A deep learning-based gold mine geological information analysis system and method

Abstract

The application discloses a kind of gold mine geological information analysis system and method based on deep learning, it is related to geological information intelligent processing technical field, including: collection mining area geological data, extract fusion space structure and time evolution characteristics;Utilize the feature mapping of antagonistic network, generate standardized geological features;It is divided into multiple time windows, combined with tectonic stress field data to build evolution sequence;Calculate the response characteristics of geological features to stress field, determine the correlation intensity of both;The correlation intensity is input into prediction function, and the probability of each time window mineralization is calculated using adaptive weight, and the distribution map is constructed to realize gold mine prediction.The method integrates multi-source geological data and spatiotemporal characteristics, improves the accuracy and reliability of gold mine prediction.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent processing of geological information, in particular to a gold mine geological information analysis system and method based on deep learning. BACKGROUND

[0002] With the development of deep learning technology, the method of geological information processing gradually develops from traditional statistical analysis to intelligent direction. Gold mine geological information analysis mainly uses geological statistical method to analyze the geological data of mining area, and predicts the ore-forming area by constructing the spatial distribution model of geological elements. However, this method is difficult to effectively process the high-dimensional characteristics of geological data, and cannot fully utilize the dynamic information in the process of geological structure evolution.

[0003] The extraction of geological features in traditional methods mainly depends on artificial experience, which is prone to problems such as incomplete feature extraction and insufficient feature expression. Due to the lack of in-depth analysis of the evolution process of geological structure, it is difficult to accurately describe the relationship between tectonic stress field and mineralization, which affects the accuracy of mineralization prediction.

[0004] In the prior art, geological features and stress field evolution features are often treated separately, and there is a lack of effective feature fusion mechanism. The standardization processing method of geological data is relatively simple, and the spatio-temporal correlation of data is not fully considered, resulting in non-standardized feature expression, which affects the reliability of subsequent analysis. SUMMARY

[0005] The purpose of the present application is to provide a gold mine geological information analysis system and method based on deep learning, which provides a gold mine geological information analysis method based on deep learning, intelligently extracts and standardizes geological features through a deep learning model, and establishes a dynamic prediction model combined with stress field evolution information, so as to improve the accuracy and reliability of gold mine mineralization prediction.

[0006] The gold mine geological information analysis method based on deep learning provided in the embodiment of the present application comprises the following steps:

[0007] Collecting geological data of the target mining area, extracting spatial structure features and time sequence evolution features and fusing them to generate complete geological features;

[0008] Mapping the complete geological features through a pre-trained adversarial network, iteratively optimizing the feature mapping parameters based on the feature similarity loss function, and generating standardized geological features;

[0009] Dividing the standardized geological features into multiple time windows, collecting tectonic stress field data, extracting principal stress direction and stress gradient to construct stress field evolution sequence;

[0010] calculate a response feature of the standardized geological feature to the stress field evolution sequence in each time window, determine a stress field geological feature correlation strength based on the response feature;

[0011] input the stress field geological feature correlation strength into a prediction function, calculate a mineralization probability score of each time window using adaptive weights, construct a mineralization probability distribution map, and determine a gold mineralization prediction result of the target mining area based on the mineralization probability distribution map.

[0012] Further, collect geological data of the target mining area, extract spatial structure features and time series evolution features and fuse them to generate complete geological features including:

[0013] Collect geological data of the target mining area, perform depth domain correction and phase extraction on the geological data of the target mining area to generate preprocessed geological data;

[0014] Extract geological body attributes based on the preprocessed geological data, introduce a connectivity judgment condition to identify structural fracture distribution, and combine spatial positioning information to calculate stress field distribution to obtain spatial structure features;

[0015] Align the preprocessed geological data in time series, calculate the deformation of the geological body and the stress change at adjacent time points, determine the stress-strain relationship according to the deformation of the geological body and the stress change, and obtain the time series evolution features;

[0016] Construct a feature tensor from the spatial structure features and the time series evolution features, introduce an evolution continuity constraint based on the stress-strain relationship in tensor decomposition to obtain fused features, and reconstruct complete geological features that satisfy the mechanical equilibrium constraint according to the fused features.

[0017] Further, map the complete geological features through a pre-trained adversarial network, iteratively optimize the feature mapping parameters based on a feature similarity loss function, and generate standardized geological features including:

[0018] Construct an adversarial network containing an encoder and a discriminator, the encoder maps the complete geological features, and the discriminator discriminates the distribution of the feature mapping results;

[0019] Pre-train the adversarial network using standard geological feature samples to obtain a pre-trained adversarial network;

[0020] Input the complete geological features into the pre-trained adversarial network for feature mapping to obtain the feature mapping results;

[0021] Calculate the distributional adversarial loss of the feature mapping results and the standard geological feature samples;

[0022] Calculate the similarity between the feature mapping results corresponding to sample points with a spatial position distance below a preset distance threshold as the feature similarity loss.

[0023] The distribution adversarial loss is combined with the feature similarity loss to obtain an optimization objective of the feature mapping parameter;

[0024] The feature mapping parameter and the discriminator parameter of the encoder are alternately optimized based on the optimization objective until the feature mapping result converges in difference with the distribution of the standard geological feature, and the converged feature mapping result is taken as the standardized geological feature.

[0025] Further, the standardized geological feature is divided into multiple time windows, tectonic stress field data is collected, the principal stress direction and the stress gradient are extracted to construct a stress field evolution sequence, including:

[0026] The weighted sum of the displacement amount and the tectonic importance weight of each tectonic unit in the standardized geological feature is calculated to obtain a tectonic activity intensity index;

[0027] According to the time variation rate of the tectonic activity intensity index, a demarcation point is determined, and the standardized geological feature is divided into multiple time windows;

[0028] In each time window, tectonic stress field data is collected, and a strain tensor is calculated according to the tectonic stress field data and is converted into a stress tensor;

[0029] Eigenvalue decomposition is performed on the stress tensor to extract the principal stress direction, the spatial variation of the stress tensor is calculated to obtain the stress gradient, and the principal stress direction and the stress gradient of adjacent time windows are processed using a time weight factor to construct a stress field evolution sequence.

[0030] Further, the weighted sum of the displacement amount and the tectonic importance weight of each tectonic unit in the standardized geological feature is calculated to obtain a tectonic activity intensity index, including:

[0031] The standardized geological feature data is projected onto the principal axis direction of the tectonic strain ellipsoid, the stress components of the tectonic unit in each principal axis direction are obtained, and the displacement amount of the tectonic unit is calculated based on the stress components;

[0032] The spatial geometric attribute and the activity period parameter of the tectonic unit are extracted, the mapping relationship between the tectonic unit attribute and the activity period is established through a stress field response function, and a tectonic importance weight is generated;

[0033] The displacement amount and the tectonic importance weight are coupled and operated in a tensor space to construct a regional tectonic activity intensity tensor;

[0034] According to the eigenvalue spatial distribution law of the regional tectonic activity intensity tensor, the weighted sum of the displacement amount and the tectonic importance weight of each tectonic unit is calculated to obtain a tectonic activity intensity index.

[0035] Further, the response characteristics of the standardized geological features in each time window corresponding to the stress field evolution sequence are calculated, and the stress field geological feature correlation strength is determined based on the response characteristics, including:

[0036] The standardized geological features are divided into multiple tectonic units in space, the stress field evolution sequence corresponding to each tectonic unit is calculated in each time window, and the tectonic unit stress field change amount is obtained;

[0037] The response characteristics are calculated according to the tectonic unit stress field change amount and the duration of the time window;

[0038] The response characteristics are decomposed to obtain the maximum amplitude of the response characteristics and the time interval required to reach the maximum amplitude, and the response characteristic coefficient is calculated based on the ratio of the maximum amplitude to the time interval;

[0039] According to the time sequence variation law of the response characteristic coefficient and the spatial distribution characteristics of the tectonic unit, the stress field geological feature correlation strength is calculated.

