Artificial intelligence-based synchronous absorption-three-dimensional fluorescence water quality monitoring, early warning and tracing system

Abstract

This invention relates to the field of industrial internet and environmental monitoring technology, and discloses an artificial intelligence-based synchronous absorption-three-dimensional fluorescence water quality monitoring, early warning, and source tracing system, comprising a physical sensing layer, a SaaS cloud platform layer, and a business decision layer. The physical sensing layer synchronously collects spectral data through a synchronous absorption-three-dimensional fluorescence spectrometer, corrects for internal filtering effects through an edge computing unit, and generates high-fidelity multidimensional spectral data. The SaaS cloud platform layer processes data based on a fingerprint database, analysis, and source tracing model, and synchronizes device status using a digital twin. The business decision layer consists of coupled holographic monitoring, dynamic early warning, precise source tracing, and closed-loop control units, enabling fingerprint anomaly detection, graded early warning, blind source separation and fingerprint matching, and hardware-software integrated control. The system achieves fully automated closed-loop monitoring, source tracing, and integrated control, improves data acquisition fidelity, provides interpretable source tracing evidence, reduces emergency response time to minutes, and provides scientific and efficient technical support for environmental supervision.

Description

Technical Field

[0001] This invention relates to the field of industrial internet and environmental monitoring technology, specifically to an artificial intelligence-based synchronous absorption-three-dimensional fluorescence water quality monitoring, early warning and source tracing system. Background Technology

[0002] Water is the source of life and a vital resource for economic and social development. Currently, water pollution is becoming increasingly prominent, especially the illegal discharges and leaks from industrial parks and key polluting enterprises, posing a serious threat to aquatic ecological security and human health. At present, ecological and environmental supervision is shifting from simply ensuring water quality meets standards to precise source tracing and risk prevention. However, the existing water quality monitoring system generally suffers from serious data and operational gaps, specifically manifested in the following ways: A capability gap exists between monitoring capabilities and the need for source tracing: Traditional water quality monitoring methods primarily rely on recording the concentrations of conventional indicators (such as chemical oxygen demand (COD), ammonia nitrogen, and total phosphorus). This concentration-only approach lacks the ability to provide a finely detailed fingerprint-like profile of the complex organic components in water bodies. When sudden pollution incidents occur, traditional monitoring equipment can only provide delayed alarms indicating exceedances, failing to answer crucial questions such as what the pollutants are and where the pollution sources originate, leading to a disconnect between monitoring and source tracing operations.

[0003] Distortion of front-end sensing data: Existing spectral water quality monitoring equipment is mostly single-function (monitoring only COD / ammonia nitrogen or only collecting three-dimensional fluorescence). More importantly, traditional equipment is generally severely affected by the internal filtration effect, and the absorption spectrum and fluorescence spectrum are not collected synchronously. This results in significant deviations in the raw data uploaded to the analysis center, making it unable to support high-precision source apportionment and AI model training—essentially, garbage in, garbage out.

[0004] The fragmented control chain and lack of a closed-loop mechanism are problematic: most water quality monitoring devices on the market currently function as data loggers. Monitoring data is simply stored in databases and cannot be automatically converted into control commands. When pollution occurs, the process from detecting anomalies, manually analyzing data, identifying the pollution source, to shutting off discharge valves often requires hours or even days of manual intervention. This situation, characterized by difficulties in pollution identification, source tracing, and delayed emergency response, urgently necessitates a system with automated closed-loop capabilities encompassing monitoring, early warning, source tracing, and coordinated control.

[0005] Therefore, developing a new water pollution early warning and source tracing system that can achieve high-fidelity data acquisition, intelligent and accurate source tracing, and automated closed-loop management is a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] The purpose of this invention is to overcome the shortcomings of the prior art and provide an artificial intelligence-based synchronous absorption-three-dimensional fluorescence water quality monitoring, early warning and source tracing system.

[0007] To achieve the above-mentioned objectives, the present invention provides the following technical solution: The AI-based synchronous absorption-three-dimensional fluorescence water quality monitoring, early warning, and source tracing system includes a physical sensing layer, a SaaS cloud platform layer, and a business decision-making layer. The physical sensing layer includes at least one synchronous absorption-three-dimensional fluorescence spectrometer, which includes a miniaturized synchronous spectral detection module. The analyzer is used to simultaneously acquire the absorption spectrum and three-dimensional fluorescence spectrum of the water body to be tested, and to use an edge computing unit to perform internal filtration effect correction on the three-dimensional fluorescence spectrum based on the absorption spectrum to generate high-fidelity multidimensional spectral data. The SaaS cloud platform layer is communicatively connected to the physical perception layer and includes a multi-user fingerprint database, a multi-dimensional spectral water quality analysis model, and a multi-dimensional spectral source tracing model. The SaaS cloud platform layer is used to receive the high-fidelity multi-dimensional spectral data, call the multi-dimensional spectral water quality analysis model to predict water quality indicators, and call the multi-dimensional spectral source tracing model to compare with the multi-user fingerprint database and output the source tracing results. The business decision-making layer, running on the SaaS cloud platform, includes multiple interconnected business intelligent agent units, including: a holographic monitoring unit for calculating the spatial distance of water quality fingerprint vectors to determine water quality fingerprint anomalies; a dynamic early warning unit for providing tiered early warnings; a precise source tracing unit for activating the AI ​​demixing engine for blind source separation and fingerprint matching; and a closed-loop control unit for generating and issuing equipment control commands and management work orders based on the source tracing results.

