Three-dimensional fluorescence coupling artificial intelligence water quality real-time monitoring and early warning system

Abstract

The invention discloses a three-dimensional fluorescence coupling artificial intelligence water quality real-time monitoring and early warning system, which comprises a three-dimensional fluorescence acquisition module used for acquiring three-dimensional fluorescence spectrum data of water quality, the three-dimensional fluorescence spectrum data is a three-dimensional matrix # imgabs0 # formed by excitation wavelength, emission wavelength and fluorescence intensity, M is the excitation wavelength point number, and the emission wavelength is the emission wavelength point number; n is the number of emission wavelength points; and the artificial intelligence processing module is electrically connected with the three-dimensional fluorescence acquisition module and comprises a three-dimensional fluorescence data preprocessing unit, a water quality evaluation model construction unit and a real-time evaluation unit. The non-linear mapping relation is constructed through the AI algorithm without depending on 10 + type single-parameter sensors such as pH electrodes, the annual average maintenance cost is reduced by 70% or above, the omission ratio of novel pollutants such as fluorescent micro-plastics is reduced to 5% or below from 40%, and the problems that index detection is lagged and sensor maintenance is tedious in a traditional method are solved.

Description

Technical Field

[0001] The present invention relates to the field of three-dimensional fluorescence technology, and particularly to a real-time water quality monitoring and early warning system that combines three-dimensional fluorescence with artificial intelligence. Background Art

[0002] As a key link in ensuring water safety, water quality monitoring in the prior art is mainly achieved through two types of methods: one is independent detection based on single-parameter sensors, such as pH electrodes, dissolved oxygen sensors, chemical oxygen demand (COD) online analyzers, etc. Water quality parameters are obtained through a combination of multiple sensors, and then a comprehensive evaluation is carried out according to national standards; the other is the use of spectral analysis techniques, such as three-dimensional fluorescence spectroscopy. By measuring the emission fluorescence intensity of water samples at different excitation wavelengths, an excitation-emission wavelength matrix (EEM) is constructed, and a quantitative relationship between three-dimensional fluorescence and a single water quality parameter (such as COD) is established based on methods such as linear regression. Then, the results of each parameter are stitched together for evaluation. This method cannot capture the synergistic effect between pollutants, resulting in a one-sided comprehensive evaluation and a lag in real-time early warning.

[0003] Due to the potential of rapid, non-destructive, and simultaneous multi-component detection, three-dimensional fluorescence spectroscopy technology has been preliminarily applied in fields such as surface water and drinking water. Traditional methods usually combine three-dimensional fluorescence data with single-parameter sensor data, and through linear algorithms such as partial least squares (PLS) and principal component regression (PCR), a regression model of fluorescence spectrum for a single index is established, and then water quality evaluation is realized by stitching the results of each index. Such technologies rely on real-time detection data of conventional indicators, and their core is an indirect mode of "regressing conventional indicators through three-dimensional fluorescence spectroscopy and then conducting water quality evaluation based on the indicators".

[0004] However, from the perspective of the prior art, the above technical path has significant limitations: First, relying on multi-parameter sensors leads to a high system complexity. It is necessary to synchronously maintain more than 10 independent devices, and the annual calibration frequency reaches more than 30 times. Both the hardware cost and the maintenance cost are extremely high; Second, the linear regression model is difficult to capture the complex pollution characteristics hidden in the three-dimensional fluorescence spectrum. The missed detection rate of new pollutants (such as microplastic fluorescent markers and emerging organic pollutants) exceeds 40%, and it cannot adapt to the non-linear changes in water quality components; In addition, traditional methods conduct comprehensive evaluation by stitching the results of single indicators, resulting in a one-sided evaluation. The recognition accuracy for complex pollution scenarios (such as mixed pollution of benzene series and phenols) is less than 70%, and the real-time performance depends on the response speed of the sensor, and the emergency early warning lags (usually more than 30 minutes).

[0005] In special scenario applications, the defects of the existing technologies are more prominent: in the detection of trace pollutants in drinking water, the detection limits of traditional fluorescence methods are generally higher than 50 ppb, which cannot meet the stringent requirements of GB5749-2022 for indicators such as benzo[a]pyrene (10 ppb); in remote areas or industrial scenarios, multi-sensor systems are limited by power consumption and communication stability and are difficult to achieve long-term continuous monitoring; for complex pollution scenarios such as chemical industrial parks, there is a lack of in-depth feature extraction and multi-modal fusion analysis of spectral data, and it is impossible to accurately trace the categories of pollutants and generate emergency response plans. Therefore, a real-time water quality monitoring and early warning system integrating three-dimensional fluorescence and artificial intelligence is proposed. Summary of the Invention

[0006] In order to overcome the above defects of the existing technologies, an embodiment of the present invention provides a real-time water quality monitoring and early warning system integrating three-dimensional fluorescence and artificial intelligence, which solves the problems of multi-sensor dependence, high maintenance cost, missed detection of new pollutants by linear models, one-sided evaluation, and poor real-time performance in the existing technologies.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] A real-time water quality monitoring and early warning system integrating three-dimensional fluorescence and artificial intelligence, comprising: a three-dimensional fluorescence acquisition module, an artificial intelligence processing module, and a real-time monitoring and early warning module;

