Rare earth element extraction separation analysis method and system based on ionic liquid
By combining physical models and neural networks, a hybrid neural network model was developed to address the problem of opaque model decisions caused by spectral complexity in LIBS technology. This model ensures interpretability and reliability in the rare earth element extraction and separation process, and guarantees the credibility of analysis in sparse or extrapolated data regions.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-15
- Publication Date
- 2026-03-24
AI Technical Summary
In existing technologies, the spectral complexity of LIBS technology during the extraction and separation of rare earth elements leads to opaque model decisions, making it difficult to explain the model's decision path and reliability.
A hybrid neural network model is adopted, combining a physical model and a neural network. The hybrid model separates spectral line interference in the spectral information and constructs a hybrid neural network model to identify and correct spectral line interference. This includes a correction model based on the basic physical model and a fully connected neural network, as well as a structured residual neural network and a multi-task learning model, to further distinguish and label spectral line interference.
It improves the transparency and traceability of the model analysis path, ensures the reliability of sample prediction results outside the training data distribution, avoids uncontrollable situations in the analysis path, and enhances the credibility and interpretability of the analysis.
Smart Images
Figure CN121324281B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of rare earth separation and analysis technology, and in particular to a method and system for rare earth element extraction, separation and analysis based on ionic liquids. Background Technology
[0002] The extraction and separation process of rare earth elements requires sophisticated analytical methods to monitor and evaluate its effectiveness. These methods are mainly divided into: rapid analytical methods for process monitoring, offline high-precision analytical methods for accurate determination of elemental content and composition, and speciation and distribution analysis methods for in-depth study of extraction and process optimization.
[0003] Among the optimized morphology and distribution analysis methods is LIBS technology, which uses high-power laser pulses to generate ionic plasmas on the sample surface and analyzes the elemental distribution morphology by analyzing the characteristic spectra emitted during the cooling process of the ionic plasmas. This technology offers advantages such as no need for sample duplication; real-time monitoring of dynamic changes in morphology during extraction and reaction; and the ability to acquire information on rare earth elements, ligands, and other related elements. However, the spectral complexity—specifically, the complex spectral background and potential interferences generated by ionic liquids—poses challenges to the analysis using this technology.
[0004] Existing technologies typically employ hardware optimization and software algorithm correction to identify and separate interfering spectra. Principal component regression (PCR) models and neural network models are commonly used in software algorithm correction to handle complex matrix effects and interference relationships. However, the decision-making process of these models is opaque and difficult to explain physically—a "black box" characteristic that prevents analysts from determining the model's decision path and reliability.
[0005] Therefore, improving the transparency of model decisions during the process of separating interference is a problem that must be solved to enhance the reliability of the analysis. Summary of the Invention
[0006] The present invention uses a hybrid neural network model consisting of a physical model and a neural network. The output predicted spectrum includes the continuous spectral profile output by the physical model and the residual spectrum output by the neural network. The elemental distribution state is analyzed based on these two spectra, making the model analysis path transparent and traceable.
[0007] The technical solution proposed in this invention is: a method and system for rare earth element extraction, separation and analysis based on ionic liquids, wherein the method includes:
[0008] The sample to be tested is an ionic liquid containing multiple rare earth complexes after an extraction reaction.
[0009] The process involves acquiring spectral information of the sample to be tested, generating an elemental distribution map of the sample, and analyzing the morphological distribution of the sample based on the elemental distribution map; including:
[0010] Construct a hybrid neural network model to separate spectral line interference in spectral information through the hybrid model;
[0011] Extract the spectral line intensity values from multiple measurement points from the interference-free spectral lines to obtain an accurate elemental distribution map.
[0012] Preferably, the construction of the hybrid neural network model, which separates spectral line interference in the spectral information through the hybrid model, includes:
[0013] Constructing a hybrid neural network model includes:
[0014] Construct a fundamental physical model; introduce neural networks to correct the shortcomings of the fundamental physical model;
[0015] Optimize the hybrid neural network model to identify weak interference in spectral line interference;
[0016] Further distinguish between weak interference and unknown interference in spectral line interference.
