A multi-component recognition method based on cataluminescence and artificial neural network

Abstract

The application relates to the technical field of analytical chemistry. A multi-component recognition method based on catalytic luminescence and an artificial neural network is provided, which comprises the following steps: a single sensing unit is used to carry out a catalytic luminescence reaction with a plurality of training samples, corresponding catalytic luminescence response signals are collected, and an initial data set is obtained; the initial data set is pretreated to obtain a pretreated data set; a component recognition initial model is constructed based on a neural network, the component recognition initial model is trained through the pretreated data set, and a component recognition model is obtained; the catalytic luminescence response signals of a sample to be detected are collected, pretreated, and the pretreated catalytic luminescence response signals of the sample to be detected are input into the component recognition model to obtain a component recognition result of the sample to be detected. The method solves the problem that existing catalytic luminescence technology is mostly limited to single-component detection, a plurality of sensing units are required for detection of a multi-component system, and the efficiency is low and it is difficult to realize simultaneous recognition of multiple components.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of analytical chemistry, in particular to a multi-component recognition method based on catalytic luminescence and artificial neural network. BACKGROUND

[0002] In the field of chemical analysis, multi-component recognition is a crucial technology that is widely used in environmental monitoring, food safety, drug analysis, and other aspects. Traditional multi-component recognition methods usually rely on complex separation techniques and the combination of multiple sensors, which not only increases the complexity of experimental operations, but also increases costs, limiting its popularity in practical applications.

[0003] In recent years, catalytic luminescence (CTL) technology has gradually emerged in the field of analytical chemistry due to its high sensitivity, high selectivity, and simple operation. Catalytic luminescence is a chemical reaction between the measured substance and the luminescent reagent triggered by the catalytic effect of the catalyst, which generates a light signal. The intensity of this light signal has a certain relationship with the concentration of the measured substance, so it can be used for quantitative analysis. However, traditional catalytic luminescence technology is mostly limited to single-component detection, and for multi-component systems, multiple sensing units are often required for detection, which not only is inefficient, but also makes it difficult to achieve simultaneous recognition of multiple components. SUMMARY

[0004] The purpose of the present application is to provide a multi-component recognition method based on catalytic luminescence and artificial neural network, aiming to solve the problem that existing catalytic luminescence technology is mostly limited to single-component detection, and for multi-component systems, multiple sensing units are required for detection, resulting in low efficiency and difficulty in achieving simultaneous recognition of multiple components.

[0005] The present application is achieved by the following technical solutions:

[0006] A multi-component recognition method based on catalytic luminescence and artificial neural network, comprising the steps of:

[0007] Performing catalytic luminescence reaction with a single sensing unit and several training samples, and collecting the catalytic luminescence response signal corresponding to each training sample to obtain an initial data set;

[0008] Pretreating the initial data set to obtain a pretreated data set;

[0009] Constructing a component recognition initial model based on a neural network, and training the component recognition initial model with the pretreated data set to obtain a component recognition model;

[0010] Collect the catalytic luminescence response signal of the to-be-detected sample, perform pretreatment, and input the pretreated catalytic luminescence response signal of the to-be-detected sample into the component recognition model to obtain a component recognition result of the to-be-detected sample.

[0011] Optionally, the preparation process of the single sensing unit is as follows:

[0012] La(NO3)3·6H2O, urea, Eu(NO3)3·6H2O and Tb(NO3)3·6H2O are added to a certain amount of ultrapure water in proportion, stirred until completely dissolved, and a mixed solution is formed;

[0013] The mixed solution is transferred to a high-pressure reaction kettle for hydrothermal reaction to generate a precursor precipitate;

[0014] After the hydrothermal reaction is completed, the high-pressure reaction kettle is cooled, the generated precursor precipitate is collected, and the precursor precipitate is washed with ultrapure water and anhydrous ethanol respectively to remove residual impurities;

[0015] The washed precursor precipitate is vacuum dried at a preset temperature to obtain a precursor solid;

[0016] The precursor solid is placed in a tube furnace, a preset heating rate is set, high-temperature calcination is performed, and finally an Eu / Tb co-doped La2O2CO3 sensing material is obtained.

[0017] Optionally, the single sensing unit adopts an Eu / Tb co-doped La2O2CO3 material; wherein the atomic ratio of the dosages of Eu and Tb to La is 0.5%-5%.