[0009] The three-dimensional fluorescence acquisition module is used to obtain three-dimensional fluorescence spectral data of water quality, and the three-dimensional fluorescence spectral data is a three-dimensional matrix composed of excitation wavelength - emission wavelength - fluorescence intensity , where M is the number of excitation wavelength points and N is the number of emission wavelength points;

[0010] The artificial intelligence processing module is electrically connected to the three-dimensional fluorescence acquisition module, and includes a three-dimensional fluorescence data preprocessing unit, a water quality evaluation model construction unit, and a real-time evaluation unit:

[0011] The three-dimensional fluorescence data preprocessing unit is used to perform denoising, baseline correction, normalization processing, and tensor decomposition feature extraction on the three-dimensional fluorescence spectral data;

[0012] The water quality evaluation model construction unit trains a water quality evaluation model based on deep learning algorithms such as convolutional neural network (CNN), recurrent neural network (RNN), or Transformer model, using historical three-dimensional fluorescence data and their corresponding comprehensive water quality evaluation results. The comprehensive water quality evaluation result is obtained through a preset evaluation rule by combining multiple actual detection indicators of water quality;

[0013] The real-time evaluation unit processes the preprocessed real-time three-dimensional fluorescence data at the second / minute level based on an online learning algorithm or streaming data processing technology, using the trained water quality evaluation model to obtain the real-time water quality evaluation result;

[0014] The real-time monitoring and early warning module is electrically connected to the artificial intelligence processing module. It compares the real-time water quality evaluation result with the preset water quality early warning threshold. When the real-time water quality evaluation result exceeds the early warning threshold, it issues a corresponding early warning signal.

[0015] Preferably, the three-dimensional fluorescence acquisition module includes an excitation light source, a fluorescence detector, an optical system, and a sample cell; the excitation light source is used to emit excitation light of different wavelengths; the optical system guides the excitation light to the water sample in the sample cell and guides the fluorescence generated by the water sample to the fluorescence detector; the fluorescence detector is used to detect the fluorescence intensity at different emission wavelengths, thereby obtaining three-dimensional fluorescence spectral data.

[0016] Preferably, the normalization processing of the three-dimensional fluorescence data preprocessing unit includes but is not limited to:

[0017] Row normalization: Normalize the excitation wavelength dimension, and the formula is: ;

[0018] Column normalization: Normalize the emission wavelength dimension, and the formula is: , or other dimension normalization methods, to make the data within a numerical range; The tensor decomposition feature extraction is specifically non-negative tensor decomposition: , where is the excitation wavelength feature vector, is the emission wavelength feature vector, R is the decomposition rank, represents the outer product operation, which is used to extract low-rank features to remove noise and redundant information.

[0019] Preferably, when training the water quality evaluation model using the deep learning algorithm, the cross-entropy loss function is used as the optimization objective, and the formula is , where K is the number of training samples, is the true label of the comprehensive water quality evaluation result, is the model prediction value, and the cross-validation method is used to divide the training set and the validation set to improve the generalization ability of the model.

[0020] Preferably, the real-time monitoring and early warning module includes an early warning threshold setting unit and an early warning signal output unit; the early warning threshold setting unit can set multiple different levels of early warning thresholds according to different water quality monitoring scenarios and requirements; the early warning signal output unit issues early warning signals through one or more of sound, light, SMS, email, or network notification, and supports linkage with the Internet of Things platform to trigger an emergency response.

[0021] Preferably, it further includes a data storage and encryption module for storing historical three-dimensional fluorescence data, water quality comprehensive evaluation results, water quality evaluation model parameters, and real-time monitoring data; the module adopts a distributed storage or database storage method and supports the AES encryption algorithm to encrypt the data during transmission and storage, meeting the data security standards in the water quality monitoring field.

[0022] Preferably, it further includes a communication module for realizing data transmission and communication between the system and a remote monitoring center or a mobile terminal; the communication module supports wired communication or wireless communication methods, and the wireless communication methods include one or more of 4G, 5G, Wi-Fi, Bluetooth, or LoRa.

[0023] Preferably, the preset evaluation rules for the water quality comprehensive evaluation results include one of the analytic hierarchy process, fuzzy comprehensive evaluation method, or principal component analysis method, which is used to comprehensively consider multiple actual detection indicators of water quality and conduct a comprehensive evaluation of water quality.

[0024] Preferably, the artificial intelligence processing module further includes a model update unit for regularly or according to the changes in real-time monitoring data, using new historical data to update and optimize the water quality evaluation model to improve the accuracy and adaptability of the model.

[0025] Preferably, the system directly constructs a non-linear mapping relationship between three-dimensional fluorescence data and water quality comprehensive evaluation results by using the three-dimensional fluorescence spectral data as an independent water quality evaluation index and directly applying deep learning algorithms.

