A coal seam lithium-rich seam detection method based on well logging curve and statistical analysis

Abstract

The application discloses a coal seam lithium-rich layer detection method based on well logging curves and statistical analysis, and belongs to the technical field of mineral resource exploration. The method comprises the following steps: selecting a series of representative coal seam positions for systematic sampling, and preparing the samples into piston samples and powder samples; determining the kaolinite and lithium content of the powder samples, and distinguishing lithium-rich and ordinary coal samples; systematically determining various geophysical response parameters of the piston samples, such as density, resistivity and wave velocity, and screening out sensitive discrimination indexes for lithium enrichment; setting a window size, calculating the moving statistical value curves of the density, resistivity, wave velocity and other well logging curves sensitive to the lithium-rich layer, calculating principal components, and using the selected principal components and a support micro machine to train a prediction model, so as to predict the lithium-rich layer in the coal seam. The application combines geophysical logging, statistical analysis, principal component analysis and a support vector machine, and provides a fast, economical and non-destructive means for lithium resource exploration in coal.

Description

Technical Field

[0001] This invention relates to the field of detecting lithium-rich coal seams, and more particularly to a method for detecting lithium-rich coal seams based on well logging curves and statistical analysis. Background Technology

[0002] The exploration of lithium resources in coal currently relies mainly on geochemical methods (such as rock sampling and laboratory analysis). Although this method can directly determine the lithium content, it has inherent drawbacks such as limited sampling density, low vertical resolution, long testing cycle, and high cost. It is difficult to achieve economical, efficient, and large-scale resource evaluation and detailed characterization of ore layers, which seriously restricts the large-scale exploration and development of coal-based lithium resources.

[0003] Meanwhile, geophysical methods, such as well logging, have advantages such as high vertical resolution, continuous non-destructive measurement, and relatively low cost, and have been widely used in coalfield exploration. This has inspired a new technical approach: since lithium in coal is mainly found in clay minerals such as kaolinite, its enrichment may lead to regular changes in the physical properties of coal seams, such as density, elasticity, electrical properties, and magnetism. Therefore, theoretically, lithium-rich ore layers can be indirectly identified and evaluated by detecting anomalies in these geophysical responses. Although some preliminary simulation studies and case analyses support this hypothesis, this technical approach still faces a key bottleneck: a lack of systematic experimental evidence and a universally applicable methodological framework.

[0004] Specifically, existing research has failed to clearly answer: (1) which geophysical parameters undergo significant changes in what characteristics due to lithium enrichment; (2) how to extract effective indicator parameters sensitive to lithium enrichment from complex well logging responses; and (3) how to establish a stable and operable data processing and interpretation process to achieve accurate identification of ore layers. Therefore, there is an urgent need to develop a detection method based on geophysical well logging data, integrating statistical theory and experimental verification, to overcome the limitations of traditional geochemical methods and provide reliable technical means for rapid, economical, and accurate exploration of lithium-rich coal seams. This invention is proposed against this background. Summary of the Invention

[0005] This solution addresses the problems and needs raised above by proposing a method for detecting lithium-rich coal seams based on well logging curves and statistical analysis. The method achieves the aforementioned technical objectives and brings about several other technical benefits due to the adoption of the following technical features.

[0006] One objective of this invention is to propose a method for detecting lithium-rich coal seams based on well logging curves and statistical analysis, comprising the following steps: S10: Within the same study block, based on the geological background and well logging response, a series of representative coal seam locations are selected for systematic sampling; S20: Prepare plunger samples and powder samples from the block samples collected in step S10; S30: Determine the kaolinite content and lithium content of the powder sample obtained in step S20; wherein the kaolinite content is determined by X-ray diffraction (XRD) technology to obtain the kaolinite content in the sample; the lithium content is determined by inductively coupled plasma mass spectrometry (ICP-MS); based on the lithium content determination results, combined with the geological definition threshold of "lithium-rich", all plunger-shaped samples are objectively and quantitatively divided into two groups: "lithium-rich coal samples" and "ordinary coal samples". S40: Determines various geophysical responses of plunger samples, including density DEN, dielectric constant DC, magnetic susceptibility MS, resistivity RES, and P-wave velocity. P and transverse wave velocity S; Based on the measurement results of the longitudinal wave velocity and the transverse wave velocity, the wave velocity ratio of the sample is calculated. P / S; S50: Statistical analysis of various geophysical response parameters of plunger samples; among which statistical parameters include median, mean, range, and deviation between median and extreme values; by comparing lithium-rich coal samples and ordinary coal samples, parameters that are statistically significant in response to the "lithium-rich" state are screened out. S60: On the actual logging curve, a depth window is set based on the typical thickness range of the target coal seam. Its length is determined according to the typical thickness range of the target coal seam and is usually not less than 10 times the sampling interval of the logging curve. The window is moved along the wellbore, and various movement statistics of the curve within the window are calculated. S70: Perform principal component analysis (PCA) on all moving average curves; select principal components with a cumulative variance contribution rate of not less than 95% as the final characteristic variables; S80: Using the obtained principal component features as input and the known "lithium-rich" or "ordinary" sample labels as output, train a support vector machine classification model; use the trained model to predict unknown well sections in the entire study block and identify potential lithium-rich ore layers.

