CN113155809B - Novel spectral detection method for ore classification and real-time quantitative analysis - Google Patents
Novel spectral detection method for ore classification and real-time quantitative analysis Download PDFInfo
- Publication number
- CN113155809B CN113155809B CN202110267879.6A CN202110267879A CN113155809B CN 113155809 B CN113155809 B CN 113155809B CN 202110267879 A CN202110267879 A CN 202110267879A CN 113155809 B CN113155809 B CN 113155809B
- Authority
- CN
- China
- Prior art keywords
- ore
- spectral data
- sample
- laser
- spectral
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/62—Systems in which the material investigated is excited whereby it emits light or causes a change in wavelength of the incident light
- G01N21/71—Systems in which the material investigated is excited whereby it emits light or causes a change in wavelength of the incident light thermally excited
- G01N21/718—Laser microanalysis, i.e. with formation of sample plasma
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Physics & Mathematics (AREA)
- Analytical Chemistry (AREA)
- Data Mining & Analysis (AREA)
- Pathology (AREA)
- Immunology (AREA)
- General Health & Medical Sciences (AREA)
- Biochemistry (AREA)
- Chemical & Material Sciences (AREA)
- Theoretical Computer Science (AREA)
- Optics & Photonics (AREA)
- Plasma & Fusion (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- Investigating, Analyzing Materials By Fluorescence Or Luminescence (AREA)
Abstract
The invention discloses a novel spectrum detection method for classifying ores and carrying out real-time quantitative analysis based on laser-induced breakdown spectroscopy, which comprises the following steps: the method comprises the steps of obtaining spectrum data of a plurality of samples after being excited by laser, firstly carrying out spectrum preprocessing to reduce spectrum volatility, secondly accurately classifying all ores through a support vector machine, reducing dimensions by using a principal component analysis method, improving the accuracy of element quantitative analysis, and then screening a partial least squares regression analysis (R-PLS) improvement algorithm through related variables, so that the root mean square error and the average relative error of a prediction set of the ores are greatly reduced. The technology removes redundant data, greatly reduces the number of modeling variables and improves the operation speed. The invention realizes quantitative analysis of iron content in the ore based on the laser-induced breakdown spectroscopy technology, and constructs a novel spectral detection method for classifying and real-time quantitative analysis of the ore by using the laser-induced breakdown spectroscopy technology.
Description
Technical Field
The invention belongs to a brand new spectral detection method for ore classification and real-time quantitative qualitative analysis, and particularly relates to a component quantitative analysis method, a test system and a processing algorithm by utilizing laser-induced breakdown spectroscopy, which are used for rapidly and conveniently classifying and quantitatively analyzing ores under the condition of low material loss.
Background
The laser-induced breakdown spectroscopy technology is a full-element detection means for analyzing the composition of a sample material, wherein a small part of the material on the surface of a sample is broken down by using pulse laser to generate plasma, the plasma is cooled to emit light, and then the spectrum emitted by the plasma is analyzed. The test means has many characteristics and advantages. Firstly, the damage to a target sample is small in the detection process of the laser-induced breakdown spectroscopy technology, only the material with the microgram level on the surface of the target sample needs to be broken down, and the method can be regarded as nondestructive detection; secondly, the testing process of the laser-induced breakdown spectroscopy technology does not need to carry out complex pretreatment on the target sample, and the treatment process does not need to use chemical reagents and has no pollution to the environment. In the test process, only pulse laser needs to be directly focused on the surface of a sample to be tested, and relevant information of the element composition can be obtained; meanwhile, the laser-induced breakdown spectroscopy technology has no specific requirements on the form of a target sample, and can be used for testing gaseous, liquid and solid samples; finally, the test process of the laser-induced breakdown spectroscopy technology is relatively quick compared with the traditional detection method because the test result is obtained by analyzing the spectral information.
