the specific contents of which are as follows.

[0087] Step S133a: Perform convolution mining on the ultrasonic frequency domain map to form an ultrasonic frequency domain convolution vector.

[0088] In the embodiments of this application, combined with Figure 6 The ultrasonic frequency domain map can be convolutionally mined to form ultrasonic frequency domain convolution vectors. This convolutional mining can be implemented using convolutional network layers.

[0089] Step S133b: The ultrasonic frequency domain map is segmented, and each local frequency domain map obtained by the segmentation process is convolutionally mined to form a local frequency domain convolution vector sequence.

[0090] In this embodiment, the ultrasonic frequency domain map can be segmented, and each segmented local frequency domain map can be subjected to convolution mining to form a local frequency domain convolution vector sequence. For example, the ultrasonic frequency domain map can be segmented to the same size, resulting in local frequency domain maps of the same size. Furthermore, the number of local frequency domain maps formed by the segmentation can be greater than or equal to two, thus the number of local frequency domain convolution vectors included in the local frequency domain convolution vector sequence is also greater than or equal to two. Additionally, in the local frequency domain convolution vector sequence, the local frequency domain maps corresponding to two adjacent local frequency domain convolution vectors are also adjacent in the ultrasonic frequency domain map. This facilitates semantic enhancement based on association relationships in subsequent steps.

[0091] Step S133c: For each local frequency domain convolution vector in the local frequency domain convolution vector sequence, based on the spatial enhancement convolution vector corresponding to the local frequency domain convolution vector, perform semantic enhancement processing on the local frequency domain convolution vector to form the local frequency domain enhancement vector corresponding to the local frequency domain convolution vector.

[0092] In this embodiment, after forming the local frequency domain convolution vector sequence, semantic enhancement processing can be performed on each local frequency domain convolution vector in the sequence, based on the spatial enhancement convolution vector corresponding to that local frequency domain convolution vector, to form a local frequency domain enhancement vector corresponding to that local frequency domain convolution vector. Specifically, the spatial enhancement convolution vector corresponding to the first local frequency domain convolution vector is that local frequency domain convolution vector, and the spatial enhancement convolution vectors corresponding to other local frequency domain convolution vectors are the local frequency domain enhancement vectors corresponding to the preceding local frequency domain convolution vector. That is, the semantic enhancement processing can be performed sequentially according to the order in the local frequency domain convolution vector sequence, such as performing semantic enhancement processing on the first local frequency domain convolution vector first, and then performing semantic enhancement processing on the second local frequency domain convolution vector. Furthermore, the semantic enhancement processing can be implemented based on an attention mechanism, that is, through the association mining of the attention mechanism, the semantic information of two adjacent local frequency domain maps can be fused.

[0093] Step S133d: Based on the local frequency domain enhancement vector corresponding to the last local frequency domain convolution vector, perform global semantic enhancement processing on the ultrasonic frequency domain convolution vector to form an ultrasonic frequency domain vector.

[0094] In this embodiment, after obtaining the local frequency domain enhancement vector corresponding to the last local frequency domain convolution vector, global semantic enhancement processing can be performed on the ultrasonic frequency domain convolution vector based on the local frequency domain enhancement vector corresponding to the last local frequency domain convolution vector to form an ultrasonic frequency domain vector. That is, based on the attention mechanism, the local frequency domain enhancement vector corresponding to the last local frequency domain convolution vector can be fused into the ultrasonic frequency domain convolution vector. In this way, it can be ensured that the formed ultrasonic frequency domain vector can take into account both the accuracy and comprehensiveness of semantic information representation.

[0095] It is understood that the specific method of semantic mining of the ultrasonic harmonic frequency domain map in step S134 above is not limited. For example, in an alternative implementation, in order to ensure the reliability of semantic mining and make the semantic information represented by the formed ultrasonic harmonic frequency domain vector richer, step S134 above may further include steps S134a, S134b, S134c and S134d, the specific contents of which are as follows.

[0096] Step S134a: Perform convolution mining on the ultrasonic harmonic frequency domain map to form an ultrasonic harmonic frequency domain convolution vector.

[0097] In this embodiment of the application, convolution mining can be performed on the ultrasonic harmonic frequency domain map to form ultrasonic harmonic frequency domain convolution vectors. Convolution mining can be implemented using convolutional network layers.