[0008] As a further description of the technical solution of the present invention, the miniaturized synchronous spectral detection module includes: a xenon lamp device, a Fresnel lens, a grating monochromator, an optical fiber coupling lens, a multimode optical fiber, a spectrometer, a cuvette, and a photodiode module; The light emitted by the xenon lamp is focused by the Fresnel lens and enters the grating monochromator, which outputs monochromatic excitation light to illuminate the cuvette. In the direction parallel to the incident light, the photodiode module detects the intensity of the transmitted light to obtain absorption information; in the direction perpendicular to the incident light, the fiber-coupled lens and the multimode fiber guide the fluorescence signal into the spectrometer to obtain fluorescence information.

[0009] As a further description of the technical solution of the present invention, the internal filtering effect correction performed by the edge computing unit specifically includes: Acquire absorption spectra acquired synchronously and original fluorescence spectrum ; According to the formula Calculate the corrected fluorescence intensity ,in, To excite the wavelength, For the emission wavelength, This indicates the absorbance at the excitation wavelength. This indicates the absorbance at the emission wavelength.

[0010] As a further description of the technical solution of the present invention, the holographic monitoring unit is configured as follows: Calculate the Mahalanobis or Euclidean distance between the current water quality fingerprint vector and the preset normal background baseline in real time; When the calculated distance exceeds a preset threshold, it is determined to be an anomaly in the water quality fingerprint and the dynamic early warning unit is triggered.

[0011] As a further description of the technical solution of the present invention, the precise traceability unit is configured as follows: Receive mixed spectral data during abnormal periods; The mixed spectral data is subjected to blind source separation using a nonnegative matrix factorization algorithm to extract independent spectral components. The extracted spectral components are compared with the multi-user fingerprint database to calculate the similarity and output structured pollution source information, which includes the pollution source industry, characteristic pollutant components, and similarity.

[0012] As a further description of the technical solution of the present invention, the closed-loop control unit is configured to perform hard control and soft control; The hard-link control includes sending instructions to a field programmable logic controller (PLC) via an industrial protocol to perform physical actions such as closing the sewage valve or starting the intercepting pump. The soft-linkage control includes: automatically generating an enforcement report containing a spectral evidence chain and source tracing conclusions, and pushing it to the regulatory platform via an application programming interface (API) to generate a pending work order.

[0013] As a further description of the technical solution of the present invention, the SaaS cloud platform layer also includes a digital twin, which is used to synchronize the physical parameters and operating status of the synchronous absorption-three-dimensional fluorescence spectrometer in real time. The system also includes a predictive maintenance module, which is used to calculate the energy decay slope and predict the remaining lifespan of the light source by comparing the current light source energy curve with the factory baseline curve in real time based on the digital twin, so as to generate a maintenance work order.

[0014] As a further description of the technical solution of the present invention, the working method of the system includes: Step A: Simultaneously acquire the absorption spectrum and three-dimensional fluorescence spectrum of the water body to be tested using a synchronous absorption-three-dimensional fluorescence spectrometer, and perform internal filtration effect correction on the fluorescence spectrum based on the absorption spectrum at the edge end to generate high-fidelity multidimensional spectral data and upload it to the SaaS cloud platform; Step B: The holographic monitoring unit of the SaaS cloud platform calculates the vector space distance between the high-fidelity multidimensional spectral data and the normal background baseline in real time. If the distance exceeds the threshold, it is determined to be an anomaly in the water quality fingerprint. Step C: The dynamic early warning unit issues graded early warnings based on the degree of anomaly and triggers the precise source tracing unit; Step D: The precise source tracing unit performs blind source separation and fingerprint matching on the mixed spectrum during the abnormal period, and outputs structured source tracing conclusions; Step E: Based on the source tracing conclusion, the closed-loop control unit automatically generates and issues equipment control commands to physically block pollution, and at the same time generates and pushes law enforcement reports and management work orders.

[0015] As a further description of the technical solution of the present invention, step D further includes: Using digital twins to simulate the theoretical spectral morphology of pollution sources after diffusion; The theoretical spectral morphology is compared with the actual collected spectral morphology to verify the accuracy of the source tracing conclusion.

[0016] The beneficial effects of this invention are as follows: 1. Achieving a fully automated closed-loop process of monitoring, tracing, and joint control: This invention upgrades traditional passive data recording equipment into an active decision-making and execution system by introducing a business intelligence agent architecture. When pollution occurs, the system can autonomously complete fingerprint anomaly detection, tiered early warning, and source analysis, and automatically execute physical blocking actions such as closing valves and starting intercepting pumps within seconds, while generating an enforcement report. This completely changes the traditional manual emergency response model, shortening the emergency response time from several days to minutes, greatly improving the timeliness and deterrent effect of environmental supervision.

[0017] 2. Significantly Improves the Fidelity and Reliability of Front-End Data Acquisition: This invention innovates from the source of optical design, achieving simultaneous acquisition of absorption spectra and three-dimensional fluorescence spectra. More importantly, by embedding an internal filtering effect correction algorithm based on synchronous absorption spectra into the hardware (edge ​​computing), precise physical compensation is performed on the fluorescence signal. This on-the-fly acquisition and correction mechanism fundamentally solves the data distortion problem caused by the internal filtering effect in traditional equipment, providing a high-fidelity, low-noise data foundation for accurate decision-making by upper-layer AI models, avoiding the inflow and outflow of substandard data.

