A method, device and storage medium for simultaneous determination of multiple components of ammonia escape

Abstract

The invention relates to the field of gas measurement, and discloses a method, a device and a storage medium for simultaneous determination of multiple components of ammonia escape. The method comprises the following steps: performing data preprocessing on the original absorbance spectrum of a modeling sample; constructing a multivariate correction basic concentration prediction model, and training the model using the preprocessed data; obtaining the preliminary predicted concentration of each component of the sample to be tested using the model; obtaining a reference absorbance spectrum by inverting the preliminary predicted concentration and the basic absorbance spectrum of the target gas; constructing a wavelength drift adaptive correction model, inputting the original absorbance spectrum into the wavelength drift adaptive correction model, and obtaining a corrected absorbance spectrum; adjusting the wavelength drift adaptive correction model parameters; obtaining a corrected spectrum; and inputting the corrected spectrum into the trained model to obtain a multi-component concentration output. The invention has the beneficial effects of eliminating cross-interference and wavelength drift influence in the concentration prediction process and improving the concentration prediction accuracy.

Description

Technical Field

[0001] The present invention relates to the field of gas measurement, and in particular to a method, a device and a storage medium for simultaneously measuring multiple components of ammonia escape. Background Art

[0002] At present, the mainstream ammonia escape monitoring methods include the national standard method "HJ533-2009 Determination of Ammonia in Ambient Air and Exhaust Gas by Nessler Reagent Spectrophotometry", tunable semiconductor laser absorption spectroscopy (TDLAS), catalytic conversion, chemiluminescence, Fourier transform infrared, chemical colorimetry, FFT-based ultraviolet spectroscopy, etc. The following are some of the methods:

[0003] National standard HJ533-2009, using dilute sulfuric acid solution to absorb NH in flue gas 3 Ammonium ions are formed, and then a yellow-brown complex is formed with Nessler's reagent. The substance has a maximum absorption peak at 420nm, and a measurement method is constructed using spectrophotometry. This method is fast and convenient to determine ammonia escape, but its accuracy is greatly affected by temperature and color development time, and a complete waste liquid recovery and treatment mechanism needs to be established to avoid environmental pollution.

[0004] The TDLAS method uses the monochromaticity of laser to measure the ammonia in the flue gas components by the absorption characteristics of specific gases. It has the advantages of wide measurement range, non-contact measurement, fast response speed and simple installation. However, its measurement results are greatly affected by the on-site flue gas conditions, installation location and other conditions, and it is difficult to calibrate online and has high maintenance costs.

[0005] The Fourier infrared method completes the measurement based on the absorption of light energy by molecular vibration and rotational energy level transitions. Different chemical bonds or functional groups absorb different energies when undergoing energy level transitions, resulting in characteristic absorption peaks at different wavelengths. The absorption spectrum signal is processed using Fourier transform to ultimately obtain an infrared absorption spectrum of transmittance or absorbance versus wave number or wavelength, and finally establish an inversion relationship with concentration. This method has high measurement accuracy, but the equipment is complex and expensive, and is mostly used for laboratory analysis and testing.

[0006] Based on the FFT ultraviolet spectroscopy, a quantitative prediction method for escaped ammonia is established by processing the characteristic absorption spectrum of substances at different ultraviolet wavelengths through Fourier transform. This method has fast analysis speed, simple sampling pretreatment, less sample damage, non-contact, and is suitable for real-time online analysis and on-site rapid inspection. In addition, in the ultraviolet band, water, CO, CO 2 , O 2 , CH 4 The interference of NH 3 、SO 2 、NO xThe characteristic absorbance spectra in the ultraviolet band overlap seriously and there is cross-interference. It is difficult to find the independent pure component absorption regions of the above gases from the spectrum after FFT transformation (there is only characteristic absorption of one gas in this region). It may be difficult to completely eliminate this influence by relying solely on FFT. In addition, when the wavelength drift occurs in the spectrum, it will also have a serious impact on the accuracy of the prediction results. Summary of the invention

[0007] The purpose of the present invention is to propose a method, device and storage medium for simultaneous determination of multiple components of ammonia escape, so as to solve the technical problem that the prior art cannot simultaneously and quickly determine the concentrations of multiple components during ammonia escape monitoring.

