A method and system for rapid identification of new pollutants in water bodies based on multifunctional sensing and AI

Abstract

This application relates to a method and system for rapid identification of new pollutants in water bodies based on multifunctional sensing and AI. The method includes: acquiring a synchronously collected raw signal dataset and extracting features from the raw signal dataset to obtain a multidimensional feature fingerprint array; inputting the multidimensional feature fingerprint array into a pre-trained classification model to obtain a list of major pollutant categories; performing concentration and morphology analysis on the pollutants in the list of major pollutant categories to obtain a pollutant morphology report; based on the multidimensional feature fingerprint array, the list of major pollutant categories, and the pollutant morphology report, dynamically inferring the coexistence relationships of the pollutants to obtain pollutant interaction behavior conclusions; and integrating the list of major pollutant categories, the pollutant morphology report, and the pollutant interaction behavior conclusions to obtain a pollutant identification report. This method can accurately identify the phase state of new pollutants in water bodies, improving pollution identification capabilities.

Description

Technical Field

[0001] This invention belongs to the field of water pollution monitoring, and in particular relates to a method and system for rapid identification of new pollutants in water bodies based on multifunctional sensing and AI. Background Technology

[0002] With the continuous development of water environment monitoring technology, various new pollutant screening technologies based on rapid sensing principles have emerged. These technologies are characterized by fast response speed, in-situ deployment, and the ability to provide continuous data, offering new possibilities for real-time monitoring and early warning of water bodies. This has spurred a shift in monitoring methods from laboratory-based "post-event tracing" to on-site "real-time sensing." Traditional technologies rely primarily on periodic on-site sampling, complex laboratory pretreatment, and offline analysis using sophisticated instruments for identifying and assessing the risks of new pollutants. This approach provides authoritative and standardized point data for environmental management and scientific research by acquiring precise chemical structures and concentrations of pollutants. However, current rapid sensing methods and traditional laboratory methods each have their limitations. Traditional methods are time-consuming and costly, and the data obtained lags significantly behind the dynamic development of pollution events. They fail to capture the rapidly changing phase migrations, morphological transformations, and interactions between multiple components of pollutants in complex aquatic environments, resulting in static blind spots in the assessment of their true environmental behavior and ecological risks. Existing rapid sensing methods are mostly independent measurements of single indicators or limited targets, making it difficult to simultaneously analyze the dynamic interaction network formed when multiple pollutants coexist, and even more difficult to quantify their phase distribution and transformation pathways. As a result, monitoring results often remain at the level of concentration alarms, making it difficult to support the understanding of pollution process mechanisms and accurate risk assessment. Summary of the Invention

[0003] Therefore, it is necessary to provide a rapid identification method and system for new water pollutants based on multifunctional sensing and AI, which can capture the rapidly changing phase migration, morphological transformation and multi-component interactions of pollutants in complex aquatic environments, thereby improving the ability to identify pollutants.

[0004] Firstly, this application provides a method for rapid identification of new pollutants in water bodies based on multifunctional sensing and AI, including: Acquire the original signal dataset acquired synchronously, and extract features from the original signal dataset to obtain a multidimensional feature fingerprint array; By inputting the multidimensional feature fingerprint array into the pre-trained classification model, a list of major pollutant categories is obtained; Concentration and speciation analysis were performed on pollutants in the list of major pollutant categories to obtain a pollutant speciation report; Based on a multidimensional feature fingerprint array, a list of major pollutant categories, and a pollutant morphology report, dynamic reasoning is performed on the coexistence relationships of pollutants to obtain conclusions on pollutant interaction behaviors. By integrating the list of major pollutant categories, pollutant speciation reports, and pollutant interaction behavior conclusions, a pollutant identification report is obtained; the pollutant identification report is used to characterize pollutant types and pollution intensity.

[0005] Furthermore, feature extraction is performed on the original signal dataset to obtain a multidimensional feature fingerprint array, including: Outliers in the original signal dataset are removed, and missing values ​​in the original signal dataset are filled in to obtain a multi-channel signal time series. Based on the environmental parameters in the original signal dataset, temperature compensation correction is performed on the multi-channel signal time series to obtain the corrected signal time series. Based on the time series of the calibrated signal, each signal stream is converted into a standard physical quantity to obtain multidimensional time series data; the signal streams include dynamic light scattering signals, multispectral signals and electrochemical signals. Statistical feature calculations are performed on multidimensional time series data to obtain a primary fusion feature vector; The initial fusion feature vector and environmental parameters are normalized to obtain a multidimensional feature fingerprint array.

[0006] Furthermore, based on the time series of the corrected signal, each signal stream is transformed into a standard physical quantity to obtain multidimensional time series data, including: Based on the time series of the correction signal, the photon autocorrelation function of the dynamic light scattering signal is inverted to obtain the particle size distribution parameters. The intensity value of each wavelength in the multispectral signal is converted into an absorbance value using the following formula, resulting in a multi-wavelength absorbance array:

[0007] in, To be at wavelength The absorbance value at that point For the sample at wavelength Measuring light intensity at that location, For reference at wavelength Measuring light intensity at that location, For dark current at wavelength The light intensity at that location was measured. Peak detection is performed on the electrochemical signal to identify the peak potential and peak current of the redox peak, and the electrochemical impedance spectrum is fitted with a preset equivalent circuit model to obtain the charge transfer resistance and double layer capacitance. By integrating particle size distribution parameters, multi-wavelength absorbance arrays, peak potential, peak current values, charge transfer resistance, double-layer capacitance, and auxiliary parameters, multidimensional time series data is obtained.

[0008] Furthermore, concentration and speciation analysis were performed on pollutants in the major pollutant category list to obtain a pollutant speciation report, including: Based on the pollutant feature association mapping table and the list of major pollutant categories, corresponding analytical features are extracted for each pollutant to obtain the pollutant feature vector mapping table. Based on the pollutant feature vector mapping table, the total concentration of pollutants was analyzed by multiple regression analysis to obtain a quantitative result table; Based on the pollutant feature vector mapping table and multidimensional feature fingerprint array, the proportion of pollutant phase distribution is estimated to obtain a phase distribution table; the phase distribution includes dissolved phase, colloidal phase and particulate phase; By integrating the quantitative results table and the phase distribution table, a pollutant speciation report is obtained.

[0009] Furthermore, based on the pollutant feature vector mapping table and the multidimensional feature fingerprint array, the proportion of pollutant phase distribution is estimated to obtain a phase allocation table, including: The distribution characteristics of various particulate matter in water bodies are extracted from a multidimensional feature fingerprint array to obtain the water body distribution characteristics. The water body distribution characteristics include the mass concentration estimation and particle size distribution characteristics of the dissolved phase, colloidal phase and particulate phase. Based on the water body distribution characteristics, the interaction strength between pollutants in the list of major pollutant categories and different phase distributions is queried to obtain the allocation coefficient matrix; Based on the total concentration, water body distribution characteristics, and distribution coefficient matrix, the concentration distribution of pollutants in each phase is calculated using the following formula:

[0010]

[0011]

[0012] in, Let i be the mass fraction of contaminant i in the dissolved phase. Let be the mass fraction of contaminant i in the colloidal phase. Let be the mass fraction of pollutant i in the particulate phase, and let i be the pollutant index. Let i be the linear partition coefficient of contaminant i between the dissolved and colloidal phases. Let i be the linear partition coefficient of contaminant i between the dissolved phase and the particulate phase. This represents the total concentration of the colloidal phase. This represents the total concentration of the particulate phase. Based on the pollutant feature vector mapping table, the concentration distribution is corrected to obtain the phase allocation table.

