Rapid detection method for water pollutants

Abstract

The invention provides a rapid detection method for water pollutants, which comprises the following steps: removing large-particle impurities and suspended solids in a water sample to be detected to obtain a target water sample; the method comprises the following steps: synchronously detecting multiple pollutant indexes of a target water sample through a multi-source sensor array to generate an initial detection data set; performing noise filtering and baseline correction on the initial detection data set to generate a target detection data set; performing space-time dimension dynamic fusion and feature decoupling on the target detection data set to generate water quality pollutant data and dynamic risk assessment indexes; and constructing a multi-dimensional risk assessment report by adopting the water quality pollutant data and the dynamic risk assessment indexes. By adopting a dynamic gradient filtering technology, impurity interference is effectively removed, dynamic risk assessment of a time dimension is realized, the accuracy and reliability of detection in a complex water quality environment are remarkably improved, and rapid and accurate water quality pollutant detection and risk assessment are realized.

Description

Technical Field

[0001] The present invention relates to the technical field of environmental monitoring, and in particular to a method for rapid detection of water pollutants. Background Art

[0002] With the acceleration of industrial development and urbanization, the water quality environment is becoming increasingly complex. There are many types of pollutants in water, including organic pollutants, heavy metal ions, microorganisms, etc., and the concentration of each pollutant fluctuates widely. At the same time, there may be a variety of interfering substances, which puts higher requirements on the detection methods of water quality pollutants.

[0003] Currently, common methods for detecting water pollutants include chromatography, spectroscopy, and electrochemistry. However, these traditional detection methods have significant shortcomings when dealing with complex and changing water quality environments. For example, while chromatography has high separation efficiency and detection accuracy, it has a long detection cycle, complex operation, and high sample pretreatment requirements, making it difficult to meet the needs of rapid detection. Spectroscopy is easily interfered with by factors such as suspended matter, color, and turbidity in the water, resulting in distorted detection signals. Electrochemical methods have stringent requirements on electrode stability and selectivity. In complex water samples containing multiple ions, electrodes are easily contaminated or poisoned, affecting the accuracy of test results. Summary of the Invention

[0004] In view of this, the present invention proposes a method for rapid detection of water pollutants to solve the technical problem that existing detection methods are easily affected by external interference when dealing with complex and changeable water quality environments, resulting in low accuracy of detection results.

[0005] The technical solution of the present invention is achieved as follows:

[0006] A method for rapid detection of water pollutants comprises the following steps:

[0007] Step S1: removing large particles of impurities and suspended matter from the water sample to be tested to obtain a target water sample;

[0008] Step S2: performing simultaneous detection of multiple pollutant indicators on the target water sample through a multi-source sensor array to generate an initial detection data set;

[0009] Step S3: performing noise filtering and baseline correction on the initial detection data set to generate a target detection data set;

[0010] Step S4: Dynamically fuse and decouple the target detection dataset in spatiotemporal dimensions to generate water quality pollutant data and dynamic risk assessment indicators;

[0011] Step S5: construct a multidimensional risk assessment report using the water pollutant data and the dynamic risk assessment indicators.

[0012] Optionally, the specific steps of step S1 are:

[0013] Step S11, passing the water sample to be tested through a dynamic gradient filtration device to remove large particle impurities to generate an initial water sample, wherein the dynamic gradient filtration device includes a nylon mesh, a polyethersulfone filter membrane, and a ceramic membrane with successively smaller mesh sizes;

[0014] Step S12, introducing the initial water sample into a magnetic graphene oxide adsorption column with surface modified polydopamine to adsorb suspended particles and hydrophobic organic pollutants to obtain an intermediate water sample;

[0015] Step S13: introducing the intermediate water sample into the microfluidic chip, utilizing the turbulent effect to peel off the magnetic graphene oxide fragments remaining in the adsorption column, and intercepting the fragments through the magnetic screen at the outlet to obtain the target water sample.

[0016] Optionally, the multi-source sensor array includes a spectral sensor module, an electrochemical sensor module, and a biosensor module;

[0017] The spectral sensor module includes a 90° scattered light detector, an ultraviolet-visible spectrophotometer sensor and a Raman spectroscopy sensor, which are used to detect characteristic spectral signals of organic pollutants and heavy metal ions;

[0018] The electrochemical sensor module includes a molecularly imprinted polymer modified glassy carbon working electrode and a self-repairing Ag / AgCl reference electrode, which is used to selectively detect heavy metal ions in a complex ionic environment;

[0019] The biosensor module comprises a PDMS microfluidic chip, a multi-enzyme immobilization detection unit and a thin film temperature control module, and is used for synchronously monitoring microbial metabolites and toxic substances.

[0020] Optionally, the specific steps of step S2 are:

[0021] Step S21: Detecting the target water sample through the spectral sensor module of the multi-source sensor array to generate a spectral detection sub-dataset;

[0022] Step S22: detecting the target water sample through the electrochemical sensor module of the multi-source sensor array to generate an electrochemical detection sub-dataset;

[0023] Step S23: Detecting the target water sample through the biosensor module of the multi-source sensor array to generate a biodetection sub-dataset;

[0024] Step S24: aligning the timestamps of the spectral detection sub-dataset, the electrochemical detection sub-dataset, and the biological detection sub-dataset to generate an initial detection data set.

[0025] Optionally, the specific steps of step S3 are:

[0026] Step S31: Using a wavelet transform algorithm and an adaptive Kalman filter algorithm to perform noise filtering and baseline correction on the initial detection data set to generate a denoised data set;

[0027] Step S32: Using a particle swarm optimization algorithm to perform nonlinear correction on the denoised data set to generate a corrected data set.

[0028] Optionally, the specific steps of step S31 are:

[0029] Step S311: Using a wavelet transform algorithm to perform multi-scale decomposition on the time domain signal of the initial detection data set to remove noise components and generate high-frequency components and low-frequency components;

[0030] Step S312: performing baseline correction on the low-frequency component using an adaptive Kalman filter algorithm to generate a low-frequency correction component;

[0031] Step S313: construct a denoising data set using the low-frequency correction component and the high-frequency component.

[0032] Optionally, the specific steps of step S32 are:

[0033] Step S321: construct a baseline drift model with the minimum mean square error between the sensor signal and the theoretical baseline as the objective function;

[0034] Step S322: using a particle swarm optimization algorithm to iteratively solve the objective function of the baseline drift model to generate optimal parameters;

[0035] Step S323: Perform nonlinear correction on the denoised data set using the optimal parameters to generate a corrected data set.

[0036] Optionally, the specific steps of step S4 are:

[0037] Step S41: Divide the target detection dataset into three-dimensional tensors, where the dimensions corresponding to the three-dimensional tensors are sensor type, time series, and pollutant index respectively;

[0038] Step S42: extracting the core tensor from the three-dimensional tensor using Tucker decomposition algorithm;

[0039] Step S43: Dynamically weight the time dimension of the core tensor through a multi-head attention mechanism and construct a matrix to generate a spatiotemporal fusion matrix;

[0040] Step S44: Decouple the spatiotemporal fusion matrix to generate water quality pollutant data and dynamic risk assessment indicators.