Nowadays, in scientific research and industrial application fields, such as many fields of biomedicine, environmental detection, harbor customs, material analysis and the like, the requirement for detecting the element components of a target sample is very urgent. Compared with traditional element testing methods such as a chemical analysis method or a high-energy ray detection method, the laser-induced breakdown spectroscopy technology has the characteristics of less damage to a sample in the testing process, more convenient preparation processes before testing such as sample preparation, shorter time consumption in the testing process and the like. Therefore, the method is more suitable for various occasions needing the sample to carry out element test. For example, in biomedical applications, etc., it is desirable that the test device be capable of being operated under a microscope. In a seaport, in order to deal with the handling capacity of a large amount of cargoes, when the cargoes are detected, a test means is needed to be convenient and quick, and the damage to the cargoes is small. In the field of material analysis, it is required that the test range of the test means is sufficiently wide and various components in a target sample can be tested and analyzed as much as possible. Due to various characteristics and advantages of the laser-induced breakdown spectroscopy technology, the requirements of various test occasions can be well met.
Most of the research at present focuses on classifying and identifying mineral components and geographical sources, and LIBS combines a chemometric method to improve the ore analysis precision. Combining classification and quantitative analysis methods, classifying the spectrums through a support vector machine, and then screening partial least square regression by adopting a correlation variable to perform quantitative analysis; in the aspect of spectrum selection, a linear relation between spectrum data and typical element content is established by depending on a principal component analysis method, spectra with correlation coefficients R larger than 0.9, 0.85, … and 0 are sequentially taken out for partial least squares regression, and a determination coefficient R is respectively calculated 2 Root Mean Square Error (RMSEP), Average Relative Error (ARE), etc. of the prediction set. After the unclassified ore is subjected to full spectrum partial least squares regression analysis, RMSEP is 3227%, ARE 31.75%. The ores ARE classified by the new method proposed by the inventor, and then correlation variable screening partial least squares regression analysis (R-PLS) is carried out, wherein RMSEP of iron ores, manganese ores and chromium ores is respectively reduced to 0.975%, 0.418% and 0.123%, and ARE is respectively reduced to 1.46%, 6.72% and 1.09%. The method shows that the accuracy of quantitative analysis is greatly improved by performing correlation variable screening partial least squares regression analysis after ore classification. Therefore, the novel spectral detection method for ore classification and real-time quantitative analysis has great advantages.
Disclosure of Invention
The invention aims to provide an ore classification and real-time quantitative analysis method and a test system, which realize ore classification and real-time quantitative analysis and detection by utilizing a laser-induced breakdown spectroscopy technology.
In one aspect, the invention provides a method for quantitative analysis of elements, comprising the following steps:
step 1: acquiring spectral data of plasma of the same type of samples after different positions are excited by laser, repeatedly measuring for 5 times at each position, and acquiring 100 sets of spectral data of each sample. Respectively and repeatedly testing N groups of the same sample, averaging the influence of uneven distribution of the elements of the sample, and further acquiring the real content of the element components in each sample;
the same type of samples refer to the same type of samples with the same components and different contents, for example, the contents of the samples are determined by national standards;
step 2: and (3) adopting a classification method combining principal component analysis and a support vector machine. Classifying the models by using a support vector machine, reducing the dimensions of all the preprocessed spectral data by adopting a principal component analysis method, randomly selecting a training set and a prediction set after reducing the dimensions, extracting the first 10 principal components of the training set to construct a feature space, training the training set by adopting 5-fold cross validation of a small sample under the feature space, and then accurately classifying ores.
And step 3: and (3) after classification is carried out by the method in the step (2), correlation variable screening partial least squares regression (R-PLS) is carried out on the spectral data. Taking the total spectral intensity as input data and the Fe content asAnd (6) regressing a target variable. The regression effect, calibration accuracy, prediction accuracy and prediction error of the model are determined by a coefficient (R) 2 ) The Root Mean Square Error (RMSEC) of the correction set, the Root Mean Square Error (RMSEP) of the prediction set and the Average Relative Error (ARE) ARE comprehensively measured.