[0013] Furthermore, based on a multidimensional feature fingerprint array, a list of major pollutant categories, and a pollutant speciation report, dynamic reasoning is performed on the coexistence relationships of pollutants to obtain conclusions about pollutant interaction behaviors, including: Based on the multidimensional feature fingerprint array and the list of major pollutant categories, the statistical correlation between the features of different pollutants is calculated to obtain the time-lag correlation coefficient matrix. Using pollutants in the list of major pollutant categories as nodes, connecting edges are constructed based on the time-lag correlation coefficient matrix. Corresponding data is extracted from the pollutant morphology report as the attributes of the nodes to generate a pollutant interaction network graph. Based on the pollutant interaction network graph, path search is performed to infer the transformation path of pollutants and obtain a list of potential transformation paths. Based on natural language generation, the list of potential transformation paths is integrated into a conclusion summary to obtain conclusions on pollutant interaction behavior.

[0014] Secondly, this application also provides a rapid identification system for new pollutants in water bodies based on multifunctional sensing and AI, including: The feature module is used to acquire the synchronously acquired raw signal dataset and extract features from the raw signal dataset to obtain a multidimensional feature fingerprint array; The classification module is used to input a multidimensional feature fingerprint array into a pre-trained classification model to obtain a list of major pollutant categories; The analysis module is used to perform concentration and speciation analysis on pollutants in the list of major pollutant categories and generate pollutant speciation reports. The reasoning module is used to dynamically reason about the coexistence relationship of pollutants based on a multidimensional feature fingerprint array, a list of major pollutant categories, and a pollutant morphology report, and to obtain conclusions on pollutant interaction behavior. The reporting module integrates the list of major pollutant categories, pollutant speciation reports, and pollutant interaction behavior conclusions to obtain a pollutant identification report; the pollutant identification report is used to characterize pollutant types and pollution intensity.

[0015] Thirdly, this application also provides a computer device including a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement any step of the method provided in the first aspect of this application.

[0016] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any step of the method provided in the first aspect of this application.

[0017] The aforementioned method and system for rapid identification of new water pollutants based on multifunctional sensing and AI acquires a synchronously collected raw signal dataset and extracts features from it to obtain a multidimensional feature fingerprint array. This multidimensional feature fingerprint array is then input into a pre-trained classification model to obtain a list of major pollutant categories. Concentration and speciation analysis is performed on the pollutants in the list to obtain a pollutant speciation report. Based on the multidimensional feature fingerprint array, the list of major pollutant categories, and the pollutant speciation report, dynamic reasoning is performed on the coexistence relationships of pollutants to obtain pollutant interaction behavior conclusions. Finally, the list of major pollutant categories, the pollutant speciation report, and the pollutant interaction behavior conclusions are integrated to obtain a pollutant identification report. This pollutant identification report is used to characterize pollutant type and pollution intensity. It can effectively capture the dynamic transformation process of water pollutants, identify pollutant migration channels and trends, and effectively predict the scope and safety of water pollution. Phase distribution analysis can effectively improve the accuracy of ecological risk prediction, thereby enhancing the ability to identify the intensity and scope of water pollution. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a schematic diagram of the process of a rapid identification method for new pollutants in water based on multifunctional sensing and AI, provided in an embodiment of the present invention. Figure 2 This is a schematic diagram of the structure of a rapid water pollutant identification system based on multifunctional sensing and AI, provided in an embodiment of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0021] In one embodiment, such as Figure 1 As shown, a method for rapid identification of new pollutants in water bodies based on multifunctional sensing and AI is provided. This embodiment illustrates the application of this method to a terminal. It is understood that this method can also be applied to a server, or to a system including both a terminal and a server, and implemented through interaction between the terminal and the server. In this embodiment, the method includes the following steps: Step 101: Obtain the original signal dataset acquired synchronously, and extract features from the original signal dataset to obtain a multidimensional feature fingerprint array.

[0022] The synchronously acquired raw signal dataset refers to a set of unprocessed raw readings obtained from multiple different types of sensors measuring the same sample at the same time point. It typically consists of signal sequences such as voltage, current, and light intensity with timestamps, and may contain noise and missing values. Feature extraction refers to the data processing procedure of calculating a stable and discriminative amount of information from the raw signal that represents its essence. A multidimensional feature fingerprint array is a structured dataset where each dimension represents a specific feature, uniquely or highly discriminatively characterizing the comprehensive properties of pollutants in the sample. The terminal receives raw signal datasets synchronously transmitted from various sensors, performs data cleaning, identifies and removes outliers that significantly deviate from the normal range, and uses interpolation algorithms to fill in missing values. Since sensor signals are susceptible to environmental temperature fluctuations, synchronously acquired environmental temperature parameters are used to mathematically correct the time series to eliminate signal drift caused by temperature changes. The raw readings from each sensor are converted into standardized physical quantities with clear significance in environmental science. Statistical feature calculations are performed on the converted multidimensional physical quantity time series, thereby condensing the sequence information over a period of time into a fusion feature representing the overall characteristics of that period. The fusion feature is combined with relevant environmental parameters and normalized to eliminate deviations caused by differences in dimensions and orders of magnitude between different features, generating a multidimensional feature fingerprint array suitable for model processing. For example, the sensor uses a gold nanorod / graphene composite material as a substrate, and by modifying the surface with specific molecularly imprinted polymers or nucleic acid aptamers, a sensor array is constructed that can specifically capture different types of pollutant molecules such as antibiotics and pesticides.

[0023] Step 102: Input the multidimensional feature fingerprint array into the pre-trained classification model to obtain a list of major pollutant categories.

[0024] Specifically, a pre-trained classification model refers to a mathematical model that has been trained using a large amount of sample data with known pollutant categories. This model has the ability to automatically determine the pollutant category based on input features. The list of major pollutant categories is a list that, after the classification model's assessment, identifies the names or category identifiers of the major pollutants present in the test sample. The terminal takes a multi-dimensional feature fingerprint array as input and directly passes it to the pre-trained classification model. Internally, the model parses, weights, and calculates the input features based on its learned complex rules and patterns, outputting the identification results in the form of multiple pollutant categories, usually accompanied by a confidence score indicating the reliability of the judgment. The model determines and generates the list of major pollutant categories according to preset rules.

[0025] Step 103: Perform concentration and speciation analysis on the pollutants in the list of major pollutant categories to obtain a pollutant speciation report.

[0026] Specifically, concentration and speciation analysis refers to determining the specific content of pollutants in a sample, as well as their distribution in different physicochemical phases, based on known pollutant categories. The pollutant speciation report is a comprehensive analytical result, containing quantitative concentration information and speciation information for each identified pollutant. The terminal quantifies the pollutant content by matching key analytical features to each pollutant in the list based on a pre-defined knowledge base. Using mathematical regression methods, these features are correlated and calculated with information in a multidimensional feature fingerprint array to estimate the total concentration of various pollutants, analyze the pollutant's state of existence, and, using the knowledge base and multidimensional feature fingerprint array, analyze and calculate the specific distribution ratio of each pollutant in different phases such as the dissolved phase, colloidal phase, and particulate phase by calling specific environmental behavior models and allocation calculation formulas.

[0027] Step 104: Based on the multidimensional feature fingerprint array, the list of major pollutant categories, and the pollutant morphology report, dynamic reasoning is performed on the coexistence relationship of pollutants to obtain conclusions on pollutant interaction behavior.

