A time-frequency electromagnetic exploration method and system for improving exploration depth and resolution
By combining a mobile remote-controlled array observation system with deep learning algorithms, the problems of observation limitations and data redundancy in traditional time-frequency electromagnetic exploration methods have been solved, enabling efficient identification of deep geological bodies and accurate construction of three-dimensional geological models.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SICHUAN TUOCHUANG DETECTION TECHNOLOGY CO LTD
- Filing Date
- 2026-03-05
- Publication Date
- 2026-06-05
AI Technical Summary
Traditional time-frequency electromagnetic exploration methods suffer from problems such as limited observation, data redundancy, inefficient inversion, and low reliability of interpretation. They are difficult to achieve large-scale, multi-directional synchronous observation, have a large amount of data that is easily affected by noise, and the inversion results are highly ambiguous, resulting in insufficient accuracy in identifying deep geological bodies.
A mobile remote-controlled array observation system equipped with multi-component sensors was used. Based on compressed sensing theory, sparse sampling was performed to construct a sparse basis matrix of time-frequency electromagnetic signals and an incoherent observation matrix. A deep reinforcement learning algorithm was combined to perform joint inversion of time-frequency dual domains. Cross-gradient constraints and Bayesian uncertainty analysis were introduced, and multi-source data from seismic, gravity and magnetic methods were fused for verification.
It enables large-scale, multi-directional, simultaneous exploration, reduces data acquisition costs, improves exploration depth and resolution, enhances the reliability of inversion results and the accuracy of geological models, and supports real-time monitoring and dynamic updates.
Smart Images
Figure CN122151220A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of exploration technology, and more specifically, to a time-frequency electromagnetic exploration method and system for improving exploration depth and resolution. Background Technology
[0002] Geophysical exploration is the core means of obtaining underground geological information. Time-frequency electromagnetic exploration technology is widely used in deep resource exploration because it has the advantages of both time-domain and frequency-domain detection.
[0003] However, traditional time-frequency electromagnetic exploration methods have significant shortcomings. Observation systems often rely on fixed ground deployments, are heavily constrained by terrain, have low data acquisition efficiency, and struggle to achieve large-scale, multi-directional synchronous observations. Signal sampling must adhere to the Nyquist sampling rate, resulting in massive data volumes, high transmission and storage costs, and susceptibility to noise interference. During the inversion process, time-domain and frequency-domain data are independent, lacking physical constraints, leading to multiple solutions in the inversion results. Traditional inversion algorithms rely on human experience, have slow convergence speeds, struggle to balance exploration depth and resolution, and suffer from insufficient accuracy in identifying deep geological bodies. Furthermore, single electromagnetic exploration data is easily affected by geological conditions, lacks cross-validation from multi-source data, and results in low reliability of geological interpretation. These problems severely limit the application of time-frequency electromagnetic exploration technology in deep resource exploration and complex geological structure detection, necessitating an intelligent exploration solution that balances exploration depth, resolution, and efficiency. Summary of the Invention
[0004] In order to overcome the problems of limited observation, data redundancy, inefficient inversion, and low reliability of interpretation in existing technologies, this invention discloses a time-frequency electromagnetic exploration method and system that can improve exploration depth and resolution and effectively solve the above-mentioned technical problems.
[0005] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows: A time-frequency electromagnetic exploration method for improving exploration depth and resolution includes the following steps: The exploration area is delineated, and a multi-directional excitation source and a mobile remote terminal array observation system are deployed. The mobile remote terminal array observation system is equipped with a multi-component electric field sensor and a magnetic field sensor to simultaneously collect time-domain electromagnetic signals and frequency-domain electromagnetic signals within the exploration area. Based on compressed sensing theory, a sparse basis matrix and an incoherent observation matrix for time-frequency electromagnetic signals are constructed. A sparse sampling method with a sampling rate much lower than that of the Nyquist is used to obtain sparse sampling data in the time domain and frequency domain. The sparse sampled data is reconstructed using a norm minimization optimization algorithm to obtain complete time-domain and frequency-domain electromagnetic data, and noise suppression preprocessing is performed on the reconstructed data. A joint time-frequency domain inversion objective function is constructed, cross-gradient constraints are introduced to establish the physical relationship between time-domain data and frequency-domain data, and a deep reinforcement learning algorithm is used to construct an intelligent inversion model. The preprocessed time-frequency domain data is input into the model to carry out joint inversion. Multi-scale features of time-frequency dual-domain data are extracted using deep learning networks, feature fusion is achieved through an attention mechanism, and the inversion process is optimized by combining physical constraints of Maxwell's equations to obtain an initial geological model. Bayesian uncertainty analysis is performed on the initial geological model to quantify the probability distribution and uncertainty information of the inversion results and output a high-precision three-dimensional geological model. By integrating multi-source geophysical data from seismic, gravity, and magnetic methods, the three-dimensional geological model is validated and corrected, and the geological bodies in the exploration area are identified and interpreted.
