CN-122017994-A - Underground pipe network micro-motion scattering imaging method, system and medium based on deep learning

CN122017994ACN 122017994 ACN122017994 ACN 122017994ACN-122017994-A

Abstract

The invention discloses a depth learning-based underground pipe network micro-motion scattering imaging method, a system and a medium, relating to the technical field of geophysical exploration and urban underground space safety detection, comprising the steps of continuously collecting urban environment background micro-motion records and segmenting, and superposing initial pure scattered wave records and background micro-motion record segments in a dynamic signal-to-noise ratio manner to construct a semisynthetic training data set; the method comprises the steps of constructing a U-Net convolutional neural network, training by utilizing a semisynthetic training data set to obtain a scattered wave characteristic extraction model, inputting the preprocessed actually measured micro-motion record into the trained scattered wave characteristic extraction model, outputting a predicted pure scattered wave field, performing offset imaging based on the predicted pure scattered wave field and a background speed model to generate a high-resolution space image, and realizing intelligent separation and imaging of weak scattered signals in a passive source micro-motion record to obtain the high-resolution space image of the urban underground pipe network and diseases.

Inventors

LI MINGLIANG
ZOU XIN
WANG YI

Assignees

安徽深地新能科技有限公司

Dates

Publication Date: 20260512
Application Date: 20260410

Claims (10)

1. The underground pipe network micro-motion scattering imaging method based on deep learning is characterized by comprising the following steps of: continuously collecting and segmenting background micro-motion records of urban environment, generating an initial pure scattered wave record containing an underground scatterer based on wave equation forward modeling, and carrying out dynamic signal-to-noise ratio superposition with background micro-motion record segments to construct a semisynthetic training data set; constructing a U-Net convolutional neural network embedded with a residual error learning unit and an attention gating mechanism, and training by utilizing the semisynthetic training data set to obtain a scattered wave characteristic extraction model for separating weak scattered wave signals; preprocessing the actual measurement jog record, inputting the actual measurement jog record into a trained scattered wave characteristic extraction model, and outputting a predicted pure scattered wave field through nonlinear mapping; and performing offset imaging based on the predicted pure scattered wave field and combining a background speed model obtained by inversion of the urban environment background jog record to generate a high-resolution spatial image of the underground pipe network and the diseases.
2. The method of claim 1, wherein urban environmental background jog records are continuously acquired using a gravity coupled millable detector array disposed on a hardened pavement; The gravity coupling type rolling detector array comprises a plurality of improved detectors, and adjacent improved detectors are connected in series through flat Kevlar cables to form a rolling towing cable; each improved detector comprises a pressure-bearing top cover (1), a aviation plug connector (2), a shell (3) and a base (4), wherein the pressure-bearing top cover (1) is positioned at the upper part of the shell (3), the base (4) is positioned at the lower part of the shell (3), the pressure-bearing top cover (1), the shell (3) and the base (4) are assembled into a sealed shell with waterproof and dustproof functions, a three-component detector is packaged in the sealed shell, and the aviation plug connector (2) penetrates through the shell (3) to be connected with the three-component detector in the interior for power supply and data transmission; the main body of the shell (3) adopts a trapezoid compression-resistant structure and is designed into a streamline or flattened low-profile wedge-shaped structure, and the base (4) is a high-density metal counterweight and is in gravity coupling with a road surface.
3. The method of claim 1, wherein the initial clean scattered wave record is generated by: Establishing an underground geological model containing circular, rectangular or irregular-form scatterers; Forward modeling is carried out on the underground geologic model containing the scatterer, so that a total wave field is obtained; Carrying out same-parameter forward modeling on the underground geologic model with the scatterer removed to obtain a background wave field; And calculating the difference value between the total wave field and the background wave field, and taking the difference value as an initial pure scattered wave record, wherein the initial pure scattered wave record only comprises hyperbolic texture characteristics generated by pipelines and is used as a target label of the U-Net convolutional neural network.
4. The method of claim 1, wherein the semi-synthetic training data set The construction formula of (2) is as follows: ; Wherein, the For the initial recording of the clean scattered wave, The segments are recorded for the background jog of the urban environment, The coefficients are adjusted for noise.
5. The method according to claim 1, wherein an attention gating module is embedded at a jump junction between an encoder and a decoder of the U-Net convolutional neural network for automatically suppressing instantaneous high-energy impulse interference generated by rolling of a vehicle and enhancing a response to hyperbolic texture features; A residual learning unit is introduced in the convolutional layer of the encoder to learn a residual mapping between the semi-synthetic training data set and the initial clean scattered wave record.
6. The method of claim 1, wherein the U-Net convolutional neural network is trained with a combined loss function The settings were as follows: ; Wherein, the The predicted pure scattered wave field is output after the semisynthetic training data set is input into a U-Net convolutional neural network; Recording for the initial clean scattered wave as a target tag corresponding to the semi-synthetic training dataset; Is that And (3) with An L1 norm loss therebetween; Is that And (3) with The multi-scale structural similarity between the two is lost, Is a weight coefficient.
7. The method of claim 1, wherein the preprocessing of the measured jog record is as follows: firstly carrying out amplitude equalization or truncation treatment on the actual measurement inching record, and then intercepting the actual measurement inching record into an actual measurement inching record fragment according to a set time window; The standardization processing is that Z-Score standardization or maximum value normalization processing is carried out on each actually measured micro-motion record segment, so that the amplitude distribution of each actually measured micro-motion record segment is consistent with the distribution characteristics of training samples in the semisynthetic training data set; and taking the normalized measured micro-motion record segment as the input of a trained scattered wave characteristic extraction model.
8. The method of claim 1, wherein the migration imaging uses kirchhoff integral migration, inverse time migration, beamforming, least squares migration, or full waveform inversion algorithms, taking the predicted clean scattered wavefield as input, calculating travel times from a background velocity model, pushing back the scattered energy at each point in time to its generated spatial location; Scattered energy from the same pipeline is focused down to the real spatial location of the subsurface anomaly.
9. A computer system comprising a memory, a processor and a computer program stored on the memory, wherein the processor executes the computer program to implement the method of any one of claims 1-8.
10. A computer readable storage medium, characterized in that it has stored thereon a number of computer programs for being called by a processor and performing the method according to any of claims 1-8.

