Abbreviation (ISO4): Prog Geophy
Editor in chief:
In January 2022, Tonga experienced a massive underwater volcanic eruption, causing significant impacts throughout the Pacific region. In this article, a precise positioning technique based on spherical harmonic expansion is used to decompose the errors along the propagation path of GPS signals. This technique allows for precise single-point positioning calculations of GPS data received at the TONG station within the Kingdom of Tonga. The results of these calculations are then utilized to study the response to the Tonga volcanic eruption.The TONG station's standard deviations in the East (E), North (N), and Up (U) directions are 1.32 cm, 0.63 cm, and 1.53 cm, respectively. During the volcanic eruption, the positional error time series data exhibits a noticeable response. In the computed coordinate time series data, there is a clear shift in the station's position after the volcanic eruption. By performing a fitting process, it is determined that the station moved in a southeast direction, with an eastward velocity of 0.0108 m/d, a southward velocity of 0.0023 m/d, and a downward velocity of 0.0149 m/d. By employing the Singular Spectrum Analysis (SSA) method to decompose the time series data, the principal components and residuals are obtained for each direction (E, N, U). In the E direction, as well as the U direction, there are noticeable periodic variations. The amplitude of the E direction is 0.045 cm, while the N direction has an amplitude of 0.05 cm. All three directions exhibit significant fluctuations around the time of the volcanic eruption, particularly around January 15th.The residuals in the N and U directions exhibit significant changes at the moment of the volcanic eruption. In particular, the U direction's residual decreased by 0.4 cm after the eruption compared to the pre-eruption stability. Additionally, the error correction of the propagation path, based on spherical harmonic expansion, shows a clear response to the errors caused by volcanic ash. The magnitude of these errors suggests that the volcanic ash has shifted in the southeast direction. Additionally, by analyzing the distribution of errors at different elevation angles, it is possible to observe the spread and ascent of volcanic ash. This demonstrates that this method can be employed for studying volcanic eruptions.
Aiming at the problem of insufficient spatial resolution of the existing high-degree gravity field models, the study takes the Colorado mountain region in the 1 cm geoid experiment initiated by the International Association of Geodesy as the research object. The spectral characteristics of the EGM2008, EIGEN_6C4, GECO, SGG-UGM-1, SGG-UGM-2 and XGM2019e_2159 models are analyzed. The six high-degree gravity field models are extended using the Residual Terrain Model (RTM) to construct a high-resolution local gravity field for the Colorado region. Finally, the validity of the RTM is checked using measured GNSS/levelling, gravity disturbance, and the Deflection of the Vertical (DOV) data. The results show that the cumulative geoid degree errors of GECO, EIGEN_6C4, SGG-UGM-1, SGG-UGM-2 and XGM2019e_2159 models are smaller than that of EGM2008 model. Before the degree 200 (long wavelength), the signal-to-noise ratios of EIGEN_6C4, GECO, SGG-UGM-1, SGG-UGM-2, and XGM2019e_2159 models show little difference and are all superior to the signal-to-noise ratio of the EGM2008 model. Between the degree 200 and 370 (medium wavelength), the SGG-UGM-2 model has the best signal-to-noise ratio, and after approximately degree 370 (short wavelength), the signal-to-noise ratio of the EIGEN_6C4 model is significantly better than the other five high-order models. The average calculation accuracy of the height anomaly, gravity disturbance, east-west vertical deflection, and north-south vertical deflection of the six high-degree gravity field models compensated by RTM has been improved by about 7.1%、47.5%、65.9%和51%, respectively. The EIEGN _6C4 model has the best accuracy, with about 7.7%、49.7%、70.4%和54.2% improvement in the calculation accuracy of the height anomalies, gravity disturbance, east-west vertical deflection, and north-south vertical deflection, respectively, which demonstrates the validity and reliability of the RTM in recovering the high-frequency gravity field.
Before and after a large earthquake, the weather often changes abnormally and produces precipitation in the earthquake area. Based on the meteorological data of surface temperature, Zenith Total Delay (ZTD) and actual precipitation 10 days before and 5 days after the earthquake happening, this paper analyzed the abnormal changes of surfaces temperature and ZTD before and after the earthquake based on the methods of Z-Score (ZS) and wavelet transform. The results shown that there were large-scale surface temperature anomalies within 5°×5° of the epicenter, and the ZS index of the epicenter was relatively large. The earthquake occurred in the decline stage after the ZS index reaches its peak. The ZTD shown several significant rises and decline changes, the rainfall occurred at the same time. The ZS index of surface temperature showed obvious similarities with the ZTD changes. The ZS index reached the peak earlier than ZTD, indicating that the abnormal changes of ZTD were highly correlated with the surface temperature. The ZTD after wavelet decomposition and reconstruction will show abnormal peak or trough signals 1~3 days before and after the earthquake happening, indicating that the intensified water vapor fluctuation is one of the indicators of precipitation. Therefore, the ZS index abnormal fluctuation of surface temperature and the ZTD peak or valley signal after wavelet decomposition and reconstruction can provide references for short-term earthquake forecast and post-earthquake secondary meteorological disaster prevention.
Experiments show that electromagnetic pulse anomalies could be observed in rock samples before rupture. In this paper, a method is proposed to identify the changes of geo-electric field pulses before earthquakes, which can be used to analyze the evolution characteristics of geo-electric field anomalies in the short and medium term before strong earthquakes. The low-frequency signal component (daily trend signal) in the geo-electric field observation data is removed, and the amplitude deviation of the high-frequency pulse signal from the normal distribution is calculated under logarithmic coordinates. The difference between the amplitude of the geo-electric field that deviates from the normal distribution and the fitting extrapolation amplitude that meets the normal distribution (ΔE) is used as the parameter to identify the pre-earthquake anomaly. The characteristics of geo-electric field pulse change before three earthquakes (≥6) in Northwest China from 2021 to 2022 are analyzed. The results show that the amplitude enhancement of geo-electric field pulse changes at the monthly prediction scale, which has a certain correlation with the occurrence of earthquakes (≥6) around the station.