[0026] Preferably, the core innovation of this system is to skip the traditional path of "fluorescence → single parameter → comprehensive score" and directly construct a non-linear mapping relationship between three-dimensional fluorescence spectra and water quality comprehensive evaluation indicators:

[0027] The preset evaluation rules fuse multiple parameters such as COD, ammonia nitrogen, and heavy metals into a single index (such as water quality categories I-V, comprehensive pollution index 0-100 points);

[0028] The deep learning model directly learns the correlation between three-dimensional fluorescence features and the overall water quality status, and the output results directly support risk decision-making.

[0029] The technical effects and advantages of the water quality real-time monitoring and early warning system of three-dimensional fluorescence coupled with artificial intelligence of the present invention:

[0030] 1. In this invention, the system directly outputs comprehensive water quality evaluation indicators (such as risk level, comprehensive index) that integrate multiple parameters through an AI model, avoiding the error accumulation of the traditional "fluorescence → single index → comprehensive score". The accuracy of identifying complex pollution is increased by 35%. Preset rules are bound to national water quality standards or scientific index models (such as WQI) to ensure the traceability of evaluation results and compliance with industry norms. A single device replaces multiple sensors, reducing the maintenance cost by 70%, and the generation time of the comprehensive evaluation result is ≤ 1 minute (traditional system > 5 minutes).

[0031] 2. In this invention, relying on machine learning / deep learning algorithms, the system supports the adaptive construction and real-time update of models, enhances the generalization ability through cross-validation, and the adaptability to unknown pollution scenarios is increased by 30%. The efficiency is more than 10 times higher than that of traditional manual judgment.

[0032] 3. In this invention, the three-dimensional fluorescence acquisition module is integratedly designed. A single set of equipment replaces multiple sensors, reducing the hardware cost by 47% - 80% and the annual maintenance cost by 60%. The communication module supports multiple modes such as 4G / LoRa. The preprocessing technology improves the spectral signal-to-noise ratio by 20% - 30%, and the accuracy of complex pollution evaluation is increased by 35% compared with traditional single-index judgment.

[0033] 4. Compared with the traditional solution, this invention realizes getting rid of the dependence on multiple sensors and completing comprehensive evaluation with a single module; getting rid of the limitation of the linear model and the identification ability of new pollutants is increased by 60%; getting rid of the bottleneck of scenario adaptation, and indicators such as detection limit, response time, and maintenance cost lead comprehensively. A monitoring system integrating high-precision acquisition, intelligent processing, and full-scenario adaptation is constructed, and the system covers all scenarios such as drinking water, industrial wastewater, and chemical industrial parks. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 It is a schematic diagram of the overall system module of a real-time water quality monitoring and early warning system that combines three-dimensional fluorescence and artificial intelligence proposed by the present invention;

[0035] Figure 2 It is a schematic diagram of the system module of the artificial intelligence processing module in a real-time water quality monitoring and early warning system that combines three-dimensional fluorescence and artificial intelligence proposed by the present invention;

[0036] Figure 3 It is a schematic diagram of the system of a real-time water quality monitoring and early warning system that combines three-dimensional fluorescence and artificial intelligence proposed by the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0037] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0038] It should be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device. Without further limitation, the elements defined by the statement "including..." do not exclude the existence of additional identical elements in the process, method, article or device including the said elements.

[0039] Embodiment 1

[0040] Reference Figures 1-3 , this embodiment provides a real-time water quality monitoring and early warning system integrating three-dimensional fluorescence and artificial intelligence for the implementation of a basic water quality monitoring system. The specific implementation content includes:

[0041] Implementation purpose: Aiming at the monitoring requirements of surface water conventional pollution, a basic version system with low cost and high compatibility is constructed to verify the monitoring ability of three-dimensional fluorescence spectroscopy combined with the CNN model for common organic matters (such as humic acid) and realize the basic real-time early warning function.

[0042] Implementation steps:

[0043] (1) Hardware construction:

[0044] Install a Hamamatsu L4992 pulsed xenon lamp as the excitation light source, connect a Hamamatsu R928 type PMT detector, and build an optical path to a 10 mm quartz cuvette sample cell. The PMT detector outputs an analog electrical signal, which is converted into a digital signal by a data acquisition card (NI USB-6366) and then transmitted to an industrial computer through a USB interface to realize the communication connection between the three-dimensional fluorescence acquisition module and the artificial intelligence processing module.

[0045] An integrated data acquisition card (such as NI USB-6366) converts the detector analog signal into a digital signal and accesses an industrial computer (Advantech UNO-3082).

[0046] (2) Data collection:

[0047] Set the excitation wavelength scanning range from 200 to 800 nm (interval 5 nm), and the emission wavelength scanning range from 250 to 650 nm (interval 2 nm). Automatically collect the fluorescence spectrum of the water sample every 10 minutes to generate a three-dimensional matrix F with a dimension of 121×201.

[0048] (3) Pretreatment process:

[0049] Wavelet denoising: Perform 3-layer decomposition of db4 wavelet on F, and reconstruct the spectrum after threshold processing of the high-frequency coefficients.

