Method and system for synchronously calibrating multiple LIBS (Laser-induced Breakdown Spectroscopy) devices based on transfer learning

Abstract

The invention discloses a method for synchronously calibrating multiple LIBS (laser-induced breakdown spectroscopy) equipment based on transfer learning. The method comprises the following steps: acquiring a calibration sample set, source equipment and equipment to be calibrated; calibrating the source equipment based on the calibration sample set to obtain a relational expression between a source model and a spectral variable; obtaining variable projection importance VIP corresponding to each spectral variable based on the relational expression; marking a coefficient corresponding to the spectral variable of which the VIP is greater than 1 as an important model coefficient; marking a coefficient corresponding to the spectral variable of which the VIP is less than or equal to 1 as a general model coefficient; obtaining a to-be-tested data set, and processing the to-be-tested data set through to-be-calibrated equipment to obtain a transfer learning data set; obtaining an important residual error based on the important model coefficient and the transfer learning data set; obtaining a learning model coefficient based on the general model coefficient, the transfer learning data set and the important residual error; and obtaining a migration model coefficient based on the important model coefficient and the learning model coefficient, and realizing calibration of the to-be-calibrated equipment. And synchronous calibration of a plurality of LIBS devices can be efficiently and conveniently realized through a small number of transfer learning samples.

Description

Technical Field

[0001] The present invention relates to the technical field of synchronous calibration of LIBS equipment, and more specifically to a synchronous calibration method and system for multiple LIBS equipment based on transfer learning. Background Art

[0002] At present, laser induced breakdown spectroscopy (LIBS) technology is based on the instantaneous breakdown of samples by strong laser pulses to generate high-temperature plasma, and analyzes the elemental composition from the spectrum emitted by the plasma. Due to its fast speed, no need for sample pretreatment, and real-time online analysis, it is widely used in many scenarios such as metal processing, chemical industry, environmental monitoring, military and aerospace fields.

[0003] With the development of LIBS technology and the expansion of its application scenarios, accurate calibration has become particularly critical. In industrial sites, multiple LIBS devices are often deployed to meet continuous and large-scale detection needs. The performance of each LIBS device may vary due to slight differences in the manufacturing process, and changes in the use environment may also affect the test results. Therefore, in order to ensure that the test results between different devices are comparable, it is particularly important to accurately calibrate each device.

[0004] However, the traditional calibration method is to perform calibration on each device individually, which is not only repetitive but also time-consuming and requires a large number of standard samples. For example, in order to cover the different concentration ranges of the elements to be measured, a large number of standard samples with different known element contents may need to be prepared. Once there are many elements involved and a wide concentration range, the number of standard samples required and the time spent on calibration will increase exponentially; in addition, each time the equipment is calibrated, the equipment's operating time will be occupied, that is, the equipment cannot perform normal production tasks during the calibration period. This leads to a decrease in production efficiency, which is particularly serious for production lines that operate 24 hours a day.

[0005] Therefore, how to efficiently and conveniently achieve synchronous calibration of multiple LIBS devices using a small amount of standard samples is an urgent problem that technicians in this field need to solve. Summary of the invention

[0006] In view of this, the present invention provides a method and system for synchronous calibration of multiple LIBS devices based on transfer learning. By comprehensively calibrating a "source" device, a "source" model for quantitative analysis is established, and then the "source" model is migrated to other devices to be migrated. With only a small amount of transfer learning samples, the synchronous calibration of multiple LIBS devices can be achieved efficiently and conveniently.

[0007] In order to achieve the above object, the present invention adopts the following technical solution:

[0008] A method for synchronous calibration of multiple LIBS devices based on transfer learning, comprising:

[0009] Obtain calibration sample set, source device and device to be calibrated;

[0010] Calibrate the source device based on the calibration sample set to obtain a relationship between a source model and a spectral variable;

[0011] Based on the relationship, obtain the variable projection importance VIP corresponding to each of the spectral variables;

[0012] The coefficients corresponding to the spectral variables with VIP>1 are marked as important model coefficients;

[0013] The coefficients corresponding to the spectral variables with VIP≤1 are marked as general model coefficients;

[0014] Acquire a data set to be tested and process it through the device to be calibrated to obtain a transfer learning data set;

[0015] Obtaining important residuals based on the important model coefficients and the transfer learning data set;

[0016] Obtaining learning model coefficients based on the general model coefficients, the transfer learning data set, and the important residuals;

[0017] The migration model coefficients are obtained based on the important model coefficients and the learning model coefficients to achieve calibration of the device to be calibrated.

