A gold ore target area optimization method and device based on elemental geochemical anomalies

Abstract

This application is applicable to the field of resource exploration technology, and provides a method and device for preferentially selecting gold ore target areas based on geochemical anomalies of elements. The method includes: obtaining geochemical sample data of multiple sampling points in the research area; training a converter model using the geochemical sample data to obtain a trained converter model; initializing the model parameters of the trained converter model to obtain a student model; performing self-distillation training on the student model using the geochemical sample data to obtain a geochemical data reconstruction model; inputting the geochemical sample data of the sampling points into the geochemical data reconstruction model for processing to obtain the reconstructed geochemical data of the sampling points; calculating the geochemical anomaly scores of each sampling point based on the reconstructed geochemical data of each sampling point, and determining the gold ore target areas in the research area based on the geochemical anomaly scores of multiple sampling points. This application can meet the need for accurately preferentially selecting target areas in gold ore exploration.

Description

Technical Field

[0001] This application belongs to the technical field of resource exploration, and particularly relates to a method and device for optimizing gold ore target areas based on geochemical anomalies of elements. Background Art

[0002] Using computer technology for geological data processing and geochemical anomaly detection is an important means of finding potential gold ore target areas. However, due to the characteristics of geochemical data such as being sparse, complex, and non-linear, there are many challenges in processing it using computer technology.

[0003] 1. Traditional machine learning methods, such as anomaly detection algorithms based on statistics (including mean square error, probability graph, etc.) and clustering analysis, factor analysis, etc., are mainly implemented in computers based on simple data statistics and basic mathematical model construction. When dealing with geochemical data, a large amount of feature engineering needs to be carried out manually to meet the requirements of model input. In a computer system, in the face of high-dimensional and sparse geochemical data, data reading, calculation, and storage operations are complex and time-consuming, the model learning ability is limited, it is easily interfered by noise, and it is difficult to effectively capture the complex spatial relationships in the data, resulting in low accuracy in optimizing gold ore target areas and being unable to accurately identify the geochemical anomaly characteristics related to gold mineralization and their spatial distribution patterns.

[0004] 2. Although there have been certain advancements in detection models based on deep learning, there are still many problems. For example, convolutional neural networks (CNNs), graph convolutional networks, autoencoders, graph convolutional autoencoders, sparse autoencoders, etc., use their powerful computing power and parallel processing capabilities in computers for automatic feature learning. However, when a CNN processes geochemical data, the setting of its convolution kernel parameters limits the capture of long-range dependence relationships and it is difficult to obtain the correlation information of elements related to gold over a large spatial range. In the computer implementation of autoencoding methods, due to focusing on minimizing the data reconstruction error, the spatial relationships and distribution characteristics of geochemical data are ignored, and it is impossible to make full use of spatial information for optimizing gold ore target areas.

[0005] 3. The Transformer model architecture has performed excellently in fields such as natural language processing, but it faces many difficulties when applied to geochemical anomaly detection and gold ore target area optimization. In a computer system, its training requires a large amount of data, while geochemical sampling data is relatively sparse, resulting in difficult convergence of model training. Moreover, the Transformer model structure is complex, and the computing process has extremely high requirements for computing resources (such as central processing units (CPUs), graphics processing units (GPUs)) and memory. For example, a large number of matrix multiplication operations are involved in calculating multi-head self-attention. This makes the generalization performance of the trained model limited in the scenario of optimizing gold ore target areas and it is difficult to accurately distinguish the areas truly related to gold.

[0006] In summary, when the prior art uses computer technology for the optimal selection of gold ore target areas, there are problems such as insufficient utilization of limited geochemical sampling data, difficult model training, weak generalization ability, and difficulty in effectively capturing spatial relationships, which cannot meet the requirements of accurately selecting target areas in gold ore exploration. Summary of the Invention

[0007] The embodiments of the present application provide a method and device for optimizing gold ore target areas based on elemental geochemical anomalies, which can solve the problem that the traditional gold ore target area optimization method cannot meet the requirements of accurately selecting target areas in gold ore exploration.

[0008] In a first aspect, the embodiments of the present application provide a method for optimizing gold ore target areas based on elemental geochemical anomalies, including:

[0009] Obtain geochemical sample data of multiple sampling points in the research area; the geochemical sample data includes concentration values of multiple elements;

[0010] Use the geochemical sample data of multiple sampling points to train a converter model to obtain a trained converter model; the converter model is used to process the geochemical sample data and output reconstructed geochemical data;

[0011] Initialize the model parameters of the trained converter model to obtain a student model;

[0012] Use the geochemical sample data of multiple sampling points to perform self-distillation training on the student model to obtain a geochemical data reconstruction model;

[0013] Input the geochemical sample data of multiple sampling points into the geochemical data reconstruction model for processing to obtain the reconstructed geochemical data of each sampling point;

[0014] Calculate the geochemical anomaly score of each sampling point based on the reconstructed geochemical data of each sampling point, and determine the gold ore target area in the research area based on the geochemical anomaly scores of multiple sampling points.

[0015] Optionally, obtaining the geochemical sample data of multiple sampling points in the research area includes:

[0016] For each sampling point, use a position encoding function to perform position encoding on the original geochemical sample data of the sampling point, and perform a vector addition operation on the position encoding result and the original geochemical sample data of the sampling point to obtain the geochemical sample data of the sampling point.

[0017] Optionally, the loss function in the process of performing self-distillation training on the student model using the geochemical sample data of multiple sampling points is:

[0018]

[0019] Among them, represents the loss function of the j-th generation model obtained in the j-th iteration of the self-distillation training process, represents the loss weight in the j-th iteration of the self-distillation training process, is related to the number of iterations, represents the number of sampling points, represents the output result of the j-th generation model for under the model parameter ; represents the geochemical sample data of the i-th sampling point, represents the geochemical sample data of other sampling points except the i-th sampling point among multiple sampling points, represents the output result of the (j - 1)-th generation model for under the model parameter ;

[0020] Optionally, the calculation formula of the loss weight is: , , , represents the initial weight.

[0021] Optionally, the reconstructed geochemical data of each sampling point includes the reconstructed concentration values of multiple elements;

[0022] Based on the reconstructed geochemical data of each sampling point, calculate the geochemical anomaly score of each sampling point, including:

[0023] Calculate the geochemical anomaly score of the i-th sampling point through the formula ;

[0024] Among them, represents the number of elements, represents the concentration value of the k-th element of the i-th sampling point, represents the reconstructed concentration value of the k-th element of the i-th sampling point.

[0025] Optionally, based on the geochemical anomaly scores of multiple sampling points, determine the gold ore target area within the study area, including:

[0026] Sort the geochemical anomaly scores of multiple sampling points in descending order;

[0027] Take the area corresponding to the sampling points of the first M geochemical anomaly scores among the sorted multiple geochemical anomaly scores as the gold ore target area within the study area.