[0041] Optionally, the specific steps of step S44 are:

[0042] Step S441: using a multi-layer fully connected network to extract high-dimensional features in the spatiotemporal fusion matrix to generate pollutant feature vectors and environmental interference feature vectors;

[0043] Step S442: Normalize the time gradient of the pollutant feature vector to generate a pollutant feature;

[0044] Step S443: weighting the spatial distribution of the pollutant feature according to the sensor type corresponding to the pollutant feature to generate a pollution weighted feature;

[0045] Step S444: asymmetrically suppress the environmental interference feature vector to generate an interference suppression feature;

[0046] Step S445: Evaluate the spatial details and temporal continuity of the pollution weighted features and the interference suppression features through a reflection image discriminator and a gradient map discriminator to generate spatial authenticity scoring data and temporal continuity scoring data;

[0047] Step S446: Asymmetric weight adjustment is performed on the pollution weighted feature and the interference suppression feature according to the spatial authenticity score data and the temporal continuity score data to construct a pollutant feature time series;

[0048] Step S447: Capture the forward accumulation and backward diffusion patterns of pollutant concentrations in the pollutant characteristic time series through a bidirectional gate cycle unit to generate water quality pollutant data and dynamic risk assessment indicators.

[0049] Optionally, the specific steps of step S5 are:

[0050] Step S51: determining an initial risk level according to the ratio of the pollutant concentration exceeding the standard in the water quality pollutant data to the ecotoxicity threshold;

[0051] Step S52: integrating the dynamic risk assessment index with the water body functional zoning data in the geographic information system, and dynamically assigning environmental sensitivity weights of different pollutants through an attention mechanism to generate a spatially weighted risk coefficient;

[0052] Step S53: Migrate the risk model in the historical pollution event database to the current scenario, correct the deviation of the initial risk level, and generate a target risk level;

[0053] Step S54: construct a three-dimensional visualization report using the target risk level and the spatially weighted risk coefficient to generate a multi-dimensional risk assessment report.

[0054] Compared with the prior art, the present invention has the following beneficial effects:

[0055] The present invention obtains a target water sample by removing large particles of impurities and suspended matter from the water sample to be tested. The target water sample is simultaneously tested for multiple pollutant indicators using a multi-source sensor array, and combined with a spatiotemporal dynamic fusion algorithm, the detection time can be shortened. The initial detection data set is subjected to noise filtering and baseline correction to generate a target detection data set. The target detection data set is then subjected to dynamic fusion and feature decoupling in the spatiotemporal dimension to generate water quality pollutant data and dynamic risk assessment indicators. Finally, the water quality pollutant data and dynamic risk assessment indicators are used to construct a multidimensional risk assessment report. By adopting dynamic gradient filtration technology, impurity interference is effectively removed, dynamic risk assessment in the time dimension is realized, the accuracy of risk assessment is improved, and the obtained detection results are highly accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only preferred embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0057] Figure 1 This is a flow chart of a method for rapid detection of water pollutants according to the present invention;

[0058] Figure 2 This is a flow chart of step S1 of a method for rapid detection of water pollutants of the present invention;

[0059] Figure 3 This is a flow chart of step S2 of a method for rapid detection of water pollutants of the present invention;

[0060] Figure 4 This is a flow chart of step S3 of a method for rapid detection of water pollutants of the present invention;

[0061] Figure 5 This is a flow chart of step S4 of a method for rapid detection of water pollutants of the present invention;

[0062] Figure 6 This is a flow chart of step S5 of a method for rapid detection of water pollutants of the present invention. DETAILED DESCRIPTION

[0063] In order to better understand the technical content of the present invention, a specific embodiment is provided below, and the present invention is further described in conjunction with the accompanying drawings.

[0064] like Figure 1 The figure shows a flow chart of a method for rapid detection of water pollutants.

[0065] In an embodiment of the present invention, the original water sample to be tested is preprocessed. Impurities and suspended matter of varying particle sizes are gradually removed from the sample through physical filtration and adsorption methods to obtain a target water sample suitable for subsequent testing. This prevents interference from large particles and suspended matter on the test results, resulting in the target water sample. A multi-source sensor array is used to simultaneously detect pollutants in the target water sample from multiple angles. Different types of sensors, based on their respective detection principles, specifically detect different pollutant indicators in the water sample. This simultaneous detection allows data on multiple pollutant indicators to be obtained in a relatively short period of time, generating an initial test dataset. Noise is removed by utilizing the characteristics of wavelet transforms and adaptive Kalman filters. The baseline is adjusted by constructing a baseline drift model and performing nonlinear correction using a particle swarm optimization algorithm. This improves the quality of the initial test dataset, resulting in a target test dataset. By removing noise and correcting the baseline, the data more accurately reflects the true presence of pollutants in the water sample. The target test dataset is analyzed and processed using methods such as tensor decomposition, attention mechanisms, and deep learning to mine the spatiotemporal characteristics and potential information within the data. By performing complex processing on the target detection dataset, valuable information is extracted from the spatial and temporal dimensions to generate water quality pollutant data and dynamic risk assessment indicators, providing a basis for subsequent risk assessment. Finally, a multidimensional risk assessment report is constructed using these data and dynamic risk assessment indicators. This addresses the technical issue of existing detection methods being susceptible to external interference and resulting in low accuracy when dealing with complex and changing water quality environments.

[0066] Preferably, Figure 2 The flowchart of step S1 of a method for rapid detection of water pollutants is shown.

[0067] In an embodiment of the present invention, the dynamic gradient filtration device is a three-stage filtration structure with decreasing pore size. A nylon mesh is used as the first layer of filtration, and the mesh size is usually 50-200 μm (for example, 100 μm), which is used to intercept large particles of impurities visible to the naked eye (such as leaves, sand and gravel). A polyethersulfone (PES) filter membrane is used as the second layer of filtration, with a pore size of 0.2-1.0 μm (for example, 0.45 μm), which is used to remove tiny suspended matter and microbial aggregates. A ceramic membrane is used as the third layer of filtration, with a pore size of 0.05-0.2 μm (for example, 0.1 μm), and is made of porous alumina or zirconia material to further intercept nano-sized particles to ensure that the turbidity of the initial water sample is less than 5. Three layers of filter material are fixed in descending order of pore size within a cylindrical filter cartridge. Fluid inlets and outlets are located at both ends of the cartridge. Water samples flow from top to bottom through the cartridge at a rate of 0.5-5 L / min. Driven by gravity or a peristaltic pump, the filter gradually removes large impurities of varying sizes (such as silt, algae, and colloids) from the water sample, preventing them from clogging subsequent detection modules or interfering with detection signals. Furthermore, the structure can be designed to be removable, facilitating regular cleaning or replacement of the filter material, adapting to water samples with varying degrees of contamination.

[0068] The magnetic graphene oxide (MOG@PDA) adsorption column with surface modified polydopamine was used to remove the suspended particles (such as colloids) and hydrophobic organic pollutants (such as polycyclic aromatic hydrocarbons and pesticide residues) remaining in the initial water sample to obtain an intermediate water sample. Polydopamine (PDA) is coated on the MOG surface through the oxidative self-polymerization reaction of dopamine. The catechol groups on its molecular chain can enhance the adsorption capacity of organic pollutants through hydrogen bonding and π-π stacking, while providing negative charge sites to adsorb positively charged metal ions. The core gives the material magnetism. After adsorption is completed, the adsorption column and the water sample can be quickly separated by an external magnetic field (such as an electromagnetic coil) to avoid secondary contamination. The specific implementation method is as follows: (1) Preparation of adsorption column: MOG@PDA particles (particle size 50-100nm) are filled in a glass column with an inner diameter of 5-10mm, the filling height is 5-10cm, and the two ends are fixed with porous glass sand cores. (2) Adsorption process: The initial water sample passes through the adsorption column at a flow rate of 1-3mL / min, with a contact time of 5-10min. The hydrophobic pollutants are adsorbed by the PDA layer, and the suspended particles are removed by the physical interception of MOG. (3) Regeneration method: After adsorption saturation, the MOG@PDA particles can be separated by a magnetic field, and the organic pollutants can be eluted with ethanol or hydrochloric acid solution to achieve material regeneration.