[0018] Preferably, before obtaining the device to be calibrated, the method further includes:

[0019] Determine whether the device to be calibrated meets the migration conditions:

[0020] Collecting spectra of the same sample based on the source device and the device to be calibrated, and correspondingly obtaining a first spectrum and a second spectrum;

[0021] If the number of peak position drifts between the first spectrum and the second spectrum does not exceed 5% of the total number of characteristic peaks, and the number of characteristic peaks with a relative deviation of the intensity of the characteristic peaks of the spectrum less than 10% is not less than 90% of the total number of characteristic peaks, then the migration condition is met;

[0022] Otherwise, the migration conditions are not met.

[0023] Preferably, calibrating the source device specifically includes:

[0024] Collecting the spectrum of the standard sample in the calibration sample set based on the source device;

[0025] The source device is calibrated based on the true value of the spectrum through the partial least square method of principal component analysis.

[0026] Preferably, the relationship between the source model and the spectral variable is specifically:

[0027] M S :Y=β0+β1x1+β2x2+...+β k x k +ε0

[0028] Among them, M S represents the source model, Y represents the response quantity, i.e., the industrial parameter of coal, β k Represents x k The corresponding model coefficient, x k represents the spectral variable corresponding to the k-th pixel position, that is, the average spectral intensity, and ε0 represents the initial residual value.

[0029] Preferably, obtaining the data set to be tested specifically includes:

[0030] Based on the calibration sample set, a preset number of samples are randomly selected to form a first sample set;

[0031] Arbitrarily obtain a certain number of samples to form a second sample set;

[0032] The data set to be tested is composed based on the first sample set and the second sample set.

[0033] Preferably, the important residual is obtained by:

[0034] Obtaining an important position i based on the important model coefficients;

[0035] Determine corresponding important spectral variables on the transfer learning dataset based on the important position i;

[0036] The predicted value y is obtained based on the important model coefficients and the important spectral variables:

[0037] y=β0+β1x1+β2x2+…+β i-1 x i-1 +β i x i , {i|VIP i >1}

[0038] Among them, β i represents important model coefficients, x i represents the i-th important spectral variable;

[0039] The important residual ε is obtained based on the predicted value y:

[0040] ε=Zy

[0041] Among them, Z represents the true value of a parameter of the sample.

[0042] Preferably, the learning model parameters are obtained, specifically including:

[0043] Obtaining a corresponding general position j based on the general model coefficients;

[0044] Determine a corresponding general spectral variable on the transfer learning dataset based on the general position j;

[0045] Based on the important residuals and the general spectral variables, the learning model coefficient β' is obtained v :

[0046] ε=β'0+β'1x1+β'2x2+…+β' v x j +ε'

[0047] Among them, x j represents the jth general spectral variable, and ε' represents the prediction residual.

[0048] Preferably, obtaining the migration model coefficients specifically includes:

[0049] Based on the important model coefficient β i and the learning model coefficient β' v The migration model coefficient β is obtained by combining t :

[0050] β t =β i +β' v .

[0051] A synchronous calibration system for multiple LIBS devices based on transfer learning, comprising: an acquisition and calibration module, a coefficient marking module, an important residual acquisition module, a coefficient acquisition module and an equipment calibration module;

[0052] The acquisition and calibration module is used to acquire a calibration sample set, a source device and a device to be calibrated; calibrate the source device based on the calibration sample set to obtain a relationship between a source model and a spectral variable;

[0053] The coefficient marking module is used to obtain the variable projection importance VIP corresponding to each of the spectral variables based on the relationship; mark the coefficients corresponding to the spectral variables with VIP>1 as important model coefficients; and mark the coefficients corresponding to the spectral variables with VIP≤1 as general model coefficients;

[0054] The important residual acquisition module is used to acquire the data set to be tested and process it through the device to be calibrated to obtain the transfer learning data set; and obtain the important residual based on the important model coefficients and the transfer learning data set;

[0055] The coefficient acquisition module is used to obtain the learning model coefficients based on the general model coefficients, the transfer learning data set and the important residuals;

[0056] The device calibration module is used to obtain the migration model coefficients based on the important model coefficients and the learning model coefficients to achieve calibration of the device to be calibrated.