[0040] Further, the stress field geological feature correlation strength is input into the prediction function, the ore-forming probability score of each time window is calculated using adaptive weights, the ore-forming probability distribution map is constructed, and the gold ore-forming prediction result of the target mining area is determined based on the ore-forming probability distribution map, including:

[0041] The feature correlation degree of adjacent time is calculated to obtain the time sequence weight, the prediction function is constructed by combining the time sequence weight and the kernel function based on the feature decomposition tectonic kernel function;

[0042] The stress field geological feature correlation strength is input into the prediction function, and the main feature and its dynamic evolution characteristics are obtained through feature decomposition;

[0043] The adaptive weight is calculated based on the direction change of the main feature in adjacent time windows;

[0044] The main feature and the adaptive weight are combined and operated in the prediction function to obtain the ore-forming probability score of each time window, and the ore-forming probability distribution map is constructed by probability superposition and spatial interpolation of the ore-forming probability score;

[0045] The probability gradient and the probability aggregation area in the ore-forming probability distribution map are extracted, and the gold ore-forming prediction result of the target mining area is determined based on the change trend of the probability gradient and the spatial distribution of the probability aggregation area.

[0046] An embodiment of the present application provides a gold mine geological information analysis system based on deep learning, the system comprising:

[0047] The geological feature extraction module is used for collecting geological data of the target mining area, extracting spatial structure features and time sequence evolution features and fusing them to generate complete geological features;

[0048] The feature standardization module is configured to map complete geological features through a pre-trained adversarial network, iteratively optimize feature mapping parameters based on a feature similarity loss function, and generate standardized geological features.

[0049] The stress field construction module is configured to divide the standardized geological features into multiple time windows, collect tectonic stress field data, extract principal stress directions and stress gradients, and construct a stress field evolution sequence.

[0050] The correlation analysis module is configured to calculate response features of the standardized geological features to the stress field evolution sequence in each time window, and determine stress field-geological feature correlation strength based on the response features.

[0051] The ore-forming prediction module is configured to input the stress field-geological feature correlation strength into a prediction function, calculate ore-forming probability scores for each time window using adaptive weights, construct an ore-forming probability distribution map, and determine a gold ore-forming prediction result for the target mining area based on the ore-forming probability distribution map.

[0052] In an embodiment of the present application, a technical solution is also provided, which is an electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor implements the steps in any of the preceding methods when executing the computer program.

[0053] In an embodiment of the present application, a technical solution is also provided, which is a computer-readable storage medium, having computer program instructions stored thereon, wherein the computer program instructions are executed by a processor to implement the steps in any of the preceding methods.

[0054] The present application can effectively overcome the dependence on artificial experience by intelligently extracting geological features through a deep learning model, realize automatic extraction and fusion of geological features, and improve the completeness and accuracy of feature expression. The adversarial network is used for feature mapping, which effectively improves the standardization degree of geological features and enhances the normalization of feature expression. The stress field evolution sequence analysis method based on time windows realizes the dynamic correlation between geological features and stress field evolution features, and can accurately capture the influence of tectonic stress field on mineralization. The introduction of adaptive weights improves the combination optimization ability of features in different time windows and enhances the expression ability of the ore-forming prediction model for time sequence features. This method has strong universality and scalability and can be applied to different types of mining area geological conditions, providing more reliable technical support for gold mine exploration.

[0055] The above description of the invention is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0056] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0057] Figure 1 A flowchart illustrating a deep learning-based method for parsing gold deposit geological information, provided as an embodiment of the present invention;

[0058] Figure 2 This is a flowchart of the geological feature standardization adversarial network in an embodiment of the present invention;

[0059] Figure 3 This is a schematic diagram of a deep learning-based gold mine geological information analysis system provided in an embodiment of the present invention. Detailed Implementation

[0060] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. In the following description, when referring to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements.

[0061] The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with one or more embodiments of this specification. It should be noted that the steps of the corresponding methods are not necessarily performed in the order shown and described in this specification in other embodiments. In some other embodiments, the methods may include more or fewer steps than described in this specification. Furthermore, a single step described in this specification may be broken down into multiple steps in other embodiments; and multiple steps described in this specification may be combined into a single step in other embodiments.

[0062] like Figure 1 As shown, Figure 1 A flowchart illustrating a deep learning-based method for analyzing gold deposit geological information is provided in this embodiment of the invention. The method includes the following steps:

[0063] Collect geological data of the target mining area, extract spatial structural features and temporal evolution features, and fuse them to generate complete geological features;

[0064] The complete geological features are mapped by a pre-trained adversarial network, the feature mapping parameters are iteratively optimized based on a feature similarity loss function, and standardized geological features are generated;

[0065] The standardized geological features are divided into multiple time windows, tectonic stress field data are collected, the principal stress direction and stress gradient are extracted to construct a stress field evolution sequence;

[0066] The response features of the standardized geological features to the stress field evolution sequence in each time window are calculated, and the stress field geological feature correlation strength is determined based on the response features;

[0067] The stress field geological feature correlation strength is input into a prediction function, the metallogenic probability score of each time window is calculated using adaptive weights, a metallogenic probability distribution map is constructed, and the gold metallogenic prediction result of the target mining area is determined based on the metallogenic probability distribution map.

[0068] In this embodiment, the geological data of the target mining area is collected, the spatial structure features and time sequence evolution features are extracted and fused, and the complete geological features are generated, including:

[0069] The geological data of the target mining area is collected, and the target mining area geological data is subjected to depth domain correction and phase extraction to generate preprocessed geological data;

[0070] Based on the preprocessed geological data, the geological body attributes are extracted, the connectivity judgment condition is introduced to identify the tectonic fracture distribution, and the stress field distribution is calculated combined with the spatial positioning information to obtain the spatial structure features;

[0071] The preprocessed geological data is subjected to time sequence alignment, the geological body deformation and stress change at adjacent time points are calculated, the stress-strain relationship is determined according to the geological body deformation and stress change, and the time sequence evolution features are obtained;

[0072] The spatial structure features and time sequence evolution features are constructed into a feature tensor, and the evolution continuity constraint based on the stress-strain relationship is introduced in the tensor decomposition to obtain the fused features, and the complete geological features satisfying the mechanical equilibrium constraint are reconstructed according to the fused features.

[0073] First, the geological data of the target mining area needs to be collected. The collection equipment includes geological radar, seismic wave detector and drilling sampling equipment, and the original geological data including underground medium wave impedance, lithology distribution, fault strike, etc. are obtained. The collected original geological data is subjected to depth domain correction processing, the noise in the geological radar and seismic wave data is filtered out using a convolutional neural network, the phase information of the data is extracted through a multi-scale decomposition algorithm, the depth-time mapping relationship is established, and the influence of sampling error on data accuracy is eliminated, thereby generating preprocessed geological data.

[0074] Based on the pre-processed geological data, the properties of the gold mine geological body are extracted. Through the deep learning model, the pre-processed data is analyzed to identify the physical parameters of different geological units, including density, elastic modulus, porosity, and permeability, etc. On this basis, the connectivity judgment condition is introduced to identify the distribution of structural fractures. In the specific implementation, a graph convolutional neural network is used to establish the connectivity graph of the geological body, and a connectivity threshold parameter is set. When the physical parameter difference between adjacent geological bodies is less than the threshold and there is a spatial contact relationship, it is determined as a connected region; when the parameter difference is greater than the threshold or there is a clear interface, it is determined as a broken region. Combined with the spatial positioning information obtained by GPS positioning and geographic information system, the stress field distribution in the region can be calculated by the finite element analysis method, and a stress tensor model can be constructed, which can represent the stress components in each direction and their effects. Through the above steps, the spatial structure characteristics of the gold mine region are obtained, including the ore body shape, the distribution of structural fracture zones, and the stress concentration region.