[0007] Preferably, in step S20, the plunger sample is cut into a plunger sample with a diameter of 35-40 mm and a height of 70-80 mm using wire cutting technology, ensuring that the two end faces of the plunger sample are parallel to and smooth with the coal seam bedding; the powder sample is obtained by collecting the debris generated during the wire cutting process and grinding it through a 200-mesh sieve.

[0008] Preferably, in step S40, obtaining multiple geophysical responses includes the following steps: S41: Density (DEN) of the test sample; S411: Use an electronic balance to measure the mass m of the plunger sample, in g; use a vernier caliper to measure the diameter d (in cm) and height H (in cm) of the plunger sample multiple times; take the average values; S412: Using the average values of diameter d and height H, according to the cylinder volume formula V calculate the volume V of the plunger-shaped sample, in cm³; then according to the density formula , calculate the density DEN of the sample, in g / cm³; S42: Test the dielectric constant DC of the sample; S421: Make the probe of the dielectric constant meter in close contact with two surfaces of the fully dried plunger-shaped sample, read and record the measured dielectric constant value; S422: Average the above measured dielectric constant values to obtain the dielectric constant DC of the sample; S43: Test the magnetic susceptibility MS of the sample; S431: Make the probe of the magnetic susceptibility meter in close contact with two bottom surfaces of the fully dried plunger-shaped sample, read and record the measured magnetic susceptibility values of the two surfaces; S432: Average the above two measured magnetic susceptibility values to obtain the magnetic susceptibility MS of the sample; S44: Test the resistivity RES of the sample; S441: Evenly apply conductive adhesive on the two bottom surfaces of the plunger sample, then stably place the sample into the four-terminal electrode fixture of the DC resistance tester, ensure good contact between the two bottom surfaces and the electrodes, connect the current terminals and voltage terminals to the corresponding ports of the bridge respectively, and test the resistance R of the sample, in Ω; S442: Take the height H of the sample in step S411 as the propagation distance L of direct current, in cm; use the average value of the multiple measured diameters d to calculate the average cross-sectional area A, in cm²; where A , use the resistivity formula , calculate the DC resistivity RES of the sample, in Ω·m; S45: Test the wave velocity of the sample , in m / s; S451: Provide an intelligent automatic ultrasonic P-wave and S-wave tester; S452: Input the height H of the plunger-shaped sample in step S411 as the propagation distance L of ultrasonic waves in the sample into the software matched with the above automatic tester; S453: Apply ultrasonic couplant on the two bottom surfaces of the sample, stably install the sample on the sample holder of the tester, ensure that the transmitting end and receiving end transducers are in close contact with the end faces of the sample, align and press the transmitting probe and receiving probe of the instrument tightly against the two end faces of the sample respectively, operate the software to trigger ultrasonic pulse emission, and obtain the ultrasonic P-wave waveform and ultrasonic S-wave waveform; S454: To ensure data reliability, data can be collected 2-3 times at the same location; S455: Read the arrival time of the longitudinal wave tp, µs and the arrival time of the first wave of the transverse wave ts, µs; S456: Using the formulas respectively and The longitudinal wave velocity was calculated. P, m / s and transverse wave velocity S, m / s; S457: Calculate wave velocity ratio P / S.

[0009] Preferably, in step S50, the steps of statistically analyzing various geophysical response parameters of the plunger sample and screening out parameters that are statistically significant in response to the "lithium-rich" state are as follows: S51: Statistical Analysis of Multiple Geophysical Response Parameters of Plunger Samples S511: The geophysical response parameters of the plunger samples obtained in step S40 are divided into two independent datasets based on lithium-rich coal samples and ordinary coal samples. The geophysical response parameter data of lithium-rich coal samples is the experimental group; the geophysical response parameter data of ordinary coal samples is the control group. S512: Calculate the median, mean, maximum, minimum, range (Max-Min), deviation between median and maximum (Med-Max), and deviation between median and minimum (Med-Min) of various geophysical response parameters for the experimental and control groups respectively. S52: Screen out parameters that are statistically significant in response to the “lithium-rich” state; S521: For each geophysical response parameter, use an appropriate statistical hypothesis testing method to determine whether there is a statistically significant difference between the experimental group and the control group; S522: For parameters that pass the significance test, further calculate their effect size to assess the practical significance of the difference; S523: Screening for sensitive parameters that show a significant response to the "lithium-rich" state.