[0028] Coexistence relationships refer to the potential interactions between multiple pollutants when they coexist in an environment. Dynamic reasoning is a process of logical inference based on the temporal correlations and statistical relationships between data. The pollutant interaction behavior conclusion is an analytical conclusion about how identified pollutants may interact. The terminal analyzes the time-varying sequences of features related to different pollutants in the multidimensional feature fingerprint array, calculates statistical correlations, and obtains a matrix representing the strength of the correlation. Using pollutants from the main pollutant category list as nodes and the calculated statistical correlations as connecting edges, and incorporating information from pollutant morphology reports as node attributes, an interaction network is constructed. Based on this network, graph analysis algorithms such as path search and community detection are used to infer potential transformation paths. The complex network relationships obtained from the analysis are summarized into a text description, i.e., the pollutant interaction behavior conclusion, through rule-based or natural language generation techniques.

[0029] Step 105: Integrate the list of major pollutant categories, pollutant morphology report, and pollutant interaction behavior conclusions to obtain a pollutant identification report; the pollutant identification report is used to characterize pollutant type and pollution intensity.

[0030] Integration refers to the systematic organization of analytical results from different levels and of different natures into a complete and coherent comprehensive document, following the logic of decision support. The pollutant identification report is an authoritative and structured report that comprehensively characterizes the type, intensity, morphology, and interactions of pollutants. The terminal collects the list of major pollutant categories, pollutant morphology reports, and pollutant interaction behavior conclusions as raw materials. Following a pre-set, professional report template, these materials are arranged, cross-references are established between content, and the arranged content is converted into a report format. The layout ensures clear charts and readable text, generating a directly usable pollutant identification report. This report is used to assist in water body condition regulation.

[0031] This embodiment provides a rapid identification method for new water pollutants based on multifunctional sensing and AI. It acquires a synchronously collected raw signal dataset and extracts features from it to obtain a multidimensional feature fingerprint array. This multidimensional feature fingerprint array is then input into a pre-trained classification model to obtain a list of major pollutant categories. Concentration and speciation analysis is performed on the pollutants in the list to obtain a pollutant speciation report. Based on the multidimensional feature fingerprint array, the list of major pollutant categories, and the pollutant speciation report, dynamic reasoning is performed on the coexistence relationships of pollutants to obtain pollutant interaction behavior conclusions. Finally, the list of major pollutant categories, the pollutant speciation report, and the pollutant interaction behavior conclusions are integrated to obtain a pollutant identification report. This pollutant identification report is used to characterize pollutant type and pollution intensity. Through these methods, the dynamic transformation process of water pollutants can be effectively captured, the migration channels and trends of pollutants can be identified, and the scope and safety of water pollution can be effectively predicted. Phase distribution can effectively improve the accuracy of ecological risk prediction, thereby enhancing the ability to identify the intensity and scope of water pollution.

[0032] In one embodiment, feature extraction is performed on the original signal dataset to obtain a multidimensional feature fingerprint array, including: Step 201: Remove outliers from the original signal dataset and fill in the missing values ​​in the original signal dataset to obtain a multi-channel signal time series.

[0033] The raw signal dataset refers to the set of raw measurements directly acquired from multiple sensors without any processing, which may contain noise, errors, and missing values. Outliers refer to erroneous data points in the signal that significantly deviate from the normal range of variation due to transient interference or equipment malfunction. Missing values ​​refer to partial data loss in a continuous time series due to data transmission failures or acquisition interruptions. A multi-channel signal time series refers to a continuous and complete set of signal data acquired synchronously from multiple channels of multiple sensors or multi-functional sensors after cleaning and imputation, presented on the time axis. The terminal identifies outliers and uses a local outlier factor method based on neighboring points to scan the signal data of each channel. It calculates the relationship between each data point and its surrounding data points and marks those points whose deviation exceeds a preset threshold as outliers. The marked outliers are removed from the data sequence, leaving a gap. The gap includes missing values ​​generated by removing outliers and missing values ​​already existing in the original data. For gaps with few single or consecutive missing points, linear interpolation is commonly used. A straight line is drawn connecting the two valid data points before and after the gap, and values ​​are taken on the line according to the location of the gap. For more complex scenarios or scenarios requiring smooth curves, spline interpolation is used. A polynomial function is used to fit the nearby valid data points to generate a smooth curve, and the missing values ​​are estimated through this curve.

[0034] Step 202: Based on the environmental parameters in the original signal dataset, perform temperature compensation correction on the multi-channel signal time series to obtain the corrected signal time series.

[0035] Specifically, environmental parameters refer to environmental physical quantities that are recorded synchronously during signal acquisition and may affect the accuracy of sensor readings; in this embodiment, temperature is specifically referred to. Temperature compensation correction is a signal processing technique designed to eliminate or reduce systematic errors caused by temperature changes in the sensor output signal. The corrected signal time series refers to multi-channel signal data where, after temperature compensation, the signal values ​​more accurately reflect the characteristics of the measured sample itself, rather than being affected by ambient temperature interference. The terminal uses a preset temperature compensation model, which describes the mathematical relationship between the sensor output signal, temperature, and the measured physical quantity. This model is determined during the sensor calibration phase. A specific temperature compensation model for each sensor channel is invoked, and the multi-channel signal time series is strictly aligned with the synchronously recorded temperature time series on the timestamps. This ensures that each signal data point corresponds to an accurate temperature value. For each data point in the signal series, the temperature value corresponding to that time point is read, and this temperature value and the current original signal value are substituted into the temperature compensation model for that channel. A correction amount is calculated and applied to the original signal value to obtain the corrected signal value at that time point. This calculation process is repeated for each point throughout the entire time series.

[0036] Step 203: Based on the time series of the calibration signal, each signal stream is converted into a standard physical quantity to obtain multidimensional time series data; the signal stream includes dynamic light scattering signal, multispectral signal and electrochemical signal.

[0037] Specifically, signal streams are calibrated data streams generated by a single sensor channel, such as dynamic light scattering signals, multispectral signals, and electrochemical signals. Standard physical quantities are physical quantities with clear definitions and units, such as particle size, absorbance, potential, and resistance. Multidimensional time series data refers to a multidimensional, timestamped data set formed after converting all signal streams into standard physical quantities. Based on the calibration curves of each sensor at the factory or in the field, the terminal converts raw electrical signals such as voltage, current, and photon counts into standard physical quantities. For dynamic light scattering signals, its photon autocorrelation function is converted into the particle size distribution of colloidal particles; for multispectral signals, the Lambert-Beer law is applied, combined with dark current and reference light intensity, to convert the raw light intensity values ​​into absorbance at various characteristic wavelengths; for electrochemical signals, by subtracting background current and comparing with standard curves, the peak current and potential in the cyclic voltammetry curve are converted into the characteristic peak height and position of specific redox reactions; and electrochemical impedance spectroscopy data is analyzed into interface parameters such as charge transfer resistance and double-layer capacitance through equivalent circuit fitting.

[0038] Step 204: Perform statistical feature calculations on the multidimensional time series data to obtain the primary fusion feature vector.

[0039] Statistical feature calculation is the process of extracting numerical indicators that summarize the statistical characteristics of a time series. The primary fusion feature vector is a one-dimensional array whose elements are various statistical feature values ​​calculated from the time series of each physical quantity, representing a condensation of the original sequence information. The terminal independently performs feature calculation operations for the time series of each physical quantity in the multi-dimensional time series data. For the time series of a single physical quantity, it iterates through a set of predefined statistical feature calculation functions. Central tendency features are obtained by calculating the arithmetic mean and median of the sequence; dispersion features are obtained by calculating the standard deviation, variance, and range; distribution shape features are obtained by calculating skewness and kurtosis; and energy features are obtained by calculating the root mean square value of the sequence. A set of feature values ​​is calculated for each physical quantity time series. In a predetermined order, all dimensions and all calculated feature values ​​are concatenated to form a very long one-dimensional array.