[0069] Remove the MOG@PDA fragments that may remain on the adsorption column to prevent them from entering the sensor array and interfering with detection (such as blocking the flow channel or generating false signals) to obtain the target water sample. A serpentine or fishbone-shaped flow channel is designed in the microfluidic chip. When the water sample flow rate is greater than 100 μL / min, the Reynolds number (Re) in the flow channel is greater than 200, forming turbulence, and the MOG@PDA fragments that may fall off the surface of the adsorption column are peeled off by shear force. A 50-100 mesh magnetic stainless steel mesh (such as 316L stainless steel nickel-plated) is embedded at the chip outlet, and the gradient magnetic field generated by a permanent magnet (magnetic field strength 0.1-0.5T) is used to adsorb and intercept magnetic fragments. Specific implementation method: (1) Chip material and structure: PDMS (polydimethylsiloxane) is used to prepare the flow channel by soft lithography technology, with a width of 200-500 μm and a depth of 100-200 μm. The inlet is connected to the outlet of the adsorption column, and the outlet is connected to the magnetic screen module. (2) Operational procedures: The intermediate water sample passes through the chip at a flow rate of 200 μL / min. The turbulent flow lasts for 1-2 min, and the debris is intercepted by the magnetic screen. The purified target water sample is collected from the outlet, and the residual metal ion concentration is <0.1 mg / L.

[0070] Preferably, the multi-source sensor array includes a spectral sensor module, an electrochemical sensor module, and a biosensor module;

[0071] The spectral sensor module includes a 90° scattered light detector, a UV-visible spectrophotometer sensor, and a Raman spectroscopy sensor, which are used to detect the characteristic spectral signals of organic pollutants and heavy metal ions;

[0072] The electrochemical sensor module includes a molecularly imprinted polymer-modified glassy carbon working electrode and a self-healing Ag / AgCl reference electrode for selective detection of heavy metal ions in complex ionic environments.

[0073] The biosensor module, including a PDMS microfluidic chip, a multi-enzyme immobilization detection unit and a thin film temperature control module, is used to simultaneously monitor microbial metabolites and toxic substances.

[0074] In the embodiment of the present invention, the multi-source sensor array realizes the detection of organic pollutants (such as COD, ammonia nitrogen) and heavy metal ions (such as Pb 2+ 、Cd 2+) for qualitative and quantitative analysis. The 90° scattered light detector is based on the Mie scattering theory. By measuring the scattered light intensity at 90°, it calculates the concentration (such as turbidity) and particle size distribution of particulate matter in the water sample. The detection range is 0.1-100NTU. The 90° scattered light detector uses a laser diode (wavelength 650nm) as a light source and a photomultiplier tube as a detector. It is integrated on one side of the flow channel and forms a 90° angle with the light source. The UV-visible spectrophotometric sensor uses the Lambert-Beer law to scan the absorbance in the wavelength range of 200-800nm ​​to detect absorbing substances (such as nitrates and aromatic hydrocarbons). The resolution is 1nm and the detection limit is as low as 0.1mg / L. The UV-visible spectrophotometric sensor uses a deuterium tungsten halogen composite light source and a miniature fiber optic spectrometer (such as Ocean Optics HR4000) to collect the spectrum. The flow cuvette has an optical path of 10mm. The Raman spectroscopy sensor detects the Raman scattering signal (displacement range 50-4000cm) generated by molecular vibration. -1 ), identify heavy metal complexes or characteristic organic functional groups (such as CH, C=O), detection limit 10 -6 The Raman spectroscopy sensor is excited by a 785nm semiconductor laser (power 50-100mW), collects signals through grating spectrometry and a charge-coupled device (CCD), and has a built-in autofocus module.

[0075] Electrochemical sensor module can specifically detect heavy metal ions (such as Pb in complex ion environment (such as high salt wastewater) 2+ 、Cd 2+ 、Hg 2+ ), and its anti-interference ability is better than that of traditional electrodes. Molecularly imprinted polymer (MIP) modified glassy carbon electrode is modified by template molecules (such as Pb 2+ ) directional polymerization forms imprinted holes, achieving specific recognition of target ions with a selectivity coefficient of >10 3 (Compared to non-target ions). Cyclic voltammetry is used to electropolymerize a molecularly imprinted polymer (MIP)-modified glassy carbon electrode onto the surface of the glassy carbon electrode. The MIP film has a thickness of 50-100 nm and a detection potential of -1.2V to 0.2V (vs Ag / AgCl). The self-healing Ag / AgCl reference electrode contains a built-in AgCl gel electrolyte. When the liquid junction is contaminated, ion diffusion through the gel automatically repairs the potential stability, with a drift of less than 5mV / 24h. The self-healing Ag / AgCl reference electrode consists of a glass tube filled with Ag / AgCl wire and KCl gel, and the tip is sealed with a porous ceramic plug, making it suitable for non-aqueous solutions.

[0076] The biosensor module monitors microbial metabolites (such as glucose and urea) and acute toxic substances (such as heavy metals and pesticides) in real time through immobilized enzyme reactions and temperature control. The PDMS (polydimethylsiloxane) microfluidic chip contains an injection channel (width 100μm), a reaction chamber (volume 10μL) and a waste liquid pool, and is encapsulated using oxygen plasma bonding technology. The immobilization method of the multi-enzyme immobilization detection unit is to fix glucose oxidase, urease, etc. to the inner wall of the reaction chamber by glutaraldehyde cross-linking, with an enzyme loading of 1-5mg / cm 2 The detection principle is as follows: (1) Glucose detection: glucose + O2 → gluconic acid + H2O2 (enzyme catalysis), the current change is detected by the hydrogen peroxide electrode, the linear range is 0.1-10mM. (2) Toxicity detection: luminescent bacteria (such as Vibrio fischeri) are fixed on the chip, and the toxic substances inhibit the luminescence intensity, and the detection limit is IC 50 The thin film temperature control module uses a polyimide heating film (50 μm thick) attached to the bottom of the chip. The reaction temperature is maintained at 30 ± 0.5 ° C by a PID controller, with a response time of less than 30 seconds.

[0077] Preferably, Figure 3 The flowchart of step S2 of a method for rapid detection of water pollutants is shown.

[0078] In an embodiment of the present invention, the characteristics of different detectors in the spectral sensor module are utilized to obtain characteristic spectral signals of organic pollutants and heavy metal ions in the target water sample, providing a data basis for subsequent pollutant analysis. The target water sample is connected to the sampling port of the spectral sensor module through a pipeline or a microfluidic channel to ensure that the water sample can flow smoothly into the detection chamber of each detector. The 90° scattered light detector selects a laser light source of appropriate wavelength (such as 650nm), sets the light source power to 5-10mW, and adjusts the sensitivity of the photodetector so that it can accurately detect the scattered light intensity. The ultraviolet-visible spectrophotometric sensor uses a deuterium tungsten composite light source, sets the wavelength scanning range to 200-800nm, the scanning interval to 1nm, and the integration time to 100-500ms to obtain accurate absorbance data. The Raman spectral sensor uses a 785nm semiconductor laser as the excitation light source, the power is set to 50-100mW, and the Raman scattering spectrum is collected by grating spectrometry and a charge-coupled device (CCD), and the spectral resolution is set to 4-8cm -1 The spectral sensor module is started, and each detector detects the water sample synchronously, converts the detected light signal into an electrical signal, and collects data in real time through the data acquisition card to generate a spectral detection sub-dataset.

