Inversion Results Research Articles

Building initial model for seismic inversion based on semi‐supervised learning

AbstractSeismic inversion is an important tool for reservoir characterization. The inversion results are significantly impacted by a reliable initial model. Conventional well interpolation methods are not able to meet the needs of seismic inversion for lateral heterogeneous reservoirs. Inspired by the sequence modelling network and seismic inversion in the Laplace–Fourier domain, we propose an initial model‐building method using semi‐supervised learning strategy. The proposed method considers spatial information to ensure the horizontal continuity of the initial model. Based on the fact that the low‐frequency components of seismic signals in the Laplace–Fourier domain are easier to obtain, we use the forward model in the Laplace–Fourier domain to replace the time‐domain forward model. The proposed workflow was validated using the Marmousi II model. Although the training was carried out on a small number of low‐frequency impedance traces, the proposed workflow was able to build low‐frequency model for the entire Marmousi II model with a correlation of 98%. Field data examples demonstrate the feasibility and effectiveness of the proposed method. For lateral heterogeneous reservoirs, the proposed method performs better than the well interpolation method. By utilizing the model obtained by the proposed method as the initial low‐frequency model of the conventional inversion method, it is possible to estimate better inversion results. The results of different combinations of training sets demonstrate the stability of the proposed method. This method may still be a viable choice if there is lateral heterogeneity underground but not much well‐logging label data.

Improved adaptive regularization for simulated annealing inversion of transient electromagnetic

Geophysical inversion usually involves ill-posed problem. Regularization is the most commonly used method to mitigate this problem. There are many regularization parameter selection methods, among which the adaptive regularization method can automatically update parameters during iteration, reducing the difficulty of parameter selection. Therefore, it is widely used in linear inversion. However, there are very few studies on the use of adaptive regularization methods in stochastic optimization algorithms. The biggest difficulty is that in stochastic optimization algorithms, the search direction of any iteration is completely random. Data fitting term and stabilizing term vary in a wide range, making it difficult for traditional methods to work. In this paper, we consider the contributions of the data fitting term and the stabilizing term in the objective function and give an improved adaptive regularization method for very fast simulated annealing (VFSA) inversion for transient electromagnetic (TEM) data. The optimized method adjusts the two terms dynamically to make them in balance. We have designed several numerical experiments, and the experimental results demonstrate that the method in this paper not only accelerates the convergence, but also the inversion results are very little affected by the initial regularization parameter. Finally, we apply this method to field data, and the inversion results show very good agreements with nearby borehole data.

Three-dimensional temperature field inversion in the northern South China Sea

A Swin Transformer oceanic three-dimensional temperature field inversion model is established for the northern South China Sea and its surrounding waters based on satellite observation data and ocean reanalysis data from 1993 to 2018. The inversion results are tested using Argo observation data. The root-mean-square errors of the monthly average temperature fields for 2017 and 2018 are 0.10 to 1.98°C and 0.11 to 1.92°C, respectively, with the maximum values mainly located near the mixing layer. The temperature inversion root-mean-square errors are generally less than 1 °C. Additionally, this article analyzes the three-dimensional structural characteristics and evolution process of a typical anticyclone eddy in the study area. The results show that the inversion results can clearly demonstrate the three-dimensional structure of the eddy and continuously and completely reproduce the process of eddy persistence and movement.

Bathymetry Inversion Using a Deep‐Learning‐Based Surrogate for Shallow Water Equations Solvers

AbstractRiver bathymetry is critical for many aspects of water resources management. We propose and demonstrate a bathymetry inversion method using a deep‐learning‐based surrogate for shallow water equations solvers. The surrogate uses the convolutional autoencoder with a shared‐encoder, separate‐decoder architecture. It encodes the input bathymetry and decodes to separate outputs for flow field variables. Utilizing the differentiability of the surrogate, a gradient‐based optimizer is used to perform bathymetry inversion. Two physically based constraints on the ranges of both bed elevation and slope have to be added as inversion loss regularizations to contract the solution space. Using the “L‐curve” criterion, a heuristic approach is proposed to determine the regularization parameters. Both the surrogate model and the inversion algorithm show good performance. The bathymetry inversion progresses in two distinctive stages, which resembles the sculptural process of initial broad‐brush calving and final fine detailing. The inversion loss due to flow prediction error and the two regularizations play dominant roles in the initial and final stages, respectively. The surrogate architecture (whether with both velocity and water surface elevation or velocity only as outputs) does not have significant impact on inversion result. The methodology proposed in this work, an example of differentiable parameter learning, can be similarly used in the inversion of other important distributed parameters such as roughness coefficient.

