A water quality pollutant identification method based on multi-modal data fusion and cognitive reasoning
Patent Information
- Application Number
- CN202610984863.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-07-03
- Publication Date
- 2026-10-30
AI Technical Summary
[0002]随着工业化进程的加快和城市化水平的提高，水环境污染问题日益严峻，对水质监测技术提出了更高的要求，传统的水质污染物识别方法主要依赖单一传感器或单一检测原理进行水质参数测量，现有水质监测系统通常采用固定流路进行水样输送和检测，无法根据水样性质、污染物类型和传感器状态动态调整检测流路，导致在面对成分复杂、污染物种类多变的水体时，检测精度和效率难以保障，部分系统虽具备流路切换功能，但切换决策缺乏智能化依据，仅依靠预设规则进行简单切换，无法实现真正意义上的自适应流路优化，为此，我们提出一种多模态数据融合与认知推理的水质污染物识别方法
1.本发明通过构建流路状态演化图谱，将每条检测流路作为图谱节点、液体传递关系作为图谱边，并采用三层图卷积网络对节点状态进行推演，能够准确预测各流路的污染风险值和可用性评分。通过充分考虑流路的历史污染状况、当前可用状态和针对特定水样的检测适配性三个维度，使每一次检测均在综合条件最优的流路上执行，有效避免了流路交叉污染对检测结果的干扰，尤其适用于成分复杂、污染物种类多变的水体检测场景；
Smart Images
Figure CN123012387A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of water quality monitoring technology, specifically to a method for identifying water pollutants using multimodal data fusion and cognitive reasoning. Background Technology
[0002] With the acceleration of industrialization and the improvement of urbanization, water pollution has become increasingly serious, placing higher demands on water quality monitoring technology. Traditional water pollutant identification methods mainly rely on a single sensor or a single detection principle to measure water quality parameters. Existing water quality monitoring systems usually use fixed flow paths for water sample transportation and detection, which cannot dynamically adjust the detection flow path according to the water sample properties, pollutant types, and sensor status. As a result, when facing water bodies with complex composition and diverse pollutant types, it is difficult to guarantee detection accuracy and efficiency. Although some systems have flow path switching functions, the switching decision lacks intelligent basis and simply relies on preset rules for simple switching, which cannot achieve true adaptive flow path optimization. To address this, we propose a water pollutant identification method based on multimodal data fusion and cognitive reasoning. Summary of the Invention
[0003] The purpose of this invention is to provide a method for identifying water pollutants using multimodal data fusion and cognitive reasoning.
[0004] To achieve the above objectives, the present invention provides the following technical solution: a water pollutant identification method based on multimodal data fusion and cognitive reasoning, comprising real-time water quality status acquisition, water sample parameter acquisition, water quality status judgment, water quality data processing, and flow path cleaning treatment. The specific steps of the water pollutant identification method are as follows: Step 1: Real-time acquisition of state parameters for each detection flow path; construction of a flow path state evolution map with flow paths as nodes and liquid transfer relationships as edges; and use graph neural networks to deduce the contamination risk value of each flow path. and usability rating ,in, , This represents the total number of flow paths; Step 2: Obtain the initial characteristic parameters of the water sample to be tested, and input them into the flow path adaptability evaluation model to obtain the adaptability score of each flow path. Based on comprehensive evaluation score The path with the highest value is selected as the target detection flow path, and a flow path switching operation is performed. , , The preset weighting coefficients satisfy the following conditions: ; Step 3: When the current state meets any of the conditions, perform a self-calibration operation. The self-calibration operation includes introducing a blank solution to obtain a baseline reference value, introducing two standard solutions of different concentrations to establish a concentration-response standard curve, and calculating the drift compensation coefficient and sensitivity correction coefficient of each sensor based on the baseline reference value and the standard curve. Step 4: Simultaneously acquire multimodal raw data of the water sample to be tested through a multimodal sensor array, preprocess it and input it into a multimodal fusion encoder. Use a cross-modal attention mechanism to generate a fusion feature tensor, and then input the fusion feature tensor into the cognitive reasoning module. Based on the preset pollutant knowledge graph, perform path search and confidence propagation, output pollutant type identification results and their confidence, and at the same time correct the concentration estimate based on the drift compensation coefficient and sensitivity correction coefficient, and output the final concentration estimate of each pollutant. Step 5: After the detection is completed, perform the flow path cleaning operation, record the operation information of this detection and cleaning, and fill the operation information back into the flow path state evolution graph, and update the attribute information of the relevant nodes and edges accordingly.