[0098] Step S134b: The ultrasonic harmonic frequency domain map is segmented, and each local harmonic frequency domain map obtained by the segmentation process is convolutionally mined to form a local harmonic convolution vector sequence.

[0099] In this embodiment, the ultrasonic harmonic frequency domain map can be segmented, and each segmented local harmonic frequency domain map can be convolutionally mined to form a local harmonic convolution vector sequence. For example, the ultrasonic harmonic frequency domain map can be segmented to the same size, ensuring that the resulting local harmonic frequency domain maps are all the same size. Furthermore, the number of local harmonic frequency domain maps formed by the segmentation can be greater than or equal to two. Thus, the number of local harmonic convolution vectors included in the local harmonic convolution vector sequence is also greater than or equal to two. Additionally, in the local harmonic convolution vector sequence, the local harmonic frequency domain maps corresponding to two adjacent local harmonic convolution vectors are also adjacent in the ultrasonic harmonic frequency domain map. This facilitates semantic enhancement based on association relationships in subsequent steps.

[0100] Step S134c: For each local harmonic convolution vector in the local harmonic convolution vector sequence, based on the spatial enhancement convolution vector corresponding to the local harmonic convolution vector, perform semantic enhancement processing on the local harmonic convolution vector to form the local frequency domain enhancement vector corresponding to the local harmonic convolution vector.

[0101] In this embodiment, after obtaining the local harmonic convolution vector sequence, semantic enhancement processing can be performed on each local harmonic convolution vector in the sequence, based on the spatial enhancement convolution vector corresponding to that local harmonic convolution vector, to form a local frequency domain enhancement vector corresponding to that local harmonic convolution vector. Specifically, the spatial enhancement convolution vector corresponding to the first local harmonic convolution vector is that local harmonic convolution vector, and the spatial enhancement convolution vectors corresponding to other local harmonic convolution vectors are the local frequency domain enhancement vectors corresponding to the preceding local harmonic convolution vector. That is, the semantic enhancement processing can be performed sequentially according to the order in the local harmonic convolution vector sequence, such as performing semantic enhancement processing on the first local harmonic convolution vector first, and then performing semantic enhancement processing on the second local harmonic convolution vector. Furthermore, the semantic enhancement processing can be implemented based on an attention mechanism, that is, through the association mining of the attention mechanism, the semantic information of two adjacent local harmonic frequency domain maps can be fused.

[0102] Step S134d: Based on the local frequency domain enhancement vector corresponding to the last local harmonic convolution vector, perform global semantic enhancement processing on the ultrasonic harmonic frequency domain convolution vector to form an ultrasonic harmonic frequency domain vector.

[0103] In this embodiment, after obtaining the local frequency domain enhancement vector corresponding to the last local harmonic convolution vector, global semantic enhancement processing can be performed on the ultrasonic harmonic frequency domain convolution vector based on the local frequency domain enhancement vector corresponding to the last local harmonic convolution vector to form an ultrasonic harmonic frequency domain vector. That is, based on an attention mechanism, the local frequency domain enhancement vector corresponding to the last local harmonic convolution vector can be fused into the ultrasonic harmonic frequency domain convolution vector. This ensures that the resulting ultrasonic harmonic frequency domain vector can balance both the accuracy and comprehensiveness of semantic information representation.

[0104] It is understood that the specific method of performing penetration characteristic semantic mining on the ultrasonic frequency domain vector in step S135 above is not limited. For example, in an alternative implementation, in order to ensure the reliability of penetration characteristic semantic mining, step S135 above may further include steps S135a, S135b, S135c and S135d, the specific contents of which are as follows.

[0105] Step S135a: Gated mapping is performed on the ultrasonic harmonic frequency domain vector to form the first gated mapping parameter.

[0106] In this embodiment, the ultrasonic harmonic frequency domain vector can be gated to form a first gated mapping parameter. For example, the ultrasonic harmonic frequency domain vector can first be linearly mapped (e.g., through a fully connected network layer), and then the result of the linear mapping can be nonlinearly activated (e.g., through a sigmoid function) to obtain the first gated mapping parameter. The parameters in the first gated mapping parameter are between 0 and 1, thus representing the importance of the corresponding position. This allows for appropriate gating adjustments in subsequent steps, such as retaining important semantic information while intercepting or suppressing unimportant or low-importance semantic information.