Efficient heat exchange of fluids within artificial thermal reservoir is the fundamental goal for the development of hot dry rock, and it is crucial to reasonably assess the heat extraction ratio of hot dry rock reservoir containing a single fracture. Based on the relationship between fluid output temperature and thermal power generation, the conventional heat extraction ratio is optimized, and the modified heat extraction ratio, i.e., heat extraction power, which considers the effects of both output temperature and power generation, is proposed. A numerical heat transfer model for fluid flowing in a single fracture using a local thermal non-equilibrium model is developed to investigate the effects of reservoir size, fracture aperture and reservoir running time of hot dry rock on the heat extraction power, and the sensitivity hierarchy is classified. The results show that, in the process of reservoir seepage heat transfer, the heat extraction power is mainly controlled by the mass flow rate, and the peak value, i.e., the optimal heat extraction power, appears with the change of the mass flow rate of the fissure flow; when the reservoir size is the same, the fracture aperture has almost no effect on the heat extraction power, the optimal heat extraction power, and the output temperature, etc., but the heat extraction power and the optimal heat extraction power will be declined with the increase of the running time. Based on the modified Morris Method, the sensitivity of reservoir size, fracture aperture and running time to the optimal heat extraction power was analyzed, and it was found that the reservoir size was highly sensitive and positively correlated with the optimal heat extraction power, the running time was generally sensitive and negatively correlated, and the fracture aperture was not sensitive. The research results can provide a reference for the evaluation of heat extraction performance of hot dry rock.
An MS6.2 earthquake struck Jishishan County, Gansu Province, China, on December 18, 2023, leading to a tremendous sand boil and causing more than 150 deaths. As the trigger of the sand boils, the ground shaking during the earthquake attracts extensive concerns. In this study, ground motions of the Jishishan earthquake were reproduced by using a stochastic finite-fault modeling approach and compared with observations. To maintain the far-field received energy independent of subfault size, two improvements were made to the source spectral model. The comparison shows that: Simulated ground motions agree well with observations in terms of the waveform, peak, and duration of the acceleration time histories at reported stations. The response spectra of the observed time series are in general well reproduced by the simulation except for the remarkably large amplitude at certain periods that may result from a site response. Both the simulated and observed PGA are higher than the predictions of the empirical model but remain close to +1 standard deviation. The contour map of the simulated PGA and PGV shows a similar pattern to the observation except for the slight underestimation near the northeast and northwest corners of the study area. The maximum intensity derived from the simulated PGA is degree Ⅺ, which is also consistent with the reported shaking intensity.
Current multiple suppression methods based on virtual events can effectively attenuate internal multiples in seismic data processing. However, challenges remain, including incomplete prediction of internal multiples and amplitude-phase mismatches between predicted and actual multiples, leading to residual artifacts after subtraction. To address these limitations, this study proposes a joint suppression strategy integrating prestack virtual events with high-precision Radon transform. The method first applies prestack virtual event construction for preliminary multiple suppression, followed by high-precision Radon transform to eliminate residual multiples, thereby minimizing subtraction-induced artifacts. Tests on synthetic models and field datasets demonstrate that compared to standalone virtual event methods, this integrated approach achieves more complete multiple removal, significantly enhancing signal-to-noise ratio and imaging quality. The method not only advances theoretical frameworks for multiple suppression but also provides a practical solution for processing seismic data in complex structural areas.
The simultaneous-source data acquisition technique allows seismic data to overlap with each other, and by exciting two or more sources simultaneously or delayed, it is possible to obtain several times more seismic data than conventional acquisition in the same time, which greatly improves the acquisition efficiency, but because the acquisition data are mixed with a large amount of confounding noise, it seriously affects the subsequent data processing and interpretation. This paper proposes a deblending method based on improved U-Net, which incorporates a dual channel attention mechanism into the original U-Net, focusing on the continuity of the reflection layer and waveform amplitude changes in seismic data, while enhancing the signal contrast in local areas and highlighting the reflection signals; The use of hybrid dilated convolution avoids partial information loss caused by pooling operations during down-sampling, ultimately achieving mixed data separation based on dual attention mechanism and hybrid dilated convolution U-Net (HDC AU-Net). The simulation data experiment results show that compared with the iterative sparse inversion method and the original U-Net method, the HDC AU-Net method has better removal effect on aliasing noise and higher separation signal-to-noise ratio. The actual data experiment further verified the reliability of the algorithm.
Time-domain fractional differentiation of seismic signals is a signal processing method in the field of differentiation. Compared with the traditional integer order differential processing and frequency domain seismic signal extension methods, the multi-scale characteristics of seismic signals can be better described and the data reliability can be improved. In order to deal with complex seismic signals with narrow frequency bands and obtain more refined seismic data, this paper proposes a seismic data extension method based on multi-level fractional differential adaptive fusion in time domain, which automatically obtains the weighted fusion coefficient by using the method of adaptive dynamic adjustment of the weighted coefficient according to the characteristics of fractional differential signals in different frequency bands in different frequency bands and the envelope of the spectrum of differentially differentiated signals of different orders. The seismic signal after frequency extension processing effectively reduces the influence of seismic wavelet band limiting. The model test and practical application show that the fractional differential operation in the time domain of the seismic signal highlights the high-frequency component of the original seismic signal and maintains the low-frequency component of the original signal. With the increase of the differential order, the main frequency of the differential signal increases, and the number of sidelobes increases while the width of the sidelobes narrows, and the ratio of the sidelobe value to the mainlobe value also increases gradually. After multi-level fractional differential adaptive fusion processing, the sidelobe amplitude can be effectively suppressed, the frequency band range can be broadened, and the ability to explain and characterize the reservoir can be provided more clearly, which can improve the ability to identify thin interbeds to a certain extent.
With the extension of seismic exploration to development, the quality of seismic data is becoming increasingly important. However, the technologies such as efficient acquisition with controllable seismic sources are limited by surface conditions, resulting in strong irregular energy interference, low signal-to-noise ratio, and no appropriate solution for obtaining the better near offset seismic traces. The existing polynomial fitting method is a method for improving signal-to-noise ratio while maintaining resolution after stacking, but its application to prestack seismic data is still limited due to the quality requirements of the data for prestack inversion. The high-order amplitude preserving Radon transform can effectively reconstruct missing seismic traces, but its application efficiency is limited. Considering the approximate formula of the Zoeppritz equation for the reflection coefficients of PP and PS waves, it can be expressed in power series form. The reflection coefficients of PP waves are approximately even polynomials, and the reflection coefficients of PS waves are approximately odd polynomials. Based on the approximation of the reflection coefficients of PP waves and PS waves, an odd or even polynomial coefficient equations are constructed within the preset incident angle range. The odd or even polynomial coefficients are obtained using the least squares fitting method, and the seismic traces involved in the fitting are continuously updated by a stepwise elimination fitting method. The polynomial coefficients of the prestack seismic traces are then continuously updated to gradually enhance the effective reflection signal of the near offset region. The analysis results of the model data indicate that the fitting accuracy of the reflection coefficient approximation formula is significantly higher than that of polynomial fitting, and the use of the mid offset traces can effectively reconstruct the reflection signals of the near offset seismic traces. Actual data also shows that it can effectively enhance the near offset seismic signal, improve and enhance the quality of the near offset seismic data.