[0005] As a further aspect of the present invention: In step one, the specific method for using a graph neural network to perform state inference based on the flow path state evolution map is as follows: a three-layer graph convolutional network is constructed. The first layer maps the initial state parameters of each node to a 128-dimensional hidden state vector. The second layer aggregates the state information of adjacent nodes through graph convolution operations. The third layer outputs the pollution risk prediction value of each node. and usability score prediction The training samples for the graph neural network are derived from historical flow path state data and corresponding actual pollution detection results.
[0006] As a further aspect of the present invention: In step two, the method for constructing the flow path adaptability evaluation model is as follows: Historical detection data is collected, including the initial characteristic parameters of the water sample to be tested in each test, the identifier of the selected detection flow path, and the recognition accuracy of that test. Using the initial characteristic parameters and the detection flow path identifier as input features, and the recognition accuracy as the prediction target, a gradient boosting decision tree algorithm is used to train and obtain the flow path adaptability evaluation model, which is then used to calculate the adaptability score of each flow path to the current water sample to be tested. .
[0007] As a further aspect of the present invention: In step three, the method for establishing the concentration-response standard curve is as follows: For each standard solution, its response values are collected on the UV-Vis absorption spectroscopy module, fluorescence spectroscopy detection module, and electrochemical sensor array, respectively. A concentration-response fitting curve is independently established for each sensor. The fitting method adopts least squares polynomial regression, and the polynomial order is second or third. During the fitting process, the concentration value of the standard solution is used as the independent variable, and the sensor response value is used as the dependent variable. Any of the following conditions are met: the cumulative number of detections of the target detection flow path since the last calibration reaches a preset number threshold; the cumulative running time of the target detection flow path since the last calibration reaches a preset duration threshold; the pollution risk value Ri of the target detection flow path exceeds a preset risk threshold; and the change rate of the turbidity value of the current water sample to be tested exceeds a preset change rate threshold.
[0008] As a further aspect of the present invention: In step three, the drift compensation coefficient is calculated by comparing the currently acquired blank solution response value with the historically stored blank solution standard response value, and calculating the response drift of each sensor at each characteristic wavelength or each detection channel. In response to the reciprocal of the drift amount As the drift compensation coefficient for the corresponding sensor, where the subscript Indicates the sensor number, subscript Indicates the characteristic wavelength or channel number; the sensitivity correction coefficient is calculated by taking the slope of the currently established standard curve. Standard slope of the standard curve stored in history Compare them by their ratio. This serves as the sensitivity correction coefficient for the corresponding sensor.
[0009] As a further aspect of the present invention: In step four, the cross-modal attention mechanism is specifically implemented as follows: UV-Vis absorption spectroscopy data, fluorescence spectroscopy data, and electrochemical sensing data are respectively used as three modal inputs. Each modality extracts its own feature vector through an independent encoder. Then, a multi-head cross-modal attention mechanism is used to calculate the mutual attention weight matrix between the modal features. Based on the attention weight matrix, the modal features are weighted and fused to generate a unified dimension fused feature tensor.
[0010] As a further aspect of the present invention: In step four, the method for constructing the pollutant knowledge graph is as follows: collect the spectral feature vectors, electrochemical response features, physicochemical property parameters and their interrelationships of known pollutants, construct a knowledge graph containing pollutant entities, feature entities and relational edges with pollutant types and feature parameters as entities and attribute associations and response associations as relational edges, and perform path search on the pollutant knowledge graph based on the fused feature tensor, calculate the confidence score of each candidate pollutant entity, and output the pollutant type with the highest confidence score as the recognition result.
[0011] As a further aspect of the present invention: in step five, the execution parameters of the cleaning operation are based on the contamination risk value of the target detection flow path. The system is dynamically adjusted based on the types of pollutants currently detected and identified. When the flow rate of the cleaning fluid exceeds the first preset threshold, the cleaning fluid flow rate and cleaning duration are increased. When the detected contaminant type belongs to the predefined adhesive contaminant category, a surfactant is added to the cleaning fluid in a preset ratio, and an additional blank solution rinsing verification operation is performed after cleaning is completed.
