Geological exploration method and system based on multi-source data fusion

Abstract

The invention relates to the technical field of geological physical exploration, discloses a geological exploration method and system based on multi-source data fusion, and aims to solve the problem that in the prior art, a single means is difficult to meet high-precision, high-efficiency and intelligent exploration requirements. The method comprises the following steps: synchronously acquiring remote sensing, geophysical, geochemical and drilling data through a multi-source data acquisition device; performing standardization processing on the heterogeneous data; geologic body features are extracted from the multi-source data and weighted fusion is carried out; constructing a physical constraint joint inversion model to optimize a geological parameter estimation value; a three-dimensional geologic structure model is generated and dynamically updated; and outputting an optimized exploration strategy in combination with the geological model and the exploration target. The system comprises a multi-source data acquisition module, a data standardization and space-time registration module, a feature fusion and joint inversion module, a dynamic three-dimensional geological modeling module and an intelligent decision support module. According to the technical scheme, effective fusion and intelligent analysis of multi-source data can be realized, and the precision and efficiency of geological exploration are remarkably improved.

Description

Technical Field

[0001] This invention belongs to the field of geophysical exploration technology, specifically relating to a geological exploration method and system based on multi-source data fusion. Background Technology

[0002] With the increasing complexity of geological exploration needs, traditional operational models relying on single methods such as drilling, geophysical exploration, or geochemical exploration are no longer sufficient to meet the requirements for high-precision, high-efficiency, and intelligent exploration. As a core component of resource exploration and environmental assessment, geological exploration is gradually shifting its technological evolution from isolated data acquisition to multi-dimensional information collaborative analysis. Currently, multi-source information such as remote sensing, geophysics, geochemistry, drilling data, and geographic information systems are widely recognized in theory as key elements for improving geological body identification capabilities and resource prediction levels; however, in practical applications, a highly efficient integration and intelligent analysis technology system has yet to be formed.

[0003] Among them, geological exploration methods based on multi-source data fusion aim to integrate heterogeneous information from different sensors and historical data through a unified framework to achieve comprehensive inference of underground structure and resource distribution. The core of this direction lies in building a multi-data processing mechanism that can accommodate spatiotemporal differences, semantic heterogeneity, and varying accuracy, and on this basis, carry out feature extraction, correlation modeling, and dynamic updates. However, existing technical solutions generally focus on optimizing local equipment functions or enhancing single data types, lacking support from a data fusion architecture that addresses the entire process and all elements.

[0004] While some existing technologies have made progress in hardware integration, such as improving sampling efficiency through integrated drilling devices or enhancing adaptability to field operations using mobile platforms, none have solved the systemic challenges in the acquisition, transmission, alignment, and interpretation of multi-source data. Specifically, this manifests as follows: various data sources are independent and in inconsistent formats, lacking standardized access and spatiotemporal registration mechanisms; data analysis remains primarily based on human experience, lacking automated feature fusion and joint inversion capabilities; and 3D geological modeling is mostly static, unable to optimize model parameters and exploration strategies in real time based on newly acquired data. These deficiencies make it difficult for existing systems to achieve the leap from "multi-source data convergence" to "intelligent geological cognition," facing bottlenecks such as low identification accuracy, slow response speed, and weak decision support in challenging scenarios such as complex tectonic zones or concealed ore body detection. A breakthrough is urgently needed in a geological exploration method and system with collaborative perception, intelligent fusion, and adaptive evolution capabilities. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a geological exploration method and system based on multi-source data fusion, which can effectively solve the problems in the background technology. To achieve the above objective, this invention provides the following technical solution: On the one hand, a geological exploration system based on multi-source data fusion, the system comprising the following components: a multi-source data acquisition module, used to simultaneously acquire heterogeneous data of the geological exploration area through remote sensing sensors, geophysical exploration equipment, geochemical analyzers, and drilling devices; a data standardization and spatiotemporal registration module, which performs format unification, coordinate transformation, and time series alignment processing on the acquired heterogeneous data; a feature fusion and joint inversion module, which extracts geological body features based on multi-source data and constructs a joint inversion model; a dynamic three-dimensional geological modeling module, which generates a three-dimensional geological structure model based on the fusion results and supports real-time parameter updates; and an intelligent decision support module, which optimizes the exploration strategy by combining the geological model and the exploration target output; the modules operate collaboratively through a data bus and computing unit.