[0008] Specifically, the present invention comprehensively considers the requirements of real-time, online, on-site application, cost, measurement accuracy, convenience, and pollution-free, and provides a method for simultaneously determining multiple components of ammonia escape based on ultraviolet differential absorption spectroscopy, comprising the following steps:

[0009] S1, performing data preprocessing on the original absorbance spectrum of the modeling sample to obtain preprocessed data;

[0010] S2. construct a multivariate correction basic concentration prediction model, and train the model using the preprocessed data to obtain a trained model;

[0011] S3, inputting the original absorbance spectrum of the sample to be tested into the trained model to obtain the preliminary predicted concentration of each component of the sample to be tested;

[0012] S4, obtaining a reference absorbance spectrum by inverting the initial predicted concentration and the basic absorbance spectrum of the target gas;

[0013] S5, constructing a wavelength drift adaptive correction model, inputting the original absorbance spectrum of the sample to be tested into the wavelength drift adaptive correction model, and obtaining a corrected absorbance spectrum;

[0014] S6, comparing the reference absorbance spectrum and the corrected absorbance spectrum, adjusting the wavelength drift adaptive correction model parameters, and obtaining a trained wavelength drift adaptive correction model;

[0015] S7, inputting the original absorbance spectrum of the sample to be tested into the trained wavelength drift adaptive correction model to obtain a corrected spectrum;

[0016] S8. Input the corrected spectrum into the trained model to obtain multi-component concentration output.

[0017] A storage medium stores instructions and data for being executed by a processor to implement a method for simultaneously determining multiple components of ammonia escape.

[0018] A device for simultaneously measuring multiple components of ammonia escape comprises: a processor and a storage medium; the processor loads and executes instructions and data in the storage medium to implement a method for simultaneously measuring multiple components of ammonia escape.

[0019] The beneficial effects provided by the present invention are: providing a method for real-time, online rapid detection of ammonia escape, based on the same ultraviolet analyzer can simultaneously detect NH 3 、SO 2 By selecting the ultraviolet spectrum band spectrometer, water, CO, CO 2 , O 2 and CH 4 The interference effects of various gas characteristics and selective multivariate correction models were used to reduce the influence of cross-interference and enhance the anti-interference ability of the method. The robustness of the method was enhanced by designing a two-level prediction process based on wavelength drift adaptive correction. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 It is a schematic flow chart of the method of the present invention;

[0021] Figure 2 It is a schematic diagram of the feature variable selection results;

[0022] Figure 3 It is a schematic diagram of the process of modeling the basic concentration prediction model;

[0023] Figure 4 It is a wavelength drift adaptive correction model prediction flow chart;

[0024] Figure 5 It is a working schematic diagram of the hardware device of the present invention. DETAILED DESCRIPTION

[0025] In order to make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention will be further described below with reference to the accompanying drawings.

[0026] Before formally explaining the present invention, relevant technical terms in the present invention are explained. In addition, the scheme of the present invention is also generally explained first for easy understanding.

[0027] Please refer to Figure 1 , Figure 1 It is a schematic diagram of the process of the present invention.

[0028] The present invention provides a method for simultaneously determining multiple components of ammonia escape, comprising the following steps:

[0029] S1, performing data preprocessing on the original absorbance spectrum of the modeling sample to obtain preprocessed data;

[0030] It should be noted that the preprocessing in step S1 includes: truncation, differentiation, feature variable selection and smoothing.

[0031] The truncation process is actually to take the union of the characteristic absorption band intervals of each component, that is, to ensure that all components have relatively obvious absorption within the truncation interval.

[0032] The process of differentiation is actually to eliminate the influence of "slow-changing" signals caused by various light scattering, thereby improving the robustness of the component absorption signal.

[0033] Characteristic variable selection is actually to select characteristic variables that are conducive to building a robust prediction model for each component.

[0034] The purpose of smoothing is to eliminate the influence of noise signals in spectral signals.

[0035] It should be noted that characteristic variable selection is one of the key steps to extract the specific characteristics of each target gas and eliminate cross-interference.

[0036] In the present invention, the feature variable selection is divided into two steps, wherein the first step is the elimination of abnormal samples, and the second step is the selection of feature variables.

