CN-121996991-A - Shallow-coverage gold mine target area prediction method and system

CN121996991ACN 121996991 ACN121996991 ACN 121996991ACN-121996991-A

Abstract

The application relates to the technical field of intelligent exploration, in particular to a shallow-coverage gold target area prediction method and a shallow-coverage gold target area prediction system, which comprise the steps of inputting multi-source mine point data of an area to be predicted into a gold target area prediction model to obtain a corresponding gold target area prediction result; the gold target area prediction model is trained by acquiring first mine point data corresponding to a first mine point and second mine point data corresponding to a second mine point, acquiring a first Au mean value and a correlation coefficient corresponding to the first mine point based on the first mine point data, acquiring a first Au mean value and a correlation coefficient corresponding to the second mine point based on the second mine point data, adjusting the first mine point data and the second mine point data based on the first Au mean value and the correlation coefficient, and acquiring a gold target area prediction model for carrying out gold target area prediction on a shallow coverage area based on the adjusted first mine point data and the adjusted second mine point data. The method effectively supplements the total sample amount and obviously reduces the risk of missed judgment and misjudgment of the target area.

Inventors

YANG XIAOQI
WANG KANGWEI
LI TIANHU
WANG SHUAI
LU YUEXIN
CHEN SHUJIE
PENG QICAI
ZHANG XIANPENG
FENG GONG
DONG QIMING
XU XIAOCHEN
FAN BAOCHENG
LI LONGBO
ZHANG JING
YANG LUKUAN
ZHANG HUISHAN
ZHAO HANSEN

Assignees

中国地质调查局西安地质调查中心(西北地质科技创新中心)
山东招金地质勘查有限公司

Dates

Publication Date: 20260508
Application Date: 20260409

Claims (10)