[0050] Baseline correction: Fit the baseline in the emission wavelength dimension using a 5th-degree polynomial to subtract the baseline drift signal.

[0051] Row normalization: Normalize the data in the excitation wavelength dimension according to the formula to generate the preprocessed matrix .

[0052] (4) Model training:

[0053] Collect 1000 groups of historical spectral data of surface water containing humic acid, and use the comprehensive water quality category (Class I-V) generated according to the GB 3838-2002 standard as the label.

[0054] Build a 3-layer CNN network: Input layer 121×201, Convolutional layer 1 (3×3 kernel, 64 channels), Convolutional layer 2 (5×5 kernel, 128 channels), Convolutional layer 3 (7×7 kernel, 256 channels), and the fully connected layer outputs a 1-dimensional score. Use the Adam optimizer to train for 50 rounds.

[0055] (5) Real-time evaluation and early warning:

[0056] The real-time data is input into the trained CNN model after preprocessing, and the water quality risk level (Class I-V) is output within 1 minute.

[0057] Preset early warning thresholds: Slight pollution, Class III (yellow early warning), Moderate pollution, Class IV (orange early warning), Severe pollution, Class V (red early warning).

[0058] Implementation effect:

[0059] Detection ability: The detection linear range for humic acid-like organic matter is 0.1 - 100 mg / L, the correlation coefficient R² = 0.95, and the lowest detection limit reaches 0.1 mg / L (better than 1 mg / L of traditional single-parameter sensors).

[0060] Real-time performance: The total time from data collection to early warning output is ≤1 minute, which improves the response speed by 3 times compared with the traditional multi-sensor system (3 - 5 minutes).

[0061] Accuracy: The recognition accuracy for water quality of class III and below (score ≤ 7 points) reaches 92%, and the false alarm rate ≤ 8% (the false alarm rate of traditional systems is 20%+).

[0062] Cost - benefit: The hardware cost is about 80,000 yuan, which is 47% lower than that of similar multi - sensor monitoring systems (150,000 yuan+), and there is no need to regularly calibrate single - parameter sensors (the traditional system needs to be calibrated 3 times a month on average).

[0063] Example 2

[0064] This example provides a real - time water quality monitoring and early warning system that couples three - dimensional fluorescence with artificial intelligence for the implementation of time - series enhanced systems. The specific implementation content includes:

[0065] Implementation purpose: Aiming at the long - term monitoring requirements of river basins, strengthening the system's ability to capture the changing trend of water quality over time, solving the problem of lag in the response of traditional methods to progressive pollution, and realizing early warning.

[0066] Implementation steps:

[0067] (1) Hardware adaptation:

[0068] Continue to use the PMT detector in Example 1, replace the optical system with a wide - spectrum filter (200 - 650nm), integrate the LoRa wireless module (SX1278 chip) into the data acquisition unit, and deploy it at mountain - river monitoring points (spacing 5km).

[0069] (2) Time - series data acquisition:

[0070] Collect three - dimensional fluorescence spectra every 5 minutes, construct a time - series data set with a 30 - minute sliding window (including 6 consecutive spectral matrices), and the input dimension is 6×121×201.

[0071] (3) Pre - processing upgrade:

[0072] Median filtering: Perform median filtering on a single - frame spectrum with a 5×5 window to remove pulse noise.

[0073] AIRPLS baseline correction: Fit the spectral baseline through an adaptive iterative algorithm to improve the signal purity under complex backgrounds.

[0074] Time - dimension normalization: Calculate the mean / standard deviation of each emission wavelength according to the daily cycle (24 hours) and perform dynamic calibration on real - time data.

[0075] (4) LSTM model construction:

[0076] Design the LSTM network: The input layer is 6×121×201, which is converted into a 14641×6 - dimensional vector through a flattening layer. The LSTM hidden layer has 128 nodes, and the fully - connected layer outputs the water quality level (Class I - V) for the next hour.

[0077] Training data: Collect the time - series spectral data of a tributary of the Yangtze River for 365 days. The label is the ammonia - nitrogen concentration level of the next day. Use the cross - entropy loss function to optimize the model.

[0078] (5)Trend prediction and early warning:

[0079] Input the real - time 30 - minute sliding - window data, and the model outputs the predicted values of the water quality level for the next 6 hours.

[0080] When the predicted level rises continuously for 2 times and exceeds the threshold, send a trend early warning to the basin management center through the LoRa network, along with the pollution - diffusion simulation curve.

[0081] Implementation effect:

[0082] Trend prediction ability: The prediction lead time for the progressive ammonia - nitrogen pollution reaches 6 - 12 hours (depending on the pollution - diffusion speed), which is more than 3 times higher than that of the traditional sensor network (2 hours in advance).

[0083] Anti - interference ability: Through dynamic calibration in the time dimension, the suppression rate of the spectral drift caused by the day - night temperature difference reaches 85%, and the false - alarm rate is reduced from 22% of the traditional system to 5.2%.

