Improved integrated learning rock debris element content analysis system and method based on laser spectrum information

Abstract

The invention discloses an improved integrated learning rock debris element content analysis system and method based on laser spectrum information. The method comprises the following steps: step 1, collecting a rock debris element LIBS spectrum data set; 2, dividing the rock debris element LIBS spectral data set into a training set and a test set, and carrying out noise reduction processing on original spectral data; 3, selecting characteristic spectral lines of the rock debris analysis elements according to the correlation coefficients, then selecting correlation interference spectral lines of the analysis elements, respectively combining the characteristic spectral lines with different correlation interference spectral lines to obtain more than one spectral line combination, and taking the spectral line combination as input of an integrated learning device; 4, constructing an ant colony optimization algorithm to optimize and improve the least square support vector machine ensemble learning base learner as an ensemble learner; and 5, designing a weighted distribution integration strategy of the base learners, and completing the optimal combination of the base learners. According to the method provided by the invention, a strong learner is constructed by combining a plurality of base learners, so that the accuracy and robustness of prediction are improved.

Description

Technical Field

[0001] The present invention relates to a method for analyzing the elemental content of cuttings, and specifically to an improved integrated learning system and method for analyzing the elemental content of cuttings based on laser spectral information. Background Art

[0002] In the field of oil and gas exploration and development, cuttings logging is a crucial task, which involves analyzing the elemental composition and its content from the cuttings samples collected during the drilling process, providing key first-hand data for the evaluation and exploitation of oil and gas resources.

[0003] In recent years, laser-induced breakdown spectroscopy (LIBS) technology, as an emerging elemental analysis method, has gradually attracted attention. It can be used for qualitative elemental detection for geochemical fingerprint analysis, sample classification and identification, as well as quantitative geochemical analysis, rock characterization through particle size analysis, and in-situ geochemical imaging. However, LIBS technology faces challenges such as matrix effects, spectral line interference, and self-absorption effects in the analysis of cuttings samples.

[0004] To overcome these challenges and improve the accuracy and reliability of LIBS technology in the analysis of cuttings elemental content, researchers have begun to explore integrated learning methods, combining the advantages of multiple algorithms to reduce the bias and uncertainty brought by a single algorithm, thereby improving the accuracy of elemental content analysis. However, existing integrated learning methods have problems such as insufficient information utilization, which cannot eliminate various spectral line interferences in cuttings samples, and complex parameter tuning, which limits their application in real-time analysis.

[0005] It should be noted that the information disclosed in the above background art section is only used to enhance the understanding of the background of the present disclosure, and thus may include information that does not constitute the prior art known to those of ordinary skill in the art. Summary of the Invention

[0006] The object of the present invention is to provide an improved integrated learning system and method for analyzing the elemental content of cuttings based on laser spectral information, which solves the defect of low accuracy of cuttings elements in the prior art, realizes high-precision content analysis of cuttings elements, and integrated learning can reduce the bias problem that may exist in a single model. The prediction results of multiple models can complement each other, reducing the error of a single model on a specific dataset.

[0007] To achieve the above object, the present invention provides an improved integrated learning method for analyzing the elemental content of cuttings based on laser spectral information, and the method includes: Step 1: Collect a LIBS spectral dataset of cuttings elements using a laser-induced breakdown spectrometer; Step 2: Divide the cuttings element LIBS spectral dataset into a spectral data training set and a spectral data test set, and perform noise reduction processing on the original spectral data; Step 3: Perform a correlation analysis on the full-spectrum data in the cuttings element LIBS spectral dataset and the cuttings element concentration. Select the characteristic spectral lines of the cuttings analysis elements based on the correlation coefficient, and then select the relevant interference spectral lines of the analysis elements. Combine the characteristic spectral lines with different relevant interference spectral lines to obtain more than one spectral line combination, and use more than one spectral line combination as the input of the ensemble learning algorithm; Step 4: Construct an ant colony optimization algorithm to optimize and improve the least squares support vector machine ensemble learning base learner as the ensemble learning algorithm: Use the least squares support vector machine LSSVM as the base learner of the ensemble learning model. The parameters of the LSSVM regression model include: the regularization parameter C , the kernel function parameter γ and the kernel function type; among them, the kernel function type is set to the radial basis function kernel, and the regularization parameter C , the kernel function parameter γ are optimized by the ant colony optimization algorithm to construct an ant colony optimization algorithm to optimize and improve the least squares support vector machine ensemble learning base learner; among them, the concentration value of the analysis element is predicted by the decision function of the LSSVM regression model, and the root mean square error RMSE is used as the performance index. According to RMSE , the smaller the value, the better the prediction performance of the model. Select the corresponding optimal regularization parameter C and the kernel function parameter γ for the input of different spectral line combinations; Step 5: Design a base learner weighted distribution integration strategy to complete the best combination of base learners, and obtain the final predicted element concentration value according to the weight distribution of the predicted values of all base learners.