Due to the absence of incident direction and azimuth direction seismic information in post-stack seismic data, conventional post-stack methods cannot use anisotropic variation characteristics to predict fractures. The response of heterogeneous fracture wave field is bidirectional in both incident direction and azimuthing direction, and its response is weak. Generally, it is hidden in the strongly reflected wave field information, and its sensitivity is not easy to be detected by conventional prestack fracture prediction method. The application effect of conventional prestack fracture prediction method is not ideal, and fracture is one of the difficulties in seismic prediction. Therefore, a 3D prestack fracture prediction method based on azimuthal radially-split gradient variation is developed in this paper. The 3D prestack seismic response amplitude equation based on azimuthal impedance factor and azimuthal anisotropy factor varying with the incident angle is derived, and based on that, the quadratic function of 3D prestack seismic response amplitude of one-variable with the sine square of the incident angle is derived at a certain azimuthal angle, and then, the derivative is taken and it gives a first order amplitude gradient equation of one-variable with the sine square of the incident angle, it is proved that the variation of the fixed azimuthal amplitude gradient is linear, the azimuthal anisotropy factor is the main factor leading to the variation of fixed azimuthal amplitude gradient, and its influence enhances with the increase of incident angle, the larger the incidence angle, the greater the anisotropy effect. It is proved that the amplitude variation with incidence-angle in fixed azimuthal direction has piecewise partitioning property, and the amplitude gradient values of different sections can be obtained by linear partitioning of the incident area, and then the amplitude gradient variation with azimuthal change can be used to predict stratum fracture with 3D prestack seismic gather. A new prestack crack prediction software is developed by programming the theoretical method of this paper. Good application effect has been achieved in the actual work area. The results show that this method is accurate and reliable, it can solve the problem of stratum microfracture prediction, and it can provide strong support for the exploration and development of complex fracture reservoirs.
Vein deposits, important carriers of metal sulfides, typically exhibit low resistivity. Transient Electromagnetic (TEM) methods have been widely employed for detecting such low-resistivity deposits, with forward modeling forming the basis of field data acquisition design and subsequent data processing and interpretation. To optimize field exploration strategies and facilitate effective data interpretation, this study constructs a series of low-resistivity vein deposit models buried beneath conductive cover layers. We use a 3D vector finite element forward modeling approach to systematically investigate the effects of transmitter loop scale, cover layer conductivity and thickness, as well as deposit conductivity, thickness, and dip angle on the resolution of TEM methods. The results show that increasing the transmitter loop scale enhances the total induced Electromotive Force (EMF) field but reduces the ratio of the total field to the background field, indicating a decrease in resolution. Using the constructed model as an example, increasing the transmitter loop scale by a factor of three leads to a 7/9 reduction in resolution. An increase in the conductivity and thickness of the cover layer significantly weakens the detection capability of TEM. Similarly, a decrease in the conductivity and thickness of vein deposits results in reduced detection capability. For instance, with a 300 m transmitter loop, increasing the cover layer conductivity by a factor of 10 leads to a 94.5% reduction in resolution, while increasing the cover layer thickness by 2.5 times reduces the resolution by 96.7%. Reducing the deposit conductivity by 49/50 lowers the resolution by 98.8%, and reducing the deposit thickness by 9/10 lowers the resolution by 96.7%. Changes in the dip angle of the deposit also affect the location of the measurement point with the strongest resolution. For example, at a 25° dip angle, the measurement point above the center of the deposit exhibits the strongest resolution, whereas at a 75° dip angle, the strongest resolution shifts to the measurement point above the deeper end of the deposit. This study shows that the physical properties of the cover layer and vein deposits significantly impact the detection capability of TEM, with these effects being more pronounced when using small-scale transmitter loops. Therefore, exploration design must strike a balance among transmitter loop scale, resolution, target dimensions, and operational efficiency.
The gold deposit in Jiuzhanggou area, Songxian County, Henan Province is a tectonic altered rock type deposit. With the continuous exploitation of the mine in recent years, the amount of resources has gradually decreased. Deep and edge prospecting is an important exploration direction at present. The near north-south trending fault structure F1 is the main ore-bearing (ore-controlling) structure in the mining area. Along the F1 of the ore-bearing (ore-controlling) structure, there is still a possibility of large-scale mineralization in the deep part. The detection depth of traditional geophysical prospecting method is limited, and the conventional large depth detection method is restricted by various factors in the study area to varying degrees, which makes it difficult to carry out deep exploration. The microtremor exploration method collects the natural source surface wave information and uses its dispersion characteristics to invert and calculate the S-wave velocity structure of the underground medium, and infers the underground geological structure and structure to achieve the purpose of exploration. It has the characteristics of high resolution, large detection depth, no electromagnetic background interference, high efficiency and low cost, and provides technical support for deep exploration in the study area. In order to explore the geological structure and metallogenic potential of the deep and edge of the research region, the microtremor exploration test was carried out at the location of the previous geophysical profile, and the apparent S-wave velocity structure below 2.5 km of the profile was obtained. There are obvious low-velocity anomalies and velocity differences on the two-dimensional micro-motion S-wave velocity section. Combined with the previous geological and geophysical survey data, the fracture morphology of the inland graben boundary, the deep extension characteristics of the ore-bearing fracture structure F1 and the intrusion range of the rock mass are inferred. The results are in good agreement with the previous geophysical profile results and the known geological profile, indicating that the micro-motion exploration can provide important geophysical information for the deep and edge prospecting work in the research region, which is of great guiding significance for the realization of the prospecting breakthrough of alternative resources.