[0075] The pre-processed geological data is processed for time series alignment to establish the corresponding relationship of data collected at different times. The dynamic time warping algorithm is used to match the geological data collected at different time points, eliminating the inconsistency of data caused by sampling time difference. The deformation of the geological body at adjacent time points is calculated, and the displacement vector field of the geological body in three-dimensional space is extracted through the deep learning model to analyze the deformation trend of the geological body. At the same time, the stress change is calculated, and the stress field data at different times is differentiated to obtain the stress increment distribution. According to the deformation of the geological body and the stress change, a non-linear mapping relationship is established through the deep neural network to determine the stress-strain relationship, which describes the deformation characteristics of the geological body under external force, and then the time evolution characteristics of the gold mine geological structure are obtained.

[0076] The spatial structure characteristics and time evolution characteristics are constructed into a feature tensor. The spatial structure characteristics include the ore body shape, the fracture distribution, and the stress field components, and the time evolution characteristics include the displacement vector field and the stress increment distribution. In the tensor decomposition process, the evolution continuity constraint based on the stress-strain relationship is introduced to ensure that the decomposition result conforms to the physical law of geological evolution. A tensor decomposition neural network is used, which contains a reconstruction error term and a continuity constraint term in the loss function. The continuity constraint is defined based on the stress-strain relationship to ensure that the evolution process of the geological body meets the mechanical equilibrium principle. By optimizing the loss function, a fusion feature is obtained, which integrates spatial structure information and time evolution information. According to the fusion feature, a generative adversarial network is used to reconstruct the complete geological features that meet the mechanical equilibrium constraint. In the reconstruction process, the mechanical equilibrium equation is introduced as a constraint condition to ensure that the reconstruction result conforms to the stress balance principle.

[0077] During feature reconstruction, a dedicated geological information decoder was designed. This decoder incorporates a multi-layer transposed convolutional network and an attention mechanism, enabling it to reconstruct a high-resolution 3D geological model from fused features. The decoder output includes information such as ore body boundaries, internal structure, mechanical properties, and evolutionary trends. To ensure the accuracy of the reconstruction results, an expert knowledge base was introduced for verification. This knowledge base contains typical gold deposit geological structures and evolutionary patterns. Similarity matching was used to compare the reconstruction results with the knowledge base, thereby improving the reliability of the analytical results.

[0078] By processing and analyzing gold mine geological data using deep learning technology, accurate analysis of complex geological structures was achieved. A comprehensive information model reflecting the geological characteristics of gold mines was constructed by integrating spatial structural features and temporal evolution characteristics. Mechanical equilibrium constraints were introduced to ensure the physical rationality of the geological analysis results. Stress-strain relationships within geological bodies were established, enabling an accurate description of the gold mine formation and evolution process. This provided a basis for gold mine exploration and development decisions, reduced exploration risks, and improved resource utilization efficiency. It also solved the problems of low accuracy and poor efficiency in traditional geological analysis methods, providing a new approach to gold mine geological information analysis.

[0079] like Figure 2 As shown, Figure 2 The geological feature standardization adversarial network process provided in the embodiments of the present invention.

[0080] In this embodiment, complete geological features are mapped using a pre-trained adversarial network. The feature mapping parameters are iteratively optimized based on a feature similarity loss function to generate standardized geological features, including:

[0081] An adversarial network is constructed, comprising an encoder and a discriminator. The encoder performs feature mapping on complete geological features, and the discriminator determines the distribution of the feature mapping results.

[0082] The adversarial network is pre-trained using standard geological feature samples to obtain a pre-trained adversarial network;

[0083] The complete geological features are input into a pre-trained adversarial network for feature mapping to obtain the feature mapping results;

[0084] Calculate the distribution of the feature mapping results against the loss of standard geological feature samples;

[0085] The similarity between the feature mapping results corresponding to sample points whose spatial location distance is less than a preset distance threshold is calculated as the feature similarity loss;

[0086] The optimization objective for the feature mapping parameters is obtained by combining the distributional adversarial loss and the feature similarity loss.

[0087] The feature mapping parameters of the encoder and the discriminator parameters are alternately optimized based on an optimization target until the difference between the feature mapping result and the distribution of the standard geological features converges, and the converged feature mapping result is taken as the standardized geological features.

[0088] First, an adversarial network architecture including an encoder and a discriminator is constructed. The encoder adopts a multi-layer convolutional neural network structure, including an input layer, multiple convolutional layers, pooling layers, and a fully connected layer. The input layer receives complete geological feature data, including lithology distribution, fault strike, vein morphology, and other three-dimensional geological information. The convolutional layers use convolution kernels of different sizes to extract multi-scale spatial features of the geological features, and the pooling layers reduce the feature dimension and extract significant features through downsampling. The fully connected layer maps the features to a latent feature space and outputs the feature mapping result. The discriminator also adopts a convolutional neural network structure, and its input is the feature mapping result output by the encoder or a standard geological feature sample. After processing through multiple convolutional and fully connected layers, it outputs a discrimination score, which indicates whether the input data conforms to the distribution characteristics of the standard geological features.

[0089] The constructed adversarial network is pre-trained using standard geological feature samples. The standard geological feature samples are derived from high-quality geological exploration data in known gold mine areas, including typical gold ore-forming environments, ore body morphology, and internal structure information. During the pre-training process, the standard geological feature samples are input into the encoder to obtain the feature mapping result, and the feature mapping result and the standard geological feature samples are input into the discriminator for discrimination. By minimizing the reconstruction error and maximizing the accuracy of the discriminator, the encoder can learn the essential characteristics of the standard geological features, and the discriminator can effectively distinguish between standard and non-standard features. Pre-training uses the batch gradient descent algorithm, each batch contains multiple standard geological feature samples, and the network parameters are updated through the backpropagation algorithm until the network converges, obtaining the pre-trained adversarial network.

[0090] Complete geological features are input into the pre-trained adversarial network for feature mapping. Complete geological features refer to comprehensive geological information formed after processing of original geological data obtained through geological exploration, including stratum structure, lithology distribution, structural features, and mineralization information in the mining area. These data are arranged according to the pre-set format and input into the pre-trained encoder. The encoder processes the input data through multiple neural networks through the forward propagation process and finally outputs the feature mapping result. The feature mapping result retains the key information of the original geological features while having a more standardized and normalized expression form.

[0091] The distribution adversarial loss is calculated between the feature mapping result and the distribution of the standard geological feature sample. The distribution adversarial loss is used to measure the difference between the distribution of the feature mapping result and the distribution of the standard geological feature sample. In a specific implementation, the feature mapping result is input into the discriminator to obtain a discrimination score of the feature mapping result by the discriminator, and the standard geological feature sample is input into the discriminator to obtain a discrimination score of the standard sample by the discriminator. The difference between the two sets of discrimination scores is the distribution adversarial loss, and the smaller the loss value is, the closer the distribution of the feature mapping result is to the distribution of the standard geological feature sample.

[0092] The feature similarity loss is calculated between the feature mapping results of the sample points with a spatial position distance below a preset distance threshold. In a geological environment, sample points with close spatial positions usually have similar geological characteristics. The preset distance threshold can be adaptively set based on a specific scene. To maintain this spatial continuity, the feature similarity loss needs to be calculated. For any two sample points with close spatial positions, their corresponding feature mapping results are extracted, and the Euclidean distance or cosine similarity between the two feature mapping results is calculated. The smaller the distance is or the higher the similarity is, the better the feature mapping maintains the spatial continuity of the original geological features. The feature similarity loss is defined as the weighted sum of the similarity indicators of all pairs of similar sample points.