A Real Space Moiré Inversion Technique and Its Practical Applications in Real Space for Lattice Reconstruction

Distinct physical properties emerge at the nanoscale in Moiré materials, such as bilayer graphene and layered material superposition. This study explores similar structural features within a second-generation nickel-based superalloy, unveiling potential formation mechanisms. Introducing the real space Moiré inversion method (RSMIM) for nanoscale imaging, combined with the transmission electron microscopy (TEM) nano-Moiré inversion method, we reveal spatial angles between specimen and reference lattices in 3D. Simultaneously, we reconstruct the Moiré pattern region to deepen us understand the phenomenon of Moiré formation. Focused on face-centered cubic structures, the research identifies six spatial angles, shedding light on Moiré patterns in the superalloy. The RSMIM not only enhances understanding but also expands 3D structure measurement capabilities. The RSMIM served to validate TEM nano-Moiré inversion results, ascertaining the spatial relative angle between lattices, and establishing a theoretical simulation model for Moiré patterns. This study marks a substantial step toward designing high-performance nanomaterials by uncovering dynamic Moiré variations.

Trans-dimensional inversion for seafloor properties for three mud depocenters on the New England shelf under dynamical oceanographic conditionsa).

This paper presents inversion results for three datasets collected on three spatially separated mud depocenters (hereafter called mud ponds) during the 2022 Seabed Characterization Experiment (SBCEX). The data considered here represent modal time-frequency (TF) dispersion as estimated from a single hydrophone. Inversion is performed using a trans-dimensional (trans-D) Bayesian inference method that jointly estimates water-column and seabed properties along with associated uncertainties. This enables successful estimation of the seafloor properties, consistent with in situ acoustic core measurements, even when the water column is dynamical and mostly unknown. A quantitative analysis is performed to (1) compare results with previous modal TF trans-D studies for one mud pond but under different oceanographic condition, and (2) inter-compare the new SBCEX22 results for the three mud ponds. Overall, the estimated mud geoacoustic properties show no significant temporal variability. Further, no significant spatial variability is found between two of the mud ponds while the estimated geoacoustic properties of the third are different. Two hypotheses, considered to be equally likely, are explored to explain this apparent spatial variability: it may be the result of actual differences in the mud properties, or the mud properties may be similar but the inversion results are driven by difference in data information content.

The Application and Research of Particle Swarm Optimization Radial Basis Function Network Driven by Edge Computing in Bogie Fault Diagnosis

The bogie system is a key system that affects the safety and quality of high-speed train operation. Constructing a fault diagnosis model for the bogie system can effectively improve the safety and comfort of high-speed train operation. The traditional modeling method uses the BP neural network to fit the temperature and other parameters of the bogie system. However, because the BP neural network is prone to fall into local minima, slow convergence and poor diagnostic accuracy, this paper proposes a radial basis function network bogie fault diagnosis model based on particle swarm optimization based on edge computing. The model combines edge computing and particle swarm optimization algorithm to improve the accuracy and real-time of bogie fault diagnosis. By implementing model training and inference on edge devices, data transmission latency has been reduced and diagnostic efficiency has been improved. A particle radial basis neural network (PSRB) was designed using a highly convergent particle swarm optimization algorithm, and the parameters of the radial basis neural network (RBF Neural Networks) were optimized using the Particle Swarm Optimization algorithm. Based on the complexity of the input parameters of the bogie system, determine the input and output parameters of the model, and use particle swarm optimization algorithm to search for the optimal values of the basis function center, width, and output layer weight threshold of the RBF neural network. The composite algorithm was applied to the fault diagnosis of the bogie system, and a particle radial basis neural network bogie fault diagnosis model was designed. The simulation and experimental results of the model show that the diagnostic model can effectively improve the identification accuracy of fault diagnosis, with a minimum error accuracy of 0.0255, saving computation time. The computation time is reduced to 2.9 seconds, eliminating the influence of non target parameters on the inversion results. This model can also be used in other systems of high-speed trains and has practical application value.