[0037] Specifically, the feature variable selection includes the following steps:

[0038] S11, removing abnormal samples of each target gas in the modeling samples to obtain a sample set after removing the abnormalities;

[0039] Step S11 is specifically as follows:

[0040] S11a: Use Monte Carlo resampling method to sample the modeling samples into K sub-sample spaces;

[0041] For the same sample space, the number of samples is fixed, and K Monte Carlo random samplings are performed. Each sampling obtains a sub-sample space, thus obtaining K sub-sample spaces.

[0042] S11b: Establish a separate multivariate correction model for each sub-sample space, perform predictions, count the residuals of each sample prediction, and calculate the mean or standard deviation of the residuals. When the mean or standard deviation of the sample residuals exceeds the preset value, remove the sample from the corresponding sub-sample space to obtain a sample set after abnormality removal.

[0043] It should be noted that the multivariate correction model adopts the PLS model. The process of establishing the model and predicting the subspace is as follows:

[0044] The processing of each sub-sample space is basically the same. The processing of one sub-sample space is taken as an example to illustrate as follows:

[0045] The samples in the subsample space are used as the training set to build a PLS model, and the samples in the non-subsample space in the same sample space are used as the test set, and the residual of each sample is predicted and calculated through the trained PLS model.

[0046] Summarize and count the PLS model prediction residuals of each sample in the K sub-sample spaces, and calculate their mean or standard deviation. According to the T-test method, set the confidence level to 0.05, screen and eliminate samples that bring significant differences, and obtain the sample set after abnormal elimination. S12, perform wavelength variable selection in the sample set after elimination, and obtain the final selected wavelength variable.

[0047] It should be noted that step S12 is specifically as follows:

[0048] S12a: After the samples after the abnormalities are eliminated are combined, the Monte Carlo sampling method is used again to obtain K sub-sample spaces;

[0049] S12b: After optimizing the wavelength variable combination for each sub-sample space using the genetic algorithm, the number of times each wavelength variable in the optimal population takes the value of 1 is counted. After the statistics are sorted, the wavelength variables with values ​​above the median are retained as the final selected wavelength variables.

[0050] The processing of each sub-sample space is basically the same. The processing of one sub-sample space is taken as an example to illustrate as follows:

[0051] The binary value array (1 and 0) of the wavelength variable size is a coded chromosome, and several chromosomes form a population. The meaning of the binary value is that when the value is 1, it means that the variable at the corresponding position is selected, otherwise it will not be used.

[0052] The initial population obtains the optimal population after an iterative process of bionic steps of biological population evolution: "selection", "crossover", "mutation", "population update" and "individual fitness evaluation".

[0053] The number of times each wavelength variable takes the value of 1 in the optimal population is counted. After the statistics are sorted, the wavelength variables with values ​​above the median are retained as the final selected wavelength variables.

[0054] Please refer to Figure 2 , Figure 2 This is a schematic diagram of the feature variable selection results.

[0055] The three line graphs represent the absorbance spectra of the three components in the 190~205nm band. It can be seen that there is a serious overlap between the characteristic absorption spectra of the three components; the three scatter plots represent the characteristic selected wavelength variables obtained by variable selection for each component.

[0056] S2. construct a multivariate correction basic concentration prediction model, and train the model using the preprocessed data to obtain a trained model;

[0057] It should be noted that the multivariate correction basic concentration prediction model is constructed according to the partial least squares method, which is specifically as follows:

[0058]

[0059]

[0060] Among them, X is the spectral signal matrix, Y is the concentration matrix, U, Q, T, P are the factor analysis decomposition matrices, u k ,q k ,t k 、p k is the factor analysis decomposition vector, E X 、E Y is the residual matrix.

[0061] Implicit regression relationship:

[0062]

[0063]

[0064] predict:

[0065]

[0066] When the model is trained using the finally selected samples, the sum of the root mean square error (the sum of the corrected root mean square error RMSEC and the predicted root mean square error RMSEP) is used as the key indicator, and the model with the minimum index is used as the output trained model.

[0067] S3, inputting the original absorbance spectrum of the sample to be tested into the trained model to obtain the preliminary predicted concentration of each component of the sample to be tested;

[0068] S4, obtaining a reference absorbance spectrum by inverting the initial predicted concentration and the basic absorbance spectrum of the target gas;

[0069] It should be noted that step S4 is specifically as follows:

[0070] S41, obtaining basic absorbance spectra of target gas at different concentrations;

[0071] The basic absorbance spectrum is the concentration-removed mean spectrum, and the concentration removal method can be achieved by subtracting the concentration information from the absorbance value of each wavelength or pixel one by one.

