Element anomaly identification method based on equidistant logarithm ratio and Canopy-K-Means clustering
By processing geochemical data using the equidistant logarithmic ratio and Canopy-K-Means clustering methods, the subjectivity and high computational cost of identifying element background and anomalies in traditional methods are solved, and more efficient element anomaly identification and distribution feature analysis are achieved.
Patent Information
- Application Number
- CN202411651572.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-19
- Publication Date
- 2025-10-10
AI Technical Summary
Traditional methods have problems of subjectivity and high computational cost when identifying geochemical element background and anomalies, and are particularly inefficient on large-scale data sets. Existing methods also have difficulty solving problems of multiple samples, multiple elements, and multiple categories at the same time.
The data were transformed by equal-interval log-ratio and combined with Canopy-K-Means clustering method. Element anomalies were identified through pre-clustering and K-means clustering. The data were processed by equal-interval log-ratio to conform to Gaussian distribution. Canopy-K-Means was used to improve clustering efficiency and identify element background and anomalies.
It achieves more accurate identification of element distribution features, improves recognition efficiency, realizes unsupervised intelligent extraction of element anomalies, and reduces computing costs.
Smart Images
Figure CN120766818A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of exploration geochemistry technology, and in particular to an element anomaly recognition method based on equidistant logarithmic ratios and Canopy-K-Means clustering. Background Art
[0002] Identifying background and anomalies is an important part of geochemical exploration. Accurately identifying background and anomaly thresholds can effectively understand the distribution characteristics of elements and provide decision support for mineral exploration. Traditionally, the identification of element background anomalies is determined by experts through statistical analysis of raw concentration data based on their own past experience, such as the mean deviation iteration method ( ), where n can be 1, 2, 2.5, 3, 4, etc. The specific value is determined by the expert's own experience, which leads to the subjectivity of the background and anomaly thresholds. Different experts have different anomaly recognition results for the same batch of data, which affects the results of anomaly recognition and target area delineation.
[0003] The original concentration data is component data. Closing operations cause the concentration to not be an independent random variable, making its distribution inconsistent with the Gaussian distribution and, therefore, inconsistent with the requirements of some mathematical statistics methods. Direct statistical analysis of the original concentration data may produce erroneous conclusions. Before analyzing the concentration component data, it needs to be preprocessed. There are three main methods for converting component data: additive log ratio (ALR), central log ratio (CLR), and equal-interval log ratio (ILR). Each method has its own advantages and disadvantages. The ALR method is simple and convenient, but the generated data is not equal-interval and has one less dimension than the original data. The CLR data is equal-interval, but it will form singular covariance. The ILR data is non-singular equal-interval and equal-dimensional, but it is more difficult to calculate and difficult to interpret.
[0004] K-means is an effective method for mining data patterns. Unsupervised machine learning can learn patterns hidden in unlabeled data. K-means clustering does not require any prior knowledge of the data and only clusters the high-dimensional space into clusters according to rules. n The data are separated into appropriate clusters according to the same characteristics. The ideal classification result is that each class of data contains only similar samples and has obvious differences from samples in other classes. However, when the data set to be clustered is large, the computational cost of traditional clustering algorithms is very high. Geochemical data sets can become large in three ways: (1) multiple samples, (2) multiple elements, and (3) multiple classes. Currently, some methods partially solve these problems, but no method can solve all three problems at the same time. For example, KD trees provide efficient EM-style clustering, but require that each sample has few elemental features. Scaling clustering can find the initial starting point well, but it is not efficient when the number of clusters is large.
[0005] To this end, a method for element anomaly recognition based on equidistant log ratio and Canopy-K-Means clustering is proposed. Summary of the Invention
[0006] The present invention aims to solve the above problems and provide an element anomaly recognition method based on equidistant logarithmic ratios and Canopy-K-Means clustering.
[0007] The present invention achieves the above-mentioned purpose through the following technical solutions: a method for identifying element anomalies based on isometric logarithmic ratios and Canopy-K-Means clustering, comprising: performing ILR conversion of chemical components; performing ILR conversion of isometric logarithmic ratios; Perform Canopy pre-clustering; Perform K-means clustering; use the concentration values corresponding to the boundary values of the K-means clustering as the background and abnormal thresholds of different levels.
[0008] The specific steps of the present invention are as follows:
[0009] S1: Record all samples as set D and calculate the logarithmic ratios of the concentrations of the components to be studied. , where m is the component to be studied and n is the number of components;
[0010] S2: Set the appropriate and , where T2 < T1;
[0011] S3: Randomly extract a sample p from the sample set D and create a canopy class containing the sample;
[0012] S4: Calculate the distance between any remaining sample q and p in D ,if , then assign sample q to the canopy class;
[0013] S5: If , then remove sample q from D;
[0014] S6: Repeat S3, S4, S5 and S6 until there are no samples in D, and count the number k and center of all canopies , where N is the number of samples in the corresponding canopy class;
[0015] S7: The number of canopy classes k and the center obtained in S6 Initialize K-means clustering;
[0016] S8: Calculate the square of the distance between each sample and each center , classify the sample into the class with the smallest distance;
[0017] S9: Update the member samples of the classification and recalculate the classification center ,in, For class The number of samples;
[0018] S10: Repeat S8 and S9 until no samples can be moved between categories, and record the center and boundary values of each category;
[0019] S11: Arrange the original concentrations corresponding to the boundary values between categories in order of size as abnormalities of different levels.