[0006] Preferably, the step of constructing the sparse basis matrix of the time-frequency electromagnetic signal and the incoherent observation matrix based on compressed sensing theory specifically includes: The sparse basis matrix is selected from one or more combinations of multi-scale transform bases, including wavelet basis, Fourier basis, and curvelet basis, which are used to realize sparse representation of time-frequency electromagnetic signals in the transform domain. The incoherent observation matrix is a random Gaussian matrix, and its finite isochronous property is verified to ensure signal reconstruction accuracy. During sparse sampling, the sampling location and observation attitude information are recorded simultaneously.
[0007] Preferably, the signal reconstruction and noise suppression preprocessing steps specifically include: The norm minimization optimization problem is solved by matching pursuit algorithm or basis pursuit algorithm. The optimization problem uses sparse sampled data as observations and constructs constraint relationships with observation matrix, sparse basis matrix and sparse coefficients. An adaptive threshold denoising algorithm is used to process the reconstructed time-frequency electromagnetic data to suppress random noise and power frequency interference.
[0008] Preferably, the intelligent inversion model constructed by the deep reinforcement learning algorithm specifically includes: The geological parameters are treated as cooperative agents by employing either the deep deterministic policy gradient algorithm or the multi-agent deep deterministic policy gradient algorithm. The inversion process is modeled as a Markov decision process. The agent learns the optimal inversion strategy through trial and error and outputs the optimal estimates of various geological parameters. The joint inversion objective function is a weighted sum of the time-domain data fitting error, the frequency-domain data fitting error, and the cross-gradient constraint term.
[0009] Preferably, the step of time-frequency dual-domain multi-scale feature fusion specifically includes: Convolutional neural networks or Transformer networks are used to extract time-series features from time-domain data and spectral features from frequency-domain data, respectively. An attention mechanism is introduced to adaptively weight and fuse multi-scale features, enhancing feature components that are sensitive to both deep targets and shallow details; Embedding Maxwell's equations physical constraint modules into deep learning networks ensures that the fused features conform to the physical laws of electromagnetic propagation, thereby improving the physical interpretability of the inversion results.
[0010] Preferably, the operation of the mobile remote-controlled array observation system specifically includes: Multiple mobile remote terminals coordinate according to a preset spatial array to form a ground-air joint observation system, and electromagnetic signals are synchronously excited by multi-directional ground excitation sources; The sensors mounted on the mobile remote terminal simultaneously measure the three components of the electric field and the three components of the magnetic field, enabling multi-component collaborative detection; The collected data is transmitted to the cloud acquisition platform in real time via a high-speed communication module, and the edge computing nodes simultaneously preprocess the transmitted data.
[0011] Preferably, the steps of the Bayesian uncertainty analysis specifically include: A large number of inversion samples were generated based on the Markov chain Monte Carlo algorithm, and the posterior probability distribution of various geological parameters was statistically analyzed. Calculate the mean, variance, and confidence interval of each geological parameter to quantify the uncertainty of the inversion results; The initial geological model is corrected by removing inversion samples with abnormal probability distributions.
[0012] Preferably, the multi-source data fusion verification step specifically includes: Establish the correlation mapping relationship between seismic, gravity, and magnetic data and electromagnetic inversion models, and construct comprehensive geophysical constraints; A weighted fusion algorithm is used to fuse geological models corresponding to multi-source data, with the weights determined based on the resolution and reliability of each exploration method. The accuracy of the fusion model was evaluated using cross-validation.
[0013] Preferably, it also includes a real-time monitoring and dynamic updating step: During the exploration process, new time-frequency electromagnetic data are continuously collected, and the established three-dimensional geological model is dynamically updated using the newly collected data. The intelligent interpretation system automatically identifies anomalous geological bodies in the updated model, classifies and labels the anomaly types and confidence levels, and generates intelligent exploration reports.