[0107] Step S135b: Perform self-attention processing on the ultrasonic harmonic frequency domain vector to form an ultrasonic attention vector, and perform gating mapping on the ultrasonic attention vector to form a second gating mapping parameter.

[0108] In this embodiment, considering the high quality requirements of the gating mechanism for the ultrasonic harmonic frequency domain vector, self-attention processing can be applied to the ultrasonic harmonic frequency domain vector to form an ultrasonic attention vector, and gating mapping can be performed on the ultrasonic attention vector to form a second gating mapping parameter. That is, self-attention processing is first used to achieve preliminary screening of the main semantic information, and then gating mapping is used to achieve high-precision mapping.

[0109] Step S135c: Merge the first gating mapping parameters and the second gating mapping parameters to form auxiliary gating mapping parameters.

[0110] In this embodiment of the application, after the first gated mapping parameter and the second gated mapping parameter are formed, the first gated mapping parameter and the second gated mapping parameter can be fused to form an auxiliary gated mapping parameter. For example, the first gated mapping parameter and the second gated mapping parameter can be averaged or weighted averaged to obtain the auxiliary gated mapping parameter.

[0111] Step S135d: Based on the auxiliary gating mapping parameters, the ultrasonic frequency domain vector is gating and adjusted to realize semantic mining of penetration characteristics and form an ultrasonic penetration characteristic vector.

[0112] In this embodiment, after obtaining the auxiliary gating mapping parameters, the ultrasonic frequency domain vector can be gated and adjusted based on the auxiliary gating mapping parameters to achieve semantic mining of penetration characteristics and form an ultrasonic penetration characteristic vector. For example, the auxiliary gating mapping parameters and the ultrasonic frequency domain vector can be multiplied bitwise to retain, suppress, or intercept semantic information of different importance levels, thereby forming an ultrasonic penetration characteristic vector.

[0113] Fourthly, regarding step S140, it should be noted that the specific method for semantically extracting the electromagnetic wave penetration characteristic vector and the ultrasonic wave penetration characteristic vector is not limited and can be selected according to actual needs.

[0114] For example, in an alternative implementation, in order to avoid overfitting due to excessive network depth while fully extracting the associated semantic information, the above step S140 may further include steps S141, S142, S143, S144 and S145, the specific contents of each step are as follows.

[0115] Step S141: Obtain the attention hidden matrix sequence.

[0116] In the embodiments of this application, combined with Figure 7 An attention hidden matrix sequence can be obtained. In this sequence, the number of non-zero elements in each attention hidden matrix gradually increases. It should be noted that each attention hidden matrix can be formed based on a combination of zero and non-zero elements (such as 1), and each attention hidden matrix can be formed during the training of a corresponding neural network model, i.e., by learning from the labels of sample electromagnetic wave data, sample ultrasonic data, and permeability characteristics data.

[0117] Step S142: Use either the electromagnetic wave penetration characteristic vector or the ultrasonic wave penetration characteristic vector as the first semantic vector and the other vector as the second semantic vector.

[0118] In this embodiment of the application, either the electromagnetic wave penetration characteristic vector or the ultrasonic wave penetration characteristic vector can be used as a first semantic vector and the other vector can be used as a second semantic vector. For example, the electromagnetic wave penetration characteristic vector can be used as the first semantic vector and the ultrasonic wave penetration characteristic vector can be used as the second semantic vector.

[0119] Step S143: In the first stage of associated semantic extraction, the original attention parameters between the first semantic vector and the second semantic vector are determined, and associated semantic extraction is performed on the second semantic vector based on the original attention parameters to form the associated semantic vector of the first stage.

[0120] In the embodiments of this application, in the first stage of associated semantic extraction, the original attention parameters between the first semantic vector and the second semantic vector (such as those obtained by dot product calculation) can be determined, and based on the original attention parameters, associated semantic extraction is performed on the second semantic vector (such as by weighted summation calculation) to form the associated semantic vector of the first stage.

[0121] Step S144: In each intermediate stage of the associated semantic extraction, the original attention parameters between the associated semantic vector of the previous stage and the second semantic vector are determined. After adjusting the original attention parameters based on the corresponding attention hidden matrix in the attention hidden matrix sequence, associated semantic extraction is performed on the second semantic vector based on the obtained target attention parameters to form the associated semantic vector of the corresponding stage.