Tar-rich coal is a kind of coal-based oil and gas resource. It is of practical significance to guarantee the supply of oil and gas resources and realize the green and low-carbon utilization of coal. Compared with traditional coal, tar-rich coal has high tar yield. However, the research on influencing factors and quantitative characterization methods of tar yield is relatively weak at present. Based on the logging data and basic experimental analysis data of key coal wells in the five coalfields of Shaanxi Province, the main influencing factors of coal tar yield are systematically analyzed, and the quantitative evaluation method of coal tar yield is tried to establish. The results shown: (1) Tar yield is mainly affected by coal quality, coalification degree and coal forming environment. A method was established to calculate the industrial components of coal with conventional logging data, identify the degree of coalification with the fixed carbon and volatile content without water and ash, and evaluate the coal formation environment with the mineral content calculated by elemental logging combined with the end-member analysis method of coal ash composition, thus realizing the quantitative characterization of the factors affecting the tar yield; (2) Under the comprehensive consideration of coal rank, coal quality and coal ash composition, the calculation results are in good agreement with the actual core analysis results, which provides a new idea for the subsequent identification and evaluation of oil-rich coal.
The calculation of water saturation in carbonate reservoirs has perennially posed a significant challenge within the logging community. The complex development of fractures and holes in the Dengying Formation on the northern slope of the central Sichuan paleo-uplift further complicates the accurate assessment of water saturation in the Dengying Formation reservoir in this region. To address this issue, this study conducts a comprehensive analysis of prior research pertaining to water saturation calculation methods, ultimately adopting the approach based on the triple porosity model. The findings reveal that: (1) The water saturation error computed through the conventional Archie formula markedly exceeds that obtained via the triple porosity model. (2) The water saturation determined by the triple porosity model derived from the Maxwell-Garnett mixing conduction law demonstrates superior accuracy compared to the triple porosity model derived from the simplistic "series-parallel" derivation. (3) The triple porosity model, accounting for the influence of fracture inclination on rock conductivity, proves more accurate than its counterpart neglecting such considerations. (4) Utilizing the triple porosity model proposed by Tian et al., satisfactory results are achieved in the calculation of water saturation for fractured-vuggy reservoirs within the Dengying Formation on the central and northern slopes of Sichuan. This investigation holds significant reference value for the accurate determination of water saturation in carbonate fractured-vuggy reservoirs.
In the eastern Tarim Basin, the Silurian Kepingtage Formation represents a crucial marine hydrocarbon-bearing system, originating within a tidal flat depositional environment. This formation has undergone multiple structural adjustments, leading to pronounced heterogeneity in the macroscopic distribution of oil and water layers, posing challenges for the exploration and development of Silurian oil reservoirs. This study leverages core samples, thin sections, core analysis, oil testing, sampling, and both conventional and Nuclear Magnetic Resonance (NMR) logging data to delineate the response characteristics of oil and water layers in conventional logging. The application of NMR logging further elucidates the intricate relationship between reservoir pore structures and hydrocarbon saturation. Through detailed analysis of well profiles across diverse locations, the study dissects the vertical and horizontal distribution patterns of oil and water layers within the Silurian Kepingtage Formation in eastern Tarim. Utilizing logging interpretation, porosity calculations, and resistivity measurements, the research assesses the macroscopic distribution of oil and water layers across various structural settings. Single-well analyses provide insights into conceptual microscopic models that explain the current differentiation of oil and water layers. The results show that during the destruction of ancient oil reservoirs, oil layers with better pore structures were the first to be damaged. The reservoir space was occupied by formation water, creating areas with better current physical properties that are now water layers, while oil layers with relatively poorer pore structures were preserved.
Mineral content logging evaluation can provide important reference data for the lithofacies classification of oil and gas reservoirs and the formulation of production construction plans. In the era of increasingly advanced big data technology, replacing the traditional logging interpretation work mode based on statistical laws with the mineral content joint inversion technology of conventional logging data is an inevitable development direction of conventional logging interpretation theory. This technology can not only fully utilize a large amount of data from new and old oilfields, improve the interpretation accuracy of complex mineral content, but also greatly improve work efficiency. However, the current joint inversion method for mineral content from conventional logging data still has many problems, such as limitations in core mineral content experimental measurement techniques, incomplete interpretation models, and poor numerical stability of inversion algorithms. This article focuses on the above three issues and conducts in-depth research one by one, proposing a log response model based on mass ratio, and providing a method for obtaining model parameters. At the same time, the VRH model is introduced into the inversion equations, and a high-precision and fast algorithm is designed. Finally, a theoretically feasible solution to the difficulty of poor numerical stability in joint inversion is proposed.
Well logging identification and quantitative characterization of source rocks are of great significance to hydrocarbon reserves evaluation. There are three types of source rocks in the Jurassic Yangxia Formation of Kuqa Depression: coal measure, dark mudstone and carboniferous mudstone. However, due to the depth of burial (average greater than 4500 m), the source rocks are too mature and affected by deep buried ground stress, the traditional ΔlogR method is difficult to apply. Based on this comprehensive use of geochemical analysis and geophysical logging data, this paper first reveals the geological characteristics of the source rocks of the Jurassic Yangxia Formation, and realizes the qualitative identification of the source rocks of different lithologies through conventional logging crossplot. The results show that the overall maturity of source rocks of Yangxia Formation is high and the quality of source rocks is medium to good. The qualitative identification of source rocks can be realized by the intersection of curves such as GR, AC, DEN, CNC and AC, GR, RT, CNC, DEN curves that are sensitive to the response of the source rock are selected to establish a quantitative prediction model of TOC content by using multi-regression analysis methods and BP artificial intelligence method, and the quantitative evaluation of the single well source rock in the study area is realized. The results are in good agreement with the measured core data. The research results are of guiding significance for comprehensive logging evaluation of deep source rock quality in Kuqa Depression.
In order to solve the problem of low data quality and unsatisfactory detection effect of Wide Field Electromagnetic Method (WFEM) caused by noise, this paper proposes a WFEM denoising method and application based on improved dung beetle optimization (Improved DBO, IDBO) and Long Short Term Memory (LSTM) network. Firstly, the Spatial Pyramid Matching (SPM) chaotic mapping, variable spiral strategy, Levy flight mechanism, adaptive t-distribution variance perturbation strategy are used to improve the IDBO algorithm. Then, the mean square error is used as the fitness function of the IDBO algorithm to optimize the hyperparameters of the LSTM algorithm. Finally, the IDBO-LSTM method is applied to the WFEM data de-noising processing. The experimental results show that the search ability of IDBO is significantly better than that of other intelligent optimization algorithms, and the LSTM algorithm optimized by IDBO has a significantly higher denoising accuracy than the probabilistic neural network(PNN) and the LSTM algorithms. The data quality of the WFEM data processed by the IDBO-LSTM method is significantly improved, and the electric field curve shape is more stable. The proposed method can provide technical support for the interpretation of electromagnetic method inversion.
