A method for reducing low latitude magnetic anomalies

Abstract

The application discloses a method for reducing low-latitude magnetic anomaly, comprising the following steps: S1, constructing a model space of a sample data set, wherein the model space is composed of an underground half space and a plurality of cuboid units, and the plurality of cuboid units are arranged in a combined form in a non-edge area of the underground half space; S2, performing traversal sampling and data amplification to obtain a sample data equivalent magnetic anomaly body; S3, acquiring a stable initial model by using a frequency domain suppression method; S4, taking the stable initial model and a magnetic anomaly with an arbitrary magnetization direction as inputs of a FCN network structure, taking a magnetic anomaly perpendicular to the magnetization direction as a label, and training the FCN reduction network structure; and S5, taking measured data and the stable initial model as inputs of the trained FCN reduction network structure to obtain a reduction prediction result. The application can obtain more accurate reduction results and provide technical support for geological interpretation.

Description

Technical Field

[0001] This invention relates to the field of magnetic exploration technology, and in particular to a method for polarization of low-latitude magnetic anomalies. Background Technology

[0002] Magnetic anomaly polarization is a crucial foundational task in magnetic data processing and interpretation. The process involves converting magnetic anomalies with arbitrary magnetization dips into anomalies perpendicular to the magnetization direction. The aim is to shift the peak value of the anomaly directly above the anomaly body, simplifying its morphology. Polarization enhances the correlation between magnetic anomalies and source distributions, facilitating geological interpretation. It can even be used directly to identify tectonic boundaries, delineate rock mass extents, and classify sedimentary basin sizes. Frequency-domain polarization is widely used due to its clear principles, ease of implementation, high computational efficiency, and proven effectiveness in many applications. However, in low latitudes and equatorial regions, the frequency-domain polarization factor exhibits a significant amplification effect, leading to instability in polarization calculations.

[0003] Existing technology discloses a latitudinal magnetic anomaly polarization method—the suppression factor method. Based on the planar and cross-sectional characteristics of the low-latitude polarization factor, it proposes a suppression factor designed to suppress and modify the "dead zone" in the θ0=D0±90°, θ0±α0 region. This factor tends to zero near θ0=D0±90°, indicating the strongest suppression effect; outside a certain range, it equals 1, meaning no suppression occurs, thus avoiding the amplification problem of the polarization factor in low-latitude magnetic anomalies. Although this method can alleviate the amplification effect of the frequency domain polarization factor of low-latitude magnetic anomalies to some extent, it still suffers from instability in the actual polarization calculation and insufficient polarization accuracy during practical polarization, resulting in a certain gap between the polarization results and the actual theoretical results. Summary of the Invention

[0004] To address the aforementioned problems, this invention aims to provide a polarization method for low-latitude magnetic anomalies. Based on the existing suppression factor polarization method, it utilizes low-latitude magnetic anomalies and FCN network structures to improve the accuracy of polarization results.

[0005] The technical solution of the present invention is as follows:

[0006] A method for polarization of low-latitude magnetic anomalies includes the following steps:

[0007] S1: Construct the model space of the sample dataset, the model space consisting of an underground half-space and multiple cuboid units, and the multiple cuboid units are arranged in combination in the non-edge area of ​​the underground half-space;

[0008] S2: Sampling is performed on various combinations of the multiple cuboid units, and the sample data with different magnetic susceptibility values ​​is amplified using the forward-modeled sample data to obtain the sample data equivalent magnetic anomaly.

[0009] S3: Obtain a stable initial model for the equivalent magnetic anomaly of the sample data using the frequency domain suppression method;

[0010] S4: Use the stable initial model and the magnetic anomaly in any magnetization direction as input to the FCN network structure, and use the magnetic anomaly in the perpendicular magnetization direction as a label to train the FCN polarization network structure.

[0011] S5: Use the measured data and the stable initial model together as input to the trained FCN polarization network structure to obtain polarization prediction results.

[0012] Preferably, in step S1, the combination of the multiple cuboid units takes the form of any one or more of the following shapes: cuboid, inclined, "L" shape, and "T" shape.

[0013] Preferably, in step S3, before obtaining a stable initial model of the equivalent magnetic anomaly of the sample data using the frequency domain suppression method, noise is added to the dataset of the equivalent magnetic anomaly of the sample data.

[0014] As a preferred method, the noise is added by adding 5% Gaussian noise to 10% of the random samples in the dataset.