[0057] It can be seen from the above technical solutions that, compared with the prior art, the present invention discloses a method and system for synchronous calibration of multiple LIBS devices based on transfer learning, which comprehensively calibrates a "source" device, establishes a "source" model for quantitative analysis, and then migrates the "source" model to other devices to be migrated. With only a small amount of transfer learning samples, synchronous calibration of multiple LIBS devices can be efficiently and conveniently achieved. It has the following beneficial effects:

[0058] 1. The method of the present invention realizes the migration application of LIBS quantitative analysis model based on variable importance analysis, greatly improves the efficiency of calibration of multiple LIBS equipment, and has great industrial application prospects.

[0059] 2. The present invention uses transfer learning to accelerate the learning process of new equipment by utilizing existing data, which is particularly beneficial for LIBS equipment with numerous parameters and complex configurations; at the same time, it significantly reduces the demand for the number of standard samples and greatly saves the time and manpower required for calibration.

[0060] 3. The present invention uses a transfer learning method, and the new device can quickly achieve performance equivalent to that of the source device through very few standard samples, which not only simplifies the operation process, but also enhances the generalization ability of the model and improves the robustness of the system.

[0061] 4. The method of the present invention not only provides a new solution for the calibration of LIBS equipment, but also opens up a new path for other detection technologies that require synchronous calibration of multiple devices. With the continuous advancement of big data and machine learning technology, this synchronous calibration method based on transfer learning will not only achieve accurate and efficient detection, but also promote the development of the entire detection technology field towards a more intelligent and automated direction.

[0062] 5. The LIBS quantitative analysis model migration method proposed in the present invention utilizes traditional machine learning methods to achieve iterative learning of the model, providing an efficient solution for the application of large-scale LIBS equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0064] Figure 1 A flow chart of a method for synchronous calibration of multiple LIBS devices based on transfer learning provided by the present invention.

[0065] Figure 2 This is a schematic diagram of the spectral analysis results of the source device and the device to be calibrated for the same sample provided by the present invention.

[0066] Figure 3 This is a schematic diagram of the source model calibration results provided by the present invention.

[0067] Figure 4 A schematic diagram of the real-time detection results of the ash content of online raw coal materials by using a migration model for the equipment to be calibrated provided by the present invention.

[0068] Figure 5 A schematic diagram of the structure of a synchronous calibration system for multiple LIBS devices based on transfer learning provided by the present invention. DETAILED DESCRIPTION

[0069] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0070] Example 1

[0071] like Figure 1 As shown, the embodiment of the present invention discloses a method for synchronous calibration of multiple LIBS devices based on transfer learning, comprising:

[0072] Obtain calibration sample set, source device and device to be calibrated;

[0073] The source device is calibrated based on the calibration sample set to obtain the relationship between the source model and the spectral variables;

[0074] Based on the relationship, the variable projection importance VIP corresponding to each spectral variable is obtained;

[0075] The coefficients corresponding to the spectral variables with VIP>1 are marked as important model coefficients;

[0076] The coefficients corresponding to the spectral variables with VIP≤1 are marked as general model coefficients;

[0077] Obtain the data set to be tested and process it through the device to be calibrated to obtain the transfer learning data set;

[0078] Obtain important residuals based on important model coefficients and transfer learning datasets;

[0079] Get the learning model coefficients based on the general model coefficients, the transfer learning dataset, and the important residuals;

[0080] The migration model coefficients are obtained based on the important model coefficients and the learning model coefficients to realize the calibration of the device to be calibrated.

[0081] Example 2

[0082] The embodiment of the present invention discloses a method for synchronous calibration of multiple LIBS devices based on transfer learning, comprising:

[0083] Get the calibration sample set, source device, and device to be calibrated:

[0084] Preferably, the calibration sample set includes multiple standard samples. In this embodiment, the calibration sample set includes 30 samples. In this embodiment, the source device and the device to be calibrated both use LIBS devices. The LIBS devices both use 300mJ pulsed lasers (Nd: YAG@1064nm), and the spectrometer detector is a linear array CCD with 6 channels × 2048 pixels.