Audio Magnetotelluric Method (AMT) is an important geophysical exploration technique widely applied in geological surveys and mineral exploration. Conductivity and magnetic permeability are two key physical parameters that influence AMT responses, and phenomena such as conductivity anisotropy and non-zero magnetic susceptibility in the Earth like magnetite are commonly observed in real geophysical scenarios. However, current mature AMT forward and inverse modeling techniques typically consider only the conductivity, with insufficient understanding of AMT responses under conditions of conductivity anisotropy and non-zero susceptibility. Therefore, we considering both arbitrary conductive anisotropy and magnetic permeability. We derived and implemented a finite element forward modeling algorithm for one-dimensional layered media in detail, and further analyzed the responses of typical models. The results indicate that when media with strong magnetic susceptibility, such as ferromagnetic medium, it is necessary to take the influence of magnetic susceptibility into consideration. Especially, low-resistive bodies may exhibit high apparent resistivity due to the effects of high magnetic susceptibility. This study provides a framework for investigating the impacts of conductive anisotropy and magnetic permeability and serves as a reference for subsequent numerical simulations of two-dimensional and three-dimensional complex scenarios.
Transient Electromagnetic Method (TEM), as a time-domain active source electromagnetic method, has been successfully applied in multiple fields and has great potential for development in deep resource exploration. To further explore whether multi-component detection is beneficial for increasing detection depth, this article first starts from the TEM three component method and derives an approximate expression for the late three component response in detail; Secondly, by combining approximate expression independent variables, the attenuation characteristics of the y and z components in deep detection under different receiving and transmitting distances are studied; Finally, taking the geological conditions of the oil and gas reservoir as a reference, a qualitative analysis was conducted on the detection depth of the two components of the magnetic field. The research results indicate that the y-component is an effective TEM method for receiving components, and the signal strength is closely related to the transmitting and receiving distance. At an appropriate transmitting and receiving distance, the y-component can compensate for the fast attenuation of the z-component in the later stage. Two component reception can improve the signal-to-noise ratio of TEM detection and reduce the multiplicity of inversion solutions.
The Ground-Airborne Transient Electromagnetic Method (GATEM), as an efficient geophysical exploration technique, exhibits significant potential in areas with complex topography. However, the influence of topographic effects on electromagnetic field distribution is pronounced, leading to the superposition of topographic anomalies and target-related anomalies in the observed data, which poses challenges for data interpretation. This paper proposes a GATEM Born approximation imaging method considering the topographic effects, aiming to eliminate topographic effects and accurately characterize the morphological features of underground electrical interfaces. Firstly, based on the theory of transient electromagnetic wavefield inverse transformation, the diffusion field can be converted into the pseudo wavefield. Then, by applying the Born approximation algorithm, rapid imaging of underground electrical interfaces is achieved. Finally, the correction algorithm is introduced to eliminate topographic effects, thereby mitigating the influence of terrain variations on the imaging results. The effectiveness of the proposed method is verified through theoretical calculations using a two-layer D-type model incorporating typical terrains such as peaks and valleys. Furthermore, both theoretical model results and field data applications confirm that the method can accurately delineate the position and morphology of subsurface electrical interfaces. This approach offers a novel and practical solution for the interpretation of GATEM data in complex topographic environments.
Antarctica and the Southern Ocean (referred to as "Antarctica") are integral parts of the Earth's system, interconnected with its northern regions through oceanic and atmospheric coupling. "Understanding the polar regions, protecting the polar regions, and utilizing the polar regions" is national development strategy of China. Over 98% of the Antarctic continent is covered by ice and snow, making it difficult to obtain subglacial geological information and hindering our understanding of the continent's geology and resource environment. Compared to potential field and electromagnetic field methods, reflection seismic exploration offers advantages in both exploration depth and resolution, which penetrates the ice sheet to detect ice layers and sub-ice geological structures, making it a crucial tool for understanding the Antarctic snow-ice-bedrock structure and playing a significant role in Antarctic exploration. This paper reviews the progress of reflection seismic exploration in Antarctica and summarizes the development history of Antarctic seismic exploration, including seismic sources, geophones, observation systems, and more. By combining actual seismic data from the Thwaites Glacier region with forward modeling, this study analyzes the characteristics of seismic wave fields under the special medium model of Antarctic snow-ice-bedrock. Additionally, the paper summarizes the applications of seismic exploration in Antarctic research and provides an analysis and outlook for China's reflection seismic exploration research in Antarctica.
The complex buried fracture zone in the N-region of the South China Sea presents significant challenges for fracture prediction due to its deep burial, weak seismic reflection signals, steep dips, and low-frequency characteristics. Conventional fracture prediction methods often fail to accurately delineate the fracture development zones and their distribution patterns. This study proposes a novel complex buried fracture zone prediction technique suitable for the area, focusing on fault signal enhancement processing. First, the seismic data undergoes structure-oriented filtering, followed by texture filtering using a grayscale co-occurrence matrix calculated along different directions. This yields fracture-sensitive data volumes in multiple orientations, which are then enhanced through coherence processing. The optimal fracture-sensitive data volume is selected, and multi-attribute fusion is performed by combining coherence enhancement and curvature. In application to the study area, this method outperforms single coherence enhancement and curvature results, providing a more detailed representation of fracture zones at various scales and clearly depicting fault characteristics, and improve the accuracy of fracture prediction.
At present, the construction of large-scale water conservancy and electric power projects has been basically completed. It has entered the post-hydropower era and is in the project operation period. The leakage and deformation of DAMS and dikes are the core problems affecting their safe operation. In recent years, a lot of application research work has been carried out on the basis of magnetoresistivity method to solve the problem of concentrated seepage of earth-rock embankment. In this paper, a new detection method, magnetic gradient method, is proposed on the basis of previous research. Firstly, the research progress at home and abroad is reviewed, the basic principle of magnetic gradient method is described, and the distribution characteristics and detection feasibility of the magnetic field vector and its gradient in the leakage channel are analyzed by theoretical model. Secondly, the technical equipment and physical test process involved in magnetic gradient method are introduced. Finally, the engineering case proves that the method can be effectively applied to the detection of seepage channels of embankment and rock, and the future research work of this method is prospected.