[0017] Preferably, the construction of the basic physical model includes:
[0018] Establish the relationship between spectral line intensity and plasma parameters and elemental concentration; that is, for a spectral line intensity consisting of energy levels... arrive Atomic or ionic spectral lines generated by transitions spectral line intensity ;
[0019] in, Indicates the efficiency factor; Elements representing emission spectral lines The concentration; This represents the Einstein spontaneous emission coefficient for the transition; Indicates the upper energy level Statistical weights; Represents element The allocation function, Indicates the local temperature of the plasma; Indicates the upper energy level The excitation energy; Represents the Boltzmann constant;
[0020] Discrete Converted into a continuous spectral profile ; Indicates at wavelength Theoretical synthesized spectral intensity at the location; This represents the contour function, Voigt function; Indicates the first The center wavelength of the spectral lines, This represents the Gaussian width parameter of the Voigt function; This represents the Lorentz width parameter of the Voigt function.
[0021] Preferably, the introduction of neural networks to correct deficiencies in the fundamental physical model includes:
[0022] An input layer is established to combine the concentration data of all elements contained in the sample to be detected into an element concentration vector. ;
[0023] A calibration model is constructed based on a fully connected neural network.
[0024] by The input is the residual spectrum, which is fed into the trained calibration model and outputs the residual spectrum. A fusion layer is constructed to add the output of the basic physical model and the output of the fully connected neural network model to obtain the predicted spectrum. .
[0025] Preferably, the optimized hybrid neural network model for identifying weak interference in spectral line interference includes:
[0026] Will Further decomposition, namely:
[0027] ;in, Indicates spectral lines The ideal spectral profile, Indicates spectral lines The self-absorption attenuation factor, , Indicates the absorption coefficient; Indicates the outline of the absorption spectral lines; Indicates spectral lines Continuous background radiation;
[0028] A structured residual neural network is introduced, that is, a multi-head residual network is introduced, where each head learns the corresponding type of physical model error;
[0029] Will The input is fed into a trained structured residual neural network, and the output is... , and ;
[0030] Then, the new predicted spectrum ;in, This represents the self-absorption residual; Indicates background residual; This indicates unmodeled residuals.
[0031] Preferably, the further distinction between weak interference and unknown interference in spectral line interference includes:
[0032] Construct a multi-task learning model, which includes a shared encoder and a multi-task decoder;
[0033] Will The input is fed into a pre-trained multi-task learning model, which outputs predicted values of plasma electron temperature and electron density. and ;
[0034] From the original spectral lines The calculated values of plasma electron temperature and electron density were independently obtained using the Boltzmann diagram method. and ;
[0035] Obtain spectral prediction error ;in, Indicates spectral lines Measured spectral line profile;
[0036] Prediction error of plasma characteristic parameters ;
[0037] From each sample to be tested Extract morphological features from them to construct the corresponding residual sample feature vectors;
[0038] Unsupervised clustering is performed on the feature vectors of the residual samples;
[0039] For each cluster obtained, the corresponding average residual spectrum is calculated, and the interference type is labeled based on the average residual spectrum.
[0040] Preferably, the unsupervised clustering of the residual sample feature vectors includes:
[0041] For a Calculate global statistical features, including:
[0042] Calculate skewness ;in, Indicates the number of samples;
[0043] Calculate kurtosis ;in, This represents the mean of the error. The standard deviation represents the error;
[0044] Calculate shape and scale features, including:
[0045] calculate autocorrelation function width ;in, Represents a set of wavelengths;
[0046] Normalized energy ratio ;in, Indicates the set of spectral lines of interest; if If the spectral interference is concentrated near the analytical line, it is either self-absorption interference or overlap interference. If so, the spectral interference is determined to be either background interference or uniformly distributed interference;
[0047] Obtain the feature vector corresponding to each sample spectral line. ;
[0048] Cluster analysis of the feature vectors of all samples to be tested was performed using the Gaussian mixture algorithm, including:
[0049] choose The sample to be tested;
[0050] Perform Gaussian mixture clustering:
[0051] Calculate probability ;in, Indicates the first The mean and covariance of each Gaussian component; Indicates the number of samples to be tested in the cluster; Indicates mixed weights;
[0052] Each sample to be tested is assigned to the cluster with the highest probability. Inside;
[0053] For each obtained cluster, the corresponding average residual spectrum is calculated, and interference type labeling is performed based on the average residual spectrum, including:
[0054] For each cluster, calculate the average residual spectrum of all samples to be tested within that cluster. ;
[0055] Output Manual labeling is then performed.