[0085] Preferably, before obtaining the device to be calibrated, the method further includes:

[0086] Determine whether the device to be calibrated meets the migration conditions:

[0087] The spectrum of the same sample is collected based on the source device and the device to be calibrated, and a first spectrum and a second spectrum are correspondingly obtained;

[0088] If the number of peak position drifts between the first spectrum and the second spectrum does not exceed 5% of the total number of characteristic peaks, and the number of characteristic peaks with a relative deviation of the intensity of the characteristic peaks of the spectrum less than 10% is not less than 90% of the total number of characteristic peaks, then the migration condition is met;

[0089] Otherwise, the migration conditions are not met.

[0090] The source device is calibrated based on the calibration sample set, and the relationship between the source model and the spectral variables is obtained:

[0091] Preferably, calibrating the source device specifically includes:

[0092] Acquiring spectra of standard samples in a calibration sample set based on a source device;

[0093] The source equipment is calibrated based on the true value of the spectrum through the partial least squares method of principal component analysis.

[0094] Preferably, the source device collects the calibration sample set to obtain a spectrum S:

[0095]

[0096] After data preprocessing, spectrum S is averaged to obtain the average spectrum

[0097]

[0098] Based on the average spectrum The relationship between the source model and the spectral variables is obtained.

[0099] Preferably, the relationship between the source model and the spectral variable is specifically:

[0100] M S :Y=β0+β1x1+β2x2+...+β k x k +ε0

[0101] Among them, M S represents the source model, Y represents the response quantity, i.e., the industrial parameter of coal, β k Represents x k The corresponding model coefficient, x k represents the spectral variable corresponding to the k-th pixel position, that is, the average spectral intensity, and ε0 represents the initial residual value.

[0102] Based on the relationship, the variable projection importance VIP corresponding to each spectral variable is obtained:

[0103]

[0104] Among them, K, a and A represent the total number of variables, the number of principal components used and the PCA-PLS dimension respectively, Wa represents the PLS (partial least squares) coefficient on the ath principal component, SSY comp,a SSY represents the sum of squares of the explanation rate of the ath principal component for the response Y. cum It represents the cumulative sum of squares explained by all principal components for the response quantity Y.

[0105] Preferably, by calculating the source model M S The variable projection importance (VIP) determines the locations of spectral variables that contribute more to the model.

[0106] The coefficients corresponding to the spectral variables with VIP>1 are marked as important model coefficients β i :{VIP i>1}, the corresponding spectral variable is located at position i:{VIP i >1}.

[0107] The coefficients corresponding to the spectral variables with VIP≤1 are labeled as the general model coefficients β j :{VIP j ≤1}, the corresponding spectral variable is located at j:{VIP j ≤1}.

[0108] Get the data set to be tested and process it through the device to be calibrated to get the transfer learning data set:

[0109] Preferably, obtaining the data set to be tested specifically includes:

[0110] Based on the calibration sample set, a preset number of samples are randomly selected to form a first sample set;

[0111] Arbitrarily obtain a certain number of samples to form a second sample set;

[0112] A data set to be tested is formed based on the first sample set and the second sample set.

[0113] Preferably, in this embodiment, the number of samples in the first sample set and the second sample set is the same.

[0114] Preferably, spectral data of the data set to be tested is collected based on the device to be calibrated to obtain a transfer learning data set.

[0115] Obtain important residuals based on important model coefficients and transfer learning dataset:

[0116] Preferably, the important residual is obtained by:

[0117] Based on the important model coefficients, the important position i is obtained accordingly;

[0118] Determine the corresponding important spectral variables on the transfer learning dataset based on the important position i;

[0119] The predicted value y is obtained based on important model coefficients and important spectral variables:

[0120] y=β0+β1x1+β2x2+…+β i-1 x i-1 +β i x i , {i|VIP i >1}

[0121] Among them, β i represents important model coefficients, x i represents the i-th important spectral variable;

[0122] Based on the predicted value y, we get the important residual ε:

[0123] ε=Zy

[0124] Among them, Z represents the true value of a parameter of the sample and is a known quantity.