[0122] In each intermediate stage of the associated semantic extraction, the original attention parameters between the associated semantic vector of the previous stage and the second semantic vector can be determined. After adjusting the original attention parameters based on the corresponding attention hidden matrix in the attention hidden matrix sequence (such as by bitwise multiplication), the associated semantics of the second semantic vector are extracted based on the obtained target attention parameters to form the associated semantic vector of the corresponding stage.

[0123] Step S145: In the last stage of the associated semantic extraction, the original attention parameters between the associated semantic vector of the previous stage and the second semantic vector are determined, and associated semantic extraction is performed on the second semantic vector based on the original attention parameters to form the associated semantic vector of the last stage, and the multidimensional penetration characteristic vector is determined based on the associated semantic vector.

[0124] In this embodiment, in the final stage of associated semantic extraction, the original attention parameters between the associated semantic vector from the previous stage and the second semantic vector can be determined. Based on these original attention parameters, associated semantic extraction is performed on the second semantic vector to form the associated semantic vector for the final stage. Furthermore, a multi-dimensional penetration characteristic vector is determined based on this associated semantic vector (e.g., the associated semantic vector can be determined as a multi-dimensional penetration characteristic vector). The difference between the number of stages of associated semantic extraction and the number of attention hiding matrices is equal to 2. That is, in the second stage of associated semantic extraction, the original attention parameters can be adjusted based on the first attention hiding matrix in the attention hiding matrix sequence. In the third stage of associated semantic extraction, the original attention parameters can be adjusted based on the second attention hiding matrix in the attention hiding matrix sequence. In the nth stage of associated semantic extraction, the original attention parameters can be adjusted based on the (n-1)th attention hiding matrix in the attention hiding matrix sequence.

[0125] Fifthly, regarding step S150, it should be noted that the specific method of semantic restoration based on the multidimensional penetration characteristic vector is not limited and can be selected according to actual needs.

[0126] For example, in an alternative implementation, step S150 described above may include: First, the multidimensional penetration characteristic vector can be fully connected to obtain a fully connected vector. When classification output is required, the size of the fully connected vector can be determined based on the number of penetration types. When linear output is required (i.e., a continuous value to represent the degree of penetration, such as 0-1), the size of the fully connected vector can be 1*1. Secondly, a linear or identity mapping is performed on the fully connected vector to obtain the target permeability characteristic data (corresponding to the linear output); or, a probability mapping is performed on the fully connected vector (such as through a function like softmax) to obtain a probability distribution. Then, the type of permeation corresponding to the highest probability in the probability distribution can be determined as the target permeability characteristic data.

[0127] In summary, the electromagnetic-acoustic joint early warning system for the permeability characteristics of deeply buried tunnels provided in this application first acquires electromagnetic wave data and ultrasonic wave data of the deeply buried tunnel; second, in the frequency domain, harmonic semantic information is used as auxiliary mining information to perform permeability characteristic semantic mining on the electromagnetic wave data, forming an electromagnetic wave permeability characteristic vector; then, in the frequency domain, harmonic semantic information is used as auxiliary mining information to perform permeability characteristic semantic mining on the ultrasonic wave data, forming an ultrasonic wave permeability characteristic vector; further, the electromagnetic wave permeability characteristic vector and the ultrasonic wave permeability characteristic vector are correlated with semantic extraction to form a multidimensional permeability characteristic vector; finally, semantic restoration is performed based on the multidimensional permeability characteristic vector to form target permeability characteristic data. Based on the above, on the one hand, because harmonic semantic information is used as auxiliary mining information in the frequency domain to perform permeability characteristic semantic mining on the electromagnetic wave data and ultrasonic wave data respectively, semantic information related to permeability characteristics can be fully mined, that is, high-precision semantic mining and characterization of permeability characteristics can be achieved, thereby improving the reliability of the corresponding permeability characteristic early warning. On the other hand, since the semantic information of permeability characteristics mined from both dimensions is further correlated and extracted, rather than simply spliced ​​or superimposed, the accuracy of semantic information representation can be further improved. This, in turn, can further enhance the reliability of subsequent permeability characteristic early warnings. Therefore, it can address the problem of relatively low reliability in existing technologies for permeability characteristic early warnings of deeply buried tunnels.