Description

Underground pipe network micro-motion scattering imaging method, system and medium based on deep learning Technical Field The invention relates to the technical field of geophysical exploration and urban underground space safety detection, in particular to an underground pipe network micro-motion scattering imaging method, system and medium based on deep learning. Background Seismic exploration technology originally originated in the oil and gas industry, and earlier mainstream technology was the reflected wave method, which uses Snell's Law to detect subsurface continuous formation boundaries (e.g., hydrocarbon reservoir formations). However, as exploration targets move from finding structures to finding details, scattered wave imaging techniques (SEISMIC SCATTERING IMAGING) based on the huygens principle have evolved. Scattering or diffraction phenomena can occur if and only if there are inhomogeneities (e.g., faults, cracks, karsts, lines) due to the subsurface and their dimensions are less than or near the seismic wavelength. The current state of the art mainly focuses on three directions of ambient noise imaging (passive source), small-scale scatterer detection, and deep-learning seismic signal processing. Despite significant progress in each branch, there is still significant technical fault in the fusion application to urban ultra shallow (0-50 m) micro pipeline targets. ① A passive source detection technique based on micro-motion; the existing mainstream micromotion technique (SPAC/HVSR) is based essentially on the layered medium assumption, mainly yielding a one-or two-dimensional Velocity Profile. It cannot focus, and for small-scale discrete bodies (Diffractor) such as pipelines and cavities, the SPAC algorithm can smooth out the scattering effect generated by the small-scale discrete bodies as high-frequency disturbance in the statistical process. ② Seismic scattered wave/diffracted wave imaging techniques; Existing scatter/diffraction imaging techniques are highly dependent on active source data (e.g., explosives or falling weight sources) with high signal-to-noise ratios. In areas such as urban arterial roads and airports where vibration and road sealing are strictly forbidden, an active source cannot be used, so that the technology cannot fall to the ground. When the method is directly applied to passive source data, the traditional mathematical method cannot extract effective scattering signals due to randomness and faintness of traffic noise sources. ③ Deep learning of the application status in seismic signal processing; Most of the current research on artificial intelligence is focused on active source reflection seismic data (petroleum industry) or natural seismic data (earthquake protection and disaster reduction). Very few studies are made on the specific scene of ultra-shallow scattered wave extraction in the urban passive source (inching) environment. The conventional AI model is mostly trained based on regular reflected waves, and the irregular scattering hyperbolic curve characteristics in the micro-motion record cannot be identified. And the lack of a disclosed passive source seismic scattering data set aiming at urban underground pipe network diseases limits the development of the direction. In summary, at present, no mature technical scheme at home and abroad is available to meet three core requirements of zero interference (passive source), high resolution (scatter imaging) and strong noise immunity (AI extraction). Disclosure of Invention Based on the technical problems in the background art, the invention provides an underground pipe network micro-scattering imaging method, an underground pipe network micro-scattering imaging system and a medium based on deep learning, which are used for realizing intelligent separation and imaging of weak scattering signals in passive source micro-recording and obtaining high-resolution space images of urban underground pipe networks and diseases. The invention provides an underground pipe network micro-motion scattering imaging method based on deep learning, which comprises the following steps: Generating an initial pure scattered wave record containing an underground scatterer based on forward wave equation, and carrying out dynamic signal-to-noise ratio superposition with a background micro-motion record segment to construct a semisynthetic training data set; constructing a U-Net convolutional neural network embedded with a residual error learning unit and an attention gating mechanism, and training by utilizing the semisynthetic training data set to obtain a scattered wave characteristic extraction model for separating weak scattered wave signals; preprocessing the actual measurement jog record, inputting the actual measurement jog record into a trained scattered wave characteristic extraction model, and outputting a predicted pure scattered wave field through nonlinear mapping; and performing offset imaging based on the predicted pure sc