[0084] Low - power transmission: The LoRa module realizes a long - distance transmission of 5 km. The power consumption of a single node is ≤100 mW. With solar power supply, it can operate continuously for 30 days in remote mountainous areas (the traditional 4G device only supports 7 days).

[0085] Data utilization rate: Through the 30 - minute sliding - window time - series data, 15% of the progressive pollution events missed by the traditional single - point detection are captured (such as the slow over - standard process of ammonia - nitrogen concentration rising from 0.8 mg / L to 1.2 mg / L).

[0086] Example 3

[0087] This example provides a real - time water quality monitoring and early - warning system that combines three - dimensional fluorescence and artificial intelligence for the implementation of a high - sensitivity drinking - water monitoring system. The specific implementation content includes:

[0088] Implementation purpose: Aiming at the monitoring requirements of trace pollutants in drinking - water sources, through hardware upgrade and algorithm optimization, improve the detection sensitivity of the system to trace substances such as disinfection by - products and polycyclic aromatic hydrocarbons, and meet the strict early - warning requirements of national standards.

[0089] Implementation steps:

[0090] (1)Deployment of high - precision hardware:

[0091] A xenon light source: Hamamatsu L10282-25 (150 W xenon lamp) and a monochromator: Horiba iHR320 (equipped with an ultraviolet grating, resolution ≤ 0.1 nm) are used as the excitation light source, paired with an Andor back-illuminated CCD detector (quantum efficiency > 90% @ 250 nm), and a full-spectrum filter set (interval 1 nm) of 150 - 600 nm is configured.

[0092] The sample cell is replaced with a quartz flow cell (volume 50 mL), and a peristaltic pump (Ismatec REGLO) is integrated to achieve continuous water sample circulation.

[0093] (2) High-sensitivity data processing:

[0094] Non-negative tensor decomposition (NTD): The three-dimensional spectral matrix F is decomposed with a rank of R = 10 to extract the core feature matrices A and B, removing noise and scattered light interference.

[0095] Analytic Hierarchy Process (AHP): A three-layer evaluation system of disinfection by-products (trihalomethanes), microbial indicators (Escherichia coli), and organic pollution (benzo[a]pyrene) is established. The weights (0.4:0.3:0.3) are determined through expert scoring, and the comprehensive water quality index (0 - 100 points) is calculated.

[0096] (3) Real-time linkage control:

[0097] The water sample spectrum is collected every 10 seconds. After preprocessing, it is input into the NTD-AHP fusion model, and the comprehensive water quality score (0 - 100 points) is output within 30 seconds.

[0098] When the score < 80 points, it triggers the SCADA system linkage: close the downstream water supply valve (response within 5 seconds), start the activated carbon filtration device (reach full-load operation within 1 minute), and at the same time send a level-three emergency response signal to the water supply dispatching center.

[0099] Implementation effect:

[0100] Trace detection ability: The detection limit for benzo[a]pyrene (national standard limit 10 ppb) reaches 5 ppb, and the detection limit for chloroform is 10 ppb. The sensitivity is 30% higher than that of the traditional fluorescence method, meeting the strict requirements of GB5749-2022.

[0101] Emergency response speed: The entire process from data collection to SCADA system linkage (closing the valve + starting the filtration) ≤ 30 seconds, among which the valve closing response time is 5 seconds (the traditional mechanical valve requires 30 seconds +), and the activated carbon filtration device reaches the full-load treatment state within 1 minute.

[0102] Comprehensive evaluation accuracy: Through the NTD-AHP fusion model, the weight allocation error of disinfection by-products, microorganisms, and organic matter is ≤5%, and the coincidence degree between the comprehensive score and the national standard full-item detection results reaches 98%.

[0103] Compliance: No missed detection events occurred during 12 consecutive months of operation, and 3 cases of sudden water source pollution were successfully warned (such as the excessive microcystin caused by the algal bloom in a certain reservoir in 2023).

[0104] Example 4

[0105] This example provides a real-time water quality monitoring and warning system that combines three-dimensional fluorescence and artificial intelligence for the implementation of a lightweight edge computing system. The specific implementation content includes:

[0106] Implementation purpose: For scenarios with high requirements for power consumption and real-time performance, such as industrial wastewater discharge outlets, build a lightweight edge system to solve the problems of high processing latency and network disconnection failure in traditional cloud processing, and achieve offline rapid monitoring.

[0107] Implementation steps:

[0108] (1) Edge node deployment:

[0109] The hardware uses Raspberry Pi 4B (4GB memory) as the edge computing unit, integrates Huawei 5G module ME909s-821, and externally connects a miniaturized three-dimensional fluorescence probe (excitation light source: LED array 250 - 400nm, detector: micro silicon photodiode).

[0110] The edge stores the data of the most recent 7 days (encrypted with AES-256), and the cloud server (Alibaba Cloud ECS) stores historical data and model parameters.

[0111] (2) Lightweight preprocessing:

[0112] Spectrum-time: According to the formula , calculate the mean value and standard deviation of each emission wavelength at time t in real time.