And 4, step 4: after the ores are classified, a partial least squares regression analysis algorithm (R-PLS) is screened through correlation variables, the Fe content in the object to be tested is finally obtained, and the accuracy of the method is greatly improved compared with that of unclassified Partial Least Squares (PLS) and classified Partial Least Squares (PLS) and that of Support Vector Machine (SVM)
The errors caused by calculation by using the conventional unclassified + full-spectrum PLS and classified + full-spectrum PLS methods are relatively large, and the inaccuracy of the calculation result mainly comes from the following aspects: the self-absorption effect of cooling atoms in the sample leads to a certain degree of non-linear dependence of the intensity of characteristic peaks in the spectral information on the actual component concentration in the sample. (II) different components interfere with each other, and characteristic peaks of the components may overlap with each other to some extent and thus influence the accuracy of the quantitative analysis result. And (III) the plasma gas generated by the sample after being excited has a certain degree of nonuniformity in space and time, so that the error of the calculation result of the traditional partial least squares regression is increased. In the quantitative analysis algorithm adopted by the device, a support vector machine is added for classification, namely: when the support vector machine classification and related variable screening partial least squares regression analysis improved algorithm is used for quantitative analysis, the error is greatly reduced, and the accuracy of accurate classification is improved.
Further preferably, the step 3 is performed as follows:
acquiring spectral data corresponding to a wavelength point at which the contribution degree meets a preset standard from the spectral data in the substance to be detected, and inputting the trained partial least squares regression model to obtain the predicted content of each component in the substance to be detected;
inputting the spectral data in the substance to be detected into the trained regression model of the support vector machine to obtain residual errors corresponding to all components in the substance to be detected;
and finally, subtracting the corresponding residual error from the predicted content of each component in the substance to be detected to obtain the content of each component.
Further preferably, the spectral data with the contribution degree meeting the preset standard in step 2 refers to the spectral data corresponding to the wavelength point with the maximum contribution degree.
Further preferably, after the spectral data is acquired in step 1, denoising processing is performed on the spectral data of each sample, where the denoising processing includes: performing background continuous spectrum removal processing on the spectral data by adopting a local weighted regression scatter point smoothing method with data window movement, wherein the data window comprises the spectral data of n wavelength points, and the value range of n is 1-j m *c 1 ，j m Indicates the total number of wavelength points obtained in one scanning channel, c 1 Is the number of scanning channels.
Further preferably, after the spectral data is acquired in step 1, denoising processing is performed on the spectral data of each sample, where the denoising processing includes: the spectral data are normalized, and the normalization calculation formula is as follows:
in formula (II)' ij Representing normalized spectral data corresponding to a jth wavelength point in an ith scan of the sample at a c-th scan channel,
representing the spectral data before normalization corresponding to the jth wavelength point in the ith scan of the sample in the c-th scan channel, j m Indicating the total number of wavelength points obtained in one scanning channel.
Further preferably, the penalty parameter c and the kernel parameter g of the support vector machine model are optimized by a particle swarm optimization.
In a second aspect, the present invention provides a test system comprising: nd of active Q-switching: YAG1064nm pulse laser, three-dimensional electric platform, four-channel fiber spectrometer (Avantes Avsdesktop USB2, Netherlands);
wherein the Nd: YAG1064nm pulse laser is used as an excitation source, and laser beams are focused on the surface of an object after passing through a lens group; the three-dimensional electric platform is used for adjusting the sample at a certain speed, so that each time the laser can act on a new sample area; the four-channel fiber optic spectrometer is used at a collection site, and light is coupled into an optical fiber through a lens and transmitted to the spectrometer.
In a third aspect, the present invention provides a computer, on which a program capable of running is installed, and when the program is executed, the processor implements the steps of the method.