1. A method for predicting a target zone of a shallow-covered gold mine, comprising: Inputting the multi-source ore point data of the region to be predicted into a pre-trained gold ore target region prediction model to obtain a corresponding gold ore target region prediction result; The gold mine target area prediction model is trained through the following steps: Acquiring first mine point data corresponding to a first mine point and second mine point data corresponding to a second mine point, wherein the first mine point data and the second mine point data comprise Au content and mine point related data corresponding to a plurality of different sampling points; Acquiring a first Au mean value and a correlation coefficient corresponding to a first mine point based on first mine point data, and acquiring a first Au mean value and a correlation coefficient corresponding to a second mine point based on second mine point data, wherein the correlation coefficient and the second correlation coefficient are the correlation degree between Au content and mine point related data; Adjusting the first mine point data and the second mine point data based on a first Au average value and a correlation coefficient corresponding to the first mine point and the second mine point to obtain adjusted first mine point data and second mine point data; and training the gold target area prediction model based on the adjusted first mine point data and second mine point data to obtain a gold target area prediction model for predicting the gold target area of the shallow coverage area.
2. The method for predicting a target zone of a shallow-covered gold mine of claim 1, wherein the mine point-related data includes As content and distance from a fracture zone; The correlation coefficient comprises a first correlation coefficient and a second correlation coefficient; Acquiring a first Au mean and a correlation coefficient corresponding to the first mine point based on the first mine point data, and acquiring a first Au mean and a correlation coefficient corresponding to the second mine point based on the second mine point data, wherein the method comprises the following steps: Performing characteristic transformation on the first mine point data and the second mine point data, wherein the Au content and the As content are transformed through a preset formula I, the distance from a fracture zone is transformed through a Z-score, and the formula I is As follows: x 1 =log 10 (x 0 +1); Wherein x 0 is a parameter value before feature transformation, and x 1 is a parameter value after feature transformation; Based on Au contents corresponding to a plurality of sampling points in the first mine point data and the second mine point data after feature transformation, obtaining a first Au average value corresponding to the first mine point and the second mine point; Constructing a first mine point first correlation coefficient and a second correlation coefficient based on the first mine point data after the feature transformation, and constructing a second mine point first correlation coefficient and a second correlation coefficient based on the second mine point data after the feature transformation; the first correlation coefficient is a degree of correlation between the Au content and the As content, and the second correlation coefficient is a degree of correlation between the Au content and the distance from the fracture zone.
3. The method for predicting a target zone of a shallow covered gold mine according to claim 2, wherein the mine point related data further comprises a gradient and Hg content in soil; acquiring a first Au mean value and a correlation coefficient corresponding to the first mine point based on the first mine point data, and acquiring a first Au mean value and a correlation coefficient corresponding to the second mine point based on the second mine point data, wherein the method further comprises: performing characteristic transformation on Hg content in soil of each sampling point in the first mine point data and the second mine point data through the formula I, and performing characteristic transformation on gradient of each sampling point in the first mine point data and the second mine point data through the Z-score; Based on Hg content and gradient in soil corresponding to each sampling point in the first mining point data and the second mining point data after feature transformation and a preset formula II, hg correction data corresponding to each sampling point is obtained, wherein the formula II is as follows: Hg Correction of =Hg Detection of ×(1+k×(Slope-5°)); Hg Detection of is Hg content in soil corresponding to a sampling point, hg Correction of is Hg correction data corresponding to the sampling point, slope is gradient corresponding to the sampling point, and k is a preset gradient correction parameter; Based on Hg correction data and Au content corresponding to each sampling point in the first mine point data and the second mine point data, respectively obtaining a third correlation coefficient corresponding to the first mine point and the second mine point, wherein the third correlation coefficient is the correlation degree between the Au content and the Hg correction data.
4. The method of claim 1, wherein adjusting the first mine data and the second mine data based on the first Au mean and the correlation coefficient corresponding to the first mine and the second mine comprises: averaging the first Au average values corresponding to the first mine points and the second mine points to obtain second Au average values corresponding to the first mine points and the second mine points; Based on a first Au average value and a second Au average value corresponding to the first mine point and the second mine point, respectively obtaining Au adjustment parameters corresponding to the first mine point and the second mine point, wherein the Au adjustment parameters are obtained by dividing the second Au average value by the first Au average value; Adjusting the Au content of each sampling point in the first mine point based on the Au adjustment parameters corresponding to the first mine point, and adjusting the Au content of each sampling point in the second mine point based on the Au adjustment parameters corresponding to the second mine point, wherein the adjusted Au content is the product of the Au adjustment parameters and the Au content; Based on the correlation coefficient corresponding to the first mine point and the Au content adjusted by each sampling point in the first mine point, the correlation coefficient corresponding to each sampling point in the first mine point is adjusted, and based on the correlation coefficient corresponding to the second mine point and the Au content adjusted by each sampling point in the second mine point, the correlation coefficient corresponding to each sampling point in the second mine point is adjusted.
5. The method for predicting a target area of a shallow-covered gold mine according to claim 2, wherein obtaining a first Au mean value corresponding to the first mine and the second mine based on Au contents corresponding to a plurality of sampling points in the first mine data and the second mine data after feature transformation comprises: acquiring a first Au average value corresponding to the first mine point based on Au contents corresponding to a plurality of sampling points in the first mine point data after feature transformation and a preset quantile mapping algorithm; Acquiring a first Au average value corresponding to the second mine point based on Au contents corresponding to a plurality of sampling points in the second mine point data after feature transformation and a preset quantile mapping algorithm; the first Au average value is a value corresponding to 50% quantiles.
6. The method of claim 4, wherein training the gold target prediction model based on the adjusted first and second mine point data to obtain the gold target prediction model for predicting the gold target in the shallow coverage area comprises: Performing mining point labeling on the adjusted first mining point data and second mining point data to label corresponding mining point labels in the first mining point data and the second mining point data; inputting the first mine point data and the second mine point data marked by the mine points into a gold mine target area prediction model to obtain prediction data corresponding to the first mine points and the second mine points; And adjusting model parameters of the gold target area prediction model based on prediction data, the mine point labels and the second Au average values corresponding to the first mine points and the second mine points to obtain the gold target area prediction model for predicting the gold target area in the shallow coverage area.
7. The method of claim 6, wherein adjusting model parameters of the gold target prediction model based on the prediction data corresponding to the first and second points, the point labels, and the second Au mean value comprises: obtaining a corresponding total loss value based on the predicted data corresponding to the first mine point and the second mine point, the mine point label and the second Au average value; And adjusting model parameters of a gold mine target area prediction model based on the total loss value and a preset BP algorithm.
8. The method for predicting a target zone of a shallow-covered gold mine of claim 7, wherein the prediction data comprises a predicted ore-bearing probability and a predicted grade; Based on the predicted data corresponding to the first mine point and the second mine point, the mine point label and the second Au average value, obtaining a corresponding total loss value comprises the following steps: Based on the predicted ore-bearing probability and the ore-point label corresponding to the first ore point and the second ore point and a preset formula III, obtaining the corresponding classification loss, wherein the formula III is as follows: ; Wherein L 1 is a classification loss, y 1 is a mining point label corresponding to a first mining point, y 2 is a mining point label corresponding to a second mining point, For the predicted mining probability corresponding to the first mine point, Predicting the ore-bearing probability corresponding to the second ore point; Obtaining a corresponding grade regression loss based on a predicted grade and a second Au average value corresponding to the first ore point and the second ore point and a preset formula IV, wherein the formula IV is as follows: ; Wherein L 2 is the grade regression loss, G 1 is the second Au average value corresponding to the first mine point, G 2 is the second Au average value corresponding to the second mine point, For the predicted grade corresponding to the first mine point, The predicted grade corresponding to the second ore point; And carrying out weighted summation on the classification loss and the grade regression loss to obtain a corresponding total loss value.
9. The method for predicting a target zone of a shallow-covered gold mine as claimed in claim 1, wherein the acquiring of the first mine data corresponding to the first mine and the second mine data corresponding to the second mine acquired in advance includes: Acquiring first mine point data corresponding to a first mine point and second mine point data corresponding to a second mine point, wherein the mine forming mechanism and the data acquisition specification of the two different mine points are consistent; the sampling coordinates of all the sampling points are unified based on a preset WGS84 coordinate system, and resampled into a10 m x 10m grid to be matched with the sampling point space.
10. A shallow covered gold target prediction system comprising a memory, a processor and a computer program stored on the memory, wherein the processor executes the computer program to implement the shallow covered gold target prediction method of any one of claims 1 to 9.