[0018] Optionally, the specific process of obtaining the initial data set by performing catalytic luminescence reaction on the single sensing unit with a plurality of training samples and collecting the catalytic luminescence response signal corresponding to each training sample is as follows:

[0019] The single sensing unit is uniformly coated on a ceramic rod in the reaction chamber;

[0020] Each training sample is subjected to gasification treatment and flows through the reaction chamber under the driving of a carrier gas, so that the training sample and the sensing material undergo catalytic luminescence reaction;

[0021] The catalytic luminescence response signal corresponding to each training sample is collected by a chemiluminescence instrument to form an initial data set containing the catalytic luminescence kinetic curves of each sample.

[0022] Optionally, the specific process of pre-processing the initial data set to obtain a pre-processed data set is as follows:

[0023] averaging sampling processing is performed on the catalytic luminescence kinetic curves of each training sample in the initial data set, and the data length of each curve is unified;

[0024] Filter processing is performed on the initial data set after the averaging sampling processing, and principal component analysis method is used for dimension reduction processing on the filtered data to extract characteristic parameters having key contribution to component identification, to obtain a feature vector corresponding to each training sample;

[0025] The feature vector corresponding to each training sample is combined into a feature vector set as a pre-processing data set.

[0026] Optionally, the specific process of constructing a component identification initial model based on a neural network and training the component identification initial model through the pre-processing data set to obtain a component identification model is as follows:

[0027] A back propagation neural network is selected as a model architecture to construct a component identification initial model; wherein the component identification initial model includes an input layer, a hidden layer and an output layer;

[0028] The hyperparameters of the component identification initial model are set, and the component identification initial model is trained according to the pre-processing data set, and the weights and biases in the network are iteratively adjusted through a back propagation algorithm, with the objective of minimizing the prediction error, until a preset training number is reached or a preset stopping condition is met, to obtain a component identification model.

[0029] Optionally, in the training process, a leave-one-out validation method is used to evaluate the performance of the model, and the hyperparameters of the component identification model are adjusted according to the evaluation results to obtain an optimal component identification model.

[0030] Optionally, the specific process of adjusting the hyperparameters of the component identification model according to the evaluation results to obtain an optimal component identification model in the training process by using the leave-one-out validation method to evaluate the performance of the model is as follows:

[0031] The pre-processing data set is divided into a training set and a validation set by using the leave-one-out validation method, and the component identification model is trained and validated for several times, and the validation results of each time are recorded;

[0032] According to the validation results, the average classification accuracy, the average quantitative accuracy and the average percentage error of non-zero true concentration of the model are calculated to evaluate the performance of the model, and the performance evaluation results are obtained;

[0033] According to the performance evaluation results, the structure, parameters and hyperparameters of the component identification model are adjusted to obtain an optimal component identification model;

[0034] The optimal component identification model is applied to multi-component identification of actual samples, the catalytic luminescence response signal of the sample to be detected is collected, and after preprocessing, the component identification result of the sample to be detected is obtained, and the accuracy and reliability of the model are verified.

[0035] Optionally, in the training process, the component identification initial model adopts a tanh function as an activation function, and adjusts the number of hidden layers and the number of neurons to optimize the model structure.

[0036] Optionally, the training sample is a mixture containing persistent organic pollutants.

[0037] The technical scheme of the present application has at least the following advantages and beneficial effects:

[0038] The efficiency of multi-component identification is improved: by using a single sensing unit combined with an advanced neural network model, simultaneous identification of multiple components is realized, avoiding the cumbersome steps of detecting multiple sensing units in traditional methods, and significantly improving the detection efficiency.

[0039] The cost is reduced: only one sensing unit is needed to complete the multi-component identification task, compared with the configuration of multiple sensing units in traditional methods, the cost of the present application is effectively controlled, which is conducive to its wide application in the fields of environmental monitoring, food safety, drug analysis, etc.

[0040] The accuracy and reliability of the identification are enhanced: by collecting the catalytic luminescence response signal of the training sample and constructing a neural network model for training, the unique characteristics of different components in the catalytic luminescence reaction can be learned, thereby realizing accurate identification of multiple components. This data-driven identification method is more accurate and reliable than traditional methods.