[0056] Preferred options also include:
[0057] Define the physical tolerance region:
[0058] Get ;
[0059] Apply nonnegativity constraints: that is, set the following constraints:
[0060] ;
[0061] Constructing a physical constraint loss function Among them, data fitting loss ; Indicates physical constraint loss; Indicates the adversarial regularization loss; Indicates hyperparameters;
[0062] ;
[0063] ;
[0064] in, ; ;
[0065] ;
[0066] ; Represents the gradient of the background spectrum;
[0067] Import a pre-trained distribution discriminator, which is built based on a neural network;
[0068] Will The input is fed into the distribution decision unit, which outputs the concentration sample distribution vector. ;
[0069] Define adversarial regularization or loss ;in, This indicates that the distribution discriminator reacts to the input. Probability estimate of belonging to the training set; Represents the concentration distribution vector of the training set; if The larger, the more it means The further outward the region; This represents a measurement function;
[0070] By fixing the distribution decision parameters, updating the hybrid neural network model parameters, we obtain the hybrid neural network model parameters that minimize physical constraints.
[0071] A rare earth element extraction and separation analysis system based on ionic liquid, the system being used to perform the aforementioned rare earth element extraction and separation analysis method based on ionic liquid.
[0072] A computer-readable storage medium storing a computer program that is executed by a processor to implement the aforementioned method for rare earth element extraction, separation and analysis based on ionic liquid.
[0073] The beneficial effects of this invention are:
[0074] This invention, by applying a physical tolerance domain, i.e., physical constraints, ensures that the predictions of the hybrid neural network model will not go out of control when encountering samples outside the training data distribution (extrapolation), thereby avoiding the problem of decreased traceability of the analysis path. Attached Figure Description
[0075] Figure 1 This is a flowchart of the rare earth element extraction and separation analysis method based on ionic liquid of the present invention. Detailed Implementation
[0076] The following description is intended to disclose the present invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious modifications will occur to those skilled in the art. The basic principles of the invention defined in the following description can be applied to other embodiments, modifications, improvements, equivalents, and other technical solutions that do not depart from the spirit and scope of the invention.
[0077] It is understood that the term "a" should be understood as "at least one" or "one or more", that is, in one embodiment, the number of an element can be one, while in another embodiment, the number of the element can be multiple, and the term "a" should not be understood as a limitation on the number.
[0078] Example 1:
[0079] refer to Figure 1 The technical solution provided by this invention is: a method and system for rare earth element extraction, separation and analysis based on ionic liquid, wherein the method includes:
[0080] Step 1: Obtain the sample to be tested, which is an ionic liquid containing multiple rare earth complexes after extraction reaction;
[0081] Step 2: Collect the spectral information of the sample to be tested, generate the elemental distribution map of the sample to be tested, and analyze the morphological distribution of the sample to be tested based on the elemental distribution map; including:
[0082] Step 2.1: Construct a hybrid neural network model to separate spectral line interference in the spectral information using the hybrid model;
[0083] Step 2.2: Extract the spectral line intensity values from multiple measurement points from the interference-free spectral lines to obtain an accurate elemental distribution map.
[0084] Step 2.1 specifically includes the following steps:
[0085] Step 2.11: Construct a hybrid neural network model, specifically as follows:
[0086] Construct a fundamental physical model; introduce neural networks to correct the shortcomings of the fundamental physical model.