[0040] Step 205: Normalize the initial fusion feature vector and environmental parameters to obtain a multidimensional feature fingerprint array.

[0041] Normalization is a data preprocessing technique that scales data proportionally to fit a specific scale. A multidimensional feature fingerprint array refers to a normalized, standardized feature vector, serving as the fingerprint of a sample. The terminal merges the initial fused feature vector with the environmental parameters to be considered, forming an extended feature vector. Each dimension of the extended vector is independently normalized. Based on prior knowledge obtained during training or statistical results of the current batch of data, the global minimum and maximum values ​​for each feature dimension are determined. For each feature value in the vector, the minimum-maximum normalization formula is used for calculation, and each feature value is linearly mapped to the [0,1] interval.

[0042] This embodiment uses a series of professional physical and chemical algorithms to transform the original voltage / current signals into standard physical quantities with clear environmental scientific significance, generating information-rich multidimensional time series data, which effectively improves the accuracy of water pollution identification.

[0043] In one embodiment, based on the time series of the corrected signal, each signal stream is converted into a standard physical quantity to obtain multidimensional time series data, including: Step 301: Based on the time series of the correction signal, perform photon autocorrelation function inversion calculation on the dynamic light scattering signal to obtain the particle size distribution parameters.

[0044] Dynamic light scattering signal refers to the calibrated light intensity fluctuation signal used to measure the Brownian motion of nano- or micro-sized particles in solution. The smaller the particle, the faster the Brownian motion, and the faster the light intensity fluctuation. The photon autocorrelation function is a mathematical function used to quantify the similarity of light intensity fluctuation signals at different time points, describing the temporal memory of the light intensity signal; the rate of decay directly reflects the speed of the particle's Brownian motion. Inversion calculation refers to the process of deriving the original parameters leading to an observation result through mathematical models and algorithms. Particle size distribution parameters are a set of parameters used to describe the size characteristics of a particle population in a sample, typically including the average particle size, particle size distribution width, and the percentage of volume or number in different size ranges. The terminal processes the time series of dynamic light scattering signals, performs mathematical autocorrelation on the light intensity signals, and calculates the average of the products of light intensity I(t) at time t and light intensity I(t+τ) at time t+τ for a series of different time delays (τ). This average is then normalized to generate an autocorrelation function curve, which starts at 1 and decays as the time delay τ increases. An inversion calculation is then performed, fitting the experimental autocorrelation function curve to a theoretical model based on the principles of light scattering physics. The theoretical model describes the shape of the autocorrelation function that should be generated by a particle population with a specific particle size distribution. Optionally, through an iterative algorithm, CO... The NTIN algorithm, or nonnegative least squares method, continuously adjusts the assumed particle size distribution parameters until the difference between the theoretical autocorrelation function calculated from the particle size distribution and the experimentally measured result is minimized. The assumed set of particle size distribution parameters is then determined as the final result. For example, baseline correction is applied to the autocorrelation function curve to ensure that it tends to stabilize at long time delays. For monodisperse or narrow distribution systems, cumulant analysis is used for fitting to directly extract the attenuation linewidth and distribution parameters. For polydisperse or wide distribution systems, the CONTIN inversion algorithm is used to perform inverse Laplace transform, and the particle size distribution is solved through regularization.

[0045] Step 302: Convert the intensity value of each wavelength in the multispectral signal into an absorbance value using the following formula to obtain a multi-wavelength absorbance array:

[0046] in, To be at wavelength The absorbance value at that point For the sample at wavelength Measuring light intensity at that location, For reference at wavelength Measuring light intensity at that location, For dark current at wavelength The light intensity at that location was measured.

[0047] Specifically, a multispectral signal refers to a calibrated light intensity measurement signal at multiple different wavelengths. Absorbance is a core physical quantity in optical analysis, representing the degree to which light is absorbed when passing through a sample. It is a dimensionless quantity and is directly proportional to the concentration of the absorbing substance. A multi-wavelength absorbance array is a data structure that records the absorbance values ​​calculated at each specific wavelength. This array forms the basis for qualitative and quantitative spectral analysis of the sample. For a specific wavelength, the terminal reads three key light intensity measurements from the calibration signal time series: the light intensity illuminating the sample, the light intensity passing through the reference solution, and the sensor's background reading when there is no light. These three light intensity values ​​are then substituted into a given formula for calculation. This data reading and formula calculation operation is repeated for all wavelengths included in the multispectral signal. The absorbance values ​​calculated at all wavelengths are then arranged in wavelength order to form a one-dimensional array, i.e., the multi-wavelength absorbance array.

[0048] Step 303: Peak detection is performed on the electrochemical signal to identify the peak potential and peak current values ​​of the redox peak, and the electrochemical impedance spectrum is fitted with a preset equivalent circuit model to obtain the charge transfer resistance and double layer capacitance.

[0049] Specifically, in this embodiment, electrochemical signals refer to two types of calibrated signals: current-voltage curves obtained from voltammetry and impedance spectroscopy data obtained from electrochemical impedance spectroscopy. Redox peaks are the peak current values ​​appearing in the voltammetric curves, corresponding to the oxidation or reduction reactions of contaminants on the electrode surface, respectively. Peak potential is the electrode potential value corresponding to the redox peak, which can be used to qualitatively identify what substance has reacted. Peak current is the current intensity of the redox peak, which, under certain conditions, is proportional to the concentration of the reactant. Electrochemical impedance spectroscopy is a technique for measuring the impedance response of a system under alternating current disturbances at different frequencies. The equivalent circuit model is a circuit diagram composed of circuit elements such as resistors and capacitors, used to simulate and describe the physicochemical processes at the electrode / solution interface. Charge transfer resistance is a component in the equivalent circuit, characterizing the ease with which an electrochemical reaction occurs on the electrode surface; the higher the resistance, the more difficult the reaction. Double-layer capacitance is a component in the equivalent circuit, characterizing the capacitive properties of the double layer at the electrode / solution interface, and is related to the interface structure and adsorption behavior. The terminal analyzes the current-voltage curve, smooths and filters the curve to reduce noise, calculates the first derivative of current with respect to voltage, and precisely locates the position of the redox peak by finding the point where the derivative is zero and the sign changes. The voltage value corresponding to this position is the peak potential, and the current value corresponding to this position is the peak current value. Nonlinear least squares fitting is performed, comparing the experimentally measured electrochemical impedance spectroscopy data with the theoretical impedance value of a preset equivalent circuit model. Through optimization algorithms, the parameter values ​​of each component in the equivalent circuit model are automatically adjusted to minimize the overall error between the theoretical impedance spectrum calculated by the model and the experimental measurement values. When the fitting converges, the obtained component parameter values ​​are determined as the result. For example, for the cyclic voltammetry curve, a background current subtraction algorithm is applied to eliminate the influence of non-Radidatic current, and then a peak detection algorithm is used to identify the peak potential and peak current values ​​of the redox peak. For the electrochemical impedance spectroscopy data, nonlinear least squares is used to fit it with a preset equivalent circuit model to extract parameters such as charge transfer resistance and double-layer capacitance.

[0050] Step 304: Integrate particle size distribution parameters, multi-wavelength absorbance arrays, peak potential, peak current value, charge transfer resistance, double layer capacitance, and auxiliary parameters to obtain multidimensional time series data.