[0093] The distribution adversarial loss and the feature similarity loss are combined to obtain an optimization objective of the feature mapping parameters. The two losses reflect different aspects of the feature mapping: the distribution adversarial loss ensures that the feature mapping result conforms to the distribution characteristics of the standard geological features, and the feature similarity loss ensures that the feature mapping maintains the spatial continuity of the original geological features. When combining the two losses, a weight coefficient is introduced to control the influence degree of each loss. Generally, the weight of the distribution adversarial loss is larger, and the weight of the feature similarity loss is smaller, to prioritize ensuring that the feature mapping result conforms to the standard distribution.

[0094] The feature mapping parameters of the encoder and the discriminator parameters are alternately optimized based on the optimization objective. The optimization process adopts an alternating iteration strategy: in the first step, the discriminator parameters are fixed, and the encoder parameters are updated to make the feature mapping result closer to the standard geological feature distribution; in the second step, the encoder parameters are fixed, and the discriminator parameters are updated to improve the ability of the discriminator to distinguish between the feature mapping result and the standard sample. The two steps are alternately performed until the difference between the feature mapping result and the distribution of the standard geological feature converges below a preset threshold. The converged feature mapping result is the standardized geological feature, which retains the key information of the original geological data and has a more standardized expression form.

[0095] The application realizes the standardized processing of gold mine geological features through the adversarial network, solves the problem of inconsistent expression forms of geological data in different mining areas, introduces the distribution adversarial loss to make the standardized geological features conform to the typical gold mine metallogenic regularity, and improves the accuracy of subsequent analysis; the feature similarity loss ensures the spatial continuity of the geological features in the standardization process, avoiding information rupture; the alternating optimization strategy effectively balances the relationship between standardization and information preservation, so that the standardized geological features not only conform to the standard distribution but also retain the key information of the original geological features; the standardized processing provides a high-quality data basis for subsequent gold mine metallogenic prediction, resource assessment and mining planning.

[0096] In the embodiment, the standardized geological features are divided into multiple time windows, tectonic stress field data are collected, the principal stress direction and stress gradient are extracted, and a stress field evolution sequence is constructed, including:

[0097] The weighted sum of the displacement amount and the tectonic importance weight of each tectonic unit in the standardized geological features is calculated to obtain a tectonic activity intensity index;

[0098] The division points are determined according to the time variation rate of the tectonic activity intensity index, and the standardized geological features are divided into multiple time windows;

[0099] In each time window, tectonic stress field data are collected, and the strain tensor is calculated according to the tectonic stress field data and converted into a stress tensor;

[0100] The principal stress direction is extracted by eigenvalue decomposition of the stress tensor, the spatial variation of the stress tensor is calculated to obtain the stress gradient, the principal stress direction and the stress gradient of adjacent time windows are processed by a time weight factor, and a stress field evolution sequence is constructed.

[0101] In the process of analyzing gold mine geological information, after obtaining the standardized geological features, the weighted sum of the displacement amount and the tectonic importance weight of each tectonic unit in the standardized geological features is calculated to obtain a tectonic activity intensity index. The tectonic unit includes geological structures such as faults, folds and joints, and the displacement amount refers to the position change amount of the tectonic unit in the process of geological evolution, which is calculated by the position difference of the tectonic unit in adjacent geological periods. The tectonic importance weight is an index for measuring the influence degree of different tectonic units on gold mineralization, which is obtained by training a deep learning model based on historical gold mineralization data. The morphological features of the tectonic unit are extracted by using a convolutional neural network, and the importance weight of different tectonic units is allocated by using an attention mechanism. In the calculation process, the displacement amount and the importance weight of the corresponding tectonic unit are multiplied, and the sum of the product results of all tectonic units is obtained to obtain the tectonic activity intensity index, which reflects the overall intensity level of tectonic activity in the study area.

[0102] The demarcation point is determined according to the time variation rate of the tectonic activity intensity index, and the standardized geological features are divided into multiple time windows. The time variation rate refers to the change speed of the tectonic activity intensity index with time, which is calculated by the difference of the tectonic activity intensity index between adjacent time points divided by the time interval. The demarcation point refers to the time when the time variation rate changes significantly, and whether the variation rate reaches a significant level is determined by setting a threshold. When the variation rate exceeds the threshold, the time point is marked as the demarcation point. The time variation rate of the tectonic activity intensity index is calculated by using the sliding window method, and the window size is determined according to the time resolution of the geological data. The demarcation point position is determined by judging whether the variation rate exceeds the preset threshold, and the standardized geological features are divided into multiple time windows, each of which represents a relatively stable tectonic activity stage.

[0103] In multiple time windows, tectonic stress field data is collected, strain tensors are calculated according to the tectonic stress field data, and stress tensors are converted from the strain tensors. The tectonic stress field data refers to a data set describing the stress distribution state in the study area, including stress size, direction and distribution characteristics and other information. The collection method includes geological structure inversion, borehole stress measurement and rock mechanics test, etc. The strain tensor is a mathematical expression describing the deformation state of the rock mass, which is calculated by the tectonic deformation characteristics. In each time window, based on the underground medium deformation data obtained by ground penetrating radar and seismic wave detection, a deep learning model is used to extract deformation characteristics, a three-dimensional deformation field is constructed, and a strain tensor is calculated. When the strain tensor is converted into the stress tensor, the mechanical properties of the rock need to be considered, including the elastic modulus, Poisson's ratio and other parameters, which are obtained by experimental measurement or empirical estimation. The conversion process uses an improved Hooke's law to establish a nonlinear mapping relationship through a deep neural network, to realize the conversion from strain to stress, and to obtain the stress tensor in each time window.

[0104] The principal stress direction is extracted by eigenvalue decomposition of the stress tensor, the stress gradient is calculated by the spatial variation of the stress tensor, and the stress field evolution sequence is constructed by processing the principal stress direction and stress gradient of adjacent time windows with a time weight factor. Eigenvalue decomposition is a mathematical method for extracting the main characteristics of the stress tensor, and the principal stress size and direction are obtained by decomposition. The stress tensor in each time window is subjected to eigenvalue decomposition to obtain three principal stress directions, corresponding to the maximum principal stress, intermediate principal stress, and minimum principal stress directions. The stress gradient refers to the rate of change of stress in space, which is obtained by calculating the stress difference between adjacent spatial positions. The spatial difference method is used to calculate the rate of change of the stress tensor in each direction in three-dimensional space to generate the stress gradient field. The time weight factor is a weight coefficient for processing adjacent time window data, which is used to smooth the time series data and reduce noise. The value of the weight factor is related to the time interval and data reliability, and is adaptively adjusted by a deep learning model. In the processing process, the time weight factor is applied to the principal stress direction and stress gradient of adjacent time windows to achieve smooth transition of the data and construct a continuous stress field evolution sequence. This sequence records the variation of the stress field in the study area over time, providing an important basis for gold ore-forming environment analysis.

[0105] On the basis of the stress field evolution sequence, a long short-term memory network is used to analyze the stress field evolution characteristics and identify the stress environment conducive to gold mineralization. The network structure includes an input layer, multiple long short-term memory layers, and an output layer. The input is the stress field evolution sequence, and the output is the mineralization environment evaluation result. The training data comes from the stress field evolution data of known gold mine areas, and the network parameters are trained through supervised learning. After training, the stress field evolution sequence of the area to be evaluated is input into the network to obtain the evaluation result of the mineralization environment of the area, including information such as mineralization possibility, potential ore body location and size.