[0041] Simplify the experimental operation: without complex separation technology and combination of multiple sensors, only a single sensing unit and a neural network model are needed to complete multi-component identification, greatly simplifying the experimental operation process and reducing the experimental difficulty.

[0042] It has a wide application prospect: since the catalytic luminescence technology itself has the advantages of high sensitivity, high selectivity and simple operation, combined with the single sensing unit multi-component identification method of the present application, it will have a broader application prospect in the fields of environmental monitoring, food safety, drug analysis, etc. BRIEF DESCRIPTION OF DRAWINGS

[0043] Figure 1 The flowchart of the multi-component identification method based on catalytic luminescence and artificial neural network of the embodiments of the present application is shown.

[0044] Figure 2A schematic diagram of the curve relationship between the rare earth ion doping ratio and the chemiluminescence intensity in three kinds of organic phosphate ester flame retardants in the embodiments of the present application is shown.

[0045] Figure 3 A schematic diagram of the curve relationship between the catalytic luminescence response signal intensity and the signal-to-noise ratio of TCP and the carrier gas flow rate in the embodiments of the present application is shown.

[0046] Figure 4 A schematic diagram of the curve relationship between the catalytic luminescence response signal intensity and the signal-to-noise ratio of TCP and the temperature in the embodiments of the present application is shown.

[0047] Figure 5 A schematic diagram of the time-signal intensity curve relationship of different OPFRs in the embodiments of the present application is shown.

[0048] Figure 6 A schematic diagram of the classification accuracy curve of the component identification initial model inputting the unfiltered curve in the embodiments of the present application is shown.

[0049] Figure 7 A schematic diagram of the quantitative accuracy curve of the component identification initial model inputting the unfiltered curve in the embodiments of the present application is shown.

[0050] Figure 8 A schematic diagram of the curve of the mean percentage error MAPE of the component identification initial model inputting the unfiltered curve in the embodiments of the present application is shown.

[0051] Figure 9 A schematic diagram of the structure of the combination of principal component analysis (PCA) and back propagation neural network (BP) in the embodiments of the present application is shown.

[0052] Figures 10-21 A schematic diagram of the curve relationship between the catalytic luminescence response signal intensity and the concentration of 6 kinds of OPFRs in the embodiments of the present application is shown.

[0053] Figure 22 A schematic diagram of the curve relationship between the catalytic luminescence response signal and the time of TBOEP in the embodiments of the present application is shown. DETAILED DESCRIPTION

[0054] The following is a specific embodiment in combination with the drawings.

[0055] Reference Figure 1 A multi-component identification method based on catalytic luminescence and artificial neural network, comprising the steps of:

[0056] Step one, a single sensing unit and a plurality of training samples are subjected to catalytic luminescence reaction, and the catalytic luminescence response signal corresponding to each training sample is collected to obtain an initial data set.

[0057] In some embodiments, the preparation process of the single sensing unit is:

[0058] La(NO3)3·6H2O, urea, Eu(NO3)3·6H2O and Tb(NO3)3·6H2O were added to a certain amount of ultrapure water in proportion, stirred until completely dissolved to form a mixed solution;

[0059] The mixed solution was transferred to a high-pressure reaction kettle for hydrothermal reaction to generate a precursor precipitate;

[0060] After the hydrothermal reaction was completed, the high-pressure reaction kettle was cooled, and the generated precursor precipitate was collected and washed with ultrapure water and anhydrous ethanol respectively to remove residual impurities;

[0061] The washed precursor precipitate was vacuum dried at a preset temperature to obtain a precursor solid;

[0062] The precursor solid was placed in a tube furnace and high-temperature calcination was performed at a preset heating rate, and finally Eu / Tb co-doped La2O2CO3 sensing material was obtained.

[0063] In some embodiments, a single sensing unit adopts Eu / Tb co-doped La2O2CO3 material; wherein the atomic ratio of the dosages of Eu and Tb to La is 0.5%-5%.