[0125] The learning model coefficients are obtained based on the general model coefficients, the transfer learning dataset, and the important residuals:

[0126] Preferably, the learning model parameters are obtained, specifically including:

[0127] Based on the general model coefficients, the corresponding general position j is obtained;

[0128] Determine the corresponding general spectral variable on the transfer learning dataset based on the general position j;

[0129] Based on the important residuals and general spectral variables, the learning model coefficients β' are obtained v :

[0130] ε=β'0+β'1x1+β'2x2+…+β' v x j +ε'

[0131] Among them, x j represents the jth general spectral variable, ε' represents the prediction residual, which is used to indicate how large the error of the model prediction is, and it must be ensured to be as close to zero as possible.

[0132] Preferably, the spectral variable x at the corresponding position i in the transfer learning dataset is i Zero: {x i |x i =0}, achieving a model coefficient β of VIP>1 i Lock and use the remaining variable x j And the important residual ε get the learning model coefficient β' v .

[0133] The migration model coefficients are obtained based on the important model coefficients and the learning model coefficients to realize the calibration of the device to be calibrated.

[0134] Preferably, obtaining the migration model coefficients specifically includes:

[0135] Based on the important model coefficient β i and learning model coefficients β' v Merge to get the migration model coefficient β t :

[0136] β t =β i +β' v .

[0137] Preferably, the device to be calibrated is calibrated based on the obtained migration model coefficients to obtain a calibrated device, and the data to be tested is input into the calibrated device to obtain corresponding test results.

[0138] Example 3

[0139] In order to verify the effectiveness of the method of the invention, this example compares the quantitative analysis results of the industrial parameters of raw coal by the online detection LIBS system.

[0140] In this embodiment, both the source device and the device to be calibrated use an online LIBS device, the online LIBS device uses a 300mJ pulse laser (Nd:YAG@1064nm), and the spectrometer detector uses a linear array CCD with 6 channels×2048 pixels.

[0141] It is determined that the optical systems of the source device and the device to be calibrated have similar spectral resolution capabilities for the same sample. The total number of characteristic peaks with a relative deviation of less than 10% in the characteristic peak intensity of the source device and the device to be calibrated (device 1, device 2) for the same sample spectrum collected is about 91%, and the number of peaks with drifting characteristic peaks accounts for 2.1% of the total number of characteristic peaks, which meets the preconditions of model migration, such as Figure 2 shown.

[0142] This embodiment uses 30 samples to calibrate the source device. The model results after calibration are as follows: Figure 3 As shown, the calibrated elements include moisture (wt.%), ash (wt.%), total sulfur (wt.%) and calorific value (MJ / kg).

[0143] Ten model migration samples are selected to form the data set to be tested, among which five calibration samples are randomly selected from the calibration sample set, and five samples are randomly selected from outside the calibration sample set. The spectral data of the data set to be tested are collected in sequence through the device to be calibrated to form a transfer learning data set.

[0144] Calculate the source model M corresponding to the source device in turn S The variable projection importance (VIP) of the ash model is used as an example in this example. There are 126 important model coefficients with VIP>1, which are marked as β i :{VIP i >1}, the corresponding spectral variable is located at position i:{VIP i >1}; there are 211 general model coefficients with VIP≤1, marked as β j :{VIP j ≤1}, the corresponding spectral variable is located at j:{VIP j >1}.

[0145] In the transfer learning dataset, the position is selected as i:{VIP i>1}, and calculate the model coefficient β i :{VIP i >1} Predicted values ​​and important residuals on 126 variables in the migration dataset.

[0146] By calculating the 10 samples in the transfer learning data set, 10 prediction values ​​and residuals are obtained as shown in Table 1:

[0147] Table 1 Prediction values ​​and residuals of the coefficients of the ash model VIP>1 on the migration data set

[0148]

[0149]

[0150] The 126 coefficients β with VIP>1 i Lock, set the important spectral variables at the corresponding i position in the transfer learning dataset to zero: {x i |x i =0}, and use the general spectral variables corresponding to the remaining position j and the important residual ε to reconstruct the machine learning model based on PCA-PLS to obtain the learning model coefficient β' v .

[0151] The important model coefficient β i and learning model coefficients β' v Merge to get the migration model coefficient β t , based on the migration model coefficient β t Calibrate device 2.