[0020] Compared with the existing technical methods, the present invention can achieve the following beneficial effects:
[0021] The present invention processes the original concentration data by equidistant logarithmic ratio, and the converted ilr(m) data obeys Gaussian distribution. Its statistical results can better reflect the element distribution characteristics than the original concentration data, and the information dimension is better than the additive logarithmic ratio. The parameters k and C of K-Means are adjusted by Canopy technology. k Preprocessing can improve the efficiency of K-Means clustering; identifying geochemical element background anomalies through K-means clustering can achieve unsupervised intelligent extraction of anomaly information. BRIEF DESCRIPTION OF THE DRAWINGS
[0022] FIG1 is a schematic diagram of the Canopy of the present invention;
[0023] Figure 2 is a flow chart of the present invention;
[0024] FIG3 is a scatter plot of the original concentration data of the embodiment;
[0025] Figure 4 shows Clustering result diagram of the data. DETAILED DESCRIPTION
[0026] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. The following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.
[0027] A method for identifying element anomalies based on equidistant logarithmic ratios and Canopy-K-Means clustering includes the following steps:
[0028] S1: Record all samples as set D and calculate the logarithmic ratios of the concentrations of the components to be studied. , where m is the component to be studied and n is the number of components;
[0029] S2: Set the appropriate and , where T2 < T1;
[0030] S3: Randomly extract a sample p from the sample set D and create a canopy class containing the sample;
[0031] S4: Calculate the distance between any remaining sample q and p in D ,if , then assign sample q to the canopy class;
[0032] S5: If , then remove sample q from D;
[0033] S6: Repeat S3, S4, S5 and S6 until there are no samples in D, and count the number k and center of all canopies , where N is the number of samples in the corresponding canopy class;
[0034] S7: The number of canopy classes k and the center obtained in S6 Initialize K-means clustering;
[0035] S8: Calculate the square of the distance between each sample and each center , classify the sample into the class with the smallest distance;
[0036] S9: Update the member samples of the classification and recalculate the classification center ,in, For the class The number of samples;
[0037] S10: Repeat S8 and S9 until no sample can be moved between classes, record the center and boundary values of each class;
[0038] S11: Arrange the original concentrations corresponding to the boundary values between classes in size as different levels of anomalies.
[0039] In addition, it should be understood that although the present specification is described in terms of embodiments, not every embodiment contains only one independent technical solution, and the description of the specification is only for the sake of clarity, and those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can be appropriately combined to form other embodiments that those skilled in the art can understand.
Claims
1. A method for identifying element anomalies based on equidistant log-ratios and Canopy-K-Means clustering, characterized by: The chemical components were converted to ILR using the compositional data theory. The equidistant logarithmic ratios ilr(m) were pre-clustered using Canopy. K-means clustering was performed on ilr(m) based on the Canopy pre-clustering results. The concentration values corresponding to the boundary values of the K-means clustering were used as the background and anomaly thresholds of different levels.
2. The element anomaly identification method according to claim 1, characterized in that The specific steps include: S1: Record all samples as set D and calculate the logarithmic ratios of the concentrations of the components to be studied. , where m is the component to be studied and n is the number of components; S2: Set the appropriate and , where T2 < T1; S3: Randomly extract a sample p from the sample set D and create a canopy class containing the sample; S4: Calculate the distance between any remaining sample q and p in D ,if , then assign sample q to the canopy class; S5: If , then remove sample q from D; S6: Repeat S3, S4, S5 and S6 until there are no samples in D, and count the number k and center of all canopies , where N is the number of samples in the corresponding canopy class; S7: The number of canopy classes k and the center obtained in S6 Initialize K-means clustering; S8: Calculate the square of the distance between each sample and each center , classify the sample into the class with the smallest distance; S9: Update the member samples of the classification and recalculate the classification center ,in, For class The number of samples; S10: Repeat S8 and S9 until no samples can be moved between categories, and record the center and boundary values of each category; S11: Arrange the original concentrations corresponding to the boundary values between categories in order of size as abnormalities of different levels.
3. The element anomaly identification method according to claim 1 or 2, characterized in that: The step S1 specifically includes selecting the element m to be studied, C i This can be the concentration of all other elements actually tested, or a constrained constant of 100% - C m .
4. The element anomaly identification method according to claim 1 or 2, characterized in that: T1 and T2 in step S2 are determined according to the actual distribution of ilr(m).
5. The element anomaly identification method according to claim 1 or 2, characterized in that: The element anomaly recognition method integrates component data ILR conversion, Canopy pre-clustering and K-means clustering technology. The recognition results are not affected by expert experience, and the element background and anomaly recognition process is intelligent and efficient.