A trace multi-ion spectral detection method for zinc liquid based on reversible feature extraction and its application
By constructing NICE and PLS models, the information lossless spectral signal separation and regression analysis in small samples is realized, which solves the problems of interference and overlap of polymetallic ion spectral signals in traditional methods, and improves the traces in the wet zinc smelting process Metal ion concentration measurement accuracy and real-time performance.
Patent Information
- Application Number
- CN202210255434.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-03-15
- Publication Date
- 2024-11-22
- Estimated Expiration
- 2042-03-15
AI Technical Summary
During the wet zinc smelting process, traditional multivariate analysis methods are difficult to effectively separate and regress the spectral signals of multiple trace metal ions in high-concentration zinc liquid in small samples, especially due to the absorption spectrum of zinc ions and other metal ions. Mutual interference and overlap will lead to large errors and loss of information, making it difficult to meet industrial needs.
Using a method based on reversible feature extraction, non-linear independent component estimation (NICE) model and partial least squares regression (PLS) model are constructed to realize lossless spectral signal separation and regression analysis of information. The specific steps include collecting the absorption spectral data of zinc liquid, building a NICE model for feature extraction, converting it into independent potential variables, and establishing a PLS regression model to predict multi-ion concentration.
Reduce detection costs, improve detection real-time performance, solve the problems of overlap and interference of multimetal ion spectral signals, realize more accurate trace metal ion concentration measurement, improve model accuracy, and meet the needs of wet zinc smelting process .
Smart Images

Figure CN114692736B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of spectral detection, and more specifically, relates to a zinc liquid trace multi-ion spectral detection method based on reversible feature extraction and an application thereof. Background Art
[0002] In the industrial process control of hydrometallurgical zinc smelting, accurate measurement of trace metal ion concentrations such as Cu, Co, Ni and Fe in high-concentration zinc liquid is the key to ensuring production stability, improving product quality and reducing production costs.
[0003] In the prior art, ultraviolet visible spectrophotometry (UV-Vis) has the advantages of simple operation, high accuracy and good reproducibility, and is one of the commonly used methods for detecting ion content. However, traditional UV-Vis detection is mainly aimed at single ions. When applied to multi-metal ions, it is necessary to pre-separate the component ions by chemical methods. However, the pre-separation steps are cumbersome, the real-time performance is poor and the cost is high.
[0004] The simultaneous determination of the concentrations of multiple metal ions using the ultraviolet absorption spectrum of a mixed solution has important economic and technological significance.
[0005] Ultraviolet absorption spectra are easy to measure; however, the ion concentration of the samples used for modeling needs to be analyzed and tested, which is quite costly and cumbersome, so there will not be too many samples for modeling. Therefore, when using ultraviolet absorption spectra for full-band modeling, the number of samples is usually much smaller than the variable dimension, which also increases the difficulty of modeling.
[0006] With the development of chemometrics, multivariate analysis methods such as partial least squares (PLS) and principal component analysis (PCR) are used to model the full-band data of the UV spectrum. These methods allow modeling when the number of samples is much smaller than the dimension of the variable, and can achieve simultaneous detection of multiple metal ions in complex mixed solutions.
[0007] In zinc solution, the solution matrix zinc ions Zn 2+ The ratio of the mass concentration of the trace ion to be measured can reach up to hundreds of thousands, so the spectral signal of the trace ion to be measured will be overshadowed by the high concentration of Zn 2+ and impurity ion signals are seriously masked; moreover, some ions to be measured (such as Cu 2+ and Ni 2+ ) are too similar in properties, and the absorption spectra almost completely overlap, which makes the spectral signal seriously nonlinear, making it difficult to perform regression prediction of ion concentration. Nonlinearity also makes it difficult to separate the spectral signals of ions with similar properties. Coupled with the interference of noise, traditional multivariate analysis methods often lose effective information when separating these ion spectral signals, further reducing the accuracy of the model.
[0008] Therefore, when the absorption spectra of various metal ions interfere with each other and seriously overlap, the traditional multivariate analysis method has large errors and is difficult to meet the process requirements. In summary, it is necessary to design a multi-ion spectral detection method that can losslessly separate the spectral information of each metal ion and perform regression analysis in the case of small samples.