[0106] By calculating the tectonic activity intensity index, the key periods in the geological evolution process of gold mines are accurately identified; based on the time variation rate of the tectonic activity intensity index, the time window is divided, improving the time resolution of geological evolution analysis; combined with stress tensor analysis and eigenvalue decomposition, the stress environment variation law in the gold mineralization process is revealed; the time weight factor processing technology is introduced to reduce noise interference in geological data and enhance the continuity and reliability of the stress field evolution sequence; the deep learning method is used to analyze the stress field evolution characteristics, realizing intelligent identification and evaluation of the gold mineralization environment.

[0107] In this embodiment, the tectonic activity intensity index is calculated by calculating the weighted sum of the displacement of each tectonic unit in the standardized geological features and the tectonic importance weight, including:

[0108] The standardized geological feature data is projected to the principal axis direction of the tectonic strain ellipsoid, stress components of the tectonic unit in each principal axis direction are obtained, and the displacement amount of the tectonic unit is calculated based on the stress components;

[0109] The spatial geometric attributes and active period parameters of the tectonic unit are extracted, a mapping relationship between the tectonic unit attributes and the active period is established through a stress field response function, and a tectonic importance weight is generated;

[0110] The displacement amount is coupled with the tectonic importance weight in a tensor space to construct a regional tectonic activity intensity tensor;

[0111] According to the eigenvalue space distribution law of the regional tectonic activity intensity tensor, the weighted sum of the displacement amount and the tectonic importance weight of each tectonic unit is calculated to obtain a tectonic activity intensity index.

[0112] In the process of analyzing gold mine geological information, after obtaining standardized geological feature data, it needs to be projected to the principal axis direction of the tectonic strain ellipsoid. The tectonic strain ellipsoid is a geometric model for describing the deformation state of geological structure, which has three mutually perpendicular principal axes, representing the maximum elongation direction, the intermediate elongation direction and the minimum elongation direction. The projection process first determines the principal axis direction of the strain ellipsoid through principal component analysis method, and then projects the standardized geological feature data to the three principal axis directions through coordinate transformation. The projection operation is realized by using a deep neural network, which contains three fully connected layers, the input is the standardized geological feature data, and the output is the projection value in each principal axis direction. Through projection, the stress components of the tectonic unit in each principal axis direction are obtained, which represent the stress size of the tectonic unit in the corresponding direction. When calculating the displacement amount of the tectonic unit based on the stress component, an improved Hooke's law is used to establish the relationship model between stress and displacement. The model considers parameters such as the elastic modulus and Poisson's ratio of rock, and through deep learning network training, a nonlinear mapping function is obtained to convert the stress component into displacement amount. When calculating the displacement amount, the displacement components in the three principal axis directions are vector synthesized to obtain the total displacement amount of the tectonic unit.

[0113] The spatial geometric properties of the tectonic units and the activity period parameters are extracted. The spatial geometric properties include the length, width, depth, strike, dip angle, and other geometric characteristics of the tectonic units, which are obtained through geological exploration data. The activity period parameters include the activity frequency, duration, intensity, and other time characteristics of the tectonic units, which are obtained through geochronology analysis. The extraction process uses a convolutional neural network to process three-dimensional geological data. The network structure includes multiple convolutional layers, pooling layers, and fully connected layers, which are specifically designed to extract geological structure features. The mapping relationship between the tectonic unit properties and the activity period is established through a stress field response function, which describes the response behavior of the tectonic units under different stress conditions. The stress field response function is implemented using a recurrent neural network, with the spatial geometric properties of the tectonic units and the activity period parameters as inputs, and the response strength of the tectonic units to the regional stress field as outputs. According to the response strength, the tectonic importance weight is generated, and the higher the weight value, the greater the impact of the tectonic unit on gold mineralization. The weight generation uses normalization processing to ensure the comparability of the weights of different tectonic units.

[0114] The displacement amount is coupled with the tectonic importance weight in the tensor space to construct the regional tectonic activity intensity tensor. The tensor space is a mathematical space that describes multi-dimensional physical quantities, which is used to represent the multi-dimensional characteristics of tectonic activity in this method. In the coupling operation process, the displacement amount is first represented as a three-dimensional tensor form, with each element corresponding to the displacement vector of a point in space. The tectonic importance weight is represented as a weight tensor, which has the same spatial dimension as the displacement tensor. The coupling operation is implemented using tensor multiplication, and the optimal tensor multiplication parameters are learned through a deep learning model. In specific implementation, a tensor network layer is designed, which receives the displacement tensor and the weight tensor as input and outputs the coupled tectonic activity intensity tensor. The tensor network layer includes a tensor decomposition module and a tensor reconstruction module. The former decomposes the input tensor into a low-dimensional representation, and the latter reconstructs the low-dimensional representation into the target tensor. In this way, a complete regional tectonic activity intensity tensor is constructed, which reflects the spatial distribution and intensity variation of tectonic activity in the study area.

[0115] According to the eigenvalue space distribution rule of the regional tectonic activity intensity tensor, the weighted sum of the displacement amount and the tectonic importance weight of each tectonic unit is calculated to obtain the tectonic activity intensity index. The eigenvalue space distribution rule refers to the distribution mode of the eigenvalue of the tectonic activity intensity tensor in space, which is obtained by tensor decomposition. The tensor decomposition adopts a high-order singular value decomposition method to decompose the tectonic activity intensity tensor into the product of a core tensor and a factor matrix. The decomposition process is optimized by a deep learning model, and the model adopts an alternating minimization strategy to train parameters. According to the decomposition result, the eigenvalue space distribution rule is extracted, and the spatial clustering analysis of the eigenvalue is performed to identify the main distribution mode of the eigenvalue. Based on the identified distribution rule, a weighted function is designed to calculate the weighted sum of the displacement amount and the tectonic importance weight of each tectonic unit. The weighted function considers factors such as the size, spatial distribution density and variation trend of the eigenvalue, and the optimal parameters are obtained through a deep learning model. The final tectonic activity intensity index is a scalar value, which reflects the overall intensity level of tectonic activity in the study area and provides a basis for subsequent time window division.

[0116] After the calculation of the tectonic activity intensity index, the time sequence deep learning model is further used to analyze the variation characteristics of the index over time. The model adopts a long short-term memory network structure and can capture the long-term dependence relationship and short-term variation mode of the tectonic activity intensity index. Through model analysis, the key period and variation trend of tectonic activity are identified, providing time dimension support for gold ore-forming environment analysis.

[0117] By projecting the standardized geological features to the principal axis direction of the tectonic strain ellipsoid, the stress state of the tectonic unit is accurately characterized; based on the mapping relationship between the tectonic unit attributes and the activity period established by the stress field response function, the accuracy of the tectonic importance weight calculation is improved; the regional tectonic activity intensity tensor constructed by tensor space coupling operation comprehensively reflects the spatial distribution characteristics of tectonic activity; the tectonic activity intensity index calculated according to the eigenvalue space distribution rule provides a reliable index for gold ore-forming environment analysis; the overall method combines deep learning technology and geological structure theory to realize intelligent analysis of gold geological information, and improves the accuracy and efficiency of gold exploration.

[0118] In this embodiment, the response characteristics of the standardized geological features to the stress field evolution sequence in each time window are calculated, and the stress field geological feature correlation strength is determined based on the response characteristics, which includes:

[0119] The standardized geological features are divided into multiple tectonic units in space, and the stress field evolution sequence corresponding to each tectonic unit is calculated in each time window to obtain the tectonic unit stress field variation;

[0120] The response characteristics are calculated according to the tectonic unit stress field variation and the duration of the time window;

[0121] The response feature is decomposed to obtain a maximum amplitude of the response feature and a time interval required to reach the maximum amplitude, and a response feature coefficient is calculated based on a ratio of the maximum amplitude to the time interval;

[0122] According to the time sequence variation law of the response feature coefficient and the spatial distribution characteristics of the structural unit, the stress field geological feature correlation strength is calculated.