[0152] The coal sample to be tested is tested in real time by the calibrated device 2, with a detection frequency of 1 Hz and a detection time of 3600 seconds. This example takes the ash content test result as an example. Figure 4 As shown, the overall average value of the real-time detection values ​​in the figure is close to the test value given by the laboratory, indicating that the migration calibration is successful.

[0153] Example 4

[0154] like Figure 5 As shown, a synchronous calibration system for multiple LIBS devices based on transfer learning includes: an acquisition and calibration module, a coefficient marking module, an important residual acquisition module, a coefficient acquisition module and an equipment calibration module;

[0155] The calibration module is used to obtain a calibration sample set, a source device and a device to be calibrated; the source device is calibrated based on the calibration sample set to obtain a relationship between a source model and a spectral variable;

[0156] A coefficient marking module is used to obtain the variable projection importance VIP corresponding to each spectral variable based on the relationship; the coefficients corresponding to the spectral variables with VIP>1 are marked as important model coefficients; the coefficients corresponding to the spectral variables with VIP≤1 are marked as general model coefficients;

[0157] An important residual acquisition module is used to acquire the data set to be tested and process it through the device to be calibrated to obtain the transfer learning data set; and obtain important residuals based on the important model coefficients and the transfer learning data set;

[0158] A coefficient acquisition module is used to obtain learning model coefficients based on general model coefficients, transfer learning datasets, and important residuals;

[0159] The device calibration module is used to obtain the migration model coefficients based on the important model coefficients and the learning model coefficients to realize the calibration of the device to be calibrated.

[0160] Through the above technical solutions, it can be known that the present invention discloses a method and system for synchronous calibration of multiple LIBS devices based on transfer learning, which comprehensively calibrates a "source" device, establishes a "source" model for quantitative analysis, and then migrates the "source" model to other devices to be migrated. With only a small amount of transfer learning samples, it is possible to efficiently and conveniently achieve synchronous calibration of multiple LIBS devices. It has the following beneficial effects:

[0161] 1. The method of the present invention realizes the migration application of LIBS quantitative analysis model based on variable importance analysis, greatly improves the efficiency of calibration of multiple LIBS equipment, and has great industrial application prospects.

[0162] 2. The present invention uses transfer learning to accelerate the learning process of new equipment by utilizing existing data, which is particularly beneficial for LIBS equipment with numerous parameters and complex configurations; at the same time, it significantly reduces the demand for the number of standard samples and greatly saves the time and manpower required for calibration.

[0163] 3. The present invention uses a transfer learning method, and the new device can quickly achieve performance equivalent to that of the source device through very few standard samples, which not only simplifies the operation process, but also enhances the generalization ability of the model and improves the robustness of the system.

[0164] 4. The method of the present invention not only provides a new solution for the calibration of LIBS equipment, but also opens up a new path for other detection technologies that require synchronous calibration of multiple devices. With the continuous advancement of big data and machine learning technology, this synchronous calibration method based on transfer learning will not only achieve accurate and efficient detection, but also promote the development of the entire detection technology field towards a more intelligent and automated direction.

[0165] 5. The LIBS quantitative analysis model migration method proposed in the present invention utilizes traditional machine learning methods to achieve iterative learning of the model, providing an efficient solution for the application of large-scale LIBS equipment.

[0166] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0167] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for synchronous calibration of multiple LIBS devices based on transfer learning, characterized in that: include: Obtain calibration sample set, source device and device to be calibrated; Calibrate the source device based on the calibration sample set to obtain a relationship between a source model and a spectral variable; Based on the relationship, obtain the variable projection importance VIP corresponding to each of the spectral variables; The coefficients corresponding to the spectral variables with VIP>1 are marked as important model coefficients; The coefficients corresponding to the spectral variables with VIP≤1 are marked as general model coefficients; Acquire a data set to be tested and process it through the device to be calibrated to obtain a transfer learning data set; Obtaining important residuals based on the important model coefficients and the transfer learning data set; Obtaining learning model coefficients based on the general model coefficients, the transfer learning data set, and the important residuals; The migration model coefficients are obtained based on the important model coefficients and the learning model coefficients to achieve calibration of the device to be calibrated.