Computation and Modelling of Seismic Refraction Tomography for Anisotropic Medium Based on Pseudo Bending Method

Seismic refraction tomography is one of the imaging techniques in geophysical methods used to remodel the near-surface velocity layer structure of the Earth. In this study, we carried out a new computational approach and modelling of seismic refraction tomography using the pseudo-bending method. The true model of the near-surface is designed to be anisotropic medium which is having a low velocity anomaly distribution. This anomaly is constructed in such a way as to be similar to a model of liquid waste away spreading, which exhibits seismic velocities ranging from 1600 m/s to 1800 m/s. Based on our computations and numerical modelling results, it was found that the ray tracing path using pseudo-bending method displays an asymmetrical trajectory when the positions of the source and geophone are exchanged. Altering the shooting configuration from direct shoot (DS) to reversed shoot (RS) also reveals a significant difference in travel time values. The results of delay time tomography inversion, which represents the difference between travel times in the true model and the initial model using the SIRT method, indicate the presence of a low velocity anomaly that can be interpreted as the distribution of liquid waste.

Extracting shallow water depth from the fusion of multi-temporal ICESat-2 data and multi-spectral imageries

With the development of satellite remote sensing and laser altimetry data, the fusion of laser altimetry data and multi-spectral images for shallow water depth inversion has become an economically convenient way. However, there are few methods that take into account the temporal dimension of data to integrate the results of water depth inversion in multiple temporal phases. In response to this issue, this article proposed a shallow water depth inversion method that integrates multi-source and multi-temporal remote sensing data. This method utilized the random forest (RF) algorithm to estimate the water depth values at different time, taking into account the contextual information of Sentinel-2 imagery and using the overall least squares as the theoretical model to fuse the multi-temporal water depth inversion results. This article took the Yongle Islands in the South China Sea as the research area and conducted the shallow water inversion experiments using ICESat-2 (Ice, Cloud and Land Elevation Satellite 2) data from 5 temporal phases and one Sentinel-2 imagery. The results show that the mean value of RMSE (root mean square error) and R2 (determination coefficient) of the proposed method using single temporal imagery was 1.53m and 0.7610, which outperformed the inversion accuracy of traditional methods that ignore image context information. The RMSE and R2 of the multi-temporal fusion model was 1.15m and 0.8622, which was 0.08m and 0.0199m higher than existing median filtering fusion method.

Land use data can improve the accuracy of carbon emission spatial inversion model

AbstractInvestigating the spatial distribution and correlation characteristics of carbon emissions would be conducive to the policy formulation for precise carbon emission spatial reduction. Firstly, a new carbon emission spatial inversion model was developed, incorporating nighttime light data and land use data. After verifying the validity and accuracy of the inversion results, the continuous carbon emission spatial data in the Beijing‐Tianjin‐Hebei Urban Agglomeration (BTHUA) were acquired from 2000 to 2019. Then, the spatial distribution and correlation characteristics were further analyzed in the BTHUA. Finally, policy recommendations were proposed for carbon emission reduction and urban sustainable development. The results showed that the built model can improve the accuracy of the carbon emission spatial inversion data. The carbon emissions were low in the northwest and high in the southeast of the BTHUA, with a noticeable expansion of the high carbon emission contiguous areas around Beijing, Tianjin, Shijiazhuang, and other prefecture‐level cities, which was consistent with the socioeconomic development pattern. The center of gravity of carbon emissions moved to the southeast, showing a relatively stable distribution. The spatial correlation degree of carbon emissions among cities gradually increased, with Beijing and Tianjin playing a prominent role. As a scientific tool, the spatial inversion model helps to produce more accurate spatial data. The results and conclusions can provide useful and scientific references for spatial analysis and regulation strategies of regional carbon emission reduction.

Inference of geoacoustic model parameters from acoustic field data: Perspectives on Geoacoustic Inversion

Estimation of parameters of geoacoustic models from acoustic field data has been a central research theme in acoustical oceanography and ocean acoustics. During the past several decades, highly efficient numerical inversion techniques have been developed that provide model parameter estimates and their uncertainties based on statistical inference methods. However, the methods are model-based and the inversions are prone to errors related to model mismatch. In any event, the inversions can generate only effective models of the true structure of the ocean bottom, which is generally highly variable over relatively small spatial scales in range and depth. There are also questions about the theory for modelling sound propagation in porous sediment media that raise doubt about the validity of inversion results. In most inversions, a visco-elastic theory is used, but is this the most appropriate propagation model? Another question is about the impact of neglecting shear waves in geoacoustic models. Most inversions assume a fluid model of the ocean bottom. This paper revisits issues that have raised questions about limitations of geoacoustic inversion methods, and discusses the impact of various mitigation measures that have been applied. The paper concludes with musings about new inversion techniques based on machine learning.