[0009] In view of this, the present invention is proposed. Summary of the invention
[0010] In order to address the problem that when detecting multiple metal ions with high concentration ratios simultaneously, the absorption spectra of each metal ion interfere with each other and seriously overlap, making it difficult to analyze and predict low-concentration ions, the present invention provides a multi-ion spectral detection method that can losslessly separate the spectral information of each metal ion and perform regression analysis in the case of small samples.
[0011] To achieve the above object, the technical solution of the present invention is as follows:
[0012] A method for detecting trace multi-ion spectroscopy in zinc liquid based on reversible feature extraction, comprising:
[0013] S1, collecting a sufficient amount of zinc liquid to be tested at the industrial site at different times, detecting the absorption spectrum of the zinc liquid to be tested under the measurement system, and at the same time, extracting a small part of the sample from the zinc liquid to be tested for chemical analysis to obtain the concentration value of each ion;
[0014] S2, constructing a reversible nonlinear independent component estimation NICE model, training the NICE model with the absorbance data of the absorption spectrum measured in step S1, and extracting independent potential features by fitting the distribution of absorbance by the model;
[0015] S3. Based on the NICE model, reversible and non-destructive feature extraction is performed on the absorbance data of the sample with ion concentration values, and the features are converted into independent latent variables. The obtained independent latent variables and the corresponding ion concentration values are used to construct a PLS regression model for predicting multi-ion concentrations;
[0016] S4. Based on the NICE model and the PLS regression model, the concentration of each metal ion in the zinc solution to be tested is predicted using the absorption spectrum of the zinc solution to be tested, thereby achieving rapid measurement of the concentration of multiple trace metal ions in the zinc solution.
[0017] In the above technical solution, in step S1, in the process of detecting the absorption spectrum of the zinc liquid to be tested under the measurement system, it includes:
[0018] S11, in the measurement system, acetic acid-sodium acetate with pH=5.5, 0.40wt% nitroso R salt and 0.01mol / L cetyltrimethylammonium bromide are respectively selected as a buffer, a color developer and a sensitizer and mixed with the zinc solution to be tested to prepare a sample solution, and the zinc solution to be tested is used as a blank reference;
[0019] S12, placing the prepared sample solution in a spectrometer, and measuring the absorbance of each point at an interval of 1 nm within a wavelength range of 400-800 nm;
[0020] S13. A small portion of the sample is analyzed using a chemical method to obtain the concentration value of each ion and mark the corresponding absorbance data.
[0021] In a specific embodiment of the present invention, in step S11,
[0022] 5.0 ml of acetic acid-sodium acetate buffer with pH=5.5, 2.5 ml of 0.40 wt% nitroso R salt color developer, 3 ml of 0.01 mol / L hexadecyltrimethylammonium bromide sensitizer and 5 ml of the zinc solution to be tested were mixed and then fixed to 25 ml to prepare a sample solution.
[0023] In the above technical solution, in step S2, the process of constructing the NICE model and training it includes:
[0024] S21. Construct NICE model to achieve reversible feature extraction;
[0025] S22, training the NICE model using the absorption spectrum measured in step S1.
[0026] In detail, in step S21, the principle is as follows:
[0027] We hope to use feature extraction to solve the problem of superposition of original absorption spectrum signals, so we hope that the extracted latent variables are independent; we hope that this process is information lossless, so we design the transformation process to be reversible. The proof of reversibility is as follows:
[0028] If the transformation function f is reversible, that is, there exists an inverse function g, we have
[0029]
[0030] Then, from the perspective of mutual information, we have
[0031] I(x;z)=I(x;f(x))=I(x;x) (2)
[0032] That is to say, as long as the transformation function f is reversible, the transformation process is information lossless.
[0033] The extracted latent variables are independent as follows:
[0034] Assuming that the original variable x is D-dimensional and the distribution of the latent variable q(z) is an independent standard Gaussian distribution, then for the original distribution q(x) we have
[0035] q(x)=∫q(z)g θ (x|z)dz (3)
[0036] The maximum likelihood method is generally used to solve the parameter θ, so it is necessary to maximize the objective
[0037]
[0038] Since q(z) is an independent standard Gaussian distribution, we have
[0039]
[0040] Among them, D is the dimension of the latent variable z. Since this process is reversible, mathematically the dimension of the latent variable z is the same as the original variable x.
[0041] Assuming that the transformation from the original variable to the latent variable is f, combining equation (1) and equation (3), we need to maximize the objective
[0042]
[0043] This optimization objective is solvable, and a NICE model is built for this objective to fit the transformation f from x to z.