[0079] In a complex ionic environment, the selective detection capability of the electrochemical sensor module is utilized to obtain the concentration information of heavy metal ions in the water sample. The working principle of the molecular imprinted polymer modified glassy carbon working electrode in the electrochemical sensor module is that the molecular imprinted polymer (MIP) is prepared by a template molecule (such as a heavy metal ion), and has cavities inside that are complementary to the spatial structure and binding sites of the template molecule. When the heavy metal ions in the target water sample come into contact with the MIP modified glassy carbon working electrode, the heavy metal ions will specifically bind to the cavities of the MIP, causing changes in the charge distribution and electron transfer process on the electrode surface. By measuring the current or potential changes of the electrode, quantitative detection of heavy metal ions can be achieved. The working principle of the self-repairing Ag / AgCl reference electrode in the electrochemical sensor module is to provide a stable potential reference for the reference electrode. The self-repairing Ag / AgCl reference electrode has a built-in AgCl gel electrolyte. When the liquid junction is contaminated, the ions in the gel can automatically repair the potential stability through diffusion, ensuring the accuracy of the working electrode potential measurement.

[0080] The electrochemical sensor module is used to detect the target water sample in the following manner: (1) Before use, the surface of the molecular imprinting polymer modified glassy carbon working electrode is cleaned and activated, such as by scanning within a certain potential range using cyclic voltammetry to remove impurities on the electrode surface and restore its activity. (2) The target water sample is injected into the detection cell of the electrochemical sensor module so that the working electrode and the reference electrode are completely immersed in the water sample. An appropriate potential or current signal is applied through the electrochemical workstation (such as differential pulse voltammetry, with a scanning potential range of -1.0V to +1.0V, a pulse amplitude of 50mV, and a pulse width of 50ms) to measure the current or potential change generated by the electrode reaction. (3) Data recording: The current-potential data output by the electrochemical workstation is recorded in real time, and the concentration of heavy metal ions is calculated based on the standard curve or electrochemical analysis method to generate an electrochemical detection sub-dataset.

[0081] Leveraging the biosensor module's sensitivity to microbial metabolites and toxic substances, it simultaneously monitors bio-related indicators in water samples and assesses the biosafety of water quality. The PDMS microfluidic chip and multi-enzyme immobilization detection unit in the biosensor module operate by immobilizing multiple enzymes (such as glucose oxidase and urease) within the reaction chamber of the PDMS microfluidic chip through physical adsorption and chemical crosslinking. When a target water sample flows into the reaction chamber, substrates (such as glucose and urea) in the sample react specifically with the immobilized enzymes, producing detectable signal substances (such as hydrogen peroxide and ammonia). By monitoring changes in the concentration of these signal substances, the substrate content can be indirectly reflected, thereby enabling the detection of microbial metabolites. The thin-film temperature control module in the biosensor module precisely controls the reaction temperature at the enzyme's optimal reaction temperature (e.g., 30±0.5°C) to ensure enzyme activity and the stability of the biological reaction. Stable temperature conditions help improve detection accuracy and repeatability.

[0082] The biosensor module is used to detect the target water sample as follows: (1) Chip preparation: Before use, clean and pre-treat the PDMS microfluidic chip to ensure that the internal channels of the chip are clean and free of impurities. Connect the reaction chamber of the immobilized enzyme to the injection channel and the waste liquid channel to ensure that the fluid path is unobstructed. (2) Detection process: The target water sample is injected into the injection port of the PDMS microfluidic chip at a constant flow rate (such as 100 μL / min) through a microinjection pump. The water sample is fully in contact with the immobilized enzyme in the reaction chamber inside the chip, and a biochemical reaction occurs. The signal substance produced by the reaction is detected by optical or electrochemical detection methods (such as using a hydrogen peroxide electrode to detect hydrogen peroxide produced by the glucose oxidase reaction, or detecting specific metabolites by fluorescence detection methods). (3) Temperature control: Start the thin film temperature control module, set the temperature to 30°C, monitor and adjust the temperature in real time, and ensure that the temperature fluctuation does not exceed ±0.5°C during the reaction. (4) Data processing: Based on the detected signal intensity, combined with the standard curve or algorithm, the concentration of microbial metabolites or the content of toxic substances in the water sample is calculated to generate a biological detection sub-dataset.

[0083] Because each sensor module has different detection speeds and data acquisition times, timestamp alignment ensures consistency across the time dimension of different detection sub-datasets, resulting in an initial detection dataset that facilitates subsequent data fusion and analysis. Within each sensor module's data acquisition system, precise timestamp information (e.g., down to the millisecond level) is added to each data point. By comparing and matching the timestamps of different sub-datasets and employing methods such as time interpolation and synchronous triggering, each sub-dataset is aligned to the same time series, ensuring accurate correspondence between detection data at the same moment.

[0084] The implementation method of timestamp alignment of data sets is as follows: (1) Timestamp addition: In the data acquisition program of the spectral sensor module, electrochemical sensor module and biosensor module, a high-precision clock module (such as a GPS-based clock synchronization module) is integrated to mark an accurate timestamp for each data point while collecting data. (2) Time alignment algorithm: An alignment algorithm based on time interpolation (such as linear interpolation and spline interpolation) is used to adjust the time series of different sub-data sets. First, the time range and sampling interval of each sub-data set are determined, and the minimum and maximum boundaries of the timestamp are found. Then, based on the sub-data set with the smallest time interval, the other sub-data sets are interpolated so that all sub-data sets have the same sampling points in time. (3) Data merging: The spectral detection sub-data set, electrochemical detection sub-data set and bio-detection sub-data set that have been timestamp aligned are merged and arranged in time series order to generate an initial detection data set containing all detection information, providing a unified data basis for subsequent data processing and analysis.

[0085] Preferably, Figure 4 The flowchart of step S3 of a method for rapid detection of water pollutants is shown.

[0086] In an embodiment of the present invention, a wavelet transform algorithm and an adaptive Kalman filter algorithm are used to remove noise interference from the initial detection data set, correct baseline drift, improve data quality, generate a denoised data set, and provide an accurate data basis for subsequent analysis. The wavelet transform algorithm, based on multi-resolution analysis theory, decomposes the time domain signal into components of different frequencies. Noise is usually concentrated in the high-frequency part. By removing or suppressing the high-frequency components, the noise can be effectively filtered out; while the low-frequency components retain the main characteristics of the signal. The adaptive Kalman filter algorithm uses a state-space model to dynamically adjust the filter parameters based on the state estimate at the previous moment and the observation value at the current moment, correct the baseline drift in the low-frequency component, and achieve an optimal estimate of the signal trend.

[0087] The implementation process of using the wavelet transform algorithm and the adaptive Kalman filter algorithm to perform noise filtering and baseline correction on the initial detection data set is as follows: (1) Data input: The initial detection data set generated in step S2 is imported into the data processing system. The data format is multidimensional data of time series, containing multi-source information such as spectroscopy, electrochemistry, and biological detection. (2) Wavelet transform processing: Select appropriate wavelet basis functions, such as Daubechies wavelet (dbN), Symlets wavelet (symN), etc., and determine the number of decomposition layers based on the data characteristics and noise characteristics, generally 3-5 layers. Perform multi-scale decomposition on the time domain signal of the initial detection data set to obtain high-frequency components and low-frequency components. Process the high-frequency components by setting thresholds (such as soft thresholds and hard thresholds) to suppress or remove noise-related high-frequency signals. (3) Adaptive Kalman filter processing: Establish a state space model for the low-frequency component and define the state equation and observation equation, where the state equation describes the dynamic changes of the signal and the observation equation represents the measurement process of the sensor. Initialize the filter parameters, including the state estimation covariance matrix and the observation noise covariance matrix. Based on the observation value at the current moment and the state estimate at the previous moment, the adaptive Kalman filter algorithm is used to update the state estimate and covariance matrix, perform baseline correction on the low-frequency component, and generate a low-frequency correction component. (4) Denoising dataset construction: The processed low-frequency correction component is reconstructed with the retained high-frequency component to generate a denoised dataset, completing the noise filtering and preliminary baseline correction of the initial detection data.