[0006] Preferably, the multi-source data acquisition module includes a remote sensing data unit, a geophysical data unit, a geochemical data unit, and a drilling data unit; the remote sensing data unit acquires land cover information and topographic data through multispectral remote sensing satellites and UAV platforms; the geophysical data unit acquires underground physical field distribution data using gravimeters, magnetometers, and seismometers; the geochemical data unit obtains elemental concentration distribution through soil and rock sampling analysis; the drilling data unit acquires core samples and records stratigraphic structure information through automated drilling equipment; each unit is equipped with a data preprocessing interface for removing noisy data and labeling spatiotemporal attributes.

[0007] Furthermore, the data standardization and spatiotemporal registration module includes a data parsing submodule, a coordinate transformation submodule, and a time alignment submodule. The data parsing submodule calls the corresponding decoding protocol according to the data type to convert heterogeneous data into a unified structured format. The coordinate transformation submodule projects spatial data from different sources to the same coordinate system based on the geographic information system and eliminates spatial resolution differences through interpolation algorithms. The time alignment submodule uses a dynamic time warping algorithm to perform timestamp matching on asynchronously acquired data sequences to ensure the consistency of multi-source data in the time dimension.

[0008] Furthermore, the feature fusion and joint inversion module adopts a multi-level feature extraction architecture. The first level extracts landform features from remote sensing data through convolutional neural networks, extracts frequency domain features from geophysical data through wavelet transform, and extracts elemental correlation features from geochemical data through principal component analysis. The second level introduces an attention mechanism to perform weighted fusion of multi-source features, generating a joint feature tensor. The third level constructs a physically constrained inversion model based on the fused features, whose objective function is defined as minimizing the residual between the predicted and observed values ​​of the multi-source data, as shown in the following formula:

[0009] Where θ represents the geological parameter vector, Fi is the forward operator for the i-th type of data, di is the observed data, and λ is the regularization coefficient; the model iteratively optimizes the geological parameter estimates through the gradient descent algorithm.

[0010] Preferably, the dynamic 3D geological modeling module adopts implicit surface reconstruction technology to convert the fused geological parameters into a 3D voxel mesh; the module has a built-in geological rule library and automatically corrects the mesh geometry and topology according to lithological distribution and structural characteristics; when new exploration data is input, the module updates the voxel attributes through the Kalman filter algorithm and adjusts the geometric shape of the geological interface in real time; the modeling results support multi-resolution rendering, including stratigraphic isosurface extraction, lithological distribution visualization, and 3D display of structural faults.

[0011] Furthermore, the intelligent decision support module integrates an exploration target parser and a strategy optimizer. The target parser automatically identifies key geological indicators, including mineralization intensity, structural complexity, and resource burial depth, based on the exploration task input by the user. The strategy optimizer generates exploration schemes based on a multi-objective programming model, whose objective function simultaneously maximizes exploration accuracy and minimizes cost and time consumption. The optimization model uses a non-dominated sorting genetic algorithm to solve for the Pareto optimal solution set and outputs drilling point layout schemes, geophysical survey line planning, and sampling density suggestions.

[0012] On the other hand, a geological exploration method based on multi-source data fusion is proposed. The specific steps of this method are as follows: Step S110, simultaneously acquire remote sensing data, geophysical data, geochemical data, and drilling data of the geological exploration area using multi-source data acquisition equipment; Step S120, standardize the acquired heterogeneous data, including data format conversion, spatial coordinate unification, and time series alignment; Step S130, extract geological body features from the multi-source data and perform feature weighted fusion using an attention mechanism; Step S140, construct a physically constrained joint inversion model based on the fused features and iteratively optimize the estimated geological parameters; Step S150, generate a three-dimensional geological structure model based on the inversion results and dynamically update the model parameters; Step S160, combine the geological model with the exploration target and output an exploration strategy scheme through a multi-objective optimization algorithm.

[0013] Preferably, the data standardization process in step S120 includes: applying radiometric calibration and atmospheric correction to remote sensing data to eliminate sensor errors and environmental interference; performing field value normalization on geophysical data to eliminate background field influence; performing elemental concentration standardization on geochemical data to eliminate dimensional differences; and resampling all data to a uniform grid scale using a spatiotemporal interpolation algorithm, with the grid resolution set to 0.5 meters to 10 meters according to the exploration accuracy requirements.