Description

Shallow-coverage gold mine target area prediction method and system Technical Field The application relates to the technical field of intelligent exploration, in particular to a shallow-coverage gold mine target area prediction method and system. Background Gold ore resources are important mineral resources, play an irreplaceable role in technological development, financial safety, high-end manufacturing and jewelry industry, and have remarkable significance in promoting regional economic development and promoting employment. As the surface outcrop is increasingly depleted, traditional prospecting spaces continue to compress, and shallow coverage areas (i.e., areas where bedrock is covered by thin layer deposits, soil, or vegetation) become key targets for finding new deposits. Although the exploration difficulty of the areas is high, the ore-forming geological conditions are often superior, and the areas have huge ore-finding potential. Therefore, the development of the high-efficiency and accurate gold prediction and investigation technology suitable for the shallow coverage area has great strategic significance for breaking through resource bottleneck, expanding gold resource reserve and guaranteeing resource safety. In the existing gold mine target area prediction technology, deep learning algorithms such as convolutional neural networks, multi-layer perceptrons and multi-mode fusion models are mostly adopted, and mining probability or potential grading results are output as the basis of target area delineation through inputting multi-source data of an area to be predicted into a model which is completed through training. However, the model training is carried out by relying on local data of a single mining area, when the known mining points of a target mining area are few and the sample size is insufficient, the model is easy to be subjected to over fitting, although some methods try to introduce cross-domain data of other mining areas to supplement samples, core characteristics of the cross-domain data are not subjected to targeted adjustment, so that distribution deviation of the data of different mining areas occurs due to fine differences (such as mineralization intensity and element migration efficiency) of an ore forming environment, the cross-domain data cannot exert a supplement effect, and negative migration interference is introduced, so that the prediction reliability of the model is reduced. Disclosure of Invention First, the technical problem to be solved In view of the above-mentioned shortcomings and disadvantages of the prior art, the present application provides a method and a system for predicting a target area of a gold mine with a shallow coverage area, which solve the technical problems of multiple training samples and long training time required and easy occurrence of reverse migration in a deep learning algorithm mode such as convolutional neural network, multi-layer perceptron, multi-mode fusion model, etc. (II) technical scheme In order to achieve the above purpose, the main technical scheme adopted by the application comprises the following steps: in a first aspect, an embodiment of the present application provides a method for predicting a target area of a gold mine with a shallow coverage area, including: Inputting the multi-source ore point data of the region to be predicted into a pre-trained gold ore target region prediction model to obtain a corresponding gold ore target region prediction result; The gold mine target area prediction model is trained through the following steps: Acquiring first mine point data corresponding to a first mine point and second mine point data corresponding to a second mine point, wherein the first mine point data and the second mine point data comprise Au content and mine point related data corresponding to a plurality of different sampling points; Acquiring a first Au mean value and a correlation coefficient corresponding to a first mine point based on first mine point data, and acquiring a first Au mean value and a correlation coefficient corresponding to a second mine point based on second mine point data, wherein the correlation coefficient and the second correlation coefficient are the correlation degree between Au content and mine point related data; Adjusting the first mine point data and the second mine point data based on a first Au average value and a correlation coefficient corresponding to the first mine point and the second mine point to obtain adjusted first mine point data and second mine point data; and training the gold target area prediction model based on the adjusted first mine point data and second mine point data to obtain a gold target area prediction model for predicting the gold target area of the shallow coverage area. Optionally, in a specific embodiment, the mine site-related data includes As content and distance from the fracture zone; The correlation coefficient comprises a first correlation coefficient and a sec