[0008] Preferably, in the step 2, a 5-point and 3rd-order convolution smoothing method is used to perform noise reduction processing that retains the change information of the original spectral data with weights.

[0009] Preferably, in the step 2, the noise reduction processing includes: The fitting formula for a point in the cuttings element LIBS spectral dataset is: (12) In formula (12), represents the spectral intensity of the spectral data at this point; represents the weight parameter fitted by the least squares method; represents the corresponding spectral intensity wavelength value; The matrix form of the 5-point fitting is expressed as: (13) In formula (13), represents the absorption intensity of 5 consecutive points; represents a 5-row and 3-column matrix composed of the 0th, 1st, and 2nd values of the coordinates of 5 consecutive wavelength values; represents the matrix form of the weight parameters fitted by the least squares method; represents the matrix form of the error term; From formula (13), The least squares solution of , is: (14) The filtered value obtained by fitting , is: (15).

[0010] Preferably, in the first step, the laser-induced breakdown spectroscopy (LIBS) acquires the LIBS spectral dataset of cuttings according to the following method: under the same experimental conditions, 30 points at different positions on the surface of each sample are collected, and the average value of the 30 points is taken as the analysis data; or / and, in the first step, the spectral data acquisition and analysis band of the laser-induced breakdown spectroscopy is 178 - 987 nm; or / and, in the second step, the spectral data is divided into a spectral data training set and a spectral data test set according to a ratio of 4:1.

[0011] Preferably, in the third step, the method for selecting the characteristic spectral lines includes: selecting the element spectral lines with a correlation result (such as the Pearson correlation coefficient r ) greater than 0.9, and then further screening out the characteristic spectral lines in combination with the resonance lines, incomplete spectral parameters, and transition probabilities less than 10 6 s -1 in the reference NIST database; excluding the spectral lines with resonance lines, or the spectral lines with incomplete transition probabilities or energy level spectral parameters in the NIST library, and selecting the spectral lines with transition probabilities less than 10 6 s -1 ; or / and, the method for selecting the relevant interference spectral lines of the analyzed element includes: selecting the elements with a significant number of spectral lines in the cuttings as interference elements, and then determining the candidate interference spectral lines according to the criterion that the wavelength difference between the characteristic spectral line of the element to be analyzed and the spectral line of the interference element does not exceed 1 nm.

[0012] Preferably, in the third step, a correlation analysis is performed on the full-spectrum data in the LIBS spectral dataset of cuttings and the concentrations of cuttings elements, and the correlation formula is as shown in formula (16): (16) In formula (16), r is the Pearson correlation coefficient; is the intensity value of the spectrum of the i th sample of a certain element; is the concentration value of the i th sample of a certain element; is the average value; is the average value; n is the number of sample points; when r is closer to 1, the positive correlation between the two variables is stronger.