[0012] As a further aspect of the present invention: In step four, the pollutant type identification results and the final concentration estimates of each pollutant are combined to generate a water quality pollutant detection report and output it to the outside world. The detection report shall at least include the pollutant type name, the concentration value of each pollutant, the identification confidence percentage, the target detection flow path number used in this detection, the self-calibration status record before this detection, and the detailed timestamp and calibration result summary of the most recent self-calibration operation.
[0013] Compared with the prior art, the beneficial effects of the present invention by adopting the above technical solution are as follows: 1. This invention constructs a flow path state evolution graph, treating each detection flow path as a graph node and the liquid transfer relationship as a graph edge. A three-layer graph convolutional network is used to deduce the node states, enabling accurate prediction of the contamination risk value and availability score of each flow path. By fully considering three dimensions—historical contamination status, current availability, and detection adaptability for specific water samples—each detection is performed on the flow path with the optimal overall conditions, effectively avoiding interference from cross-contamination and making it particularly suitable for water body detection scenarios with complex compositions and diverse pollutant types. 2. This invention obtains baseline reference values by introducing a blank solution and establishes a concentration-response standard curve by introducing at least two standard solutions of different concentrations. It can simultaneously calculate the drift compensation coefficient and sensitivity correction coefficient of each sensor, realizing the coordinated correction of sensor baseline drift and sensitivity attenuation. The calibration effect is significantly better than the traditional single-point calibration or fixed-period calibration method, effectively ensuring the detection accuracy of the sensor during long-term continuous operation. 3. This invention introduces a cognitive reasoning module based on a pollutant knowledge graph, combining features extracted by deep learning with domain knowledge. By performing path search and confidence propagation on the graph, it infers the pollutant type, giving the identification results the dual advantages of data-driven and knowledge-driven approaches, significantly improving the accuracy and interpretability of pollutant identification in complex aquatic environments. Attached Figure Description
[0014] Figure 1 This is a flowchart of the water pollutant identification method in an embodiment of the present invention. Detailed Implementation
[0015] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings. It should be noted that the description of these embodiments is for the purpose of helping to understand the present invention, but does not constitute a limitation of the present invention.
[0016] Furthermore, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.
[0017] Please see the appendix Figure 1 This invention discloses a water pollutant identification method based on multimodal data fusion and cognitive reasoning, comprising real-time water quality status acquisition, water sample parameter acquisition, water quality status judgment, water quality data processing, and flow path cleaning. The specific steps of the water pollutant identification method are as follows: Step 1: Real-time acquisition of state parameters for each detection flow path; construction of a flow path state evolution map with flow paths as nodes and liquid transfer relationships as edges; and use graph neural networks to deduce the contamination risk value of each flow path. and usability rating ,in, , This represents the total number of flow paths; Step 2: Obtain the initial characteristic parameters of the water sample to be tested, and input them into the flow path adaptability evaluation model to obtain the adaptability score of each flow path. Based on comprehensive evaluation score The path with the highest value is selected as the target detection flow path, and a flow path switching operation is performed. , , The preset weighting coefficients satisfy the following conditions: ; Step 3: When the current state meets any of the conditions, perform a self-calibration operation. The self-calibration operation includes introducing a blank solution to obtain a baseline reference value, and establishing a concentration-response standard curve with two standard solutions of different concentrations. The drift compensation coefficient and sensitivity correction coefficient of each sensor are calculated based on the baseline reference value and the standard curve. Step 4: Simultaneously collect multimodal raw data of the water sample to be tested through a multimodal sensor array, preprocess it and input it into a multimodal fusion encoder, perform path search and confidence propagation based on a preset pollutant knowledge graph, and output the pollutant type identification result and its confidence level. Step 5: After the detection is completed, perform the flow path cleaning operation, record the operation information of this detection and cleaning, and fill the operation information back into the flow path state evolution graph, and update the attribute information of the relevant nodes and edges accordingly. Example