[0088] To address the potential nonlinear baseline drift in the denoised dataset, a particle swarm optimization algorithm is used to find optimal parameters, perform nonlinear correction on the data, and generate a corrected dataset, further improving data accuracy. The baseline drift model is a mathematical model constructed with the minimum mean square error between the sensor signal and the theoretical baseline as the objective function, describing the baseline drift characteristics of the data. The theoretical baseline is pre-set based on sensor characteristics and measurement conditions, such as a stable zero potential or a specific concentration-signal curve. The particle swarm optimization algorithm simulates the foraging behavior of a flock of birds, treating the parameters of the baseline drift model as particle positions. Through an iterative search in the solution space, the particles' positions and velocities are continuously updated, aiming to minimize the objective function and find the optimal parameter combination.

[0089] The implementation method of nonlinear correction of denoised data sets using particle swarm optimization algorithm is as follows:

[0090] (1) Baseline drift model construction: Based on the characteristics of the denoised dataset and the sensor type, an appropriate nonlinear function (such as a polynomial function, exponential function, or piecewise function) is selected to construct the baseline drift model. The sum of the squares of the differences between the sensor signal and the theoretical baseline is used as the objective function, namely the minimum mean square error (MSE). The goal is to minimize the MSE by adjusting the model parameters.

[0091] (2) Particle Swarm Optimization Algorithm Initialization: 1) Set the number of particles (e.g., 20-50), the maximum number of iterations (e.g., 100-200), the inertia weight, the learning factor, and other parameters. 2) Randomly initialize the position and velocity of each particle in the parameter space. The position of each particle corresponds to a set of parameters of the baseline drift model.

[0092] (3) Iterative solution: 1) Calculate the objective function value corresponding to each particle position (i.e., the MSE under the current parameters), and record the particle's historical optimal position and global optimal position. 2) Update the particle's velocity and position based on the particle's historical optimal position, global optimal position, and current velocity. The velocity update formula combines inertia, self-cognition, and social learning factors, while the position update is based on the new velocity. 3) Repeat the above process until the maximum number of iterations is reached or the convergence condition is met (e.g., the change in the objective function value is less than a threshold), and the optimal parameter combination is obtained.

[0093] (4) Nonlinear correction: Substitute the optimal parameters obtained by the particle swarm optimization algorithm into the baseline drift model to perform nonlinear correction on the denoised data set. Calculate the corrected signal value based on the model, generate a corrected data set, and complete the refined processing of the data.

[0094] Preferably, the specific steps of step S31 are:

[0095] Step S311: Use a wavelet transform algorithm to perform multi-scale decomposition on the time domain signal of the initial detection data set to remove noise components and generate high-frequency components and low-frequency components;

[0096] Step S312: performing baseline correction on the low-frequency component using an adaptive Kalman filter algorithm to generate a low-frequency correction component;

[0097] Step S313: Use the low-frequency correction component and the high-frequency component to construct a denoised data set.

[0098] In an embodiment of the present invention, an appropriate wavelet basis function, such as db4 or sym8, is selected based on the data characteristics and noise level. The number of decomposition layers is determined experimentally or empirically, generally selecting a number that effectively separates noise while preserving signal characteristics. For example, for spectral signals containing random noise, a four-layer decomposition can be selected. The selected wavelet basis functions are used to perform a multi-scale decomposition of the time domain signal of the initial detection dataset. At each decomposition layer, the signal is decomposed into a low-frequency approximation component and a high-frequency detail component. The low-frequency approximation component can then be decomposed into the next layer. After multiple layers of decomposition, high-frequency components at different scales and a final low-frequency component are obtained. The decomposed high-frequency components are then thresholded. Common thresholding methods include hard thresholding and soft thresholding. The hard thresholding method sets coefficients with absolute values ​​less than the threshold to zero and retains coefficients greater than the threshold. The soft thresholding method shrinks coefficients greater than the threshold to a fixed value toward zero. By properly setting the threshold, the noise component in the high-frequency component is removed or suppressed, resulting in denoised high-frequency and low-frequency components.

[0099] The implementation process of using the adaptive Kalman filter algorithm to perform baseline correction on the low-frequency component is as follows:

[0100] (1) Establish a state space model: define the state equation and observation equation of the low-frequency component. The state equation describes the dynamic change law of the low-frequency signal, such as x k =Ax k-1 +w k-1 , where x k is the state at the current moment; A is the state transfer matrix; w k-1 is the process noise; the observation equation represents the relationship between sensor measurement and state, such as z k =Hx k +v k , where z k is the observation value; H is the observation matrix; v k is the observation noise.

[0101] (2) Initialization parameters: Setting the initial state estimate Initial state estimation covariance matrix P0, process noise covariance matrix Q and observation noise covariance matrix R. These parameters can be set based on experience or prior knowledge, or estimated through data preprocessing.

[0102] (3) Iterative correction: At each time step k, according to the current observation value z k and the state estimate at the previous moment The adaptive Kalman filter algorithm is used to perform the following calculations:

[0103] 1) Prediction step: Calculate the state prediction value and the predicted covariance matrix P k|k-1 =APk-1|k-1 A T +Q.

[0104] 2) Update step: Calculate the filter gain K k =P k|k-1 H T (HP k|k-1 H T +R) -1 , update the state estimate Update the covariance matrix P k|k = = (IK k H)P k|k-1 .

[0105] (4) Generate low-frequency correction component: After multiple iterations, the corrected low-frequency component estimate, i.e., the low-frequency correction component, is obtained, completing the correction of the low-frequency signal baseline.

[0106] The low-frequency correction components, which have undergone wavelet transform denoising and adaptive Kalman filter baseline correction, are reconstructed with the retained high-frequency components to restore the signal's time-domain characteristics, generating a denoised dataset after noise removal and baseline correction. This is achieved by performing an inverse wavelet transform on the low-frequency correction components and the high-frequency components after thresholding using the same wavelet basis functions used during decomposition. The low-frequency and high-frequency components are then gradually merged in the reverse order of decomposition to reconstruct the time-domain signal. The reconstructed signal obtained by the inverse wavelet transform is then organized according to the format and order of the original data to generate a denoised dataset. This dataset, containing the noise-removed and baseline-corrected detection data, can be used for subsequent analysis and processing.

[0107] Preferably, the specific steps of step S32 are:

[0108] Step S321: construct a baseline drift model with the minimum mean square error between the sensor signal and the theoretical baseline as the objective function;

[0109] Step S322: using a particle swarm optimization algorithm to iteratively solve the objective function of the baseline drift model to generate optimal parameters;

[0110] Step S323: Perform nonlinear correction on the denoised data set using the optimal parameters to generate a corrected data set.