[0028] Optionally, after the step of using the geochemical sample data of multiple sampling points to perform self-distillation training on the student model to obtain a geochemical data reconstruction model, the gold ore target area optimization method further includes:

[0029] Calculating the AUC value of the geochemical data reconstruction model by using the geochemical sample data of multiple positive samples and multiple negative samples; the positive samples are known gold ore points, and the negative samples are known non-gold ore points;

[0030] If the AUC value is less than the preset threshold, adjust the model parameters of the geochemical data reconstruction model, and use the geochemical data reconstruction model with adjusted model parameters as the student model, and return to execute the step of using the geochemical sample data of multiple sampling points to perform self-distillation training on the student model to obtain a geochemical data reconstruction model, until the AUC value is greater than or equal to the preset threshold, then enter the step of inputting the geochemical sample data of multiple sampling points into the geochemical data reconstruction model for processing to obtain the reconstructed geochemical data of each sampling point;

[0031] If the AUC value is greater than or equal to the preset threshold, enter the step of inputting the geochemical sample data of multiple sampling points into the geochemical data reconstruction model for processing to obtain the reconstructed geochemical data of each sampling point.

[0032] Optionally, calculating the AUC value of the geochemical data reconstruction model by using the geochemical sample data of multiple positive samples and multiple negative samples includes:

[0033] Using the geochemical data reconstruction model to process the geochemical sample data of each positive sample and each negative sample respectively to obtain the reconstructed geochemical data of each positive sample and each negative sample;

[0034] Obtaining the geochemical anomaly scores of each positive sample and each negative sample based on the reconstructed geochemical data of each positive sample and each negative sample;

[0035] Calculating the AUC value of the geochemical data reconstruction model by using the geochemical anomaly scores of multiple positive samples and multiple negative samples.

[0036] Optionally, the calculation formula for the AUC value of the geochemical data reconstruction model is:

[0037]

[0038] Wherein, represents the AUC value of the geochemical data reconstruction model, represents the number of positive samples, represents the number of negative samples, represents the indicator function, , denotes the geochemical anomaly score of the \(t\)-th positive sample, denotes the geochemical anomaly score of the \(q\)-th negative sample.

[0039] In a second aspect, an embodiment of the present application provides a terminal device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the above-mentioned method for optimizing gold ore target areas based on elemental geochemical anomalies is implemented.

[0040] The above solution of the present application has the following beneficial effects:

[0041] In the embodiment of the present application, based on the trained transformer model, a progressive rebirth mechanism is applied to initialize the model parameters of the trained transformer model to obtain a student model, which prompts the student model to re-learn while retaining some prior knowledge. Since the progressive rebirth mechanism allows for accurate identification of geochemical anomalies related to gold mineralization through continuous iterative optimization under sparse geochemical data conditions, thereby achieving precise optimization of gold ore target areas and overcoming the problem of low accuracy in target area prediction using traditional methods with limited data. Secondly, in terms of knowledge transfer and model performance improvement, through the progressive rebirth mechanism, the model can gradually accumulate and transfer knowledge related to gold ore, effectively enhancing the model's understanding ability of complex geochemical patterns, thus improving the accuracy and reliability of gold ore target area optimization and meeting the requirements for accurately optimizing target areas in gold ore exploration.

[0042] Other beneficial effects of the present application will be described in detail in the subsequent specific implementation section. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0044] Figure 1 is a flowchart of a method for optimizing gold ore target areas based on elemental geochemical anomalies provided by an embodiment of the present application;

[0045] Figure 2 is a change curve diagram of the AUC value of each generation of reborn models in the reborn framework in an example of the present application;

[0046] Figure 3 is an ROC curve diagram of the model in an example of the present application;

[0047] Figure 4Schematic structural diagram of a terminal device provided by an embodiment of the present application. Detailed implementation manners

[0048] In the following description, specific details such as specific system architectures and technologies are presented for the purpose of illustration rather than limitation, so as to thoroughly understand the embodiments of the present application. However, those skilled in the art should clearly understand that the present application can also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to avoid unnecessary details from interfering with the description of the present application.

[0049] It should be understood that when used in the specification of the present application and the appended claims, the term "comprising" indicates the presence of the described features, wholes, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components, and / or their combinations.

[0050] It should also be understood that the term "and / or" used in the specification of the present application and the appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0051] As used in the specification of the present application and the appended claims, the term "if" can be interpreted as "when", "once", "in response to determining", or "in response to detecting" according to the context. Similarly, the phrase "if determined" or "if detecting [the described condition or event]" can be interpreted as meaning "once determined", "in response to determining", "once detecting [the described condition or event]", or "in response to detecting [the described condition or event]" according to the context.

[0052] In addition, in the description of the specification of the present application and the appended claims, the terms "first", "second", "third", etc. are only used for distinguishing descriptions and cannot be understood as indicating or implying relative importance.

[0053] The reference to "an embodiment" or "some embodiments" etc. described in the specification of the present application means that a specific feature, structure, or characteristic described in connection with the embodiment is included in one or more embodiments of the present application. Thus, statements such as "in an embodiment", "in some embodiments", "in other some embodiments", "in still other embodiments", etc. that appear in different places in this specification do not necessarily refer to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in other ways. The terms "comprising", "including", "having", and their variants all mean "including but not limited to", unless otherwise specifically emphasized in other ways.

[0054] Aiming at the problem that the traditional gold ore target area optimization method cannot meet the requirement of accurately optimizing the target area in gold ore exploration, the embodiment of the present application provides a gold ore target area optimization method based on elemental geochemical anomalies. This method initializes the model parameters of the trained transformer model to obtain a student model by applying a progressive rebirth mechanism on the basis of the trained transformer model, prompting the student model to relearn while retaining some previous knowledge. Since the progressive rebirth mechanism allows for accurately identifying geochemical anomalies related to gold mineralization through continuous iterative optimization under sparse geochemical data conditions, and then realizing precise gold ore target area optimization, overcoming the problem of low accuracy in target area prediction using traditional methods with limited data. Secondly, in terms of knowledge transfer and model performance improvement, through the progressive rebirth mechanism, the model can gradually accumulate and transfer knowledge related to gold ore, effectively improving the model's ability to understand complex geochemical patterns, thereby improving the accuracy and reliability of gold ore target area optimization and meeting the requirement of accurately optimizing the target area in gold ore exploration.

[0055] The following provides an exemplary description of the gold ore target area optimization method based on elemental geochemical anomalies provided by the present application in combination with specific embodiments.

[0056] As Figure 1 shown, the gold ore target area optimization method based on elemental geochemical anomalies provided by the embodiment of the present application includes the following steps:

[0057] Step 11: Obtain geochemical sample data of multiple sampling points in the study area.

[0058] The above-mentioned study area is the area where gold ore target area optimization is required. As a preferred example, the study area can be divided into grids, and a sampling point can be set in each grid unit, and the original geochemical sample data of each sampling point is collected. The original geochemical sample data includes concentration values of multiple elements, such as concentration values of elements like gold (Au), arsenic (As), antimony (Sb), mercury (Hg), etc. The geochemical sample data in step 11 above is obtained by embedding position encoding in the original geochemical sample data. It can be understood that the geochemical sample data, like the original geochemical sample data, also includes concentration values of multiple elements.