[0044] The NICE model constructed consists of multiple additive coupling layers, and it is known that it is reversible, so the data is lossless during the change process; and in this way, the Jacobian matrix of each additive coupling layer is
[0045]
[0046] This ensures that the determinant of each additive coupling layer is is 1.
[0047] The NICE model consists of multiple additively coupled layers; by the chain rule, the determinant of the combination of multiple additively coupled layers is still 1.
[0048] In addition, the NICE model also designs a scale transformation layer at the end to compress the manifold, so
[0049]
[0050] Therefore, the Jacobian matrix of the scale transformation layer is
[0051]
[0052] Then the determinant of the scale transformation layer is Π i s i , combined with formula (6), under the NICE model
[0053]
[0054] Construct a network model and set the loss function to (10), that is, complete the construction of the NICE model. It can be found that the feature extraction process is information-free and the generated latent variables are independent of each other.
[0055] In a specific embodiment of the present invention, in step S22, the NICE model is trained using the absorption spectrum measured in step S1, including:
[0056] First, n additive coupling layers are constructed, the activation function is the linear rectification function ReLU, the optimizer selects the Adam algorithm, and the maximum number of iterations is m times; the NICE model belongs to unsupervised learning, does not require ion concentration label values, and directly uses the absorption spectrum measured in S1 as the input of the network model to train the NICE model; after the NICE model training is completed, the transformation f from the absorption spectrum x to the latent variable z is determined.
[0057] In the above technical solution, in step S3, based on the NICE model, reversible and non-destructive feature extraction is performed on the absorbance data of the sample with ion concentration values, and converted into independent latent variables, and the obtained independent latent variables and the corresponding ion concentration values are used to construct a PLS regression model for predicting multi-ion concentrations, including:
[0058] S31, based on the trained NICE model, the labeled sample absorbance X L Transformed into latent variable Z L ;
[0059] S32, using the new variable Z L The corresponding ion concentration Y L , establish a PLS regression model.
[0060] The solution process is as follows:
[0061] The independent variable Z L With dependent variable Y L Perform principal component decomposition, and we have
[0062]
[0063] Where P and V are Z L With Y L The loading matrix, i.e., the eigenvector matrix, T and U are ZL With Y L The projection onto the loading matrix, i.e. the principal components.
[0064] The algorithm is based on Z L With Y L Assuming linear correlation, if we want T to maximize the explanatory power of U, we can see from the canonical correlation analysis that T and U satisfy
[0065] U=TB (12)
[0066] Combining (8), we have
[0067] Z L P T =TPP T =T (13)
[0068] According to equations (11), (12), and (13), we have
[0069] Y L =UV=Z L P T BV (14)
[0070] By solving the values of P, B and V and selecting the appropriate number of components, the latent variable Z can be obtained. L and ion concentration Y L The regression model is established based on the relationship between
[0071] In a specific embodiment of the present invention, in step S32, selecting a suitable number of components for PLS specifically includes:
[0072] S321, using the K-fold cross validation method for the data used to train the model, and using the mean square prediction error of the validation set as an indicator for evaluating the quality of the PLS regression model;
[0073] S322. Design a PLS regression model with a minimum number of components of 2 and a maximum number of components of 80. Gradually increase the number of components, calculate the prediction mean square error of the validation set, and select the number of components with the minimum error as the model parameter.
[0074] In the above technical solution, in step S4, based on the NICE model and the PLS regression model, the absorption spectrum of the zinc solution to be tested is used to predict the concentration of each metal ion in the zinc solution to be tested, thereby realizing rapid measurement of the concentration of multiple trace metal ions in the zinc solution, including:
[0075] S41, taking the zinc solution to be tested, detecting its absorption spectrum under the measurement system at a wavelength of 400-800nm, and obtaining absorbance data x;
[0076] S42, based on the trained NICE model, the absorbance data x is used as the model input, and the independent latent variable z is obtained by feature extraction;
[0077] S43. Based on the trained PLS model, the latent variable z is used as input, the regression relationship determined by the model is used to determine the concentration y of each metal ion, and the concentration of each metal ion is output.
[0078] In a specific embodiment of the present invention, in step S4, the plurality of trace metal ions include at least Cu, Co, Ni and Fe.
[0079] On the other hand, the present invention also provides the application of the above-mentioned zinc liquid trace multi-ion spectrum detection method based on reversible feature extraction in the detection of trace metal ion concentration in high-concentration zinc liquid.