[0014] Furthermore, in step S130, feature fusion employs a multi-scale convolutional neural network architecture; the network input is a multi-source data tensor, the first convolutional layer extracts local texture features, and the second convolutional layer expands the receptive field through dilated convolution to capture regional structural features; a gating mechanism is introduced in the feature weighting stage to dynamically adjust the fusion weights based on feature confidence, and the weight calculation formula is as follows:

[0015] Where wi is the feature weight of the i-th class, confi is the feature confidence score, and α is the scaling factor; the fused feature tensor is mapped to a geological semantic vector through a fully connected layer.

[0016] In addition, the joint inversion model in step S140 introduces multi-physics coupling constraints; the gravity inversion sub-model constructs the Poisson equation based on density distribution and gravity anomaly data; the magnetic inversion sub-model constructs the Laplace equation through magnetic susceptibility distribution and magnetic field data; each sub-model is coupled through cross gradient terms to ensure the consistency of the inversion results in spatial distribution; the inversion process is solved using the conjugate gradient method, and the iteration termination condition is that the residual decrease rate is less than 1e-6 or the number of iterations exceeds 500.

[0017] Preferably, in step S150, the three-dimensional geological modeling uses the moving cube algorithm to extract the geological interface; the model update mechanism is based on incremental learning, and when new drilling data is input, the lithology distribution probability is corrected by the Kriging interpolation algorithm, and the geometric shape of the geological interface is evolved in real time by the level set method; the model supports uncertainty quantification and generates confidence intervals for geological parameters through Monte Carlo simulation.

[0018] Furthermore, in step S160, the exploration strategy optimization introduces a multi-agent collaborative decision-making framework; each agent is responsible for a type of exploration task, including drilling point assessment, geophysical survey line planning, and sampling scheme generation; agents share geological confidence and cost constraints through message passing mechanism, and finally allocate exploration resources through auction algorithm to ensure the optimal global objective.

[0019] Compared with existing technologies, this invention has the following advantages: 1. By standardizing and registering multi-source data, it solves the problem of heterogeneous data access and alignment, improving data utilization efficiency and consistency; 2. By adopting multi-level feature fusion and physical constraint joint inversion, it significantly improves the accuracy of geological parameter estimation and inversion stability; 3. Dynamic 3D geological modeling supports real-time data-driven updates, overcoming the shortcomings of poor adaptability of static models; 4. The intelligent decision support system, through multi-objective optimization and multi-agent collaboration, realizes the automation and optimization of exploration strategies, greatly reducing the need for manual intervention and operating costs. Attached Figure Description

[0020] Figure 1This is a schematic diagram of the overall technical architecture of the geological exploration method and system based on multi-source data fusion proposed in this invention; Figure 2 This is a schematic diagram of the core principle framework of multi-level feature fusion and joint inversion in this invention; Figure 3 This is a schematic diagram of the multi-agent collaborative decision-making logic framework of the intelligent decision support module in this invention. Detailed Implementation

[0021] To further illustrate the technical means and effects adopted by the present invention to achieve the intended purpose, the specific embodiments according to the present invention will be described in detail below with reference to the accompanying drawings and preferred embodiments.

[0022] A geological exploration system based on multi-source data fusion includes the following components: a multi-source data acquisition module for simultaneously acquiring heterogeneous data of the geological exploration area through remote sensing sensors, geophysical exploration equipment, geochemical analyzers, and drilling equipment; a data standardization and spatiotemporal registration module for unifying the format, transforming coordinates, and aligning the time series of the acquired heterogeneous data; a feature fusion and joint inversion module for extracting geological body features based on multi-source data and constructing a joint inversion model; a dynamic 3D geological modeling module for generating a 3D geological structure model based on the fusion results and supporting real-time parameter updates; and an intelligent decision support module for optimizing exploration strategies by combining the geological model with exploration target outputs. All modules operate collaboratively through a data bus and computing unit.

[0023] The multi-source data acquisition module includes a remote sensing data unit, a geophysical data unit, a geochemical data unit, and a drilling data unit. The remote sensing data unit acquires land cover information and topographic data through multispectral remote sensing satellites and UAV platforms. The geophysical data unit uses gravimeters, magnetometers, and seismometers to collect data on the distribution of underground physical fields. The geochemical data unit obtains elemental concentration distribution through soil and rock sampling and analysis. The drilling data unit obtains core samples and records stratigraphic structure information through automated drilling equipment. Each unit is equipped with a data preprocessing interface to remove noisy data and label spatiotemporal attributes.