[0064] In some embodiments, Eu / Tb co-doped La2O2CO3 sensing material is prepared by a hydrothermal synthesis method. 2.6g La(NO3)3·6H2O, 1.2g urea, Eu(NO3)3·6H2O and Tb(NO3)3·6H2O (the atomic ratio of the dosages of Eu and Tb to La is 0.5%, 1%, 2%, 3%, 5%, and the products are represented as x% Eu / Tb co-doped La2O2CO3, respectively) were added to 30ml ultrapure water, and stirred and dispersed at room temperature until completely dissolved. Then the dispersion was transferred to a high-pressure reaction kettle, and hydrothermal reaction was carried out in an oven at 170°C for 6h. After the reaction, the white precursor was collected by centrifugation, and washed with ultrapure water and anhydrous ethanol for 3 times, and vacuum dried at 60°C for 24h. Finally, the dried solid was calcined in a tube furnace at 550°C for 3h (atmosphere condition: air, heating rate: 10°C / min), and the obtained white solid was Eu / Tb co-doped La2O2CO3 sensing material. In addition, the sensing material without Eu / Tb doping does not need to add Eu(NO3)3·6H2O and Tb(NO3)3·6H2O, and the remaining steps are the same as the above operation steps.

[0065] In some embodiments, the specific process of performing catalytic luminescence reaction with a single sensing unit and a plurality of training samples, and collecting the catalytic luminescence response signal corresponding to each training sample to obtain an initial data set is as follows:

[0066] Uniformly coating a single sensing unit on a ceramic rod in a reaction chamber;

[0067] Each training sample is vaporized and flowed through the reaction chamber driven by carrier gas, so that the training sample and the sensing material undergo catalytic luminescence reaction;

[0068] The catalytic luminescence response signal corresponding to each training sample is collected by a chemiluminescence instrument to form an initial data set containing the catalytic luminescence kinetic curves of each sample.

[0069] In some embodiments, the training sample can be a mixture containing persistent organic pollutants (POPs), where the POPs can be organophosphate flame retardants (OPFRs). CTL experiments on OPFRs were conducted using a self-built CTL apparatus, which primarily consists of a BPCL-2-TGC ultraweak chemiluminescence analyzer, a self-developed quartz reaction tube, a carrier gas flow path, and a voltage controller for controlling the reaction temperature. First, 30 mg of pre-prepared, optimized Eu / Tb co-doped La2O2CO3 sensing material was dispersed in ultrapure water and evenly applied to a ceramic rod. The rod was placed in a quartz reaction tube connected to a heating device, serving as the reaction chamber. The OPFR vapor to be tested flowed through the reaction chamber driven by a carrier gas (dry air), whereupon a CTL reaction occurred. The catalytic luminescence response signal corresponding to the organophosphate flame retardants (OPFRs) was collected using the BPCL-2-TGC ultraweak chemiluminescence analyzer with an integration time of 0.1 s and an operating voltage of -0.82 kV. The catalytic luminescence wavelengths of OPFRs were obtained through a series of filters (460 nm, 490 nm, 520 nm, 555 nm, 575 nm, 620 nm, and 640 nm).

[0070] In order to achieve the ideal catalytic activity, the main factors affecting the CTL response intensity were investigated. First, the effect of different Eu / Tb dosages on the catalytic luminescence response signal of organophosphate flame retardants (OPFRs) was investigated, and three representative organophosphate flame retardants (OPFRs) including alkyl TEP (triethyl phosphate), chlorinated TCEP (tris(2-carboxyethyl)phosphine) and aryl-substituted TCP (tricalcium phosphate) were selected as the analysis objects. Figure 2 As shown in the figure, the catalytic luminescence response signal of the rare earth ion doped material is significantly higher than that of the undoped material. The catalytic luminescence response signal is strongest when the Eu / Tb dosage is 1% of La, indicating the highest catalytic activity of the sensing material at this time. Therefore, a 1% Eu / Tb co-doped La2O2CO3 sensing material was selected. Experimental conditions such as the CTL carrier gas flow rate and detection temperature were optimized. Taking TEP with the highest signal intensity as an example,Figure 3 As shown, the catalytic luminescence response signal intensity and signal-to-noise ratio (S / N) of the system under different flow rates (100 mL / min, 200 mL / min, 300 mL / min, 400 mL / min and 500 mL / min) were tested and calculated respectively, and it can be found that both the catalytic luminescence response signal intensity and the signal-to-noise ratio increase first and then decrease with the increase of the flow rate, and the catalytic luminescence response signal intensity and the signal-to-noise ratio reach the highest at the carrier gas flow rate of 300 mL / min. This is because the diffusion of TCP is relatively slow at a lower flow rate, and the reaction is insufficient at a too high flow rate. Accordingly, the carrier gas flow rate of 300 mL / min is selected as the optimal condition. As shown in Figure 4 As shown, the catalytic luminescence response signal intensity increases with the increase of the detection temperature from 193°C to 273°C, and the signal-to-noise ratio reaches the optimum at 263°C, so 263°C is selected as the optimal detection temperature. In summary, all subsequent experiments are carried out under the conditions of the dosage of the sensing material being 1% of La, the carrier gas flow rate being 300 mL / min, and the detection temperature being 263°C. Six representative OPFRs, TEP, TNBP (tributyl phosphate), TBOEP (tris (2-butoxy) ethyl phosphate), TCEP, TCPP (tris (2-chloropropyl) phosphate) and TCP, are selected as single components of each compound and four multi-component mixtures of TCPP and TNBP, TCEP and TEP, TCEP and TCP, and TEP and TCP as training samples. The catalytic luminescence response signals of each training sample at the concentrations of 1.76 μg / mL, 3.52 μg / mL, 5.38 μg / mL, 7.04 μg / mL and 8.8 μg / mL are detected, and 10 parallel experiments are carried out. As shown in Figure 5 As shown, the time-signal intensity curves of different OPFRs obtained are used as initial data to establish an initial data set.

