Geophysical prospecting earth deep detection system and method based on artificial intelligence

Abstract

The invention discloses a geophysical exploration earth deep detection system and method based on artificial intelligence, and the system comprises a multi-source data collection subsystem, an intelligent preprocessing subsystem, an AI inversion decision subsystem, a dynamic interpretation subsystem and a visual output subsystem, and all subsystems achieve the data interaction and cooperative work through a high-speed communication network. According to the method, a full-process intelligent framework of collection-preprocessing-inversion-interpretation-visualization is constructed, a distributed geophysical prospecting sensor array, edge computing nodes, a multi-model AI inversion engine and a three-dimensional visualization terminal are integrated, seismic wave, electromagnetic, gravity and magnetic force multi-modal data are synchronously collected, enhanced preprocessing and multi-target optimization inversion are carried out, and the multi-modal data of the seismic wave, the electromagnetic, the gravity and the magnetic force are obtained. High-precision extraction of deep geological parameters is achieved, the technical problems that a traditional method is low in data processing efficiency and high in inversion multiplicity of solutions are solved, and the detection precision is improved.

Description

Technical Field

[0001] This invention relates to the field of geophysical exploration of the deep earth, specifically to a geophysical exploration system and method based on artificial intelligence. Background Technology

[0002] Deep Earth exploration is the core foundation for mineral resource exploration, geological disaster early warning, and underground space development. However, current traditional exploration technologies still face many bottlenecks under complex geological conditions.

[0003] On the one hand, traditional exploration relies on a single geophysical method (such as seismic exploration or electrical exploration), which makes it difficult to fully capture the diverse characteristics of deep geological bodies. Moreover, the lack of a precise synchronization mechanism for multi-source data acquisition leads to large time-series errors, resulting in insufficient data integrity and consistency. This, coupled with problems such as deep signal attenuation and noise interference, further reduces the quality of the original data.

[0004] On the other hand, the data preprocessing stage often adopts traditional noise reduction and outlier removal methods, which are insufficient for feature mining of multimodal data. Moreover, the different dimensions of different types of data are very different, making it difficult to achieve effective fusion and causing interference to subsequent inversion calculations. At the same time, traditional inversion models are mostly single fixed architectures, lacking the ability to adapt to different geological scenarios and unable to continuously absorb new exploration data to optimize themselves. They are prone to catastrophic forgetting, resulting in strong multiple solutions and low accuracy in inversion, making it difficult to accurately obtain core geological parameters such as stratum thickness and rock density.

[0005] Furthermore, the geological interpretation process relies heavily on human experience, the analysis of inversion results is highly subjective, and there is a lack of dynamic correction mechanisms. It is difficult to quickly integrate real-time exploration data with new geological data, resulting in insufficient accuracy and delayed updates in interpretation conclusions. At the same time, exploration results are mostly presented in two-dimensional reports or simple models, with low visualization, data formats that are incompatible with mainstream exploration software, and low sharing efficiency, which restricts the promotion and application of the results.

[0006] Furthermore, traditional systems lack a robust fault diagnosis and adaptive adjustment mechanism. Problems such as sensor failure and data transmission interruption can easily lead to interruptions in the detection process, further affecting detection efficiency and stability.

[0007] These problems collectively lead to the shortcomings of traditional deep Earth exploration, such as low accuracy, poor efficiency, weak adaptability to different scenarios, and insufficient practicality of results. They are difficult to meet the current demand for high-precision and intelligent exploration in fields such as deep mineral exploration and geological disaster early warning. There is an urgent need for an innovative exploration technology solution that integrates multi-source data fusion, intelligent processing, dynamic interpretation, and stable operation. Summary of the Invention

[0008] To address the problems mentioned in the background art, the present invention aims to provide an artificial intelligence-based geophysical exploration system and method for deep earth exploration, which has the advantages of multi-source data fusion, intelligent processing, dynamic interpretation and stable operation, and solves the problems of insufficient data integrity, poor preprocessing effect, low inversion accuracy, reliance on human experience for interpretation, weak practicality of results and insufficient system stability of traditional exploration methods.

[0009] To achieve the above objectives, the present invention provides the following technical solution: an artificial intelligence-based geophysical deep earth exploration system, comprising a multi-source data acquisition subsystem, an intelligent preprocessing subsystem, an AI inversion decision subsystem, a dynamic interpretation subsystem, and a visualization output subsystem, wherein each subsystem achieves data interaction and collaborative work through a high-speed communication network; The multi-source data acquisition subsystem consists of a distributed geophysical sensor array, a data transmission module, and a synchronization control unit. The geophysical sensor array includes seismic wave sensors, electromagnetic sensors, gravity sensors, and magnetic sensors, which are used to synchronously acquire multimodal geophysical data of the deep earth. The synchronization control unit realizes the timing synchronization of the acquisition of each sensor, and the data transmission module transmits the raw data through a 5G / fiber optic network. The intelligent preprocessing subsystem is deployed on edge computing nodes and includes a data cleaning module, an outlier removal module, a multimodal feature enhancement module, and a data standardization module. It is used to enhance and preprocess the raw geophysical data and output highly reliable structured data. The AI ​​inversion decision subsystem includes a pre-trained inversion model library, an incremental learning module, and a multi-objective optimization module. The pre-trained inversion model library includes seismic wave inversion models, electromagnetic parameter inversion models, and gravity-magnetic joint inversion models based on deep learning. The incremental learning module continuously absorbs new detection data to optimize model parameters. The multi-objective optimization module generates the optimal inversion result with the goal of maximizing detection accuracy and minimizing data fitting error. The dynamic interpretation subsystem includes a geological body identification module, a parameter correlation analysis module, and a dynamic correction module. Based on the inversion results and known geological data, it intelligently identifies deep geological structures, mineral distribution, and potential hazards, and dynamically corrects the interpretation conclusions. The visualization output subsystem includes a 3D modeling module, a results display module, and a data export module. It transforms the inversion results and interpretation conclusions into a 3D geological model, supporting multi-dimensional visualization and standardized data export.