[0087] The construction of the fundamental physical model, based on the basic principles of LIBS spectral generation—namely, the relationship between spectral line intensity and plasma parameters and elemental concentration—includes the following steps:
[0088] Establish the relationship between spectral line intensity and plasma parameters and elemental concentration; that is, for a spectral line intensity consisting of energy levels... arrive Atomic or ionic spectral lines generated by transitions spectral line intensity ;
[0089] in, This represents the efficiency factor (which is usually considered a constant). Elements representing emission spectral lines The concentration; This represents the Einstein spontaneous emission coefficient for the transition; Indicates the upper energy level Statistical weights; Represents element The allocation function, Indicates the local temperature of the plasma; Indicates the upper energy level The excitation energy; Represents the Boltzmann constant;
[0090] Discrete Converted into a continuous spectral profile ; Indicates at wavelength Theoretical synthesized spectral intensity at the location; This represents the contour function, Voigt function; Indicates the first The center wavelength of the spectral lines, This represents the Gaussian width parameter of the Voigt function; This represents the Lorentz width parameter of the Voigt function.
[0091] The process of introducing neural networks to correct the deficiencies of the fundamental physical model includes the following steps:
[0092] An input layer is established to combine the concentration data of all elements contained in the sample to be detected into an element concentration vector. ;
[0093] A calibration model is constructed based on a fully connected neural network; the task of the fully connected neural network is to learn the mapping from the concentration vector C to the physical model residuals.
[0094] With concentration vector The input is the residual spectrum, which is fed into the trained calibration model and outputs the residual spectrum. ;
[0095] A fusion layer is constructed to add the output of the basic physics model and the output of the fully connected neural network model to obtain the predicted spectrum. .
[0096] For any predicted spectrum, we can clearly decompose it into... (Physics section) and (Data-driven residual part) That is, to perform contribution decomposition.
[0097] if This indicates that the physical model accurately describes the data; if... A large contribution decomposition clearly indicates which bands exhibit significant effects not captured by the physical model, such as strong self-absorption or unknown interference. This improves the traceability of the model analysis path, thereby enhancing the transparency of model decision-making.
[0098] Some parts are generated entirely by interpretable physical formulas. Researchers can examine each of them, for example, by analyzing... Changes can reveal system stability through analysis. It allows us to understand the state of the plasma.
[0099] Step 2.12: Optimize the hybrid neural network model to identify weak interference in spectral line interference. This specifically includes the following:
[0100] Will Further decomposition, namely:
[0101] ;in, Indicates spectral lines The ideal spectral profile, Indicates spectral lines The self-absorption attenuation factor, , Indicates the absorption coefficient; Indicates the outline of the absorption spectral lines; Indicates spectral lines Continuous background radiation;
[0102] A structured residual neural network is introduced, that is, a multi-head residual network is introduced, where each head learns the corresponding type of physical model error;
[0103] Will The input is fed into a trained structured residual neural network, and the output is... , and ;
[0104] Then, the new predicted spectrum ;in, This represents the self-absorption residual; Indicates background residual; This indicates unmodeled residuals.
[0105] Step 2.13: Further distinguish between weak interference and unknown interference in spectral line interference. This specifically includes the following steps:
[0106] Step 2.130: Construct a multi-task learning model, which includes a shared encoder and a multi-task decoder;
[0107] Step 2.131, The input is fed into a pre-trained multi-task learning model, which outputs predicted values of plasma electron temperature and electron density. and ;
[0108] Step 2.132, from the original spectral lines The calculated values of plasma electron temperature and electron density were independently obtained using the Boltzmann diagram method. and ;
[0109] Step 2.133: Obtain the spectral prediction error ;in, Indicates spectral lines Measured spectral line profile;
[0110] Step 2.134: Obtain the prediction error of plasma characteristic parameters. ;
[0111] Step 2.135, from each sample to be tested The morphological features are extracted to form the corresponding residual sample feature vector.