[0013] Preferably, in the fourth step, the method for constructing the ant colony optimization algorithm to optimize and improve the least squares support vector machine ensemble learning base learner includes: (1) Initialize the ant colony optimization algorithm, set the ant colony size, pheromone evaporation rate, pheromone importance degree, and heuristic factor, and define the search range for the regularization parameter C and the kernel function parameter γ of the LSSVM; (2) Perform ant colony optimization algorithm iteration, randomly assign the position of each ant, that is, randomly select the initial values of C and γ , and start iterative training of the LSSVM model. The decision function predicted by the LSSVM model is formula (17): (17) In formula (17), M is the number of training samples; is the feature vector of the j th sample input combined spectral line in the training samples; is the Lagrange multiplier, and each corresponds to a training sample; is the kernel function, which is used to calculate the similarity between the new input and the training sample . The radial basis function kernel is selected as the kernel function, and the kernel function formula is formula (18); b is the bias term, which is used to adjust the position of the decision function; f ( x ) is the calculation result of the decision function, that is, the predicted value of the target variable; (18) In formula (18), is a positive parameter that determines the width of the kernel function; is the kernel function value between two input vectors x and x′, where x and x′ represent the intensity value features of two different samples respectively; (3) Calculate the predicted concentration value based on the decision function and use the root mean square error RMSE As a performance indicator, select the best C and γ; (19) In formula (19), It is k The actual concentration value of the test sample element, is the model's predicted concentration value, P is the number of test samples.

[0014] Preferably, the ant colony size is set to 50; the pheromone evaporation rate is set to 0.5; the pheromone importance is set to 2; the heuristic factor is set to 2; the regularization parameter is set to C The search range is defined as (0.01, 100), and the kernel function parameters are γ The search range is defined as (0.01, 100).

[0015] Preferably, in step 5, different weights are assigned to each base learner. The weights are set according to the performance of the base learners. The prediction result is the weighted average of the prediction results of each base learner. The weight assignment formula is as shown in formula (20): (20) In formula (20), is the weight coefficient of each base learner, satisfying , n is the number of base learners; The predicted values of all base learners are multiplied by their respective weights, then summed up and finally divided by the sum of the weights. The value obtained by weighted average is the final element concentration prediction value.

[0016] An improved integrated learning system for analyzing the elemental content of cuttings based on laser spectral information according to the present invention. The system adopts the described analysis method and includes: a LIBS spectrum acquisition module for obtaining a LIBS spectrum dataset of cuttings elements collected by a laser-induced breakdown spectrometer; a noise reduction processing module connected to the LIBS spectrum acquisition module for performing noise reduction processing on the LIBS spectrum dataset of cuttings elements; a spectral line screening module connected to the noise reduction processing module for selecting characteristic spectral lines of cuttings analysis elements and related interference spectral lines of cuttings analysis elements from the LIBS spectrum dataset of cuttings elements processed by the noise reduction processing module; a data analysis module that uses an ant colony optimization algorithm to optimize and improve the least squares support vector machine integrated learning base learner as an integrated learner. It is connected to the spectral line screening module for combining the characteristic spectral lines with different related interference spectral lines respectively to obtain more than one spectral line combination, taking more than one spectral line combination as the input of the integrated learner respectively, predicting the concentration value of the analysis element through the decision function of the LSSVM regression model, and using the root mean square error RMSE as a performance index, based on RMSE the smaller the value, the better the prediction performance of the model. Select the optimal regularization parameter C and kernel function parameter γ for the input of different spectral line combinations; and an integrated strategy module connected to the data analysis module for performing weighted distribution integration on the base learners optimized with different spectral line combinations as inputs, and obtaining the final predicted elemental concentration value based on the predicted values of all base learners through weight distribution.

[0017] The improved integrated learning system and method for analyzing the elemental content of cuttings based on laser spectral information according to the present invention solve the defect of low accuracy of cuttings elements in the prior art and have the following advantages: (1) In the method of the present invention, a strong learner is constructed by combining multiple base learners to improve the prediction accuracy and robustness. Different learners may capture different data features and patterns, providing more comprehensive predictions, and can reduce the overfitting problem that may occur in a single model. The integrated learning combines multiple combinations of characteristic spectral lines and interference spectral lines as inputs, and is more capable of fully capturing and extracting LIBS spectral information, improving the analysis accuracy of the model;