[0051] Auxiliary parameters refer to other parameters needed during the analysis process, such as timestamps, temperature, and other environmental or experimental condition information. Integration refers to the collection and organization of data from different sources and with different properties according to a certain structure and order. Multidimensional time series data is a multidimensional data set with timestamps, where each dimension represents a specific physicochemical parameter, collectively forming a comprehensive description of the sample. The terminal ensures that all parameters to be integrated correspond to the same time point or the same measurement batch, creating a structured data object. For each time point, all parameters representing the sample characteristics at that moment are packaged together. These parameters include particle size distribution parameters obtained from dynamic light scattering signals, multi-wavelength absorbance arrays obtained from multispectral signals, peak potential and peak current values ​​extracted from voltammetry signals, charge transfer resistance and double-layer capacitance obtained from impedance spectroscopy fitting, and related auxiliary parameters. Multiple data packets generated at different time points are arranged in chronological order to form a multidimensional time series data that evolves over time.

[0052] This embodiment integrates parameters from different sensors, representing different physicochemical meanings, into a unified, structured multidimensional time series data, maximizing the preservation of comprehensive information of the sample and providing a complete data foundation for subsequent feature extraction and machine learning models. By calculating physical quantities, the state of pollution is effectively captured, improving the accuracy of pollution identification.

[0053] In one embodiment, concentration and speciation analysis are performed on pollutants in the major pollutant category list to obtain a pollutant speciation report, including: Step 401: Based on the pollutant feature association mapping table and the list of major pollutant categories, extract the corresponding analytical features for each pollutant to obtain the pollutant feature vector mapping table.

[0054] The pollutant characteristic association mapping table is a pre-defined knowledge base, a database or lookup table, storing the identity characteristics of various known pollutants. For each pollutant, the table records its unique characteristic data used for identification and quantitative analysis. Optionally, this may include its unique spectral absorption peak position, standard electrochemical redox potential, specific molecular weight range, etc. The list of major pollutant categories is the result of the previous classification model and is a list of the names of the major pollutants detected in the sample. Analytical features refer to the key parameters or data segments used for quantitative and speciation analysis of specific pollutants, retrieved from the pollutant characteristic association mapping table. The pollutant feature vector mapping table is an enhanced list in which each pollutant on the list is associated with a corresponding set of characteristic data used for subsequent precise analysis. The terminal reads the list of major pollutant categories and obtains the names of the pollutants to be analyzed. For each pollutant name on the list, it is used as a keyword to perform a precise search in a large knowledge base called the pollutant feature association mapping table, similar to a dictionary search. All pre-stored analytical features of the pollutant are retrieved from the knowledge base. For example, for pesticide A, the following are extracted: its standard absorption coefficients at 235nm and 280nm in the ultraviolet spectrum, its peak potential value of the characteristic reduction peak in cyclic voltammetry (-0.75V), and its typical distribution coefficients among different phases in the water body. All the dispersed feature parameters extracted for the pollutant are combined into an ordered, structured data set, which is called the feature vector of the pollutant. For each pollutant on the list, a new table is generated. Each row of the table corresponds to a pollutant. Each row contains not only the pollutant name but also the complete feature vector constructed for the pollutant, which is the pollutant feature vector mapping table.

[0055] Step 402: Based on the pollutant feature vector mapping table, the total concentration of pollutants is analyzed by multiple regression analysis to obtain a quantitative result table.

[0056] Specifically, multiple regression analysis is a statistical modeling method used to establish a mathematical relationship between a dependent variable and multiple independent variables. In this embodiment, the dependent variable is the total concentration of pollutants, and the independent variables are various characteristic signals related to pollutants in a multidimensional feature fingerprint array. Total concentration refers to the total amount of pollutants in the sample, regardless of their phase. The quantitative results table is a table containing multiple columns of data, clearly listing the name of each identified pollutant and its corresponding calculated total concentration value. The terminal invokes a preset or multi-task learned multiple regression model. The mathematical form of this model can be linear or nonlinear. The model describes the following relationship: the measured comprehensive signal is equal to the sum of the signals contributed by various pollutants according to their concentration proportions. The pollutant feature vector mapping table provides the standard signal information required for model calculation. The multidimensional feature fingerprint array obtained from actual measurement is the input signal of the model. The operation of multiple regression analysis is to solve an optimization problem, to find a set of concentration values ​​such that when these concentration values ​​are multiplied by their corresponding standard signals and superimposed, the overall difference between the theoretical comprehensive signal obtained and the multidimensional feature fingerprint array actually measured in the experiment is minimized. When the optimization algorithm finds the set of concentration values ​​that minimizes the error, the calculation stops. This set of concentration values ​​is mapped to the pollutant names to generate a clear quantitative result table.

[0057] Step 403: Based on the pollutant feature vector mapping table and the multidimensional feature fingerprint array, estimate the proportion of the phase distribution of pollutants to obtain the phase distribution table; the phase distribution includes dissolved phase, colloidal phase and particulate phase.

[0058] Specifically, phase distribution refers to the allocation of pollutants in different physicochemical phases within a water body, primarily including the dissolved phase, colloidal phase, and particulate phase. A phase distribution table is a table detailing the mass fraction or concentration ratio of each pollutant in the dissolved, colloidal, and particulate phases. The terminal retrieves key parameters for each pollutant from the pollutant feature vector mapping table, including the linear partition coefficients between the dissolved and colloidal phases and between the dissolved and particulate phases, representing the pollutant's affinity for different phases. Background information of the sample water body is extracted from the multidimensional feature fingerprint array, and the total concentration of the colloidal and particulate phases in the water body is inverted using signals such as dynamic light scattering. This describes the availability of carriers for pollutant adsorption. For each pollutant, the total concentration, partition coefficients, and total phase concentration of the water body are substituted into a classic linear equilibrium distribution model. This formula allows for the precise calculation of the pollutant's mass fraction in the dissolved phase, colloidal phase, and particulate phase. This calculation is repeated for all pollutants, and the results are compiled into a table to obtain the phase distribution table.

[0059] Step 404: Integrate the quantitative results table and the phase distribution table to obtain the pollutant speciation report.

[0060] Integration refers to combining different data tables with inherent logical connections according to a certain structure and format to form a comprehensive document with complete content and clear organization. The pollutant speciation report is a comprehensive analytical report that not only lists the types and total concentrations of pollutants but also details the specific distribution of each pollutant in different phases. The terminal uses the pollutant name as the primary key to link the data in the quantitative results table and the phase allocation table. Following a preset report template, it arranges this linked data and outputs the arranged content as a complete and readable document—the pollutant speciation report.

[0061] This embodiment provides an in-depth understanding of the pollution status through pollutant speciation reports, which serves as a direct basis for conducting environmental risk assessments and developing remediation strategies, thereby improving the accuracy of pollution identification.

[0062] In one embodiment, based on a pollutant feature vector mapping table and a multidimensional feature fingerprint array, the proportion of pollutant phase distribution is estimated to obtain a phase allocation table, including: Step 501: Extract the distribution characteristics of various types of particulate matter in the water body from the multidimensional feature fingerprint array to obtain the water body distribution characteristics; the water body distribution characteristics include the mass concentration estimation and particle size distribution characteristics of the dissolved phase, colloidal phase and particulate phase.