[0059] Exemplarily, the geochemical sample data of multiple sampling points can be denoted as the geochemical sample data set , , representing the geochemical sample data of the i-th sampling point, including concentration values of multiple elements, , representing the number of sampling points.

[0060] In some embodiments of the present application, the specific implementation manner of the above step 11 may be: for each sampling point, use a position encoding function to perform position encoding on the original geochemical sample data of the sampling point, and perform a vector addition operation on the position encoding result and the original geochemical sample data of the sampling point to obtain the geochemical sample data of the sampling point.

[0061] Specifically, the original geochemical sample data can be preprocessed to meet the input requirements of computer data processing. The data set composed of multiple element concentration values included in each sampling point is organized into a format suitable for reading by a computer model and regarded as an unordered sequence for input into the model. On this basis, position encoding is added to assist the model in understanding the spatial information of the data. In the computer data processing flow, position encoding is crucial for the model to understand the spatial information of geochemical data. The generation of position encoding relies on computer algorithms. Through a learnable relative position encoding function (such as a function based on a multi-layer perceptron (MLP)), the relative position coordinates of geochemical samples are mapped into position encoding and added to the original geochemical sample data. Specifically, calculate the relative position coordinates of each sampling point with respect to the geochemical sample to be reconstructed (i.e., the target sample, and each point will take turns as the target sample), map them through the position encoding function to obtain the position encoding, and after adding this encoding to the original high-dimensional geochemical data (i.e., the original geochemical sample data), the final geochemical sample data (i.e., the geochemical sample data in step 11) is formed. This process utilizes the powerful computing and data processing capabilities of the computer, enhances the model's perception ability of the data spatial relationship, and helps to more accurately identify the spatial characteristics of geochemical anomalies related to gold mineralization in subsequent steps.

[0062] The calculation formula of the position encoding function is:

[0063] .

[0064] Where, represents the position encoding result of the original geochemical sample data (which can be understood as the original geochemical sample data of the current sampling point of the position encoding function) with respect to the geochemical sample to be reconstructed , represents the original geochemical sample data to be reconstructed The spatial coordinates stored in the computer memory, represents the spatial coordinates of the original geochemical sample data of the remaining sampling points in the computer system. is a multi-layer perceptron (MLP) function, and its specific calculation process is as follows:

[0065] When the computer executes this function, first, for the input relative position coordinates , it will be fed into the first linear layer. This linear layer performs a linear transformation on the input data through its transformation matrix and bias , that is, calculate . Then, after passing through the ReLU layer for non-linear activation, we get . Next, it passes through the second linear layer, and uses its transformation matrix and bias to perform a further linear transformation, and finally obtain the function output :

[0066]

[0067] These parameters related to the MLP ( , ) will be trained end-to-end together with the parameters of the transformer model in the computer. During the training process, the computer uses the backpropagation algorithm to optimize and adjust these parameters according to the loss function of the model. Using this feed-forward network, the computer system maps the relative two-dimensional position coordinates into position encodings. To be compatible with the original geochemical data in terms of data dimensions, the position encodings will be mapped to the L dimension. In the computer memory, the high-dimensional geochemical data and the generated position encodings are vector-added to finally form a format suitable for model input.

[0068] After such a learnable relative position encoding process, the computer system can obtain a series of advantages. On the one hand, the learnable function enables the computer to adaptively learn the representation of the target geochemical sample space relationship according to the characteristics of the data in the geochemical anomaly identification task. When the computer processes data, by continuously adjusting the parameters in the function , it can better capture the spatial dependence relationship between samples, thereby enhancing the model's understanding ability of the spatial structure of geochemical data. On the other hand, the relative coordinates generate unique position encodings for each geochemical sample, which provides a diverse sample set for training the end-to-end transformer model using a single geochemical data set in the computer. During the training process, the computer can use these different sample sets to better learn the data features, improve the generalization ability of the model, make it adaptable to different geochemical data distribution situations, and thus show more accurate and stable performance in the geochemical anomaly detection and target area optimization tasks.

[0069] In order to deal with the problem of small data volume, random data masking operation is introduced. Using computer random algorithm, some values ​​in the data are randomly masked. This not only increases the diversity of data and reduces the risk of model overfitting, but also takes advantage of the variable number of elements in the self-attention mechanism, adapts to the requirements of variable-length data input in computer data processing, and prompts the model to learn the potential spatial structure of the data, so as to better discover the abnormal patterns related to gold mines hidden in geochemical data.

[0070] Step 12: training the converter model using geochemical sample data from multiple sampling points to obtain a trained converter model.

[0071] The above transformer model (Transformer) is used to process geochemical sample data and output reconstructed geochemical data. It should be noted that when the transformer model is used to process the geochemical sample data of a certain sampling point (hereinafter referred to as the target geochemical sample data for the convenience of description) and output the reconstructed geochemical data of the sampling point, the geochemical sample data of the sampling point and the geochemical sample data of other sampling points need to be output to the transformer model for processing.

[0072] In some embodiments of the present application, geochemical sample data from multiple sampling points can be used to perform unsupervised training on the transformer model to obtain a trained transformer model. It can be understood that Transformer is a commonly used transformer model, and the model can be trained using a conventional unsupervised training method. That is, the embodiments of the present application can use geochemical sample data from multiple sampling points as training data and use a conventional unsupervised training method of the transformer model to train the model.

[0073] The following is an exemplary explanation of the Transformer training process.

[0074] After the preprocessed data is input into the Transformer model, the computer system begins to execute a series of complex data processing operations. First, the multi-head self-attention mechanism is used to extract the features of each data point (i.e., sampling point). By calculating the relationships between queries, keys, and values, the long-range dependencies between data points are explored. The specific calculation process is as follows: the input data is linearly projected to obtain features, and then transformed through the scaled dot-product attention function to calculate the attention scores between each sample and other samples, thereby extracting features. Next, a fully connected network is used to further extract high-level features. This process involves multiple complex matrix operations and non-linear transformations, such as residual connections and layer normalization operations, to avoid the problems of gradient vanishing or explosion and achieve deep feature extraction. The entire process is carried out in the computing unit of the computer (such as CPU or GPU), making full use of the computing power of the computer to process complex data relationships, so as to mine the geochemical feature information related to gold ore formation and accurately conduct target area optimization.

[0075] Specifically, the training process of the Transformer includes the following three parts:

[0076] (1) The computer extracts the characteristics of the spatial variation of geochemical samples: The computer receives an unordered sequence of geochemical data, uses the multi-head self-attention mechanism to extract the features of each data point, and captures the long-range dependencies between data points. Then, the fully connected network algorithm is used to further extract high-level features. This step enables the computer system to extract the spatial structure features of multivariate geochemical patterns and the compositional relationships between elements.