2. According to a method for synchronous calibration of multiple LIBS devices based on transfer learning according to claim 1, it is characterized in that: Before obtaining the device to be calibrated, the following steps are also required: Determine whether the device to be calibrated meets the migration conditions: Collecting spectra of the same sample based on the source device and the device to be calibrated, and correspondingly obtaining a first spectrum and a second spectrum; If the number of peak position drifts between the first spectrum and the second spectrum does not exceed 5% of the total number of characteristic peaks, and the number of characteristic peaks with a relative deviation of the intensity of the characteristic peaks of the spectrum less than 10% is not less than 90% of the total number of characteristic peaks, then the migration condition is met; Otherwise, the migration conditions are not met.

3. The method for synchronous calibration of multiple LIBS devices based on transfer learning according to claim 1, characterized in that: Calibrating the source device specifically includes: Collecting the spectrum of the standard sample in the calibration sample set based on the source device; The source device is calibrated based on the true value of the spectrum through the partial least square method of principal component analysis.

4. The method for synchronous calibration of multiple LIBS devices based on transfer learning according to claim 1, characterized in that: The relationship between the source model and the spectral variable is specifically: M S :Y=β0+β1x1+β2x2+...+β k x k +ε0 Among them, M S represents the source model, Y represents the response quantity, i.e., the industrial parameter of coal, β k Represents x k The corresponding model coefficient, x k represents the spectral variable corresponding to the k-th pixel position, that is, the average spectral intensity, and ε0 represents the initial residual value.

5. The method for synchronous calibration of multiple LIBS devices based on transfer learning according to claim 1, characterized in that: Get the data set to be tested, including: Based on the calibration sample set, a preset number of samples are randomly selected to form a first sample set; Arbitrarily obtain a certain number of samples to form a second sample set; The data set to be tested is composed based on the first sample set and the second sample set.

6. The method for synchronous calibration of multiple LIBS devices based on transfer learning according to claim 1, characterized in that: The important residual is obtained by: Obtaining an important position i based on the important model coefficients; Determine corresponding important spectral variables on the transfer learning dataset based on the important position i; The predicted value y is obtained based on the important model coefficients and the important spectral variables: y=β0+β1x1+β2x2+…+β i-1 x i-1 +b i x i VIP i >1} Among them, β i represents important model coefficients, x i represents the i-th important spectral variable; The important residual ε is obtained based on the predicted value y: ε=Zy Among them, Z represents the true value of a parameter of the sample.

7. The method for synchronous calibration of multiple LIBS devices based on transfer learning according to claim 6, characterized in that: Get the learning model parameters, including: Obtaining a corresponding general position j based on the general model coefficients; Determine a corresponding general spectral variable on the transfer learning dataset based on the general position j; Based on the important residuals and the general spectral variables, the learning model coefficient β' is obtained v : ε=β'0+β'1x1+β'2x2+…+β' v x j +e' Among them, x j represents the jth general spectral variable, and ε' represents the prediction residual.

8. The method for synchronous calibration of multiple LIBS devices based on transfer learning according to claim 7, characterized in that: Obtaining the migration model coefficients specifically includes: Based on the important model coefficient β i and the learning model coefficient β' v The migration model coefficient β is obtained by combining t : β t ＝β i +β' v 。 9. A synchronous calibration system for multiple LIBS devices based on transfer learning, characterized in that: include: Acquisition calibration module, coefficient marking module, important residual acquisition module, coefficient acquisition module and equipment calibration module; The acquisition and calibration module is used to acquire a calibration sample set, a source device and a device to be calibrated; calibrate the source device based on the calibration sample set to obtain a relationship between a source model and a spectral variable; The coefficient marking module is used to obtain the variable projection importance VIP corresponding to each of the spectral variables based on the relationship; mark the coefficients corresponding to the spectral variables with VIP>1 as important model coefficients; and mark the coefficients corresponding to the spectral variables with VIP≤1 as general model coefficients; The important residual acquisition module is used to acquire the data set to be tested and process it through the device to be calibrated to obtain the transfer learning data set; Obtaining important residuals based on the important model coefficients and the transfer learning data set; The coefficient acquisition module is used to obtain the learning model coefficients based on the general model coefficients, the transfer learning data set and the important residuals; The device calibration module is used to obtain the migration model coefficients based on the important model coefficients and the learning model coefficients to achieve calibration of the device to be calibrated.

Images