[0024] The data standardization and spatiotemporal registration module includes a data parsing submodule, a coordinate transformation submodule, and a time alignment submodule. The data parsing submodule calls the corresponding decoding protocol according to the data type to convert heterogeneous data into a unified structured format. The coordinate transformation submodule projects spatial data from different sources to the same coordinate system based on the geographic information system and eliminates spatial resolution differences through interpolation algorithms. The time alignment submodule uses a dynamic time warping algorithm to perform timestamp matching on asynchronously acquired data sequences to ensure the consistency of multi-source data in the time dimension.

[0025] Furthermore, the feature fusion and joint inversion module adopts a multi-level feature extraction architecture. The first level extracts landform features from remote sensing data through convolutional neural networks, extracts frequency domain features from geophysical data through wavelet transform, and extracts elemental correlation features from geochemical data through principal component analysis. The second level introduces an attention mechanism to perform weighted fusion of multi-source features, generating a joint feature tensor. The third level constructs a physically constrained inversion model based on the fused features, whose objective function is defined as minimizing the residual between the predicted and observed values ​​of the multi-source data, as shown in the following formula:

[0026] Where θ represents the geological parameter vector, Fi is the forward modeling operator for the i-th type of data, di is the observed data, and λ is the regularization coefficient. This model iteratively optimizes the geological parameter estimates using a gradient descent algorithm. For gravity data, Fi is a forward modeling process that converts the density distribution of the geological model into gravity anomalies based on the principle of superposition of cuboid elements (or more complex geometries). Its physical model can be expressed as [supplement with specific physical formulas or algorithm pseudocode]. For magnetic data, Fi is a forward modeling process that converts the magnetic susceptibility distribution of the geological model into magnetic anomalies based on a dipole model (or a more complex magnetization model). These different types of forward modeling operators, through a unified interface or data structure, take the fused geological parameter θ as input to generate corresponding multi-source data prediction values. In the objective function, the residuals between the multi-source data prediction value Fi(θ) and the observed value di are calculated using a weighted average (e.g., weighted according to the data signal-to-noise ratio or contribution) or a multi-task learning method.

[0027] The dynamic 3D geological modeling module employs implicit surface reconstruction technology to convert the fused geological parameters into a 3D voxel mesh. The module includes a built-in geological rule base, which uses an expert system architecture based on production rules. Rules are encoded in IF-THEN format, such as 'IF adjacent voxel lithological difference index > 0.8 AND no fault evidence THEN insert fault interface'. The forward inference engine continuously checks whether the geological model state triggers rule conditions through a pattern matching mechanism. When multiple rules are simultaneously satisfied, a preset priority mechanism (e.g., geometric constraints take precedence over lithological constraints) or a weighted voting mechanism is used to resolve rule conflicts. After the rules are executed, the voxel mesh is corrected using geometric topology optimization algorithms (such as Delaunay triangulation reconstruction or mesh smoothing), and the consistency of the correction results is verified (such as Euler characteristic number check). The mesh geometry and topology are automatically corrected according to lithological distribution and structural characteristics. The geological rule base contains multiple executable rules, such as: Rule 1 - Lithological abrupt change constraint (forced insertion of fault interfaces when the lithological difference index between adjacent voxels is >0.8); Rule 2 - Stratigraphic attitude constraint (marked as steeply dipping structures when the interface dip angle is >60°). Rule execution uses a forward inference engine, triggering topology correction by traversing the voxel neighborhood relationships. When new exploration data is input, the module updates voxel attributes using a Kalman filter algorithm and adjusts the geological interface geometry in real time. The modeling results support multi-resolution rendering, including stratigraphic isosurface extraction, lithological distribution visualization, and three-dimensional display of structural faults.