[0111] In an embodiment of the present invention, an objective function is constructed based on the difference between the sensor signal and the theoretical baseline. The sensor signal is made as close to the theoretical baseline as possible by adjusting the model parameters. The minimum mean square error is used as an indicator to measure the difference between the two, reflecting the degree of fit of the model to the data. The specific model construction process is: (1) Determine the theoretical baseline: Determine the theoretical baseline according to the type of sensor and the measurement principle. For example, for the voltage signal measured by the electrochemical sensor, the stable potential in the absence of the measured substance can be used as the theoretical baseline; for the spectral sensor, the spectral curve under the background of a pure solvent can be used as the theoretical baseline. (2) Select the model form: According to the characteristics of the data and the characteristics of the baseline drift, select a suitable nonlinear function form to construct the baseline drift model. Common model forms include polynomial models (such as quadratic polynomials, exponential models, piecewise function models, etc.). (3) Construct the objective function: Define the objective function as the minimum mean square error (MSE) between the sensor signal and the theoretical baseline. The formula is Where n is the number of data points; y i is the actual sensor signal value; The signal value predicted by the baseline drift model. The goal is to minimize the MSE by adjusting the model parameters, thereby accurately describing and correcting the baseline drift.

[0112] The particle swarm optimization algorithm is used to search for the optimal parameter combination in parameter space to minimize the baseline drift model's objective function (minimum mean square error), thereby improving the model's accuracy in fitting the data and achieving effective nonlinear correction of the data. The particle swarm optimization algorithm simulates the behavior of a flock of birds searching for food in space, treating each particle as a potential solution in the solution space (i.e., a parameter combination for the baseline drift model). Particles continuously adjust their positions based on their own experience and group information, moving closer to the optimal solution. Ultimately, the parameter combination that minimizes the objective function is found, generating the optimal parameters.

[0113] The baseline drift model determined by the optimal parameters is used to calculate each data point in the denoised data set, and the original data is adjusted according to the baseline value predicted by the model so that the data conforms to the theoretical baseline and nonlinear correction is achieved. The specific nonlinear correction process is as follows: (1) The optimal parameters obtained by the particle swarm optimization algorithm are substituted into the baseline drift model constructed in step S321 to determine the specific model expression. (2) For each data point in the denoised data set, the corresponding theoretical baseline value is calculated according to the baseline drift model. The original data point is compared with the theoretical baseline value, and the original data is corrected by subtraction or other appropriate operations to obtain the corrected data point. (3) After all the data points in the denoised data set have been corrected, they are sorted according to the order and format of the original data to generate a corrected data set. This data set contains the detection data after noise filtering, baseline correction and nonlinear correction, and can be used for subsequent spatiotemporal dimension analysis and risk assessment operations.

[0114] Preferably, Figure 5 The flowchart of step S4 of a method for rapid detection of water pollutants is shown.

[0115] In an embodiment of the present invention, the target detection data set obtained after processing in step S3 is sorted out to clarify the sensor type (spectral sensor module, electrochemical sensor module, biosensor module, etc.), time series (the order of time points of detection), and pollutant indicators (such as heavy metal ion concentration, organic pollutant content, etc.) contained in the data. Use a programming language or tool that supports tensor operations (such as the NumPy library and PyTorch framework in Python) to construct the sorted data into a three-dimensional tensor. Assuming that there are a types of sensors, the time series contains b time points, and there are c pollutant indicators, the dimension of the constructed three-dimensional tensor is a×b×c, and each tensor element corresponds to a pollutant indicator detection value at a specific sensor type and time point.

[0116] Select an implementation of the Tucker decomposition algorithm in the existing tensor decomposition algorithm library, such as the TensorLy library. Set appropriate decomposition parameters based on the scale and characteristics of the target detection dataset. For example, to determine the core tensor dimension to be retained, the appropriate dimension value can generally be determined by cross-validation and other methods to compress the data while retaining key information as much as possible. Input the three-dimensional tensor obtained in step S41 into the Tucker decomposition algorithm, and the algorithm automatically calculates and outputs the core tensor and the corresponding factor matrix. The dimension of the core tensor is usually smaller than the original three-dimensional tensor, but it contains the core features of the data.

[0117] A multi-head attention mechanism model is built based on a deep learning framework (such as TensorFlow, PyTorch). The number of heads is set (such as 8-16 heads), and each head calculates attention in a different subspace. The core tensor obtained in step S42 is used as the input of the multi-head attention mechanism, and the time dimension of the core tensor is processed. During the calculation process, each head calculates the attention score for each element of the time dimension based on the learned parameters. The attention scores calculated by each head are weighted and summed to obtain the dynamic weight of each time point. The time dimension of the core tensor is weighted according to these weights, and then fused with other dimensional information to construct a spatiotemporal fusion matrix. This matrix comprehensively considers the spatiotemporal characteristics of the data, providing richer information for subsequent analysis.

[0118] Leveraging the powerful feature extraction capabilities of a multi-layer, fully connected network, high-dimensional features are extracted from the spatiotemporal fusion matrix and separated into feature vectors reflecting pollutant information and feature vectors representing environmental interference, achieving a preliminary decoupling of data features. Specifically, the multi-layer, fully connected network abstracts and extracts features from the input data layer by layer through the connection of multiple neuron layers. Different neurons learn different characteristic patterns during training, enabling the extraction of valuable high-dimensional features from the complex spatiotemporal fusion matrix. These features are then classified according to their properties, resulting in water quality pollutant data and dynamic risk assessment indicators.

[0119] Preferably, the specific steps of step S44 are:

[0120] Step S441: Use a multi-layer fully connected network to extract high-dimensional features in the spatiotemporal fusion matrix to generate pollutant feature vectors and environmental interference feature vectors;

[0121] Step S442: Normalize the time gradient of the pollutant feature vector to generate a pollutant feature;

[0122] Step S443: weighting the spatial distribution of the pollutant features according to the sensor types corresponding to the pollutant features to generate pollution weighted features;

[0123] Step S444: Asymmetrically suppress the environmental interference feature vector to generate an interference suppression feature;

[0124] Step S445: Evaluate the spatial details and temporal continuity of the pollution weighted features and the interference suppression features through a reflection image discriminator and a gradient map discriminator to generate spatial authenticity score data and temporal continuity score data;

[0125] Step S446: Asymmetric weight adjustment is performed on the pollution weighted features and the interference suppression features according to the spatial authenticity score data and the temporal continuity score data to construct a pollutant feature time series;

[0126] Step S447: Capture the forward accumulation and backward diffusion patterns of pollutant concentrations in the pollutant characteristic time series through a bidirectional gate cycle unit to generate water quality pollutant data and dynamic risk assessment indicators.

[0127] In an embodiment of the present invention, a multi-layer fully connected network structure is designed to determine the number of layers (e.g., 3-5 layers) and the number of neurons in each layer. For example, the first layer can be set to 256 neurons, and the subsequent layers can be gradually reduced to 64 neurons. The spatiotemporal fusion matrix obtained in step S43 is used as the input of the multi-layer fully connected network, and the network is trained using labeled training data (if historical data is available for training). During the training process, the network learns to extract high-dimensional features and outputs two feature vectors, corresponding to the pollutant feature vector and the environmental interference feature vector, respectively.

[0128] To calculate the gradient of the pollutant signature vector in the time dimension, numerical differentiation methods such as forward differencing, backward differencing, or central differencing can be used to obtain the time gradient value at each time point. A normalization formula, such as the minimum-maximum normalization formula, is used to map the time gradient value to a specified interval to generate the normalized pollutant signature.

[0129] A weight is assigned to each sensor type based on factors such as sensor performance parameters and the accuracy of historical detection data. For example, spectral sensors are more accurate in detecting certain organic pollutants and can be assigned a higher weight; biosensors, on the other hand, are more advantageous in detecting microbial indicators and are accordingly assigned an appropriate weight. The pollutant signatures obtained in step S442 are weighted according to their corresponding sensor types and the corresponding weights. The characteristic values ​​of each pollutant indicator under different sensors are weighted and summed to obtain a weighted pollution signature.