[0072] S42. Invert the corresponding basic absorbance reference spectrum based on the preliminary concentration of each target component.

[0073] S5, constructing a wavelength drift adaptive correction model, inputting the original absorbance spectrum of the sample to be tested into the wavelength drift adaptive correction model, and obtaining a corrected absorbance spectrum;

[0074] It should be noted that the correction model in the present invention is as follows:

[0075]

[0076]

[0077] in, is the corrected absorbance spectrum i The correction value at each wavelength or pixel point, is the interpolation function, x is the wavelength or pixel vector of the original absorbance spectrum, y is the absorbance vector of the original absorbance spectrum, is the absorbance value at the i-th wavelength or pixel point in the original absorbance spectrum, k is the expansion correction factor, b is the translation correction factor, k , b is the initialization parameter, N is the total number of wavelengths or pixels in the absorbance spectrum; k 1 , k 2 , b 1 , b 2 is the preset value, where k 1 , k 2 , b 1 , b 2 It is set based on practical experience and defines a predetermined range of the translation correction factor and the telescopic correction factor. Data within this range are credible values ​​or acceptable values.

[0078] S6, comparing the reference absorbance spectrum and the corrected absorbance spectrum, adjusting the wavelength drift adaptive correction model parameters, and obtaining a trained wavelength drift adaptive correction model;

[0079] It should be noted that in step S6, the reference absorbance spectrum and the corrected absorbance spectrum are compared using a similarity measurement method. If the similarity between the two exceeds a preset value, it indicates that the wavelength drift adaptive correction model parameters are appropriate.

[0080] The similarity evaluation formula is as follows:

[0081]

[0082] in, R is the similarity measurement parameter, Similarity () indicates the use of multiple evaluation indicators for comprehensive voting calculation, is the basic absorbance reference spectrum, is the corrected absorbance spectrum.

[0083] Adjust model parameters, specifically: k , b Increase or decrease the corresponding step size Δ k , Δ b . k 1 , k 2 Limit the value range of k, b 1 、b 2 The value range of b is limited. S7, inputting the original absorbance spectrum of the sample to be tested into the trained wavelength drift adaptive correction model to obtain a corrected spectrum;

[0084] S8, inputting the corrected spectrum into the trained model to obtain multi-component concentration output;

[0085] Generally speaking, the concentration prediction in the present invention is divided into two levels, wherein the first level is a multivariate correction basic concentration prediction model. Figure 3 , Figure 3 It is a schematic diagram of the process of building a basic concentration prediction model.

[0086] The cross-interference effect of each target gas is eliminated by integrating characteristic variable selection and multivariate correction model building. The optimal basic concentration prediction model is output with the sum of predicted root mean square errors as the evaluation index.

[0087] Please refer to Figure 4 , Figure 4 It is the prediction flow chart of wavelength drift adaptive correction model.

[0088] Concentration prediction is performed at two levels.

[0089] The preliminary concentration is obtained from the first-level prediction. In the first-level concentration prediction, the wavelength drift adaptive correction model is integrated, which can improve the robustness of the prediction model.

[0090] In the second stage, the final prediction output still uses the basic concentration prediction model. The initial predicted concentration values ​​of each component and the corresponding stored basic spectrum are used to invert to obtain the real-time reference sum spectrum, and the correction model is optimized to obtain the corrected spectrum. The corrected spectrum is used as the input of the basic concentration prediction model to obtain the final concentration prediction result. Through the above overall steps, the present invention can achieve more accurate concentration prediction while eliminating cross interference.

[0091] As an embodiment, the present invention provides a multi-component (NO, SO 2 and NH 3 ) and the fast prediction results are shown in Table 1.

[0092] Table 1 Prediction table of mixed gas sample concentration

[0093]

[0094] It can be seen from Table 1 that for the prediction of three mixed gas with different concentrations, the absolute error for NO is within 3.5 ppm, and for SO 2 and NH 3 The absolute error is within 1ppm.

[0095] See also Figure 5 , Figure 5 It is a schematic diagram of the working of the hardware device of an embodiment of the present invention, and the hardware device specifically includes: a multi-component simultaneous determination device 401 for ammonia escape, a processor 402 and a storage medium 403.