[0123] The embodiment first divides the standardized geological features into multiple structural units in space. The structural unit refers to a spatial region with similar geological characteristics, including geological structures such as faults, folds, and joint zones. The division process uses a hierarchical clustering algorithm, taking the physical properties and geometric characteristics of the geological body as the clustering basis. The physical property parameters in the standardized geological features are extracted, including lithology, density, elastic modulus, etc.; the geometric characteristic parameters are extracted, including spatial position, shape, strike and dip angle, etc.; these parameters are input into a depth clustering model composed of an autoencoder and a clustering layer. The autoencoder is responsible for feature dimension reduction and extraction, and the clustering layer performs spatial division according to the reduced features. The model training uses an unsupervised learning method, which jointly optimizes the network parameters by minimizing the reconstruction error and clustering loss. After training, the trained model is applied to the standardized geological features, and the space is divided into multiple structural units, each unit having relatively independent geological characteristics.

[0124] The stress field evolution sequence corresponding to each structural unit is calculated in each time window to obtain the structural unit stress field change amount. The time window is a time period determined according to the time variation rate of the tectonic activity intensity index, and each window represents a relatively stable tectonic activity stage. The stress field evolution sequence is a data sequence describing the change of the stress field with time, containing the principal stress direction and stress gradient information. In the calculation process, for each structural unit, the stress field data in each time window is extracted, and the data is processed by time series analysis method to generate the stress field evolution sequence. The convolutional long short-term memory network is used to extract the spatio-temporal features, which combines the spatial feature extraction capability of convolutional neural network and the time series processing capability of long short-term memory network. The network input is the stress field data of the structural unit at different time points, and the output is the stress field evolution sequence. By calculating the stress field difference value between adjacent time points, the structural unit stress field change amount is obtained, which reflects the response degree of the structural unit to the stress field change.

[0125] The stress field change amount of the tectonic unit is first calculated with the duration of the time window to obtain a response rate, and the response rate is multiplied by a tectonic importance weight to obtain a response feature. The response rate refers to the ratio of the stress field change amount of the tectonic unit to the duration of the time window, and represents the response speed of the tectonic unit to the stress field change. When calculating, the stress field change amount of the tectonic unit is divided by the duration of the time window to obtain the response rate. The tectonic importance weight is an index for measuring the influence degree of the tectonic unit on gold mineralization, and is obtained by the foregoing method. The response rate is multiplied by the tectonic importance weight to obtain the response feature. The response feature comprehensively considers the response speed and importance of the tectonic unit, and can more comprehensively reflect the role of the tectonic unit in the gold mineralization process.

[0126] The response feature is decomposed to obtain a maximum amplitude of the response feature and a time interval required to reach the maximum amplitude, and a response feature coefficient is calculated based on the ratio of the maximum amplitude to the time interval. The response feature decomposition uses a wavelet transform method to analyze the response feature in both time and frequency domains. A discrete wavelet transform is used to perform multi-scale decomposition on the response feature to extract information of different frequency components. The wavelet transform parameters are optimized by a deep learning model to improve the decomposition accuracy. The maximum amplitude of the response feature, i.e., the maximum value of the response feature in the time series, is extracted from the decomposition result, and the time interval required to reach the maximum amplitude from the initial state is also recorded. The maximum amplitude is divided by the time interval to obtain the response feature coefficient. The coefficient reflects the efficiency of the tectonic unit in responding to stress changes, and the higher the coefficient, the more sensitive the tectonic unit is to stress changes, and the stronger the correlation with gold mineralization activities.

[0127] According to the time series variation law of the response feature coefficient and the spatial distribution characteristics of the tectonic unit, the stress field geological feature correlation strength is calculated. The time series variation law refers to the change pattern of the response feature coefficient with time, which is extracted by time series analysis method. A long short-term memory network is used to analyze the time series data of the response feature coefficient to extract periodicity, trend and suddenness, etc. The spatial distribution characteristics of the tectonic unit refer to the positional relationship and spatial arrangement pattern of the tectonic unit in the study area, which are obtained by spatial statistical method analysis. A graph convolutional neural network is used to establish a spatial relationship model between the tectonic units to extract the spatial distribution characteristics. The time series variation law and the spatial distribution characteristics are combined, and the spatio-temporal information is fused through an attention mechanism to calculate the stress field geological feature correlation strength. The correlation strength represents the interaction degree between the stress field evolution and the geological features, and is an important index for evaluating the gold mineralization potential.

[0128] After the calculation of the stress field geological feature correlation strength, the spatial distribution pattern of the correlation strength is further analyzed by a deep learning model to identify high correlation strength areas, which are usually favorable positions for gold mineralization. The model uses a three-dimensional convolutional neural network structure to capture the three-dimensional spatial distribution characteristics of the correlation strength and output gold mineralization prediction results.

[0129] By dividing the standardized geological features into multiple structural units, a fine description of complex geological bodies is achieved. The stress field variation of the structural units is calculated to reveal the response characteristics of the structural units to the stress field variation, and the response efficiency of the structure is accurately quantified. The correlation model between the stress field and the geological features is established by combining the time series variation and spatial distribution characteristics of the response feature coefficients. The overall method integrates deep learning technology and geological structure theory to realize intelligent analysis of gold geological information, which can more comprehensively capture the interaction characteristics of geological structure and stress field, reveal the control factors of gold mineralization, reduce exploration risk, and improve resource utilization.

[0130] In this embodiment, the stress field geological feature correlation strength is input into the prediction function, the adaptive weight is used to calculate the mineralization probability score of each time window, the mineralization probability distribution map is constructed, and the gold mineralization prediction results of the target mining area based on the mineralization probability distribution map include:

[0131] The feature correlation degree of adjacent time is calculated to obtain the time sequence weight, the time sequence weight and the kernel function are combined to construct the prediction function based on the feature decomposition kernel function;

[0132] The stress field geological feature correlation strength is input into the prediction function, and the main feature and its dynamic evolution characteristics are obtained by feature decomposition;

[0133] The adaptive weight is calculated based on the directional change of the main feature in adjacent time windows;

[0134] The main feature and the adaptive weight are combined in the prediction function to obtain the mineralization probability score of each time window, and the mineralization probability distribution map is constructed by probability superposition and spatial interpolation of the mineralization probability score;

[0135] The probability gradient and probability aggregation area in the mineralization probability distribution map are extracted, and the gold mineralization prediction results of the target mining area are determined based on the change trend of the probability gradient and the spatial distribution of the probability aggregation area.

[0136] In the process of gold deposit geological information analysis, the feature correlation degree between adjacent time points is calculated to obtain the time sequence weight. The feature correlation degree refers to the similarity between geological features at different time points, reflecting the continuity of geological evolution. In the calculation process, the stress field geological feature correlation strength data of adjacent time windows is extracted, and the feature correlation degree is calculated through the depth correlation analysis network. The network adopts a two-way structure, and the feature data of adjacent time points are input into the two ways respectively. The feature representation is extracted through multi-layer convolution and self-attention mechanism, and then the cosine similarity between the two features is calculated as the feature correlation degree. The higher the feature correlation degree, the more continuous the geological evolution, and the greater the corresponding time sequence weight. The time sequence weight is normalized to ensure that the weight sum is one. Based on feature decomposition, the kernel function is constructed. Feature decomposition is to decompose high-dimensional feature data into a linear combination of principal components through principal component analysis. The kernel function is a nonlinear function defined in the feature space, which is used to capture the internal structure of the feature data. In the construction process, the stress field geological feature correlation strength data is decomposed into principal components, and the radial basis kernel function is constructed based on the principal components. This function can effectively express the nonlinear relationship in the feature space. The time sequence weight and the kernel function are combined to construct the prediction function. The combination method is to multiply the output result of the kernel function by the corresponding time sequence weight, and then sum to obtain the prediction function value. The prediction function realizes the modeling of the time evolution of geological features, providing a basis for ore-forming probability prediction.