Advantageous effects
1. The support vector machine classification and related variable screening partial least squares regression analysis improved algorithm is characterized in that firstly, a Lauda criterion is adopted to eliminate abnormal spectra, secondly, a classification model is constructed by adopting a principal component analysis and support vector machine method, and the ores are classified quickly and accurately. Then, a partial least square regression method is screened through the correlation variables, so that the regression effect, the calibration precision, the prediction precision and the prediction error of the model are respectively improved by using a determination coefficient (R) 2 ) The Root Mean Square Error (RMSEC) of the correction set, the Root Mean Square Error (RMSEP) of the prediction set, and the Average Relative Error (ARE). Realizes the quantitative analysis of the components, and is a novel spectral detection method for ore classification and real-time quantitative analysis.
2. The testing system constructed by the invention uses the nanosecond laser as an excitation light source, so that the laser induced breakdown spectroscopy maintains smaller volume on the premise of realizing the excitation of all-component elements.
Drawings
FIG. 1 is a schematic structural diagram of a test system provided in an embodiment of the present invention;
FIG. 2 is a calibration curve of Fe content obtained using full spectrum partial least squares regression on 35 ores before and after classification (a) for all ores; (b) iron ore; (c) manganese ore; (d) chromium ore;
FIG. 3 is a calibration curve of Fe content obtained by classifying 35 ores and then performing partial least squares regression analysis on the classified ores by correlation variable screening. (a) Iron ore; (b) manganese ore; (c) chromium ore; (d) all the ore.
Detailed Description
The present invention will be further described with reference to the following examples. As shown in fig. 1, the embodiment of the present invention provides a testing system for ore classification and timing quantitative analysis based on nanosecond laser-induced breakdown spectroscopy, which is specifically described below as an example, but it should be understood that the testing system of the present invention is not limited to the embodiment.
In this embodiment, the test system includes: nd at 1064nm wavelength: YAG nanosecond laser, three-dimensional electric platform, four-channel fiber spectrometer.
The energy and the frequency of the laser are adjustable, and after being focused by the lens, the laser is focused on the surface of the sample. The movable microscope system can be used as a bearing original piece, the position of a sample is adjusted on three axes, focusing is assisted, and new points as many as possible are collected according to a certain stepping displacement sample. The collection system couples the light emitted by the plasma into the optical fiber, enters the spectrometer, is connected with the PC, and carries out the next processing.
Providing above-mentioned test system based on this embodiment, operating personnel only need place the sample that awaits measuring in laser location indicating position, can carry out elemental composition content measurement to the sample, then classify and real-time quantitative analysis to the ore.
It should be understood that in other possible embodiments, Nd: the YAG nanosecond laser, the three-dimensional motorized displacement stage, and the collection fiber can be adjusted accordingly, but at least the components should be guaranteed to perform their basic functions, such as Nd: the YAG picosecond laser beam focuses on the surface of the substance to be measured, the optical fiber spectrometer can collect spectral data, and the three-position electric displacement table has enough moving space.
The samples used in this application were 35 ores, of which 14 iron ores, 12 manganese ores and 9 chromium ores are listed in table 1. The samples were pressed into small round cakes with a thickness of 2mm and a diameter of 13mm using a mechanical powder tablet press of 15 MPa.
TABLE 1 sample types and Fe contents of iron, manganese and chromium ores
Table 1Sample type and Fe content of the iron ore,manganese ore and chromium ore
In order to realize the measurement of the component content, the invention provides a component quantitative analysis method based on laser-induced breakdown spectroscopy, which comprises the following steps:
step 1: and acquiring the spectral data of the plasma after a plurality of samples of the same type are excited by laser, repeating the N groups respectively, and acquiring the real content of components in each sample.