[0096] A device 401 for simultaneous determination of multiple components of ammonia escape: The device 401 for simultaneous determination of multiple components of ammonia escape implements a method for simultaneous determination of multiple components of ammonia escape.

[0097] Processor 402: The processor 402 loads and executes the instructions and data in the storage medium 403 to implement the method for simultaneously determining multiple components of ammonia escape.

[0098] Storage medium 403: The storage medium 403 stores instructions and data; the storage medium 403 stores instructions and data for being executed by the processor 402 to implement the ammonia escape multi-component simultaneous determination method.

[0099] The beneficial effects of the present invention are: providing a method for real-time, online rapid detection of ammonia escape, which can simultaneously detect NH 3 、SO 2 By selecting the ultraviolet spectrum band spectrometer, water, CO, CO 2 , O 2and CH 4 The interference effects of various gas characteristics and selective multivariate correction models were used to reduce the influence of cross-interference and enhance the anti-interference ability of the method. The robustness of the method was enhanced by designing a two-level prediction process based on wavelength drift adaptive correction.

[0100] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for simultaneous determination of multiple components of ammonia escape, characterized in that: The method comprises the following steps: S1, performing data preprocessing on the original absorbance spectrum of the modeling sample to obtain preprocessed data; The preprocessing in step S1 includes: truncation, difference, feature variable selection and smoothing; The feature variable selection comprises the following steps: S11, removing abnormal samples of each target gas in the modeling samples to obtain a sample set after removing the abnormalities; S12, performing wavelength variable selection in the sample set after elimination to obtain a finally selected wavelength variable; Step S11 is specifically as follows: S11a: Use Monte Carlo resampling method to sample the modeling samples into K sub-sample spaces; S11b: Establish a separate multivariate correction model for each sub-sample space, perform predictions, count the residuals of each sample prediction, and calculate the mean or standard deviation of the residuals. When the mean or standard deviation of the sample residuals exceeds a preset value, remove the sample from the corresponding sub-sample space to obtain a sample set after abnormal elimination. Step S12 is specifically as follows: S12a: After the samples after the abnormalities are eliminated are combined, the Monte Carlo sampling method is used again to obtain K sub-sample spaces; S12b: After optimizing the wavelength variable combination by using the genetic algorithm for each sub-sample space, the number of times each wavelength variable in the optimal population takes the value of 1 is counted. After the statistics are sorted, the wavelength variables with values ​​above the median are retained as the final selected wavelength variables; S2. construct a multivariate correction basic concentration prediction model, and train the model using the preprocessed data to obtain a trained model; S3, inputting the original absorbance spectrum of the sample to be tested into the trained model to obtain the preliminary predicted concentration of each component of the sample to be tested; S4, obtaining a reference absorbance spectrum by inverting the initial predicted concentration and the basic absorbance spectrum of the target gas; S5, constructing a wavelength drift adaptive correction model, wherein the wavelength drift adaptive correction model adopts translation correction and telescopic correction, inputting the original absorbance spectrum of the sample to be tested into the wavelength drift adaptive correction model, and obtaining a corrected absorbance spectrum; S6, comparing the reference absorbance spectrum and the corrected absorbance spectrum, adjusting the wavelength drift adaptive correction model parameters, and obtaining a trained wavelength drift adaptive correction model; S7, inputting the original absorbance spectrum of the sample to be tested into the trained wavelength drift adaptive correction model to obtain a corrected spectrum; S8. Input the corrected spectrum into the trained model to obtain multi-component concentration output.

2. A method for simultaneous determination of multiple components of ammonia escape as claimed in claim 1, characterized in that: In step S6, the reference absorbance spectrum and the corrected absorbance spectrum are compared using a similarity measurement method. If the similarity between the two exceeds a preset value, it indicates that the wavelength drift adaptive correction model parameters are appropriate.

3. A storage medium, characterized in that: The storage medium stores instructions and data for being executed by a processor to implement a method for simultaneously determining multiple components of ammonia escape as described in any one of claims 1 to 2.

4. A multi-component simultaneous determination device for ammonia escape, characterized in that: include: A processor and a storage medium; the processor loads and executes instructions and data in the storage medium to implement a method for simultaneously determining multiple components of ammonia escape as described in any one of claims 1 to 2.

Images