To enhance the comprehension of the relevant laws governing the geoelectric field of levee leakage models under artificial current source excitation, and to improve the realism of numerical simulations, this study establishes coupled equations for the multi-physical fields involved, including the water current field, ion diffusion field, and stabilized electric field of the levee leakage model. Additionally, coupled conversion equations for the multi-physical attributes of porosity, moisture content, mineralization, and resistivity are formulated. The three-dimensional simulation of the geoelectric field in the levee leakage model is achieved utilizing a finite-infinite element multi-physical field numerical simulation platform. By leveraging this platform, the coupling of geoelectric field leakage is accurately simulated. Through dimensional numerical simulations of a typical dam leakage model, it is demonstrated that the electric field resulting from dam leakage comprises three components: the filtered electric field, the diffusion adsorption electric field, and the stabilized electric field, all of which are realistically reproduced. In the stabilized electric field, the dominant current density induced by the collector effect reaches amplitudes of several thousand millivolts, while the filtered and adsorbed electric fields play an enhancing role with amplitudes ranging from a few tens of microvolts to a few tens of millivolts. These findings provide a theoretical basis for the rapid monitoring of dam leakage channels.
It is currently challenging to accurately detect overburden cavities ahead of shield tunneling in subway tunnels. Given the shallow depth of subway tunnels, this study investigates the principle of super-ahead detection of overburden cavities in shield tunnels using small coil transient electromagnetic detection technology, as well as the transient electromagnetic anomaly characteristics. Numerical simulation studies show that when the transient electromagnetic coil is located directly above the cavity, an area resembling a "butterfly" is generated in the electric field response. The closer the distance to the cavity, the greater the electric field response influenced by the cavity. Due to the rapid attenuation of the transient electric field in the cavity, differences exist in the apparent resistivity responses, allowing for the identification of underground cavities. Through engineering practice, the transient electromagnetic anomaly characteristics of dry cavities mainly manifest as isolated, discontinuous, and easily closed-loop electromagnetic curves, with high resistivity anomalies in apparent resistivity data. Wet cavities exhibit transient electromagnetic anomaly characteristics with some continuity in the electromagnetic curves, often showing convex or concave features, and low resistivity anomalies in apparent resistivity data. Ground transient electromagnetic methods can effectively identify shallow underground cavities.
Grotto temples are invaluable cultural heritage and crucial physical resources for studying the history of Buddhism and ancient sculptural art in China. However, they are susceptible to natural environmental factors like weathering and erosion, which can lead to the development of joints, fractures, and karst cavities in the rock of the grotto ceilings. These geological issues can cause water leakage and structural instability, threatening the long-term preservation of the caves and their wall sculptures. To tackle these challenges, we employed a combined approach using ground-coupled shielded antennas with center frequencies of 400 MHz and 270 MHz. This method utilizes a grid layout for comprehensive non-destructive detection of concealed geological structures in the ceiling of a specific grotto temple. The collected Ground Penetrating Radar (GPR) data from the grid lines were then analyzed using three-dimensional (3D) visualization techniques. The results reveal that the integrated use of 400 MHz and 270 MHz antennas not only allows for the detailed delineation of geological anomalies, such as lithological interfaces, fractures, and karst cavities within the cave ceilings, but also clearly outlines the topographic relief of the rock mass. Furthermore, the 3D visualization of the GPR data enables viewing and analyzing the 3D radar data volume from various angles and dimensions, providing an intuitive representation of the rock structure, anomalies, and other relevant information. This significantly enhances the efficiency and accuracy of data interpretation. The application of GPR non-destructive detection technology is highly beneficial for analyzing the geological causes and pathways of water leakage in grotto temples. It provides an important basis for water hazard management, cultural relic preservation, and reinforcement efforts.
Faults, as one of the main geological structures, are crucial for analyzing subsurface structures and determining oil and gas enrichment areas. Traditional methods face challenges in the efficiency of seismic data feature extraction and the accuracy of fault identification. This article first outlines the background of seismic fault identification and the limitations of traditional methods, and then explores the application of deep learning methods in this field. In deep learning, fault identification is regarded as an image processing task, usually trained in a supervised learning manner. Data, model, and loss function are the three core elements of supervised deep learning. Data is the foundation for training deep learning models, and the quality, diversity, and representativeness of data are crucial for the training and generalization ability of the model; the model can establish a nonlinear expression of the relationship between input and output, used to learn patterns and rules in the data; the loss function is used to quantify the difference between the model's predictions and the true labels, and a good loss function can guide the model towards more accurate optimization continuously. This paper first introduces the training dataset and methods of data fusion and difference optimization, then discusses the effectiveness of different deep learning models such as Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), and Transformer models in seismic fault identification, and finally analyzes the impact of different loss functions. This paper summarizes the current performance, advantages, and challenges of deep learning methods in seismic fault identification and provides an outlook on possible future research directions.
Artificial intelligence technology is one of the most important development directions of detailed reservoir description in the future. Detailed reservoir description provides a high-quality platform and foundation for the development and application of artificial intelligence technology. Artificial intelligence also provides a powerful tool and way for the development and progress of fine reservoir description from digitization to intelligence. The research status, advantages and disadvantages of artificial intelligence technology application in fine reservoir description at home and abroad are compared.The application of artificial intelligence technology almost covers all aspects of detailed reservoir description, mainly including fine stratigraphic division and comparison based on analogy learning, fine interpretation of volcanic reservoir structure based on ant colony algorithm, sedimentary microfacies and reservoir configuration division and identification of expert system, fine logging secondary interpretation based on artificial neural network, fine reservoir evaluation based on grey system theory, training image establishment and multi-point geostatistical modeling based on machine learning, knowledge discovery and data mining reservoir flow unit research, fine reservoir description result management platform based on knowledge system, etc. Finally, 10 problems existing in the application of artificial intelligence technology in fine reservoir description and 10 development directions in the future are pointed out.