In this embodiment, the spectral signals are collected by a fiber optic spectrometer. The fiber optic spectrometer collects spectral information of plasma luminescence generated after a sample is excited. Each collection will yield a complete spectrum of data comprising the intensity values of the plasma emission of the sample at each wavelength, arranged in order of wavelength, with each scan yielding j m *c 1 Spectral data of data points, j m Indicates the total number of wavelength points obtained in one scanning channel, c 1 For the number of scan channels, there are 4 scan channels, j, in the present embodiment m Equal to 2048 for a total of 8192 data points. In this embodiment, each position is repeatedly measured 5 times, N is repeated 20 times for each sample, and then 100 pieces of complete spectral information are obtained, and the 100 pieces of spectral information are arranged into a matrix, that is, an original spectral data matrix X, X obtained by scanning and testing one sample by the apparatus is obtained ij For its matrix elements, i represents the number of scanning tests, and j represents the order of intensity values in wavelength order in the spectral data acquired by the fiber spectrometer, the intensity values beingThe order of the values corresponds one-to-one to the values of their respective wavelengths, so that the value of j in fact also represents information on the wavelength represented by the intensity value. Thus, each component in the matrix X represents the jth intensity value obtained from the ith of the 100 tests of the fiber optic spectrometer. In other possible embodiments, N is a positive integer.
In this embodiment, it is preferable to perform denoising processing on the collected spectral data, including removing the background continuous spectrum and normalizing processing.
And (3) removing a disordered spectrum:
in order to avoid the influence of the abnormal spectrum on the measurement result, the abnormal spectrum is removed by adopting a Laplace criterion. The Larita criterion generally assumes that the measured data has normal distribution, calculates the standard deviation of the data, and demarcates a probability interval according to requirements, and finally eliminates the gross errors exceeding the probability interval.
The algorithm is realized as follows:
(1) x for collected LIBS spectrum ij Is represented by X ij The spectrum data of the jth sample point of the sample with the number i is indicated, the spectrum intensity of each channel is summed, and the spectrum corresponding to the median of the total light intensity of each channel is taken as the center point of the sample
(2) Determining the distance (i.e., Euclidean distance) of each sample point from the center point
) Fig. 2(a) shows a euclidean distance distribution diagram corresponding to the spectral data of the sample. It is then normalized (0-1).
(3) In order to more intuitively see the probability density distribution rule of the distance, the relation between the segmentation distance and the frequency can be drawn.
(4) Calculating residual error
And calculating the standard deviation sigma according to Bessel formula, if a certain measured value D j Residual error v of j Satisfy the requirements of
Then consider v to be j Bad values containing gross error values should be rejected.
After the abnormal spectrum is removed, the volatility of most spectral lines is reduced. And then processing the data with the abnormal spectrum removed by adopting spectrum background integral intensity normalization, peak position drift correction and missing peak completion.
Normalization treatment:
the corrected linear spectrum data is subjected to spectrum uncertainty processing by a spectrum background integral intensity normalization method. The background spectrum intensity of each channel is independently normalized, so that the uncertainty of the spectrum is reduced, and the analysis of the concentration difference of each component in the subsequent steps is facilitated. Spectral anomalies in the line spectrum, such as spectral shifts, occasional noise spectra, etc., are also removed in this step. The calculation formula is as follows:
in the formula (I), the compound is shown in the specification,
representing normalized spectral data corresponding to a jth wavelength point in an ith scan of the sample at a c-th scan channel,
represents the spectrum data before normalization corresponding to the jth wavelength point in the ith scanning of the sample in the jth scanning channel, i.e. the spectrum data S in this embodiment ik Expanded data, N x K spectral data S ik Expanded and then N (j) m *c 1 ) A matrix of sizes of the components of the image,
is a matrix element, and then the spectral data of each scanning channel is used as a listAnd respectively carrying out normalization processing on the elements.
Step 2: and (3) classifying by using a support vector machine and combining principal component analysis: because the number of spectral characteristic peaks in ore is very large, the number of spectral characteristic peaks is similar, but the spectral line intensity and the peak position have obvious difference. And (3) carrying out principal component analysis to reduce dimensions by using 8192 points in each of 3500 groups of spectra which are all the preprocessed spectral data (3500 x 8192), wherein the accumulated interpretation rate of the first M principal components reaches more than 92%, and the first M principal components can be considered to cover most information of the ore spectral data. After dimensionality reduction, the proportion relationship between the training set and the prediction set is randomly selected as follows: 25: 10, extracting the first 10 main components of the training set to construct a feature space, and training the training set by adopting a 5-fold cross-validation method of a small sample under the feature space, wherein the classification accuracy of the modeling set and the prediction set is 100%. The classification method combining principal component analysis and a support vector machine can realize the rapid and high-accuracy classification and identification of ores.