[0014] Preferably, a time-frequency electromagnetic exploration system for improving exploration depth and resolution includes: The excitation module includes multi-directional ground excitation sources to generate electromagnetic excitation signals of different frequencies and intensities. The observation module is a mobile remote array observation system, equipped with a multi-component electric field sensor, a magnetic field sensor and a high-speed communication unit, used for sparse acquisition of time-frequency electromagnetic signals and real-time transmission. The data processing module is built on an open-source distributed operating system to create a cloud-edge collaborative platform. Edge nodes are responsible for data preprocessing and signal reconstruction, while cloud nodes are responsible for deep inversion and feature fusion. The intelligent inversion module integrates deep reinforcement learning algorithms and physically constrained deep learning networks for joint time-frequency dual-domain inversion and uncertainty analysis. The interpretation and visualization module is used for multi-source data fusion verification, automatic anomaly identification, and output of 3D geological models and intelligent exploration reports.
[0015] Compared with existing technologies, the advantages of this invention are as follows: This technology employs a mobile remote-controlled array observation system, where multiple mobile remote-controlled terminals coordinate according to a preset array, equipped with multi-component sensors to synchronously acquire time-frequency electromagnetic signals, overcoming terrain limitations in observation and achieving joint multi-directional detection, thus improving data acquisition efficiency and coverage. Because it uses a sparse sampling method with a sampling rate far lower than the Nyquist sampling rate based on compressed sensing theory, combined with a norm minimization optimization algorithm to reconstruct complete data, it reduces data transmission and storage costs while suppressing interference through adaptive threshold denoising, ensuring data quality. Furthermore, by constructing a joint time-frequency dual-domain inversion objective function and introducing cross-gradient constraints to establish the physical... By combining deep reinforcement learning-based intelligent inversion models with the fusion of Maxwell's equations' physical constraints and attention mechanisms, the inversion ambiguity is effectively reduced, while simultaneously considering both the depth of deep geological body exploration and the detail resolution of shallow areas. By quantifying the probability distribution of the inversion results through Bayesian uncertainty analysis and eliminating outliers, and then cross-validating the results with multi-source data from seismic, gravity, and magnetic methods, the accuracy and reliability of the 3D geological model are improved. Furthermore, the real-time monitoring and dynamic update mechanism can optimize the model based on newly acquired data, enabling automatic identification and labeling of anomalous geological bodies, further enhancing the intelligence level and interpretation efficiency of exploration, and providing technical support for deep resource exploration and the detection of complex geological structures. Attached Figure Description
[0016] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary. For those skilled in the art, other embodiments can be derived from the provided drawings without creative effort.
[0017] Figure 1A flowchart illustrating the steps of a time-frequency electromagnetic exploration method to improve exploration depth and resolution; Figure 2 This is a structural diagram of a time-frequency electromagnetic exploration system designed to improve exploration depth and resolution. Detailed Implementation
[0018] The accompanying drawings are for illustrative purposes only and should not be construed as limiting the scope of this patent. To better illustrate this embodiment, some parts in the accompanying drawings may be omitted, enlarged, or reduced, and do not represent the actual product dimensions; It will be understood by those skilled in the art that certain well-known structures and their descriptions may be omitted in the accompanying drawings.
[0019] The embodiments of this application will be further described in detail below with reference to the accompanying drawings and examples.
[0020] It is understood that the specific embodiments described herein are merely illustrative of the embodiments of this application and are not intended to limit the embodiments of this application. Furthermore, it should be noted that, for ease of description, the accompanying drawings only show the parts related to the embodiments of this application, not all structures. Those skilled in the art, after reading this specification, should be able to realize that any combination of technical features can constitute an optional implementation method, provided that the technical features do not contradict each other.
[0021] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first," "second," etc., are generally of the same class and the number of objects is not limited; for example, a first object can be one or more. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship. In the description of this application, "multiple" means two or more, and "several" means one or more.
[0022] Example 1 This embodiment addresses the core problems of traditional time-frequency electromagnetic exploration in deep resource exploration and complex geological structure detection, such as limited observation, data redundancy, strong inversion ambiguity, and difficulty in balancing resolution and exploration depth. It discloses a high-performance time-frequency electromagnetic exploration scheme that integrates mobile remote terminal array observation, compressed sensing, intelligent inversion, and multi-source data fusion.