[0080] Compared with the prior art, the present invention has the following advantages:
[0081] (1) Compared with the method of pre-separating each component ion by chemical method and then detecting the concentration of each ion one by one, the trace multi-ion spectral detection method of zinc liquid based on reversible feature extraction provided by the present invention greatly reduces the detection cost and improves the real-time detection, thus providing a guarantee for the hydrometallurgical zinc smelting process;
[0082] (2) Compared with the method of simultaneously detecting multiple metal ions using traditional multivariate analysis methods, the zinc liquid trace multi-ion spectral detection method based on reversible feature extraction provided by the present invention first decomposes the signal into independent features, thereby solving the problem of serious overlap and mutual interference of the absorption spectra between the ions to be measured. The reversible and non-destructive feature extraction process also ensures that the effective information for distinguishing ions with similar properties will not be lost, thus achieving better detection results. BRIEF DESCRIPTION OF THE DRAWINGS
[0083] Figure 1 A schematic flow chart of a method for detecting trace multi-ion spectroscopy in zinc liquid based on reversible feature extraction provided in an embodiment of the present invention;
[0084] Figure 2 It is an additive coupling layer of the NICE model of the zinc liquid trace multi-ion spectroscopy detection method based on reversible feature extraction provided in an embodiment of the present invention;
[0085] Figure 3 The network structure of the NICE model of the zinc liquid trace multi-ion spectroscopy detection method based on reversible feature extraction provided in the embodiment of the present invention;
[0086] Figure 4 A graph showing the variation of the mean square error of each ion prediction with the number of components in the zinc liquid trace multi-ion spectral detection method based on reversible feature extraction provided in an embodiment of the present invention;
[0087] In the figure:
[0088] x is the input signal, z is the latent variable, and m is the activation function. DETAILED DESCRIPTION
[0089] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with embodiments.
[0090] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.
[0091] In the examples, unless otherwise specified, all the means used are conventional means in the art.
[0092] As used herein, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion. For example, a composition, step, method, article, or apparatus that comprises the listed elements is not necessarily limited to only those elements but may include other elements not expressly listed or inherent to such composition, step, method, article, or apparatus.
[0093] In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.
[0094] A method for detecting trace multi-ion spectroscopy in zinc solution based on reversible feature extraction, such as Figure 1 As shown, including:
[0095] S1, collecting a sufficient amount of zinc liquid to be tested at the industrial site at different times, detecting the absorption spectrum of the zinc liquid to be tested under the measurement system, and at the same time, extracting a small part of the sample from the zinc liquid to be tested for chemical analysis to obtain the concentration value of each ion;
[0096] S2, constructing a reversible nonlinear independent component estimation NICE model, training the NICE model with the absorbance data of the absorption spectrum measured in step S1, and extracting independent potential features by fitting the distribution of absorbance by the model;
[0097] S3. Based on the NICE model, reversible and non-destructive feature extraction is performed on the absorbance data of the sample with ion concentration values, and the features are converted into independent latent variables. The obtained independent latent variables and the corresponding ion concentration values are used to construct a PLS regression model for predicting multi-ion concentrations;
[0098] S4. Based on the NICE model and the PLS regression model, the concentration of each metal ion in the zinc solution to be tested is predicted using the absorption spectrum of the zinc solution to be tested, thereby realizing rapid measurement of the concentration of multiple trace metal (Zn, Co, Ni, Fe) ions in the zinc solution.
[0099] Taking 300 groups of zinc liquid samples collected on site as an example, the absorption spectra of these zinc liquids were measured under the measurement system, and the ion concentration values of 50 groups of samples were chemically analyzed; by constructing a reversible NICE network model, the absorbance data was extracted without information loss; then, a regression PLS regression model was established based on the extracted new potential features and the corresponding ion concentrations; finally, based on the determined NICE and PLS models, rapid detection of trace multi-metal ions in zinc liquid was achieved.
[0100] The specific implementation plan is as follows:
[0101] 1. Measure the absorption spectrum of this batch of zinc liquid samples under the measurement system, and select 50 groups for testing to obtain the concentration values of each ion, including:
[0102] According to the properties of Cu, Co, Ni and Fe, the color developer is 0.40wt% nitroso R salt, the buffer is acetic acid-sodium acetate with pH=5.5, and the sensitizer is 0.01mol / L cetyltrimethylammonium bromide (CTMAB) in the measurement system; 5.0ml acetic acid-sodium acetate, 2.5ml nitroso R salt, 3ml CTMAB and 5ml test solution are mixed, and the mixed solution is fixed to 25ml to prepare a sample solution, and the test solution blank is used as a reference.