[0010] Preferably, in step S60, the moving statistics of the curve within the window include: moving mean, moving median, moving range, moving standard deviation, moving coefficient of variation, moving skewness, and moving kurtosis.

[0011] Compared with the prior art, the present invention has the following beneficial effects: 1) A complete technological closed loop is formed, and the methodology is reliable and highly operable. This invention constructs a complete technical chain from "geological sampling - experimental measurement - statistical screening - feature construction - intelligent prediction". The method uses standardized sample preparation (S10-S20) and high-precision geochemical-geophysical experimental analysis (S30-S40) as its data foundation, with subsequent statistical analysis (S50), feature engineering (S60-S70), and machine learning modeling (S80) all based on this reliable data. This design allows the entire method to move beyond theoretical concepts, possessing clear experimental support and repeatable operational steps, providing a practical and feasible technical path for large-scale, engineered exploration of coal-bearing lithium resources.

[0012] 2) Reveal and utilize the core detection mechanism to achieve targeted indirect detection. The fundamental innovation of this method lies in the systematic verification and application of the core mechanism: "lithium enrichment in kaolinite → causing regular changes in the physical properties of coal and petrology → manifesting as specific geophysical response anomalies." The correlation analysis of steps S30 and S40 quantitatively confirms this causal chain; step S50, through rigorous statistical testing, selects the physical parameters (such as density and resistivity) most sensitive to the "lithium-rich" state. This transforms the indirect identification of lithium-rich layers using conventional well logging curves from "empirical speculation" to "scientific judgment with clear physical basis and statistical evidence," greatly enhancing the rigor and reliability of the detection logic.

[0013] 3) Effective identification indicators improve detection accuracy. Unlike traditional methods that directly use the absolute values ​​of well logging curves, this invention creatively introduces "moving window statistics" in step S60 to generate a series of high-order statistical curves reflecting local formation structure and heterogeneity, such as mean, range, standard deviation, and skewness. Step S70 further purifies and reduces the dimensionality of these high-dimensional features through principal component analysis (PCA). Practice has shown that statistical quantities such as "range" and "median-extreme deviation," which reflect the dispersion and distribution of data, are often more sensitive to lithium-rich coal seams. This shift from "single response value" to "multidimensional statistical feature spectrum" greatly enriches the effective discriminative information, providing high-quality input for subsequent support vector machine (SVM) models, thereby achieving higher accuracy and resolution in the identification of lithium-rich ore seams.

[0014] 4) It enables a transformation in exploration methods that are efficient, low-cost, and non-destructive, with significant application value. The most direct effect of this invention is the successful transformation of the exploration model from relying on geochemical analysis, which involves intensive sampling, is destructive, costly, and time-consuming, to utilizing conventional geophysical logging data, which is continuous, non-destructive, and relatively inexpensive. Through in-depth analysis and intelligent interpretation of existing logging data, rapid prediction and evaluation of lithium-rich ore layers in unknown well sections can be achieved (S80). This not only significantly shortens the exploration cycle and reduces exploration costs, but also enables continuous evaluation of the entire well section, avoiding the risk of missed ore deposits. It has significant practical application value for rapidly evaluating the potential of coal-bearing lithium resources and delineating favorable target areas.

[0015] 5) The technological approach has broad scalability and significant strategic importance. This method, for the first time, systematically combines mature geophysical logging techniques, multivariate statistical theory, and cutting-edge machine learning algorithms, and has been successfully applied to the exploration of strategic coal-bearing metallic minerals. This technical framework has strong scalability; its principles and methods can be transferred to the detection and research of other associated minerals such as gallium, aluminum, and rare earth elements in coal-bearing strata. Therefore, this invention not only provides key technical support for the exploration of coal-bearing lithium resources, but also helps to improve the overall exploration technology level and guarantee capability of unconventional strategic mineral resources in my country, possessing significant industry-leading value and strategic significance. Attached Figure Description

[0016] Figure 1 This is a flowchart of a method for detecting lithium-rich coal seams based on well logging curves and geostatistical analysis, according to the present invention.

[0017] Figure 2 This is a diagram illustrating the processing method for irregular samples according to the present invention.

[0018] Figure 3 This is an X-ray diffraction pattern used to determine the kaolinite content of a sample in an embodiment of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. The described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the described embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0020] like Figure 1 The diagram shown is an overall flowchart of the method described in this invention, illustrating the complete technical process from geological sampling, experimental measurement, statistical analysis, feature engineering to intelligent prediction. The following embodiments will elaborate on each step in conjunction with this flowchart.