[0028] Furthermore, the intelligent decision support module integrates an exploration target parser and a strategy optimizer. The target parser automatically identifies key geological indicators, including mineralization intensity, structural complexity, and resource burial depth, based on the user-input exploration task. The strategy optimizer generates exploration schemes based on a multi-objective programming model, whose objective function simultaneously maximizes exploration accuracy and minimizes cost and time consumption. The optimization model uses a non-dominated sorting genetic algorithm (NSGA-II) to solve for the Pareto optimal solution set. The population size is set to 200, the crossover probability is 0.85, and the mutation probability is 0.01. Geological parameters (such as mineralization intensity) are normalized and mapped to the [0,1] interval before being input into the algorithm. The objective function weights are preset according to the exploration task type as follows: accuracy weight 0.6, cost weight 0.3, and time weight 0.1. The output includes a drilling point layout scheme, geophysical survey line planning, and sampling density suggestions. The exploration accuracy objective function is defined as: 1 - (The mean squared error of the deviation between predicted geological parameters and actual validation data), where the deviation is calculated through cross-validation or historical data simulation; the cost objective function is defined as: Σ(number of drilled wells * cost per well) + Σ(length of geophysical survey line * cost per unit length) + Σ(number of sampling points * cost per unit sampling); the time consumption objective function is defined as: data acquisition cycle + data processing cycle + decision generation cycle. In the multi-agent collaborative decision-making framework, each agent (such as drilling agent, geophysical agent, and sampling agent) communicates through a defined RESTful API or message queue protocol. The shared geological confidence can be represented as the probability distribution or entropy value of geological parameters, and the cost constraint is the budget ceiling of each agent. The auction algorithm allocates exploration resources through an iterative bidding mechanism (e.g., agents submit resource requirements and expected returns to the central coordinating agent based on their own utility functions, and the central coordinating agent allocates resources and adjusts prices according to the global objective) to ensure that the global objective (such as maximizing overall returns) is optimal.

[0029] On the other hand, a geological exploration method based on multi-source data fusion includes the following steps: Step S110, simultaneously acquiring remote sensing data, geophysical data, geochemical data, and drilling data of the geological exploration area through multi-source data acquisition equipment; Step S120, standardizing the acquired heterogeneous data, including data format conversion, spatial coordinate unification, and time series alignment; Step S130, extracting geological body features from the multi-source data and performing feature weighted fusion through an attention mechanism; Step S140, constructing a physically constrained joint inversion model based on the fused features and iteratively optimizing the geological parameter estimates; Step S150, generating a three-dimensional geological structure model based on the inversion results and dynamically updating the model parameters; and Step S160, combining the geological model with the exploration target and outputting an exploration strategy scheme through a multi-objective optimization algorithm.

[0030] Preferably, the data standardization process in step S120 includes: applying radiometric calibration and atmospheric correction to remote sensing data to eliminate sensor errors and environmental interference; performing field value normalization on geophysical data to eliminate background field influence; performing elemental concentration standardization on geochemical data to eliminate dimensional differences; and resampling all data to a uniform grid scale using a spatiotemporal interpolation algorithm, with the grid resolution set to 0.5 meters to 10 meters according to the exploration accuracy requirements.

[0031] Furthermore, in step S130, feature fusion employs a multi-scale convolutional neural network architecture; the network input is a multi-source data tensor, the first convolutional layer extracts local texture features, and the second convolutional layer expands the receptive field through dilated convolution to capture regional structural features; a gating mechanism is introduced in the feature weighting stage to dynamically adjust the fusion weights based on feature confidence, and the weight calculation formula is as follows:

[0032] Where wi is the feature weight of the i-th class, confi is the feature confidence score, and α is the scaling factor; the fused feature tensor is mapped to a geological semantic vector through a fully connected layer, and the feature confidence score confi is calculated through cross-validation: the features from each data source are input into the preset geological classification model, and the F1-score of the model on the validation set is used as the confi value; the scaling factor α is taken as an empirical value of 0.75.

[0033] In addition, the joint inversion model in step S140 introduces multi-physics coupling constraints; the gravity inversion sub-model constructs the Poisson equation based on density distribution and gravity anomaly data; the magnetic inversion sub-model constructs the Laplace equation through magnetic susceptibility distribution and magnetic field data; each sub-model is coupled through cross gradient terms to ensure the consistency of the inversion results in spatial distribution; the inversion process is solved using the conjugate gradient method, and the iteration termination condition is that the residual decrease rate is less than 1e-6 or the number of iterations exceeds 500.

[0034] Preferably, in step S150, the three-dimensional geological modeling uses the moving cube algorithm to extract the geological interface; the model update mechanism is based on incremental learning, and when new drilling data is input, the lithology distribution probability is corrected by the Kriging interpolation algorithm, and the geometric shape of the geological interface is evolved in real time by the level set method; the model supports uncertainty quantification and generates confidence intervals for geological parameters through Monte Carlo simulation.

[0035] Furthermore, in step S160, the exploration strategy optimization introduces a multi-agent collaborative decision-making framework; each agent is responsible for a type of exploration task, including drilling point assessment, geophysical survey line planning, and sampling scheme generation; agents share geological confidence and cost constraints through message passing mechanism, and finally allocate exploration resources through auction algorithm to ensure the optimal global objective.