[0010] As a preferred embodiment of the present invention, the fault diagnosis and adaptive adjustment module monitors the sensor working status, data transmission link and model operation in real time. When a sensor failure occurs, it automatically switches to a backup sensor; when data transmission is interrupted, it starts the local cache of the edge node; when the model inversion accuracy decreases, it triggers the incremental learning process to ensure the stable operation of the system.

[0011] Correspondingly, the present invention also provides a geophysical exploration method for deep Earth exploration based on artificial intelligence, the steps of which are as follows: S1: Synchronous acquisition of multi-source geophysical data: Seismic wave data, electromagnetic data, gravity data, and magnetic data are acquired synchronously through a distributed geophysical sensor array. The acquisition timing is calibrated by the synchronous control unit and uploaded to the intelligent preprocessing subsystem through the data transmission module. S2: Multimodal data augmentation and preprocessing: The original geophysical data is cleaned, outlier removed, feature enhanced and standardized to obtain structured data; S3: AI Model Training and Optimization: Based on historical geophysical data and known geological information, the deep learning model in the inversion model library is initially trained. The model parameters are optimized by absorbing new scene data through the incremental learning module to ensure that the model is adapted to the geological conditions of different exploration areas. S4: Intelligent Inversion Calculation: Input the preprocessed structured data into the trained AI inversion model, and combine it with a multi-objective optimization algorithm to invert and obtain the geological parameters of the deep earth. S5: Dynamic Geological Interpretation: Based on the geological parameters obtained from inversion and combined with regional geological background data, it intelligently identifies the types of deep geological bodies, the distribution of mineral resources and potential geological hazards, and dynamically corrects the interpretation results. S6: 3D visualization output: Construct a 3D geological model from the inversion results and interpretation conclusions, and display it in multiple dimensions through a visualization terminal, supporting data export and result sharing.

[0012] As a preferred embodiment of the present invention, S1-1: Seismic wave data acquisition: acquiring P-wave and S-wave propagation signals through a detector array, with the sampling frequency set to 250-1000Hz, and recording parameters such as signal amplitude, phase, and propagation time; S1-2: Electromagnetic data acquisition: Using a combination of controlled source electromagnetic method and natural field electromagnetic method, electromagnetic signals with a frequency range of 1Hz-10kHz are acquired, and electric field strength, magnetic field strength and impedance parameters are recorded. S1-3: Gravity data acquisition: Collect abnormal gravity acceleration data using a gravimeter, with a sampling interval of 1-5 seconds, and simultaneously record the latitude, longitude, and elevation information of the acquisition point; S1-4: Magnetic Data Acquisition: Collect total intensity and component anomaly data of the geomagnetic field using a magnetometer with a resolution ≤0.1nT to ensure sensitivity to weak magnetic anomalies and simultaneously calibrate instrument errors, thereby eliminating instrument drift and environmental interference; S1-5: Synchronization control: GPS timing and clock synchronization technology are used to ensure that the timing error of each sensor acquisition is ≤1ms, and data transmission adopts an encryption protocol to ensure data security.

[0013] As a preferred embodiment of the present invention, S2-1: Data cleaning: Random noise in the data is removed by using the sliding window method, and invalid data is deleted by data consistency verification; S2-2: Outlier Removal: An improved isolated forest algorithm combined with statistical tests (Grubbs test) is used to identify and remove systematic outliers caused by geological anomalies and random outliers caused by instrument errors. S2-3: Multimodal feature enhancement: Extract time-domain and frequency-domain features from seismic wave data, extract impedance and polarization features from electromagnetic data, extract gradient and anomaly boundary features from gravity and magnetic data, and construct a multi-dimensional feature set; S2-4: Data Standardization: The Z-score normalization algorithm is used to map geophysical data of different dimensions to the same distribution range, eliminating the difference in dimensions and obtaining structured data.

[0014] As a preferred embodiment of the present invention, S3-1: Model construction: The seismic wave inversion model adopts the U-Net improved network, the electromagnetic parameter inversion model adopts the deep belief network, and the gravity and magnetic joint inversion model adopts the multi-task convolutional neural network. S3-2: Initial Training: Using historical geophysical data and geological parameter labels, train the model until the loss function converges; S3-3: Incremental optimization: The Elastic Weight Consolidation (EWC) algorithm is used to continuously absorb data from new exploration areas and validation results, fine-tune model parameters, and avoid catastrophic forgetting. S3-4: Model Evaluation: The accuracy of the model inversion is evaluated by the root mean square error (RMSE) and the correlation coefficient (R²), ensuring that RMSE ≤ 5% and R² ≥ 0.9.