[0015] The beneficial effects of this invention are:

[0016] This invention obtains equivalent magnetic anomalies by traversing and sampling the underground grid model space and amplifying the data. Adding 5% Gaussian noise to 10% of the random sample pairs in the dataset improves the generalization performance of the trained network. Using a stable initial model and magnetic anomalies with arbitrary magnetization directions as input to the FCN network structure, and magnetic anomalies perpendicular to the magnetization direction as labels, the resulting FCN polarization network structure is trained. Then, measured data and the stable initial model are used as input to this FCN polarization network structure for polarization prediction. This solves the problems of unstable polarization calculations and insufficient accuracy of low-latitude magnetic anomalies in existing technologies, significantly improving the effectiveness, robustness, and accuracy of polarization calculation results. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A flowchart for model-constrained magnetic anomaly FCN polarization;

[0019] Figure 2 This is a schematic diagram of a magnetic anomaly in the underground space of the model.

[0020] Figure 3 This is a perspective view of the combined model and a schematic diagram of the forward modeling results of the magnetic anomaly;

[0021] Figure 4 A schematic diagram of the magnetic anomaly FCN polarization network structure constrained by the model;

[0022] Figure 5 A schematic diagram comparing the coefficients of determination of the extreme results of different methods under different signal-to-noise ratio conditions;

[0023] Figure 6 Schematic diagrams of magnetic anomalies with different signal-to-noise ratios and their polarization results using different polarization methods;

[0024] Figure 7 A schematic diagram of magnetic anomalies in a sea area of ​​East Asia and the polarization results of different polarization methods;

[0025] Figure 8 A schematic diagram showing the magnetic anomaly in a certain area of ​​Hebei Province and the polarization results of different polarization methods; Detailed Implementation

[0026] The present invention will be further described below with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and technical features described in this application can be combined with each other. It should also be pointed out that, unless otherwise indicated, all technical and scientific terms used in this application have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. The terms "comprising" or "including" and similar words used in this invention refer to elements or objects preceding the word that encompass the elements or objects listed following the word and their equivalents, without excluding other elements or objects.

[0027] like Figure 1 As shown, the present invention provides a method for polarization of low-latitude magnetic anomalies, comprising the following steps:

[0028] S1: Construct a model space for the sample dataset. The model space consists of an underground half-space and multiple cuboid units, and the multiple cuboid units are arranged in combination in the non-edge region of the underground half-space.

[0029] In one specific embodiment, when constructing the model space of the sample dataset, the underground space is divided into 63×63×10 cuboid units, each cuboid being 1m×1m×1m in size. A Cartesian coordinate system is used, with the z-axis pointing downwards. The number of points used to calculate the planar magnetic anomaly data is 64×64, and the sampled data points are the projections of the underground half-space grid points onto the horizontal plane. For example... Figure 2 As shown, multiple cuboid units are arranged in combination in the non-edge area of ​​the underground semi-space, with the following types: Type I is a cuboid type, Type II is an inclined body model, Type III is an "L" shape, and Type IV is a "T" shape. In this embodiment, using these combination types can enhance the universality of the model training results.

[0030] S2: Sampling is performed on various combinations of the multiple cuboid units, and the sample data with different magnetic susceptibility values ​​is amplified using the forward-modeled sample data to obtain the sample data equivalent magnetic anomaly.

[0031] In one specific embodiment, the combined model and the forward modeling results of the magnetic anomaly are as follows: Figure 3 As shown, where, Figure 3 (a) is a schematic diagram of the perspective result of the combined model. Figure 3 (b) is a schematic diagram of the magnetic anomaly. Figure 3 (c) is a schematic diagram of the forward modeling results of magnetic anomalies. It should be noted that traversal sampling and forward modeling of magnetic anomalies are existing technologies, and the specific methods will not be described in detail here.

[0032] S3: Obtain a stable initial model for the equivalent magnetic anomaly of the sample data using the frequency domain suppression method.

[0033] In one specific embodiment, before obtaining a stable initial model of the equivalent magnetic anomaly of the sample data using a frequency domain suppression method, noise is added to the dataset of the equivalent magnetic anomaly of the sample data. Optionally, when adding noise, 5% Gaussian noise is added to 10% of the random samples in the dataset.

[0034] It should be noted that the frequency domain suppression method is existing technology, namely the "low-latitude magnetic anomaly polarization method - suppression factor method" mentioned in the background technology. Therefore, the specific steps for obtaining a stable initial model using this method will not be described here.