[0137] The stress field geological feature correlation strength is input into the prediction function, and the main features and their dynamic evolution characteristics are obtained through feature decomposition. The stress field geological feature correlation strength is an index calculated in the previous step, reflecting the degree of interaction between the stress field and the geological features. The index is input into the prediction function to obtain the corresponding function output value. Feature decomposition uses the singular value decomposition method to decompose the output result of the prediction function into a combination of a series of feature vectors and eigenvalues. The eigenvalue size represents the importance of the corresponding feature vector, and the largest eigenvalues are selected as the main features. The main features reflect the most critical geological factors in the gold deposit forming process. The dynamic evolution characteristics refer to the variation of the main features over time, which are obtained by analyzing the differences in the main features of different time windows. A recurrent neural network is designed to extract the time series pattern of the main features, and the network structure contains long short-term memory units that can capture long-term dependencies and output the dynamic evolution characteristics of the main features.

[0138] The adaptive weight is calculated based on the direction change of the principal feature in adjacent time windows. The direction change refers to the degree of change in the direction of the principal feature vector in the feature space, which is obtained by calculating the angle between the principal feature vectors of adjacent time windows. The smaller the angle, the more gentle the change in the principal feature, and the larger the corresponding adaptive weight; the larger the angle, the more dramatic the change in the principal feature, and the smaller the corresponding adaptive weight. During the calculation process, an attention mechanism is used to automatically adjust the weight distribution. This mechanism dynamically allocates weight values by learning the correlation degree of the principal features of adjacent time windows. The implementation of the attention mechanism uses a multi-head self-attention network, which can capture the dependency between different positions and improve the accuracy of weight calculation. The introduction of adaptive weight enables the prediction model to automatically adjust the prediction strategy according to the stability of geological evolution, improving the reliability of the prediction results.

[0139] The principal feature and the adaptive weight are combined in the prediction function to obtain the ore-forming probability score of each time window. The combination operation is to multiply the principal feature vector by the corresponding adaptive weight, and then calculate the ore-forming probability score through the prediction function. During the calculation process, a deep neural network is designed to realize the nonlinear combination of features and weights. This network contains multiple fully connected layers and activation functions, and can capture the complex relationship between features and weights. The ore-forming probability score is a value between zero and one, representing the likelihood of each location in the study area becoming a gold mine. The ore-forming probability score is subjected to probability stacking and spatial interpolation to construct the ore-forming probability distribution map. Probability stacking is to combine the ore-forming probability scores of different time windows according to certain rules to form a comprehensive evaluation result. Spatial interpolation is to estimate the value of unknown points between discrete sampling points to generate a continuous spatial distribution. The Kriging interpolation method is used in combination with deep learning technology. This method takes into account the spatial autocorrelation and can generate a smooth and continuous probability distribution map. The ore-forming probability distribution map is displayed in the form of a heat map, with color depth representing the level of ore-forming probability, providing an intuitive reference for gold mine exploration.

[0140] This study extracts probability gradients and probability clustering regions from the mineralization probability distribution map. Based on the changing trends of the probability gradients and the spatial distribution of these clustering regions, it determines the gold mineralization prediction results for the target mining area. The probability gradient refers to the rate of change of the mineralization probability in space, obtained by calculating the probability differences between adjacent locations. The probability gradient calculation is implemented using a convolutional neural network, which extracts the changing characteristics of the probability distribution through gradient operators. Probability clustering regions refer to spatially continuous areas with high mineralization probabilities, identified using clustering algorithms. The density peak clustering method is employed, which can automatically identify irregularly shaped clusters and is suitable for identifying clustering regions in complex geological environments. The changing trends of the probability gradients are analyzed to determine mineralization boundaries and directions; areas with steep gradient changes are typically mineralization boundaries, and areas pointed to by the gradient direction may be potential orebody centers. The spatial distribution of probability clustering regions is used to analyze the scale and morphology of the mineralization; the area and shape of the clustering regions reflect the scale and morphological characteristics of potential orebodies. Considering the characteristics of both the probability gradients and probability clustering regions, a deep learning model is used for comprehensive evaluation, outputting the gold mineralization prediction results for the target mining area, including information on mineralization location, scale, morphology, and grade predictions.

[0141] By calculating the temporal weights of the correlation between features at adjacent time points, an accurate description of the continuity of geological evolution is achieved. A high-precision mineralization prediction model is established based on the combination of kernel functions constructed from eigenvalue decomposition and temporal weights. The introduction of principal features and their dynamic evolutionary characteristics reveals key geological factors and their changing patterns in the gold mineralization process. Adaptive weights calculated based on changes in the direction of principal features improve the predictive model's adaptability to geological instability. A mineralization probability distribution map constructed through probability superposition and spatial interpolation visually displays the spatial distribution of gold mineralization potential. Combining probability gradient and probability clustering region analysis, accurate predictions of the location, scale, and morphology of gold deposits are achieved. The overall method integrates deep learning technology and geological theory, realizing intelligent analysis of the entire process from geological data to mineralization prediction, significantly improving the efficiency and success rate of gold exploration, reducing exploration costs, and providing a scientific basis for the efficient development and utilization of gold resources.

[0142] like Figure 3 As shown, Figure 3 A schematic diagram of a deep learning-based gold mine geological information analysis system provided in this embodiment of the invention, the system comprising:

[0143] The geological feature extraction module is used to collect geological data of the target mining area, extract spatial structure features and temporal evolution features, and fuse them to generate complete geological features;

[0144] The feature standardization module is configured to map complete geological features through a pre-trained adversarial network, iteratively optimize feature mapping parameters based on a feature similarity loss function, and generate standardized geological features.

[0145] The stress field construction module is configured to divide the standardized geological features into multiple time windows, collect tectonic stress field data, extract a principal stress direction and a stress gradient, and construct a stress field evolution sequence.

[0146] The correlation analysis module is configured to calculate response features of the standardized geological features to the stress field evolution sequence in each time window, determine stress field-geological feature correlation strength based on the response features.

[0147] The ore-forming prediction module is configured to input the stress field-geological feature correlation strength into a prediction function, calculate an ore-forming probability score of each time window using adaptive weights, construct an ore-forming probability distribution map, and determine a gold ore-forming prediction result of a target mining area based on the ore-forming probability distribution map.

[0148] A technical solution provided in the embodiments of the present application is an electronic device, which comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the steps in any of the foregoing methods when executing the computer program.

[0149] A technical solution provided in the embodiments of the present application is a computer readable storage medium, which stores a computer program, and the processor implements the steps in any of the foregoing methods when executing the computer program.

[0150] The above-described specific embodiments are preferred embodiments of the present application, and the specific implementation range of the present application is not limited by the above-described specific embodiments, and the scope of the present application includes but is not limited to the above-described specific embodiments, and equivalent changes made according to the shape and structure of the present application are within the protection scope of the present application.