[0015] As a preferred embodiment of the present invention, S4-1: Data input: The structured data is input into the corresponding inversion model according to the modality classification, and the batch processing mechanism is used to improve the computational efficiency; S4-2: Multi-objective optimization: Constructing the optimization function ,in , , Weighting coefficients (satisfying) + + =1), T is the inversion time. This is the maximum allowable time. S4-3: Inversion Calculation: Accelerates the inversion process through GPU parallel computing, and outputs core geological parameters such as stratum thickness, rock density, resistivity, and wave velocity; S4-4: Result Verification: Cross-validate the inversion results by combining known borehole data and regional geological maps, and eliminate invalid inversion results.

[0016] As a preferred embodiment of the present invention, S5-1: Geological body identification: Based on inversion parameters, rock types such as sedimentary rocks, igneous rocks, and metamorphic rocks are identified through a pre-trained geological body classification model, and the distribution range of ore bodies is delineated; S5-2: Parameter Correlation Analysis: Bayesian network analysis is used to analyze the correlation between geological parameters and mineral resources and disaster risks, and to quantify the probability of resource reserves and the risk level of disaster occurrence. S5-3: Dynamic Correction: By combining real-time detection data with newly acquired geological data, the parameters of the interpretation model are dynamically adjusted and the geological interpretation conclusions are updated. S5-4: Result Verification: Verify the interpretation results through methods such as field geological surveys and borehole verification to ensure that the interpretation accuracy rate is ≥85%.

[0017] As a preferred embodiment of the present invention, S6-1: Three-dimensional modeling: VTK is used to construct a three-dimensional geological model to restore the spatial distribution characteristics of deep geological structures; S6-2: Visualization: Supports slice viewing, transparency adjustment, and multi-view switching to display the distribution of geological parameters, geological body boundaries, resource distribution range, and disaster-prone areas; S6-3: Data Export: Supports exporting standardized data formats, compatible with mainstream geological exploration software; S6-4: Results Sharing: Accessibility sharing of exploration results is achieved through a cloud platform, supporting access and viewing from multiple terminals.

[0018] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention constructs a fully intelligent architecture encompassing "acquisition-preprocessing-inversion-interpretation-visualization," integrating a distributed geophysical sensor array, edge computing nodes, a multi-model AI inversion engine, and a 3D visualization terminal. It simultaneously acquires multi-modal data including seismic waves, electromagnetic, gravity, and magnetic forces. Through enhanced preprocessing and multi-objective optimized inversion, it achieves high-precision extraction of deep geological parameters, solving the technical pain points of low data processing efficiency and strong inversion ambiguity in traditional methods, and improving detection accuracy.

[0019] 2. This invention continuously adapts to new geological conditions through an incremental learning module (EWC algorithm), and the dynamic interpretation subsystem combines AI intelligent recognition with field verification to correct interpretation conclusions. The fault adaptive adjustment module ensures stable system operation, while supporting standardized data export and cloud sharing. It eliminates the reliance on human experience in traditional exploration, significantly improves the intelligence level, scenario adaptability, and practicality of results, and can widely meet the needs of multiple scenarios such as mineral exploration and disaster early warning. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the system architecture of the present invention; Figure 2This is a schematic diagram of the method flow of the present invention; Figure 3 This is a schematic diagram of the multi-source data acquisition subsystem structure of the present invention; Figure 4 This is a schematic diagram of the AI ​​inversion model structure of the present invention; Figure 5 This is a schematic diagram illustrating the three-dimensional visualization results of the present invention; Figure 6 This is a block diagram illustrating the combined use of the multi-source data acquisition subsystem and the fault diagnosis and adaptive adjustment module of the present invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] like Figures 1 to 6 As shown, the present invention provides an artificial intelligence-based geophysical deep earth exploration system, including a multi-source data acquisition subsystem, an intelligent preprocessing subsystem, an AI inversion decision subsystem, a dynamic interpretation subsystem, and a visualization output subsystem. Each subsystem achieves data interaction and collaborative work through a high-speed communication network. The multi-source data acquisition subsystem consists of a distributed geophysical sensor array, a data transmission module, and a synchronization control unit. Its core function is to synchronously acquire multimodal geophysical data. The geophysical sensor array includes seismic wave sensors (detectors), electromagnetic sensors, gravity sensors, and magnetic sensors. Deployed according to a preset detection grid, it covers the target detection area and realizes the synchronous acquisition of seismic wave, electromagnetic, gravity, and magnetic data. It can also unify and standardize the output data format of each sensor and preset data fields (such as acquisition timestamp, sensor number, parameter type, numerical unit, etc.), avoiding the cumbersome format conversion problem during subsequent data integration. Synchronization control unit: Based on GPS timing technology and high-precision clock synchronization algorithm, it ensures that the acquisition timing error of each sensor is ≤1ms, which solves the problem of subsequent inversion deviation caused by the misalignment of traditional data acquisition timing, and enables multimodal data to have time dimension matching, which is convenient for subsequent fusion processing; Data transmission module: It adopts a transmission method combining 5G communication and fiber optic network, which supports high-speed and stable transmission of massive data. It also has data encryption function to ensure the security of detection data. During the transmission process, a unified data encapsulation protocol is used and a data verification code is attached to facilitate the subsequent receiving end to quickly verify the integrity of the data and avoid repeated collection due to data loss or damage.