[0018] (2) In the method of the present invention, optimizing the least squares support vector machine through the ant colony optimization algorithm can effectively help find the optimal or near-optimal LSSVM parameter combination, thereby improving the prediction performance of the LSSVM model. Moreover, using the ant colony optimization algorithm to automatically optimize the LSSVM parameters can reduce or eliminate the need for manual parameter selection, saving time and effort; (3) In the method of the present invention, by using the characteristic spectral lines combined with the spectral lines of interfering elements as input, the influence of interfering elements on the analytical element can be effectively reduced, thereby improving the accuracy of analysis; (4) In the method of the present invention, by using the 5-point and 3rd-order convolution smoothing method, the noise can be effectively reduced while the change information in the original data is well retained, and important spectral features will not be lost due to over-smoothing; (5) In the method of the present invention, under the same experimental conditions, 30 points at different positions are collected on the surface of each sample, and the average value of the 30 points is taken as the analysis data to compensate for the non-uniformity of the sample. Description of the Drawings

[0019] Figure 1 It is the flow chart of the improved integrated learning method for analyzing the content of cuttings elements based on laser spectral information of the present invention.

[0020] Figure 2 It is the specific flow chart of the integrated learning model of the least squares support vector machine improved by the ant colony optimization algorithm of the present invention.

[0021] Figure 3 It is the comparison chart of the analysis results of the content of Si element in cuttings between the analysis model of cuttings elements of the present invention and other models. Specific Embodiments

[0022] Next, the technical solutions in the embodiments of the present invention will be described clearly and completely. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0023] The features mentioned in the present invention can be combined arbitrarily as long as there is no contradiction in the combination of these features. All possible combinations should be considered as the scope described in this specification. Each feature disclosed in the specification can be replaced by any alternative feature that can provide the same, equivalent or similar purpose. Therefore, unless otherwise specified, the disclosed features are only general examples of equivalent or similar features.

[0024] Embodiment 1 An improved integrated learning method for analyzing the content of cuttings elements based on laser spectral information, see Figure 1 , the method includes: Step 1: Use a laser-induced breakdown spectrometer to collect the LIBS spectral dataset of cuttings elements.

[0025] Specifically, a LIBS spectral dataset of 100 national standard cuttings samples was collected using a laser-induced breakdown spectrometer. The acquisition and analysis band of the spectral data was 178 - 987 nm, with a total of 8,192 wavelength points. To compensate for the non-uniformity of the samples, 30 points at different positions on the surface of each sample were collected under the same experimental conditions, and the average value of the 30 points was taken as the analysis data.

[0026] Step 2: Divide the collected cuttings element LIBS spectral dataset into a spectral data training set and a spectral data test set in a ratio of 4:1, and perform noise reduction processing on the original spectral data by retaining the weighted variation information according to the 5-point 3-time convolution smoothing method.

[0027] Specifically, the fitting formula for a point in the cuttings element LIBS spectral dataset is: (12) Where, represents the spectral intensity of the spectral data at this point, represents the weight parameter fitted by the least squares method, represents the corresponding spectral intensity of the wavelength value.

[0028] The matrix form for 5-point fitting can be expressed as: (13) Where, represents the absorption intensity of 5 consecutive points, represents a 5-row 3-column matrix composed of the 0th, 1st, and 2nd values of the coordinates of 5 consecutive wavelength values, represents the matrix form of the weight parameter fitted by the least squares method, represents the matrix form of the error term. From the above formula (13), the least squares solution of can be obtained:

[0029] (14) The fitted filtered value is: (15) In formulas (14) and (15), the superscript T represents the transpose.

[0030] The method provided by the present invention can effectively reduce noise while better retaining the variation information in the original data by adopting the 5-point 3-time convolution smoothing method, and will not cause loss of important spectral features due to excessive smoothing.

[0031] Step 3: Select the characteristic spectral lines of the cuttings analysis elements based on the correlation coefficient, and then select the relevant interference spectral lines of the analysis elements. Multiple spectral line combinations composed of these two types of spectral lines are used as the input of the ensemble learning algorithm.