And 3, step 3: correlation variable screening partial least squares regression method
Because the support vector machine can realize accurate classification of all ores, a partial least squares regression (R-PLS) method is used for carrying out correlation variable screening on the classified spectral data. The method removes redundant data, greatly reduces the number of modeling variables and improves the operation speed. The regression effect, the calibration accuracy, the prediction accuracy and the prediction error of the model ARE comprehensively measured by a determination coefficient (R2), the Root Mean Square Error (RMSEC) of a correction set, the Root Mean Square Error (RMSEP) of a prediction set and an Average Relative Error (ARE).
The method comprises the following specific steps:
and screening a partial least squares regression method based on the trained support vector machine and the correlation variable, and obtaining the content of the element to be detected by using the spectral data of the substance to be detected. The method comprises the following specific steps:
1. and classifying by using a support vector machine and combining principal component analysis, wherein the classification comprises the following specific steps:
the selection criterion of the number of principal components of principal component analysis is as follows: the modeling effect is better along with the increase of the number of the main components. When the number of principal components is small, the prediction accuracy is insufficient, so that the RMSEP value becomes small and the prediction accuracy increases as the number of principal components increases, but when the number of principal components is too large, overfitting occurs, the RMSEP value becomes large and the prediction accuracy decreases. Therefore, the selection of the number of the main components is determined by the RMSEP minimum method.
2. And (3) carrying out correlation variable screening, specifically as follows:
firstly, establishing a linear relation between spectrum data and Fe element content, respectively selecting spectrum data with correlation coefficients R larger than 0.9, 0.85, …, 0.05 and 0, and sequentially performing partial least squares regression analysis.
When the selected correlation coefficient is small (R is equal to 0-0.5), the Root Mean Square Error (RMSEP) of the prediction set changes along with the change of R, and the change rule is not obvious, but the RMSEP is larger overall. When R is larger than 0.5, RMSEP shows a tendency to decrease and then increase with the change in R because when R is too small, a peak having a weak correlation with Fe element content is selected so much that analysis accuracy is insufficient. When R is too large, overfitting is easy to cause and the tolerance capability is weak due to too few selected variables, and the analysis accuracy is also influenced. The selection basis for the optimal prediction effect of the correlation variable screening partial least squares regression analysis is as follows: the correlation coefficient corresponding to the RMSEP minimum is selected.
Claims (4)
1. A novel detection method for ore classification and real-time quantitative analysis based on laser-induced breakdown spectroscopy is characterized by comprising the following steps: the method comprises the following steps:
step 1: pressing the ore samples into small round cakes with the thickness of 2mm and the diameter of 13mm by using a 15Mpa mechanical powder cake press, acquiring spectral data of plasmas of the same type of samples after different positions are excited by laser, repeatedly measuring for 5 times at each position, acquiring 100 sets of spectral data of each type of samples, repeatedly testing N sets of the same type of samples respectively, averaging the influence of uneven element distribution of the ore samples, and further acquiring the real content of element components in each ore sample;
step 2: removing abnormal spectral data of each ore sample by using Lauda criterion, assuming that the data has normal distribution by using the Lauda criterion, calculating the standard deviation of the data, defining a probability interval according to requirements, and finally removing gross errors exceeding the probability interval, correcting peak position drift and completing missing peaks;
the abnormal spectrum elimination algorithm is realized as follows:
(1) x for collected LIBS spectrum ij Is represented by X ij The spectral data of the jth sample point of the ore sample with the number i is designated, the spectral intensity of each channel is summed, and the spectrum corresponding to the median of the total light intensity of each channel is taken as the center point of the ore sample