[0018] It was applied to an automatic surface water quality monitoring station. I. System Configuration: In this embodiment, the multimodal sensing array includes: a UV-Vis absorption spectroscopy detection module (wavelength range 190nm~1100nm, resolution 1nm), a fluorescence spectroscopy detection module (excitation wavelength 200nm~600nm, emission wavelength 250nm~700nm), an electrochemical sensor array (including pH electrode, dissolved oxygen electrode, conductivity electrode and redox potential electrode), a temperature sensor (accuracy ±0.1℃) and a pH sensor (accuracy ±0.01pH). The flow path switching matrix consists of 5 parallel detection flow paths, each detection flow path is independently equipped with an injection valve, a metering pump, a detection cell and a waste discharge valve. The standard solution storage unit includes a blank solution (ultrapure water) and a mixed standard solution of low, medium and high concentrations. II. Flow path state perception and evolution map construction: The status parameters of the five detection flow paths are collected in real time. Taking flow path No. 1 as an example, its current status parameters are as follows: the liquid type of the current operation is identified as "water sample to be tested", the liquid type of the last operation is identified as "standard solution B", the number of cleanings since the last operation is 2, the idle time since the last operation is 15 minutes, the cumulative number of operations is 186, the cumulative running time is 312 hours, the cleanliness index of the current detection pool is 0.92, the full score is 1.0, and the type of residual liquid in the current flow path is identified as "none". Five detection flow paths were used as nodes to construct a flow path state evolution graph. Node attributes included the historical liquid type sequence (last 50 operation records) and historical contaminant contact records for each flow path. Edge attributes included the liquid transfer timestamp and transfer amount between flow paths. A three-layer graph convolutional network was used for state inference. The first layer mapped the initial state parameters of each node to a 128-dimensional hidden state vector. The second layer aggregated the state information of adjacent nodes through graph convolution. The third layer output the contamination risk value and availability score of each node. The calculated contamination risk values for the five flow paths were as follows: =0.12、 =0.08、 =0.21、 =0.05、 =0.15, usability scores are respectively =0.85、 =0.92、 =0.78、 =0.95、 =0.82; III. Adaptive Flow Path Switching Decision: Obtain the initial characteristic parameters of the water sample to be tested: turbidity 3.2 NTU, conductivity 485 μS / cm, preliminary pH 7.3. Input the above parameters into the flow path adaptability evaluation model trained using the gradient boosting decision tree algorithm, and calculate the adaptability score of each flow path. =0.78、 =0.65、 =0.88、 =0.72、 =0.81; Set weight coefficients =0.35、 =0.30、 =0.35, calculate the comprehensive optimization score for each flow path: =0.35×(1-0.12)+0.30×0.85+0.35×0.78=0.308+0.255+0.273=0.836; =0.35×(1-0.08)+0.30×0.92+0.35×0.65=0.322+0.276+0.228=0.826; =0.35×(1-0.21)+0.30×0.78+0.35×0.88=0.2765+0.234+0.308=0.8185; =0.35×(1-0.05)+0.30×0.95+0.35×0.72=0.3325+0.285+0.252=0.8695; =0.35×(1-0.15)+0.30×0.82+0.35×0.81=0.2975+0.246+0.2835=0.827; The highest overall optimization score was for flow path 4. =0.8695), the control flow path switching matrix switches the water sample to be tested to the 4th detection flow path; IV. Self-calibration judgment: The calibration trigger conditions are as follows: the cumulative number of tests since the last calibration of flow path 4 is 42, which is less than the preset threshold of 50; the cumulative running time is 68 hours, which is less than the preset time threshold of 100 hours; the pollution risk value R_4 = 0.05, which is less than the preset risk threshold of 0.20; and the turbidity change rate of the current water sample to be tested is 5%, which is less than the preset change rate threshold of 10%. None of these conditions are met, so self-calibration will not be performed this time, and the test will proceed directly. V. Multimodal Detection Data Acquisition and Fusion Inference: The water sample to be tested was introduced into the No. 4 flow path, and multimodal raw data were collected simultaneously: UV-Vis absorption spectrum: absorbance 0.385 at 220 nm, absorbance 0.212 at 254 nm, absorbance 0.156 at 280 nm; fluorescence spectrum: fluorescence intensity 1523 a.u at excitation 280 nm / emission 340 nm; electrochemical data: pH 7.32, dissolved oxygen 6.8 mg / L, conductivity 488 μS / cm, redox potential 312 mV, temperature 22.5℃. After preprocessing the above data, it is input into a multimodal fusion encoder. Feature vectors are extracted from the three modalities by independent encoders. A multi-head cross-modal attention mechanism (8-head attention) is used to calculate the mutual attention weight matrix and perform weighted fusion to generate a 512-dimensional fusion feature tensor. The fusion feature tensor is input into the cognitive reasoning module, which performs path search and confidence propagation on a knowledge graph containing 326 pollutants. The output identification results are: ammonia nitrogen (confidence 0.94), total phosphorus (confidence 0.87), and COD (confidence 0.91). After correction by drift