[0023] The intelligent preprocessing subsystem includes a data cleaning module, an outlier removal module, a multimodal feature enhancement module, and a data standardization module. Deployed on edge computing nodes, it reduces data transmission latency and performs enhanced preprocessing on the raw geophysical data. Data cleaning module: Uses sliding window method to remove random noise, deletes invalid data such as zero values ​​and extreme values ​​caused by sensor failure through data consistency verification, and synchronously marks data cleaning logs to facilitate subsequent data processing traceability; Outlier Removal Module: Combining the improved isolated forest algorithm with Grubbs' statistical test, it accurately identifies and removes random outliers caused by instrument errors and systematic outliers caused by geological anomalies, retaining effective geological information and avoiding interference from outlier data with subsequent inversion model training and calculation; Multimodal feature enhancement module: Extracts specific features for different types of geophysical data (time domain and frequency domain features for seismic wave data, impedance and polarization features for electromagnetic data, and gradient and boundary features for gravity and magnetic data), constructs a multi-dimensional feature set, and directly provides suitable input features for subsequent AI inversion models, reducing the burden of model data preprocessing; Data standardization module: The Z-score normalization algorithm is used to map geophysical data of different dimensions and magnitudes to the same distribution range, eliminate the difference in dimensions, and output highly reliable structured data, thereby ensuring that subsequent multimodal data can be directly fused and input into the inversion model.

[0024] The AI ​​inversion decision subsystem includes a pre-trained inversion model library, an incremental learning module, and a multi-objective optimization module. As the core decision-making unit of the system, it realizes the intelligent inversion of geological parameters. The pre-trained inversion model library contains three core models: seismic wave inversion model (U-Net improved network + attention mechanism), electromagnetic parameter inversion model (deep belief network DBN), and gravity and magnetic joint inversion model (multi-task convolutional neural network MT-CNN). Each model is adapted to the inversion requirements of different modal data. At the same time, the input layer of each model is adapted to the structured data format output by the intelligent preprocessing subsystem, without the need for additional data conversion. Incremental learning module: Employs the Elastic Weight Consolidation (EWC) algorithm to continuously absorb data from new exploration areas and borehole verification results, fine-tune model parameters, and protect existing knowledge while adapting to new geological conditions to avoid catastrophic amnesia. At the same time, the new data format is consistent with historical training data, ensuring the continuity of model training. Multi-objective optimization module: Constructs an optimization function with the objectives of detection accuracy, data fit, and computational efficiency, balances inversion accuracy and time consumption, generates the optimal inversion result, and stores the geological parameters output by inversion in a unified format for easy direct call by the subsequent dynamic interpretation subsystem.

[0025] The dynamic interpretation subsystem includes a geological body identification module, a parameter correlation analysis module, and a dynamic correction module. Based on the inversion results and known geological data, it intelligently identifies deep geological structures, mineral distributions, and potential hazards, and dynamically corrects the interpretation conclusions. Geological body identification module: It adopts a pre-trained ResNet-50 classification model and combines the geological parameters (density, resistivity, wave velocity) obtained by inversion to identify rock types such as sedimentary rocks, igneous rocks, and metamorphic rocks, and delineate the boundaries of geological bodies such as ore bodies, faults, and aquifers. The output identification results are accompanied by confidence scores for subsequent verification and correction. Parameter Correlation Analysis Module: This module uses Bayesian networks to analyze the correlation between geological parameters and mineral resource reserves and potential geological hazards (such as fault activity and underground karst), quantifies the probability of resource reserves and the level of hazard risk, and the analysis results are presented in a standardized report format, including information such as data source, correlation logic, and quantitative indicators. Dynamic correction module: By combining real-time detection data and newly acquired geological data (such as borehole data and regional geological maps), the module dynamically adjusts the parameters of the interpretation model, updates the geological interpretation conclusions, improves the accuracy of the interpretation, and supports the import of raw data, preprocessed data and inversion results from previous stages for retrospective analysis, ensuring the verifiability of the interpretation conclusions.

[0026] The visualization output subsystem includes a 3D modeling module, a results display module, and a data export module. It transforms the inversion results and interpretation conclusions into a 3D geological model, supporting multi-dimensional visualization and standardized data export. 3D Modeling Module: Uses VTK tools to build 3D geological models, restoring the spatial characteristics of deep geological structures, geological body distribution, resource distribution, and disaster-prone areas; Results display module: Supports viewing model slices, adjusting transparency, and switching between multiple perspectives, intuitively displaying key information such as the distribution of geological parameters and the boundaries of geological bodies. At the same time, the display interface can link raw data, preprocessed data, and inversion results, and facilitates cross-stage comparison and analysis for users. Data export module: Supports exporting standardized data formats such as SEGY, ASCII, and Shapefile, and is compatible with mainstream geological exploration software such as Surfer and MapGIS. It also enables permission-based sharing of exploration results through a cloud platform. In addition, the exported data comes with complete metadata information (such as collection time, exploration area, processing flow, etc.), which makes it easy for users to quickly understand the data background and use it directly.

[0027] The fault diagnosis and adaptive adjustment module monitors the system's operating status in real time. Status monitoring: Real-time monitoring of sensor operating status, data transmission link stability, and AI model operation; Adaptive adjustment: When a sensor fails, it automatically switches to a backup sensor to ensure the continuity of data collection; when data transmission is interrupted, it activates the local cache of the edge node to restore the connection and resume transmission, avoiding data loss that could affect subsequent processing; when the model inversion accuracy decreases, it triggers the incremental learning process and marks abnormal situations, making it easier for staff to trace the root cause of the problem, ensuring the smooth flow of data throughout the system, and providing reliable data support for subsequent stages.