[0112] Step 2.136: Perform unsupervised clustering on the residual sample feature vectors, specifically including the following steps:
[0113] For a Calculate global statistical features, including:
[0114] Calculate skewness (a measure of the symmetry of the error distribution). ;in, Indicates the number of samples;
[0115] Calculate kurtosis (a measure of the sharpness of the error distribution). ;in, This represents the mean of the error. The standard deviation represents the error;
[0116] Calculate shape and scale features, including:
[0117] calculate autocorrelation function width ;in, Represents a set of wavelengths;
[0118] Normalized energy ratio ;in, Indicates the set of spectral lines of interest; if If the spectral interference is concentrated near the analytical line, it is either self-absorption interference or overlap interference. If so, the spectral interference is determined to be either background interference or uniformly distributed interference;
[0119] Obtain the feature vector corresponding to each sample spectral line. ;
[0120] Cluster analysis of the feature vectors of all samples to be tested was performed using the Gaussian mixture algorithm, including:
[0121] choose The sample to be tested;
[0122] Perform Gaussian mixture clustering:
[0123] Calculate probability ;in, Indicates the first The mean and covariance of each Gaussian component; Indicates the number of samples to be tested in the cluster; Indicates mixed weights;
[0124] Each sample to be tested is assigned to the cluster with the highest probability. Inside.
[0125] Step 2.137: For each obtained cluster, calculate the corresponding average residual spectrum, and label the interference type based on the average residual spectrum. This includes the following steps:
[0126] For each cluster, calculate the average residual spectrum of all samples to be tested within that cluster. ;
[0127] Output Manual labeling is then performed.
[0128] For example, if If a strong spectral line has a symmetrical V-shaped depression at its center, it is labeled as weak self-absorption.
[0129] if If a wide, gentle bulge appears in a wide bandwidth region, it is labeled as CN molecular band interference;
[0130] if If sharp positive and negative peaks appear at multiple positions, it is marked as positional spectral line overlap;
[0131] if Featureless shifts across the full spectrum, and If the value is large, it is marked as an inaccurate background estimation.
[0132] Example 2:
[0133] In sparse or extrapolated regions (samples outside the training data distribution), the behavior of hybrid neural network models may become uncontrollable and uninterpretable. Therefore, based on Example 1, we propose the following technical solution:
[0134] Define the physical tolerance region:
[0135] Get ;
[0136] Apply nonnegativity constraints: that is, set the following constraints:
[0137] ;
[0138] Constructing a physical constraint loss function Among them, data fitting loss This is used to ensure that the model can reproduce the observed data; Indicates physical constraint loss; This represents the adversarial regularization loss, used to constrain outward propagation behavior; Indicates hyperparameters;
[0139] ;
[0140] ;in, ; ;
[0141] ;
[0142] ; Represents the gradient of the background spectrum;
[0143] Import a pre-trained distribution discriminator, which is built based on a neural network;
[0144] Will The input is fed into the distribution decision unit, which outputs the concentration sample distribution vector. ;
[0145] Define adversarial regularization or loss ;in, This indicates that the distribution discriminator reacts to the input. Probability estimate of belonging to the training set; Represents the concentration distribution vector of the training set; if The larger, the more it means The further outward the region; This represents a measurement function used to measure the difference between the predicted spectrum and the predictions of the pure physics model; for example, it's the mean squared error function. This loss term implies that when the main model makes a prediction on an uncertain, extrapolated input, the output should not deviate too far from the prediction of the pure physics model. As an anchor value for extrapolation.
[0146] With the distribution judge parameters fixed, the hybrid neural network model parameters are updated to obtain the hybrid neural network model parameters that minimize physical constraints. The model parameters include the connection weights and biases of each sub-neural network.
[0147] Through the above constraint framework, we enhance the reliability of traceability in two ways: direct constraints, Directly ensure that the decomposed parts, for example Even in sparse regions of data, it adheres to basic physical laws (nonnegativity, etc.), greatly increasing the reliability of the decomposition.
[0148] This ensures that the model's overall behavior in the extrapolation region does not deviate from the theoretical physical values. Since the theoretical values are fully interpretable, this provides an interpretable benchmark for traceability in the extrapolation region, allowing analysts to make comparisons. And understand what corrections were made to the physical predictions during extrapolation, thereby reducing the possibility of spurious decompositions in the model that are incomprehensible due to extrapolation.