[0018] 3. Providing interpretable, chemically sound source tracing evidence: Unlike traditional black-box AI source tracing methods, this invention introduces blind source separation technology and a digital twin simulation verification mechanism into the source tracing unit. The system can separate independent chemical fingerprints from the spectra of mixed water bodies and match them with an industry-specific fingerprint database, outputting a structured conclusion that similar pollutants are disperse blue dye with a 99% matching degree. This source tracing result possesses a visualized spectral evidence chain, providing direct, scientific, and interpretable evidence for environmental law enforcement. Attached Figure Description

[0019] The present invention will now be further described with reference to the accompanying drawings.

[0020] Figure 1 This is a partial structural schematic diagram of the water pollution early warning and source tracing system based on three-dimensional fluorescent water quality fingerprinting provided by the present invention; Figure 2 This is a schematic diagram of the optical structure of the synchronous absorption-three-dimensional fluorescence spectrometer described in this invention; Figure 3 This is a schematic diagram of the intelligent water quality traceability interface deployed on a mobile device as an example of the process of this invention; Figure 4 This is a schematic diagram of the intelligent water quality traceability interface deployed on a PC, illustrating the process of this invention. Figure 1 ; Figure 5 This is a schematic diagram of the intelligent water quality traceability interface deployed on a PC, illustrating the process of this invention. Figure 2 ; Figure 6 This is a schematic diagram of the intelligent water quality traceability interface deployed on a PC, illustrating the process of this invention. Figure 3 .

[0021] Explanation of reference numerals in the attached figures: 1-Xenon lamp device; 2-Fresnel lens; 3-Grating monochromator; 4-Fiber optic coupling lens; 5-Multimode fiber; 6-Spectrometer; 7-Cuvette; 8-Photodiode module. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Example 1: System Overall Architecture Please see Figure 1As shown, this embodiment provides an artificial intelligence-based simultaneous absorption-three-dimensional fluorescence water quality monitoring, early warning, and source tracing system. The system adopts a three-layer architecture of edge-cloud-intelligence, including a physical sensing layer, a SaaS cloud platform layer, and a business decision-making layer. The physical sensing layer includes at least one synchronous absorption-three-dimensional fluorescence spectrometer, which includes a miniaturized synchronous spectral detection module. The analyzer is used to simultaneously acquire the absorption spectrum and three-dimensional fluorescence spectrum of the water body to be tested, and to use an edge computing unit to perform internal filtration effect correction on the three-dimensional fluorescence spectrum based on the absorption spectrum to generate high-fidelity multidimensional spectral data. The physical sensing layer comprises multiple simultaneous absorption-3D fluorescence spectrometers deployed at the main discharge outlets of industrial parks, downstream of key pollution sources, or at river sections. Each analyzer integrates an embedded edge computing motherboard.

[0024] The SaaS cloud platform layer is communicatively connected to the physical perception layer and includes a multi-user fingerprint database, a multi-dimensional spectral water quality analysis model, and a multi-dimensional spectral source tracing model. The SaaS cloud platform layer is used to receive the high-fidelity multi-dimensional spectral data, call the multi-dimensional spectral water quality analysis model to predict water quality indicators, and call the multi-dimensional spectral source tracing model to compare with the multi-user fingerprint database and output the source tracing results. The SaaS cloud platform layer is responsible for receiving data from all front-end devices. Within the cloud platform, a MySQL database is deployed to store raw data, a Redis cache is used for high-speed read and write operations, and a distributed file system is used to store a large 3D fluorescence fingerprint library. The core of the cloud platform is a pre-trained multidimensional spectral water quality analysis model and a source tracing model.

[0025] The business decision-making layer, running on the SaaS cloud platform, includes multiple interconnected business intelligent agent units, including: a holographic monitoring unit for calculating the spatial distance of water quality fingerprint vectors to determine water quality fingerprint anomalies; a dynamic early warning unit for providing tiered early warnings; a precise source tracing unit for activating the AI ​​demixing engine for blind source separation and fingerprint matching; and a closed-loop control unit for generating and issuing equipment control commands and management work orders based on the source tracing results.

[0026] The business decision-making layer is the core of this invention's innovation; it is a multi-agent system running on a cloud platform. This agent continuously monitors new data flowing into the data pipeline and invokes the cloud platform's model services to execute logic across four stages: monitoring, early warning, source tracing, and joint control.

[0027] Example 2: Synchronous Absorption-Three-Dimensional Fluorescence Spectrometer and Edge Correction 2.1 Miniaturized Synchronous Spectroscopic Detection Module Please see Figure 2As shown, the miniaturized synchronous spectral detection module includes: a xenon lamp device 1, a Fresnel lens 2, a grating monochromator 3, an optical fiber coupling lens 4, a multimode optical fiber 5, a spectrometer 6, a cuvette 7, and a photodiode module 8; The light emitted by the xenon lamp device 1 is focused by the Fresnel lens 2 and enters the grating monochromator 3. The grating monochromator 3 outputs monochromatic excitation light to illuminate the cuvette 7. Parallel to the incident light direction, the photodiode module 8 detects the transmitted light intensity to obtain absorption information; perpendicular to the incident light direction, the fiber optic coupling lens 4 and the multimode fiber 5 guide the fluorescence signal into the spectrometer 6 to obtain fluorescence information.