[0021] A method for detecting lithium-rich coal seams based on well logging curves and geostatistical analysis according to the present invention includes the following steps: S10: Within the same study block, based on the geological background and well logging response, a series of representative coal seam locations are selected for systematic sampling; S20: The sample is processed into two sizes to meet different testing conditions; the collected block sample is required to be prepared into plunger sample and powder sample; the block sample is cut into plunger sample with a diameter of 35-40 mm and a height of 70-80 mm using wire cutting technology, ensuring that the two end faces of the plunger sample are parallel to and smooth with the coal seam bedding; the debris generated during the wire cutting process is collected and ground through a 200-mesh sieve to obtain powder sample; like Figure 2 The diagram shows the preparation of standard plunger and powder samples from irregular raw coal samples by wire cutting, illustrating the key steps and final form of sample processing. S30: Determine the kaolinite content and lithium content of the powder sample; wherein the kaolinite content is determined by X-ray diffraction (XRD) technique, and the XRD pattern of sample 9-XJ01-9 is as follows. Figure 3 As shown, the spectrum was compared and analyzed with the standard diffraction spectrum of the International Center for Diffraction Data (ICDD). By identifying the characteristic diffraction peaks of kaolinite, the content of kaolinite in the sample was obtained. The lithium content was determined by inductively coupled plasma mass spectrometry (ICP-MS). Based on the lithium content determination results and combined with the geological definition threshold of "lithium-rich", all plunger-shaped samples were objectively and quantitatively divided into two groups: "lithium-rich coal samples" and "ordinary coal samples". S40: Measure various geophysical responses of the plunger sample; wherein the various geophysical responses include density DEN, dielectric constant DC, magnetic susceptibility MS, resistivity RES, and longitudinal wave velocity. P and transverse wave velocity S; Based on the measurement results of the longitudinal wave velocity and the transverse wave velocity, the wave velocity ratio of the sample is calculated. P / S; S50: Statistical analysis of various geophysical response parameters of plunger samples; among which statistical parameters include median, mean, range, and deviation between median and extreme values; by comparing lithium-rich coal samples and ordinary coal samples, parameters that are statistically significant in response to the "lithium-rich" state are screened out. S60: On the actual logging curve, a depth window is set based on the typical thickness range of the target coal seam. Its length is determined according to the typical thickness range of the target coal seam and is usually not less than 10 times the sampling interval of the logging curve. The window is moved along the wellbore, and various movement statistics of the curve within the window are calculated. S70: Perform principal component analysis (PCA) on all moving average curves; select principal components with a cumulative variance contribution rate of not less than 95% as the final characteristic variables; S80: Using the obtained principal component features as input and the known "lithium-rich" or "ordinary" sample labels as output, train a support vector machine classification model; use the trained model to predict unknown well sections in the entire study block and identify potential lithium-rich ore layers.

[0022] Specifically, in step S30, a specific lithium content threshold (T) is determined by comprehensively considering the existing background values ​​of lithium content in coal in the study area, the global average lithium content in coal (generally considered to be about 20-30 μg / g), and the potential economic boundary grade for industrial use. This threshold is used to distinguish between "lithium-rich" and "ordinary" coal samples. This threshold should be specified as a clear numerical value in the method (for example, T can be set to 80 μg / g, 100 μg / g, or adjusted according to the specific mining area).

[0023] By quantitatively determining the kaolinite and lithium content in coal powder samples, the geological hypothesis that "kaolinite is the key lithium-bearing mineral" was verified. Based on the lithium content determination results, the prepared plunger samples were objectively and quantitatively classified, providing an accurate geochemical scale and grouping basis for the subsequent establishment of a correlation model between geophysical response and "lithium-rich" state.