[0035] S4: As Figure 4 As shown, the stable initial model and the magnetic anomaly in any magnetization direction are used together as input to the FCN network structure, and the magnetic anomaly in the perpendicular magnetization direction is used as a label to train the FCN polarization network structure.

[0036] In this invention, based on the stable initial model, the FCN polarization network structure can be given a general direction, which can minimize unnecessary features in the deep learning fitting process.

[0037] S5: Use the measured data and the stable initial model together as input to the trained FCN polarization network structure to obtain polarization prediction results.

[0038] In one specific embodiment, different methods are used to perform polarization prediction on the same dataset, and the results are as follows: Figure 5 As shown, where, Figure 5 (a) is a schematic diagram comparing the coefficients of determination of the polarization results of the training model with noise-free datasets as training samples. Figure 5 (b) is a diagram showing the comparison of the determination coefficients of the polarization results of the training model with training samples containing noisy datasets. FCN-Magnetic anomalies (FCN-Mag) represent the determination coefficients of the polarization results of magnetic anomalies FCN, FCN-Initial-Magnetic anomalies (FCN-Initial-Mag) represent the determination coefficients of the polarization results of FCN jointly driven by "stable initial model & magnetic anomaly", FCN-Initial represents the determination coefficients of the polarization results of FCN with stable initial model, and Initial represents the determination coefficients of stable initial model. The larger the value of the determination coefficient, the better the polarization result.

[0039] from Figure 5 It can be seen that, when the training samples are free of noise, the magnetic anomaly-driven FCN exhibits the best polarization results when the signal-to-noise ratio (SNR) is less than 30, followed by the FCN polarization results driven by the "stable initial model & magnetic anomaly" of this invention. When the SNR is greater than 30, the FCN polarization results driven by the "stable initial model & magnetic anomaly" of this invention are the best. In the presence of noise, the FCN polarization results driven by the "stable initial model & magnetic anomaly" of this invention are the best, and the coefficient of determination increases with increasing SNR.

[0040] In one specific embodiment, for Figure 3 The magnetic anomaly in (b) and the magnetic anomaly after adding the signal-to-noise ratio were polarized using different methods (FCN-Mag, FCN-Initial-Mag, FCN-Initial, Initial). The results are shown in Table 1. Figure 6 As shown, where Figure 6 (a) is Figure 3 (b) is a schematic diagram showing the results of magnetic anomalies with different signal-to-noise ratios. Figure 6 (b) shows the results of the suppression factor method for polarization. Figure 6 (c) shows the FCN polarization results driven by magnetic anomaly. Figure 6(d) shows the FCN polarization results driven by the initial model. Figure 6 (e) shows the FCN polarization results jointly driven by "magnetic anomaly & initial model". In this embodiment, the FCN polarization networks are all trained using a sample dataset containing noise.

[0041] Table 1. Determination coefficients of polarization results for different polarization methods

[0042] Polarization method Primitive magnetic anomaly Magnetic anomaly with SNR=40 Magnetic anomaly with SNR=20 Magnetic anomaly with SNR=10 Initial 0.9339 0.9339 0.8461 0.8289 FCN-Mag 0.9446 0.9445 0.9437 0.9032 FCN-Initial 0.9691 0.9691 0.8853 0.8681 FCN-Initial-Mag 0.9869 0.9868 0.9609 0.9544

[0043] From Table 1 and Figure 6 It can be seen that, regardless of whether the signal-to-noise ratio (SNR) is absent or different, the pole-setting results of the "stable initial model & magnetic anomaly" combined-drive FCN method of this invention are the best. It shows improvement over the FCN-Mag, FCN-Initial, and Initial pole-setting methods. The pole-setting results of this invention, with a coefficient of determination as high as 98% when the SNR is absent and SNR = 40, and still greater than 95% when the SNR = 10, demonstrate that this invention has good generalization performance and better effectiveness, accuracy, and robustness.

[0044] In one specific embodiment, the present invention is used to perform polarization on magnetic measurement data of a certain sea area in East Asia at low latitudes and aeromagnetic data of a certain area in Baoding, Hebei Province at mid-to-high latitudes.