[0103] Then, for the prepared sample solution, the absorbance of each point was measured every 1 nm in the range of 400-800 nm in the spectrometer. Then, for 50 groups of samples, the concentration of each ion was analyzed by chemical methods, and the corresponding absorbance data were marked.
[0104] The samples were then divided using the Kennard-Stone (KS) algorithm, with 10 groups of labeled samples used as test sets and the rest used as training sets and validation sets for modeling.
[0105] 2. To achieve information lossless feature extraction, build and train a reversible NICE model
[0106] We hope to solve the problem of superposition of original absorption spectrum signals by feature extraction, so we hope that the extracted latent variables are independent. We hope that this process is information lossless, so we design the transformation process to be reversible. The proof of reversibility is as follows:
[0107] If the transformation function f is reversible, that is, there exists an inverse function g, we have
[0108]
[0109] Then from the perspective of mutual information, we have
[0110] I(x;z)=I(x;f(x))=I(x;x) (2)
[0111] That is to say, as long as the transformation function f is reversible, the transformation process is information lossless.
[0112] The extracted latent variables are independent as shown below:
[0113] Assuming that the original variable x is D-dimensional and the distribution of the latent variable q(z) is an independent standard Gaussian distribution, then for the original distribution q(x) we have
[0114] q(x)=∫q(z)q θ (x|z)dz (3)
[0115] The maximum likelihood method is generally used to solve the parameter θ, so it is necessary to maximize the objective
[0116]
[0117] Since q(z) is an independent standard Gaussian distribution, we have
[0118]
[0119] Among them, D is the dimension of the latent variable z. Since this process is reversible, mathematically the dimension of the latent variable z is the same as the original variable x.
[0120] Assuming that the transformation from the original variable to the latent variable is f, combining equation (3) and equation (5), we need to maximize the objective
[0121]
[0122] This optimization objective is solvable, and a NICE network model is built for this objective to fit the transformation f from x to z.
[0123] The NICE network model constructed consists of multiple additive coupling layers, each of which is Figure 2 shown by Figure 2 It can be seen that the network structure is reversible, so the data is lossless during the change process; and in this way, the Jacobian matrix of each additive coupling layer is
[0124]
[0125] This ensures that the determinant of each additive coupling layer is is 1.
[0126] The NICE model consists of multiple additively coupled layers, such as Figure 2 As shown in Figure 2, by the chain rule, the determinant of the combination of multiple additive coupling layers is still 1.
[0127] In addition, the NICE model also designs a scale transformation layer at the end to compress the manifold, so
[0128]
[0129] Therefore, the Jacobian matrix of the scale transformation layer is
[0130]
[0131] Then the determinant of the scale transformation layer is Π i s i , combined with formula (6), under the NICE model
[0132]
[0133] Build as Figure 3 The network model shown in the figure is set to the loss function as (10), that is, the NICE model is built. It can be found that the feature extraction process is information-free and the generated latent variables are independent of each other.
[0134] In the above experiment, four additive coupling layers were first constructed, the activation function was the linear rectifier function (ReLU), the optimizer selected the (adaptive momentum estimation) Adam algorithm, and the maximum number of iterations was 300. The NICE model belongs to unsupervised learning, so it does not require ion concentration label values. The measured absorption spectrum is directly used as the input of the network model to train the network model. After the NICE model training is completed, the transformation f from the absorption spectrum x to the latent variable z is determined.
[0135] 3. Based on the trained NICE model, the absorbance data of 40 groups of labeled samples with ion concentration values belonging to the training set are converted into independent latent variables, and a PLS regression model for predicting multi-ion concentration is constructed.
[0136] Based on the trained NICE model, the labeled sample absorbance X L Transformed into latent variable Z L , using the new variable Z L The corresponding ion concentration Y L , establish a PLS regression model.