Tectonic stress around the South Caspian basin deduced from earthquake focal mechanisms

ABSTRACT In this paper, we study the tectonic stress around the South Caspian Basin (SCB), which includes the Kopeh Dagh, Alborz, Talesh, eastern Greater Caucasus mountain belts, the Apsheron sill, and the Balkhan-Ashkabad fault zone. We apply the stress inversion to focal mechanisms of 410 mainshocks that have occurred over the last 69 years. These mechanisms indicate that the surrounding fault zones of the SCB exhibit diverse types of faulting, ranging from thrust to strike-slip, normal, and their combinations. The results of the stress inversion align with the kinematics of the major fault zones bounding the SCB and emphasize the spatial heterogeneity of the stress field in this region. The region is predominantly under compression, but transpressive and strike-slip regimes are also present. This highlights the role of obliquely oriented basement faults with respect to the maximum horizontal compressive stress (S Hmax ) in accommodating deformation through convergent zones. The orientation of the S Hmax is in the NE to NNE direction in the Kopeh Dagh, Alborz, Talesh, and Ashkabad-Balkhan fault zones, being rotated to NNE-N in the Greater Caucasus. The orientation of the S Hmax relative to the convergence direction of the Arabian and Eurasian plates indicates that the Arabian-Eurasian oblique convergence-derived tectonic stress is the primary contributor to the total stress and deformation in this region.

Diagnosing ratio of electron density to collision frequency of plasma surrounding scaled model in a shock tube using low-frequency alternating magnetic field phase shift

A non-contact low-frequency (LF) method of diagnosing the plasma surrounding a scaled model in a shock tube is proposed. This method utilizes the phase shift occurring after the transmission of an LF alternating magnetic field through the plasma to directly measure the ratio of the plasma loop average electron density to collision frequency. An equivalent circuit model is used to analyze the relationship of the phase shift of the magnetic field component of LF electromagnetic waves with the plasma electron density and collision frequency. The applicable range of the LF method on a given plasma scale is analyzed. The upper diagnostic limit for the ratio of the electron density (unit: m−3) to collision frequency (unit: Hz) exceeds 1 × 1011, enabling an electron density to exceed 1 × 1020 m−3 and a collision frequency to be less than 1 GHz. In this work, the feasibility of using the LF phase shift to implement the plasma diagnosis is also assessed. Diagnosis experiments on shock tube equipment are conducted by using both the electrostatic probe method and LF method. By comparing the diagnostic results of the two methods, the inversion results are relatively consistent with each other, thereby preliminarily verifying the feasibility of the LF method. The ratio of the electron density to the collision frequency has a relatively uniform distribution during the plasma stabilization. The LF diagnostic path is a loop around the model, which is suitable for diagnosing the plasma that surrounds the model. Finally, the causes of diagnostic discrepancy between the two methods are analyzed. The proposed method provides a new avenue for diagnosing high-density enveloping plasma.

Cohesive approach for determining porosity and P-impedance in carbonate rocks using seismic attributes and inversion analysis

AbstractThe assessment of hydrocarbon flow through seismic and well-log data presents a persistent challenge in determining porosity. The acoustic impedance section provides a visual representation of the layers, while the raw seismic data showcase the subsurface reflectors that exist within the rock layers. The accuracy of acoustic impedance is widely acknowledged to surpass that of seismic data as a representation of reality. The primary objective of this study is to convert seismic reflector data into acoustic impedance values, which provide insights into the layer properties based on lithology. This approach enhances the accuracy of seismic inversion results by aligning them more closely with actual geological conditions. Seismic inversion is employed to ascertain the physical characteristics of the rock, including acoustic impedance and porosity. Carbonate reservoirs are recognised for their complex pore structures and heterogeneity, which present difficulties in their characterisation. The objective of this research is to predict the porosity and identify the reservoir within the dense carbonate reservoirs in Central Luconia, Sarawak. These objectives are achieved by employing a porosity and acoustic impedance cross-plot and improved precision and predictability through the integration of seismic attribute interpretation and deterministic seismic inversions. The uniqueness of our approach stems from the incorporation of various geophysical techniques to detect reservoirs that have hydrocarbon deposits. A correlation is observed between seismic inversion acoustic impedance and porosity within the zone of interest, indicating an estimated porosity range of 10–35%. The analysed area demonstrates the possibility of containing a hydrocarbon based on the observed relationship between porosity and impedance, as well as the outcomes of the inversion analysis.