[0024] Specifically, in step S40, obtaining multiple geophysical responses includes the following steps: S41: Density (DEN) of the test sample; S411: Measure the mass m (in g) of the plunger sample using an electronic balance, and measure the diameter d (in cm) and height H (in cm) of the plunger sample multiple times using vernier calipers; take the average value. S412: Using the average of the diameter d and the height H, according to the formula for the volume of a cylinder, V... Calculate the volume V (unit: cm³) of the plunger-shaped sample, and then apply the density formula. Calculate the density DEN of the sample (unit: g / cm³). S42: Dielectric constant DC of the test sample; S421: Use the probe of the dielectric constant meter to make close contact with both sides of the fully dried plunger-shaped sample, and read and record the dielectric constant value; S422: Average the above dielectric constant measurements to obtain the dielectric constant DC of the sample; S43: Magnetic susceptibility MS of the test sample; S431: Use the probe of the magnetic susceptibility meter to make close contact with the two surfaces of the fully dried plunger-shaped sample, and read and record the magnetic susceptibility values ​​of the two surfaces. S432: Take the average of the two measured magnetic susceptibility values above to obtain the magnetic susceptibility MS of the sample; S44: Test the resistivity RES of the sample; S441: Evenly apply conductive adhesive on the two bottom surfaces of the plunger sample, then stably place the sample into the four-terminal electrode fixture of a DC resistance tester, ensure good contact between the two bottom surfaces and the electrodes, connect the current terminals and voltage terminals to the corresponding ports of the bridge respectively, and test the resistance R (unit: Ω) of the sample; S442: Take the height H of the sample in step S411 as the propagation distance L (unit: cm) of direct current; use the average value of multiple measured diameters d to calculate the average cross-sectional area A (unit: cm²); where A , use the resistivity formula to calculate the DC resistivity RES (unit: Ω·m) of the sample; S45: Test the wave velocity of the sample (unit: m / s); S451: Provide an intelligent automatic ultrasonic P-wave and S-wave tester; S452: Input the height H of the plunger-shaped sample in step S411 as the propagation distance L of the ultrasonic wave in the sample into the software supporting the above-mentioned automatic tester; S453: Apply ultrasonic coupling agent on the two bottom surfaces of the sample, stably install the sample on the sample holder of the tester, ensure that the transmitting end and receiving end transducers are in close contact with the end faces of the sample, align and press the transmitting probe and receiving probe of the instrument against the two end faces of the sample respectively, operate the software to trigger ultrasonic pulse emission, and obtain the ultrasonic P-wave waveform and ultrasonic S-wave waveform; S454: To ensure data reliability, the acquisition can be repeated 2-3 times at the same position; S455: Read the P-wave arrival time tp (unit: µs) and the first arrival time ts of the S-wave (unit: µs); S456: Respectively through the formula and calculate the P-wave velocity P (unit: m / s) and S-wave velocity S (unit: m / s), where the unit of the propagation distance L is cm; S457: Calculate the wave velocity ratio P / S.

[0025] Specifically, system tests and analysis results show that there are significant differences in the mineral composition and geophysical response test data between lithium-rich coal and ordinary coal, the response data of lithium-rich coal has higher discreteness, while the response data of ordinary coal is relatively concentrated.

[0026] Specifically, in step S50, the steps for statistically analyzing various geophysical response parameters of the plunger sample and screening out parameters that are statistically significant in response to the "lithium-rich" state are as follows: S51: Statistical analysis of various geophysical response parameters of plunger samples; S511: The geophysical response parameters of the plunger samples obtained in step S40 are divided into two independent datasets based on lithium-rich coal samples and ordinary coal samples (determined by the lithium content determination results in step S30); the geophysical response parameter data of lithium-rich coal seam samples is the experimental group; the geophysical response parameter data of ordinary coal seam samples is the control group. S512: For each geophysical response parameter (density DEN, dielectric constant DC, magnetic susceptibility MS, resistivity RES, P-wave velocity) P, transverse wave velocity S, wave speed ratio P / S), for the experimental group and the control group, respectively, calculate various statistical parameters; Statistical parameters include median, mean, range, absolute deviation between median and maximum, absolute deviation between median and minimum, relative deviation between median and maximum, and relative deviation between median and minimum.

[0027] S52: Screen out parameters that are statistically significant in response to the “lithium-rich” state; S521: For each geophysical response parameter, use an appropriate statistical hypothesis testing method to determine whether there is a statistically significant difference between the experimental group and the control group; the selection of the testing method should be based on the data characteristics; Normality and homogeneity of variance tests: First, the Shapiro-Wilk test is used to determine whether the two sets of data conform to a normal distribution, and the Levene's test is used to determine whether the variances are homogeneous. Parametric test: If both sets of data conform to a normal distribution and have homogeneity of variance, use the independent samples t-test to compare the differences between the means of the two sets; Nonparametric test: If the data does not conform to a normal distribution or has unequal variances, the Mann-Whitney U test is used to compare the difference in medians between the two data distributions; Test criteria: The significance level was set at α = 0.05. When the p-value obtained from the hypothesis test was less than 0.05, the parameter was considered to have a statistically significant difference between the experimental group and the control group. S522: The parameters that pass the significance test are further calculated to assess the actual significance of the difference; For the t-test, calculate the Cohen d-value; For the Mann-Whitney U test, calculate the rank-bicol correlation coefficient r; An effect size threshold is set: d ≥ 0.5 or |r| ≥ 0.3 is considered a moderate or higher effect size, indicating that the difference has practical application value; S523: Select sensitive parameters that show a significant response to the "lithium-rich" state based on the following conditions: Statistical significance condition: p-value < 0.05; Effect size condition: The effect size reaches a moderate or higher level (d≥0.5 or |r|≥0.3). Geological consistency condition: The direction of parameter variation is consistent with the geological mechanism (kaolinite lithium enrichment mechanism) determined in step S30; Stability condition: The relative difference of the median of the parameter between the experimental group and the control group is ≥10%, and the range ratio (range of the experimental group / range of the control group) is in the range of 0.8-1.2, indicating that the difference is stable and not caused by outliers; S524: For the sensitive parameters that pass the screening, prioritize them according to the following rules: Level 1 sensitivity parameters: simultaneously satisfying p-value < 0.01, large effect size (d ≥ 0.8 or |r| ≥ 0.5), and median relative difference ≥ 15%; Secondary sensitivity parameters: p-value < 0.05, moderate effect size (d ≥ 0.5 or |r| ≥ 0.3), and median relative difference ≥ 10%; Level 3 sensitivity parameter: Only the p-value < 0.05 is met, but the effect size or degree of difference does not meet the above criteria; S525: Organize the selected sensitive parameters into a list according to priority, and record their statistical characteristics (median, mean, range, etc. of the experimental group and the control group), hypothesis testing results (test method, statistic, p-value), and effect size; this list will serve as the direct basis for determining the type of logging curve that needs to be statistically analyzed by moving window in step S60.