compensation coefficient and sensitivity correction coefficient, the output concentration estimates are: ammonia nitrogen 0.82 mg / L, total phosphorus 0.15 mg / L, and COD 18.6 mg / L. VI. Post-inspection cleaning and status update: After the test is completed, the control flow path switching matrix switches the cleaning solution to flow path 4 for cleaning. The cleaning flow rate is 5 mL / min and the duration is 3 minutes. The operation information is recorded (test time: June 28, 2026, 14:32; flow path number: "04"; water sample source: "river section A"; test results: "ammonia nitrogen 0.82 mg / L, total phosphorus 0.15 mg / L, COD 18.6 mg / L"; cleaning execution status). The data is then backfilled into the flow path state evolution map, and the attribute information of node 4 and related edges is updated. Finally, a test report is generated, which includes pollutant type, concentration value, confidence percentage, flow path number, calibration status record, and most recent calibration timestamp. The report is then uploaded to the environmental monitoring center platform via wireless communication.
[0019] Based on the above embodiments, the method of the present invention successfully selects the fourth flow path with the highest comprehensive score from five parallel detection flow paths as the target detection flow path through flow path state evolution map and graph neural network inference. Without performing self-calibration operation, it simultaneously collects ultraviolet-visible absorption spectrum, fluorescence spectrum and electrochemical sensing data through multimodal sensor array. After fusion by cross-modal attention mechanism and combined with pollutant knowledge graph for inference, it verifies that the method of the present invention can achieve adaptive selection of flow paths and high-confidence accurate identification of multiple pollutants in actual surface water monitoring scenarios.
[0020] While the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the invention. Any variations and modifications can be made by those skilled in the art without departing from the spirit and scope of the invention. Therefore, any modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention, without departing from the scope of the invention, fall within the protection scope defined by the claims of the present invention.
Claims
1. A method for identifying water pollutants using multimodal data fusion and cognitive reasoning, comprising real-time water quality status acquisition, water sample parameter acquisition, water quality status judgment, water quality data processing, and flow path cleaning, characterized in that: The specific steps of the water pollutant identification method are as follows: Step 1: Real-time acquisition of state parameters for each detection flow path; construction of a flow path state evolution map with flow paths as nodes and liquid transfer relationships as edges; and use graph neural networks to deduce the contamination risk value of each flow path. and usability rating ,in, , This represents the total number of flow paths; Step 2: Obtain the initial characteristic parameters of the water sample to be tested, and input them into the flow path adaptability evaluation model to obtain the adaptability score of each flow path. Based on comprehensive evaluation score The path with the highest value is selected as the target detection flow path, and a flow path switching operation is performed. , , The preset weighting coefficients satisfy the following conditions: ; Step 3: When the current state meets any of the conditions, perform a self-calibration operation. The self-calibration operation includes introducing a blank solution to obtain a baseline reference value, and establishing a concentration-response standard curve with two standard solutions of different concentrations. The drift compensation coefficient and sensitivity correction coefficient of each sensor are calculated based on the baseline reference value and the standard curve. Step 4: Simultaneously collect multimodal raw data of the water sample to be tested through a multimodal sensor array, preprocess it and input it into a multimodal fusion encoder, perform path search and confidence propagation based on a preset pollutant knowledge graph, and output the pollutant type identification result and its confidence level. Step 5: After the detection is completed, perform the flow path cleaning operation, record the operation information of this detection and cleaning, and fill the operation information back into the flow path state evolution graph, and update the attribute information of the relevant nodes and edges accordingly.
2. The water pollutant identification method based on multimodal data fusion and cognitive reasoning according to claim 1, characterized in that: In step one, the specific method for using a graph neural network to perform state inference based on the flow path state evolution map is as follows: a three-layer graph convolutional network is constructed. The first layer maps the initial state parameters of each node to a 128-dimensional hidden state vector. The second layer aggregates the state information of neighboring nodes through graph convolution operations. The third layer outputs the pollution risk prediction value of each node. and usability score prediction The training samples for the graph neural network are derived from historical flow path state data and corresponding actual pollution detection results.