[0071] Step two, pre-process the initial data set to obtain a pre-processed data set.

[0072] In some embodiments, the specific process is as follows:

[0073] Average sampling processing is performed on the catalytic luminescence kinetics curves of each training sample in the initial data set to unify the data length of each curve;

[0074] Filtering processing is performed on the initial data set after the average sampling processing, and the dimension reduction processing is performed on the data after the filtering processing by using the principal component analysis method to extract the characteristic parameters which have key contributions to the component identification, to obtain the characteristic vector corresponding to each training sample;

[0075] The characteristic vectors corresponding to each training sample are combined to form a characteristic vector set as the pre-processed data set.

[0076] According to the catalytic luminescence kinetics curve number, it can be found that due to the inevitable noise of catalytic luminescence response signal in the measurement process, the spectral vector of each training sample looks very much after visualizing. Assuming that the "burr" caused by noise does not contain the characteristic information we need, filtering is performed on the catalytic luminescence response signal to obtain a more "smooth" curve, and the unfiltered curve and the filtered curve are used to establish the data set input component identification initial model respectively. 10 times of repeated experiments are carried out for comparison. In 10 experiments, the classification accuracy (such as Figure 6 ) and the quantitative accuracy (such as Figure 7 ) of the component identification initial model inputting the unfiltered curve are always lower than those of the component identification initial model inputting the filtered curve, and the average percentage error MAPE (such as Figure 8 ) of the component identification initial model inputting the unfiltered curve is always higher than that of the component identification initial model inputting the filtered curve. The average classification accuracy decreases to 95.41%, the average quantitative accuracy decreases to 84.44%, and the average percentage error MAPE increases to 0.13. This is enough to show that the noise in the catalytic luminescence response signal has little practical significance for the identification of OPFRs, and the more essential information is still contained in the main waveform of the model. The uniform transformation of the spectrum shape during filtering is also acceptable. At the same time, the filtered catalytic luminescence kinetics curve is more conducive to the subsequent model step focusing on more obvious waveforms respectively, rather than noise.

[0077] The spectral vector of the catalytic luminescence kinetics curve of the training sample is processed by average sampling to have a length of 64. The initial vector processed by filtering without dimension reduction is still very long and uneven in length. Due to the high dimension of the feature space, the distance between any two samples in the feature space is extremely far, which causes the curse of dimensionality. Therefore, principal component analysis (PCA) and back propagation neural network (BP) can be used to solve this problem (as shown in Figure 9 ). Its advantage is that PCA is a clear reversible orthogonal transformation. Assuming that the original data is X, the projection matrix composed of the characteristic vector is P, then the reduced matrix Y can be calculated by Y=X*P, and the original matrix can also be calculated by the reduced matrix Y. At the same time, PCA is to transform a set of possibly related variables into a set of linearly independent variables through orthogonal transformation, which lays the foundation for the extracted feature parameters to be independent of each other. When the sample set is mapped to low latitude, the characteristics of the sample set are kept different. Finally, the first 6 dimensions after dimension reduction are used as input features.

[0078] Step three, based on neural network to construct component identification initial model, and through the preprocessed data set to train the component identification initial model, to obtain the component identification model.