[0028] By conducting systematic statistical analysis and hypothesis testing on the geophysical response parameters of lithium-rich coal seam samples and ordinary coal seam samples, we screened out discriminant indicators that have significant indicative significance for lithium enrichment status, providing key input features for subsequent intelligent analysis of well logging curves.

[0029] Specifically, in step S60, the steps for calculating various movement statistics of the curve within the window are as follows: S61: Input data preparation and preprocessing; S611: Obtain the conventional logging digital curves for the target well section. The logging type should be the geophysical parameters that are sensitive to the "lithium-rich" state selected in step S50. S612: Perform necessary preprocessing on well logging curves to eliminate instrument noise and local abrupt changes to ensure data quality; S62: Determine the parameters for moving the window; S621: Using core or geological logging data from existing boreholes in the study area, statistically analyze the thickness distribution of lithium-rich coal seams. Set the window length W to 0.5 to 1.5 times the typical thickness of the target coal seam, and not less than 10 times the sampling interval of the logging curve. Preferably, the window length W should match the peak thickness of the lithium-rich coal seam thickness distribution in the study area (e.g., 0.5 m, 1.0 m) to ensure statistical significance. S622: To generate continuous characteristic curves, the moving step size S is usually set to the original sampling interval of the logging curve (e.g., 0.125 m, 0.1 m, or 0.05 m). S63: Along the depth direction, starting from the top of the well section, slide a window of length W downwards with a step size S, and calculate the sequence of original logging data points falling into the window; For each window position, calculate the moving mean, moving median, moving range, moving standard deviation, moving coefficient of variation, moving skewness, and moving kurtosis; S64: Take the various statistical values ​​calculated at the center depth point d of each window as the new characteristic value of that depth point; repeat the sliding calculation until the bottom of the well section, and finally generate a series of corresponding depth-aligned moving statistical value curves for each original sensitive logging curve; The output is a multidimensional feature dataset with the following data structure: indexed by the depth sequence, each depth point corresponds to a high-dimensional feature vector composed of various moving statistics of all original sensitive logging curves; this dataset will be used as the direct input for principal component analysis in step S70.

[0030] Specifically, in step S70, the steps of using the principal component analysis (PCA) method are as follows: Align all moving statistics curves output from step S60 by depth to construct an initial feature matrix X; perform Z-score standardization on each column (each feature) of matrix X to obtain the standardized feature matrix Z; calculate the correlation matrix R of the standardized matrix Z, and solve for the eigenvalues ​​and eigenvectors of R. Calculate the variance contribution rate and cumulative variance contribution rate of each principal component and determine the number of principal components. The variance contribution rate of the kth principal component is the proportion of the eigenvalue to the sum of all eigenvalues. Based on the preset threshold, select the minimum K value that makes the cumulative variance contribution rate reach or exceed 95% for the first time. Construct the principal component feature matrix F (score matrix), and output the principal component score matrix F as the final set of feature variables.