[0128] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. An electromagnetic acoustic combined early warning system for the permeability characteristics of deeply buried tunnels, characterized in that, The system includes a processor and a memory, the processor executing a computer program stored in the memory to implement an electromagnetic-acoustic joint early warning method for the permeability characteristics of deeply buried tunnels, wherein the electromagnetic-acoustic joint early warning method for the permeability characteristics of deeply buried tunnels includes: Acquire electromagnetic wave data and ultrasonic data generated from data collection of deeply buried tunnels; In the frequency domain, harmonic semantic information is used as auxiliary mining information to perform penetration characteristic semantic mining on the electromagnetic wave data, forming an electromagnetic wave penetration characteristic vector. In the process of electromagnetic wave propagation, the nonlinear effect caused by the change in permeability leads to the appearance of harmonics. In the frequency domain, harmonic semantic information is used as auxiliary mining information to perform penetration characteristic semantic mining on the ultrasonic data, forming an ultrasonic penetration characteristic vector. In the process of ultrasonic propagation, the nonlinear effect caused by the change in permeability leads to the appearance of harmonics. The electromagnetic wave penetration characteristic vector and the ultrasonic wave penetration characteristic vector are correlated and semantically extracted to form a multidimensional penetration characteristic vector. Semantic reconstruction is performed based on the multidimensional permeability characteristic vector to form target permeability characteristic data. The target permeability characteristic data is used to reflect the permeability of the deep-buried tunnel and serves as the basis for corresponding early warning operations.

2. The electromagnetic acoustic combined early warning system for the permeability characteristics of deeply buried tunnels according to claim 1, characterized in that, The step of using harmonic semantic information as auxiliary mining information in the frequency domain to perform penetration characteristic semantic mining on the electromagnetic wave data and form an electromagnetic wave penetration characteristic vector includes: The electromagnetic wave data in the time domain is subjected to Fourier processing to form an electromagnetic wave frequency domain diagram, and the fundamental frequency of the electromagnetic wave is identified from the electromagnetic wave frequency domain diagram. Based on the fundamental frequency of the electromagnetic wave, at least one harmonic frequency of the electromagnetic wave frequency domain diagram is determined. Based on the at least one harmonic frequency, the electromagnetic wave frequency domain diagram is masked to form an electromagnetic harmonic frequency domain diagram, wherein the electromagnetic harmonic frequency domain diagram has information about the at least one harmonic frequency but does not have information about other frequencies. Semantic mining is performed on the electromagnetic wave frequency domain graph to form an electromagnetic wave frequency domain vector; Semantic mining is performed on the electromagnetic harmonic frequency domain graph to form electromagnetic harmonic frequency domain vectors; Using the electromagnetic harmonic frequency domain vector as auxiliary information, the electromagnetic wave frequency domain vector is subjected to penetration characteristic semantic mining to form an electromagnetic wave penetration characteristic vector.

3. The electromagnetic acoustic combined early warning system for the permeability characteristics of deeply buried tunnels according to claim 2, characterized in that, The step of semantic mining of the electromagnetic wave frequency domain graph to form an electromagnetic wave frequency domain vector includes: Convolution mining is performed on the electromagnetic wave frequency domain graph to form an electromagnetic wave frequency domain convolution vector; The electromagnetic wave frequency domain map is segmented, and each local frequency domain map obtained by the segmentation process is subjected to convolution mining to form a local frequency domain convolution vector sequence. For each local frequency domain convolution vector in the local frequency domain convolution vector sequence, based on the spatial enhancement convolution vector corresponding to the local frequency domain convolution vector, semantic enhancement processing is performed on the local frequency domain convolution vector to form the local frequency domain enhancement vector corresponding to the local frequency domain convolution vector. Here, the spatial enhancement convolution vector corresponding to the first local frequency domain convolution vector is the local frequency domain convolution vector, and the spatial enhancement convolution vectors corresponding to other local frequency domain convolution vectors are the local frequency domain enhancement vectors corresponding to the previous local frequency domain convolution vector. Based on the local frequency domain enhancement vector corresponding to the last local frequency domain convolution vector, global semantic enhancement processing is performed on the electromagnetic wave frequency domain convolution vector to form an electromagnetic wave frequency domain vector.