[0032] Taking the selection of six elements, namely Si, Ca, Al, Na, and Mg, as an example for the cuttings elements, the full-spectrum data in the LIBS spectral dataset of the cuttings elements is analyzed for correlation with the concentrations of the cuttings elements. The correlation formula is shown in Equation (16): (16) where r is the Pearson correlation coefficient, is the intensity value of the spectrum of the i th sample of a certain element, is the concentration value of the i th sample of a certain element (the concentration is known, such as the national standard cuttings sample, whose concentration is known), is 's average value, is 's average value, N is the number of sample points. When r is closer to 1, the positive correlation between the two variables is stronger.

[0033] Select the spectral lines of elements with a correlation result greater than 0.9, and then continue to screen the spectral lines in combination with the resonance lines, incomplete spectral parameters (transition probability, energy level), and transition probability less than 10 6 s -1 and other conditions in the NIST (National Institute of Standards and Technology) database. Specifically, if a spectral line has a resonance line, or the transition probability and energy level spectral parameters of a certain spectral line in the NIST library are incomplete, these spectral lines need to be excluded, and spectral lines with a transition probability less than 10 6 s -1 are selected. The characteristic spectral lines obtained by screening the six elements are shown in Table 1.

[0034] Table 1 Characteristic Spectral Lines Taking the Si element in Table 1 as an example, the selection of alternative interference spectral lines is as follows: When the Si element is the analysis element, select the elements with a significant number of spectral lines in the cuttings as interference elements: Ca, Al, Na, Mg, V, Ti, Mn, Fe. Determine the alternative interference spectral lines according to the criterion that the wavelength difference between the characteristic spectral lines of the Si element and the spectral lines of other interference elements does not exceed 1 nm. The alternative interference spectral lines obtained are shown in Table 2 below.

[0035] Table 2 Interference Spectral Lines The characteristic spectral lines of the analysis element Si and the spectral lines of each interfering element are respectively combined as the input of the ensemble learning machine. For example, combination 1 is the spectral lines of the Si element, combination 2 is the spectral lines of the Si element + the spectral lines of the Al interfering element, combination 3 is the spectral lines of the Si element + the spectral lines of the Mg interfering element, and so on. There are a total of 6 combined inputs.

[0036] The method provided by the present invention can effectively reduce the influence of interfering elements on the analysis element by using the characteristic spectral lines combined with the spectral lines of interfering elements as the input, thereby improving the accuracy of the analysis.

[0037] Step 4: Construct an ant colony optimization algorithm to optimize and improve the least squares support vector machine (LSSVM) ensemble learning base learner.

[0038] The least squares support vector machine (LSSVM) is used as the base learner of the ensemble learning model. The parameters of the LSSVM regression model include the regularization parameter C and the kernel function parameter γ as well as the kernel function type. Among them, the kernel function type is set to the radial basis function kernel (RBF Kernel), and the regularization parameter C and the kernel function parameter γ are optimized for parameter search by the ant colony optimization algorithm. The specific method for constructing an ant colony optimization algorithm to optimize and improve the least squares support vector machine (LSSVM) ensemble learning base learner includes:

[0039] First, initialize the ant colony optimization algorithm, set parameters such as the ant colony size (num_ants = 50), pheromone evaporation rate (evaporation_rate = 0.5), pheromone importance degree (alpha = 2), heuristic factor (beta = 2), etc., and define the search range for the regularization parameter C of LSSVM (0.01, 100) and the kernel function parameter γ of LSSVM (0.01, 100). Then, perform the iteration of the ant colony optimization algorithm, randomly assign the positions of each ant, that is, randomly select the initial values of C and γ , and start the iterative training of the LSSVM model. The decision function predicted by the LSSVM model is Equation (17):

[0040] (17) In Equation (17), M is the number of training samples; is the feature vector of the combined spectral lines of the j -th sample input in the training samples; is the Lagrange multiplier, and each corresponds to a training sample; is the kernel function, used to calculate the similarity between the new input and the training sample In the present invention, the kernel function is selected as the radial basis function kernel (RBF Kernel), and its kernel function formula is Equation (18); b is the bias term, used to adjust the position of the decision function; f ( x ) is the calculation result of the decision function, that is, the value of the predicted target variable (i.e., the predicted concentration value). Through the decision function, LSSVM can perform regression prediction on the sample data.