[0063] Among them, water body distribution characteristics are a quantitative description of the physicochemical background of the sample water body itself, specifically referring to the overall situation of substances in the three physical forms of the water body: dissolved phase, colloidal phase, and particulate phase. Mass concentration estimation refers to the estimated value of the total mass concentration of substances in the dissolved phase, colloidal phase, and particulate phase. Particle size distribution characteristics refer to the description of the size and distribution range of substances in the colloidal and particulate phases. The terminal extracts information about the background phase of the water body from the complex integrated signal, focusing on features from the dynamic light scattering sensor in the multidimensional feature fingerprint array. These features include the decay rate of the light intensity autocorrelation function and the scattered light intensity, which are related to the size and concentration of particles in the water body. An inversion algorithm is run to substitute the feature values ​​into a mathematical model based on the physical principle of light scattering, thereby estimating the total concentration of the colloidal phase and the total concentration of the particulate phase, as well as the particle size distribution characteristics of the colloidal and particulate phases. Utilizing the multispectral absorbance features in the multidimensional feature fingerprint array, the scattering and absorption characteristics of dissolved substances, colloids, and particles differ. By analyzing the absorbance and signal baseline drift at specific wavelengths, the concentration of the dissolved phase can be estimated, and the concentration estimation results of the colloidal / particulate phase can be cross-validated and corrected.

[0064] Step 502: Based on the water body distribution characteristics, query the interaction strength between pollutants in the list of major pollutant categories and different phase distributions to obtain the allocation coefficient matrix.

[0065] In this embodiment, the interaction strength specifically refers to the strength of the adsorption / desorption between pollutant molecules and solid phase surfaces such as colloids and particulate matter. The partition coefficient is a physicochemical parameter that quantitatively describes the interaction strength. It represents the ratio of the pollutant concentration in the solid phase to its concentration in the dissolved phase under equilibrium conditions. The larger the coefficient value, the easier it is for the pollutant to partition from the dissolved phase to the corresponding solid phase. The partition coefficient matrix is ​​a two-dimensional table or data structure. The rows of the matrix represent each pollutant in the list of major pollutant categories, the columns represent different partition coefficients, and each element in the matrix is ​​a specific partition coefficient value for a particular pollutant. The terminal accesses a pre-defined pollutant characteristic database containing various physicochemical properties of pollutants, namely a pollutant characteristic association mapping table. The database stores a large number of standard partition coefficient values ​​of known pollutants. It traverses each pollutant in the list of major pollutant categories. For each pollutant, its name is used as an index to query its inherent linear partition coefficients of the dissolved phase to the colloidal phase and the dissolved phase to the particulate phase under standard conditions in the knowledge base. All the partition coefficient values ​​retrieved are organized into a structured table according to the order of pollutants, which is the partition coefficient matrix. It clearly lists the affinity of each pollutant for the colloidal phase and the particulate phase.

[0066] Step 503: Based on the total concentration, water body distribution characteristics, and distribution coefficient matrix, calculate the concentration distribution of pollutants in each phase using the following formula:

[0067]

[0068]

[0069] in, Let i be the mass fraction of contaminant i in the dissolved phase. Let be the mass fraction of contaminant i in the colloidal phase. Let be the mass fraction of pollutant i in the particulate phase, and let i be the pollutant index. Let i be the linear partition coefficient of contaminant i between the dissolved and colloidal phases. Let i be the linear partition coefficient of contaminant i between the dissolved phase and the particulate phase. This represents the total concentration of the colloidal phase. This represents the total concentration of the particulate phase.

[0070] Specifically, total concentration refers to the total amount of pollutants in the sample, which is the calculation result in the quantitative results table. Concentration distribution is the mass fraction, which refers to the proportion of the total amount of pollutants present in the dissolved phase, colloidal phase, and particulate phase, respectively. The sum of the three is 1, i.e., 100%. For each pollutant on the list, the terminal reads its total concentration from the quantitative results table; reads the total concentration of the colloidal phase and the total concentration of the particulate phase; and reads the distribution coefficient of the pollutant from the distribution coefficient matrix. Substituting the above four parameters into the three given mathematical formulas, precise calculations are performed. The parameter preparation and formula calculation operations are repeated for each pollutant in the list of major pollutant categories.

[0071] Step 504: Based on the pollutant feature vector mapping table, the concentration distribution is corrected to obtain the phase allocation table.

[0072] The correction process involves revising the theoretical concentration distribution calculated based on an ideal model to better reflect the complexities of actual water bodies. The phase distribution table is a clear table listing the final, more accurate mass fraction or concentration of each pollutant in the dissolved, colloidal, and particulate phases after correction. The terminal accesses the pollutant feature vector mapping table, which stores phase distribution correction factors or more complex correction functions for various pollutants under different environmental conditions. These factors are obtained through fitting extensive experimental data. Based on the actual environmental parameters of the current water body, appropriate correction factors are selected or calculated for each pollutant from the feature vector mapping table. These correction factors are used to correct the calculated theoretical mass fractions, and the corrected phase distribution results for all pollutants are organized into a structured table. This table is the final phase distribution table, reflecting the predicted values ​​of pollutant phase distribution under conditions closer to reality.

[0073] This embodiment decomposes the total concentration of pollutants into different phases, obtaining the theoretical mass fraction of each pollutant in the dissolved phase, colloidal phase, and particulate phase. By introducing environmental-specific corrections, it compensates for the shortcomings of ideal models, making the resulting phase allocation table more practical and providing a more reliable basis for accurately assessing the environmental behavior of pollutants, thereby improving the accuracy of pollution identification.

[0074] In one embodiment, based on a multidimensional feature fingerprint array, a list of major pollutant categories, and a pollutant morphology report, dynamic reasoning is performed on the coexistence relationships of pollutants to obtain conclusions on pollutant interaction behaviors, including: Step 601: Based on the multidimensional feature fingerprint array and the list of major pollutant categories, calculate the statistical correlation between the features of different pollutants to obtain the time-delay correlation coefficient matrix.

[0075] Statistical correlation refers to the use of mathematical methods to measure the existence and strength of linear or nonlinear relationships between two or more variables. The time-lag correlation coefficient is a special type of correlation coefficient that calculates not only the correlation between two time series at the same point in time, but also the correlation between one series and another at different time lags. This helps to discover whether the change of one variable always leads or lags another. The time-lag correlation coefficient matrix is ​​a square matrix where rows and columns represent pollutants from the list of major pollutant categories. Each element in the matrix is ​​a numerical value representing the correlation coefficient between two corresponding pollutants at a specific time lag in their characteristic time series. The matrix quantifies the dynamic correlation strength between all pollutants. For each pollutant in the list of major pollutant categories, the terminal extracts the feature time series that best represents the concentration change of the pollutant from the multidimensional feature fingerprint array. For any two different pollutants in the list, the terminal obtains the feature time series, calculates the cross-correlation coefficient of the two series under different time delays, and generates a curve of correlation coefficient changing with time delay τ. The terminal finds the correlation coefficient with the largest absolute value and its corresponding time delay value from the curve. The maximum value represents the strongest possible association between the two series at a certain time difference. The strongest time delay correlation coefficient of each pair of pollutants is filled into a table. The row index of the table is pollutant A, the column index is pollutant B, and the value of the intersection point is the association strength between the pollutants, forming a complete time delay correlation coefficient matrix.

[0076] Step 602: Using pollutants in the list of major pollutant categories as nodes, construct connection edges based on the time-lag correlation coefficient matrix, extract corresponding data from the pollutant morphology report as node attributes, and generate a pollutant interaction network graph.

[0077] Specifically, in graph theory, a node represents an entity. In this embodiment, each node represents a pollutant from the list of major pollutant categories. A connecting edge is a line connecting two nodes in graph theory, representing a relationship between the nodes. In this embodiment, the edge is determined by the time-delay correlation coefficient matrix. Node attributes are additional information attached to a node, used to describe the characteristics of that node. The pollutant interaction network graph is a data structure composed of nodes, edges, and node attributes, visually representing the interaction relationships between pollutants. The terminal creates a node for each pollutant in the list of major pollutant categories, using the name of the pollutant as the unique identifier of the node. From the pollutant morphology report, relevant attribute information is extracted and attached to each pollutant node. For example, attributes include: the total concentration of the pollutant, and its distribution ratio in the dissolved phase, colloidal phase, and particulate phase. Attributes describe the state of each node. The time-delay correlation coefficient matrix is ​​read, and for each non-zero correlation coefficient value in the matrix, a connecting edge is created between the two pollutant nodes corresponding to that value. The edge itself can have weights and directions.