[0113] Feature dimensionality reduction: Compress the three-dimensional matrix into a 20-dimensional feature vector through principal component analysis (PCA) to reduce the computational complexity.

[0114] (3) Model optimization and offline operation:

[0115] Train a lightweight Transformer model in the cloud: Compress the parameters by 70% through knowledge distillation technology, with a 20-dimensional input layer, 6-layer encoder, and output water quality level (1 - 5 levels).

[0116] After the edge device receives the cloud model, it supports the offline mode: when the network is disconnected, it calls the latest model cached locally to evaluate real-time data (processing time < 10 seconds), and stores the results to be synchronized to the cloud after the network is restored.

[0117] (4)Low-power communication:

[0118] When the network is connected normally, upload the original spectrum (once per hour) and the evaluation results at the edge device (once per minute) through 5G; when the network is disconnected, it automatically switches to the LoRa low-power mode and only sends warning signals (power consumption drops to 50mW).

[0119] Implementation effects:

[0120] Offline operation ability: In a network-disconnected environment, it can rely on the 7-day data and lightweight model stored locally to maintain monitoring. The offline processing time < 10 seconds per time, and the battery life reaches 72 hours (the traditional cloud system fails when the network is disconnected).

[0121] Low-power advantage: The operating power consumption of the edge device (Raspberry Pi 4B) is 5W, only 10% of that of a traditional industrial computer (50W+). With a 5W solar panel, 24-hour monitoring of remote factory areas can be achieved.

[0122] Improved real-time performance: The 5G transmission delay ≤ 2 seconds, which is 10 times higher than that of traditional cloud processing (20 seconds+); when the network is disconnected, warning signals are sent through LoRa, and the power consumption drops to 50mW (the power consumption of traditional wireless modules is 200mW+).

[0123] Balance between cost and computing power: After the model parameters are compressed by 70%, the CPU utilization rate of the edge device < 30%, and the single-node cost is 12,000 yuan, only 24% of the traditional edge computing solution (50,000 yuan+).

[0124] Example 5

[0125] This example provides a real-time water quality monitoring and warning system integrating three-dimensional fluorescence and artificial intelligence for the implementation of a multi-modal fusion monitoring system. The specific implementation content includes:

[0126] Implementation purpose: For the complex pollution scenario in chemical industrial parks, integrate three-dimensional fluorescence and ultraviolet-visible spectrum data to solve the problem of insufficient ability of single-spectrum to distinguish multi-component pollutants, and achieve accurate traceability and hierarchical warning.

[0127] Implementation steps:

[0128] (1)Multi-modal data acquisition:

[0129] Add a Shimadzu UV-2600 ultraviolet-visible spectrometer (scanning range: 200 - 800 nm, interval: 1 nm), and collect the same water sample in parallel with a three-dimensional fluorescence probe (excitation light source: xenon lamp + monochromator) to generate a fluorescence matrix F (121×201) and an absorption spectrum vector A (601 dimensions).

[0130] (2)Feature-level fusion processing:

[0131] Tensor splicing: Flatten F into a 24321-dimensional vector, and concatenate it with A to form a 24922-dimensional input feature.

[0132] Dual-channel CNN design: Construct two parallel convolutional branches (fluorescence channel + absorption channel), extract spectral features respectively and then fuse them through a fully connected layer, and output the pollutant categories (benzene series / phenols / aldehydes, etc.) and the comprehensive pollution index.

[0133] (3)Federated learning model update:

[0134] Local training models at each monitoring station: Update the model parameters using local data (after desensitization), and only upload the gradient information to the central server.

[0135] Cloud aggregation: Aggregate the gradients of all stations every 24 hours, generate a global model and distribute it to ensure data privacy (for example, the original spectral data of chemical industry enterprise water quality is not uploaded).

[0136] (4)Intelligent warning expansion:

[0137] Real-time display on the mobile APP: Three-dimensional fluorescence spectrogram, UV-Vis absorption curve, pollutant categories and diffusion paths (based on GIS map).

[0138] Connect to the emergency platform: When detecting a combined pollution of benzene series + phenols (comprehensive pollution index > 85), automatically send the pollution source location information and recommended disposal plan (such as starting a neutralization reaction device) to the Ministry of Emergency Management.

[0139] Implementation effects:

[0140] Recognition rate of combined pollution: The discrimination accuracy rate for combined pollutions such as benzene series + phenols, aldehydes + polycyclic aromatic hydrocarbons, etc. reaches 95%, which is 12% higher than that of a single three-dimensional fluorescence model (83%), and more than 10 kinds of mixed pollutants are successfully recognized.

[0141] Privacy protection: Through federated learning technology, the risk of local data leakage is reduced by 90%, and only encrypted gradient information is uploaded (the original spectral data is not uploaded), improving data security.

[0142] Emergency response efficiency: After connecting to the platform of the Ministry of Emergency Management, the time from pollution identification to disposal plan generation is ≤ 2 minutes (traditional manual analysis takes 30 minutes+), and the error of automatically calculating the dosage of sodium bicarbonate for phenol pollution is ≤ 5%.