Interval velocity is a key parameter for obtaining high signal-to-noise ratio, high fidelity, and high-resolution seismic profiles. In recent years, with the help of the nonlinear mapping ability of deep learning, interval velocity modeling method based on seismic reflection waveform data of different types has development. However, current research is mainly focused on supervised learning, i.e., training a large number of "seismic waveform data-velocity model labeled" data pairs into the input network.This approach faces two drawbacks: first, the high cost and large amount of data for obtaining the actual subsurface velocity structure as a label, and second, the problem that the accuracy of velocity modeling depends directly on the nonlinear mapping ability of the neural network after supervised training.Therefore, in this paper, we propose a semi-supervised self-training velocity modeling method, which trains the initial supervised network with a small number of "seismic waveform data-velocity model label" data pairs, and then generates velocity pseudo-labels using unlabeled seismic waveform data. The "seismic waveform data-velocity model label" data pair and the "unlabeled seismic waveform data-velocity model pseudo-label" data pair are mixed to retrain the network and the velocity model pseudo-labels are iteratively updated using the new network model until the network model is updated by the velocity pseudo-label self-training, and then semi-supervised self-training is realized to improve the accuracy and generalization of the velocity modeling, accuracy and generalization.Meanwhile, in order to compress the spatial and channel redundancy in the convolutional neural network and to improve its performance, spatial and channel reconstruction convolution (SCConv) was used to construct the SCConv-Unet network.Finally, in order to verify that the semi-supervised self-training method is suitable for velocity modeling and can improve the accuracy of velocity modeling, numerical experiments are conducted using a fault velocity model, a horizontal laminar velocity model, and a velocity model containing velocity anomalies.The experimental results show that the accuracy of the semi-supervised self-training velocity modeling method can further improve the velocity modeling accuracy of the supervised learning method; making full use of the potential of the velocity-free labeled seismic waveform data can effectively improve the velocity modeling accuracy and reduce the cost of dataset production. In addition, the SCConv-Unet network shows good generalization ability and nonlinear mapping ability, which helps to accelerate the convergence speed of semi-supervised iterative training.
The acoustic remote detection technology can effectively detect geological structures within tens of meters around the well, which holds significant importance for predicting oil distribution and evaluating reservoir productivity. Currently, fault identification in acoustic remote detection imaging largely relies on manual operation. This paper proposes an automatic fault identification method that integrates the greedy algorithm with the ant colony algorithm. The method uses the capping method to correct outliers, improving the clarity of the imaging. It employs an edge detection method based on pheromone tagging to highlight fault features, utilizes a greedy algorithm for preliminary fault identification, and optimizes the greedy identification results through correlation analysis and horizontal stretching. Based on the results of the greedy identification, the ant colony algorithm is used for secondary identification, addressing the issue of the greedy algorithm getting stuck in local optima when tracking complex faults, while also avoiding the accuracy loss caused by the random search of the ant colony algorithm. Practical applications show that the fault identification method, which utilizes the complementary advantages of greedy algorithm and ant colony algorithm, can accurately identify faults in acoustic remote detection.
The lithology of the transitional facies strata between sea and land in the eastern margin of the Ordos Basin is complex, and the rock mechanics parameters and rock physics response are complex, resulting in insufficient prediction accuracy of rock mechanics parameter prediction models based on traditional methods.The rock mechanics test and the matching acoustic wave and density test were adopt, the parameters of rock sample density, acoustic wave velocity, strength and elastic in the study area were gained, and then carry out the prediction research of rock mechanics parameters based on traditional methods. On this basis, multiple algorithms are selected to build different types of rock mechanics parameter prediction models; According to the prediction effect of different algorithm models, an intelligent fusion prediction model is constructed; The research results show that the traditional methods cannot accurately predict the rock mechanics parameters of the marine continental transitional facies strata in the eastern margin of the Ordos Basin. Different machine learning algorithms have different prediction effects on different types of rock mechanics parameters, and the average relative error is more than 20%, that is, a single machine learning algorithm model is difficult to achieve synchronous and accurate prediction of different types of rock mechanics parameters; The intelligent fusion model has a high prediction accuracy for different types of rock mechanics parameters. The average relative errors of the test set and the training set are about 8.14% and 13.02%, respectively, indicating that the model has achieved synchronous and accurate prediction of different types of rock mechanics parameters; This model is applied to the calculation of horizontal principal stress in the research area, and the relative error between the predicted value and the measured value is relatively small, reflecting that the model can be applied to improve the accuracy of the calculation of horizontal principal stress in the formation.
Accurate identification and classification of lithofacies provide essential support for reservoir evaluation, fluid identification, and reservoir characterization, serving as a critical factor in locating high-quality reservoir development zones and favorable hydrocarbon accumulation areas. Core observation and thin-section analysis are the primary sources of first-hand data for direct lithofacies identification. However, due to limitations in core availability and high analytical costs, lithofacies recognition often requires the integration of well logging data. Compared to traditional well logging-based lithofacies identification methods, deep learning offers the advantages of automated and efficient lithofacies recognition, with improved interpretative accuracy and reduced uncertainty. To address these challenges, this study reviews the application of deep learning in lithofacies recognition using well logging data, systematically summarizing the research findings from two aspects: application conditions and effectiveness. Specifically, the study focuses on lithofacies recognition models based on conventional well logging and models integrating conventional and electrical imaging logging. Drawing on previous research, this paper proposes a preliminary intelligent lithofacies recognition model tailored to the complexities of carbonate lithofacies. Finally, it highlights the challenges in applying intelligent recognition models to well logging lithofacies identification and discusses future development trends in this field.
In actual logging operations, various factors such ascomplex wellbore environment, complex geological structures, and wellbore collapses can lead to the issue of missing logging curves, and the cost of remeasuring data is high. To address the issue of missing logging curves, this paper first employs Singular Spectrum Analysis (SSA) to decompose the original logging curves, utilizing the more correlated components for more efficient curve completion. Furthermore, a logging curves completion model based on graph attention network incorporating Multi-Head Attention Mechanism and Bidirectional Gated Recurrent Units (GAT-MABiGRU) is proposed. In the completion experiments for the RHOB and DT logging curves, results show that the GAT-MABiGRU model based on SSA outperforms Support Vector Regression (SVR), Multi-Layer Perceptron (MLP), Long Short-Term Memory Network (LSTM), and Temporal Convolutional Network (TCN) in terms of Root Mean Squard Error(RMSE), Mean Absolute Error(MAE), and coefficient of determination(R2). Ablation experiments and blind well experiments further verify the effectiveness of incorporating SSA and GAT modules in improving the model's prediction accuracy, providing a new method for logging data completion.