Claims

1. A deep learning-based gold mine geological information analysis method, characterized in that, The method comprises the following steps: Collecting geological data of a target mining area, extracting spatial structure features and time sequence evolution features and fusing them to generate complete geological features; Mapping the complete geological features through a pre-trained adversarial network, iteratively optimizing feature mapping parameters based on a feature similarity loss function, and generating standardized geological features; Dividing the standardized geological features into multiple time windows, collecting tectonic stress field data, extracting principal stress directions and stress gradients to construct a stress field evolution sequence; Calculating the response features of the standardized geological features to the stress field evolution sequence in each time window, and determining the stress field geological feature correlation strength based on the response features; Inputting the stress field geological feature correlation strength into a prediction function, calculating the ore-forming probability score of each time window using adaptive weights, constructing an ore-forming probability distribution map, and determining the gold ore-forming prediction result of the target mining area based on the ore-forming probability distribution map.

2. The method of claim 1, wherein, Collecting geological data of a target mining area, extracting spatial structure features and time sequence evolution features and fusing them to generate complete geological features includes: Collecting geological data of a target mining area, performing depth domain correction and phase extraction on the target mining area geological data to generate preprocessed geological data; Extracting geological body attributes based on the preprocessed geological data, introducing a connectivity judgment condition to identify the distribution of structural fractures, and combining spatial positioning information to calculate stress field distribution to obtain spatial structure features; Performing time sequence alignment on the preprocessed geological data, calculating the deformation amount of the geological body and the stress change amount at adjacent time points, determining the stress-strain relationship according to the deformation amount of the geological body and the stress change amount, and obtaining the time sequence evolution features; Constructing a feature tensor from the spatial structure features and the time sequence evolution features, introducing an evolution continuity constraint based on the stress-strain relationship in tensor decomposition to obtain fused features, and reconstructing complete geological features that satisfy the mechanical equilibrium constraint according to the fused features.

3. The method of claim 1, wherein, Mapping the complete geological features through a pre-trained adversarial network, iteratively optimizing feature mapping parameters based on a feature similarity loss function, and generating standardized geological features includes: Constructing an adversarial network containing an encoder and a discriminator, the encoder performing feature mapping on the complete geological features, and the discriminator determining the distribution of the feature mapping results; Pre-training the adversarial network using standard geological feature samples to obtain a pre-trained adversarial network; Inputting the complete geological features into the pre-trained adversarial network for feature mapping to obtain feature mapping results; Calculating the distribution adversarial loss of the feature mapping results and the standard geological feature samples; Calculating the similarity between the feature mapping results corresponding to sample points with a spatial position distance below a preset distance threshold as the feature similarity loss; Combining the distribution adversarial loss and the feature similarity loss to obtain the optimization target of the feature mapping parameters; Based on the optimization target, alternately optimizing the feature mapping parameters of the encoder and the discriminator parameters until the distribution difference between the feature mapping results and the standard geological features converges, and taking the converged feature mapping results as the standardized geological features.

4. The method of claim 1, wherein, Dividing the standardized geological features into multiple time windows, collecting tectonic stress field data, extracting principal stress directions and stress gradients to construct a stress field evolution sequence includes: The tectonic activity intensity index is obtained by calculating the weighted sum of the displacement amount of each tectonic unit in the standardized geological feature and the tectonic importance weight; The boundary points are determined according to the time variation rate of the tectonic activity intensity index, and the standardized geological feature is divided into multiple time windows; In each time window, the tectonic stress field data is collected, the strain tensor is calculated according to the tectonic stress field data, and the stress tensor is converted; The principal stress direction is extracted by performing eigenvalue decomposition on the stress tensor, the spatial variation of the stress tensor is calculated to obtain the stress gradient, the principal stress direction and the stress gradient of adjacent time windows are processed by using a time weight factor, and a stress field evolution sequence is constructed.

5. The method of claim 4, wherein, The tectonic activity intensity index is obtained by calculating the weighted sum of the displacement amount of each tectonic unit in the standardized geological feature and the tectonic importance weight includes: The stress components of the tectonic unit in each principal axis direction are obtained by projecting the standardized geological feature data onto the principal axis direction of the tectonic strain ellipsoid, and the displacement amount of the tectonic unit is calculated based on the stress components; The spatial geometric attributes and activity period parameters of the tectonic unit are extracted, the mapping relationship between the tectonic unit attributes and the activity period is established through a stress field response function, and the tectonic importance weight is generated; The displacement amount in the tensor space is coupled with the tectonic importance weight to construct a regional tectonic activity intensity tensor; According to the spatial distribution law of the characteristic value of the regional tectonic activity intensity tensor, the weighted sum of the displacement amount of each tectonic unit and the tectonic importance weight is calculated to obtain the tectonic activity intensity index.

6. The method of claim 1, wherein, The response characteristics of the standardized geological feature in each time window to the stress field evolution sequence are calculated, and the stress field geological feature correlation strength is determined based on the response characteristics includes: The standardized geological feature is divided into multiple tectonic units in space, and the stress field evolution sequence corresponding to each tectonic unit is calculated in each time window to obtain the tectonic unit stress field variation; The response characteristics are decomposed to obtain the maximum amplitude of the response characteristics and the time interval required to reach the maximum amplitude, and the response characteristic coefficient is calculated based on the ratio of the maximum amplitude to the time interval; According to the time sequence variation law of the response characteristic coefficient and the spatial distribution characteristics of the tectonic unit, the stress field geological feature correlation strength is calculated. The stress field geological feature correlation strength is input into the prediction function, the adaptive weight is used to calculate the ore-forming probability score of each time window, the ore-forming probability distribution map is constructed, and the gold ore-forming prediction result of the target mining area is determined based on the ore-forming probability distribution map includes:

7. The method of claim 1, wherein, The feature correlation degree of adjacent time points is calculated to obtain the time sequence weight, the prediction function is constructed by combining the time sequence weight and the kernel function based on the feature decomposition tectonic kernel function; The stress field geological feature correlation strength is input into the prediction function, and the main feature and its dynamic evolution characteristics are obtained by feature decomposition; The adaptive weight is calculated based on the direction change of the main feature in adjacent time windows; The main feature and the adaptive weight are combined and operated in the prediction function to obtain the ore-forming probability score of each time window, and the ore-forming probability score is subjected to probability superposition and spatial interpolation to construct the ore-forming probability distribution map. ​ The probability gradient and the probability aggregation area in the metallogenic probability distribution map are extracted, and the gold ore metallogenic prediction result of the target mining area is determined based on the change trend of the probability gradient and the spatial distribution of the probability aggregation area.

8. A deep learning-based gold mine geological information analysis system for implementing the method of any one of claims 1-7. The system comprises: The geological feature extraction module is configured to collect geological data of the target mining area, extract spatial structure features and time sequence evolution features, and fuse the features to generate complete geological features. The feature standardization module is configured to perform feature mapping on the complete geological features by using a pre-trained adversarial network, iteratively optimize feature mapping parameters based on a feature similarity loss function, and generate standardized geological features. The stress field construction module is configured to divide the standardized geological features into a plurality of time windows, collect tectonic stress field data, extract principal stress directions and stress gradients, and construct a stress field evolution sequence. The correlation analysis module is configured to calculate response features of the standardized geological features to the stress field evolution sequence in each time window, and determine stress field-geological feature correlation strength based on the response features. The metallogenic prediction module is configured to input the stress field-geological feature correlation strength into a prediction function, calculate a metallogenic probability score of each time window using adaptive weights, construct a metallogenic probability distribution map, and determine the gold ore metallogenic prediction result of the target mining area based on the metallogenic probability distribution map.

9. An electronic device, comprising: The computer readable storage medium stores computer program instructions, and the computer program instructions are executed by the processor to implement the steps in the method of any one of claims 1 to 7. The computer readable storage medium stores computer program instructions, and the computer program instructions are executed by the processor to implement the steps in the method of any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, ​

Images