[0130] By calculating certain statistics (such as variance and absolute value) of the environmental interference feature vector, the importance of each feature is assessed. Based on the importance assessment results, an asymmetric suppression function is designed. For example, for interference features with high importance, a larger suppression coefficient is used for weighting; for interference features with low importance, a smaller suppression coefficient is used or no suppression is performed, thus generating an interference suppression feature.

[0131] Based on deep learning models (such as convolutional neural networks), a reflectance image discriminator and a gradient map discriminator are constructed. The reflectance image discriminator processes two-dimensional spatial feature data, while the gradient map discriminator processes gradient data in the temporal dimension. Contamination-weighted features and interference suppression features are input into the reflectance image discriminator and the gradient map discriminator, respectively. The reflectance image discriminator outputs a spatial authenticity score, which measures the spatial rationality of the feature; the gradient map discriminator outputs a temporal continuity score, which measures whether the feature changes continuously and rationally over time, thereby generating corresponding scoring data.

[0132] Based on the spatial authenticity score and temporal continuity score data, weights are calculated for the pollution-weighted and interference-rejection features. For example, features with high spatial authenticity and temporal continuity scores are assigned higher weights, while features with low scores are assigned lower weights. The pollution-weighted and interference-rejection features are weighted and combined according to the calculated weights to obtain adjusted feature values. These feature values ​​are arranged in chronological order to construct a pollutant feature time series.

[0133] Construct a bidirectional gate recurrent unit model based on the deep learning framework, and set the appropriate number of hidden layer neurons (such as 128-256) and the number of layers (such as 1-2 layers). Use the pollutant characteristic time series obtained in step S446 as the input of the bidirectional gate recurrent unit, and use historical data to train the model (if any). After the training is completed, the model outputs the predicted trend of pollutant concentration changes, and generates water quality pollutant data (such as pollutant concentration values ​​at different time points) based on these trends. At the same time, combined with the preset risk assessment rules and thresholds, calculate dynamic risk assessment indicators (such as risk level, risk probability, etc.).

[0134] Preferably, Figure 6 The flowchart of step S5 of a method for rapid detection of water pollutants is shown.

[0135] In an embodiment of the present invention, the ecotoxicity threshold refers to the critical concentration at which a pollutant has a harmful effect on an ecosystem. The actual detected pollutant concentration is compared with the threshold, and the excess multiple is calculated. Different risk levels are assigned according to the preset excess multiple intervals, thereby achieving a preliminary classification of water quality risks. Water quality pollutant data is obtained to clarify the actual detected concentration of each pollutant. At the same time, an ecotoxicity threshold database is established to include ecotoxicity thresholds corresponding to different pollutants (such as heavy metals, organic pollutants, etc.). The database can be constructed and updated based on international standards (such as WHO standards), national environmental quality standards (such as China's "Surface Water Environmental Quality Standards"), etc. For each pollutant, the excess multiple is calculated using the formula "exceedance = (actual concentration of pollutant - ecotoxicity threshold) / ecotoxicity threshold". If the calculated result is less than 0, it means that the standard is not exceeded and is recorded as 0.

[0136] Multiple exceedance multiples are preset, and each interval is assigned an initial risk level. For example: if the exceedance multiple is ≤ 0.5, the initial risk level is "low"; if the exceedance multiple is 0.5 < ≤ 1.5, the initial risk level is "medium"; if the exceedance multiple is 1.5 < ≤ 3, the initial risk level is "high"; if the exceedance multiple is > 3, the initial risk level is "very high". If multiple pollutants are tested, a comprehensive assessment method can be used to determine the final initial risk level. For example, the highest risk level among all pollutants can be used as the comprehensive initial risk level for the water sample; or different weights can be assigned according to the degree of harm of the pollutants to calculate the weighted average risk level.

[0137] By combining geospatial information and pollutant characteristics, and considering the varying sensitivities of different water functional zones to pollutants, a dynamic weight is assigned to each pollutant, generating a spatially weighted risk coefficient that better reflects actual environmental risks. Water functional zoning data in the Geographic Information System (GIS) (e.g., drinking water sources, fishery water areas, industrial water areas, etc.) clearly defines the different water quality requirements for different regions. The attention mechanism automatically learns and dynamically adjusts the weights of different pollutants in risk assessment based on water functional zoning and pollutant characteristics, ensuring that the risk coefficient is more aligned with actual environmental needs.

[0138] The dynamic risk assessment indicators generated in step S447 (e.g., pollutant concentration trends, risk probabilities, etc.) are spatially correlated with the water function zoning data in the GIS. For example, the coordinates of the test point are matched with the water function zone boundaries in the GIS layer to determine the type of water function zone to which the test point belongs. A data fusion model is established to integrate the dynamic risk assessment indicators and water function zoning data into a single data structure to facilitate subsequent analysis.

[0139] A pollutant-water functional zone sensitivity matrix is ​​constructed, pre-setting the basic sensitivity weights of different pollutants in different water functional zones. For example, the basic sensitivity weights of heavy metal pollutants in drinking water sources are set to high (e.g., 0.8-1.0), and those in industrial water areas are set to medium (e.g., 0.4-0.6). Using an attention mechanism model (e.g., a Transformer-based attention network), with dynamic risk assessment indicators and water functional zoning data as input, the model is trained to learn the importance of different pollutants in the current water functional zone. During training, the model automatically adjusts weights based on the input data, ensuring that the weights reflect the actual environmental sensitivity of the pollutants in the current scenario. The dynamic environmental sensitivity weight for each pollutant is output, ranging from 0 to 1, with larger values ​​indicating greater environmental sensitivity in the current water functional zone. For each pollutant, its dynamic risk assessment indicator (e.g., risk probability) is multiplied by the corresponding environmental sensitivity weight. The results for all pollutants are then summed to obtain the spatially weighted risk coefficient.

[0140] Leveraging the experience and patterns accumulated from historical pollution incidents, we refine the initial risk rating, compensating for assessment biases caused by current data limitations and ensuring the final risk rating more accurately reflects the actual risk situation. The historical pollution incident database records the risk development process and ultimate impact of similar pollution scenarios. Using transfer learning techniques, we apply risk patterns from historical data to the current detection scenario, compare the current data with historical patterns, and adjust the initial risk rating accordingly.

[0141] Collect and organize historical pollution event data, including information such as the time, location, type of pollutant, concentration changes, and ecological impacts of the incident. Clean, label, and classify the data to build a structured historical pollution event database. Extract features of water quality pollutant data and dynamic risk assessment indicators of the current detection scene. The extracted features include pollutant concentration, change trend, spatial location, etc. Search the historical pollution event database for historical events with similar characteristics to the current scene. Similarity calculation methods (such as cosine similarity, Euclidean distance, etc.) can be used to calculate the similarity between the current scene characteristics and the historical event characteristics, and screen out historical events with higher similarity as references.

[0142] Analyze the risk development patterns and final risk levels of the selected historical events, and determine a correction strategy for the initial risk level based on the characteristics of the current scenario. For example, if similar pollutant concentrations and trends in historical events ultimately lead to an escalation in the risk level, the current initial risk level will be increased accordingly. Conversely, if historical events indicate that the risk was effectively controlled, the initial risk level can be appropriately lowered. Based on the correction strategy, the initial risk level determined in step S51 is adjusted to generate a target risk level. Limits can be set during the adjustment process to prevent over-correction.