A unified empirical method for predicting both vertical and horizontal ground displacements induced by tunnel excavation

An empirical model for describing the soil movements induced by tunneling is proposed, then the mathematical relationship between horizontal displacement and vertical displacement is obtained. By analyzing 25 sets of data from field engineering and 35 sets of data from model tests, the formula for the maximum settlement Sv, max( z) is optimized to adapt to different ground conditions. The Modified Gaussian function is developed by introducing the existing width coefficient of the settlement trough i( z) and the optimized Sv, max( z) to describe the surface and subsurface vertical displacement. Subsequently, the formula for the horizontal displacement is derived based on the Modified Gaussian function. H( z) representing the position of the oriented point of the soil movement is a variable in the formula for horizontal displacement. Based on measured results, a logarithmic function and a quadratic polynomial function are proposed to describe the variation of H( z) with depth in the clay stratum and sand stratum, respectively. Then, the rationality of the proposed method is validated by 4 sets of in-situ data and 4 sets of test data. Finally, the implementation process of the proposed method is illustrated, and the inversion results of the ground displacement field in Beijing Metro Line 12 are presented.

High‐resolution reservoir prediction method based on data‐driven and model‐based approaches

AbstractThe Jiyang depression in the southeastern part of the Bohai Bay Basin has a relatively large scale set of shale oil in the Paleogene Shahejie Formation, but the complex internal components lead to narrow frequency bands, low resolution and difficulty in reservoir information extraction. Impedance is important information for reservoir characterization, and how to predict high‐resolution impedance using available information is particularly important. Deep learning, known for its effectiveness in addressing non‐linear problems, has found extensive applications in various fields of oil and gas exploration. However, the challenges of overfitting and poor generalization persist due to the limited availability of training datasets. Besides, existing methods often use networks to solve a single problem in fact, deep learning can deal with a series of problems intelligently. In order to partially solve the above problems, an intelligent storage prediction network framework is proposed in this paper. Physical information is introduced to realize data‐driven and model‐based approaches, thus solving the problem of difficult construction of training datasets. The processing part accomplishes the high‐resolution processing of seismic records, thus solving the problems of narrow bandwidth and low resolution. Initial model constraints are introduced so as to obtain more stable inversion results. Finally, the well data is compared and analysed to identify and predict the lithology and complete the intelligent prediction of unconventional reservoirs. The results are compared with the traditional model‐driven inversion method, revealing that the approach presented in this paper exhibits higher resolution in predicting dolomite. This contributes to the establishment of a robust data foundation for reservoir evaluation.

A facies‐constrained geostatistical seismic inversion method based on multi‐scale sparse representation

AbstractGeostatistical seismic inversion is an important method for establishing high‐resolution reservoir parameter models. There is no accurate representation method for reservoir structural features, and prior information about structural features cannot be incorporated into geostatistical inversion. Based on the assumption of the sparsity of stratigraphic sedimentary features, the same type of structural feature is used to represent the sedimentary pattern of reservoirs within the same facies. Different sparse representation patterns are used to represent the differences in sedimentary patterns between facies. Although changes in depositional environment might result in the multi‐scale characteristics of geological structures for varying sedimentary rhythms, this paper proposes a facies‐constrained geostatistical inversion method based on multi‐scale sparse representation to better accommodate such situation. Using the method of sparse representation combined with wavelet transform, the multi‐scale sedimentary structural features of reservoirs are learned from well‐logging data. Seismic facies and multi‐scale features are used as prior information for geostatistical inversion. Further, the likelihood function is constructed using seismic data to obtain the posterior probability distribution of reservoir parameters. Finally, the accurate inversion result is obtained by using multi‐scale sparse representation as a constraint in the posterior probability distribution of reservoir parameters. Compared with conventional geostatistical methods, this algorithm can better match the structural features of reservoir parameters with varying geological conditions. Field data tests have shown the effectiveness of this method in improving the accuracy and resolution of reservoir parameter structural features.