[0077] Suppose the geochemical sample data set , where the geochemical sample data of each data point contains the concentration values of multiple elements. Taking as the target sample (i.e., the target geochemical sample data), and the remaining samples are denoted as . After positional encoding and data masking, is input into the Transformer model as an unordered sequence. is linearly projected to obtain the feature , capturing the interactions between the geochemical elements of a single geochemical sample. Then, the feature forms the initial queries, keys, and values. Subsequently, the keys, values, and queries are transformed through the multi-head scaled dot-product attention function to explore the spatial dependencies between the given geochemical samples. For the -th head, the feature of the -th geochemical sample after the output transformation of the self-attention mechanism is :

[0078]

[0079] Among them, is the query value obtained by performing linear projection, and is the key and value obtained by performing linear projection, and their dimensions are both , being any point in the geochemical sample dataset, By performing linear projection, features are obtained. is the feature set of geochemical samples. Finally, for the features of each head concatenation and projection are performed to obtain the final features extracted by multi-head attention :

[0080]

[0081] In the above formula, is the number of heads of the attention mechanism, is the operation of concatenating and projecting the features of all heads, where the projection uses a fully connected layer. Denote this transformation as , that is:

[0082]

[0083] Subsequently, the features extracted by multi-head attention are passed through residual connection and layer normalization to avoid gradient vanishing or gradient explosion.

[0084]

[0085] In the above formula, is the high-level feature after layer normalization, is the layer normalization function.

[0086] Then continue to use a fully connected network to further extract high-level features, and then pass through residual connection and layer normalization again:

[0087]

[0088] In the above formula, is the further extracted high-level feature, is the fully connected network. The output i.e., completes one extraction of spatial variability characteristics. Then After passing through the multi-head scaled dot product attention function, residual connection, layer normalization, fully connected network, residual connection, and layer normalization again, feature extraction is performed again. This is repeated N times to form a deep feature extraction process.

[0089] (2)Reconstruct the geochemical signal based on the extracted features: In the computer system, the interaction between the target sample features and the overall features is realized through a specific algorithm. Specifically, calculate the attention function between the target sample features and the remaining sample features, and apply a linear transformation. Then, after multiple residual connections and layer normalization processes, the interaction of the spatial features and element combination information of the geochemical sample space is completed, and finally the reconstruction result of the target geochemical sample is obtained. This reconstruction process depends on the computer's efficient data processing ability and precise algorithm calculation. By comparing the original data and the reconstructed data, it helps to identify the geochemical anomaly areas that may be related to gold deposits, which serves as the basis for target area optimization.

[0090] The features extracted by the computer system in the previous step are denoted as , which are the deep features of the geochemical signal in the overall dataset. Then, the features of the target sample are denoted as , and the computer system performs the interaction between the features of the target geochemical sample and the overall features :

[0091]

[0092] Among them, is the attention function between the features of the target sample and the features of the remaining samples , represents the linear transformation applied by the computer system to the features of the remaining samples. The computer system uses this transformation to capture the spatial dependence of the geochemical sample for the target sample and extract complex features to reconstruct the geochemical background of the target sample.

[0093] For the result produced after the interaction with the overall features, the computer system performs residual connection and layer normalization processing to avoid gradient disappearance or gradient explosion in the model:

[0094]

[0095] Then, after passing through the transformation of function M, residual connection, and layer normalization again, is obtained:

[0096]

[0097] In the above formula, function M is the multi-head attention function.

[0098] The computer system performs the conversions of the fully connected network, residual connection, and layer normalization again to obtain :

[0099]

[0100] In the above formula, is the fully connected network.

[0101] Thus, from to , the interaction of the geochemical sample space characteristics and element combination information is completed in the computer. This process is similar to the feature extraction stage, and the computer system will repeat it N times. Finally, the final output result - the reconstructed geochemical data of the target geochemical sample data is obtained.

[0102] The computer system will perform the above operations for each sample, perform feature interaction with the overall data, and generate a reconstructed geochemical signal. In this way, the reconstruction of all geochemical samples is completed. Its reconstruction result reflects the characteristics of the geochemical background, provides a benchmark for subsequent anomaly detection, and becomes favorable information for gold ore target area selection.

[0103] (3) Aiming to minimize the difference between the original data and the reconstructed data, the L1 loss is used to quantify the difference. During the training process executed by the calculator system, with the help of computer optimization algorithms (such as the Adam optimizer) and the backpropagation algorithm, the objective function value (i.e., the loss value) is continuously calculated, and the model parameters are adjusted according to the calculation results. Through the rapid processing and iterative calculation of a large amount of data, the computer gradually updates the model, enabling the model to have the basic ability to identify the geochemical background, laying a foundation for the subsequent training of the regeneration framework to more accurately screen out the anomaly information related to the gold ore target area from the geochemical data.

[0104] Represent the initial model (i.e., the trained transformer model) as : . Among them represents the learnable parameters. Among them, the training objective function (i.e., the loss function) of the transformer model is:[[]]

[0105]

[0106] Among them, is the target geochemical sample data, is the output result of the transformer model for under the model parameters , is the loss function of the transformer model, and the remaining samples are denoted as , It is a transformer model.

[0107] Step 13: Initialize the model parameters of the trained transformer model to obtain a student model.

[0108] In some embodiments of the present application, after the transformer model is trained, the computer system performs a parameter reset operation to generate a first-generation reborn model (i.e., the above-mentioned student model, which can also be called a newborn model). This process involves re-initializing the model parameters, prompting the model to re-learn while retaining some prior knowledge, so as to improve the generalization ability of the model.

[0109] Step 14: Use the geochemical sample data of multiple sampling points to perform self-distillation training on the student model to obtain a geochemical data reconstruction model.

[0110] In some embodiments of the present application, a self-knowledge distillation mechanism (i.e., self-distillation training) is introduced. The first-generation reborn model is regarded as the "teacher model", and its output is used as a guiding signal for the training of the newborn model. During the computer processing, the newborn model learns the output of the "teacher model" to obtain more abundant geochemical background feature knowledge, thereby enhancing the learning ability of complex geochemical patterns. In particular, it helps to identify complex feature patterns related to gold ore formation, providing a more accurate basis for the optimal selection of gold ore target areas.

[0111] Specifically, the initial model After being reconstructed and trained by the computer system, the parameters are reset, and the model is reset to the first-generation reborn model . The initial model The reconstruction result of the geochemical data is expressed as , which can be regarded as the initial geochemical background signal extracted by the model. This background still requires further training and cannot be directly used for gold ore target area selection. Therefore, when the computer system trains the model , it uses the output and the original geochemical data together to guide the training of the model. The output result of

[0112] will be considered to be closer to the real geochemical background signal and more conducive to the selection of gold ore target areas. It can be understood that the self-distillation training needs to perform multiple iterations on the first-generation reborn model

[0113] to obtain a geochemical data reconstruction model that meets the performance requirements.