[0028] The artificial intelligence-based geophysical deep-earth exploration method of the present invention is based on the above system and includes the following steps: S1: Synchronous acquisition of multi-source geophysical data: Seismic wave data, electromagnetic data, gravity data, and magnetic data are acquired synchronously through a distributed geophysical sensor array. The acquisition timing is calibrated by the synchronous control unit and then uploaded to the intelligent preprocessing subsystem through the data transmission module. S2: Multimodal data augmentation and preprocessing: It can clean, remove outliers, enhance features and standardize raw geophysical data to obtain structured data; S3: AI Model Training and Optimization: Based on historical geophysical data and known geological information, the deep learning model in the inversion model library is initially trained. The model parameters are optimized by absorbing new scene data through the incremental learning module, thereby ensuring that the model is adapted to the geological conditions of different exploration areas. S4: Intelligent Inversion Calculation: Input the pre-processed structured data into the trained AI inversion model, and combine it with a multi-objective optimization algorithm to invert the geological parameters of the deep Earth (stratum thickness, rock density, resistivity, wave velocity). S5: Dynamic geological interpretation: Based on the geological parameters obtained by inversion and combined with regional geological background data, it can intelligently identify the types of deep geological bodies, the distribution of mineral resources and potential geological hazards, thereby dynamically correcting the interpretation results; S6: 3D visualization output: Construct a 3D geological model from the inversion results and interpretation conclusions, display it in multiple dimensions through a visualization terminal, and support data export and results sharing.

[0029] refer to Figure 3 S1-1: Seismic wave data acquisition: The detector array acquires the propagation signals of P-waves and S-waves, with a sampling frequency of 250-1000Hz, and records parameters such as signal amplitude, phase, and propagation time; For example, shallow exploration (such as engineering geology) may use a higher sampling rate (such as 1000 Hz) to capture high-frequency signals; Deep exploration (such as oil and gas resources) may use a lower sampling rate (such as 250-500 Hz) because deep reflection waves have a lower frequency.

[0030] S1-2: Electromagnetic data acquisition: Combining the controlled source electromagnetic method and the natural field electromagnetic method, electromagnetic signals in the frequency range of 1Hz-10kHz are acquired, and electric field strength, magnetic field strength and impedance parameters are recorded. For example: Low frequency range (1Hz-1kHz): suitable for deep exploration, such as monitoring crustal structure or natural gas hydrate reservoirs, where electromagnetic waves have strong penetrating power. High-frequency range (1kHz-10kHz): Focuses on fine imaging of shallow areas, such as hydrogeological or engineering geological surveys, and can capture high-resistivity bodies or conductive anomalies.

[0031] S1-3: Gravity data acquisition: The gravimeter acquires abnormal gravity acceleration data at a sampling interval of 1-5 seconds, and simultaneously records the latitude, longitude, and elevation information of the acquisition point; S1-4: Magnetic data acquisition: The magnetometer acquires the total intensity and component anomaly data of the geomagnetic field with a resolution ≤0.1nT, and simultaneously calibrates the instrument error; S1-5: Data Upload: The collected raw data is encrypted by the data transmission module and then uploaded to the intelligent preprocessing subsystem.

[0032] As a technical optimization of the present invention, the synchronous acquisition and encrypted transmission of multi-source geophysical data ensures the integrity, timeliness and security of the data, providing high-quality raw data support for subsequent intelligent processing and avoiding detection errors caused by data misalignment or loss.

[0033] refer to Figure 1 S2-1: Data cleaning: Sliding window method to remove random noise, consistency check to delete invalid data; S2-2: Outlier Removal: Improved Isolation Forest Algorithm + Grubbs Test to remove random and systematic outliers; First, the Isolation Forest algorithm is used to perform preliminary anomaly detection on the multidimensional data to identify potential anomalous data points and their occurrence times. Then, the Grubbs test is applied to the anomalous subsets or key dimensions detected by the Isolation Forest for secondary verification or further analysis to improve detection accuracy.

[0034] S2-3: Multimodal Feature Enhancement: Extracting specific features from various types of data and constructing a multi-dimensional feature set; S2-4: Data Standardization: Z-score normalization to eliminate dimensional differences and output structured data.

[0035] As a technical optimization scheme of the present invention, the preprocessing is enhanced through multiple steps, which effectively improves the purity and feature characterization ability of geophysical data, eliminates noise and dimensional interference, provides highly reliable structured data for AI inversion models, and lays the foundation for accurate inversion.

[0036] refer to Figure 4 S3-1: Model Building: Building seismic wave inversion model, electromagnetic parameter inversion model, and gravity and magnetic joint inversion model; S3-2: Initial training: Using historical geophysical data (including known borehole verification data) and geological parameter labels, train the model until the loss function converges (loss value ≤ 0.03). S3-3: Incremental optimization: The EWC algorithm incorporates new exploration data and validation results, fine-tunes model parameters, and adapts to new geological conditions; S3-4: Model Evaluation: Evaluate the model accuracy using RMSE (Root Mean Square Error) and R² (Correlation Coefficient), ensuring RMSE ≤ 5% and R² ≥ 0.9.

[0037] As a technical optimization scheme of the present invention, the model can be dynamically adapted to different geological conditions through multi-model adaptation and incremental learning, avoiding catastrophic forgetting. At the same time, the reliability and stability of the model inversion results are guaranteed through rigorous accuracy evaluation.