[0079] In some embodiments, the component identification initial model is constructed based on a neural network, and the component identification initial model is trained through a preprocessed dataset, and the specific process is as follows:

[0080] A back propagation neural network is selected as the model architecture to construct the component identification initial model; wherein the component identification initial model comprises an input layer, a hidden layer and an output layer; the hyperparameters of the component identification initial model are set, and the component identification initial model is trained according to the preprocessed dataset, the weights and biases in the network are iteratively adjusted through the back propagation algorithm, and the prediction error is minimized as the target, until the preset training number is reached or the preset stopping condition is met, to obtain the component identification model. The back propagation neural network is selected as the model architecture of the component identification initial model, because the back propagation neural network is more suitable for the simple dataset of OPFRs. According to the dataset, the back propagation neural network is optimized, and after optimization, 5 layers of hidden layers are included, and the number of neurons is 48, 72, 54, 36 and 24 respectively. In order to avoid negative numbers after data standardization, a tanh function with a mapping range including all real numbers is selected as the activation function. The learning rate is set to 0.01, and is reduced to 0.5 times of the previous value every 500 rounds, and the training is stopped after 1200 rounds and the indicators are recorded. The advantage of neural network compared with other regression algorithms or ordinary machine learning methods is that it can more simply predict the concentration of multiple compounds and establish the relationship between different compounds. The key is that the distortion of the cataluminescence kinetic curve in the multi-component system may not be linear, which is also the reason why traditional methods are more difficult to face multi-component identification. When multiple compounds coexist, spectral distortion not only occurs when the compound type changes, but also occurs when the compound concentration changes. In this case, the back propagation neural network will have better fitting effect. In order to meet the actual requirements of multi-component identification and clearly express the detection results, the component identification initial model has 6 layers of output layer, which can output the concentration detection results of 6 kinds of compounds at the same time, to ensure that the component identification initial model can simultaneously evaluate and detect the concentration of 6 OPFRs for each unknown sample, so that the final output result can give the component number of the unknown sample and the qualitative and quantitative analysis results of each component.

[0081] In some embodiments, in the training process, the leave-one-out method is used to evaluate the performance of the model, the hyperparameters of the component identification model are adjusted according to the evaluation results, and the optimal component identification model is obtained.

[0082] In some embodiments, in the training process, the leave-one-out method is used to evaluate the performance of the model, the hyperparameters of the component identification model are adjusted according to the evaluation results, and the optimal component identification model is obtained.

[0083] The preprocessed data set is divided into a training set and a validation set by using leave-one-out validation method, the component identification model is trained and validated for several times, and the validation result of each time is recorded; according to the validation result, the average classification accuracy, the average quantitative accuracy and the average percentage error of non-zero real concentration of the model are calculated, the performance of the model is evaluated, the performance evaluation result is obtained, and the prediction effect of the model can be better measured by multiple data; according to the performance evaluation result, the structure, parameters and hyperparameters of the component identification model are adjusted to obtain the optimal component identification model; the optimal component identification model is applied to the multi-component identification of actual samples, the catalytic luminescence response signal of the sample to be detected is collected, and after preprocessing, the model is input to obtain the component identification result of the sample to be detected, and the accuracy and reliability of the model are verified.

[0084] According to the 6-layer output result given by the trained component identification model, the simultaneous detection results of 6 OPFRs (TEP, TNBP, TBOEP, TCEP, TCPP and TCP) of the training set and the validation set are drawn respectively. Through the component identification model, 10 experiments of multi-component OPFRs are carried out to evaluate the generality of the model. The classification accuracy, quantitative accuracy and average percentage error (MAPE) results of 10 experiments are shown in Table 1 as follows:

[0085] Table 1: Results of 10 experiments of multi-component OPFRs of component identification model

[0086]

[0087] From Table 1, it can be seen that the average classification accuracy of multi-component identification of 6 OPFRs is 98.38%, and the classification accuracy of each training is above 94.59%. The average quantitative accuracy is 94.42%, and the quantitative accuracy of each training is above 90.20%. The average MAPE is 0.08, and the MAPE of each training is below 0.10. Based on this, it is verified that the constructed component identification model can quickly and accurately realize the multi-component identification and quantitative analysis of 6 OPFRs based on the catalytic luminescence kinetic curve, including but not limited to the quantitative analysis of concentration. This identification strategy well avoids the shortcomings of insufficient multi-component identification ability in existing methods, and further embodies the great potential of the component identification model for multi-component OPFRs identification.

[0088] At the same time, the analysis performance of the component identification model is also investigated. For example, Figures 10 to 21As shown, the catalytic luminescence response signal intensity of the six OPFRs showed good linear relationship with different concentrations. Through calculation, the detection limits of TBOEP, TEP, TCP, TNBP, TCPP and TCEP were 0.16 μg / mL, 0.67 μg / mL, 1.51 μg / mL, 0.49 μg / mL, 0.31 μg / mL and 0.99 μg / mL, respectively. In addition, the stability of the material is very important for gas sensing. The catalytic luminescence response signal of 1% Eu / Tb doped La2O2CO3 surface TBOEP was detected, as shown in Figure 22 The results showed that the sensing material exhibited excellent stability within 5 days, with RSD (relative standard deviation) less than 5.30%, which provided a strong guarantee for long-term identification and detection of OPFRs in future practical applications.

[0089] Step four, collecting the catalytic luminescence response signal of the sample to be detected, preprocessing, and inputting the catalytic luminescence response signal of the preprocessed sample to be detected into the component recognition model to obtain the component recognition result of the sample to be detected.

[0090] In some embodiments, according to actual needs, prepare the mixture sample to be detected, ensure that the quantity and state of the sample meet the experimental requirements; ensure that the state of the experimental device (such as the CTL device) is consistent with that during training, including the flow rate of the carrier gas, the detection temperature, etc.; gasify the sample to be detected, and make it flow through the reaction chamber equipped with a single sensing unit under the driving of the carrier gas, so that the sample and the sensing material undergo a catalytic luminescence reaction; collect the catalytic luminescence response signal corresponding to the sample to be detected by the chemiluminescence instrument, and ensure that the integration time, working voltage and other parameters are consistent with those during training. Average sampling processing is performed on the collected catalytic luminescence kinetic curve, the data length of each curve is unified, and consistency with the data format of the training set is ensured; the same filtering method as during training is used to filter the catalytic luminescence response signal, remove noise, and obtain a smoother curve; principal component analysis method is used to reduce the dimension of the data after filtering, extract characteristic parameters that have key contributions to component identification, and obtain the characteristic vector corresponding to the sample to be detected. The characteristic vector of the preprocessed sample to be detected is used to form input data, and the data format is ensured to be consistent with that during training; the input data is input into the component identification model that has been trained, and the model will automatically make a prediction; the component identification result of the sample to be detected is read from the model output, including qualitative and quantitative analysis results of each component. Compare the component identification result predicted by the model with other methods in the laboratory, such as standard detection methods or verified alternative methods, to verify the accuracy and reliability of the model; according to the comparison result, evaluate the classification accuracy, quantitative accuracy and average percentage error of the model, and ensure that the performance of the model in actual application is stable and reliable. Apply the verified model to the multi-component identification of actual samples, continuously collect the catalytic luminescence response signal of new samples to be detected and input the model, to realize the rapid and accurate identification of multi-component pollutants; according to the feedback and data accumulation in actual application, continuously optimize the structure and parameters of the model, and improve the identification accuracy and generalization ability of the model.

Claims

1. A multi-component identification method based on catalytic luminescence and artificial neural network, characterized in that: Including steps: A single sensor unit is used to perform a catalytic luminescence reaction with several training samples, and the catalytic luminescence response signal corresponding to each training sample is collected to obtain an initial data set; The preparation process of the single sensing unit is as follows: Add La(NO3)3⋅6H2O, urea, Eu(NO3)3⋅6H2O and Tb(NO3)3⋅6H2O in proportion to a certain amount of ultrapure water and stir until completely dissolved to form a mixed solution; The mixed solution is transferred to a high-pressure reactor for hydrothermal reaction to generate a precursor precipitate; After the hydrothermal reaction is completed, the autoclave is cooled, the generated precursor precipitate is collected, and the precursor precipitate is washed with ultrapure water and anhydrous ethanol respectively to remove residual impurities; The washed precursor precipitate is vacuum dried at a preset temperature to obtain a precursor solid; The precursor solid is placed in a tube furnace and calcined at a preset heating rate to obtain a Eu / Tb co-doped La2O2CO3 sensing material. The single sensing unit uses the Eu / Tb co-doped La2O2CO3 material. The atomic ratio of the amount of Eu and Tb added to La is 0.5%-5%, respectively. Preprocessing the initial data set to obtain a preprocessed data set; An initial component recognition model is constructed based on a neural network, and the initial component recognition model is trained by preprocessing the data set to obtain a component recognition model; The catalytic luminescence response signal of the sample to be detected is collected and preprocessed, and the preprocessed catalytic luminescence response signal of the sample to be detected is input into the component recognition model to obtain the component recognition result of the sample to be detected.

2. The multi-component identification method based on catalytic luminescence and artificial neural network according to claim 1, characterized in that: The specific process of performing catalytic luminescence reaction on a single sensor unit and a plurality of training samples and collecting the catalytic luminescence response signal corresponding to each training sample to obtain the initial data set is as follows: Uniformly coating a single sensing unit on a ceramic rod in a reaction chamber; Each training sample is vaporized and flowed through the reaction chamber driven by carrier gas, so that the training sample and the sensing material undergo catalytic luminescence reaction; The catalytic luminescence response signal corresponding to each training sample is collected by a chemiluminescence instrument to form an initial data set containing the catalytic luminescence kinetic curves of each sample.

3. The multi-component identification method based on catalytic luminescence and artificial neural network according to claim 2, characterized in that: The specific process of preprocessing the initial data set to obtain the preprocessed data set is as follows: The catalytic luminescence kinetic curves of each training sample in the initial data set were averaged and the data length of each curve was unified. The initial data set after average sampling is filtered, and the principal component analysis method is used to reduce the dimension of the filtered data, extract the characteristic parameters that have a key contribution to component identification, and obtain the characteristic vector corresponding to each training sample; The feature vectors corresponding to each training sample are combined into a feature vector set as a preprocessing data set.

4. The multi-component identification method based on catalytic luminescence and artificial neural network according to claim 1, characterized in that: The specific process of constructing the initial component recognition model based on the neural network and training the initial component recognition model by preprocessing the data set to obtain the component recognition model is as follows: A back-propagation neural network was selected as the model architecture to construct an initial component identification model. The initial component identification model includes an input layer, a hidden layer, and an output layer. The hyperparameters of the initial component identification model are set, and the initial component identification model is trained based on the preprocessed dataset. The weights and biases in the network are iteratively adjusted through the back propagation algorithm with the goal of minimizing the prediction error until the preset number of training rounds is reached or the preset stopping condition is met, and the component identification model is obtained.

5. The multi-component identification method based on catalytic luminescence and artificial neural network according to claim 4, characterized in that: During the training process, the leave-one-out validation method is used to evaluate the performance of the model. The hyperparameters of the component recognition model are adjusted according to the evaluation results, and the optimal component recognition model is obtained.

6. The multi-component identification method based on catalytic luminescence and artificial neural network according to claim 5, characterized in that: During the training process, the leave-one-out validation method is used to evaluate the performance of the model. The hyperparameters of the component identification model are adjusted according to the evaluation results. The specific process of obtaining the optimal component identification model is as follows: The leave-one-out validation method was used to divide the preprocessed data set into a training set and a validation set. The component recognition model was trained and validated several times, and the validation results were recorded each time. Based on the validation results, the average classification accuracy, average quantitative accuracy, and average percentage error of non-zero true concentration of the model were calculated to evaluate the performance of the model and obtain the performance evaluation results; According to the performance evaluation results, the structure, parameters and hyperparameters of the component identification model are adjusted to obtain the optimal component identification model; The optimal component recognition model is applied to the multi-component recognition of actual samples. By collecting the catalytic luminescence response signals of the samples to be tested, preprocessing them and inputting them into the model, the component recognition results of the samples to be tested are obtained, and the accuracy and reliability of the model are verified.

7. The multi-component identification method based on catalytic luminescence and artificial neural network according to claim 4, characterized in that: During the training process of the component identification initial model, the tanh function is used as the activation function, and the model structure is optimized by adjusting the number of hidden layers and the number of neurons.

8. The multi-component identification method based on catalytic luminescence and artificial neural network according to any one of claims 1 to 7, characterized in that: The training samples are mixtures containing persistent organic pollutants.

Images