[0023] Please see Figure 1 A time-frequency electromagnetic exploration method for improving exploration depth and resolution includes the following steps: The exploration area is delineated, and a multi-directional excitation source and a mobile remote terminal array observation system are deployed. The mobile remote terminal array observation system is equipped with a multi-component electric field sensor and a magnetic field sensor to simultaneously collect time-domain electromagnetic signals and frequency-domain electromagnetic signals within the exploration area. Based on compressed sensing theory, a sparse basis matrix and an incoherent observation matrix for time-frequency electromagnetic signals are constructed. A sparse sampling method with a sampling rate much lower than that of the Nyquist is used to obtain sparse sampling data in the time domain and frequency domain. The sparse sampled data is reconstructed using a norm minimization optimization algorithm to obtain complete time-domain and frequency-domain electromagnetic data, and noise suppression preprocessing is performed on the reconstructed data. A joint time-frequency domain inversion objective function is constructed, cross-gradient constraints are introduced to establish the physical relationship between time-domain data and frequency-domain data, and a deep reinforcement learning algorithm is used to construct an intelligent inversion model. The preprocessed time-frequency domain data is input into the model to carry out joint inversion. Multi-scale features of time-frequency dual-domain data are extracted using deep learning networks, feature fusion is achieved through an attention mechanism, and the inversion process is optimized by combining physical constraints of Maxwell's equations to obtain an initial geological model. Bayesian uncertainty analysis is performed on the initial geological model to quantify the probability distribution and uncertainty information of the inversion results and output a high-precision three-dimensional geological model. By integrating multi-source geophysical data from seismic, gravity, and magnetic methods, the three-dimensional geological model is validated and corrected, and the geological bodies in the exploration area are identified and interpreted.
[0024] The steps for constructing the sparse basis matrix and incoherent observation matrix of the time-frequency electromagnetic signal based on compressed sensing theory specifically include: The sparse basis matrix is selected from one or more combinations of multi-scale transform bases, including wavelet basis, Fourier basis, and curvelet basis, which are used to realize sparse representation of time-frequency electromagnetic signals in the transform domain. The incoherent observation matrix is a random Gaussian matrix, and its finite isochronous property is verified to ensure signal reconstruction accuracy. During sparse sampling, the sampling location and observation attitude information are recorded simultaneously.
[0025] The signal reconstruction and noise suppression preprocessing steps specifically include: The norm minimization optimization problem is solved by matching pursuit algorithm or basis pursuit algorithm. The optimization problem uses sparse sampled data as observations and constructs constraint relationships with observation matrix, sparse basis matrix and sparse coefficients. An adaptive threshold denoising algorithm is used to process the reconstructed time-frequency electromagnetic data to suppress random noise and power frequency interference.
[0026] The intelligent inversion model constructed by the deep reinforcement learning algorithm specifically includes: The geological parameters are treated as cooperative agents by employing either the deep deterministic policy gradient algorithm or the multi-agent deep deterministic policy gradient algorithm. The inversion process is modeled as a Markov decision process. The agent learns the optimal inversion strategy through trial and error and outputs the optimal estimates of various geological parameters. The joint inversion objective function is a weighted sum of the time-domain data fitting error, the frequency-domain data fitting error, and the cross-gradient constraint term.
[0027] The steps of the time-frequency dual-domain multi-scale feature fusion specifically include: Convolutional neural networks or Transformer networks are used to extract time-series features from time-domain data and spectral features from frequency-domain data, respectively. An attention mechanism is introduced to adaptively weight and fuse multi-scale features, enhancing feature components that are sensitive to both deep targets and shallow details; Embedding Maxwell's equations physical constraint modules into deep learning networks ensures that the fused features conform to the physical laws of electromagnetic propagation, thereby improving the physical interpretability of the inversion results.
[0028] The working process of the mobile remote terminal array observation system specifically includes: Multiple mobile remote terminals coordinate according to a preset spatial array to form a ground-air joint observation system, and electromagnetic signals are synchronously excited by multi-directional ground excitation sources; The sensors mounted on the mobile remote terminal simultaneously measure the three components of the electric field and the three components of the magnetic field, enabling multi-component collaborative detection; The collected data is transmitted to the cloud acquisition platform in real time via a high-speed communication module, and the edge computing nodes simultaneously preprocess the transmitted data.
[0029] The steps of the Bayesian uncertainty analysis specifically include: A large number of inversion samples were generated based on the Markov chain Monte Carlo algorithm, and the posterior probability distribution of various geological parameters was statistically analyzed. Calculate the mean, variance, and confidence interval of each geological parameter to quantify the uncertainty of the inversion results; The initial geological model is corrected by removing inversion samples with abnormal probability distributions.
[0030] The steps for multi-source data fusion verification specifically include: Establish the correlation mapping relationship between seismic, gravity, and magnetic data and electromagnetic inversion models, and construct comprehensive geophysical constraints; A weighted fusion algorithm is used to fuse geological models corresponding to multi-source data, with the weights determined based on the resolution and reliability of each exploration method. The accuracy of the fusion model was evaluated using cross-validation.