[0038] refer to Figure 4 S4-1: Data Input: Input structured data into the corresponding inversion model according to modality classification, and use a batch processing mechanism to improve computational efficiency; S4-2: Multi-objective optimization: Constructing the optimization function ,in , , Weighting coefficients (satisfying) + + =1), T is the inversion time. This is the maximum allowable time. S4-3: Inversion Calculation: Accelerates the inversion process through GPU parallel computing, and outputs core geological parameters such as stratum thickness, rock density (error ≤3%), resistivity (error ≤8%), and wave velocity (error ≤4%). S4-4: Result Verification: Cross-validate the inversion results by combining known borehole data and regional geological maps, and eliminate invalid inversion results.

[0039] As a technical optimization scheme of the present invention, multi-objective optimization and GPU parallel computing are used to improve computational efficiency while ensuring inversion accuracy. Combined with the cross-validation mechanism, effective results are further screened, thereby significantly reducing the error caused by multiple solutions in inversion.

[0040] refer to Figure 1 S5-1: Geological body identification: Based on inversion parameters, the pre-trained geological body classification model (ResNet-50) is used to identify rock types such as sedimentary rocks, igneous rocks, and metamorphic rocks, and to delineate the distribution range of ore bodies; S5-2: Parameter Correlation Analysis: Bayesian network analysis is used to analyze the correlation between geological parameters and mineral resources and disaster risks, and to quantify the probability of resource reserves and the risk level of disaster occurrence. S5-3: Dynamic Correction: By combining real-time detection data with newly acquired geological data, the parameters of the interpretation model are dynamically adjusted and the geological interpretation conclusions are updated. S5-4: Result Verification: Verify the interpretation results through methods such as field geological surveys and borehole verification to ensure that the interpretation accuracy rate is ≥85%.

[0041] As a technical optimization of the present invention, AI intelligent recognition and dynamic correction are used to get rid of the dependence on human experience, quantify resources and disaster risks, improve the objectivity and accuracy of geological interpretation, and further ensure the credibility of interpretation results through field verification.

[0042] refer to Figure 5 S6-1: 3D Modeling: VTK (Visualization Toolkit) is used to construct a 3D geological model, which can restore the spatial distribution characteristics of deep geological structures; S6-2: Visualization: Supports slice viewing, transparency adjustment, and multi-view switching, and can display the distribution of geological parameters, geological body boundaries, resource distribution range, and disaster-prone areas; S6-3: Data Export: Supports exporting standardized data formats (SEGY, ASCII, Shapefile) and is compatible with mainstream geological exploration software; S6-4: Results Sharing: Accessibility sharing of detection results is achieved through a cloud platform, supporting access and viewing by multiple terminals.

[0043] As a technical optimization of the present invention, the deep geological features are presented intuitively through three-dimensional visualization and standardized sharing, which makes it convenient for exploration personnel to quickly grasp the exploration results. At the same time, it is compatible with mainstream software and multi-terminal access, which improves the practicality and dissemination efficiency of the results.

[0044] The working principle and usage process of this invention are as follows: First, the distributed geophysical sensor array (including seismic wave sensors, electromagnetic sensors, gravity sensors, magnetic sensors, a backup sensor array, and auxiliary monitoring sensors) of the multi-source data acquisition subsystem is deployed according to the detection grid. Under the calibration of the synchronization control unit (GPS timing, error ≤1ms), it synchronously acquires raw data of seismic waves, electromagnetic waves, gravity, and magnetic forces in multiple modes. The gravity sensor acquisition accuracy is ≤0.01mGal, and the magnetic sensor resolution is ≤0.1nT. The auxiliary monitoring sensor synchronously records elevation and positioning information. At this time, the acquired data is monitored in real time by the fault diagnosis module. When a sensor fails, the backup sensor is automatically switched. Then, the raw data is uploaded to the intelligent preprocessing subsystem via the data output interface through the data transmission module (5G + fiber optic, encrypted). The system then proceeds to the intelligent preprocessing subsystem, deployed on edge computing nodes. The data cleaning module uses a sliding window method to remove random noise and delete invalid data. Simultaneously, the outlier removal module combines an improved isolated forest algorithm with the Grubbs test to remove random and systematic outliers. The multimodal feature enhancement module extracts data-specific features to construct a multi-dimensional feature set. The data standardization module uses the Z-score normalization algorithm to eliminate dimensional differences, thus outputting highly reliable structured data. The AI ​​inversion decision subsystem receives the structured data and processes it according to data modality using pre-trained inversion model libraries, including seismic wave inversion models (U-Net + attention mechanism), electromagnetic parameter inversion models (DBN), and gravity-magnetic joint inversion models (MT-CNN). The multi-objective optimization module then performs optimization based on an optimization function. To balance accuracy and efficiency, the inversion calculation module uses GPU parallel acceleration to obtain geological parameters such as stratigraphic thickness and rock density. The result verification module then performs cross-validation using borehole data and geological maps. Simultaneously, the incremental learning module employs the EWC algorithm to absorb new data and optimize model parameters, avoiding catastrophic forgetting. The dynamic interpretation subsystem's geological body identification module uses a ResNet-50 model to identify rock types and delineate orebody boundaries. The parameter correlation analysis module uses a Bayesian network to quantify resource reserve probability and disaster risk level. The dynamic correction module updates interpretation conclusions based on real-time data and new geological information. The result verification module ensures an interpretation accuracy of ≥85% through field surveys and borehole verification. If the interpretation is satisfactory, the result is output; otherwise, it is returned to the AI ​​for feedback. The decision-making subsystem performs a re-inversion; the 3D modeling module of the visualization output subsystem uses VTK tools to build a 3D geological model, the results display module supports visualization operations such as slice viewing and multi-view switching, the data export module exports standardized format data such as SEGY, ASCII, and Shapefile, and the cloud sharing module achieves permission-based sharing through the cloud platform, ultimately forming a 3D geological model, standardized data, and shared results. New data generated during the application of results and continuous monitoring is returned to the multi-source data acquisition subsystem to start the next round of exploration. The fault diagnosis and adaptive adjustment module monitors the system operation throughout the process, starts local caching at edge nodes when data transmission is interrupted, and triggers an incremental learning process when the model inversion accuracy decreases, ensuring stable system operation.