[0114]

[0115] Among them, represents the loss function of the j-th generation model obtained in the j-th iteration during the self-distillation training process, represents the reconstruction loss , represents the rebirth loss , represents the loss weight in the j-th iteration during the self-distillation training process, which is related to the number of iterations, represents the number of sampling points, represents the output result of the j-th generation model with respect to under the model parameters , represents the geochemical sample data of the i-th sampling point, represents the geochemical sample data of other sampling points except the i-th sampling point among multiple sampling points, represents the output result of the (j - 1)-th generation model with respect to under the model parameters . The model parameters are the model parameters of the j-th generation model obtained in the j-th iteration, and the model parameters are the model parameters of the (j - 1)-th generation model obtained in the (j - 1)-th iteration. j is a positive integer greater than or equal to 1 (for the sake of description, the j-th generation model obtained in the j-th iteration is called the j-th generation reborn model ), is the parameter learned by the first generation reborn model on the computer, and they are obtained by resetting the parameters of the initial model and then re-learning the model. Therefore, the models and have the same model structure but different parameters learned on the computer. It can be understood that is . It should be noted that after the model , the parameters will be reset again to generate a new model , and its training is guided by the output of and the geochemical sample data. In the computer system, this iterative process will be repeated multiple times (it can be set that the training termination condition is that the number of iterations reaches a preset number, or the loss value of the model meets the preset loss requirement value), and finally a continuously regenerated network framework is formed.

[0116] During the training of the newborn model, the computer uses the output of the previous-generation model as an additional reference signal, enabling the newborn model to gradually approach the output distribution of the previous-generation model while learning the original geochemical data. The reconstruction loss is controlled by adjusting the rebirth coefficient (i.e., the loss weight mentioned above, for example, adjusted by linear increase). and the rebirth loss In the total loss, the computer calculates and adjusts the loss weight in each iteration according to the set algorithm and coefficient values to ensure the stable convergence of the model and continuously improve the reconstruction accuracy, thereby more accurately locating potential gold ore target areas. Therefore, the training objective of the newborn model is to minimize the reconstruction loss and the rebirth loss .

[0117] Specifically, the calculation formula for the above loss weight is: , , , represents the initial weight, which can be set according to the actual situation. As the number of iterations (i.e., the number of rebirths) increases, the value of the loss weight increases linearly. This dynamic adjustment strategy ensures that in the later stage of training, as the understanding of the data deepens, the model can rely more on the knowledge of the output of the previous-generation model, improving the accuracy and convergence speed of the model.

[0118] It can be seen that during the self-distillation training process, the computer repeatedly executes the training process of the reborn model and performs multiple iterative optimizations. The training of each generation of the model is jointly guided by the output of the previous-generation model and the geochemical sample data, forming a continuously regenerated network framework in the computer system. During the iterative process, the computer gradually increases the rebirth loss weight (according to the set linear increase rule) for each generation and reduces the learning rate according to the convergence characteristics of the model. By repeatedly processing a large amount of data and adjusting parameters, the computer helps the model gradually approach the global optimal solution, effectively reducing the risk of overfitting, enabling the model to continuously optimize the judgment of geochemical anomaly areas during the gold ore target area selection process, and improving the accuracy and reliability of the selection.

[0119] In the gold ore target area selection reborn network framework, this means that as the training progresses, the guiding role of the teacher model (the previous-generation model) on the newborn model (the current-generation model) gradually increases. From the perspective of computer data processing, when calculating the loss function, the gradual increase in the distillation loss coefficient (i.e., the loss weight) makes the model pay more attention to learning the features and information extracted by the previous-generation model, thereby finding the optimal solution faster in the data space and improving the convergence speed of the model. In the computer's memory, by effectively using the output knowledge of the previous generation, the model can better focus on the geochemical feature information related to gold mineralization, improving the accuracy and reliability of gold ore target area selection.

[0120] The learning rate decreases generation by generation during training: During the model training process, as the number of generations increases, the set value of the learning rate in the computer system decreases generation by generation. This strategy is to enable the subsequent model to be trained at a lower learning rate after achieving the performance of the previous generation model, thereby helping the model to get closer to the global optimal solution during the optimization process in the computer and effectively reducing the risk of overfitting.

[0121] Since the subsequent generation network models inherit the knowledge stored in the previous generation network models in the computer memory, their convergence speed during training is faster than that of the first-generation network. In the computer program implementation, when it is detected that the model performance reaches or approaches that of the previous generation model, the learning rate reduction mechanism will be triggered. There are mainly two reasons for this adjustment:

[0122] Firstly, to adapt to the fast convergence characteristics of the model learning process. At the initial stage of computer training, a larger learning rate helps the model quickly explore the data space. However, as the training progresses and the model gradually approaches the optimal solution region, an overly large learning rate may cause the model to skip the optimal solution. Therefore, decreasing the learning rate generation by generation allows the model to more finely adjust the model parameters while accelerating the convergence speed, avoiding missing the global optimal solution. In geochemical data processing, this means that the model can more accurately learn the complex features and relationships contained in the data, such as the subtle correlations between element concentrations and the spatial distribution patterns of elements related to gold mineralization.

[0123] Secondly, to enable the model to continue learning at a lower learning rate after reaching the performance level of the previous generation. A lower learning rate in the computer makes the update amplitude of the model parameters smaller, thus being more sensitive to the detailed information in the data. When processing geochemical data, this fine-tuning ability helps the model capture more subtle abnormal features, enhances its ability to learn the complex knowledge contained in the data, and further improves the accuracy and robustness of the model in anomaly detection and gold ore target area optimization tasks. In the computer implementation process, specific rules and algorithms are set to control the generation-by-generation decrease of the learning rate. For example, exponential decay or linear decay can be adopted. According to different decay strategies and parameter settings, the model can better balance the convergence speed and model performance during training to achieve the best target area selection effect.

[0124] It should be noted that to ensure that the model can meet the requirements of accurately optimizing target areas in gold ore exploration, it is also necessary to calculate the correlation between the model reconstruction results and the distribution of known gold ore points. Specifically, after the step of self-distilling and training the student model using the geochemical sample data of multiple sampling points to obtain the geochemical data reconstruction model, the above gold ore target area optimization method further includes: calculating the AUC value of the geochemical data reconstruction model using the geochemical sample data of multiple positive samples and multiple negative samples. If the AUC value is greater than or equal to the preset threshold, proceed to step 15, input the geochemical sample data of multiple sampling points into the geochemical data reconstruction model for processing to obtain the reconstructed geochemical data of each sampling point. If the AUC value is less than the preset threshold, adjust the model parameters of the geochemical data reconstruction model, and use the geochemical data reconstruction model with adjusted model parameters as the student model in step 14, return to execute step 14, use the geochemical sample data of multiple sampling points to perform self-distilling training on the student model to obtain the geochemical data reconstruction model, until the AUC value is greater than or equal to the preset threshold, then proceed to step 15, input the geochemical sample data of multiple sampling points into the geochemical data reconstruction model for processing to obtain the reconstructed geochemical data of each sampling point. Among them, the positive samples are known gold ore points, and the negative samples are known non-gold ore points.