[0028] Working principle: The continuous broadband light emitted by the xenon lamp device 1 is efficiently focused and compressed into a spot by a combination of a specially designed internal reflector bowl and Fresnel lens 2, and then enters the entrance slit of the Czerny-Turner structure grating monochromator 3. This structural design greatly reduces the space occupied by the optical path.

[0029] The grating monochromator 3 can output monochromatic excitation light of a specific wavelength (e.g., starting from 200 nm with a step size of 5 nm) by rotating the internal grating. This monochromatic light is then irradiated onto a cuvette 7 filled with the water sample to be tested.

[0030] In the direction parallel to the incident light, photodiode module 8 detects the intensity of the transmitted light. By comparing the incident light intensity and the transmitted light intensity, the absorption spectrum information at that wavelength can be calculated.

[0031] The fluorescence signal generated by the water sample being excited in a direction perpendicular to the incident light is collected by a large numerical aperture fiber-coupled lens 4 and transmitted to the spectrometer 6 via a multimode fiber optic cable 5. The spectrometer 6 records the emission spectrum corresponding to the excitation wavelength.

[0032] By sequentially outputting all excitation wavelengths within the range of 200-800 nm (in 5 nm steps) using a grating monochromator 3 and repeating the above detection process, a complete transmission spectrum matrix (absorption spectrum) and a three-dimensional fluorescence spectrum matrix can be synchronously constructed within minutes. This design achieves truly synchronous acquisition, avoiding errors introduced by changes in water quality due to time differences.

[0033] 2.2 Edge Computing Internal Filtering Effect Correction The filtering effect is a common interference in fluorescence spectroscopy analysis, where absorbing substances in the solution absorb part of the excitation or emission light, resulting in a lower detected fluorescence intensity and thus distorting the fingerprint characteristics. Traditional methods often perform post-correction at the data analysis stage, with limited effectiveness.

[0034] This invention innovatively implements in-situ correction at the hardware level (edge ​​computing unit). The edge computing unit executes the following steps in real time: Data is acquired from photodiode module 8 to calculate the absorption spectrum. .

[0035] Raw fluorescence spectral data were acquired from spectrometer 6. ,in To excite the wavelength, The wavelength is the emission wavelength.

[0036] The correction factor was calculated using synchronously acquired absorption spectral data. Due to the internal filtering effect, which is influenced by the absorption of both excitation and emission light, the corrected fluorescence intensity... Calculated using the following formula:

[0037] in, It is the absorbance at the excitation wavelength. It is the absorbance at the emitted light wavelength. This formula uses an exponential term to amplify and compensate for the original fluorescence.

[0038] The revised The absorbed spectra collected simultaneously are packaged and uploaded to the SaaS cloud platform via a 4G / 5G network.

[0039] By employing this edge computing strategy of collecting and processing data simultaneously, the data uploaded to the cloud is already a high-fidelity water quality fingerprint that has undergone physical compensation, greatly improving the accuracy of subsequent analysis.

[0040] Example 3: SaaS Cloud Platform and Multidimensional Spectral Model After receiving data from the front end, the cloud platform processes it using a multidimensional spectral analysis model and a source tracing model.

[0041] 3.1 Multidimensional Spectral Water Quality Analysis Model This model is built upon a dual-channel convolutional neural network (CNN) specifically designed for processing one-dimensional absorption spectra and two-dimensional / three-dimensional fluorescence spectra (which can be viewed as two-dimensional images or matrices). The network structure includes: The first channel (one-dimensional CNN) receives an absorption spectrum vector from 200-800 nm as input. This channel consists of multiple one-dimensional convolutional and pooling layers, used to extract local features in the absorption spectrum, such as shoulder peaks or slope changes of specific organic compounds.

[0042] The second channel (2D CNN): receives a 3D fluorescence spectrum as input, which is the excitation-emission matrix (EEM). This channel uses a residual network (ResNet) structure, which can effectively extract complex spatial features such as fluorescence peak positions, intensities, Rayleigh scattering, and Raman scattering from the EEM image.

[0043] Feature Fusion Layer: An attention mechanism is employed to weightedly fuse the high-dimensional feature vectors output from the two channels. The attention mechanism can automatically learn the spectral features most important for the final water quality index prediction (for example, for chemical oxygen demand (COD) prediction, which quickly estimates the COD value in the water body based on the synchronously collected and corrected spectral data (absorption spectrum + three-dimensional fluorescence spectrum), it may rely more on the absorption spectrum; for specific fluorescent whitening agents, it relies more on the three-dimensional fluorescence features).

[0044] Multi-task learning output layer: The fused feature vector is input to multiple parallel fully connected networks (task-specific towers). Each tower is responsible for predicting a water quality indicator, such as chemical oxygen demand (COD), total organic carbon (TOC), total nitrogen (TN), and the concentration of a specific pollutant (such as benzene compounds and polycyclic aromatic hydrocarbons). The loss function employs a dynamic weighted average strategy, adaptively adjusting the weights of each task during training to achieve collaborative optimization.