Near-surface imaging by joint inversion of ERT and seismic traveltime data with guided FCM clustering

We present a novel joint inversion strategy for electrical resistivity tomography (ERT) and seismic first-arrival traveltime data with guided fuzzy c-means (GFCM) clustering coupling to increase the accuracy of inversion results and improve the characterization of heterogeneous near-surface materials. The physical parameter correlation between the resistivity and seismic velocity is enforced using a GFCM clustering coupling term in the objective function, which also includes misfit and regularization terms. The physical property models estimated by joint inversion are guided by the clustering centers of a priori petrophysical parameters. A limited-memory quasi-Newton approach is used to optimize the objective function. To validate the proposed approach and describe its implementation in detail, tests were first performed using two synthetic models. The stratigraphic structures recovered by joint inversion with GFCM clustering coupling are more consistent with the true model than those inverted by individual inversion, cross-gradient and fuzzy c-means (FCM) joint inversion. Furthermore, the physical parameters estimated using GFCM joint inversion are significantly more accurate than those estimated using individual inversion and the other two joint inversion methods. The joint inversion algorithm was applied to field data from the Liangzhu site in Hangzhou, China. The depth and shape of the stratigraphic interface imaged by GFCM joint inversion are consistent with the drilling results. However, when using individual inversion and the other two joint inversion methods, the recovered stratigraphic structures have some differences with the drilling data. Therefore, GFCM joint inversion is an effective and reliable method for near-surface fine imaging.

Quasi-2D inversion of surface large fixed-loop transient electromagnetic sounding data

Abstract In many cases, 1D inversion is still an important step in transient electromagnetic data processing. Potential issues may arise in the calculation of apparent resistivity using induced electromotive force (EMF) due to overshoot and the presence of multi-valued functions. Obtaining reliable and consistent inversion results using a uniform half-space as the initial model is challenging, especially when aiming for efficient inversion. Focusing on these problems, we use the land-based transient electromagnetic (TEM) sounding data, which was acquired by using a large fixed-loop transmitter, and adopt a quasi-2D inversion scheme to generate improved images of the subsurface resistivity structure. First, we have considered directly using magnetic field data or converting induced EMF into magnetic field, and then calculating the apparent resistivity over the whole zone. Next, a resistivity profile that varies with depth is obtained through fast smoke ring imaging. This profile serves as the initial model for the subsequent optimal inversion. The inversion scheme uses a nonlinear least-squares method, incorporating lateral and vertical constraints, to produce a quasi-2D subsurface image. The potentiality of the proposed methodology has been exemplified through the interpretation of synthetic data derived from a 3D intricate resistivity model, as well as field data obtained from a TEM survey conducted in a coalmine field. In both cases, the inversion process yields quasi-2D subsurface images that exhibit a reasonable level of accuracy. These images appear to be less moulded by 3D effects and demonstrate a satisfactory level of agreement with the known target area.

MT2DInv-Unet: A 2D magnetotelluric inversion method based on deep-learning technology

Traditional gradient-based inversion methods usually suffer from the problems of falling into local minima and relying heavily on initial guesses. Deep-learning methods have received increasing attention due to their excellent nonlinear fitting ability. However, given the recent application of deep-learning methods in the field of magnetotelluric (MT) inversion, there are currently challenges associated with achieving high inversion resolution and extracting sufficient features. We develop a neural network model (called MT2DInv-Unet) based on the deformable convolution for 2D MT inversion to approximate the nonlinear mapping from the MT response data to the resistivity model. The deformable convolution is achieved by adding an offset to each sample point of the conventional convolution operation, which extracts hidden relationships and allows the flexible adjustment of the size and shape of the feature region. Meanwhile, we design the network structure with multiscale residual blocks, which effectively extract the multiscale features of the MT response data. This design not only enhances the network performance but also alleviates issues such as vanishing gradients and network degradation. The results of synthetic models indicate that our network inversion method has stable convergence, good robustness, and generalization performance, and it performs better than the fully convolutional neural network and U-Net network. Finally, the inversion results of field data show that MT2DInv-Unet can effectively obtain a reliable underground resistivity structure and has a good application prospect in MT inversion.