[0149] The present invention also provides a rare earth element extraction and separation analysis system based on ionic liquid, the system being used to perform the aforementioned rare earth element extraction and separation analysis method based on ionic liquid.
[0150] The present invention also provides a computer-readable storage medium storing a computer program that is executed by a processor to implement the described method for rare earth element extraction, separation and analysis based on ionic liquid.
[0151] The processes described above with reference to the flowcharts in the embodiments disclosed in this invention can be implemented as computer software programs. The embodiments disclosed in this invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication component, and / or installed from a removable medium. When the computer program is executed by a central processing unit (CPU), it performs the functions defined in the methods of this application. It should be noted that the computer-readable medium described above in this application can be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. Computer-readable storage media can be, for example, but not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wire segments, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in connection with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on a computer-readable medium may be transmitted using any suitable medium, including but not limited to: wireless segments, wire segments, optical fibers, RF, etc., or any suitable combination thereof.
[0152] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation that may be implemented in systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.
[0153] Those skilled in the art should understand that the embodiments of the present invention described above and shown in the accompanying drawings are merely examples and do not limit the present invention. The purpose of the present invention has been fully and effectively achieved. The functions and structural principles of the present invention have been shown and explained in the embodiments. Without departing from the principles described, the implementation of the present invention may have any changes or modifications.
Claims
1. A rare earth element extraction and separation analysis method based on ionic liquid, characterized in that, The method includes: The sample to be tested is an ionic liquid containing multiple rare earth complexes after an extraction reaction. The process involves acquiring spectral information of the sample to be tested, generating an elemental distribution map of the sample, and analyzing the morphological distribution of the sample based on the elemental distribution map; including: Construct a hybrid neural network model to separate spectral line interference in spectral information; the construction of the hybrid neural network model includes: Constructing a fundamental physical model; introducing neural networks to correct the shortcomings of the fundamental physical model; including: establishing the relationship between spectral line intensity and plasma parameters and elemental concentration; that is, for a spectral line intensity consisting of energy levels arrive Atomic or ionic spectral lines generated by transitions spectral line intensity ; in, Indicates the efficiency factor; Elements representing emission spectral lines The concentration; This represents the Einstein spontaneous emission coefficient for the transition; Indicates the upper energy level Statistical weights; Represents element The allocation function, Indicates the local temperature of the plasma; Indicates the upper energy level The excitation energy; Represents the Boltzmann constant; Discrete Converted into a continuous spectral profile ; Indicates at wavelength Theoretical synthesized spectral intensity at the location; Contour function function; Indicates the first The center wavelength of the spectral lines, express The Gaussian width parameter of the function; express The Lorentz width parameter of the function; Establish the relationship between spectral line intensity and plasma parameters and elemental concentration; that is, for a spectral line intensity consisting of energy levels... arrive Atomic or ionic spectral lines generated by transitions spectral line intensity ; in, Indicates the efficiency factor; Elements representing emission spectral lines The concentration; This represents the Einstein spontaneous emission coefficient for the transition; Indicates the upper energy level Statistical weights; Represents element The allocation function, Indicates the local temperature of the plasma; Indicates the upper energy level The excitation energy; Represents the Boltzmann constant; Optimize the hybrid neural network model to identify weak interference in spectral line interference, including: Further decomposition, namely: ;in, Indicates spectral lines The ideal spectral profile, Indicates spectral lines The self-absorption attenuation factor, , Indicates the absorption coefficient; Indicates the outline of the absorption spectral lines; Indicates spectral lines Continuous background radiation; A structured residual neural network is introduced, that is, a multi-head residual network is introduced, where each head learns the corresponding type of physical model error; Will The input is fed into a trained structured residual neural network, and the output is... , and ; Then, the new predicted spectrum ;in, This represents the self-absorption residual; Indicates background residual; This indicates unmodeled residuals; Further distinguish between weak interference and unknown interference in spectral line interference; Extract the spectral line intensity values from multiple measurement points from the interference-free spectral lines to obtain an accurate elemental distribution map.