3. A method for identifying water pollutants using multimodal data fusion and cognitive reasoning according to claim 2, characterized in that: In step two, the method for constructing the flow path suitability evaluation model is as follows: Historical detection data is collected, including the initial characteristic parameters of the water sample to be tested in each test, the identifier of the selected detection flow path, and the recognition accuracy of that test. Using the initial characteristic parameters and the detection flow path identifier as input features, and the recognition accuracy as the prediction target, a gradient boosting decision tree algorithm is used to train the flow path suitability evaluation model, which is then used to calculate the suitability score of each flow path for the current water sample to be tested. .
4. A water pollutant identification method based on multimodal data fusion and cognitive reasoning according to claim 3, characterized in that: In step three, the method for establishing the concentration-response standard curve is as follows: For each standard solution, its response values are collected on the UV-Vis absorption spectroscopy module, fluorescence spectroscopy detection module, and electrochemical sensor array, respectively. A concentration-response fitting curve is independently established for each sensor. The fitting method adopts least squares polynomial regression, and the polynomial order is second or third. In the fitting process, the concentration value of the standard solution is used as the independent variable, and the sensor response value is used as the dependent variable. Any of the following conditions are met: the cumulative number of detections of the target detection flow path since the last calibration reaches a preset number threshold; the cumulative running time of the target detection flow path since the last calibration reaches a preset duration threshold; the pollution risk value Ri of the target detection flow path exceeds a preset risk threshold; and the change rate of the turbidity value of the current water sample to be tested exceeds a preset change rate threshold.
5. A water pollutant identification method based on multimodal data fusion and cognitive reasoning according to claim 4, characterized in that: In step three, the drift compensation coefficient is calculated as follows: the currently acquired blank solution response value is compared with the historically stored blank solution standard response value, and the response drift of each sensor at each characteristic wavelength or each detection channel is calculated. In response to the reciprocal of the drift amount As the drift compensation coefficient for the corresponding sensor, where the subscript Indicates the sensor number, subscript Indicates the characteristic wavelength or channel number; the sensitivity correction coefficient is calculated by taking the slope of the currently established standard curve. Standard slope of the standard curve stored in history Compare them by their ratio. This serves as the sensitivity correction coefficient for the corresponding sensor.
6. A method for identifying water pollutants using multimodal data fusion and cognitive reasoning according to claim 5, characterized in that: In step four, the cross-modal attention mechanism is implemented as follows: UV-Vis absorption spectroscopy data, fluorescence spectroscopy data, and electrochemical sensing data are used as three modal inputs, respectively. Each modality extracts its own feature vector through an independent encoder. Then, a multi-head cross-modal attention mechanism is used to calculate the mutual attention weight matrix between the modal features. Based on the attention weight matrix, the modal features are weighted and fused to generate a unified dimension fused feature tensor.
7. A method for identifying water pollutants using multimodal data fusion and cognitive reasoning according to claim 5, characterized in that: In step four, the method for constructing the pollutant knowledge graph is as follows: collect the spectral feature vectors, electrochemical response features, physicochemical property parameters and their interrelationships of known pollutants, construct a knowledge graph containing pollutant entities, feature entities and relation edges with pollutant types and feature parameters as entities and attribute associations and response associations as relation edges, and perform path search on the pollutant knowledge graph based on the fused feature tensor, calculate the confidence score of each candidate pollutant entity, and output the pollutant type with the highest confidence score as the recognition result.
8. A method for identifying water pollutants using multimodal data fusion and cognitive reasoning according to claim 5, characterized in that: In step five, the execution parameters of the cleaning operation are based on the contamination risk value of the target detection flow path. It is dynamically adjusted based on the types of pollutants currently detected and identified.
9. A method for identifying water pollutants using multimodal data fusion and cognitive reasoning according to claim 8, characterized in that: In step four, the pollutant type identification results and the final concentration estimates of each pollutant are combined to generate a water quality pollutant detection report and output it to the outside world. The detection report includes the pollutant type name, the concentration value of each pollutant, the identification confidence percentage, the target detection flow path number used in this detection, the self-calibration status record before this detection, and the detailed timestamp and calibration result summary of the most recent self-calibration operation.