[0036] In metal mineral resource exploration areas, multi-source data acquisition modules synchronously acquire geological data through multi-platform detection equipment deployed in the exploration area. (See also...) Figure 1 The remote sensing data unit acquired land cover information, including vegetation distribution density (0.85), surface humidity index (0.32), and topographic elevation data, through multispectral remote sensing satellites and UAV platforms, with a sampling frequency of 256 points per square kilometer. The geophysical data unit used a high-precision gravimeter to measure Bouguer gravity anomalies with a measurement accuracy of 0.1 milligal, and a proton magnetometer to collect geomagnetic field strength data with a sensitivity of 0.1 nanotesla. It also used a distributed seismograph to record reflected wave signals generated by artificial seismic sources, with a sampling rate of 2000 Hz. The geochemical data unit acquired soil samples through systematic grid sampling, with each sampling point covering an area of ​​100 square meters. X-ray fluorescence spectrometry was used to analyze the concentration of 35 elements, with a detection limit of 1 part per million. The drilling data unit used a fully hydraulic core drilling rig to obtain continuous core samples at depths up to 500 meters, while recording drilling rate, cuttings characteristics, and groundwater outflow parameters. Each data unit is equipped with a dedicated preprocessing interface to perform tidal correction on gravity data, diurnal variation correction on magnetic data, and background value removal on geochemical data, ensuring that the quality of the raw data meets the requirements of subsequent processing.

[0037] The data standardization and spatiotemporal registration module implements strict consistency processing for the acquired heterogeneous data. The data parsing submodule calls the appropriate decoding protocol based on the data type: remote sensing data is parsed in GeoTIFF format to extract pixel values ​​and georeferenced information; geophysical data is parsed in SEG-Y format to extract trace information and sampling sequences; geochemical data is parsed in CSV format to extract elemental concentration matrices; and drilling data is parsed using the WITSML standard to extract lithological descriptions and well logging curves. The coordinate transformation submodule uniformly transforms multi-source spatial data to the UTM coordinate system and uses a bilinear interpolation algorithm to resample data of different resolutions to a unified grid, with the grid size set to 2 meters × 2 meters according to the exploration stage. The time alignment submodule employs the Dynamic Time Warping (DTW) algorithm. First, it performs Z-score standardization on the asynchronous data sequence to eliminate dimensional differences. Second, it calculates the minimum cumulative distance path by setting path constraints (e.g., the Sakoe-Chiba Band window width is 20% of the sampling period) to generate a time offset mapping table. Finally, it performs cubic spline interpolation resampling on the low-frequency data sequence based on the mapping table, achieving a time synchronization accuracy ≤0.1 seconds. The time alignment submodule uses the DTW algorithm to match asynchronously acquired data sequences, establishing a unified time axis, and controlling the time synchronization error accuracy within 0.1 seconds. The DTW algorithm directly resamples high-frequency data (e.g., seismic data, sampling rate 2000Hz) and completes the time series of low-frequency data (e.g., drilling data, sampling interval ≥10 seconds) through cubic spline interpolation, ensuring that the time synchronization error of all data sources is ≤0.1 seconds.

[0038] The feature fusion and joint inversion module implements multi-level feature extraction and fusion processing. In the first-level feature extraction, a convolutional neural network performs convolution operations on multispectral remote sensing data with a kernel size of 3×3 and a stride of 1, extracting linear surface structures and ring anomaly features; wavelet transform performs a 5-level decomposition on gravity data, extracting detail coefficients to represent local density changes; principal component analysis performs covariance matrix eigenvalue decomposition on geochemical data, retaining principal component features with a cumulative contribution rate of 95%. The second-level feature fusion introduces a multi-head attention mechanism, setting 8 attention heads to perform weighted fusion of multi-source features. Feature confidence scores are calculated through cross-validation, and the scaling factor is set to 0.75. The third-level joint inversion constructs a physically constrained inversion model. The forward modeling operators include gravity forward modeling based on the superposition of cuboid elements and magnetic forward modeling based on the dipole model. The inversion parameters include density contrast, magnetic susceptibility, and elastic parameters. The regularization coefficient is determined to be 0.01 by the L-curve method. The quasi-Newton method is used for iterative optimization, with the iteration step size adaptively adjusted and the residual tolerance set to 1e-6.