[0045] 3.2 Multidimensional Spectral Source Tracing Model and Digital Twin The source tracing model is essentially a convolutional neural network based on metric learning. It maps the high-fidelity 3D fluorescent fingerprint of the input water sample to a high-dimensional embedding space. Simultaneously, a multi-user fingerprint database (containing hundreds of typical wastewater fingerprints from dozens of industries such as chemical, printing and dyeing, papermaking, and electroplating) is also mapped to the same space. By calculating the cosine similarity between the fingerprint to be tested and all fingerprints in the database, the top-N possible pollution sources with the highest similarity are output. For example: 1. Printing and dyeing industry - disperse blue dye, similarity 99.2%; 2. Chemical industry - aniline, similarity 75.1%.

[0046] Digital twin mapping and predictive maintenance: This system creates a real-time updated digital twin for each online device in the cloud. This twin includes not only a 3D model of the device, but also synchronizes the device's physical parameters in real time (such as xenon lamp spectral energy distribution, photodiode gain, temperature, humidity, etc.).

[0047] Predictive maintenance: The digital twin compares the current xenon lamp's emission spectrum energy curve with the factory baseline curve in real time. By calculating the energy decay slope, when it is predicted that the light source lifespan will reach the critical threshold in 30 days, the system automatically generates and sends a xenon lamp replacement work order to the maintenance personnel's mobile phone.

[0048] Simulated source tracing verification: When the source tracing model outputs a conclusion (e.g., pollution originates from upstream factory A), the digital twin can simulate the spectral morphology of pollutants diffusing to downstream monitoring points based on the hydrodynamic model and the fingerprint characteristics of factory A. The simulated spectrum is compared with the actually detected spectrum. If they match closely, the credibility of the source tracing conclusion is greatly enhanced; if they do not match, a re-tracing is triggered.

[0049] Example 4: Monitoring-Early Warning-Tracing-Joint Control Business Intelligent Agent 4.1 The holographic monitoring unit is configured as follows: Calculate the Mahalanobis or Euclidean distance between the current water quality fingerprint vector and the preset normal background baseline in real time; When the calculated distance exceeds a preset threshold, it is determined to be an anomaly in the water quality fingerprint and the dynamic early warning unit is triggered.

[0050] The holographic monitoring unit does not monitor a single chemical oxygen demand (COD) value, but rather a high-dimensional vector. It calculates in real time the Mahalanobis distance between the current water quality fingerprint vector (e.g., a high-dimensional vector composed of absorption spectra and three-dimensional fluorescence EEM) and the normal background baseline vector.

[0051] Logical reasoning: Mahalanobis distance can effectively eliminate the influence of dimensionality and variable correlation. At a certain time t, the system calculates the Mahalanobis distance. Set the threshold M.

[0052] Action: When When <M, it is considered a normal fluctuation. When the value is greater than M, the threshold is exceeded. The monitoring unit determines that the water quality fingerprint has changed and immediately packages the spectral data at the abnormal time (from t-10 minutes to t+10 minutes) and sends it to the dynamic early warning unit.

[0053] 4.2 Dynamic Early Warning Unit The early warning unit received the abnormal data packet.

[0054] Level 1 Warning (Attention): If Only between M and M1, with minimal concentration fluctuations, the system only logs the data and displays a notification on the large screen.

[0055] Level 2 Warning (Abnormal): If If the value is greater than M1, it is considered a significant anomaly. The early warning unit will trigger a level-two early warning and perform the following actions: Instruct the front-end equipment to switch from the regular 30-minute / sampling mode to a high-frequency sampling mode (1 minute / sampling) to capture the entire process of the contamination clump.

[0056] A notification was sent to the park's operations and maintenance personnel via WeChat / DingTalk: "[Abnormal] The fingerprint of the water body at the main discharge outlet has changed. High-frequency monitoring has been initiated. Please pay attention."

[0057] Level 3 Warning (Accident): Awaiting results from the source tracing unit. If confirmed to be a toxic or hazardous substance, the warning level will be upgraded to Level 3.

[0058] 4.3 Precise Traceability Unit Once activated by the early warning unit, the precise source tracing unit immediately launches its core engine—the AI ​​demixing and matching engine. This engine aims to separate the spectral characteristics of a single pollution source from the complex spectral signals representing mixed pollutants and identify its identity. The specific process is divided into two closely coupled sub-stages: the blind source separation stage and the fingerprint matching stage.

[0059] The precise traceability unit is configured as follows: Receive mixed spectral data during abnormal periods; The mixed spectral data is subjected to blind source separation using a nonnegative matrix factorization algorithm to extract independent spectral components. The extracted spectral components are compared with the multi-user fingerprint database to calculate the similarity and output structured pollution source information, which includes the pollution source industry, characteristic pollutant components, and similarity.

[0060] The specific process of the blind source separation stage: During the abnormal period, the system continuously collected multidimensional spectral data (fusion of absorption spectrum and three-dimensional fluorescence spectrum) at m time points. The data at each time point can be represented as an n-dimensional vector. Combine the vectors from these m time points column-wise to construct a nonnegative matrix V of size n×m. Each element in matrix V... Both represent the signal intensity at the j-th time and the i-th spectral channel. Since spectral intensity (absorbance, fluorescence intensity) are non-negative physical quantities, matrix V has inherent non-negativity.