[0031] Specifically, step S80 is as follows: S81: Construct the training dataset; S811: Extract depth point samples with complete principal component features (output from S70) from known boreholes where steps S10-S70 have been completed; calculate the K principal component scores (PC1, PC2, ..., PC5) for each depth point sample. K ) as feature vectors; S812: Based on the classification results of step S30, assign a class label to each depth point; lithium-rich samples are labeled as positive (+1), and ordinary samples are labeled as negative (-1), and construct the training dataset; S82: Data processing and sample balancing; S821: Normalize each principal component feature in the training dataset to unify its value range to the interval [0,1] or [-1,1], so as to eliminate the influence of the difference in numerical scale of different principal components on the SVM model. S822: To address the imbalance issue in coal-based lithium resources where "lithium-rich" samples are typically far fewer than "ordinary" samples, a higher misclassification penalty weight C is set for positive class samples (lithium-rich) in the SVM objective function. + ; S823: Perform reasonable interpolation on minority class (lithium-rich) samples in the feature space to generate synthetic samples to balance the class distribution; S83: Construct a support vector machine model; The training dataset is divided into N subsets. N-1 subsets are used as training subsets and 1 subset is used as validation subset in turn. The performance indicators of the model on the validation set are calculated, such as the average classification accuracy of more than 95% and the average F1 score of more than 0.9. The optimal parameter combination is then selected. S84: Model Training and Evaluation; S841: Train the final SVM classification model using the optimal parameter combination and the entire training dataset; S842: Evaluate model performance on independent test datasets; for example, confusion matrix analysis, with a particular focus on the ability to identify “lithium-rich” classes; S85: Application for predicting unknown well sections; S851: For the logging data of the unknown well section to be predicted, execute steps S60 (moving window statistics) and S70 (principal component analysis) in sequence to obtain the K-dimensional principal component feature vector of each depth point; S852: Normalize the features to be predicted to the same scale; S853: Input the normalized feature vector into the trained SVM model to obtain the predicted class label (lithium-rich / normal) and decision function value for each depth point; S854: Convert the SVM decision function value into a posterior probability estimate to obtain the probability value of each depth point belonging to the "lithium-rich" category; S86: Compare the prediction results with data such as geological logging and core description to ensure that the predicted lithium-rich layers are geologically reasonable in terms of lithology and sedimentary facies; S87: Model Update and Iteration; S871: When experimentally proven lithium-rich or ordinary samples are obtained in new boreholes, they are added to the training dataset. S872: Periodically or when prediction performance deteriorates, re-execute steps S82-S84 to retrain the model and optimize parameters; S873: Establish a "sample-feature-model" knowledge base for research blocks to enable continuous learning and performance improvement of models.

[0032] By combining the powerful machine learning algorithm Support Vector Machine (SVM) with previously extracted principal component features and known geochemical analysis results, an intelligent classification model capable of automatically identifying lithium-rich coal seams was constructed. This method not only enables rapid evaluation of existing boreholes but also allows for high-precision prediction of unknown sections in new boreholes, significantly improving the efficiency and accuracy of coal-bearing lithium resource exploration and providing a reliable technical means for intelligent mineral resource exploration.

[0033] The foregoing description, with reference to preferred embodiments, details an exemplary implementation of the method for detecting lithium-rich coal seams based on well logging curves and geostatistical analysis proposed in this invention. However, those skilled in the art will understand that various modifications and alterations can be made to the above specific embodiments without departing from the concept of this invention, and various combinations can be made to the various technical features and structures proposed in this invention without exceeding the protection scope of this invention, which is determined by the appended claims.

Claims

1. A method for detecting lithium-rich coal seams based on well logging curves and statistical analysis, characterized in that, Specifically, the following steps are included: S10: Within the same study block, based on the geological background and well logging response, a series of representative coal seam locations are selected for systematic sampling; S20: Prepare plunger samples and powder samples from the samples collected in step S10; S30: Determine the kaolinite content and lithium content of the powder sample obtained in step S20; wherein the kaolinite content is determined by X-ray diffraction technology to obtain the kaolinite content in the sample; the lithium content is determined by inductively coupled plasma mass spectrometry; based on the lithium content determination results and combined with the geological definition threshold of "lithium-rich", all plunger-shaped samples are divided into two groups: "lithium-rich coal sample" and "ordinary coal sample". S40: Determine various geophysical responses of plunger samples, including density DEN, dielectric constant DC, magnetic susceptibility MS, resistivity RES, and P-wave velocity. P and transverse wave velocity S, based on the measurement results of the longitudinal wave velocity and the transverse wave velocity, calculate the wave velocity ratio of the sample. P / S; S50: Statistical analysis of various geophysical response parameters of plunger samples; including various statistical parameters such as median, mean, range, and deviation between median and extreme values; by comparing lithium-rich coal samples and ordinary coal samples, parameters that are statistically significant in response to the "lithium-rich" state are screened out. S60: On the actual logging curve, a depth window is set based on the typical thickness range of the target coal seam. Its length is determined according to the typical thickness range of the target coal seam and is not less than 10 times the sampling interval of the logging curve. The window is moved along the wellbore, and various movement statistics of the curve within the window are calculated. S70: Perform principal component analysis on all moving average curves; select principal components with a cumulative variance contribution rate of not less than 95% as the final characteristic variables; S80: Using the obtained principal component features as input and the known "lithium-rich" or "ordinary" sample labels as output, train a support vector machine classification model; use the trained model to predict unknown well sections in the entire study block and identify potential lithium-rich mineral layers.

2. The method for detecting lithium-rich coal seams based on well logging curves and statistical analysis according to claim 1, characterized in that, In step S20, the plunger sample is cut into a plunger sample with a diameter of 35-40 mm and a height of 70-80 mm using wire cutting technology, ensuring that the two end faces of the plunger sample are parallel to and smooth with the coal seam bedding; the powder sample is obtained by collecting the debris generated during the wire cutting process and grinding it through a 200-mesh sieve.