[0031] It also includes real-time monitoring and dynamic updating steps: During the exploration process, new time-frequency electromagnetic data are continuously collected, and the established three-dimensional geological model is dynamically updated using the newly collected data. The intelligent interpretation system automatically identifies anomalous geological bodies in the updated model, classifies and labels the anomaly types and confidence levels, and generates intelligent exploration reports.
[0032] In practice, the delineation of exploration areas and the deployment of systems include: clearly defining the boundaries of the exploration area using geographic information software, marking the topography and known geological control points to provide a reference for subsequent system deployment and data interpretation; installing controllable source electromagnetic transmitters in multiple directions around the exploration area, debugging the equipment to ensure that all parameters meet the requirements, and controlling the position error within a reasonable range by calibrating the transmitter position through positioning; after the mobile remote terminal is powered on and preheated, performing sensor zero-point calibration and sensitivity testing and calibration to ensure that the array layout accuracy meets the requirements; and the ground control center monitoring the status and communication stability of the mobile remote terminal in real time.
[0033] For sparse acquisition of time-frequency electromagnetic signals, a combination of multi-scale transform bases is selected as the sparse basis matrix. Through multi-scale transformation, the time-frequency electromagnetic signals are sparsely represented in the transform domain, ensuring that the signal sparsity meets the reconstruction requirements. For the construction of the incoherent observation matrix, a random Gaussian matrix is used as the observation matrix. Numerical verification is performed to ensure that it satisfies the finite equidistant property, thus ensuring the accuracy of signal reconstruction. The mobile remote terminal array synchronously acquires time-domain and frequency-domain electromagnetic signals. Data is recorded in real time during the sampling process, and data quality reports are generated periodically.
[0034] Signal reconstruction and noise suppression preprocessing employs a norm minimization optimization algorithm (matching pursuit algorithm or basis pursuit algorithm) to solve the optimization problem. Constraint relationships are constructed based on sparse sampled data, observation matrix, sparse basis matrix, and sparse coefficients to reconstruct complete time-frequency electromagnetic data. An adaptive threshold denoising algorithm is then used to automatically adjust the denoising threshold by calculating the local variance of the data, effectively suppressing random noise and power frequency interference, and ensuring the integrity and signal-to-noise ratio of the effective signal.
[0035] The time-frequency dual-domain joint intelligent inversion method uses an objective function that includes time-domain data fitting error, frequency-domain data fitting error, and cross-gradient constraint terms. By rationally allocating weights, it establishes a physical correlation between time-domain and frequency-domain data, reducing the ambiguity of inversion. A deep reinforcement learning algorithm is employed, treating geological parameters as agents and modeling the inversion process as a Markov decision process. The agent learns through trial and error to obtain the optimal inversion strategy and outputs the optimal estimates of various geological parameters. An early stopping mechanism is introduced during training to avoid overfitting. A deep learning network extracts time-series features from the time-domain data and spectral features from the frequency-domain data, respectively. An attention mechanism is introduced for adaptive weighted fusion, strengthening the feature components sensitive to deep targets and shallow details. A Maxwell's equation physical constraint module is embedded to verify the physical rationality of the inversion results in real time. Preprocessed time-frequency dual-domain data is input into the trained intelligent inversion model. After iterative calculation, an initial 3D geological model is output. The model's grid resolution meets the requirements for geological body identification and covers the target depth range of the exploration area.
[0036] Bayesian uncertainty analysis is used to generate a large number of inversion samples based on the Markov chain Monte Carlo algorithm. Parallel computing is used to improve efficiency. Each sample meets the physical constraints of geological parameters. The posterior probability distribution of each geological parameter is statistically analyzed, and the mean, variance and confidence interval are calculated to clarify the reliable range of the inversion results. Inversion samples with abnormal probability distributions are removed, and the initial geological model is iteratively corrected to reduce the fitting error between the model and the observation data.
[0037] Multi-source data fusion verification involves collecting geophysical data such as seismic, gravity, and magnetic data from the exploration area. Through coordinate transformation and data resampling, the spatial resolution is ensured to be consistent with the electromagnetic inversion model. A correlation mapping relationship is established between data from different exploration methods and the parameters of the electromagnetic inversion model. Comprehensive geophysical constraints are constructed, and a weighted fusion algorithm is used to fuse the geological models corresponding to the multi-source data based on the resolution and reliability of each exploration method. The accuracy of the fused model is evaluated through cross-validation to ensure the reliability of the model.