4. The electromagnetic acoustic combined early warning system for the permeability characteristics of deeply buried tunnels according to claim 2, characterized in that, The step of semantic mining of the electromagnetic harmonic frequency domain map to form electromagnetic harmonic frequency domain vectors includes: Convolution mining is performed on the electromagnetic harmonic frequency domain graph to form electromagnetic harmonic frequency domain convolution vectors; The electromagnetic harmonic frequency domain map is segmented, and each local harmonic frequency domain map obtained by the segmentation process is convolutionally mined to form a local harmonic convolution vector sequence. For each local harmonic convolution vector in the local harmonic convolution vector sequence, based on the spatial enhancement convolution vector corresponding to the local harmonic convolution vector, semantic enhancement processing is performed on the local harmonic convolution vector to form the local frequency domain enhancement vector corresponding to the local harmonic convolution vector. Here, the spatial enhancement convolution vector corresponding to the first local harmonic convolution vector is the local harmonic convolution vector, and the spatial enhancement convolution vectors corresponding to other local harmonic convolution vectors are the local frequency domain enhancement vectors corresponding to the previous local harmonic convolution vector. Based on the local frequency domain enhancement vector corresponding to the last local harmonic convolution vector, global semantic enhancement processing is performed on the electromagnetic harmonic frequency domain convolution vector to form an electromagnetic harmonic frequency domain vector.

5. The electromagnetic acoustic combined early warning system for the permeability characteristics of deeply buried tunnels according to claim 2, characterized in that, The step of using the electromagnetic harmonic frequency domain vector as auxiliary mining information to perform penetration characteristic semantic mining on the electromagnetic wave frequency domain vector to form an electromagnetic wave penetration characteristic vector includes: The electromagnetic harmonic frequency domain vector is gated and mapped to form the first gated mapping parameter; The electromagnetic harmonic frequency domain vector is subjected to self-attention processing to form an electromagnetic wave attention vector, and the electromagnetic wave attention vector is subjected to gated mapping to form a second gated mapping parameter. The first gating mapping parameter and the second gating mapping parameter are combined to form an auxiliary gating mapping parameter; Based on the auxiliary gating mapping parameters, the electromagnetic wave frequency domain vector is gating and adjusted to achieve semantic mining of penetration characteristics and form an electromagnetic wave penetration characteristic vector.

6. The electromagnetic acoustic combined early warning system for the permeability characteristics of deeply buried tunnels according to claim 1, characterized in that, The step of using harmonic semantic information as auxiliary mining information in the frequency domain to perform penetration characteristic semantic mining on the ultrasonic data and form an ultrasonic penetration characteristic vector includes: The ultrasonic data in the time domain is subjected to Fourier processing to form an ultrasonic frequency domain diagram, and the fundamental frequency of the ultrasonic wave is identified from the ultrasonic frequency domain diagram. Based on the fundamental frequency of the ultrasonic wave, at least one harmonic frequency of the ultrasonic frequency domain diagram is determined. Based on the at least one harmonic frequency, the ultrasonic frequency domain map is masked to form an ultrasonic harmonic frequency domain map, wherein the ultrasonic harmonic frequency domain map has information of the at least one harmonic frequency but does not have information of other frequencies. Semantic mining is performed on the ultrasonic frequency domain map to form an ultrasonic frequency domain vector; Semantic mining is performed on the ultrasonic harmonic frequency domain map to form ultrasonic harmonic frequency domain vectors; Using the ultrasonic harmonic frequency domain vector as auxiliary mining information, the ultrasonic frequency domain vector is subjected to semantic mining of penetration characteristics to form an ultrasonic penetration characteristic vector.

7. The electromagnetic acoustic combined early warning system for the permeability characteristics of deeply buried tunnels according to claim 6, characterized in that, The step of semantic mining the ultrasonic frequency domain map to form an ultrasonic frequency domain vector includes: Convolution mining is performed on the ultrasonic frequency domain map to form an ultrasonic frequency domain convolution vector; The ultrasonic frequency domain map is segmented, and each local frequency domain map obtained by the segmentation process is subjected to convolution mining to form a local frequency domain convolution vector sequence. For each local frequency domain convolution vector in the local frequency domain convolution vector sequence, based on the spatial enhancement convolution vector corresponding to the local frequency domain convolution vector, semantic enhancement processing is performed on the local frequency domain convolution vector to form the local frequency domain enhancement vector corresponding to the local frequency domain convolution vector. Here, the spatial enhancement convolution vector corresponding to the first local frequency domain convolution vector is the local frequency domain convolution vector, and the spatial enhancement convolution vectors corresponding to other local frequency domain convolution vectors are the local frequency domain enhancement vectors corresponding to the previous local frequency domain convolution vector. Based on the local frequency domain enhancement vector corresponding to the last local frequency domain convolution vector, global semantic enhancement processing is performed on the ultrasonic frequency domain convolution vector to form an ultrasonic frequency domain vector.