[0061] Perform NMF decomposition to find two nonnegative matrices W and H such that their product approximates the original matrix V as closely as possible. ; Where k is the preset number of sources (the number of potential pollution sources), the value of which can be estimated dynamically through principal component analysis (PCA) or based on the complexity of the anomalous spectrum. Size is Each column This represents an unmixed spectral basis vector, i.e., the characteristic fingerprint of the r-th potential pollution source. This fingerprint is a mathematical estimate of the spectrum of a pure substance or a single source.

[0062] Size is Each line This represents the abundance (relative concentration) of the corresponding r-th spectral basis vector as a function of time.

[0063] Initialize random nonnegative matrices W and H, then iterate continuously using a multiplication update rule to minimize the reconstruction error. A commonly used loss function is the square of the Euclidean distance.

[0064] After iterative convergence, the obtained Each column in the matrix is ​​an independent, physically interpretable spectral fingerprint, corresponding to a potential source of contamination or background component. Meanwhile, Each row of the matrix reveals the peak time and diffusion process of each potential pollution source over time, which helps to determine the occurrence time and duration of illegal discharge events.

[0065] The system transmits the results of the NMF algorithm, namely the k separated spectral basis vectors (each column of the W matrix) and their corresponding abundance curves (each row of the H matrix), as candidate pollution source fingerprints and pollution process dynamics to the fingerprint matching unit in the next stage.

[0066] The fingerprint matching stage involves obtaining several candidate spectral basis vectors through blind source separation, after which the precise source tracing unit enters the fingerprint matching stage. The core task of this stage is to verify the identity of each candidate fingerprint and compare it with a database of known contamination source fingerprints.

[0067] For each candidate spectral basis vector obtained from NMF unmixing (With a length of n), the system first standardizes the data to eliminate the influence of concentration dimensions and retain pure waveform (fingerprint) features.

[0068] The "multi-user fingerprint database" built in the SaaS cloud platform of this invention stores a massive amount of standard fingerprints of known pollution sources. These fingerprints cover hundreds of typical wastewaters and their characteristic pollutants (such as benzene compounds, polycyclic aromatic hydrocarbons, disperse dyes, organochlorine pesticides, etc.) from dozens of industries, including chemical, printing and dyeing, papermaking, electroplating, and pharmaceutical.

[0069] Calculate the cosine similarity S between the candidate fingerprint vector to be matched and the embedding vector of each known fingerprint in the fingerprint database. The value of S ranges from [-1, 1]. The closer the value is to 1, the more consistent the directions of the two vectors are, that is, the more similar the two spectral fingerprints are.

[0070] Based on the similarity calculation results, the top-ranked candidate sources (e.g., Top-3) are selected in descending order. Combined with information from the abundance matrix H, the start time, peak time, and duration of pollution are determined.

[0071] Ultimately, the conclusions from the precise source tracing unit were pushed to the closed-loop joint control unit.

[0072] 4.4 Closed-loop control unit The closed-loop control unit is configured to perform both hardware and software control. The hard-link control includes sending instructions to a field programmable logic controller (PLC) via an industrial protocol to perform physical actions such as closing the sewage valve or starting the intercepting pump. The soft-linkage control includes: automatically generating an enforcement report containing a spectral evidence chain and source tracing conclusions, and pushing it to the regulatory platform via an application programming interface (API) to generate a pending work order.

[0073] As a key executor of the business closed loop, the joint control unit will perform two types of operations—hard joint control and soft joint control—after receiving the conclusion from the traceability unit.

[0074] Hardware-based control (equipment operation): The following command is sent to the PLC controller of the downstream sewage pipeline pumping station via the MQTT protocol: ACTION: CLOSE_VALVE, VALVE_ID: ZP-01, REASON: Aniline leakage. The entire process takes less than 2 seconds.

[0075] After receiving the instruction, the pump station PLC automatically closes the electric valve connecting the main discharge outlet of the park to the municipal pipeline network, thus trapping the polluted water within the park.

[0076] At the same time, it links with nearby video surveillance pan-tilt units, automatically adjusts the camera's preset position, points it at the drainage outlet of Factory B, and activates recording and snapshot functions to save key video evidence.

[0077] Soft Linkage (Management Collaboration): The system automatically invokes its report generation service to generate a standardized "Pollution Source Tracing Enforcement Report" based on spectral evidence chains, source tracing conclusions, and on-site video screenshots.

[0078] The report and pending work orders are pushed to the smart supervision platform of the superior ecological and environmental bureau and the mobile terminal (App) of law enforcement personnel through the API interface.

[0079] The system automatically sends text messages and app push notifications containing legal liability notices to factory managers and park environmental protection managers.

[0080] Through the close coupling of the above four units, the system of the present invention achieves a fully automated business closed loop from detecting water quality anomalies to physically blocking pollution and generating law enforcement evidence without human intervention.

[0081] Example 5: Overall System Workflow Step A: Simultaneously acquire the absorption spectrum and three-dimensional fluorescence spectrum of the water body to be tested using a synchronous absorption-three-dimensional fluorescence spectrometer, and perform internal filtration effect correction on the fluorescence spectrum based on the absorption spectrum at the edge end to generate high-fidelity multidimensional spectral data and upload it to the SaaS cloud platform; Step B: The holographic monitoring unit of the SaaS cloud platform calculates the vector space distance between the high-fidelity multidimensional spectral data and the normal background baseline in real time. If the distance exceeds the threshold, it is determined to be an anomaly in the water quality fingerprint. Step C: The dynamic early warning unit issues graded early warnings based on the degree of anomaly and triggers the precise source tracing unit; Step D: The precise source tracing unit performs blind source separation and fingerprint matching on the mixed spectrum during the abnormal period, and outputs structured source tracing conclusions; Step D further includes: Using digital twins to simulate the theoretical spectral morphology of pollution sources after diffusion; The theoretical spectral morphology is compared with the actual collected spectral morphology to verify the accuracy of the source tracing conclusion.