[0039] The dynamic 3D geological modeling module constructs a geological structure model based on fused inversion results. The moving cube algorithm discretizes the exploration area into a voxel grid with voxel sizes of 1m × 1m × 1m, and stratigraphic interfaces are generated through isosurface extraction. The geological rule base contains 16 geological constraint rules, including the sequence law of sedimentary stratigraphy and fault cutting relationships, automatically correcting grid topology consistency. When new drilling data is input, the sequential Gaussian simulation method is used to update the lithology probability distribution, and the geological interface evolves using the level set method. The interface evolution rate function is calculated based on lithological difference, with a convergence threshold set to 0.001. The model supports multi-resolution visualization, including stratigraphic isosurface extraction with a 5-meter spacing, lithology distribution using RGB color mapping, and structural faults displayed with transparent rendering showing fault displacement and dip.

[0040] The intelligent decision support module generates optimized solutions based on exploration objectives.

[0041] See Figure 3 The target parser analyzes the user-input mineral exploration task, identifying key geological indicators including a mineralization intensity threshold of 0.15%, a structural complexity index of 0.68, and a resource burial depth range of 50-300 meters. The strategy optimizer constructs a multi-objective programming model, with the objective function simultaneously maximizing exploration accuracy and minimizing cost and time consumption. Constraints include a budget limit of 1 million yuan and a project duration of 90 days. A non-dominated sorting genetic algorithm is used to solve for the Pareto optimal solution set. The population size is set to 200, with a crossover probability of 0.85 and a mutation probability of 0.01. After 500 generations, the optimal exploration plan is output, including 32 suggested drilling points, 8 geophysical survey lines, and a sampling density of 50 points per square kilometer.

[0042] In oil and gas field exploration and development blocks, the multi-source data acquisition module adjusts the data acquisition scheme to meet the needs of reservoir characteristic description. The remote sensing data unit acquires surface hydrocarbon micro-leakage anomalies through hyperspectral remote sensing, with a spectral resolution of 10 nanometers; the geophysical data unit uses a 3D seismic acquisition system to acquire pre-stack seismic data, with a cell size of 25m × 25m and 64 coverage times; the geochemical data unit obtains C1-C5 hydrocarbon concentrations through soil adsorbed hydrocarbon analysis; and the drilling data unit acquires resistivity, sonic transit time, and natural gamma curves through logging while drilling. In the data standardization process, the seismic data undergoes amplitude recovery and deconvolution processing, while the geochemical data undergoes background value standardization.

[0043] The feature fusion stage employs a multi-scale convolutional neural network architecture. The first convolutional layer uses a 3×3 convolutional kernel to extract local features, while the second convolutional layer uses dilated convolution with a dilation rate of 2 to extract large-scale constructed features. Feature weighted fusion utilizes a gating mechanism, with feature confidence calculated using the reciprocal of the prediction error and a scaling factor of 1.2. The joint inversion model incorporates coupling between elastic wave impedance inversion and electromagnetic inversion, with a cross-gradient weight of 0.5. The conjugate gradient method is used for solution, converging after 200 iterations.

[0044] The 3D geological modeling employs a stochastic modeling method based on geostatistics, using variogram analysis to determine spatial correlation. The primary range is set to 350 meters, and the secondary range to 280 meters. Model updates utilize ensemble Kalman filtering, updating reservoir parameter distributions after assimilating new drilling data. The intelligent decision-making module addresses well location optimization by employing a mixed-integer programming model, considering drilling costs, expected production, and pipeline constraints, to output the optimal well location layout.

Claims

1. A geological exploration system based on multi-source data fusion, characterized in that, The system includes the following components: a multi-source data acquisition module, used to simultaneously acquire heterogeneous data of the geological exploration area through remote sensing sensors, geophysical exploration equipment, geochemical analyzers and drilling equipment; The data standardization and spatiotemporal registration module performs format unification, coordinate transformation, and time series alignment processing on the collected heterogeneous data; the feature fusion and joint inversion module extracts geological body features based on multi-source data and constructs a joint inversion model; the dynamic 3D geological modeling module generates a 3D geological structure model based on the fusion results and supports real-time parameter updates; and the intelligent decision support module combines the geological model with the exploration target output to optimize the exploration strategy. The multi-source data acquisition module, data standardization and spatiotemporal registration module, feature fusion and joint inversion module, dynamic 3D geological modeling module, and intelligent decision support module are all connected to the central computing unit through a data bus and operate collaboratively under the unified scheduling of the central computing unit.