[0078] Step 603: Based on the pollutant interaction network graph, perform path search to infer the transformation path of pollutants and obtain a list of potential transformation paths.

[0079] Specifically, path search is an algorithm in graph theory used to find pathways connecting one node to another in a graph via a series of edges. A transformation path refers to a possible route through which one pollutant transforms into one or more other pollutants via processes such as chemical reactions or biodegradation. A potential transformation path list is a textual or structured list that shows one or more possible pollutant transformation chains discovered in the interactive network using a path search algorithm. The terminal sets the rules for path search; for example, rules could include considering only strongly correlated edges with weights above a certain threshold; following the direction of edges; and limiting the maximum path length. The graph search algorithm is run on the pollutant interactive network graph to attempt to find all paths in the network that satisfy the rules set in step 1 and connect different pollutants. For all searched paths, they are sorted and filtered based on factors such as total path weight and path length, retaining the most reliable paths. The filtered paths are recorded in a structured manner to form a potential transformation path list, with each path information including the path sequence and related confidence indicators.

[0080] Step 604: Based on natural language generation, the list of potential transformation paths is integrated into a conclusion summary to obtain the conclusion on pollutant interaction behavior.

[0081] Natural language generation (NLP) is a branch of artificial intelligence, referring to the technology that automatically converts structured data into fluent and readable natural language text. A conclusion summary is a concise and generalized textual statement of the analysis results. The pollutant interaction behavior conclusion is one or more paragraphs of textual summary that clearly elucidates the main interactions and potential transformation pathways between pollutants. The terminal analyzes the list of potential transformation pathways, determines the core content to be reported, identifies the most important transformation pathways, and the pollutants involved in these pathways and their morphological attributes. Based on a predefined report template, the text structure is planned. For each transformation pathway to be described, NLP technology is used to fill the elements of the pathway into a pre-defined sentence template. All generated sentences are arranged according to the planned structure, ensuring clear pronoun references and appropriate use of conjunctions, ultimately forming a coherent and fluent text summary, i.e., the pollutant interaction behavior conclusion.

[0082] This embodiment effectively reveals the possible indirect transformation relationships and complex action chains between pollutants by constructing a knowledge graph structure and performing path search, which helps to predict the environmental behavior of pollution.

[0083] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0084] Based on the same inventive concept, this application also provides a system for rapidly identifying new water pollutants based on multifunctional sensing and AI, used to implement the aforementioned method for rapid identification of new water pollutants based on multifunctional sensing and AI. The solution provided by this system is similar to the solution described in the above method. Therefore, the specific limitations of one or more embodiments of the system for rapidly identifying new water pollutants based on multifunctional sensing and AI provided below can be found in the limitations of the method for rapidly identifying new water pollutants based on multifunctional sensing and AI described above, and will not be repeated here.

[0085] In one exemplary embodiment, such as Figure 2 As shown, a rapid identification system 700 for new water pollutants based on multifunctional sensing and AI is provided, including: The feature module 701 is used to acquire the synchronously acquired raw signal dataset and extract features from the raw signal dataset to obtain a multidimensional feature fingerprint array; The classification module 702 is used to input the multidimensional feature fingerprint array into the pre-trained classification model to obtain a list of major pollutant categories; Analysis module 703 is used to perform concentration and speciation analysis on pollutants in the list of major pollutant categories and obtain a pollutant speciation report. The reasoning module 704 is used to dynamically reason about the coexistence relationship of pollutants based on a multidimensional feature fingerprint array, a list of major pollutant categories, and a pollutant morphology report, and to obtain conclusions on pollutant interaction behavior. Report module 705 is used to integrate the list of major pollutant categories, pollutant speciation report and pollutant interaction behavior conclusions to obtain a pollutant identification report; the pollutant identification report is used to characterize pollutant type and pollution intensity.

[0086] Furthermore, feature module 701 is also used for: Outliers in the original signal dataset are removed, and missing values ​​in the original signal dataset are filled in to obtain a multi-channel signal time series. Based on the environmental parameters in the original signal dataset, temperature compensation correction is performed on the multi-channel signal time series to obtain the corrected signal time series. Based on the time series of the calibrated signal, each signal stream is converted into a standard physical quantity to obtain multidimensional time series data; the signal streams include dynamic light scattering signals, multispectral signals and electrochemical signals. Statistical feature calculations are performed on multidimensional time series data to obtain a primary fusion feature vector; The initial fusion feature vector and environmental parameters are normalized to obtain a multidimensional feature fingerprint array.

[0087] Furthermore, feature module 701 is also used for: Based on the time series of the correction signal, the photon autocorrelation function of the dynamic light scattering signal is inverted to obtain the particle size distribution parameters. The intensity value of each wavelength in the multispectral signal is converted into an absorbance value using the following formula, resulting in a multi-wavelength absorbance array:

[0088] in, To be at wavelength The absorbance value at that point For the sample at wavelength Measuring light intensity at that location, For reference at wavelength Measuring light intensity at that location, For dark current at wavelength The light intensity at that location was measured. Peak detection is performed on the electrochemical signal to identify the peak potential and peak current of the redox peak, and the electrochemical impedance spectrum is fitted with a preset equivalent circuit model to obtain the charge transfer resistance and double layer capacitance. By integrating particle size distribution parameters, multi-wavelength absorbance arrays, peak potential, peak current values, charge transfer resistance, double-layer capacitance, and auxiliary parameters, multidimensional time series data is obtained.

[0089] Furthermore, the analysis module 703 is also used for: Based on the pollutant feature association mapping table and the list of major pollutant categories, corresponding analytical features are extracted for each pollutant to obtain the pollutant feature vector mapping table. Based on the pollutant feature vector mapping table, the total concentration of pollutants was analyzed by multiple regression analysis to obtain a quantitative result table; Based on the pollutant feature vector mapping table and multidimensional feature fingerprint array, the proportion of pollutant phase distribution is estimated to obtain a phase distribution table; the phase distribution includes dissolved phase, colloidal phase and particulate phase; By integrating the quantitative results table and the phase distribution table, a pollutant speciation report is obtained.

[0090] Furthermore, the analysis module 703 is also used for: The distribution characteristics of various particulate matter in water bodies are extracted from a multidimensional feature fingerprint array to obtain the water body distribution characteristics. The water body distribution characteristics include the mass concentration estimation and particle size distribution characteristics of the dissolved phase, colloidal phase and particulate phase. Based on the water body distribution characteristics, the interaction strength between pollutants in the list of major pollutant categories and different phase distributions is queried to obtain the allocation coefficient matrix; Based on the total concentration, water body distribution characteristics, and distribution coefficient matrix, the concentration distribution of pollutants in each phase is calculated using the following formula:

[0091]

[0092]

[0093] in, Let i be the mass fraction of contaminant i in the dissolved phase. Let be the mass fraction of contaminant i in the colloidal phase. Let be the mass fraction of pollutant i in the particulate phase, and let i be the pollutant index. Let i be the linear partition coefficient of contaminant i between the dissolved and colloidal phases. Let i be the linear partition coefficient of contaminant i between the dissolved phase and the particulate phase. This represents the total concentration of the colloidal phase. This represents the total concentration of the particulate phase. Based on the pollutant feature vector mapping table, the concentration distribution is corrected to obtain the phase allocation table.