[0038] Real-time monitoring and dynamic updates are implemented, with periodic supplementation of time-frequency electromagnetic data during exploration. Sampling is appropriately intensified in anomalous areas to obtain richer geological information. The newly acquired data is used to iteratively update the 3D geological model, optimize geological parameter estimates, and improve the resolution of anomalous areas. An intelligent interpretation system automatically identifies anomalous geological bodies in the updated model, classifies and labels them according to type and confidence level, and automatically integrates data and results from the entire exploration process to generate a standardized exploration report containing core conclusions and recommendations for reference in subsequent work.
[0039] Compared to traditional time-frequency electromagnetic exploration methods, this method significantly increases exploration depth and effectively identifies deep geological targets. Both lateral and vertical resolutions meet the requirements for fine exploration, clearly distinguishing adjacent geological bodies and accurately presenting their spatial distribution characteristics. Sparse sampling greatly reduces data volume, lowering transmission and storage costs. Intelligent inversion significantly shortens inversion time and improves exploration efficiency. Multi-source data fusion improves the accuracy of geological interpretation and reduces exploration risks.
[0040] Example 2 Please see Figure 2 A time-frequency electromagnetic exploration system for improving exploration depth and resolution, comprising: The excitation module includes multi-directional ground excitation sources to generate electromagnetic excitation signals of different frequencies and intensities. The observation module is a mobile remote array observation system, equipped with a multi-component electric field sensor, a magnetic field sensor and a high-speed communication unit, used for sparse acquisition of time-frequency electromagnetic signals and real-time transmission. The data processing module is built on an open-source distributed operating system to create a cloud-edge collaborative platform. Edge nodes are responsible for data preprocessing and signal reconstruction, while cloud nodes are responsible for deep inversion and feature fusion. The intelligent inversion module integrates deep reinforcement learning algorithms and physically constrained deep learning networks for joint time-frequency dual-domain inversion and uncertainty analysis. The interpretation and visualization module is used for multi-source data fusion verification, automatic anomaly identification, and output of 3D geological models and intelligent exploration reports.
[0041] The excitation module uses multiple controllable electromagnetic transmitters, supports wide frequency range excitation, continuously adjustable emission current, and has a frequency scanning working mode, which can generate electromagnetic signals of different intensities to cover geological targets at different depths.
[0042] Excitation sources are deployed in multiple directions around the exploration area to form a surrounding excitation pattern, ensuring that electromagnetic signals uniformly cover the entire exploration area and guaranteeing the signal reception strength of deep geological bodies.
[0043] The excitation frequency covers the low to high frequency range, the excitation duration at each frequency point is moderate, the signal type is sine wave, and key data such as excitation timing, frequency and current intensity are recorded synchronously.
[0044] The observation module (mobile remote array observation system) uses an industrial-grade mobile remote terminal, equipped with a high-precision positioning module and attitude measurement unit. The mobile remote terminal is equipped with a multi-component electromagnetic sensor kit, including a three-component electric field sensor and a three-component magnetic field sensor, which can synchronously acquire time-frequency electromagnetic signals. The sensor response bandwidth covers the excitation frequency range.
[0045] The mobile remote terminal is equipped with a high-speed communication module to transmit the collected data to the cloud data collection platform in real time; it also has a built-in local storage unit to cache the raw data in case of network interruption.
[0046] Multiple mobile remote terminals form a joint observation network according to a preset array to ensure that there are no data blind spots in the exploration area and that the spatial sampling density meets the inversion accuracy requirements.
[0047] The data processing module (cloud-edge collaborative platform) is deployed on the mobile workstation at the exploration site and is responsible for preprocessing tasks such as real-time data reception, format conversion, outlier removal, sparse sampling, and signal reconstruction.
[0048] It adopts a distributed cluster architecture, is built on an open-source distributed operating system, and integrates distributed storage and parallel computing frameworks, and is responsible for deep inversion, feature fusion and multi-source data processing.
[0049] It comes with a compatible programming environment and tool library, including related software packages for signal processing, deep learning, and data visualization, and supports modular algorithm calls and parallel computing.
[0050] The intelligent inversion module integrates a deep reinforcement learning algorithm (selectable single-agent or multi-agent mode), and the feature extraction network adopts a hybrid architecture of CNN and Transformer, embedding a Maxwell equation physical constraint module to ensure that the inversion results conform to physical laws.
[0051] It integrates the Markov chain Monte Carlo algorithm, supports parallel generation of inversion samples, and improves the efficiency of uncertainty analysis.
[0052] Set reasonable training iterations, learning rate, and convergence threshold, and use an appropriate optimizer to optimize the objective function.
[0053] The interpretation and visualization module features an interactive interface that supports real-time rendering of 3D geological models, multi-dimensional slice analysis, anomaly area annotation, and parameter query functions.