[0125] Among them, the calculation process of the AUC value of the geochemical data reconstruction model is as follows: First, use the geochemical data reconstruction model to process the geochemical sample data of each positive sample and each negative sample respectively to obtain the reconstructed geochemical data of each positive sample and each negative sample (that is, input the geochemical sample data of the positive sample into the geochemical data reconstruction model for processing to obtain the reconstructed geochemical data of the positive sample; input the geochemical sample data of the negative sample into the geochemical data reconstruction model for processing to obtain the reconstructed geochemical data of the negative sample). Then, based on the reconstructed geochemical data of each positive sample and each negative sample, obtain the geochemical anomaly scores of each positive sample and each negative sample. Finally, calculate the AUC value of the geochemical data reconstruction model using the geochemical anomaly scores of multiple positive samples and multiple negative samples.

[0126] The calculation formula for the AUC value of the geochemical data reconstruction model is:

[0127]

[0128] Among them, represents the AUC value of the geochemical data reconstruction model, represents the number of positive samples, represents the number of negative samples, represents the indicator function, , denotes the geochemical anomaly score of the \(t\)-th positive sample, denotes the geochemical anomaly score of the \(q\)-th negative sample.

[0129] The calculation formula of Calculate the geochemical anomaly score of the \(t\)-th positive sample . Among them, denotes the number of elements, denotes the concentration value of the \(k\)-th element in the \(t\)-th positive sample, denotes the reconstructed concentration value of the \(k\)-th element in the \(t\)-th positive sample.

[0130] The calculation formula of

[0131] Calculate the geochemical anomaly score of the \(q\)-th negative sample . Among them, denotes the number of elements, denotes the concentration value of the \(k\)-th element in the \(q\)-th negative sample, denotes the reconstructed concentration value of the \(k\)-th element in the \(q\)-th negative sample.

[0132] It should be noted that the AUC value is used as a quantitative index for the spatial correlation between the anomaly score and the distribution of known gold ore points. By comparing the false positive rate and the true positive rate, the accuracy and robustness of the model in geochemical anomaly detection are evaluated. The computer calculates the anomaly scores of positive and negative samples according to the set algorithm, and then obtains the AUC value. The higher the AUC value, the better the detection result of the model, and the model performance can be accurately evaluated.

[0133] Step 15: Input the geochemical sample data of multiple sampling points into the geochemical data reconstruction model for processing to obtain the reconstructed geochemical data of each sampling point.

[0134] After successfully completing the training and optimization process of the model, it enters the key stage of gold ore target area optimization. First, use the finally trained and complete geochemical data reconstruction model to process the input geochemical sample data to obtain the reconstructed geochemical data of each sampling point. The reconstructed geochemical data of each sampling point includes the reconstructed concentration values of multiple elements.

[0135] Step 16: Calculate the geochemical anomaly score of each sampling point based on the reconstructed geochemical data of each sampling point, and determine the gold ore target area in the study area based on the geochemical anomaly scores of multiple sampling points.

[0136] In a computer system, for the geochemical sample data of each sampling point, the difference between the original elemental concentration value and the model-reconstructed elemental concentration value is accurately calculated to determine the geochemical anomaly score. Specifically, the implementation method for calculating the geochemical anomaly score of each sampling point based on the reconstructed geochemical data of each sampling point is as follows:

[0137] Through the formula Calculate the geochemical anomaly score of the i-th sampling point ; In the formula, Represents the number of elements, Represents the concentration value of the k-th element at the i-th sampling point, Represents the reconstructed concentration value of the k-th element at the i-th sampling point.

[0138] After calculating the geochemical anomaly scores of each sampling point, the geochemical anomaly scores of multiple sampling points can be sorted in descending order, and the areas corresponding to the first M geochemical anomaly scores among the sorted multiple geochemical anomaly scores are used as the gold ore target areas within the study area. These anomaly areas may contain potential information about gold mineralization. Exemplarily, the value of M can be 5.

[0139] In some embodiments of the present application, based on the selected gold ore target areas, geochemical data and other relevant information in this area and its surrounding areas are further collected, supplemented into the original dataset, and then enter the data preprocessing stage again to restart the entire model training and target area selection process. In a computer system, this process realizes the iterative optimization of the gold ore target areas. With the continuous addition of new data and the repeated training of the model, the model can more accurately identify and select more potential gold ore target areas, gradually improving the accuracy and reliability of gold ore target area selection.

[0140] In practical applications, based on the identified geochemical anomaly areas (i.e., the gold ore target areas mentioned above), relevant information in multiple aspects such as geology, topography, and regional structure can be fully integrated, and further screening and evaluation work can be carried out in combination with the known gold ore distribution information and the fracture zone distribution. In a computer system, a special gold ore target area selection algorithm is preset. This algorithm comprehensively considers the following key factors to calculate the possibility score of each anomaly area becoming a gold ore target area:

[0141] Element combination characteristics: Analyze according to specific element combination patterns closely related to known gold ore-forming models. For example, for the enrichment or depletion ratio relationships of certain specific elements, if they conform to the element symbiotic combination rules in the typical gold ore-forming process, corresponding weight increases are given when calculating scores. For example, elements such as gold (Au), arsenic (As), antimony (Sb), and mercury (Hg) often have specific symbiotic relationships in some gold ore-forming systems. When the element combinations in the anomaly area show characteristics that conform to the ore-forming rules, the possibility score of this area becoming a gold ore target area will be increased.

[0142] Anomaly intensity: Anomaly intensity reflects the significance of geochemical anomalies and is an important indicator for judging the potential of gold ore target areas. In the computer, the anomaly intensity is quantified by calculating the degree of deviation of element concentrations in the anomaly area from the background value. The greater the anomaly intensity, the more intense the possible ore-forming process in this area, and a higher score will be given when calculating the possibility score. For areas where the gold element concentration is much higher than the background value and has a large concentration gradient change, its high potential value will be reflected in the score calculation.

[0143] Spatial distribution pattern: Study the spatial distribution form and rules of the anomaly area. If the anomaly area shows spatial distribution characteristics related to known gold ore-forming structures or geological bodies, such as distribution along fault zones, fold axes, or specific stratigraphic interfaces, it will be given key consideration when calculating scores. For example, an anomaly area showing a linear or zonal distribution and consistent with the strike of the regional fault zone may imply the association between the ore fluid migration channel and the favorable ore-forming part, thus increasing its possibility score of becoming a gold ore target area.

[0144] Regional tectonic information: Regional tectonics plays an important controlling role in the formation and distribution of gold deposits. In computer analysis, factors such as the tectonic position, tectonic type, and tectonic complexity of the anomaly area are investigated in detail. For example, anomaly areas located in favorable tectonic positions such as large fault intersections, anticlinal cores, or ductile shear zones will obtain higher weights when calculating the possibility score because they are more conducive to the accumulation and precipitation of ore-forming materials.