[0041] (18) In Equation (18), is a positive parameter that determines the width of the kernel function; is the kernel function value between two input vectors x and x′, where x and x′ represent the intensity value features of two different samples respectively.

[0042] Finally, the predicted concentration value is calculated according to the decision function, and the root mean square error (Root Mean Squared Error, RMSE ) is used as the performance index. RMSE The smaller the C value, the better the prediction performance of the model. The root mean square error formula is as shown in Equation (19), and based on this, the optimal

[0043] (19) In Equation (19), is the actual concentration value of the k -th test sample element (such as the national standard cuttings sample, whose concentration is known), is the predicted concentration value of the model, P is the number of test samples.

[0044] According to this process, taking the Si element as an example, with 6 combined inputs, 6 LSSVM base learner models with optimal parameters are determined and trained, and the predicted results corresponding to the inputs are obtained.

[0045] The method provided by the present invention can effectively help find the optimal or near-optimal LSSVM parameter combination by optimizing the least squares support vector machine through the ant colony optimization algorithm, thereby improving the prediction performance of the LSSVM model. In addition, using the ant colony optimization algorithm to automatically optimize the LSSVM parameters can reduce or eliminate the need for manual parameter selection, saving time and effort.

[0046] Step Five: Design the weighted allocation integration strategy of the base learners to complete the best combination of the base learners.

[0047] Different weights are assigned to each base learner, and the weights are set according to the performance of the base learner (root mean square error RMSE ). The prediction result is the weighted average of the prediction results of each base learner. The weight assignment formula is as shown in Equation (20):

[0048] (20) is the weighted coefficient of each base learner, satisfying , n is the number of base learners.

[0049] Multiply the predicted values of all base learners by their respective weights, then sum them, and finally divide by the sum of the weights. The value obtained through weighted average is the final predicted value of the element concentration.

[0050] The method provided by the present invention constructs a strong learner by combining multiple base learners to improve the prediction accuracy and robustness. Different learners may capture different data features and patterns, providing a more comprehensive prediction, which can reduce the overfitting problem that may occur in a single model. The ensemble learning combines multiple combinations of characteristic spectral lines and interference spectral lines as inputs, which can more fully capture and extract the LIBS spectral information and improve the analysis accuracy of the model.

[0051] As Figure 3 shown, it is a comparison chart of the analysis results of the cuttings element content analysis model of the present invention and other models for the Si element content in cuttings. It can be seen that compared with the commonly used PLS (partial least squares regression), SVM (support vector machine) methods, and a single LSSVM (least squares support vector machine) method, the method of the present invention has the best fitting effect, so the method of the present invention performs better in predicting the element concentration.

[0052] Although the content of the present invention has been introduced in detail through the above preferred embodiments, it should be recognized that the above description should not be considered as a limitation of the present invention. After those skilled in the art have read the above content, various modifications and substitutions to the present invention will be obvious. Therefore, the protection scope of the present invention should be defined by the appended claims.