[0045] In one area, the original magnetic data point spacing in a low-latitude East Asian sea region was approximately 2 km, and the re-gridted point spacing was 1.1 km. The magnetization declination and magnetization dip angle within the survey area were both near 0°. A uniform magnetization declination and dip angle were used for polarization processing throughout the area. The magnetic anomaly and polarization results for this sea region are as follows: Figure 7 As shown, where Figure 7 (a) shows magnetic survey data for a certain area. The coordinates in the figure represent the number of data points, and the distance between adjacent points is 1.1 km. Figure 7 (b) shows the results of the suppression factor method for polarization. Figure 7 (c) shows the FCN polarization results driven by magnetic anomaly. Figure 7 (d) shows the FCN polarization results driven by the initial model. Figure 7 (e) shows the FCN polarization results driven by the combined "magnetic anomaly & initial model".

[0046] from Figure 7It can be seen that, before and after polarization using each method, the magnetic anomalies remain stable without distortion, but their morphology and amplitude have changed significantly. Overall, the polarization results using the FCN method are largely consistent with those using the suppression factor method. The FCN polarization results obtained by directly using magnetic anomalies show multiple closed loops of positive anomalies. Due to the unique characteristics of the East Asian seas, no geological or other geophysical data were available for corroboration. However, the main change in the polarization results of the other two methods compared to the original magnetic anomalies is the conversion from "positive / negative" anomalies to "negative / positive" anomalies, which is more consistent with the "phase inversion" concept proposed in the literature. The polarization results driven by both "data and prior information" provide richer details compared to those driven by the initial model.

[0047] A region in Baoding, Hebei Province, located in the mid-to-high latitudes, has a magnetization tilt of 58° and a magnetization declination of -7°. The magnetic anomaly and its polarization results for this region are as follows: Figure 8 As shown, where Figure 8 (a) shows magnetic anomaly data with a grid spacing of 100m. Figure 8 (b) is the result of the pseudo-tilt method for polarization. Figure 8 (c) shows the FCN polarization results driven by magnetic anomaly. Figure 8 (d) shows the FCN polarization results driven by the initial model. Figure 8 (e) shows the FCN polarization results driven by the combined "magnetic anomaly & initial model".

[0048] from Figure 8 It can be seen that the four methods yielded similar polarization results, and all four methods effectively polarized two large-scale, high-intensity anomalies (both above 100 nT) in the region. This is mainly manifested in two aspects: firstly, the two anomaly morphologies shifted northward; and secondly, the anomaly on the left side of the figure essentially converted both positive and negative anomalies into positive anomalies. The distribution of these two anomalies coincides with the distribution of northwest-trending and near-north-south-trending faults in the region. Furthermore, the locations of the two magnetic anomalies also show a good correspondence with the currently mined skarn-type metallic deposits. This demonstrates the good versatility of this invention, achieving the same polarization effect as traditional methods even in mid-to-high latitudes.

[0049] In summary, this invention can solve the problems of unstable polarization calculation and insufficient polarization accuracy in low-latitude magnetic anomalies in the original polarization method, and significantly improves the effectiveness, robustness and accuracy of the polarization results (from about 80% to more than 98%).

[0050] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for reducing low latitude magnetic anomalies, characterized in that, The method comprises the following steps: S1: constructing a model space of a sample data set, the model space being composed of an underground half space and a plurality of cuboid units, and the plurality of cuboid units being arranged in a combined form in a non-edge region of the underground half space; S2: performing traversal sampling on various combination types of the plurality of cuboid units, and performing sample data amplification of different magnetic susceptibility sizes by using the forward-simulated sample data to obtain sample data equivalent magnetic anomaly bodies; S3: obtaining a stable initial model by using a frequency domain suppression method on the sample data equivalent magnetic anomaly bodies; S4: taking the stable initial model and a magnetic anomaly of an arbitrary magnetization direction as inputs of an FCN network structure, taking a magnetic anomaly perpendicular to the magnetization direction as a label, and training the FCN network structure; S5: taking the measured data and the stable initial model as inputs of the trained FCN network structure to obtain a reduction-to-pole prediction result.

2. The method of reduction to the pole of low latitude magnetic anomalies of claim 1, wherein, In step S1, the combination form of the plurality of cuboid units is any one or more of a cuboid type, an inclined body type, an "L" type, and a "T" type.

3. The method of reduction to the pole of low latitude magnetic anomalies of claim 1, wherein, In step S3, before obtaining the stable initial model by using the frequency domain suppression method on the sample data equivalent magnetic anomaly bodies, noise is added to the data set of the sample data equivalent magnetic anomaly bodies.

4. The method of reduction to the pole of low latitude magnetic anomalies of claim 3, wherein, The specific method of adding noise is to add 5% Gaussian noise to 10% random samples in the data set.

Images