[0137] The solution process is as follows:
[0138] The independent variable Z L With dependent variable Y L Perform principal component decomposition, and we have
[0139]
[0140] Where P and V are Z L With YL The loading matrix, i.e., the eigenvector matrix, T and U are Z L With Y L The projection on the loading matrix is the principal component. The algorithm is based on Z L With Y L Assuming linear correlation, if we want T to maximize the explanatory power of U, we can see from the canonical correlation analysis that T and U satisfy
[0141] U=TB (12)
[0142] Combining (8), we have
[0143] Z L P T =TPP T =T (13)
[0144] According to equations (11), (12), and (13), we have
[0145] Y L =UV=Z L P T BV (14)
[0146] By solving the values of P, B and V and selecting the appropriate number of components, the latent variable Z can be obtained. L and ion concentration Y L The regression model is established based on the relationship between
[0147] Furthermore, in the above steps, the appropriate number of components is selected for the PLS algorithm, specifically including:
[0148] The K (K = 5) fold cross validation method is used for the data used for training the model, and the mean square prediction error of the validation set is used as an indicator to evaluate the quality of the PLS model. The minimum number of components in the PLS model is designed to be 2, and the maximum is 80. The number of components is gradually increased, and the prediction mean square error of the validation set is calculated. The changes in the prediction mean square error of the four ions Cu, Co, Ni, and Fe are shown in the figure below. Figure 4 As shown, the number of components when the error of the four ions is minimum is 23, which is used as the model parameter.
[0149] 4. Based on the NICE model and PLS model, the absorption spectrum of the mixed solution is used to predict the concentration of each metal ion in the mixed solution.
[0150] Obtain the solution to be tested, make it into a sample solution under a given measurement system, measure the absorption spectrum of the sample solution at a wavelength of 400-800nm, and obtain the absorbance data x. Based on the trained NICE model, the absorbance data x is used as the model input, and the feature extraction obtains the independent latent variable z. Based on the trained PLS model, the latent variable z is used as the input, and the regression relationship determined by the model is used to determine the concentration y of each metal ion, and the concentration of each metal ion is output.
[0151] Analysis of experimental results: In order to verify the effectiveness of the method proposed in this patent, the PLS method, MC-UVEPLS and CARS PLS, which are commonly used algorithms for simultaneous analysis of multiple components in ultraviolet spectra, were compared and verified. The symmetric absolute percentage error SMAPE, the root mean square of the prediction error RMSEP and the determination coefficient R2 were used as evaluation indicators of the model. The results are shown in Table 1 below.
[0152] Table 1 Comparison of results of different methods
[0153]
[0154]
[0155] As can be seen from Table 1, the PLS model is most seriously affected by ion overlap interference and noise. The prediction of each ion is poor in both RMESP and SMAPE. 2 It also shows that the interpretability of the model is poor. CARS PLS and MC-UVEPLS, an algorithm based on wavelength selection improvement, can be seen that the prediction effect on ions with similar properties such as Cu and Ni is poor, especially in the prediction of Cu ions, because the wavelength selection process loses a lot of information, and the prediction effect is even worse than PLS. The regression method (NICE-PLS) based on reversible feature extraction proposed by the present invention not only solves the problem that the absorption spectra of each metal ion interfere with each other and are difficult to separate due to severe overlap, but also ensures that effective information will not be lost without loss of information, so it is better than other algorithms in the prediction of Cu, Co, Ni, and Fe ions, especially in the prediction of ions with similar properties, compared with other methods, the effect is significantly improved. Rapid detection of trace polyions in zinc liquid is thus achieved, providing a guarantee for hydrometallurgical copper smelting process.
[0156] The above-mentioned embodiments only express several implementation modes of the present invention, and the description thereof is relatively specific and detailed, but it should not be understood as limiting the scope of the invention patent.
[0157] It should be noted that, for those skilled in the art, several modifications and improvements can be made without departing from the concept of the present invention, and these modifications and improvements all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.