[0054] It integrates weighted fusion and cross-validation algorithms to support automatic identification and conversion of seismic, gravity, and magnetic data in various formats.
[0055] It supports the automatic generation of intelligent exploration reports, including core modules such as exploration overview, data quality analysis, inversion results, model verification, and anomaly identification conclusions, and can be exported in a variety of commonly used formats.
[0056] The excitation module communicates with cloud nodes via the network, receives excitation parameter commands, and uploads its working status. The mobile remote terminal array of the observation module establishes a real-time connection with edge nodes through a high-speed network, transmitting acquired data and attitude information. Edge nodes and cloud nodes are connected through a high-speed communication link to achieve data uploading and command issuance. The intelligent inversion module and interpretation module are deployed in the cloud, calling data processing results through internal interfaces to output the final exploration results.
[0057] The cloud node sends excitation parameters to the excitation module → the excitation source synchronously excites electromagnetic signals → the mobile remote terminal array collects time-frequency electromagnetic signals → the edge node preprocesses and reconstructs the data → the cloud node performs intelligent inversion and feature fusion → the interpretation module completes multi-source data verification and anomaly identification → the three-dimensional geological model and exploration report are output to the user.
[0058] This embodiment fully discloses the module composition, connection relationships, and core configuration of a time-frequency electromagnetic exploration system, and elaborates in detail the operation steps, algorithm logic, and key processes of the exploration method. It achieves standardization and reproducibility of the entire process of observation, acquisition, processing, inversion, and interpretation. The system adopts mobile remote terminal array observation to overcome terrain limitations, compressed sensing technology to reduce data redundancy, intelligent inversion algorithm to balance exploration depth and resolution, and multi-source data fusion to improve interpretation reliability. It effectively solves the core problems of traditional methods. This solution is applicable to various exploration scenarios and provides a complete technology for the intelligent upgrading of geophysical exploration technology.
[0059] The same or similar labels correspond to the same or similar parts; The terms used to describe positional relationships in the accompanying drawings are for illustrative purposes only and should not be construed as limiting this patent. Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the implementation of the present invention. For those skilled in the art, other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all implementation methods here. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the claims of the present invention.
Claims
1. A time-frequency electromagnetic exploration method for improving exploration depth and resolution, characterized in that, Includes the following steps: The exploration area is delineated, and a multi-directional excitation source and a mobile remote terminal array observation system are deployed. The mobile remote terminal array observation system is equipped with a multi-component electric field sensor and a magnetic field sensor to simultaneously collect time-domain electromagnetic signals and frequency-domain electromagnetic signals within the exploration area. Based on compressed sensing theory, a sparse basis matrix and an incoherent observation matrix for time-frequency electromagnetic signals are constructed. A sparse sampling method with a sampling rate much lower than that of the Nyquist is used to obtain sparse sampling data in the time domain and frequency domain. The sparse sampled data is reconstructed using a norm minimization optimization algorithm to obtain complete time-domain and frequency-domain electromagnetic data, and noise suppression preprocessing is performed on the reconstructed data. A joint time-frequency domain inversion objective function is constructed, cross-gradient constraints are introduced to establish the physical relationship between time-domain data and frequency-domain data, and a deep reinforcement learning algorithm is used to construct an intelligent inversion model. The preprocessed time-frequency domain data is input into the model to carry out joint inversion. Multi-scale features of time-frequency dual-domain data are extracted using deep learning networks, feature fusion is achieved through an attention mechanism, and the inversion process is optimized by combining physical constraints of Maxwell's equations to obtain an initial geological model. Bayesian uncertainty analysis is performed on the initial geological model to quantify the probability distribution and uncertainty information of the inversion results and output a high-precision three-dimensional geological model. By integrating multi-source geophysical data from seismic, gravity, and magnetic methods, the three-dimensional geological model is validated and corrected, and the geological bodies in the exploration area are identified and interpreted.
2. The time-frequency electromagnetic exploration method according to claim 1, characterized in that, The steps for constructing the sparse basis matrix and incoherent observation matrix of the time-frequency electromagnetic signal based on compressed sensing theory specifically include: The sparse basis matrix is selected from one or more combinations of multi-scale transform bases, including wavelet basis, Fourier basis, and curvelet basis, which are used to realize sparse representation of time-frequency electromagnetic signals in the transform domain. The incoherent observation matrix is a random Gaussian matrix, and its finite isochronous property is verified to ensure signal reconstruction accuracy. During sparse sampling, the sampling location and observation attitude information are recorded simultaneously.