[0082] Step E: Based on the source tracing conclusion, the closed-loop control unit automatically generates and issues equipment control commands to physically block pollution, and at the same time generates and pushes law enforcement reports and management work orders.

[0083] Please see Figure 3-5 To provide a more complete understanding of this invention, a workflow example from device deployment to a complete event handling is provided below: S1: Install the synchronous absorption-three-dimensional fluorescence spectrometer described in this invention at the target water area (the main outlet of the stormwater pipe network in an industrial park). The system runs continuously for 72 hours, collecting water quality spectral data under normal operating conditions. Using statistical process control methods, establish the normal background baseline fingerprint vector and the corresponding Mahalanobis distance control limits (M=3, M1=4.5) for this location.

[0084] S2: The device enters online monitoring mode. Every 15 minutes, the device automatically completes a simultaneous acquisition of absorption and three-dimensional fluorescence spectra in the 200-800nm ​​range. The edge computing unit immediately performs internal filtering effect correction and generates high-fidelity data pairs. , And upload it to the SaaS cloud platform.

[0085] S3: The cloud-based holographic monitoring unit calculates the Mahalanobis distance between the current fingerprint and the baseline in real time. The calculated Mahalanobis distance is given at a specific time T on a certain day. =5.2 > 4.5, triggering a water quality fingerprint anomaly event. The dynamic early warning unit issues a level two warning (abnormal) based on the degree of anomaly and instructs the front-end equipment to adjust the sampling frequency to 1 minute / time.

[0086] S4: The precise source tracing unit is activated. It extracts high-frequency spectral data for 20 minutes before and after T and runs a blind source separation algorithm. The algorithm successfully separates three spectral components. After comparison with the multidimensional spectral source tracing model, one of the components has a 98.7% match with the target industry feature fingerprint database, and its abundance coefficient reaches its peak at T+5 minutes. The source tracing conclusion output is: the main pollution source is the characteristic wastewater of the corresponding industry, the characteristic pollutant is identified, and the suspected upstream pollution source is located.

[0087] S5: The closed-loop control unit receives the conclusion and initiates hard control: a closing command is sent to the PLC of the emergency intercepting gate located downstream of the discharge outlet. The gate closes completely within 1.5 seconds to prevent pollution from spreading to the outer river. Simultaneously, the monitoring camera at the discharge outlet of upstream A dyeing and printing factory is activated to retrieve video footage from before and after 10:00 AM, indicating any abnormal tanker truck activity at the discharge outlet.

[0088] Soft Linkage Control: The system automatically generates an "Emergency Report on Pollution Source Tracing" that includes fingerprint comparison images, pollutant concentration change curves, video screenshots, and source tracing conclusions. It is then pushed to the park management committee and the municipal ecological environment bureau's law enforcement team system via API, and generates a pending work order, which is assigned to the corresponding law enforcement grid member.

[0089] S6: After the pollution incident is handled, law enforcement personnel confirm the accuracy of the source tracing results. The system adds the complete spectral data of the incident and the confirmed pollution source labels as positive samples to the training dataset. During low-load periods at night, the SaaS cloud platform automatically triggers a model retraining task to further optimize the recognition accuracy and generalization ability of the multidimensional spectral source tracing model.

[0090] It should be noted that the formulas in this application are all dimensionless and numerical calculations. The formulas are obtained by software simulation based on a large amount of data and are the closest to the real situation. The thresholds, coefficients, standard values ​​and allowable values ​​involved in this application are all empirical values ​​and are selected by those skilled in the art according to the actual situation.

[0091] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. An AI-based synchronous absorption-three-dimensional fluorescence water quality monitoring, early warning, and source tracing system, characterized in that: It includes the physical perception layer, the SaaS cloud platform layer, and the business decision-making layer: The physical sensing layer includes at least one synchronous absorption-three-dimensional fluorescence spectrometer, which includes a miniaturized synchronous spectral detection module. The analyzer is used to simultaneously acquire the absorption spectrum and three-dimensional fluorescence spectrum of the water body to be tested, and to use an edge computing unit to perform internal filtration effect correction on the three-dimensional fluorescence spectrum based on the absorption spectrum to generate high-fidelity multidimensional spectral data. The SaaS cloud platform layer is communicatively connected to the physical perception layer and includes a multi-user fingerprint database, a multi-dimensional spectral water quality analysis model, and a multi-dimensional spectral source tracing model. The SaaS cloud platform layer is used to receive the high-fidelity multi-dimensional spectral data, call the multi-dimensional spectral water quality analysis model to predict water quality indicators, and call the multi-dimensional spectral source tracing model to compare with the multi-user fingerprint database and output the source tracing results. The business decision-making layer, running on the SaaS cloud platform, includes multiple interconnected business intelligent agent units, including: a holographic monitoring unit for calculating the spatial distance of water quality fingerprint vectors to determine water quality fingerprint anomalies; a dynamic early warning unit for providing tiered early warnings; a precise source tracing unit for activating the AI ​​demixing engine for blind source separation and fingerprint matching; and a closed-loop control unit for generating and issuing equipment control commands and management work orders based on the source tracing results.