3. The method for detecting lithium-rich coal seams based on well logging curves and statistical analysis according to claim 1, characterized in that, Step S40, obtaining multiple geophysical responses, includes the following steps: S41: Density (DEN) of the test sample; S411: Use an electronic balance to measure the mass m, g of the plunger sample; and use vernier calipers to measure the diameter d, cm and height H, cm of the plunger sample multiple times; take the average value. S412: Using the average of the diameter d and the height H, according to the formula for the volume of a cylinder, V... Calculate the volume V, cm³, of the plunger-shaped sample; then apply the density formula. Calculate the density of the sample, DEN, in g / cm³; S42: Dielectric constant DC of the test sample; S421: Use the probe of the dielectric constant meter to make close contact with both sides of the fully dried plunger-shaped sample, and read and record the dielectric constant value; S422: Average the above dielectric constant measurements to obtain the dielectric constant DC of the sample; S43: Magnetic susceptibility MS of the test sample; S431: Use the probe of the magnetic susceptibility meter to make close contact with the two bottom surfaces of the fully dried plunger-shaped sample, and read and record the magnetic susceptibility values ​​of the two surfaces. S432: The magnetic susceptibility values ​​of the above two measurements are averaged to obtain the magnetic susceptibility MS of the sample; S44: Resistivity RES of the test sample; S441: Evenly apply conductive adhesive on two bottom surfaces of the plunger sample, then stably place the sample into the four-terminal electrode fixture of the DC resistance tester to ensure good contact between the two bottom surfaces and the electrodes, connect the current terminals and voltage terminals to the corresponding ports of the bridge respectively, and test the resistance R of the sample, in Ω; S442: Take the height H of the sample in step S411 as the propagation distance L of the direct current, in cm; use the average value of the measured diameters d; calculate the average cross-sectional area A, in cm²; where A Using the resistivity formula Calculate the DC resistivity RES of the sample, in Ω·m; S45: Wave velocity of the test sample , m / s; S451: Provide an intelligent ultrasonic longitudinal and transverse wave automatic tester; S452: Input the height H of the plunger-shaped sample obtained in step S411 as the propagation distance L of ultrasonic waves in the sample into the software matched with the above-mentioned automatic tester; S453: Apply ultrasonic couplant on two bottom surfaces of the sample, stably mount the sample on the sample holder of the tester to ensure that the transmitting-end and receiving-end transducers are in close contact with the end faces of the sample, align and press the transmitting probe and the receiving probe of the instrument against the two end faces of the sample respectively, operate the software to trigger ultrasonic pulse emission, and acquire the ultrasonic longitudinal wave waveform and the ultrasonic transverse wave waveform; S454: To ensure data reliability, the acquisition can be repeated 2-3 times at the same position; S455: Read the arrival time tp of the longitudinal wave, in μs, and the arrival time ts of the first arrival of the transverse wave, in μs; S456: Using the formulas respectively and The longitudinal wave velocity was calculated. P, m / s and transverse wave velocity S, m / s; S457: Calculate wave velocity ratio P / S.

4. The method for detecting lithium-rich coal seams based on well logging curves and statistical analysis according to claim 1, characterized in that, In step S50, the steps of statistically analyzing multiple geophysical response statistical parameters of the plunger sample and screening out parameters with statistically significant response to the "lithium-rich" state are as follows: S51: Statistically analyze multiple geophysical response statistical parameters of the plunger sample; S511: Divide the geophysical response parameters of the plunger sample obtained in step S40 into two independent data sets according to lithium-rich coal samples and ordinary coal samples, wherein the geophysical response parameter data of lithium-rich coal samples is set as the experimental group, and the geophysical response parameter data of ordinary coal samples is set as the control group; S512: Calculate parameters including the median, mean, maximum value, minimum value, range, deviation between the median and the maximum value, and deviation between the median and the minimum value for each geophysical response parameter of the experimental group and the control group respectively; S52: Screen out parameters with statistically significant response to the "lithium-rich" state; S521: For each geophysical response parameter, adopt an appropriate statistical hypothesis testing method to determine whether there is a statistically significant difference between the experimental group and the control group; S522: For the parameters that pass the significance test, further calculate their effect size to evaluate the practical significance of the difference; S523: Screen out sensitive parameters with significant response to the "lithium-rich" state.

5. The method for detecting lithium-rich coal seams based on well logging curves and statistical analysis according to claim 1, characterized in that, In step S60, the moving statistical values of the curve within the window include: moving mean, moving median, moving range, moving standard deviation, moving coefficient of variation, moving skewness and moving kurtosis.

Images