3. The time-frequency electromagnetic exploration method according to claim 1, characterized in that, The signal reconstruction and noise suppression preprocessing steps specifically include: The norm minimization optimization problem is solved by matching pursuit algorithm or basis pursuit algorithm. The optimization problem uses sparse sampled data as observations and constructs constraint relationships with observation matrix, sparse basis matrix and sparse coefficients. An adaptive threshold denoising algorithm is used to process the reconstructed time-frequency electromagnetic data to suppress random noise and power frequency interference.
4. The time-frequency electromagnetic exploration method according to claim 1, characterized in that, The intelligent inversion model constructed by the deep reinforcement learning algorithm specifically includes: The geological parameters are treated as cooperative agents by employing either the deep deterministic policy gradient algorithm or the multi-agent deep deterministic policy gradient algorithm. The inversion process is modeled as a Markov decision process. The agent learns the optimal inversion strategy through trial and error and outputs the optimal estimates of various geological parameters. The joint inversion objective function is a weighted sum of the time-domain data fitting error, the frequency-domain data fitting error, and the cross-gradient constraint term.
5. The time-frequency electromagnetic exploration method according to claim 1, characterized in that, The steps of the time-frequency dual-domain multi-scale feature fusion specifically include: Convolutional neural networks or Transformer networks are used to extract time-series features from time-domain data and spectral features from frequency-domain data, respectively. An attention mechanism is introduced to adaptively weight and fuse multi-scale features, enhancing feature components that are sensitive to both deep targets and shallow details; Embedding Maxwell's equations physical constraint modules into deep learning networks ensures that the fused features conform to the physical laws of electromagnetic propagation, thereby improving the physical interpretability of the inversion results.
6. The time-frequency electromagnetic exploration method according to claim 1, characterized in that, The working process of the mobile remote terminal array observation system specifically includes: Multiple mobile remote terminals coordinate according to a preset spatial array to form a ground-air joint observation system, and electromagnetic signals are synchronously excited by multi-directional ground excitation sources; The sensors mounted on the mobile remote terminal simultaneously measure the three components of the electric field and the three components of the magnetic field, enabling multi-component collaborative detection; The collected data is transmitted to the cloud acquisition platform in real time via a high-speed communication module, and the edge computing nodes simultaneously preprocess the transmitted data.
7. The time-frequency electromagnetic exploration method according to claim 1, characterized in that, The steps of the Bayesian uncertainty analysis specifically include: A large number of inversion samples were generated based on the Markov chain Monte Carlo algorithm, and the posterior probability distribution of various geological parameters was statistically analyzed. Calculate the mean, variance, and confidence interval of each geological parameter to quantify the uncertainty of the inversion results; The initial geological model is corrected by removing inversion samples with abnormal probability distributions.
8. The time-frequency electromagnetic exploration method according to claim 1, characterized in that, The steps for multi-source data fusion verification specifically include: Establish the correlation mapping relationship between seismic, gravity, and magnetic data and electromagnetic inversion models, and construct comprehensive geophysical constraints; A weighted fusion algorithm is used to fuse geological models corresponding to multi-source data, with the weights determined based on the resolution and reliability of each exploration method. The accuracy of the fusion model was evaluated using cross-validation.
9. The time-frequency electromagnetic exploration method according to claim 1, characterized in that, It also includes real-time monitoring and dynamic updating steps: During the exploration process, new time-frequency electromagnetic data are continuously collected, and the established three-dimensional geological model is dynamically updated using the newly collected data. The intelligent interpretation system automatically identifies anomalous geological bodies in the updated model, classifies and labels the anomaly types and confidence levels, and generates intelligent exploration reports.
10. A time-frequency electromagnetic exploration system for improving exploration depth and resolution, characterized in that, The time-frequency electromagnetic exploration method applied to any one of claims 1-9 includes: The excitation module includes multi-directional ground excitation sources for generating electromagnetic excitation signals of different frequencies and intensities. The observation module is a mobile remote array observation system, equipped with a multi-component electric field sensor, a magnetic field sensor and a high-speed communication unit, used for sparse acquisition of time-frequency electromagnetic signals and real-time transmission. The data processing module is built on an open-source distributed operating system to create a cloud-edge collaborative platform. Edge nodes are responsible for data preprocessing and signal reconstruction, while cloud nodes are responsible for deep inversion and feature fusion. The intelligent inversion module integrates deep reinforcement learning algorithms and physically constrained deep learning networks for joint time-frequency dual-domain inversion and uncertainty analysis. The interpretation and visualization module is used for multi-source data fusion verification, automatic anomaly identification, and output of 3D geological models and intelligent exploration reports.