Claims

1. An improved integrated learning method for analyzing the element content of cuttings based on laser spectral information, characterized in that The method includes: Step 1: Collect the LIBS spectral dataset of cuttings elements using a laser-induced breakdown spectrometer; Step 2: Divide the LIBS spectral dataset of cuttings elements into a spectral data training set and a spectral data test set, and perform noise reduction processing on the original spectral data; Step 3: Perform a correlation analysis on the full-spectrum data in the LIBS spectral dataset of cuttings elements and the cuttings element concentration. Select the characteristic spectral lines of the cuttings analysis elements based on the correlation coefficient, and then select the relevant interference spectral lines of the analysis elements. Combine the characteristic spectral lines with different relevant interference spectral lines to obtain more than one spectral line combination, and use more than one spectral line combination as the input of the ensemble learning algorithm respectively; Step 4: Construct an ant colony optimization algorithm to optimize and improve the least squares support vector machine ensemble learning base learner as the ensemble learner: Use the least squares support vector machine LSSVM as the base learner of the ensemble learning model. The parameters of the LSSVM regression model include: regularization parameter C , kernel function parameter γ and kernel function type; among them, the kernel function type is set to the radial basis function kernel, and the regularization parameter C , kernel function parameter γ are optimized by the ant colony optimization algorithm to construct an ant colony optimization algorithm to optimize and improve the least squares support vector machine ensemble learning base learner; among them, the concentration value of the analysis element is predicted by the decision function of the LSSVM regression model, and the root mean square error RMSE is used as the performance index. According to RMSE , the smaller the value, the better the prediction performance of the model. The corresponding optimal regularization parameter C and kernel function parameter γ are selected for the input of different spectral line combinations; Step 5: Design a weighted allocation integration strategy for the base learners to complete the optimal combination of the base learners, and obtain the final predicted value of the element concentration according to the weight allocation from the predicted values of all the base learners.

2. The analysis method according to claim 1, characterized in that, In Step 2, a 5-point 3rd-order convolutional smoothing method is used to perform noise reduction processing that retains the variation information of the original spectral data with weights.

3. The analysis method according to claim 2, wherein In Step 2, the noise reduction processing includes: The fitting formula for a point in the LIBS spectral dataset of cuttings elements is: （12） In formula (12), represents the spectral intensity of the spectral data at this point; represents the weight parameter fitted by the least squares method; represents the corresponding spectral intensity of the wavelength value; The matrix form representation for 5-point fitting is: （13） In Equation (13), represents the absorption intensity of five consecutive points; represents a 5-row and 3-column matrix composed of the 0th, 1st, and 2nd values of the coordinates of five consecutive wavelength values; represents the matrix form of the weight parameters fitted by the least squares method; represents the matrix form of the error term; From Equation (13), we get The least-squares solution of is as follows: （14） The fitted filtering value , is as follows: （15） In formulas (14) and (15), the superscript T denotes transpose.

4. The analysis method according to claim 1, characterized in that, In Step 1, the LIBS spectral dataset of cuttings elements collected by the laser-induced breakdown spectrometer is obtained according to the following method: Under the same experimental conditions, 30 points at different positions are collected on the surface of each sample, and the average value of the 30 points is taken as the analysis data; Or / and, in Step 1, the spectral data acquisition and analysis band of the laser-induced breakdown spectrometer is 178~987nm; Or / and, in Step 2, it is divided into the spectral data training set and the spectral data test set in a ratio of 4:

1.

5. The analysis method according to claim 1, characterized in that In the third step, the method for selecting the characteristic spectral lines includes: selecting the elemental spectral lines with a correlation result greater than 0.9, and then further screening out the characteristic spectral lines in combination with the resonance lines, incomplete spectral parameters, and transition probabilities less than 10 in the reference NIST database; excluding the spectral lines with resonance lines or incomplete transition probabilities and energy level spectral parameters of a certain spectral line in the NIST library, and selecting the spectral lines with transition probabilities less than 10 6 s -1 conditions; excluding the spectral lines with resonance lines or incomplete transition probabilities and energy level spectral parameters of a certain spectral line in the NIST library, and selecting the spectral lines with transition probabilities less than 10 6 s -1 ; Or / and, the method for selecting the relevant interference spectral lines of the analysis elements includes: Select the elements with a significant number of spectral lines in the cuttings as interference elements, and then determine the alternative interference spectral lines according to the criterion that the wavelength difference between the characteristic spectral lines of the element to be analyzed and the spectral lines of the interference elements does not exceed 1nm.

6. The analysis method according to claim 1, wherein In Step 3, perform a correlation analysis on the full-spectrum data in the LIBS spectral dataset of cuttings elements and the cuttings element concentration. The correlation formula is as shown in Equation (16): （16） In Equation (16), r is the Pearson correlation coefficient; is the intensity value of the spectrum of the i -th sample of a certain element; is the concentration value of the i -th sample of a certain element; is the average value; is the average value; n is the number of sample points; when r is closer to 1, the positive correlation between the two variables is stronger.