8. The electromagnetic acoustic combined early warning system for the permeability characteristics of deeply buried tunnels according to claim 6, characterized in that, The step of semantic mining of the ultrasonic harmonic frequency domain map to form ultrasonic harmonic frequency domain vectors includes: Convolution mining is performed on the ultrasonic harmonic frequency domain map to form ultrasonic harmonic frequency domain convolution vectors; The ultrasonic harmonic frequency domain map is segmented, and each local harmonic frequency domain map obtained by the segmentation process is convolutionally mined to form a local harmonic convolution vector sequence. For each local harmonic convolution vector in the local harmonic convolution vector sequence, based on the spatial enhancement convolution vector corresponding to the local harmonic convolution vector, semantic enhancement processing is performed on the local harmonic convolution vector to form the local frequency domain enhancement vector corresponding to the local harmonic convolution vector. Here, the spatial enhancement convolution vector corresponding to the first local harmonic convolution vector is the local harmonic convolution vector, and the spatial enhancement convolution vectors corresponding to other local harmonic convolution vectors are the local frequency domain enhancement vectors corresponding to the previous local harmonic convolution vector. Based on the local frequency domain enhancement vector corresponding to the last local harmonic convolution vector, global semantic enhancement processing is performed on the ultrasonic harmonic frequency domain convolution vector to form an ultrasonic harmonic frequency domain vector.

9. The electromagnetic acoustic combined early warning system for the permeability characteristics of deeply buried tunnels according to claim 6, characterized in that, The step of using the ultrasonic harmonic frequency domain vector as auxiliary mining information to perform penetration characteristic semantic mining on the ultrasonic frequency domain vector to form an ultrasonic penetration characteristic vector includes: The ultrasonic harmonic frequency domain vector is gated and mapped to form the first gated mapping parameter; The ultrasonic harmonic frequency domain vector is subjected to self-attention processing to form an ultrasonic attention vector, and the ultrasonic attention vector is subjected to gating mapping to form a second gating mapping parameter. The first gating mapping parameter and the second gating mapping parameter are combined to form an auxiliary gating mapping parameter; Based on the auxiliary gating mapping parameters, the ultrasonic frequency domain vector is gating and adjusted to achieve semantic mining of penetration characteristics and form an ultrasonic penetration characteristic vector.

10. The electromagnetic acoustic combined early warning system for the permeability characteristics of deeply buried tunnels according to any one of claims 1-9, characterized in that, The step of performing semantic association extraction on the electromagnetic wave penetration characteristic vector and the ultrasonic wave penetration characteristic vector to form a multidimensional penetration characteristic vector includes: Obtain a sequence of attention hiding matrices, wherein the number of non-zero elements in each attention hiding matrix gradually increases in the sequence; Take either the electromagnetic wave penetration characteristic vector or the ultrasonic wave penetration characteristic vector as the first semantic vector and the other vector as the second semantic vector. In the first stage of associated semantic extraction, the original attention parameters between the first semantic vector and the second semantic vector are determined, and associated semantic extraction is performed on the second semantic vector based on the original attention parameters to form the associated semantic vector of the first stage. In each intermediate stage of the associated semantic extraction, the original attention parameters between the associated semantic vector of the previous stage and the second semantic vector are determined. After adjusting the original attention parameters based on the corresponding attention hidden matrix in the attention hidden matrix sequence, the associated semantics of the second semantic vector are extracted based on the obtained target attention parameters to form the associated semantic vector of the corresponding stage. In the final stage of the associated semantic extraction, the original attention parameters between the associated semantic vector of the previous stage and the second semantic vector are determined, and associated semantic extraction is performed on the second semantic vector based on the original attention parameters to form the associated semantic vector of the final stage. Furthermore, a multidimensional penetration characteristic vector is determined based on the associated semantic vector, wherein the difference between the number of stages of associated semantic extraction and the number of attention hiding matrices is equal to 2.

Images