2. The AI-based synchronous absorption-three-dimensional fluorescence water quality monitoring, early warning, and source tracing system according to claim 1, characterized in that, The miniaturized synchronous spectral detection module includes: a xenon lamp device (1), a Fresnel lens (2), a grating monochromator (3), an optical fiber coupling lens (4), a multimode optical fiber (5), a spectrometer (6), a cuvette (7), and a photodiode module (8); The light emitted by the xenon lamp device (1) is focused by the Fresnel lens (2) and enters the grating monochromator (3), and the grating monochromator (3) outputs monochromatic excitation light to illuminate the cuvette (7); Parallel to the incident light direction, the photodiode module (8) detects the transmitted light intensity to obtain absorption information; perpendicular to the incident light direction, the fiber-coupled lens (4) and the multimode fiber (5) guide the fluorescence signal into the spectrometer (6) to obtain fluorescence information.

3. The AI-based synchronous absorption-three-dimensional fluorescence water quality monitoring, early warning, and source tracing system according to claim 1, characterized in that, The internal filtering effect correction performed by the edge computing unit specifically includes: Acquire absorption spectra acquired synchronously and original fluorescence spectrum ; According to the formula Calculate the corrected fluorescence intensity ,in, To excite the wavelength, For the emission wavelength, This indicates the absorbance at the excitation wavelength. This indicates the absorbance at the emission wavelength.

4. The AI-based synchronous absorption-three-dimensional fluorescence water quality monitoring, early warning, and source tracing system according to claim 1, characterized in that, The holographic monitoring unit is configured as follows: Calculate the Mahalanobis or Euclidean distance between the current water quality fingerprint vector and the preset normal background baseline in real time; When the calculated distance exceeds a preset threshold, it is determined to be an anomaly in the water quality fingerprint and the dynamic early warning unit is triggered.

5. The AI-based synchronous absorption-three-dimensional fluorescence water quality monitoring, early warning, and source tracing system according to claim 1, characterized in that, The precise traceability unit is configured as follows: Receive mixed spectral data during abnormal periods; The mixed spectral data is subjected to blind source separation using a nonnegative matrix factorization algorithm to extract independent spectral components. The extracted spectral components are compared with the multi-user fingerprint database to calculate the similarity and output structured pollution source information, which includes the pollution source industry, characteristic pollutant components, and similarity.

6. The AI-based synchronous absorption-three-dimensional fluorescence water quality monitoring, early warning, and source tracing system according to claim 1, characterized in that, The closed-loop control unit is configured to perform both hardware and software control. The hard-link control includes sending instructions to a field programmable logic controller (PLC) via an industrial protocol to perform physical actions such as closing the sewage valve or starting the intercepting pump. The soft-linkage control includes: automatically generating an enforcement report containing a spectral evidence chain and source tracing conclusions, and pushing it to the regulatory platform via an application programming interface (API) to generate a pending work order.

7. The AI-based synchronous absorption-three-dimensional fluorescence water quality monitoring, early warning, and source tracing system according to claim 1, characterized in that, The SaaS cloud platform layer also includes a digital twin, which is used to synchronize the physical parameters and operating status of the synchronous absorption-three-dimensional fluorescence spectrometer in real time. The system also includes a predictive maintenance module, which is used to calculate the energy decay slope and predict the remaining lifespan of the light source by comparing the current light source energy curve with the factory baseline curve in real time based on the digital twin, so as to generate a maintenance work order.

8. The AI-based synchronous absorption-three-dimensional fluorescence water quality monitoring, early warning, and source tracing system according to claim 1, characterized in that, The system's operating method includes: Step A: Simultaneously acquire the absorption spectrum and three-dimensional fluorescence spectrum of the water body to be tested using a synchronous absorption-three-dimensional fluorescence spectrometer, and perform internal filtration effect correction on the fluorescence spectrum based on the absorption spectrum at the edge end to generate high-fidelity multidimensional spectral data and upload it to the SaaS cloud platform; Step B: The holographic monitoring unit of the SaaS cloud platform calculates the vector space distance between the high-fidelity multidimensional spectral data and the normal background baseline in real time. If the distance exceeds the threshold, it is determined to be an anomaly in the water quality fingerprint. Step C: The dynamic early warning unit issues graded early warnings based on the degree of anomaly and triggers the precise source tracing unit; Step D: The precise source tracing unit performs blind source separation and fingerprint matching on the mixed spectrum during the abnormal period, and outputs structured source tracing conclusions; Step E: Based on the source tracing conclusion, the closed-loop control unit automatically generates and issues equipment control commands to physically block pollution, and at the same time generates and pushes law enforcement reports and management work orders.

9. The artificial intelligence-based synchronous absorption-three-dimensional fluorescence water quality monitoring, early warning, and source tracing system according to claim 8, characterized in that, Step D further includes: Using digital twins to simulate the theoretical spectral morphology of pollution sources after diffusion; The theoretical spectral morphology is compared with the actual collected spectral morphology to verify the accuracy of the source tracing conclusion.

Images