(2) Determining the distance of each sample point from the central point, i.e. the Euclidean distance
Then normalizing the obtained product to be (0-1);
(3) calculating residual error
And calculating the standard deviation sigma according to Bessel formula, if a certain measured value D j Residual error v of j Satisfy the requirement of
The bad value containing the large error value is considered to be eliminated;
and step 3: the classification method combining principal component analysis and a support vector machine is adopted, classification is firstly carried out, and then dimension reduction is carried out: classifying the models by using a support vector machine, reducing the dimensions of all the preprocessed spectral data by adopting a principal component analysis method, randomly selecting a training set and a prediction set after reducing the dimensions, extracting the first 10 principal components of the training set to construct a feature space, training the training set by adopting 5-fold cross validation of a small sample under the feature space, then classifying the modeling set and the prediction set with the accuracy rate of 100%, and accurately classifying the ore samples;
and 4, step 4: after classification is carried out by the method in the step 3, a prediction model of a correlation variable screening partial least squares regression method is established for the spectral data, all spectral intensities ARE used as input data, the content of Fe is used as a regression target variable, and the regression effect, the calibration precision, the prediction precision and the prediction error of the model ARE comprehensively measured by a determination coefficient R2, the root mean square error RMSEC of a correction set, the root mean square error RMSEP of the prediction set and the average relative error ARE respectively; the method comprises the following specific steps:
acquiring spectral data corresponding to a wavelength point with contribution degree meeting a preset standard from spectral data in an ore sample, and inputting the trained predictive model of the correlation variable screening partial least squares regression method to obtain the predicted content of each component in the ore sample;
inputting spectral data in the ore sample into a trained regression model of a support vector machine to obtain residual errors corresponding to all components in the ore sample;
finally, subtracting the corresponding residual error from the predicted content of each component in the ore sample to obtain the content of each component;
and 5: the ore samples are classified and then subjected to correlation variable screening partial least square regression method, the content of Fe in the ore samples is finally obtained, and the accuracy is greatly improved compared with the method that correlation variable screening partial least square regression method is directly carried out without classification and the method that correlation variable screening partial least square regression method is carried out after classification.
2. The method of claim 1, wherein: all spectral data of the ore sample in step 2: the method comprises the steps of screening abnormal spectral data, eliminating fluctuation of reduced spectral lines, normalizing spectral background integral intensity, correcting peak position drift and completing missing peaks.
3. The method of claim 1, wherein: after accurate classification is carried out by using a support vector machine, a predictive model of a correlation variable screening partial least squares regression method is established for the spectral data, all spectral intensities are used as input data, and the content of Fe is used as a regression target variable.
4. A test system, characterized by: the method comprises the following steps: nd for active Q-switching: YAG1064nm pulse laser, three-dimensional electric platform, four-channel fiber spectrometer;
wherein the Nd: YAG1064nm pulse laser as excitation light source, the laser beam is focused on the surface of the object;
the four-channel fiber optic spectrometer is used for collecting the spectral data of plasma generated by exciting an ore sample, and the spectral data is used for acquiring the content of Fe element in the ore sample according to the method of any one of claims 1 to 3.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110267879.6A CN113155809B (en) | 2021-03-12 | 2021-03-12 | Novel spectral detection method for ore classification and real-time quantitative analysis |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110267879.6A CN113155809B (en) | 2021-03-12 | 2021-03-12 | Novel spectral detection method for ore classification and real-time quantitative analysis |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113155809A CN113155809A (en) | 2021-07-23 |