[0094] Furthermore, the inference module 704 is also used for: Based on the multidimensional feature fingerprint array and the list of major pollutant categories, the statistical correlation between the features of different pollutants is calculated to obtain the time-lag correlation coefficient matrix. Using pollutants in the list of major pollutant categories as nodes, connecting edges are constructed based on the time-lag correlation coefficient matrix. Corresponding data is extracted from the pollutant morphology report as the attributes of the nodes to generate a pollutant interaction network graph. Based on the pollutant interaction network graph, path search is performed to infer the transformation path of pollutants and obtain a list of potential transformation paths. Based on natural language generation, the list of potential transformation paths is integrated into a conclusion summary to obtain conclusions on pollutant interaction behavior.

[0095] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of the aforementioned method for rapid identification of new pollutants in water bodies based on multifunctional sensing and AI.

[0096] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0097] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0098] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A method for rapid identification of new pollutants in water bodies based on multifunctional sensing and AI, characterized in that, The method includes: Acquire the original signal dataset acquired synchronously, and extract features from the original signal dataset to obtain a multidimensional feature fingerprint array; The multidimensional feature fingerprint array is input into the pre-trained classification model to obtain a list of major pollutant categories; Concentration and speciation analysis were performed on the pollutants in the list of major pollutant categories to obtain a pollutant speciation report; Based on the multidimensional feature fingerprint array, the list of major pollutant categories, and the pollutant morphology report, dynamic reasoning is performed on the coexistence relationship of the pollutants to obtain conclusions on pollutant interaction behavior. By integrating the list of major pollutant categories, the pollutant morphology report, and the pollutant interaction behavior conclusions, a pollutant identification report is obtained; the pollutant identification report is used to characterize pollutant type and pollution intensity.

2. The method according to claim 1, characterized in that, The step of extracting features from the original signal dataset to obtain a multidimensional feature fingerprint array includes: Outliers in the original signal dataset are removed, and missing values ​​in the original signal dataset are filled in to obtain a multi-channel signal time series. Based on the environmental parameters in the original signal dataset, the multi-channel signal time series is temperature-compensated and corrected to obtain the corrected signal time series. Based on the time series of the correction signal, each signal stream is converted into a standard physical quantity to obtain multidimensional time series data; the signal stream includes dynamic light scattering signal, multispectral signal and electrochemical signal. Statistical feature calculations are performed on the multidimensional time series data to obtain a primary fusion feature vector; The primary fusion feature vector and the environmental parameters are normalized to obtain the multidimensional feature fingerprint array.

3. The method according to claim 2, characterized in that, Based on the corrected signal time series, each signal stream is converted into a standard physical quantity to obtain multidimensional time series data, including: Based on the time series of the correction signal, the photon autocorrelation function of the dynamic light scattering signal is inverted to obtain the particle size distribution parameters. The intensity value of each wavelength in the multispectral signal is converted into an absorbance value using the following formula, resulting in a multi-wavelength absorbance array: in, To be at wavelength The absorbance value at that point For the sample at wavelength Measuring light intensity at that location, For reference at wavelength Measuring light intensity at that location, For dark current at wavelength The light intensity at that location was measured. Peak detection is performed on the electrochemical signal to identify the peak potential and peak current values ​​of the redox peak, and the electrochemical impedance spectrum and a preset equivalent circuit model are fitted to obtain the charge transfer resistance and double layer capacitance. By integrating the particle size distribution parameters, the multi-wavelength absorbance array, the peak potential, the peak current value, the charge transfer resistance, the double-layer capacitance, and auxiliary parameters, the multidimensional time series data is obtained.

4. The method according to claim 1, characterized in that, The process of analyzing the concentration and speciation of pollutants in the list of major pollutant categories to obtain a pollutant speciation report includes: Based on the pollutant feature association mapping table and the list of major pollutant categories, corresponding analytical features are extracted for each pollutant to obtain a pollutant feature vector mapping table. Based on the pollutant feature vector mapping table, the total concentration of the pollutants is analyzed by multiple regression analysis to obtain a quantitative result table; Based on the pollutant feature vector mapping table and the multidimensional feature fingerprint array, the proportion of the phase distribution of the pollutants is estimated to obtain a phase distribution table; the phase distribution includes dissolved phase, colloidal phase and particulate phase; By integrating the quantitative results table and the phase allocation table, the pollutant speciation report is obtained.

5. The method according to claim 4, characterized in that, The step of estimating the proportion of phase distribution of pollutants based on the pollutant feature vector mapping table and the multidimensional feature fingerprint array to obtain a phase distribution table includes: The distribution characteristics of various particles in the water body are extracted from the multidimensional feature fingerprint array to obtain the water body distribution characteristics; the water body distribution characteristics include the mass concentration estimation and particle size distribution characteristics of the dissolved phase, the colloidal phase and the particulate phase; Based on the water body distribution characteristics, the interaction strength between the pollutants in the list of major pollutant categories and different phase distributions is queried to obtain the allocation coefficient matrix; Based on the total concentration, the water body distribution characteristics, and the distribution coefficient matrix, the concentration distribution of the pollutant in each phase is calculated using the following formula: in, Let i be the mass fraction of contaminant i in the dissolved phase. Let be the mass fraction of contaminant i in the colloidal phase. Let be the mass fraction of pollutant i in the particulate phase, and let i be the pollutant index. Let i be the linear partition coefficient of contaminant i between the dissolved and colloidal phases. Let i be the linear partition coefficient of contaminant i between the dissolved phase and the particulate phase. This represents the total concentration of the colloidal phase. This represents the total concentration of the particulate phase. Based on the pollutant feature vector mapping table, the concentration distribution is corrected to obtain the phase allocation table.

6. The method according to claim 1, characterized in that, The method of dynamically reasoning about the coexistence relationships of pollutants based on the multidimensional feature fingerprint array, the list of major pollutant categories, and the pollutant morphology report, to obtain pollutant interaction behavior conclusions, includes: Based on the multidimensional feature fingerprint array and the list of major pollutant categories, the statistical correlation between the features of different pollutants is calculated to obtain a time-lag correlation coefficient matrix. Using the pollutants in the list of major pollutant categories as nodes, connection edges are constructed based on the time-delay correlation coefficient matrix, and corresponding data is extracted from the pollutant morphology report as attributes of the nodes to generate a pollutant interaction network graph; Based on the pollutant interaction network graph, path search is performed to infer the transformation path of the pollutants and obtain a list of potential transformation paths; Based on natural language generation, the list of potential transformation paths is integrated into a conclusion summary to obtain the conclusion on the pollutant interaction behavior.

7. A rapid identification system for new pollutants in water bodies based on multifunctional sensing and AI, characterized in that, The system includes: The feature module is used to acquire the synchronously acquired raw signal dataset and extract features from the raw signal dataset to obtain a multidimensional feature fingerprint array; The classification module is used to input the multidimensional feature fingerprint array into the pre-trained classification model to obtain a list of major pollutant categories; The analysis module is used to perform concentration and speciation analysis on pollutants in the list of major pollutant categories and obtain a pollutant speciation report. The reasoning module is used to dynamically reason about the coexistence relationship of pollutants based on the multidimensional feature fingerprint array, the list of major pollutant categories, and the pollutant morphology report, and to obtain conclusions on pollutant interaction behavior. The reporting module is used to integrate the list of major pollutant categories, the pollutant morphology report, and the pollutant interaction behavior conclusions to obtain a pollutant identification report; the pollutant identification report is used to characterize the pollutant type and pollution intensity.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images