[0143] Using three-dimensional visualization technology, data such as the target risk level, spatially weighted risk coefficient, and geographic spatial information are mapped to a three-dimensional coordinate system, and the risk differences between different regions are displayed through the color, shape, size, and other attributes of the graphics. Specifically, using three-dimensional visualization technology, data such as the target risk level, spatially weighted risk coefficient, and geographic spatial information are mapped to a three-dimensional coordinate system, and the risk differences between different regions are displayed through the color, shape, size, and other attributes of the graphics. Use professional three-dimensional visualization software (such as ArcGIS Pro, Blender, etc.) or programming libraries (Python's Plotly, Mayavi, etc.) to build a three-dimensional visualization model. In three-dimensional space, the location of the detection point is determined by geographic coordinates, the target risk level is mapped to the color attribute (such as red for high risk and green for low risk), and the spatially weighted risk coefficient is mapped to the size of the graphic (the larger the value, the larger the graphic). Geographic information such as water body functional zoning boundaries and terrain can be added as a background to enhance the readability of the visualization effect. Add interactive features to the 3D visualization report, such as hovering the mouse to display detailed risk information (including pollutant type, concentration, and risk level), zooming and panning functions, and switching between different perspectives, to facilitate users to observe and analyze risk distribution from different angles. The constructed 3D visualization model can be exported as an image, video, or interactive webpage format, integrated with relevant text descriptions (such as assessment methods, data sources, and risk response recommendations) to generate a multidimensional risk assessment report. The report can be published and shared through online platforms, mobile applications, and other channels.

[0144] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for rapid detection of water pollutants, characterized in that: The following steps are involved: Step S1: removing large particles of impurities and suspended matter from the water sample to be tested to obtain a target water sample; Step S2: performing simultaneous detection of multiple pollutant indicators on the target water sample through a multi-source sensor array to generate an initial detection data set; Step S3: performing noise filtering and baseline correction on the initial detection data set to generate a target detection data set; Step S4: Dynamically fuse and decouple the target detection dataset in spatiotemporal dimensions to generate water quality pollutant data and dynamic risk assessment indicators; Step S5: construct a multidimensional risk assessment report using the water pollutant data and the dynamic risk assessment indicators.

2. The method according to claim 1, characterized in that The specific steps of step S1 are: Step S11, passing the water sample to be tested through a dynamic gradient filtration device to remove large particle impurities to generate an initial water sample, wherein the dynamic gradient filtration device includes a nylon mesh, a polyethersulfone filter membrane, and a ceramic membrane with successively smaller mesh sizes; Step S12, introducing the initial water sample into a magnetic graphene oxide adsorption column with surface modified polydopamine to adsorb suspended particles and hydrophobic organic pollutants to obtain an intermediate water sample; Step S13: introducing the intermediate water sample into the microfluidic chip, utilizing the turbulent effect to peel off the magnetic graphene oxide fragments remaining in the adsorption column, and intercepting the fragments through the magnetic screen at the outlet to obtain the target water sample.

3. The method according to claim 1, characterized in that The multi-source sensor array includes a spectral sensor module, an electrochemical sensor module and a biosensor module; The spectral sensor module includes a 90° scattered light detector, an ultraviolet-visible spectrophotometer sensor and a Raman spectroscopy sensor, which are used to detect characteristic spectral signals of organic pollutants and heavy metal ions; The electrochemical sensor module includes a molecularly imprinted polymer modified glassy carbon working electrode and a self-repairing Ag / AgCl reference electrode, which is used to selectively detect heavy metal ions in a complex ionic environment; The biosensor module comprises a PDMS microfluidic chip, a multi-enzyme immobilization detection unit and a thin film temperature control module, and is used for synchronously monitoring microbial metabolites and toxic substances.

4. The method according to claim 1, wherein The specific steps of step S2 are: Step S21: Detecting the target water sample through the spectral sensor module of the multi-source sensor array to generate a spectral detection sub-dataset; Step S22: detecting the target water sample through the electrochemical sensor module of the multi-source sensor array to generate an electrochemical detection sub-dataset; Step S23: Detecting the target water sample through the biosensor module of the multi-source sensor array to generate a biodetection sub-dataset; Step S24: aligning the timestamps of the spectral detection sub-dataset, the electrochemical detection sub-dataset, and the biological detection sub-dataset to generate an initial detection data set.

5. The method according to claim 1, wherein The specific steps of step S3 are: Step S31: Using a wavelet transform algorithm and an adaptive Kalman filter algorithm to perform noise filtering and baseline correction on the initial detection data set to generate a denoised data set; Step S32: Using a particle swarm optimization algorithm to perform nonlinear correction on the denoised data set to generate a corrected data set.

6. The method according to claim 5, characterized in that The specific steps of step S31 are: Step S311: Using a wavelet transform algorithm to perform multi-scale decomposition on the time domain signal of the initial detection data set to remove noise components and generate high-frequency components and low-frequency components; Step S312: performing baseline correction on the low-frequency component using an adaptive Kalman filter algorithm to generate a low-frequency correction component; Step S313: construct a denoising data set using the low-frequency correction component and the high-frequency component.

7. The method according to claim 5 or 6, characterized in that The specific steps of step S32 are: Step S321: construct a baseline drift model with the minimum mean square error between the sensor signal and the theoretical baseline as the objective function; Step S322: using a particle swarm optimization algorithm to iteratively solve the objective function of the baseline drift model to generate optimal parameters; Step S323: Perform nonlinear correction on the denoised data set using the optimal parameters to generate a corrected data set.

8. The method according to claim 1, characterized in that The specific steps of step S4 are: Step S41: Divide the target detection dataset into three-dimensional tensors, where the dimensions corresponding to the three-dimensional tensors are sensor type, time series, and pollutant index respectively; Step S42: extracting the core tensor from the three-dimensional tensor using Tucker decomposition algorithm; Step S43: Dynamically weight the time dimension of the core tensor through a multi-head attention mechanism and construct a matrix to generate a spatiotemporal fusion matrix; Step S44: Decouple the spatiotemporal fusion matrix to generate water quality pollutant data and dynamic risk assessment indicators.

9. The method according to claim 8, characterized in that The specific steps of step S44 are: Step S441: using a multi-layer fully connected network to extract high-dimensional features in the spatiotemporal fusion matrix to generate pollutant feature vectors and environmental interference feature vectors; Step S442: Normalize the time gradient of the pollutant feature vector to generate a pollutant feature; Step S443: weighting the spatial distribution of the pollutant feature according to the sensor type corresponding to the pollutant feature to generate a pollution weighted feature; Step S444: asymmetrically suppress the environmental interference feature vector to generate an interference suppression feature; Step S445: Evaluate the spatial details and temporal continuity of the pollution weighted features and the interference suppression features through a reflection image discriminator and a gradient map discriminator to generate spatial authenticity scoring data and temporal continuity scoring data; Step S446: Asymmetric weight adjustment is performed on the pollution weighted feature and the interference suppression feature according to the spatial authenticity score data and the temporal continuity score data to construct a pollutant feature time series; Step S447: Capture the forward accumulation and backward diffusion patterns of pollutant concentrations in the pollutant characteristic time series through a bidirectional gate cycle unit to generate water quality pollutant data and dynamic risk assessment indicators.

10. The method according to claim 1, characterized in that The specific steps of step S5 are: Step S51: determining an initial risk level according to the ratio of the pollutant concentration exceeding the standard in the water quality pollutant data to the ecotoxicity threshold; Step S52: integrating the dynamic risk assessment index with the water body functional zoning data in the geographic information system, and dynamically assigning environmental sensitivity weights of different pollutants through an attention mechanism to generate a spatially weighted risk coefficient; Step S53: Migrate the risk model in the historical pollution event database to the current scenario, correct the deviation of the initial risk level, and generate a target risk level; Step S54: construct a three-dimensional visualization report using the target risk level and the spatially weighted risk coefficient to generate a multi-dimensional risk assessment report.

Images