[0145] In addition, considering that the areas within 5 km around gold ore points and fault zones may have been fully explored or have geological conditions unfavorable for ore formation (such as the dispersion of ore-forming elements or the destruction of geological structures), when calculating the possibility score, the anomaly areas within this range are downgraded to reduce their possibility of being selected as gold ore target areas.

[0146] By comprehensively incorporating all the above factors into the gold ore target area optimization algorithm, the computer system calculates a comprehensive, scientific, and reasonable possibility score for each anomaly area. Finally, based on the calculated possibility scores, all anomaly areas are ranked. In the computer system, through an efficient sorting algorithm, the scores of different areas are quickly compared to determine their priority order. Areas with higher scores are considered to have a higher gold ore-forming potential and are thus preferentially selected as the most likely gold ore target areas. Finally, the computer system outputs these optimization results, presenting them to geological exploration personnel in an intuitive and clear manner, providing valuable decision-making references for gold ore exploration work, helping to improve the efficiency and accuracy of gold ore exploration, and reducing exploration costs and risks.

[0147] The following provides an exemplary illustration of the gold ore target area optimization method of the present application in combination with specific examples.

[0148] Taking the Jiaoxi-North gold ore concentration area in Shandong Province as an example, various element stream sediment geochemical measurement values in this area are selected as the original data. In a computer environment (such as the pytorch framework (the pytorch framework is an open-source deep learning framework for machine learning and deep learning), with a graphics card configuration of Nvidia RTX 3090Ti 128G), a converter network of the progressive rebirth framework is used for geochemical anomaly detection and gold ore target area optimization. First, the original data is preprocessed, including operations such as data format adjustment, position encoding addition, and random data masking. Then, the processed data is input into the model for training, and the model adopts a specific layer structure and multi-head attention mechanism (8-layer structure and 2-head attention mechanism).

[0149] During the training process, the computer records relevant metrics of each generation of the model, such as the receiver operating characteristic (ROC) curve and the area under the curve (AUC) value of the geochemical anomaly recognition result, the gold ore target area optimization score, etc. As Figure 2 shown, as the number of model rebirths increases, the AUC generally shows an improving trend. This represents that the accuracy of geochemical anomaly recognition gradually improves from a certain level of the initial model, and the gold ore target area optimization score also gradually increases. As Figure 3 and Table 1 show, after multiple iterations, the selected gold ore target areas have a high correlation with the distribution of known ore points, initially proving the feasibility and effectiveness of the present application in gold ore target area optimization. The entire experimental process fully relies on the powerful computing power and data processing ability of the computer. From data collection, preprocessing, model training to target area optimization and iterative optimization, it is all completed in a large computer workstation with a graphics processing unit (GPU). Utilizing the parallel processing ability of the graphics processing unit to perform scientific calculations and data processing of the model significantly improves the accuracy and reliability of the model optimized by the rebirth framework in target area optimization. It demonstrates the great potential of the present invention in practical gold ore exploration applications.Figure 3 In it, the abscissa is the False Positive Rate, the ordinate is the True Positive Rate, and the ROC Curve represents the ROC curve. Only the positions of the sampling points corresponding to some gold ore target areas are shown in Table 1. The x-axis coordinate and the y-axis coordinate are the abscissa and the ordinate of the sampling points in the plane rectangular coordinate system.

[0150] Table 1 Preferred Scores of Gold Ore Target Areas

[0151]

[0152] In summary, compared with the prior art, the present application has significant advantages: in the aspect of gold ore target area optimization, the progressive rebirth framework of the present application allows for more accurate identification of geochemical anomalies related to gold mineralization through continuous iterative optimization under the condition of sparse geochemical data, thereby achieving more precise gold ore target area optimization and overcoming the problem of low accuracy in target area prediction using traditional methods with limited data. Secondly, in terms of knowledge transfer and model performance improvement, through the progressive rebirth mechanism, the model can gradually accumulate and transfer knowledge related to gold mines, effectively improving the model's understanding ability of complex geochemical patterns, thus improving the accuracy and reliability of gold ore target area optimization. Finally, the present application has good scalability and adaptability. It is not only applicable to gold ore target area optimization, but can also be adjusted according to different combinations of ore-forming elements and geological conditions and applied to the target area optimization of other metal ore types or geological targets, providing a general and effective data processing and target area optimization method for the field of geological exploration.

[0153] The following will exemplarily describe the terminal device provided by the present application in combination with specific embodiments.

[0154] As Figure 4 shown, an embodiment of the present application provides a terminal device. As Figure 4 shown, the terminal device D10 of this embodiment includes: at least one processor D100 ( Figure 4 only one processor is shown in the figure), a memory D101, and a computer program D102 stored in the memory D101 and executable on the at least one processor D100. When the processor D100 executes the computer program D102, it implements the steps in any of the above method embodiments.

[0155] Specifically, when the processor D100 executes the computer program D102, based on the trained transformer model, by applying the progressive rebirth mechanism, the model parameters of the trained transformer model are initialized to obtain the student model, which prompts the student model to relearn while retaining some prior knowledge. Since the progressive rebirth mechanism allows for the accurate identification of geochemical anomalies related to gold mineralization through continuous iterative optimization under sparse geochemical data conditions, and then realizes the precise optimization of gold ore target areas, overcoming the problem of low accuracy in target area prediction using traditional methods with limited data. Secondly, in terms of knowledge transfer and model performance improvement, through the progressive rebirth mechanism, the model can gradually accumulate and transfer knowledge related to gold, effectively enhancing the model's ability to understand complex geochemical patterns, thereby improving the accuracy and reliability of gold ore target area optimization and meeting the requirements for accurately optimizing target areas in gold exploration.

[0156] The so-called processor D100 may be a central processing unit (CPU, Central Processing Unit), and this processor D100 may also be other general-purpose processors, digital signal processors (DSP, Digital Signal Processor), application specific integrated circuits (ASIC, Application Specific Integrated Circuit), field-programmable gate arrays (FPGA, Field-Programmable Gate Array), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or this processor may also be any conventional processor, etc.

[0157] In some embodiments, the memory D101 may be an internal storage unit of the terminal device D10, such as the hard disk or memory of the terminal device D10. In other embodiments, the memory D101 may also be an external storage device of the terminal device D10, such as a plug-in hard disk, smart media card (SMC, SmartMedia Card), secure digital (SD, Secure Digital) card, flash card (Flash Card), etc. equipped on the terminal device D10. Further, the memory D101 may also include both the internal storage unit of the terminal device D10 and the external storage device. The memory D101 is used to store the operating system, application programs, boot loader (BootLoader), data, and other programs, such as the program code of the computer program, etc. The memory D101 may also be used to temporarily store data that has been output or will be output.

[0158] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the above division of each functional unit and module is used as an example. In actual applications, the above functions can be allocated to different functional units and modules according to needs, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. Each functional unit and module in the embodiment can be integrated into a processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above integrated unit can be implemented in the form of hardware or in the form of a software functional unit. In addition, the specific names of each functional unit and module are only for the convenience of mutual distinction and do not limit the protection scope of this application. The specific working processes of the units and modules in the above system can refer to the corresponding processes in the foregoing method embodiments and will not be repeated here.