Accurately identifying the interlayer in the reservoir is crucial for the fine characterization of the reservoir and the exploration of remaining oil potential. In order to better utilize logging data and improve the efficiency and accuracy of interlayer partitioning, this paper proposes a interlayer partitioning method based on the BA-Catboost algorithm. In the study, the general methods for identifying and dividing interlayer were compared and analyzed. In response to the difficulties of low efficiency and easy errors in manual division, the technical route of BA-Catboost algorithm was optimized and constructed. By using core logging and other data to identify interlayers and classify their types, ADASYN method was used to increase the sample size of interlayers, and high correlation logging curves such as GR, SP, and AC were selected as feature parameters. Based on the BA-Catboost algorithm, a classification model was trained and established, with model training and testing accuracies of 96.7% and 98.9%, respectively. Using a classification model to identify the feature fuzzy and difficult to divide interlayers, 62 groups of muddy interlayers, 20 groups of calcareous interlayers, and 59 groups of physical interlayers were identified. On this basis, the distribution characteristics of interlayer planes were studied, and it was found that interlayer planes were more developed in the Y2 and Y3 sub layers, showing a high frequency and density of interlayer distribution in the southeast region and a low density in the central and western regions on the plane. The use of this method to divide the interlayer makes up for the lack of understanding in previous production and development processes. Subsequently, by adjusting injection and production measures, using methods such as hole filling and increasing water injection volume, the production increase effect can be achieved. The research results show that the BA-Catboost algorithm has better performance than similar algorithms. The classification model established by this method has good training and testing effects, and is used for fine recognition and automatic classification of interlayer, improving recognition accuracy and efficiency. It can effectively guide production and development work and has application value in J area of Longdong oilfield.
With its powerful feature extraction ability, deep learning has shown great potential in various fields and provides new ideas for solving various complex problems. Deep learning models often require a large amount of labeled data for training, but in practice, limited logging data are obtained due to cost, resulting in insufficient training samples. Therefore, this paper proposes a CNN-BiLSTM based semi-supervised learning method for seismic wave impedance inversion. The interpolation resampling technique is used to augment the wave impedance, and then a semi-supervised learning strategy is introduced to train the augmented data, and the unlabeled data information is used to improve the generalization ability and performance of the model. The Marmusi-2 model test shows that it can achieve better inversion results with only a small amount of data augmentation, which verifies the effectiveness of the method in the case of small samples.
Accurate identification of microseismic events is the basis of data processing in microseismic monitoring. To address the issue of low accuracy in identifying microseismic events using deep learning methods, this paper firstly constructs a basic semi-supervised Generative Adversarial Network (GAN) classification model based on downhole microseismic monitoring data. The model consists of a generator for simulating the distribution of real data and a discriminator for identifying microseismic events. Next, layer normalization is introduced to reduce the training loss of the discriminator. Meanwhile, a convolutional interpolation method is applied to the generator to improve its ability of autonomously learning and extracting detailed signal features. In order to verify the effectiveness of the proposed method, actual microseismic data from fracturing monitoring is used as the dataset for training and testing the model. Experimental results indicate that the identification method based on semi-supervised GAN outperforms the identification method based on convolutional neural network in terms of accuracy and precision. Compared with the latter model, the former model has faster convergence and more stable training results. The accuracy of the test set for the improved semi-supervised GAN identification model can reach 97%, and all the test indicators of this model have been improved. The improved method can better learn the shape features of microseismic events, effectively identifying microseismic event samples, which increases the identification rate of microseismic events based on neural network classification models.
Affected by near-surface absorption, seismic wave energy attenuation and phase distortion lead to a significant reduction in the resolution and signal-to-noise ratio of seismic data. Conventional inverse Q filtering methods have problems such as unstable amplitude compensation and difficult parameter selection. To address these problems, a new unsupervised learning inverse Q filtering high-resolution processing method is proposed. This method integrates the Deep Learning (DL) framework and the seismic wave absorption attenuation theory based on unconditional numerical stability, and provides a DL inverse Q filtering strategy that does not require training labels and avoids numerical instability in amplitude compensation. First, a DL network is constructed, the data to be compensated is input into the network, and the network output is used as the compensated result. Then, the predicted compensation result is sent to the attenuation kernel matrix constructed by the near-surface Q model for forward attenuation. Next, the error between the attenuated seismic data obtained by forward modeling and the original data to be compensated is used to reversely adjust the network parameters, and the error is minimized by iteratively optimizing the network parameters, and the final compensation result is output. In the entire training network prediction process, no data labels are required, and the effect of unsupervised autonomous learning is achieved. The application results of theoretical model data and actual pre-stack seismic data show that compared with the conventional inverse Q filtering method, the unsupervised method can effectively compensate the amplitude energy of seismic signals and has high numerical stability. This method improves the resolution and signal-to-noise ratio of seismic records.
K Singular Value Decomposition (KSVD) dictionary learning has been successfully applied in the field of seismic data denoising. In order to make more efficient, full and accurate use of data, this paper proposes a Monte Carlo-filter dictionary learning (MC-FDL) method based on Monte Carlo segmentation for seismic data denoising, which uses Monte Carlo segmentation method to obtain pre-trained dictionaries, this enables the dictionary to learn the characteristics of the signal to a greater extent. First, the variance of all the data blocks is calculated, a uniformly distributed random number is generated for each data block, and if the variance of the block is greater than the random number, the block is selected as an atom in the dictionary. Then the dictionary is trained and updated by KSVD algorithm, and the interference of noise in dictionary atoms is filtered by median filter to make the dictionary more prominent in the feature of seismic data. After that, the trained dictionary is used for seismic data denoising, and the denoising performance of the proposed method is tested by synthetic and real data. Finally, the influence of different block selection methods and different data preprocessing methods on the denoising results is discussed in detail, and the block selection strategy of the proposed method is determined, and the future work is prospected.
The airborne transient electromagnetic method is an important electromagnetic exploration technology, which obtains the information about the electrical structure of the earth through inversion.But due to the slow progress of the forward, the inversion will consume a significant amount of time.Aiming at the problem of long computation time of traditional forward method, this paper proposes a deep learning-based fast forward method for airborne transient electromagnetic.The method first uses the traditional finite volume method to calculate the induced electromotive force of a large number of different ground resistivity models to form a training dataset; then it designs a ResNet-UNet deep neural network; then it trains the network with the training dataset; finally, it inputs the ground resistivity model into the trained neural network to obtain the forward results.In order to verify the accuracy and efficiency of the method, the forward results of the ResNet-UNet deep neural network and the traditional finite volume method are compared.The experimental results show that the average relative errors of the entire validation set is less than 1.2%, with 87% of the average relative errors falling within the range of 0.1% to 0.3%, and the speed of deep neural network forward is about 2934 times higher than that of the traditional finite volume method, which significantly improves the forward efficiency of the airborne transient electromagnetic.The method is capable of fast forward of airborne transient electromagnetic, which can be put into the existing inversion framework to accelerate the inversion speed of large datasets.
ISSN 1004-2903 (Print)
Started from
Published by:
