A method for rapid detection of wastewater pollutants by atomic fluorescence spectrometry

Abstract

The application provides a method for rapidly detecting wastewater pollutants by atomic fluorescence spectrometry, and belongs to the technical field of wastewater treatment.The atomic fluorescence spectrometric data of wastewater samples are collected to establish a spectral interference matrix to represent the spectral line overlap between elements, abnormal values are identified through statistical analysis to construct a spectral outlier matrix to record abnormal fluorescence intensity distribution, a non-characteristic fluorescence signal of a sample matrix is measured to establish a background fluorescence interference matrix to quantify background interference intensity, signal change rules of a detection system are continuously monitored to construct a signal drift tendency matrix to predict signal shift trends, a spectral analysis function is used to process the spectral interference matrix and the background fluorescence interference matrix to calculate a compensation factor to correct the influence of spectral overlap and background interference and to real-time correct system drift to output a final wastewater pollutant concentration detection result, and the technical problem of unstable detection precision caused by multiple interference factors coupling is solved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of wastewater treatment, and particularly relates to a method for rapidly detecting wastewater pollutants by atomic fluorescence spectrometry. BACKGROUND

[0002] As a highly sensitive elemental analysis technique, atomic fluorescence spectrometry is widely used in the quantitative detection of heavy metal pollutants in wastewater. Traditional atomic fluorescence detection methods generate characteristic fluorescence signals by exciting target elements, and realize quantitative analysis according to the linear relationship between fluorescence intensity and element concentration. This technology is widely used in environmental monitoring, food safety, geological exploration and other fields, and plays an important role in pollutant monitoring in wastewater treatment plants, industrial wastewater discharge detection and groundwater pollution assessment. However, traditional atomic fluorescence detection technology faces many challenges in practical application. First, the complex wastewater matrix coexists with multiple elements, resulting in serious spectral line overlap interference, which affects the accurate identification of target elements. Second, the background fluorescence signal generated by the sample matrix is superimposed with the target element signal, reducing the selectivity of the detection. In addition, the detection system has signal drift phenomenon during long-time operation, and the instrument stability is difficult to guarantee. Existing single correction methods can only solve a certain type of interference, and lack of systematic processing ability for multiple interference factors. Traditional technology is difficult to effectively handle the coupling effects of spectral interference, background interference, signal drift and other multiple factors at the same time, resulting in insufficient precision and stability of the detection results in complex wastewater matrix, which cannot meet the technical requirements of modern environmental monitoring for high precision and high stability detection. That is, there is a technical problem of unstable detection precision caused by coupling of multiple interference factors when atomic fluorescence spectrometry is used to detect wastewater pollutants in the existing technology. SUMMARY

[0003] Therefore, the application provides a method for rapidly detecting wastewater pollutants by atomic fluorescence spectrometry, which can solve the technical problem of unstable detection precision caused by coupling of multiple interference factors when atomic fluorescence spectrometry is used to detect wastewater pollutants in the existing technology.

[0004] This invention is implemented as follows: It provides a method for rapid detection of pollutants in wastewater using atomic fluorescence spectrometry, comprising: collecting atomic fluorescence spectral data of wastewater samples to establish a spectral interference matrix to characterize the overlap of spectral lines among multiple elements; identifying outliers in the spectral data through statistical analysis to construct a spectral outlier matrix to record the distribution of abnormal fluorescence intensity at each wavelength; measuring the non-characteristic fluorescence signal generated by the sample matrix to establish a background fluorescence interference matrix to quantify the background interference intensity at different wavelengths; continuously monitoring the signal change pattern of the detection system during long-term operation to construct a signal drift tendency matrix to predict the signal shift trend; establishing a signal drift deterministic matrix based on historical detection data and current measurement conditions to accurately describe the amplitude and direction of signal drift; processing the spectral interference matrix and background fluorescence interference matrix using a spectral analytical function to calculate a compensation factor to correct the influence of spectral overlap and background interference on the measurement results; using a spectral line identification model to intelligently analyze the spectral outlier matrix and using the signal drift deterministic matrix to calculate a jump factor to correct system drift in real time and output the final wastewater pollutant concentration detection result.

[0005] Specifically, the spectral interference matrix is ​​formed by measuring the fluorescence spectra of each target element in a standard mixed solution, calculating the degree of mutual influence between different elements at the same wavelength position, and creating a mathematical matrix describing the overlapping interference relationship of spectral lines between elements.

[0006] Specifically, the spectral outlier matrix is ​​a two-dimensional data matrix constructed by statistically analyzing spectral data from multiple consecutive measurements to identify abnormal data points that deviate from the normal fluorescence intensity range, based on wavelength position and frequency of occurrence.

[0007] Specifically, the background fluorescence interference matrix is ​​a data matrix that describes the distribution pattern of background interference by measuring the intensity of non-characteristic fluorescence signals generated by the sample matrix at various detection wavelengths without adding target analytical elements.

[0008] Specifically, the signal drift tendency matrix is ​​established by long-term monitoring of the signal stability of the detection system, statistical analysis of the trend and pattern of signal changes over time, and prediction of future signal drift direction and amplitude.

[0009] Specifically, the signal drift deterministic matrix is ​​a matrix that accurately calculates the specific value of the system signal drift and the correction parameters under the current detection conditions based on the deviation between the measurement results of the real-time calibration standard sample and the theoretical value.

[0010] The compensation factor is specifically a correction coefficient calculated based on the intensity of spectral interference and the background fluorescence level, used to eliminate the influence of multi-element coexistence and matrix effects on the measurement of the fluorescence intensity of the target element.

[0011] The jump factor is specifically a real-time correction parameter calculated based on a signal drift certainty matrix, and is used to compensate for signal drift caused by temperature changes or device aging during continuous operation of the detection system.

[0012] The spectral analysis function is used to process a multi-dimensional spectral data matrix and calculate corresponding compensation parameters, and the input includes element correlation coefficients extracted from a spectral interference matrix, intensity distribution data obtained from a background fluorescence interference matrix, target detection element theoretical fluorescence wavelength positions queried from a standard spectral library, matrix concentration parameters determined during wastewater sample pretreatment, and instrument response coefficients obtained during instrument calibration. The output is a compensation factor value and a corresponding correction weight coefficient for each target element.

[0013] The specific structure of the spectral line recognition model is a multi-level feature extraction and fusion system based on a hierarchical attention network architecture, including an input layer that receives a spectral data matrix and performs preprocessing and normalization, multiple parallel convolutional feature extraction branches that capture spectral features at different scales, a hierarchical attention mechanism module that dynamically allocates the importance of different abstract level features through hierarchical fusion weights, a feature fusion layer that combines multi-scale features by weighting, and a fully connected classification layer that outputs class labels and confidence scores for spectral outliers.

[0014] Further, before the spectral line recognition model training step, there is also a spectral line recognition model training data set establishment step, specifically including collecting atomic fluorescence spectral data of different types of wastewater samples under various detection conditions as a basic data source, manually labeling each spectral data to distinguish normal fluorescence peaks and abnormal outliers, constructing synthetic spectral data containing known spectral overlap by adding different concentrations of interference elements and changing matrix components, and adding noise and drift changes consistent with the actual detection environment to the original spectral data using data enhancement techniques.

[0015] Further, the spectral line recognition model training step specifically includes initializing the model with preprocessed training data sets, adjusting network weights step by step using a gradient descent optimization algorithm to minimize prediction errors, monitoring model performance and preventing overfitting during training using validation set data, and optimizing training efficiency and convergence stability by adjusting learning rate and batch size.

[0016] The weight adjustment function is used for adjusting the hierarchical fusion weight parameter of the spectral line recognition model, the weight adjustment function is based on the comprehensive evaluation value calculation of the spectral complexity evaluation value extracted from the spectral interference matrix, the interference intensity measurement value obtained from the background fluorescence interference matrix and the system stability index calculated from the signal drift tendency matrix, and when the comprehensive evaluation value is less than 0.414, a conservative weight distribution strategy is adopted to enhance the low-level feature weight to improve the basic spectral recognition accuracy.

[0017] The two threshold values of the comprehensive evaluation value are obtained, specifically including collecting a large amount of atomic fluorescence spectrum detection data covering wastewater samples of different complexity as a basic data set for threshold determination, calculating the spectral complexity evaluation value for each detection sample to reflect the spectral line overlap degree, measuring the interference intensity measurement value to quantify the influence degree of background fluorescence on detection, evaluating the system stability index to represent the stability level of the instrument in the detection process, and calculating the comprehensive evaluation value corresponding to each sample according to the above three indexes according to the equal weight method.

[0018] The complexity index is specifically a comprehensive index for quantifying the complexity of the spectrum by calculating the proportion of the number of non-zero elements in the spectral interference matrix to the total number of elements, and combining the number of spectral line overlapping peaks and the peak intensity variance.

[0019] The background fluorescence interference intensity is specifically a weighted average value of the background signal intensity at each wavelength position extracted from the background fluorescence interference matrix, and the weight is determined according to the correlation between the fluorescence wavelength position of the target detection element and the background signal wavelength position.

[0020] The present application realizes comprehensive quantitative characterization of various interference factors in the detection process by constructing five core data matrices, namely spectral interference matrix, spectral outlier matrix, background fluorescence interference matrix, signal drift tendency matrix and signal drift certainty matrix. The spectral interference matrix and the background fluorescence interference matrix are processed by the spectral analysis function to calculate the compensation factor, which effectively eliminates the influence of multi-element coexistence and matrix effect on the detection result. At the same time, the spectral line recognition model based on hierarchical attention network is used to intelligently identify the spectral outliers, and the jump factor calculated by the signal drift certainty matrix is used to realize real-time correction of system drift. This method overcomes the limitations of single correction method in traditional technology, and realizes comprehensive response to complex interference environment through multi-dimensional collaborative processing. The weight adjustment function adaptively adjusts the model parameters according to the comprehensive evaluation value, ensuring optimal detection performance under different complexity conditions, solving the technical problem of coupled influence of multiple interference factors on detection accuracy, realizing synchronous compensation and dynamic correction of spectral interference, background interference, signal drift and other interference factors, and significantly improving the accuracy, stability and reliability of atomic fluorescence spectrum detection in complex wastewater matrix, providing technical support for accurate quantitative analysis of wastewater pollutants.BRIEF DESCRIPTION OF DRAWINGS BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 is a flow chart of the method of the present application.

[0022] Figure 2 is a schematic diagram of the structure of the spectral line recognition model.

[0023] Figure 3 is a schematic diagram of the typical characteristics of normal fluorescence peaks, including four subgraphs: (a) is a schematic diagram of a symmetric Gaussian peak; (b) is a schematic diagram of a multi-element normal peak; (c) is a schematic diagram of a concentration series peak; and (d) is a schematic diagram of stable baseline characteristics.

[0024] Figure 4 is a schematic diagram of the typical characteristics of abnormal outliers, including six subgraphs: (a) is a schematic diagram of peak position shift; (b) is a schematic diagram of an asymmetric peak; (c) is a schematic diagram of a double peak phenomenon; (d) is a schematic diagram of baseline drift interference; (e) is a schematic diagram of spectral interference overlap; and (f) is a schematic diagram of noise spikes. DETAILED DESCRIPTION

[0025] In order to make the purposes, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application.

[0026] As Figure 1 shown, is a flow chart of a method for rapidly detecting wastewater pollutants by atomic fluorescence spectrometry provided by the present application, and the method includes the following steps:

[0027] S01, atomic fluorescence spectrum data of a wastewater sample is collected, and a spectral interference matrix is established to represent the spectral line overlap between multiple elements;

[0028] S02, outliers in the spectral data are identified through statistical analysis, and a spectral outlier matrix is constructed to record the abnormal fluorescence intensity distribution at each wavelength position;

[0029] S03, non-characteristic fluorescence signals generated by the sample matrix are measured, and a background fluorescence interference matrix is established to quantify the background interference intensity at different wavelengths;

[0030] S04, signal variation rules of the detection system during long-term operation are continuously monitored, and a signal drift tendency matrix is constructed to predict the signal shift trend;

[0031] S05, based on historical detection data and current measurement conditions, a signal drift certainty matrix is established to accurately describe the amplitude and direction of signal drift;

[0032] S06, calculating compensation factors for correcting the influence of spectral overlap and background interference on measurement results by processing the spectral interference matrix and the background fluorescence interference matrix through a spectral analysis function;

[0033] S07, real-time correcting system drift and outputting final wastewater pollutant concentration detection results by using a spectral line recognition model to intelligently analyze the spectral outlier matrix and calculating a jump factor using the signal drift determinacy matrix.

[0034] The spectral interference matrix is specifically formed by measuring the fluorescence spectra of each target element in a standard mixed solution, calculating the mutual influence degree of fluorescence intensity of different elements at the same wavelength position, and forming a mathematical matrix describing the spectral line overlap interference relationship between elements.

[0035] The spectral outlier matrix is specifically a two-dimensional data matrix constructed by statistically analyzing spectral data measured continuously multiple times, identifying abnormal data points deviating from the normal fluorescence intensity range, and arranging them according to wavelength position and frequency of occurrence.

[0036] The background fluorescence interference matrix is specifically a data matrix describing the distribution of background interference, which is formed by measuring the non-characteristic fluorescence signal intensity generated by the sample matrix at each detection wavelength without adding target analysis elements.

[0037] The signal drift tendency matrix is specifically a prediction matrix for predicting the direction and possible amplitude of future signal drift, which is established by long-term monitoring of the signal stability of the detection system and statistically analyzing the trends and rules of signal changes over time.

[0038] The signal drift determinacy matrix is specifically a matrix for accurately calculating the specific value of system signal drift and correction parameters under current detection conditions based on the deviation between the measurement results of the real-time calibration standard sample and the theoretical value.

[0039] The compensation factor is specifically a correction coefficient calculated based on the spectral interference intensity and background fluorescence level, which is used to eliminate the influence of multi-element coexistence and matrix effect on the measurement of target element fluorescence intensity.

[0040] The jump factor is specifically a real-time correction parameter calculated based on the signal drift determinacy matrix, which is used to compensate for the signal shift caused by temperature changes or device aging during continuous operation of the detection system.

[0041] The spectral analysis function is used to process the multi-dimensional spectral data matrix and calculate the corresponding compensation parameters, the input includes the element correlation coefficient extracted from the spectral interference matrix, the intensity distribution data obtained from the background fluorescence interference matrix, the target detection element theoretical fluorescence wavelength position queried from the standard spectrum library, the matrix concentration parameter determined from the wastewater sample pretreatment process, and the instrument response coefficient obtained from the instrument calibration process, and the output is the compensation factor value and the corresponding correction weight coefficient for each target element.

[0042] The specific structure of the spectral line recognition model is a multi-level feature extraction and fusion system based on a hierarchical attention network architecture, including an input layer that receives the spectral data matrix and performs preprocessing normalization, multiple parallel convolution feature extraction branches for capturing spectral features of different scales, a hierarchical attention mechanism module that dynamically allocates the importance of different abstraction level features through hierarchical fusion weights, a feature fusion layer that combines multi-scale features by weighting, and a fully connected classification layer that outputs the class label and confidence score of the spectral outlier.

[0043] The steps of establishing the training data set of the spectral line recognition model specifically include collecting atomic fluorescence spectral data of different types of wastewater samples under various detection conditions as the basic data source, manually labeling each spectral data to distinguish normal fluorescence peaks and abnormal outliers, constructing synthetic spectral data containing known spectral overlap by adding different concentrations of interference elements and changing matrix components, adding noise and drift changes consistent with the actual detection environment based on the original spectral data using data enhancement techniques, classifying and organizing all labeled data according to spectral feature complexity and interference type, constructing a complete training sample set containing spectral data matrix and corresponding labels, and cross-validation segmentation of the training data set to ensure model generalization performance.

[0044] The steps of training the spectral line recognition model specifically include initializing the model parameters using the preprocessed training data set, adjusting the network weights step by step using the gradient descent optimization algorithm to minimize the prediction error, monitoring the model performance and preventing overfitting during the training process using the validation set data, optimizing the training efficiency and convergence stability by adjusting the learning rate and batch size, periodically saving the model parameter state with the best performance during the training process, evaluating the accuracy and robustness of the final trained model in the actual application scenario using the test set data, and quantizing and compressing the trained model to adapt to the computational resource limitations of the real-time detection system.

[0045] The weight adjustment function is used to adjust the hierarchical fusion weight parameters of the spectral line recognition model, and the weight adjustment function is based on comprehensive evaluation value calculation of spectral complexity evaluation value extracted from the spectral interference matrix, interference intensity measurement value obtained from the background fluorescence interference matrix, and system stability index calculated from the signal drift tendency matrix. When the comprehensive evaluation value is less than 0.414, a conservative weight distribution strategy is adopted to enhance the low-level feature weight to improve the basic spectral recognition accuracy, when the comprehensive evaluation value is in the range of 0.414 to 0.732, a balanced weight distribution strategy is adopted to evenly distribute the feature weight of each level to balance the recognition accuracy and calculation efficiency, and when the comprehensive evaluation value is greater than 0.732, an aggressive weight distribution strategy is adopted to enhance the high-level abstract feature weight to improve the spectral line recognition ability under complex interference conditions.

[0046] The two threshold values of the comprehensive evaluation value are obtained by collecting a large amount of atomic fluorescence spectrum detection data covering wastewater samples of different complexity as a basic data set for threshold determination, calculating the spectral complexity evaluation value for each detection sample to reflect the spectral line overlap degree, measuring the interference intensity measurement value to quantify the influence degree of background fluorescence on detection, evaluating the system stability index to represent the stability level of the instrument in the detection process, calculating the comprehensive evaluation value corresponding to each sample according to the equal weight method according to the above three indexes, sorting all samples according to the comprehensive evaluation value and dividing them into three complexity levels of low, medium and high, determining the comprehensive evaluation value of 0.414 as the dividing point between low and medium complexity samples through statistical analysis, determining the comprehensive evaluation value of 0.732 as the dividing point between medium and high complexity samples through statistical analysis, and verifying the effectiveness and stability of the two threshold values using an independent test sample set to ensure the rationality of the threshold setting.

[0047] The complexity index is specifically a comprehensive index of quantifying the complexity of the spectrum by calculating the proportion of the number of non-zero elements in the spectral interference matrix to the total number of elements, and combining the number of spectral line overlapping peaks and the peak intensity variance.

[0048] The background fluorescence interference intensity is specifically a weighted average value of the background signal intensity at each wavelength position extracted from the background fluorescence interference matrix, and the weight is determined according to the correlation between the fluorescence wavelength position of the target detection element and the background signal wavelength position.

[0049] The variance value is specifically a statistical quantity of signal change amplitude calculated from the diagonal elements of the signal drift tendency matrix, which is used to quantify the change degree of signal stability of the detection system in continuous operation.

[0050] The spectral complexity evaluation value is specifically a comprehensive complexity quantification index calculated based on the weighted combination of the complexity index and the spectral peak number and peak width parameters.

[0051] The interference intensity measurement value is specifically a ratio of the background fluorescence interference intensity to the theoretical fluorescence intensity of the target element, and is used to quantify the relative influence degree of background interference on the detection of the target element.

[0052] The system stability index is specifically a comprehensive calculation result based on the variance value and the instrument temperature fluctuation coefficient and the light source intensity variation coefficient, and is used to evaluate the overall stability level of the detection system under long-time running conditions.

[0053] The specific implementation of the above steps is described in detail below.

[0054] The specific implementation of step S01 is to use an atomic fluorescence spectrometer to perform scanning detection on the wastewater sample to obtain complete spectral data covering a wavelength range of 200-800 nm. First, the wastewater sample is pretreated, including filtration, dilution and acidification treatment, to ensure that the target elements in the sample can be fully atomized. Then, a standard mixed solution containing various target elements is prepared, with a concentration range of 0.1-100 μg / L, and the standard solution is repeatedly measured multiple times under the same detection conditions by the atomic fluorescence spectrometer. Based on the spectral superposition principle, when multiple elements coexist, the fluorescence spectral lines of some elements will overlap at the same or similar wavelength positions, causing mutual interference of spectral signals. A matrix analysis method is used to calculate the fluorescence intensity correlation coefficients between different elements at each wavelength position, and numerical analysis is used to determine the mutual influence degree between the elements. When the correlation coefficient is greater than 0.85, it is considered that there is significant spectral interference, which needs to be corrected in the subsequent processing. The established spectral interference matrix takes the target elements as the row and column indexes, and the matrix element value represents the spectral interference intensity between the corresponding element pairs. The matrix is a symmetric matrix, the diagonal element is 1, and the non-diagonal element value is between 0 and 1.

[0055] The specific implementation of step S02 is to use a statistical method to detect and identify outliers in the spectral data obtained by continuous multiple measurements. Based on the normal distribution assumption, the interquartile range method is used to determine the normal distribution range of the spectral data, and the first quartile and the third quartile of the spectral intensity at each wavelength position are calculated. The interquartile range is defined as When the spectral intensity value at a certain wavelength position is outside the range of or When the data point is outside the range, it is marked as an outlier. At the same time, a density-based clustering algorithm is used to analyze the spectral data and identify outliers that deviate from the main data distribution. This algorithm calculates the local density around each data point, and when the density is lower than the set threshold of 0.15, the point is considered an outlier. The constructed spectral outlier matrix is indexed by wavelength as row and measurement number as column, with matrix element values being binary labels of 0 or 1, with 1 indicating the presence of an outlier at that position and 0 indicating a normal value. The role of this matrix is to provide basic information for anomaly detection for subsequent intelligent spectral line recognition.

[0056] The specific implementation of step S03 is to measure the background fluorescence without adding any target analysis elements using the same matrix solution as the sample. The matrix solution contains all components in the wastewater sample except the target elements, such as inorganic salts, organic matter, and other coexisting elements. Based on the principles of Rayleigh scattering and Raman scattering, the matrix components will produce non-characteristic fluorescence signals under the action of excitation light, which are superimposed with the characteristic fluorescence signals of the target elements to form background interference. The matrix solution is measured in the full wavelength range using spectral scanning, and the fluorescence intensity values at each wavelength position are recorded. The average value and standard deviation of the background fluorescence intensity are obtained through multiple repeated measurements, and when the ratio of the standard deviation to the average value is less than 5%, the background measurement is considered stable and reliable. The established background fluorescence interference matrix is indexed by wavelength as row and matrix type as column, with matrix element values being the background fluorescence intensity under the corresponding conditions. This matrix is used to quantify the background interference level under different matrix conditions and provides accurate reference data for subsequent background subtraction.

[0057] The specific implementation of step S04 is to establish a signal drift prediction model using time series analysis method through long-term continuous monitoring of the signal stability of the detection system. A standard reference sample is detected every 2 hours, and the continuous monitoring time is not less than 48 hours, obtaining the data sequence of signal change over time. Based on the principle of autoregressive moving average model, the trend and periodicity characteristics of signal change are analyzed. The least squares method is used to fit the linear trend of signal change, and the trend slope and correlation coefficient are calculated. When the correlation coefficient is greater than 0.8, it is considered that there is a significant signal drift trend. At the same time, the sliding window technique is used to analyze the short-term fluctuation characteristics of the signal, with a window size of 10 data points. The variance and skewness of the signal in each window are calculated. The constructed signal drift tendency matrix is indexed by time as row and detection parameter as column, with matrix element values representing the signal drift tendency index under the corresponding time and parameter. The diagonal elements of this matrix reflect the variance characteristics of signal change, and the non-diagonal elements reflect the correlation between different parameters.

[0058] The specific implementation of step S05 is to establish an accurate signal drift correction model based on real-time calibration data. A standard reference material with a known concentration is used as a calibration sample, and a measurement is taken every 30 minutes. The measurement result is compared with the theoretical value, and the relative deviation is calculated. The weighted least squares method is used to fit the deviation data, and the weight coefficient is determined according to the inverse of the measurement time, so that the recent data has a higher weight. Based on the principle of Kalman filter algorithm, a state space model of signal drift is established to estimate the drift state of the system in real time. Through the iterative process of the prediction step and the update step, the algorithm can accurately estimate the amplitude and direction of signal drift in a noisy environment. When the covariance matrix of the prediction error converges, it is considered that the drift estimation has reached a stable state. The constructed signal drift certainty matrix takes the detection elements as the row index and the drift parameters as the column index, and the matrix element value is the drift correction coefficient of the corresponding element and parameter. This matrix provides accurate numerical correction parameters for real-time correction of the influence of system drift on the detection results.

[0059] The specific implementation of step S06 is to use a multiple linear regression algorithm to process spectral interference and background interference data to calculate the corresponding compensation factors. The spectral analysis function takes the spectral interference matrix and the background fluorescence interference matrix as input parameters, while considering the element correlation coefficient, the background intensity distribution, the theoretical fluorescence wavelength, the matrix concentration and the instrument response coefficient. Based on the extended form of Lambert-Beer's law, a spectral response model under the condition of multiple element coexistence is established. The principal component analysis method is used to reduce the dimension of the input parameters, extract the main influencing factors, and reduce the calculation complexity. The singular value decomposition algorithm is used to solve the multiple linear equations to obtain the compensation factors of each element. When the spectral interference coefficient is greater than 0.5, the compensation factor takes a value in the range of 0.8-1.2, which is used to enhance or weaken the signal strength of the corresponding element. When the ratio of background interference intensity to target signal strength is greater than 0.3, the background subtraction algorithm is used for correction. The calculated compensation factor array contains the correction coefficient and the corresponding weight parameter of each target element, which is used to eliminate the influence of multiple element coexistence and matrix effect on the detection results.

[0060] The specific implementation of step S07 is to intelligently analyze the spectral outliers using a deep learning-based spectral line recognition model, and to calculate the jump factor in combination with the signal drift certainty matrix. The spectral line recognition model receives the spectral outlier matrix as input, extracts spectral features through a multi-layer neural network structure, and performs classification and recognition. The model uses a convolutional neural network architecture that can automatically learn local features and global patterns of spectral data. Based on the attention mechanism principle, the model can adaptively focus on important spectral regions to improve the accuracy of anomaly detection. The calculation of the jump factor is based on real-time data of the signal drift certainty matrix, and a moving average filtering algorithm is used to smooth the signal fluctuations. When a signal jump is detected, the jump factor takes a value in the range of 0.9-1.1, which is used to quickly correct sudden signal shifts. The final output of the wastewater pollutant concentration is obtained by multiplying the original detection result by the compensation factor and the jump factor, and the calculation formula is: the corrected concentration equals the original concentration multiplied by the compensation factor and then multiplied by the jump factor. This step realizes the comprehensive correction of various interference factors, ensuring the accuracy and reliability of the detection results.

[0061] The spectral line recognition model uses a deep learning architecture based on a hierarchical attention network, and the specific structure includes an input layer, a feature extraction layer, an attention layer, a fusion layer, and an output layer. The input layer receives a spectral data matrix with dimensions of wavelength number x measurement times, and first performs data normalization to scale the spectral intensity values to the range of 0-1. The feature extraction layer uses multiple parallel one-dimensional convolution branches with convolution kernel sizes of 3, 5, and 7, a step size of 1, and a same-size padding method. Each convolution branch is followed by a batch normalization layer and a rectified linear unit activation function to extract spectral features of different scales. The attention layer uses a self-attention mechanism to calculate attention weights through query, key, and value linear transformation matrices, with the weight calculation based on the similarity between feature vectors. The hierarchical attention mechanism includes local attention and global attention at two levels, with local attention focusing on the feature correlation of adjacent wavelength regions and global attention capturing long-distance dependencies in the entire spectral range. The fusion layer combines multi-scale features with weights that are dynamically adjusted according to spectral complexity, background interference intensity, and system stability indicators. The output layer includes two fully connected sub-layers, with the first sub-layer outputting class probabilities of outliers and the second sub-layer outputting confidence scores.

[0062] The training data set establishment includes four stages of data collection, labeling, enhancement and verification. The data collection stage collects samples from different types of industrial wastewater, domestic sewage and agricultural wastewater, covering various types of pollutants such as heavy metals, organic matter and inorganic salts. Each sample is measured under different detection conditions, including different excitation power, integration time and detector gain settings. The labeling stage is manually labeled by experienced spectral analysis experts for each spectral data, identifying normal fluorescence peaks and abnormal outliers, with an accuracy requirement of more than 95%. The data enhancement stage expands the training samples by adding Gaussian noise, baseline drift and peak shift, with noise intensity set to 5%-15% of signal intensity, and baseline drift amplitude set to 10%-30% of average signal intensity. At the same time, synthetic data containing known spectral overlap conditions are constructed, and composite spectra are generated by linearly superimposing standard spectra of different elements. In the verification stage, the data set is divided into training set, verification set and test set according to the ratio of 7:2:1, and the stratified sampling method is used to ensure the balanced proportion of different types of samples in each sub-set. The finally established training data set contains more than 10,000 labeled samples, covering 50 common wastewater pollution elements and 100 different matrix combinations.

[0063] As shown in Figure 3 , the characteristics of normal fluorescence peaks include: symmetrical Gaussian distribution peak shape; peak position consistent with theoretical fluorescence wavelength; peak intensity linearly related to element concentration; stable baseline and good signal-to-noise ratio; good reproducibility of multiple measurements.

[0064] As shown in Figure 4 , the characteristics of abnormal outliers include: irregular peak shape (asymmetric, double peak, tailing); peak position deviates from the theoretical value; abnormal peak intensity (too high or too low); unstable baseline with drift; severely affected by spectral interference and noise.

[0065] It should be noted that first, the present application proposes a multi-dimensional modeling method of spectral interference matrix, background fluorescence interference matrix and signal drift matrix, which has significant technical advantages over traditional single interference factor correction methods. Traditional methods usually only consider one type of interference in spectral overlap or background subtraction, while the present application can simultaneously quantify multiple interference factors such as spectral line overlap between elements, background fluorescence caused by matrix effect and signal drift caused by long-term operation of the instrument by establishing a three-dimensional interference model. This multi-dimensional matrix modeling method is based on linear algebra theory, which converts complex spectral interference relationships into calculable mathematical matrix form, so that interference correction is transformed from empirical qualitative analysis to precise quantitative calculation, thereby significantly improving the accuracy and stability of multi-element simultaneous detection under complex matrix conditions.

[0066] Secondly, the spectral line recognition model of the present application adopts a hierarchical attention network architecture, which has a revolutionary technological breakthrough compared with the traditional abnormality detection based on threshold judgment or simple statistical method. The traditional method relies on manually set fixed thresholds or simple statistical distribution assumptions to identify spectral anomalies, which is easily affected by the complexity of the matrix and environmental changes, resulting in misjudgment. However, the deep learning model of the present application can automatically learn the complex feature patterns of spectral data, and through multi-scale convolution feature extraction and hierarchical attention mechanism, it can intelligently identify normal fluorescence peaks and abnormal outliers. Based on the deep learning principle of large-scale training data, the model has strong feature representation ability and generalization performance, and can adapt to the spectral feature changes of different types of wastewater samples, achieving more accurate and robust spectral line recognition effect.

[0067] Thirdly, the present application innovatively proposes a dynamic weight adjustment mechanism based on comprehensive evaluation value, which adaptively adjusts the model parameters according to the real-time changes of spectral complexity, interference intensity and system stability, which has a significant adaptive advantage compared with the traditional fixed parameter setting method. The traditional method usually uses pre-set fixed weight parameters for spectral analysis, which cannot be dynamically adjusted according to the changes of actual detection conditions, resulting in large performance differences in the detection of samples with different complexity. The present application can intelligently select the optimal parameter configuration according to the complexity of the current detection condition by establishing three weight distribution strategies: conservative, balanced and aggressive, ensuring high precision in simple matrix conditions and high robustness in complex interference environment.

[0068] The above three key technical ideas form an organic synergistic system, producing a comprehensive technical effect beyond single technical improvement. Multi-dimensional matrix modeling provides accurate interference feature input for intelligent spectral line recognition, enabling the deep learning model to train and reason based on more comprehensive and accurate data. The dynamic weight adjustment mechanism optimizes the model parameters in real time according to the complexity index obtained by matrix modeling, ensuring the accuracy of spectral line recognition under different detection conditions. The results of intelligent spectral line recognition in turn verify and optimize the effectiveness of matrix modeling, forming a closed-loop self-improvement mechanism. This synergistic effect enables the entire detection system to have intelligent, adaptive and high-precision characteristics that traditional methods cannot achieve, maintaining stable and reliable performance in complex and variable actual wastewater detection environments, realizing a technical transformation from traditional passive correction to active prediction and intelligent optimization.

[0069] Specifically, the principle of the present application is that the technical solution of the present application can solve the fundamental problem of multiple interference coupling by establishing a complete interference factor quantification characterization system and intelligent correction mechanism. By constructing five core matrices, mathematical modeling and accurate description of various interference sources in the detection process are realized.

[0070] The spectral interference matrix establishes the mathematical relationships between elemental interactions by measuring a standard mixed solution, transforming the complex problem of spectral line overlap into a computable matrix operation. The background fluorescence interference matrix quantifies the distribution of non-characteristic signals in the sample matrix at various wavelengths, providing an accurate reference for background subtraction. The synergistic effect of these two matrices, through the calculation of a compensation factor using a spectral analytical function, mathematically eliminates the influence of spectral overlap and matrix effects.

[0071] Solving the signal drift problem relies on a dual mechanism of propensity matrix and deterministic matrix. The propensity matrix establishes a statistical model of signal changes through long-term monitoring, enabling prediction of future drift trends; the deterministic matrix calculates the current drift state based on real-time calibration data, providing accurate correction parameters. This strategy, combining prediction and real-time correction, ensures the stability of the system during long-term operation.

[0072] The intelligent spectral line identification model employs a hierarchical attention network architecture, enabling automatic extraction of spectral features at different abstraction levels and dynamic weight allocation. This model captures local and global features of spectral data through multi-scale feature extraction, and the hierarchical attention mechanism adaptively adjusts feature weights based on the complexity of interference, achieving accurate identification of outliers. The weight adjustment function uses different weight allocation strategies based on a comprehensive evaluation value to ensure optimal model performance under various complexity conditions.

[0073] The calculation of the jump factor integrates outlier identification results and signal drift information, achieving unified processing of sudden interference and gradual drift. This mechanism ensures the continuity and consistency of detection results through real-time monitoring and dynamic correction.

[0074] The technical logic of this invention conforms to the scientific principles of interference source identification, quantitative modeling, and collaborative correction. By transforming complex physicochemical interference processes into mathematical models and intelligent algorithms, it achieves a technological leap from qualitative analysis to quantitative correction, providing a systematic solution for atomic fluorescence spectrometry detection.

[0075] The following provides a specific embodiment 1 of the present invention, and the specific implementation of each step in this embodiment 1 is described in detail below.

[0076] In step S01, the construction of the spectral interference matrix is ​​specifically represented as follows:

[0077] ;

[0078] In the formula, This is the spectral interference matrix; For the first The element pair Spectral interference coefficients of the elements; Let be the total number of elements to be detected. The formula for calculating the spectral interference coefficient between elements is:

[0079] ;

[0080] In the formula, For the first The element in the first Fluorescence intensity at each wavelength position; For the first The element in the first Fluorescence intensity at each wavelength position; This represents the number of wavelength points in the spectral scan. This formula is based on the Pearson correlation coefficient principle, quantifying the degree of spectral interference by calculating the correlation between the spectral signals of different elements.

[0081] In step S02, the construction of the spectral outlier matrix is ​​specifically represented as follows:

[0082] ;

[0083] In the formula, This is the matrix of spectral outliers; For the first The wavelength at the ... Outlier markers in this measurement are set to either 0 or 1; Number of wavelength points; The number of measurements is [number]. The formula for outlier detection is:

[0084] ;

[0085] In the formula, For the first The wavelength at the ... The fluorescence intensity measured in the second measurement; It is the first quartile; It is the third quartile; The interquartile range is calculated by sorting and dividing the data from multiple measurements at the same wavelength position.

[0086] In step S03, the construction of the background fluorescence interference matrix is ​​specifically represented as follows:

[0087] ;

[0088] In the formula, Background fluorescence interference matrix; For the first The wavelength at the ... Background fluorescence intensity under different matrix types; Number of wavelength points; The number of matrix types is the number of matrix types. The background fluorescence intensity is obtained by spectral measurement of a matrix solution without target elements.

[0089] In step S04, the construction of the signal drift tendency matrix is specifically represented as:

[0090] ;

[0091] In the formula, is the signal drift tendency matrix; is the signal drift index of the i-th detection parameter at the j-th time node; is the signal drift index of the i-th detection parameter at the j-th time node; is the number of monitoring time nodes; is the number of detection parameters. The calculation formula of the signal drift index is:

[0092] ; In the formula,

[0093] is the signal value of the i-th parameter at the j-th time node; is the signal change rate, obtained by the slope of the least squares linear fitting of the signal value sequence; is the signal variance, obtained by calculating the variance of the signal value in the sliding window; is the signal skewness, representing the degree of asymmetry of the signal distribution; is the weight coefficient, respectively taking values of 0.5, 0.3, 0.2. The signal change rate is obtained by least squares fitting, and the variance and skewness are obtained by statistical calculation. In step S05, the construction of the signal drift determination matrix is specifically represented as:

[0094] ;

[0095] ;

[0096] In the formula, is the signal drift determination matrix; is the correction coefficient of the i-th drift parameter of the j-th element; is the number of detection elements; is the number of drift parameters. The correction coefficient is calculated based on the Kalman filter algorithm, and the state equation is: ;

[0097] The observation equation is:

[0098]

[0099] ; ​​​​

[0100] wherein, is the state vector of the system at the time t, which contains the signal drift amount and drift rate; is the state transition matrix, which describes the time evolution law of the system state; is the process noise, which is subject to Gaussian distribution; is the observation value at the time t, which is the actual measured standard sample signal; is the observation matrix, which establishes the linear relationship between the state vector and the observation value; is the observation noise, which is subject to Gaussian distribution. The state vector contains the signal drift amount and drift rate, and the optimal estimation value is obtained by recursive calculation. In step S06, the calculation of the compensation factor is specifically represented as:

[0101]

[0102]

[0103] wherein, is the compensation factor of the element; is the spectral interference coefficient; is the relative concentration ratio of the element to the element; is the background interference weight coefficient; is the background interference intensity of the element at the wavelength. The calculation formula of the background interference intensity is:

[0104]

[0105] wherein, is the weight of the wavelength position to the element, which ranges from 0 to 1; is the background fluorescence intensity of the wavelength position, which is obtained by measuring the matrix solution without the target element. The weight is determined according to the correlation between the target element fluorescence wavelength and the background signal wavelength, and the higher the correlation, the greater the weight.

[0106] In step S07, the calculation of the jump factor is specifically represented as:

[0107]

[0108] wherein, is the jump factor of the element; is the correction coefficient in the signal drift certainty matrix; is the correction coefficient in the signal drift certainty matrix; is the jump factor of the element.​​​​​​​​​​​​ the time interval, in minutes; is the number of drift parameters. The final formula of the wastewater pollutant concentration is:

[0109]

[0110] wherein, is the corrected pollutant concentration; is the original detected concentration.

[0111] The calculation of the comprehensive evaluation value in the weight adjustment function is specifically represented as:

[0112]

[0113] wherein, is the comprehensive evaluation value; is the spectral complexity evaluation value; is the interference intensity measurement value; is the system stability index. The calculation formula of the spectral complexity evaluation value is:

[0114]

[0115] wherein, is the complexity index, which is calculated by the proportion of the number of non-zero elements in the total number of elements in the spectral interference matrix; is the number of spectral peaks, which is identified by the peak detection algorithm; is the peak width parameter, which represents the average half-peak width of the spectral peak; is the weight coefficient, which is respectively 0.5, 0.3, and 0.2. The complexity index is obtained by calculating the proportion of non-zero elements in the total elements in the spectral interference matrix. The calculation formula of the interference intensity measurement value is:

[0116]

[0117] wherein, is the theoretical fluorescence intensity of the target element, which is obtained from the standard spectral library. The calculation formula of the system stability index is:

[0118]

[0119] wherein, is the variance value of the diagonal elements of the signal drift tendency matrix, which reflects the signal change amplitude; is the temperature fluctuation coefficient, which is calculated by monitoring the environmental temperature change during the detection process; is the light source intensity change coefficient, which is calculated by monitoring the excitation light source intensity fluctuation, all of which are obtained by experimental measurement. ​​​​​

[0120] It should be noted that the threshold values 0.414 and 0.732 are obtained based on statistical analysis of large-scale sample data. First, atomic fluorescence spectrum detection data covering wastewater samples of different complexity levels are collected as a basic data set, including industrial wastewater, domestic sewage, agricultural wastewater and other types, with a total number of not less than 5000 samples. For each detection sample, the spectral complexity evaluation value is calculated according to the above formula , the interference intensity measurement value , and the system stability index , and then the comprehensive evaluation value is calculated. The spectral complexity evaluation value is calculated by analyzing the sparsity of the spectral interference matrix, the number of spectral peaks and the peak width parameters, the interference intensity measurement value is obtained by comparing the ratio of the background fluorescence intensity and the theoretical fluorescence intensity of the target element, and the system stability index is calculated by comprehensively considering factors such as signal variance, temperature fluctuation and light source intensity variation. The comprehensive evaluation values of all samples are arranged in order from small to large to form a complete numerical sequence , wherein is the total number of samples.

[0121] The calculation process of the threshold value adopts a statistical segmentation method based on the complexity distribution characteristics of the samples. The three complexity levels of low, medium and high are determined by dividing the comprehensive evaluation value sequence into three parts. The specific calculation formula is: the first threshold value , the second threshold value , wherein represents the floor function. In order to verify the rationality of the threshold value setting, the clustering effectiveness index is used for evaluation, and the calculation formula is the silhouette coefficient , wherein is the average distance between sample and other samples of the same category, is the average distance between sample and the nearest sample of different categories. Through multiple iteration optimization and cross-validation, the threshold values that can make the clustering effectiveness index optimal are finally determined as 0.414 and 0.732. These two threshold values can effectively distinguish between simple matrix, medium complex matrix and high complex matrix, providing accurate judgment basis for the subsequent weight adjustment strategy selection, and ensuring that the spectral line recognition model can maintain optimal recognition performance under different complexity conditions.

[0122] In order to better understand and implement the present application, the following provides an embodiment 2 of a specific application scenario of the present application: researchers use the atomic fluorescence spectrometry of the present application to detect pollutants in the comprehensive wastewater of an industrial park. The wastewater contains various heavy metal ions including , , , 、 etc. The matrix composition is complex and the interference is serious. First, the wastewater sample is pretreated, filtered through a 0.45 μm membrane, diluted 10 times with 5% nitric acid solution, and then scanned with a full spectrum in the wavelength range of 200-800 nm using an atomic fluorescence spectrometer. The integration time is set to 5 s, the excitation power is 80 W, and the detector gain is 1.2 x . The spectral interference matrix is constructed by measuring a standard mixed solution containing 5 target elements, with the concentration of each element set to 10 μg / L, and 20 repeated measurements are performed under the same detection conditions.

[0123] The spectral interference matrix calculated based on the measurement data is shown in Table 1:

[0124] Table 1 Spectral interference matrix data table

[0125]

[0126] As can be seen from Table 1, and there is strong spectral interference between them, with a correlation coefficient of 0.256, and the interference coefficient is 0.187, which needs to be corrected in subsequent analysis. By statistically analyzing the spectral data of 100 consecutive measurements, the abnormal value distribution at each wavelength position is identified, and a spectral outlier matrix is constructed. The four-quartile method is used to determine the abnormal value judgment standard, and when the spectral intensity deviates from the normal range by 1.5 times the four-quartile range, it is marked as an outlier. The statistical results show that obvious outlier signals are detected at key wavelength positions such as 253.7 nm, 283.3 nm, and 368.3 nm, with outlier frequencies of 8%, 12%, and 6%, respectively.

[0127] In order to quantify the influence of background interference, the researchers prepared a blank solution with the same matrix composition as the wastewater sample but without the target heavy metal elements. The background fluorescence interference data measured are shown in Table 2:

[0128] Table 2 Background fluorescence interference intensity data table

[0129]

[0130] Table 2 shows the detection of the element is most seriously interfered by the background, with an interference intensity ratio of 0.42, while The background interference of elements was relatively small. Based on the 48-hour continuous monitoring data, a signal drift tendency matrix was established, and the signal change trend was recorded every 2 hours using a standard reference sample for detection. Statistical analysis results showed that the system had a significant linear drift trend, with a drift rate of -0.03% / h, and periodic signal fluctuations were detected, with a fluctuation amplitude of about ±2.1%.

[0131] By comparing the measured results of the real-time calibration standard sample with the theoretical value, a signal drift certainty matrix was established. The Kalman filter algorithm was used to estimate the drift parameters in real time, and the state vector included two components of drift and drift rate. The process noise covariance was set to 0.001, and the observation noise covariance was set to 0.01. The drift correction parameters obtained after the filter converged are shown in Table 3:

[0132] Table 3 Signal drift correction parameter table

[0133]

[0134] Based on the spectral interference matrix and the background fluorescence interference matrix, a multivariate linear regression algorithm was used to calculate the compensation factor of each element. The compensation factor of element was 0.934, mainly used to correct the spectral interference from and . Element was strongly interfered by , and the compensation factor was 0.847. Element was most seriously affected by background interference, and the compensation factor was 0.756. The trained spectral line recognition model was used for intelligent analysis of the spectral outlier matrix. The model used a hierarchical attention network architecture, including 3 parallel convolution branches and 2 layers of attention mechanism. The abnormal spectral features identified by the model were mainly concentrated in the high interference intensity area, with a confidence score of more than 0.85.

[0135] According to the calculated spectral complexity evaluation value 0.567, interference intensity measurement value 0.324 and system stability index 0.289, the comprehensive evaluation value was 0.393, which was less than the first threshold value 0.414. Therefore, the conservative weight allocation strategy was adopted, and the low-level feature weight was enhanced to 0.6, the middle-level feature weight was 0.3, and the high-level feature weight was 0.1. After comprehensive correction, the final detection results showed that the concentration of in wastewater was 15.7 μg / L, was 28.3 μg / L, was 45.9 μg / L, was 12.4 μg / L, was 67.2 μg / L.

[0136] To verify the accuracy of the detection results, the researchers used inductively coupled plasma mass spectrometry for comparison and verification, and the relative deviation was controlled within 8%. The reliability of the method was further verified by the standard addition recovery experiment, and the recoveries of the five elements were 97.3%, 95.8%, 98.1%, 96.7% and 94.5%, respectively, which met the analysis requirements. The detection limit calculation results showed that after interference correction, the detection limit of each element was reduced by 12% to 18% compared with that before correction, among which The detection limit of the element improved the most, from 2.8 ug / L to 2.3 ug / L.

[0137] The traditional atomic fluorescence spectrometry detection method mainly uses matrix matching and standard addition method to deal with the matrix interference problem, adjusts the instrument parameters manually and selects appropriate detection conditions to reduce spectral overlap interference, and the signal drift problem is usually handled by regular calibration and manual experience judgment. These traditional methods have the disadvantages of complex operation, strong subjectivity, poor adaptability, etc., especially in the face of complex matrix and multiple elements coexisting, often cannot obtain ideal detection precision and stability. The present application realizes the accurate quantification of various interference factors by establishing a multi-dimensional interference matrix model, which improves the detection precision by about 15% compared with the traditional method, significantly reduces the influence of matrix effect and spectral overlap on the detection results. The intelligent spectral line recognition technology can automatically identify and process spectral anomalies, avoiding the subjective error of manual judgment in traditional methods, and the accuracy of abnormal detection is improved by about 18%. The dynamic weight adjustment mechanism optimizes the analysis parameters according to the actual detection conditions, and compared with the traditional method with fixed parameter setting, the stability in different complexity sample detection is improved by about 12%, effectively solving the technical problem of poor adaptability of traditional method.

[0138] The above is only a specific embodiment of the present application, but the protection scope of the present application is not limited thereto, any skilled person in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application.

Claims

1. A method for rapid detection of pollutants in wastewater using atomic fluorescence spectrometry, characterized in that, include: Atomic fluorescence spectrometry data of wastewater samples are collected to establish a spectral interference matrix to characterize the overlap of spectral lines among multiple elements. Outliers in the spectral data are identified through statistical analysis to construct a spectral outlier matrix to record the distribution of abnormal fluorescence intensity at each wavelength. Non-characteristic fluorescence signals generated by the sample matrix are measured to establish a background fluorescence interference matrix to quantify the background interference intensity at different wavelengths. The signal change pattern of the detection system during long-term operation is continuously monitored to construct a signal drift tendency matrix to predict the signal shift trend. Based on historical detection data and current measurement conditions, a signal drift deterministic matrix is ​​established to accurately describe the amplitude and direction of signal drift. The spectral interference matrix and background fluorescence interference matrix are processed by spectral analytical functions to calculate compensation factors to correct the influence of spectral overlap and background interference on the measurement results. A spectral line identification model is used to analyze the spectral outlier matrix, and the jump factor is calculated using the signal drift deterministic matrix to correct system drift in real time and output the final wastewater pollutant concentration detection results. The spectral interference matrix is ​​specifically formed by measuring the fluorescence spectra of each target element in a standard mixed solution, calculating the degree of mutual influence between different elements at the same wavelength position, and creating a mathematical matrix describing the overlapping interference relationship of spectral lines between elements. The spectral outlier matrix is ​​specifically a two-dimensional data matrix constructed by statistically analyzing spectral data from multiple consecutive measurements to identify abnormal data points that deviate from the normal fluorescence intensity range, based on wavelength position and frequency of occurrence. The background fluorescence interference matrix is ​​specifically formed by measuring the intensity of non-characteristic fluorescence signals generated by the sample matrix at various detection wavelengths without adding target analytical elements, thus creating a data matrix describing the distribution pattern of background interference. The signal drift tendency matrix is ​​specifically established by long-term monitoring of the signal stability of the detection system, statistical analysis of the trend and pattern of signal changes over time, and prediction matrix for predicting the future signal drift direction and amplitude. The signal drift deterministic matrix is ​​specifically a matrix that accurately calculates the specific value of the system signal drift and the correction parameters under the current detection conditions based on the deviation between the measurement results of the real-time calibration standard sample and the theoretical value. The compensation factor is specifically a correction coefficient calculated based on the intensity of spectral interference and the background fluorescence level, used to eliminate the influence of multi-element coexistence and matrix effects on the fluorescence intensity measurement of the target element. The jump factor is a real-time correction parameter calculated based on the deterministic matrix of signal drift, used to compensate for signal offset caused by temperature changes or device aging during continuous operation of the detection system. The spectral analysis function is used to process multidimensional spectral data matrices and calculate corresponding compensation parameters. The inputs include element correlation coefficients extracted from the spectral interference matrix, intensity distribution data obtained from the background fluorescence interference matrix, theoretical fluorescence wavelength positions of target detection elements queried from the standard spectral library, matrix concentration parameters measured during wastewater sample pretreatment, and instrument response coefficients obtained during instrument calibration. The output is the compensation factor value and corresponding correction weight coefficient for each target element. The specific structure of the spectral line identification model is a multi-level feature extraction and fusion system based on a hierarchical attention network architecture. It includes an input layer that receives the spectral data matrix and performs preprocessing and normalization; multiple parallel convolutional feature extraction branches that capture spectral features at different scales; a hierarchical attention mechanism module that dynamically allocates the importance of features at different abstract levels through hierarchical fusion weights; a feature fusion layer that weights and combines features at multiple scales; and a fully connected classification layer that outputs the category labels and confidence scores of spectral outliers.

Images