[0159] An embodiment of this application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps in the foregoing method embodiments can be implemented.

[0160] An embodiment of this application provides a computer program product. When the computer program product runs on a terminal device, the terminal device is caused to execute the steps in the foregoing method embodiments.

[0161] If the above integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, to implement all or part of the processes in the foregoing method embodiments of this application, a computer program can be used to instruct relevant hardware to complete. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by a processor, the steps in the foregoing method embodiments can be implemented. Among them, the computer program includes computer program code, and the computer program code can be in the form of source code, object code, executable file or some intermediate form, etc. The computer-readable medium can at least include: any entity or device that can carry the computer program code to the terminal device, recording medium, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electrical carrier signal, telecommunication signal, and software distribution medium. For example, a USB flash drive, a mobile hard disk, a magnetic disk, or an optical disc, etc. In some jurisdictions, according to legislation and patent practice, the computer-readable medium cannot be an electrical carrier signal and a telecommunication signal.

[0162] In the above embodiments, the descriptions of the various embodiments each have their own emphasis. For parts not detailed or recorded in a certain embodiment, reference may be made to the relevant descriptions of other embodiments.

[0163] Those of ordinary skill in the art will realize that the units and algorithm steps of the examples described in combination with the embodiments disclosed herein can be implemented by electronic hardware, or by a combination of computer software and electronic hardware. Whether these functions are executed in hardware or software depends on the specific application and design constraints of the technical solution. Skilled professionals can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of this application.

[0164] In the embodiments provided in this application, it should be understood that the disclosed method can be implemented in other ways. For example, the method / device embodiments described above are merely illustrative. For example, the division of the modules or units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be through some interfaces, and the indirect couplings or communication connections of devices or units can be in electrical, mechanical or other forms.

[0165] The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they can be located in one place, or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0166] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them; although this application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included in the protection scope of this application.

Claims

1. A gold ore target area optimization method based on elemental geochemical anomalies, characterized in that Including: Obtaining geochemical sample data of multiple sampling points within a research area; The geochemical sample data includes concentration values of multiple elements; Using the geochemical sample data of the multiple sampling points to train a converter model, obtaining a trained converter model; the converter model is used to process the geochemical sample data and output reconstructed geochemical data; Initializing the model parameters of the trained converter model to obtain a student model; Using the geochemical sample data of the multiple sampling points to perform self-distillation training on the student model, obtaining a geochemical data reconstruction model; Inputting the geochemical sample data of the multiple sampling points into the geochemical data reconstruction model for processing, obtaining the reconstructed geochemical data of each sampling point; Calculating the geochemical anomaly score of each sampling point based on the reconstructed geochemical data of each sampling point, and determining the gold ore target area within the research area based on the geochemical anomaly scores of the multiple sampling points; The loss function during the process of using the geochemical sample data of the multiple sampling points to perform self-distillation training on the student model is: ； Among them, represents the loss function of the j-th generation model obtained in the j-th iteration during the self-distillation training process, represents the loss weight in the j-th iteration during the self-distillation training process, which is related to the number of iterations, represents the number of sampling points, represents the output result of the j-th generation model with respect to under the model parameters ; represents the geochemical sample data of the i-th sampling point, represents the geochemical sample data of other sampling points except the i-th sampling point among multiple sampling points, represents the output result of the (j - 1)-th generation model with respect to under the model parameters .

2. The gold ore target area optimization method according to claim 1, wherein The obtaining of the geochemical sample data of multiple sampling points within the research area includes: For each sampling point respectively, using a position encoding function to perform position encoding on the original geochemical sample data of the sampling point, and performing a vector addition operation on the position encoding result and the original geochemical sample data of the sampling point, obtaining the geochemical sample data of the sampling point.

3. The gold ore target area optimization method according to claim 1, wherein The loss weight has the following calculation formula: , , , where represents the initial weight.

4. The gold ore target area optimization method according to claim 1, wherein The reconstructed geochemical data of each sampling point includes reconstructed concentration values of multiple elements; The calculating of the geochemical anomaly score of each sampling point based on the reconstructed geochemical data of each sampling point includes: Calculate the geochemical anomaly score of the i-th sampling point through the formula ;​ Among them, represents the quantity of the said element, represents the concentration value of the k-th element at the i-th sampling point, represents the reconstructed concentration value of the k-th element at the i-th sampling point.

5. The gold ore target area optimization method according to claim 4, wherein The determining of the gold ore target area within the research area based on the geochemical anomaly scores of the multiple sampling points includes: Sorting the geochemical anomaly scores of the multiple sampling points in descending order; Taking the areas corresponding to the sampling points of the first M geochemical anomaly scores among the sorted multiple geochemical anomaly scores as the gold ore target area within the research area.

6. The gold ore target area optimization method according to claim 1, wherein After the step of using the geochemical sample data of the multiple sampling points to perform self-distillation training on the student model to obtain a geochemical data reconstruction model, the gold ore target area optimization method further includes: Calculating the AUC value of the geochemical data reconstruction model using the geochemical sample data of multiple positive samples and multiple negative samples; the positive samples are known gold ore points, and the negative samples are known non-gold ore points; If the AUC value is less than the preset threshold, adjust the model parameters of the geochemical data reconstruction model, and use the geochemical data reconstruction model with adjusted model parameters as the student model. Return to execute the step of self-distilling and training the student model using the geochemical sample data of the multiple sampling points to obtain the geochemical data reconstruction model, until the AUC value is greater than or equal to the preset threshold, then enter the step of inputting the geochemical sample data of the multiple sampling points into the geochemical data reconstruction model for processing to obtain the reconstructed geochemical data of each sampling point; If the AUC value is greater than or equal to the preset threshold, enter the step of inputting the geochemical sample data of the multiple sampling points into the geochemical data reconstruction model for processing to obtain the reconstructed geochemical data of each sampling point.

7. The gold ore target area optimization method according to claim 6, characterized in that, Calculating the AUC value of the geochemical data reconstruction model using the geochemical sample data of multiple positive samples and multiple negative samples includes: Using the geochemical data reconstruction model to process the geochemical sample data of each positive sample and each negative sample respectively to obtain the reconstructed geochemical data of each positive sample and each negative sample; Obtaining the geochemical anomaly scores of each positive sample and each negative sample based on the reconstructed geochemical data of each positive sample and each negative sample; Calculating the AUC value of the geochemical data reconstruction model using the geochemical anomaly scores of the multiple positive samples and the multiple negative samples.

8. The gold ore target area optimization method according to claim 7, characterized in that The calculation formula for the AUC value of the geochemical data reconstruction model is: Among them, represents the AUC value of the geochemical data reconstruction model, represents the number of positive samples, represents the number of negative samples, represents the indicator function, , represents the geochemical anomaly score of the t-th positive sample, represents the geochemical anomaly score of the q-th negative sample.

9. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the gold ore target area optimization method based on elemental geochemical anomalies according to any one of claims 1 to 8.

Images