Claims
1. A method for detecting trace multi-ion spectroscopy in zinc liquid based on reversible feature extraction, characterized in that: include: S1, collecting a sufficient amount of zinc liquid to be tested at the industrial site at different times, detecting the absorption spectrum of the zinc liquid to be tested under the measurement system, and at the same time, extracting a small part of the sample from the zinc liquid to be tested for chemical analysis to obtain the concentration value of each ion; S2, constructing a reversible nonlinear independent component estimation NICE model, training the NICE model with the absorbance data of the absorption spectrum measured in step S1, and extracting independent potential features by fitting the distribution of absorbance by the model; S3. Based on the NICE model, reversible and non-destructive feature extraction is performed on the absorbance data of the sample with ion concentration values, and the features are converted into independent latent variables. The obtained independent latent variables and the corresponding ion concentration values are used to construct a PLS regression model for predicting multi-ion concentrations; S4, based on the NICE model and the PLS regression model, using the absorption spectrum of the zinc solution to be tested, predicting the concentration of each metal ion in the zinc solution to be tested, thereby realizing rapid measurement of the concentration of multiple trace metal ions in the zinc solution; In step S2, the process of constructing and training the NICE model includes: S21. Construct NICE model to achieve reversible feature extraction; S22, training the NICE model using the absorption spectrum measured in step S1; In step S22, the NICE model is trained using the absorption spectrum measured in step S1, including: First, n additive coupling layers are constructed, the activation function is the linear rectification function ReLU, the optimizer selects the Adam algorithm, and the maximum number of iterations is m; the NICE model belongs to unsupervised learning, does not require ion concentration label values, and directly uses the absorption spectrum measured in S1 as the input of the network model to train the NICE model; after the NICE model training is completed, the transformation f from the absorption spectrum x to the latent variable z is determined; In step S3, it includes: S31, based on the trained NICE model, the labeled sample absorbance X L Transformed into latent variable Z L ; S32, using the new variable Z L The corresponding ion concentration Y L , establish a PLS regression model.
2. The zinc liquid trace multi-ion spectrum detection method based on reversible feature extraction according to claim 1 is characterized in that: In step S1, the process of detecting the absorption spectrum of the zinc liquid to be tested under the measurement system includes: S11, in the measurement system, acetic acid-sodium acetate with pH=5.5, 0.40 wt% of nitroso R salt and 0.01 mol / L of cetyltrimethylammonium bromide are respectively selected as a buffer, a color developer and a sensitizer and mixed with the zinc solution to be tested to prepare a sample solution, and the zinc solution to be tested is used as a blank reference; S12, placing the prepared sample solution in a spectrometer, and measuring the absorbance of each point at an interval of 1 nm within a wavelength range of 400-800 nm; S13. A small portion of the sample is analyzed using a chemical method to obtain the concentration value of each ion and mark the corresponding absorbance data.
3. The zinc liquid trace multi-ion spectrum detection method based on reversible feature extraction according to claim 2 is characterized in that: In step S11, The sample solution was prepared by mixing 5.0 ml of acetic acid-sodium acetate buffer with a pH of 5.5, 2.5 ml of 0.40 wt% nitroso R salt color developer, 3 ml of 0.01 mol / L hexadecyltrimethylammonium bromide sensitizer and 5 ml of the zinc solution to be tested, and then diluting the volume to 25 ml.
4. The method for detecting trace multi-ion spectroscopy in zinc liquid based on reversible feature extraction according to claim 1, characterized in that: In step S32, the appropriate number of components is selected for PLS, specifically including: S321, using the K-fold cross validation method for the data used to train the model, and using the mean square prediction error of the validation set as an indicator for evaluating the quality of the PLS regression model; S322. Design a PLS regression model with a minimum number of components of 2 and a maximum number of components of 80. Gradually increase the number of components, calculate the prediction mean square error of the validation set, and select the number of components with the minimum error as the model parameter.
5. The zinc liquid trace multi-ion spectrum detection method based on reversible feature extraction according to claim 1 is characterized in that: In step S4, it includes: S41, taking the zinc solution to be tested, detecting its absorption spectrum under the measurement system at a wavelength of 400-800 nm, and obtaining absorbance data x; S42, based on the trained NICE model, the absorbance data x is used as the model input, and the independent latent variable z is obtained by feature extraction; S43. Based on the trained PLS model, the latent variable z is used as input, the regression relationship determined by the model is used to determine the concentration y of each metal ion, and the concentration of each metal ion is output.
6. The method for detecting trace multi-ion spectroscopy in zinc liquid based on reversible feature extraction according to claim 1, characterized in that: In step S4, the plurality of trace metal ions include at least Cu, Co, Ni and Fe.
7. Application of the zinc liquid trace multi-ion spectroscopy detection method based on reversible feature extraction as described in any one of claims 1 to 6 in the detection of trace metal ion concentration in high-concentration zinc liquid.
Citation Information
Patent Citations
Method for predicting trace metal ion concentration in zinc solution based on subsection building model
CN109187392A
Method, device, and program for nondestructive measurement of nitrate ion concentration
JP2019101040A
Method and apparatus for real time monitoring of industrial electrolytes
US20050183958A1