[0143] Visualization ability: The mobile APP can display the fused spectrum of three-dimensional fluorescence fingerprint and UV-Vis absorption curve in real time, support the simulation of pollution diffusion path (based on GIS map, with an accuracy of 100m), and assist decision-makers to quickly locate the pollution source.

[0144] Comparative Example 1

[0145] This comparative example provides a traditional water quality monitoring system, and the specific implementation details include:

[0146] Implementation purpose: To verify the limitations of traditional methods that rely on single sensors and linear regression methods, and to compare the advantages of the present invention in comprehensive monitoring capabilities.

[0147] Implementation steps:

[0148] (1) Deployment of sensor array:

[0149] Install 10 independent sensors such as pH electrode (Mettler Toledo SG8), dissolved oxygen sensor (Hach LDO), and COD on-line analyzer (HACH DR2800) to collect conventional indicators respectively.

[0150] (2) Data processing:

[0151] Perform linear regression on specific wavelength points of three-dimensional fluorescence spectrum (such as Ex = 254, Em = 450) to fit empirical formulas for indicators such as COD and ammonia nitrogen (such as COD = 0.8F + 0.3F').

[0152] Summarize the data of each sensor through a PLC controller (Siemens S7-1200) and calculate the single indicator value.

[0153] (3) Threshold warning:

[0154] Preset single indicator thresholds (such as COD > 50mg / L, ammonia nitrogen > 1.5mg / L). When triggered, send a relay signal through the RS485 bus to control the flashing of the on-site alarm light.

[0155] Traditional methods need to rely on multi-sensor calibration, cannot identify the comprehensive pollution information hidden in the spectrum, and only support independent warning of single indicators, with a false alarm rate as high as 25%, and the comprehensive performance is significantly lower than that of the system of the present invention.

[0156] Compared with Examples 1-5 and Comparative Example 1, the present invention breaks through the technical bottleneck of traditional monitoring systems that rely on single sensors and linear models through the deep coupling of three-dimensional fluorescence spectrum and artificial intelligence, and realizes:

[0157] Index system: It has been converted from splicing single parameters to spectral comprehensive evaluation, and the water quality grade can be determined without returning to conventional indicators.

[0158] Scene adaptability: It covers the whole scene from drinking water to industrial wastewater, from conventional monitoring to extreme environments through 5 major embodiments, and each solution provides exclusive technologies for specific pain points.

[0159] Breakthrough in performance boundaries: It comprehensively surpasses traditional systems in key indicators such as detection limit (ppb level), response speed (second-level linkage), false alarm rate (≤10%), and maintenance cost (reduced by 80%). In particular, it fills the technical gaps in fields such as new pollutant identification, trend prediction, and emergency linkage.

[0160] In contrast, due to the limitations of the technical path, the monitoring ability of traditional systems has significantly lagged behind in complex water quality environments. Through continuous algorithm innovation and scenario-based design, the present invention constructs a more intelligent and reliable water quality monitoring and early warning system.

[0161] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any other combination. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product.

[0162] Those of ordinary skill in the art can realize that the modules and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are executed in hardware or software depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods for each specific application to implement the described functions, but such implementation should not be considered to exceed the scope of this application.

[0163] In addition, in each embodiment of the present application, the functional modules can be integrated into one processing module, or each module can exist physically alone, or two or more modules can be integrated into one module.

[0164] The above is only the specific implementation manner of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art within the technical scope disclosed in this application can easily think of changes or substitutions, which should all be covered within the protection scope of this application. Therefore, the protection scope of this application should be subject to the protection scope of the claims.

[0165] Finally: The above is only the preferred embodiment of the present invention and is not used to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should all be included within the protection scope of the present invention.

Claims

1. A real-time water quality monitoring and early warning system integrating three-dimensional fluorescence and artificial intelligence It is characterized in that Including: three-dimensional fluorescence acquisition module, artificial intelligence processing module, real-time monitoring and early warning module; Three-dimensional fluorescence acquisition module, which is used to obtain three-dimensional fluorescence spectral data of water quality, and the three-dimensional fluorescence spectral data is a three-dimensional matrix composed of excitation wavelength - emission wavelength - fluorescence intensity , where M is the number of excitation wavelength points and N is the number of emission wavelength points; The artificial intelligence processing module is connected to the three-dimensional fluorescence acquisition module for receiving three-dimensional fluorescence spectrum data, including a three-dimensional fluorescence data preprocessing unit, a water quality evaluation model building unit and a real-time evaluation unit: The three-dimensional fluorescence data preprocessing unit is used to perform denoising, baseline correction, normalization and tensor decomposition feature extraction on the three-dimensional fluorescence spectrum data; The water quality evaluation model construction unit is based on a deep learning algorithm such as a convolutional neural network (CNN), a recurrent neural network (RNN) or a Transformer model, and uses historical three-dimensional fluorescence data and its corresponding comprehensive water quality evaluation index to train a water quality evaluation model; The comprehensive water quality evaluation index is a single quantitative value or category label generated by preset evaluation rules based on the measured values of multiple key water quality parameters, and is used to characterize the overall state or risk level of water quality; The preset evaluation rules include water quality classification standards or water quality comprehensive index calculation models; The real-time evaluation unit processes the pre-processed real-time three-dimensional fluorescence data in seconds or minutes based on an online learning algorithm or streaming data processing technology using a trained water quality evaluation model to obtain a real-time water quality evaluation result; The real-time monitoring and early warning module is electrically connected to the artificial intelligence processing module, and compares the real-time water quality evaluation result with the preset water quality early warning threshold. When the real-time water quality evaluation result exceeds the early warning threshold, a corresponding early warning signal is issued.