[0045] In summary, this AI-based geophysical deep earth exploration system and method, through a fully intelligent architecture encompassing "acquisition-preprocessing-inversion-interpretation-visualization," innovatively integrates multi-source geophysical data synchronous acquisition, multi-modal data enhancement preprocessing, multi-model AI intelligent inversion, dynamic geological interpretation, and 3D visualization sharing technologies. This enables high-precision and high-efficiency exploration of deep earth geological structures, mineral resources, and potential hazards. Its core innovations lie in enhancing the completeness of exploration information through multi-source data fusion and mining, improving model scenario adaptability through incremental learning and multi-objective optimization, eliminating reliance on human experience through AI intelligent interpretation, ensuring result credibility through dynamic correction and cross-validation, and enhancing system stability through fault adaptive adjustment. It provides an intelligent, precise, and efficient solution for deep earth exploration, effectively contributing to the digital and intelligent development of geophysical exploration equipment.

[0046] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0047] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A geophysical deep-earth exploration system based on artificial intelligence, characterized in that: It includes a multi-source data acquisition subsystem, an intelligent preprocessing subsystem, an AI inversion decision-making subsystem, a dynamic interpretation subsystem, and a visualization output subsystem. Each subsystem achieves data interaction and collaborative work through a high-speed communication network. The multi-source data acquisition subsystem consists of a distributed geophysical sensor array, a data transmission module, and a synchronization control unit. The geophysical sensor array includes seismic wave sensors, electromagnetic sensors, gravity sensors, and magnetic sensors, which are used to synchronously acquire multimodal geophysical data of the deep earth. The synchronization control unit realizes the timing synchronization of the acquisition of each sensor, and the data transmission module transmits the raw data through a 5G / fiber optic network. The intelligent preprocessing subsystem is deployed on edge computing nodes and includes a data cleaning module, an outlier removal module, a multimodal feature enhancement module, and a data standardization module. It is used to enhance and preprocess the raw geophysical data and output highly reliable structured data. The AI ​​inversion decision subsystem includes a pre-trained inversion model library, an incremental learning module, and a multi-objective optimization module. The pre-trained inversion model library includes seismic wave inversion models, electromagnetic parameter inversion models, and gravity-magnetic joint inversion models based on deep learning. The incremental learning module continuously absorbs new detection data to optimize model parameters. The multi-objective optimization module generates the optimal inversion result with the goal of maximizing detection accuracy and minimizing data fitting error. The dynamic interpretation subsystem includes a geological body identification module, a parameter correlation analysis module, and a dynamic correction module. Based on the inversion results and known geological data, it intelligently identifies deep geological structures, mineral distribution, and potential hazards, and dynamically corrects the interpretation conclusions. The visualization output subsystem includes a 3D modeling module, a results display module, and a data export module. It transforms the inversion results and interpretation conclusions into a 3D geological model, supporting multi-dimensional visualization and standardized data export.

2. A geophysical method for deep Earth exploration based on artificial intelligence, characterized in that, Includes the following steps: S1: Synchronous acquisition of multi-source geophysical data: Seismic wave data, electromagnetic data, gravity data, and magnetic data are acquired synchronously through a distributed geophysical sensor array. The acquisition timing is calibrated by the synchronous control unit and uploaded to the intelligent preprocessing subsystem through the data transmission module. S2: Multimodal data augmentation and preprocessing: The original geophysical data is cleaned, outlier removed, feature enhanced and standardized to obtain structured data; S3: AI Model Training and Optimization: Based on historical geophysical data and known geological information, the deep learning model in the inversion model library is initially trained. The model parameters are optimized by absorbing new scene data through the incremental learning module to ensure that the model is adapted to the geological conditions of different exploration areas. S4: Intelligent Inversion Calculation: Input the pre-processed structured data into the trained AI inversion model, and combine it with a multi-objective optimization algorithm to invert the deep geological parameters of the earth (stratum thickness, rock density, resistivity, wave velocity). S5: Dynamic Geological Interpretation: Based on the geological parameters obtained from inversion and combined with regional geological background data, it intelligently identifies the types of deep geological bodies, the distribution of mineral resources and potential geological hazards, and dynamically corrects the interpretation results. S6: 3D visualization output: Construct a 3D geological model from the inversion results and interpretation conclusions, and display it in multiple dimensions through a visualization terminal, supporting data export and result sharing.