7. The analysis method according to claim 1, wherein In Step 4, the method for constructing the ant colony optimization algorithm to optimize and improve the least squares support vector machine ensemble learning base learner includes: (1) Initialize the ant colony optimization algorithm, set the ant colony size, pheromone evaporation rate, pheromone importance degree, and heuristic factor, and set the regularization parameter C and kernel function parameter γ of LSSVM, and define the search range. (2) Perform the iteration of the ant colony optimization algorithm, randomly assign the position of each ant, that is, randomly select C and γ the initial value of, and start the iterative training of the LSSVM model. The decision function predicted by the LSSVM model is Equation (17): （17） In formula (17), M is the number of training samples; is the feature vector of the j -th sample input combined spectral line in the training samples; is the Lagrange multiplier, and each corresponds to a training sample; is the kernel function, which is used to calculate the similarity between the new input and the training sample . The kernel function is selected as the radial basis function kernel, and the kernel function formula is formula (18); b is the bias term, which is used to adjust the position of the decision function; f ( x ) is the calculation result of the decision function, that is, the value of the predicted target variable; （18） In Equation (18), is a positive parameter that determines the width of the kernel function; is the kernel function value between two input vectors x and x′, where x and x′ represent the intensity value features of two different samples respectively; (3) Calculate the predicted concentration value according to the decision function, and use the root mean square error RMSE as the performance index to select the optimal C and γ; （19） In formula (19), is the actual concentration value of the k th test sample element, is the predicted concentration value of the model, P is the number of test samples.

8. The analysis method according to claim 7, wherein The ant colony size is set to 50; the pheromone evaporation rate is set to 0.5; the importance degree of pheromone is set to 2; the heuristic factor is set to 2; the regularization parameter C search range is defined as (0.01, 100), and the kernel function parameter γ search range is defined as (0.01, 100).

9. The analysis method according to any one of claims 1 to 8, characterized in that, In Step 5, different weights are assigned to each base learner. The weights are set according to the performance of the base learners. The prediction result is the weighted average of the prediction results of each base learner. The weight assignment formula is as shown in Equation (20): （20） In formula (20), is the weighting coefficient of each base learner, satisfying , n is the number of base learners; Multiply the predicted values of all the base learners by their respective weights, then sum them up, and finally divide by the sum of the weights. The value obtained through weighted average is the final predicted value of the element concentration.

10. An improved integrated learning cuttings element content analysis system based on laser spectral information, characterized in that, The system adopts the analysis method described in any one of claims 1~9, including: A LIBS spectral acquisition module, which is used to obtain the LIBS spectral dataset of cuttings elements collected by a laser-induced breakdown spectrometer; A noise reduction processing module, which is connected to the LIBS spectral acquisition module and is used to perform noise reduction processing on the cuttings element LIBS spectral data set; A spectral line screening module, which is connected to the noise reduction processing module and is used to select the characteristic spectral lines of the cuttings analysis elements and the related interference spectral lines of the cuttings analysis elements from the cuttings element LIBS spectral data set processed by the noise reduction processing module; The data analysis module uses the ant colony optimization algorithm to optimize and improve the least squares support vector machine integrated learning base learner as the integrated learning machine. It is connected to the spectral line screening module and is used to respectively combine the characteristic spectral lines with different relevant interference spectral lines to obtain more than one spectral line combination, use more than one spectral line combination as the input of the integrated learning machine respectively, predict and analyze the concentration value of the element through the decision function of the LSSVM regression model, and use the root mean square error RMSE as the performance index, and based on RMSE the smaller the value, the better the prediction performance of the model, and select the optimal regularization parameter C and kernel function parameter γ ; and An integrated strategy module, which is connected to the data analysis module and is used to perform weighted distribution integration on the base learners optimized with different spectral line combinations as inputs, and obtain the final element concentration prediction value based on the predicted values of all base learners through weight distribution.

Images