2. The real-time water quality monitoring and early warning system with three-dimensional fluorescence coupling artificial intelligence according to claim 1, characterized in that, The three-dimensional fluorescence acquisition module includes an excitation light source, a fluorescence detector, an optical system and a sample pool; the excitation light source is used to emit excitation light of different wavelengths; the optical system guides the excitation light to the water sample in the sample pool, and guides the fluorescence generated by the water sample to the fluorescence detector; the fluorescence detector is used to detect the fluorescence intensity at different emission wavelengths, thereby obtaining three-dimensional fluorescence spectrum data.

3. The real-time water quality monitoring and early warning system integrating three-dimensional fluorescence and artificial intelligence according to claim 1, characterized in that, The normalization processing of the three-dimensional fluorescence data preprocessing unit includes but is not limited to: Row normalization: Normalize the excitation wavelength dimension, and the formula is: ; Column normalization: Normalize the emission wavelength dimension, with the formula: , or other dimension normalization methods, to make the data fall within a numerical range; The tensor decomposition feature extraction is specifically non - negative tensor decomposition: , where is the excitation wavelength feature vector, is the emission wavelength feature vector, R is the decomposition rank, represents the outer product operation, used to extract low - rank features to remove noise and redundant information.

4. The real-time water quality monitoring and early warning system coupled with 3D fluorescence and artificial intelligence according to claim 1, characterized in that When training the water quality evaluation model using the deep learning algorithm, the cross-entropy loss function is used as the optimization objective, and the formula is , where K is the number of training samples, is the true label of the comprehensive water quality evaluation result, is the model prediction value. The cross-validation method is used to divide the training set and the validation set to improve the generalization ability of the model.

5. The real-time water quality monitoring and early warning system coupling three-dimensional fluorescence with artificial intelligence according to claim 1, characterized in that The real-time monitoring and early warning module includes an early warning threshold setting unit and an early warning signal output unit; the early warning threshold setting unit can set multiple early warning thresholds of different levels according to different water quality monitoring scenarios and needs; the early warning signal output unit sends out early warning signals through one or more methods such as sound, light, text message, email or network notification, and supports linkage with the Internet of Things platform to trigger emergency response.

6. The real-time water quality monitoring and early warning system coupling three-dimensional fluorescence with artificial intelligence according to claim 1, characterized in that It also includes a data storage and encryption module for storing historical three-dimensional fluorescence data, comprehensive water quality evaluation results, water quality evaluation model parameters and real-time monitoring data; the module adopts distributed storage or database storage, and supports AES encryption algorithm to encrypt transmitted and stored data, which meets the data security standards in the field of water quality monitoring.

7. The real-time water quality monitoring and early warning system coupled with artificial intelligence by three-dimensional fluorescence as described in claim 1, characterized in that, It also includes a communication module for realizing data transmission and communication between the system and a remote monitoring center or a mobile terminal; the communication module supports wired communication or wireless communication methods, and the wireless communication methods include one or more of 4G, 5G, Wi-Fi, Bluetooth or LoRa.

8. The real-time water quality monitoring and early warning system with three-dimensional fluorescence coupling artificial intelligence according to claim 1, characterized in that, The preset evaluation rules for the comprehensive water quality evaluation indicators include at least one of the water quality category grading standards, water quality comprehensive index calculation models, Analytic Hierarchy Process (AHP), Fuzzy Comprehensive Evaluation Method, or Principal Component Analysis (PCA), which are used to fuse the measured values of multiple water quality parameters into a single quantitative value or category label.

9. The real-time water quality monitoring and early warning system coupled with three-dimensional fluorescence and artificial intelligence according to claim 1, characterized in that, The artificial intelligence processing module further includes a model update unit, which is used to update and optimize the water quality evaluation model regularly or according to the changes in real-time monitoring data, using new historical data to improve the accuracy and adaptability of the model.

10. A real-time water quality monitoring and early warning system integrating three-dimensional fluorescence and artificial intelligence according to any one of claims 1-9, characterized in that, The system directly constructs a non-linear mapping relationship between three-dimensional fluorescence data and the comprehensive water quality evaluation results by using three-dimensional fluorescence spectral data as an independent water quality evaluation indicator and directly applying deep learning algorithms.

Images