2. The method for rare earth element extraction, separation, and analysis based on ionic liquid according to claim 1, characterized in that, The introduction of neural networks to correct the deficiencies of the fundamental physical model includes: An input layer is established to combine the concentration data of all elements contained in the sample to be detected into an element concentration vector. ; A calibration model is constructed based on a fully connected neural network. by The input is the residual spectrum, which is fed into the trained calibration model and outputs the residual spectrum. ; A fusion layer is constructed to add the output of the basic physics model and the output of the fully connected neural network model to obtain the predicted spectrum. .
3. The method for rare earth element extraction, separation, and analysis based on ionic liquid according to claim 2, characterized in that, The further distinction between weak interference and unknown interference in spectral line interference includes: Construct a multi-task learning model, which includes a shared encoder and a multi-task decoder; Will The input is fed into a pre-trained multi-task learning model, which outputs predicted values for plasma electron temperature and electron density. and ; From the original spectral lines The calculated values of plasma electron temperature and electron density were independently obtained using the Boltzmann diagram method. and ; Obtain spectral prediction error ;in, Indicates spectral lines Measured spectral line profile; Prediction error of plasma characteristic parameters From each sample to be tested Extract morphological features to construct the corresponding residual sample feature vector; Unsupervised clustering is performed on the feature vectors of the residual samples; For each cluster obtained, the corresponding average residual spectrum is calculated, and the interference type is labeled based on the average residual spectrum.
4. The method for rare earth element extraction, separation and analysis based on ionic liquid according to claim 3, characterized in that, The unsupervised clustering of the residual sample feature vectors includes: For a Calculate global statistical features, including: Calculate skewness ;in, Indicate the number of samples; calculate kurtosis. ;in, This represents the mean of the error. The standard deviation of the error is represented; shape and scale characteristics are calculated, including: calculate width of autocorrelation function ;in, Represents the set of wavelengths; normalized energy ratio ;in, Indicates the set of spectral lines of interest; if If the spectral interference is concentrated near the analytical line, it is either self-absorption interference or overlap interference. If so, the spectral interference is determined to be either background interference or uniformly distributed interference; Obtain the feature vector corresponding to each sample spectral line. ; Cluster analysis of the feature vectors of all samples to be tested was performed using the Gaussian mixture algorithm, including: choose The sample to be tested; Perform Gaussian mixture clustering: Calculate probability ;in, Indicates the first The mean and covariance of each Gaussian component; Indicates the number of samples to be tested in the cluster; Indicates mixed weights; Each sample to be tested is assigned to the cluster with the highest probability. Inside; For each obtained cluster, the corresponding average residual spectrum is calculated, and interference type labeling is performed based on the average residual spectrum, including: For each cluster, calculate the average residual spectrum of all samples to be tested within that cluster. ; Output Manual labeling is then performed.
5. The method for rare earth element extraction, separation, and analysis based on ionic liquid according to claim 4, characterized in that, Also includes: Define the physical tolerance region: Get ; Apply nonnegativity constraints: that is, set the following constraints: ； Constructing a physical constraint loss function Among them, data fitting loss ; Indicates physical constraint loss; Indicates the loss of adversarial regularization; Indicates hyperparameters; ； ;in, ; ; ; ; Represents the gradient of the background spectrum; Import a pre-trained distribution discriminator, which is built based on a neural network; Will The input is fed into the distribution decision unit, which outputs the concentration sample distribution vector. ; Define adversarial regularization or loss ;in, This indicates that the distribution discriminator reacts to the input. Probability estimate of belonging to the training set; Represents the concentration distribution vector of the training set; if The larger, the more it means The further outward the region; This represents a measurement function; By fixing the distribution decision parameters, updating the hybrid neural network model parameters, we obtain the hybrid neural network model parameters that minimize physical constraints.
6. A rare earth element extraction, separation, and analysis system based on ionic liquids, characterized in that, The system is used to perform the rare earth element extraction and separation analysis method based on ionic liquid as described in any one of claims 1-5.
7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that is executed by a processor to implement the rare earth element extraction and separation analysis method based on ionic liquid as described in any one of claims 1-5.
Citation Information
Patent Citations
Soil component detection method, equipment and system
CN114813709A