2. The geological exploration system based on multi-source data fusion according to claim 1, characterized in that, The multi-source data acquisition module includes a remote sensing data unit, a geophysical data unit, a geochemical data unit, and a drilling data unit. The remote sensing data unit acquires land cover information and topographic data through multispectral remote sensing satellites and UAV platforms. The geophysical data unit uses gravimeters, magnetometers, and seismometers to collect underground physical field distribution data. The geochemical data unit obtains elemental concentration distribution through soil and rock sampling and analysis. The drilling data unit obtains core samples and records stratigraphic structure information through automated drilling equipment. Each unit is equipped with a data preprocessing interface for removing noisy data and labeling spatiotemporal attributes.

3. The geological exploration system based on multi-source data fusion according to claim 1, characterized in that, The data standardization and spatiotemporal registration module includes a data parsing submodule, a coordinate transformation submodule, and a time alignment submodule; the data parsing submodule calls the corresponding decoding protocol according to the data type to convert heterogeneous data into a unified structured format; The coordinate transformation submodule projects spatial data from different sources to the same coordinate system based on the geographic information system and eliminates spatial resolution differences through interpolation algorithms; the time alignment submodule uses a dynamic time warping algorithm to match the timestamps of asynchronously acquired data sequences to ensure the consistency of multi-source data in the time dimension.

4. The geological exploration system based on multi-source data fusion according to claim 1, characterized in that, The feature fusion and joint inversion module adopts a multi-level feature extraction architecture; The first layer extracts landform features from remote sensing data through convolutional neural networks, extracts frequency domain features from geophysical data through wavelet transform, and extracts elemental correlation features from geochemical data through principal component analysis. The second level introduces an attention mechanism to weightedly fuse multi-source features and generate a joint feature tensor; The third level constructs a physically constrained inversion model based on fused features. Its objective function is defined as minimizing the residual between the predicted and observed values ​​from the multi-source data. The specific formula is as follows: , where θ represents the geological parameter vector, Fi is the forward operator for the i-th type of data, di is the observed data, and λ is the regularization coefficient; the model iteratively optimizes the geological parameter estimates using the gradient descent algorithm.

5. The geological exploration system based on multi-source data fusion according to claim 1, characterized in that, The dynamic 3D geological modeling module uses implicit surface reconstruction technology to convert the fused geological parameters into a 3D voxel mesh. The module has a built-in geological rule library that automatically corrects the mesh geometry and topology based on lithological distribution and structural characteristics. When new exploration data is input, the module updates the voxel properties through the Kalman filter algorithm and adjusts the geological interface geometry in real time. The modeling results support multi-resolution rendering, including stratigraphic isosurface extraction, lithological distribution visualization, and three-dimensional display of structural faults.

6. The geological exploration system based on multi-source data fusion according to claim 1, characterized in that, The intelligent decision support module integrates an exploration target parser and a strategy optimizer. The target parser automatically identifies key geological indicators, including mineralization intensity, structural complexity, and resource burial depth, based on the exploration task input by the user. The strategy optimizer generates exploration schemes based on a multi-objective programming model, whose objective function simultaneously maximizes exploration accuracy and minimizes cost and time consumption. The optimization model uses a non-dominated sorting genetic algorithm to solve for the Pareto optimal solution set and outputs drilling point layout schemes, geophysical survey line planning, and sampling density suggestions.

7. A geological exploration method based on multi-source data fusion, characterized in that, The method includes the following steps: S110 synchronously acquires remote sensing data, geophysical data, geochemical data, and drilling data of the geological exploration area through multi-source data acquisition equipment; S120 standardizes the collected heterogeneous data, including data format conversion, spatial coordinate unification, and time series alignment. S130 extracts geological body features from multi-source data and performs feature weighted fusion through an attention mechanism; S140, a physical constraint joint inversion model is constructed based on fusion features, and the geological parameter estimates are iteratively optimized; S150 generates a three-dimensional geological structure model based on the inversion results and dynamically updates the model parameters; S160 combines geological models and exploration objectives to output exploration strategy schemes through multi-objective optimization algorithms.

8. The geological exploration method based on multi-source data fusion according to claim 7, characterized in that, In step S130, feature fusion adopts a multi-scale convolutional neural network architecture; the network input is a multi-source data tensor, the first convolutional layer extracts local texture features, and the second convolutional layer expands the receptive field through dilated convolution to capture regional structural features. A gating mechanism is introduced in the feature weighting stage to dynamically adjust the fusion weights based on feature confidence. The weight calculation formula is as follows: Where wi is the feature weight of the i-th class, confi is the feature confidence score, and α is the scaling factor; The fused feature tensor is mapped to a geological semantic vector through a fully connected layer.

Images