3. The method for deep Earth exploration based on artificial intelligence according to claim 2, characterized in that: S1-1: Seismic wave data acquisition: Acquire P-wave and S-wave propagation signals through a detector array, with the sampling frequency set to 250-1000Hz, and record parameters such as signal amplitude, phase, and propagation time; S1-2: Electromagnetic data acquisition: Using a combination of controlled source electromagnetic method and natural field electromagnetic method, electromagnetic signals with a frequency range of 1Hz-10kHz are acquired, and electric field strength, magnetic field strength and impedance parameters are recorded. S1-3: Gravity data acquisition: Collect abnormal gravity acceleration data using a gravimeter, with a sampling interval of 1-5 seconds, and simultaneously record the latitude, longitude, and elevation information of the acquisition point; S1-4: Magnetic data acquisition: Collect total intensity and component anomaly data of the geomagnetic field using a magnetometer, with a resolution ≤0.1nT, and simultaneously calibrate the instrument error; S1-5: Synchronization control: GPS timing and clock synchronization technology are used to ensure that the timing error of each sensor acquisition is ≤1ms, and data transmission adopts an encryption protocol to ensure data security.

4. The method for deep Earth exploration based on artificial intelligence according to claim 2, characterized in that: S2-1: Data cleaning: Use the sliding window method to remove random noise from the data, and delete invalid data (such as zero values ​​and extreme values ​​caused by sensor failure) through data consistency verification. S2-2: Outlier Removal: An improved isolated forest algorithm combined with statistical tests (Grubbs test) is used to identify and remove systematic outliers caused by geological anomalies and random outliers caused by instrument errors. S2-3: Multimodal feature enhancement: Extract time-domain features (peak value, energy, dominant frequency) and frequency-domain features (spectral distribution, phase spectrum) from seismic wave data; extract impedance features and polarization features from electromagnetic data; extract gradient features and anomaly boundary features from gravity and magnetic data; and construct a multi-dimensional feature set. S2-4: Data Standardization: The Z-score normalization algorithm is used to map geophysical data of different dimensions to the same distribution range, eliminating the difference in dimensions and obtaining structured data.

5. The method for deep Earth exploration based on artificial intelligence according to claim 2, characterized in that: S3-1: Model Construction: The seismic wave inversion model adopts the U-Net improved network (with added attention mechanism), the electromagnetic parameter inversion model adopts the deep belief network (DBN), and the gravity and magnetic joint inversion model adopts the multi-task convolutional neural network (MT-CNN). S3-2: Initial training: Using historical geophysical data (including known borehole verification data) and geological parameter labels, train the model until the loss function converges (loss value ≤ 0.03). S3-3: Incremental optimization: The Elastic Weight Consolidation (EWC) algorithm is used to continuously absorb data from new exploration areas and validation results, fine-tune model parameters, and avoid catastrophic forgetting. S3-4: Model Evaluation: The accuracy of the model inversion is evaluated by the root mean square error (RMSE) and the correlation coefficient (R²), ensuring that RMSE ≤ 5% and R² ≥ 0.

9.

6. The method for deep Earth exploration based on artificial intelligence according to claim 2, characterized in that: S4-1: Data Input: Input structured data into the corresponding inversion model according to modality classification, and use a batch processing mechanism to improve computational efficiency; S4-2: Multi-objective optimization: Constructing the optimization function ,in , , Weighting coefficients (satisfying) + + =1), T is the inversion time. This is the maximum allowable time. S4-3: Inversion Calculation: Accelerates the inversion process through GPU parallel computing, and outputs core geological parameters such as stratum thickness, rock density (error ≤3%), resistivity (error ≤8%), and wave velocity (error ≤4%). S4-4: Result Verification: Cross-validate the inversion results by combining known borehole data and regional geological maps, and eliminate invalid inversion results.

7. The method for deep Earth exploration based on artificial intelligence according to claim 2, characterized in that: S5-1: Geological body identification: Based on inversion parameters, the pre-trained geological body classification model (ResNet-50) is used to identify rock types such as sedimentary rocks, igneous rocks, and metamorphic rocks, and to delineate the distribution range of ore bodies; S5-2: Parameter Correlation Analysis: Bayesian network analysis is used to analyze the correlation between geological parameters and mineral resources and disaster risks, and to quantify the probability of resource reserves and the risk level of disaster occurrence. S5-3: Dynamic Correction: By combining real-time detection data with newly acquired geological data, the parameters of the interpretation model are dynamically adjusted and the geological interpretation conclusions are updated. S5-4: Result Verification: Verify the interpretation results through methods such as field geological surveys and borehole verification to ensure that the interpretation accuracy rate is ≥85%.

8. The method for deep Earth exploration based on artificial intelligence according to claim 2, characterized in that: S6-1: 3D Modeling: VTK (Visualization Toolkit) is used to construct a 3D geological model to restore the spatial distribution characteristics of deep geological structures; S6-2: Visualization: Supports slice viewing, transparency adjustment, and multi-view switching to display the distribution of geological parameters, geological body boundaries, resource distribution range, and disaster-prone areas; S6-3: Data Export: Supports exporting standardized data formats (SEGY, ASCII, Shapefile), compatible with mainstream geological exploration software; S6-4: Results Sharing: Accessibility sharing of exploration results is achieved through a cloud platform, supporting access and viewing from multiple terminals.

9. The artificial intelligence-based geophysical deep-earth exploration system according to claim 1, characterized in that, It also includes a fault diagnosis and adaptive adjustment module, which monitors the sensor working status, data transmission link and model operation in real time. When a sensor failure occurs, it automatically switches to a backup sensor; when data transmission is interrupted, it starts the local cache of the edge node; when the model inversion accuracy decreases, it triggers the incremental learning process to ensure the stable operation of the system.

Images