| CN113155809B true CN113155809B (en) | 2022-08-19 |
Family
ID=76886894
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110267879.6A Active CN113155809B (en) | 2021-03-12 | 2021-03-12 | Novel spectral detection method for ore classification and real-time quantitative analysis |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113155809B (en) |
Families Citing this family (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113740277A (en) * | 2021-10-15 | 2021-12-03 | 北方民族大学 | Environment safety analysis method based on spectral multi-component analysis |
| CN114002204B (en) * | 2021-10-15 | 2023-01-06 | 华中科技大学 | Laser-induced breakdown spectroscopy analysis method based on spectral dithering |
| CN114850658B (en) * | 2022-05-17 | 2024-05-24 | 北京工业大学 | Laser parameter automatic regulation and control method based on reflection spectrum |
| WO2023236540A1 (en) * | 2023-01-17 | 2023-12-14 | 山东大学 | Ore constituent analysis apparatus and method |
| CN116148212B (en) * | 2023-04-21 | 2023-07-07 | 矿冶科技集团有限公司 | Method for rapidly determining types and contents of clay minerals in ores based on near infrared spectrum analysis |
| CN117269148B (en) * | 2023-11-23 | 2024-02-09 | 山东智谷碳素研究院有限公司 | Petroleum coke component analysis and detection method based on LIBS technology |
-
2021
- 2021-03-12 CN CN202110267879.6A patent/CN113155809B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN113155809A (en) | 2021-07-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113155809B (en) | Novel spectral detection method for ore classification and real-time quantitative analysis | |
| EP2710353B1 (en) | SPECTROSCOPIC APPARATUS AND METHOD of DETERMINING COMPONENTS PRESENT IN A SAMPLE | |
| CN103487411B (en) | A kind of random forests algorithm is in conjunction with the method for Laser-induced Breakdown Spectroscopy identification type of steels | |
| CN106814061B (en) | A method of improving laser induced breakdown spectroscopy overlap peak accuracy of quantitative analysis | |
| US10309831B2 (en) | Dynamic calibration method for echelle spectrometer in laser-induced breakdown spectroscopy | |
| CN108333171B (en) | Method for detecting content of trace elements in milk powder based on laser-induced breakdown spectroscopy | |
| CN104483292B (en) | A kind of method that use multiline ratio method improves laser microprobe analysis accuracy | |
| CN104697966A (en) | Method for quantitatively analyzing chromium and manganese in steel based on least square support vector machine algorithm combined with laser-induced breakdown spectroscopy | |
| Dai et al. | Quantitative determination of Al–Cu–Mg–Fe–Ni aluminum alloy using laser-induced breakdown spectroscopy combined with LASSO–LSSVM regression | |
| CN113340874B (en) | Quantitative analysis method based on combination ridge regression and recursive feature elimination | |
| CN107220625B (en) | LIBS spectrum automatic identification and classification method for sample | |
| CN112834485B (en) | Non-calibration method for quantitative analysis of laser-induced breakdown spectroscopy elements | |
| Zou et al. | Online simultaneous determination of H 2 O and KCl in potash with LIBS coupled to convolutional and back-propagation neural networks | |
| Harhira et al. | Advanced laser-induced breakdown spectroscopy (LIBS) sensor for gold mining | |
| EP2366991A1 (en) | Probabilistic scoring for components of a mixture | |
| CN110751048B (en) | Laser probe classification method and device based on image characteristic automatic spectral line selection | |
| CN105717094B (en) | A kind of metal element content analysis method based on large database concept identification | |
| CN111272735B (en) | Detection method of laser-induced breakdown spectroscopy | |
| CN105717093B (en) | A kind of cement characteristics analysis method based on large database concept identification | |
| CN116026808A (en) | Raman spectrum discrimination method and system | |
| CN113588597A (en) | Method for improving analysis precision of furnace slag | |
| Mao et al. | Reduction of spectral interferences and noise effects in laser ablation molecular isotopic spectrometry with partial least square regression–a computer simulation study | |
| CN111044504B (en) | Coal quality analysis method considering uncertainty of laser-induced breakdown spectroscopy | |
| CN113567417A (en) | Method for identifying peanut oil production place based on Raman spectrum fingerprint analysis technology | |
| Awasthi et al. | New Insights on Modern Age Coins by Calibration-Free Laser-Induced Breakdown Spectroscopy Method and Chemometric Approaches |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |