A water quality pollution monitoring method and device fusing multi-source data and a storage medium

Abstract

The application provides a water quality pollution monitoring method and device fusing multi-source data and a storage medium, and relates to the technical field of water quality monitoring. The method deeply integrates multi-source water quality data and monitoring point spatial topological information through key steps such as data preprocessing, cross-modal feature fusion, pollution concentration estimation and graph neural network tracing, and constructs a full-process automatic monitoring system. The application realizes end-to-end processing from data standardization to pollution emission source positioning, effectively breaks through the barrier of multi-source heterogeneous data, fully excavates the complementary value and spatio-temporal correlation law of different modal data, improves the integrity, accuracy and efficiency of water quality pollution monitoring, accurately locates the position of the emission source and the pollution diffusion path, provides full-link and targeted technical support for water quality pollution control and management decision-making, and solves the problems of low utilization rate of traditional monitoring data and insufficient tracing accuracy.

Description

Technical Field

[0001] This invention relates to the field of water quality monitoring technology, specifically to a method, device, and storage medium for monitoring water pollution by integrating multi-source data. Background Technology

[0002] Currently, water pollution has become a major challenge in the global environmental protection field. The accelerated pace of industrialization and urbanization has further exacerbated water pollution problems. Anthropogenic pollution sources such as illegal discharge of industrial wastewater, agricultural non-point source pollution, and urban sewage discharge have led to increasingly complex compositions of pollutants such as heavy metals, organic pollutants, and pathogenic microorganisms in water bodies. This not only disrupts the balance of aquatic ecosystems but also threatens the sustainable use of water resources and human health.

[0003] To address water pollution, water environment monitoring technology has emerged as a core tool for pollution prevention and control. However, traditional monitoring methods have significant limitations: First, while manual sampling and laboratory physicochemical analysis can achieve precise quantification, they require cumbersome sample pretreatment and expensive instruments, resulting in poor monitoring timeliness and limited spatial coverage, making it difficult to capture dynamic changes in pollution and sudden pollution events. Second, single monitoring technologies have inherent shortcomings—biological monitoring is greatly affected by the environment and biological life cycle and is time-consuming, so it can only be used as a supplementary means; while remote sensing monitoring can cover large areas of water, it cannot accurately identify the specific components and quantitative concentrations of pollutants, requiring reliance on conventional monitoring. Third, multi-source monitoring data suffers from spatiotemporal inconsistencies, heterogeneous formats, and dimensional differences, making it difficult for traditional technologies to effectively integrate the data, leading to fragmented monitoring information and failing to provide comprehensive support for pollution source tracing and remediation.

[0004] In recent years, although technologies such as online monitoring sensors, surface-enhanced Raman scattering (SERS), and the Internet of Things have been gradually applied to water environment monitoring, achieving breakthroughs in real-time performance and sensitivity, the core bottleneck in current technological development remains how to further integrate the advantages of multi-source data, improve monitoring accuracy and reliability, and simultaneously achieve full-chain coverage from "pollution monitoring" to "precise location of pollution sources." Furthermore, existing monitoring systems lack intelligent analysis capabilities, making it difficult to mine pollution propagation patterns and trends based on monitoring data. This results in a lack of scientific and efficient technical support for pollution prevention and control decisions, failing to meet the urgent needs of water resource protection and ecosystem health maintenance. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a water pollution monitoring method, device and storage medium that integrates multi-source data to address the shortcomings of the prior art.

[0006] The technical solution of this invention to solve the above-mentioned technical problems is as follows: A water pollution monitoring method integrating multi-source data, comprising the following steps:

[0007] S1: Import multi-source raw water quality data collected from multiple monitoring points, perform spatiotemporal alignment processing on the multi-source raw water quality data, and perform standardization processing on the aligned multi-source raw water quality data to obtain a standardized water quality dataset.

[0008] S2: Construct a cross-modal feature fusion model for water pollution, and perform feature fusion processing on the standardized water quality dataset based on the cross-modal feature fusion model for water pollution to obtain fused features;

[0009] S3: Import spatial topology information from multiple monitoring points, combine the fused features with the spatial topology information to obtain comprehensive features, and construct an adaptive pollution concentration estimation model. Estimate the comprehensive features based on the adaptive pollution concentration estimation model and output the pollution concentration estimation value corresponding to each monitoring point.

[0010] S4: Construct a graph structure with multiple monitoring points as nodes and water connectivity as edges. Embed the estimated pollution concentration values ​​corresponding to each monitoring point as graph node features into the graph structure to construct a graph neural network source tracing model. Based on the graph neural network source tracing model, learn the pollution propagation correlation between nodes in the graph structure. Based on the pollution propagation correlation, output the probability that each node is a pollution emission source by classifying nodes. Locate the pollution emission source based on the probability of the pollution emission source.

[0011] Another technical solution of the present invention to solve the above-mentioned technical problems is as follows: A water pollution monitoring device that integrates multi-source data, comprising:

[0012] The data preprocessing module is used to import multi-source raw water quality data collected from multiple monitoring points, perform spatiotemporal alignment processing on the multi-source raw water quality data, and perform standardization processing on the aligned multi-source raw water quality data to obtain a standardized water quality dataset.

[0013] The cross-modal feature fusion module is used to construct a cross-modal feature fusion model for water pollution, and to perform feature fusion processing on the standardized water quality dataset based on the cross-modal feature fusion model for water pollution to obtain fused features;

[0014] The pollution concentration estimation module is used to import spatial topology information from multiple monitoring points, concatenate the fused features with the spatial topology information to obtain comprehensive features, construct an adaptive pollution concentration estimation model, estimate the comprehensive features based on the adaptive pollution concentration estimation model, and output the pollution concentration estimation value corresponding to each monitoring point.

[0015] The pollution source tracing and location module is used to construct a graph structure with multiple monitoring points as nodes and water body connectivity as edges. The estimated pollution concentration values ​​corresponding to each monitoring point are embedded into the graph structure as graph node features to construct a graph neural network source tracing model. Based on the graph neural network source tracing model, the module learns the pollution propagation correlation between nodes in the graph structure, and outputs the probability that each node is a pollution emission source based on the pollution propagation correlation through node classification. Based on the probability of pollution emission source, the module locates the pollution emission source.

[0016] Another technical solution of the present invention to solve the above-mentioned technical problems is as follows: a water pollution monitoring device that integrates multi-source data, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it executes the above-mentioned water pollution monitoring method that integrates multi-source data.

[0017] Another technical solution of the present invention to solve the above-mentioned technical problems is as follows: the computer-readable storage medium stores a computer program, and when the computer program runs, it controls the device where the computer-readable storage medium is located to execute the above-mentioned water pollution monitoring method that integrates multi-source data.

[0018] The beneficial effects of this invention are: by integrating multi-source water quality data and spatial topology information through multiple main steps such as data preprocessing, cross-modal feature fusion, pollution concentration estimation and graph neural network source tracing, a fully automated monitoring system is constructed, realizing end-to-end processing from data standardization to emission source location, breaking down multi-source data barriers, improving the integrity and efficiency of water pollution monitoring, and providing full-link technical support for pollution control. Attached Figure Description

[0019] Figure 1 A flowchart of a water pollution monitoring method provided in an embodiment of the present invention;

[0020] Figure 2 This is a block diagram of a water pollution monitoring device provided in an embodiment of the present invention. Detailed Implementation

[0021] The principles and features of the present invention are described below with reference to the accompanying drawings. The examples given are only for explaining the present invention and are not intended to limit the scope of the present invention.

[0022] Example 1

[0023] like Figure 1 As shown in the figure, this invention provides a water pollution monitoring method that integrates multi-source data, including the following steps:

[0024] S1. Import multi-source raw water quality data collected from multiple monitoring points, perform spatiotemporal alignment processing on the multi-source raw water quality data, and standardize the aligned multi-source raw water quality data to obtain a standardized water quality dataset.

[0025] S2. Construct a cross-modal feature fusion model for water pollution, and perform feature fusion processing on the standardized water quality dataset based on the cross-modal feature fusion model for water pollution to obtain fused features;

[0026] S3. Import the spatial topology information of multiple monitoring points, splice the fused features with the spatial topology information to obtain comprehensive features, and construct an adaptive pollution concentration estimation model. Estimate the comprehensive features based on the adaptive pollution concentration estimation model and output the pollution concentration estimation value corresponding to each monitoring point.

[0027] S4. Construct a graph structure with multiple monitoring points as nodes and water connectivity as edges. Embed the estimated pollution concentration values ​​corresponding to each monitoring point as graph node features into the graph structure to construct a graph neural network source tracing model. Based on the graph neural network source tracing model, learn the pollution propagation correlation between nodes in the graph structure. Based on the pollution propagation correlation, output the probability that each node is a pollution emission source by classifying nodes. Based on the probability of pollution emission sources, locate the pollution emission source.

[0028] In the above embodiments, by performing multiple main steps such as data preprocessing, cross-modal feature fusion, pollution concentration estimation, and graph neural network source tracing on the original data of multi-source water quality, multi-source water quality data and spatial topology information are integrated to build a fully automated monitoring system. This system achieves end-to-end processing from data standardization to emission source location, breaks down multi-source data barriers, improves the integrity and efficiency of water pollution monitoring, and provides full-link technical support for pollution control.

[0029] Preferably, the process involves importing multi-source raw water quality data collected from multiple monitoring points, performing spatiotemporal alignment on the multi-source raw water quality data, and then standardizing the aligned multi-source raw water quality data to obtain a standardized water quality dataset. This includes the following specific steps:

[0030] S1.1 Import multi-source raw water quality data collected from multiple monitoring points. Classify and label the imported multi-source raw water quality data according to data type and monitoring point affiliation to obtain a classified and labeled data set. The data types include real-time sensor monitoring data, satellite remote sensing data, hydrological and meteorological data, and historical pollution event data. Each data entry is associated with a unique monitoring point identifier, data type label, and collection time and space stamp (e.g., collection time, collection latitude and longitude).

[0031] S1.2. Extract time series of different types of data from each monitoring point based on the classification and labeling data set, and use the set time as the unified time granularity to adaptively interpolate and supplement the data of each missing time node in the classification and labeling data set using the temporal correlation of the time series, so as to obtain a multi-type data subset of each monitoring point with time synchronization.

[0032] S1.3. Using the latitude and longitude of each monitoring point as a reference, the spatial nearest neighbor interpolation method is used to correct the low spatial resolution positional deviation of each data in the multi-type data subset, so as to obtain a multi-source data set after spatiotemporal two-dimensional alignment, so that all data entries are accurately matched to the spatial coordinates of the corresponding monitoring points.

[0033] S1.4. Adaptive multi-source data format conversion model is used to convert the various data in different formats (e.g., CSV, HDF5, TIFF, etc.) in the multi-source data set into a structured data table format to obtain a standardized water quality dataset; wherein, the structured data table format includes monitoring point identifier, collection time, spatial coordinates and data type label.

[0034] In step S1.4, the adaptive multi-source data format conversion model can be the AMF-Conv conversion model.

[0035] In the above embodiments, multi-source heterogeneous water quality raw data are preprocessed in all dimensions through steps such as classification and labeling, spatiotemporal alignment, format conversion, and standardization. This solves the problems of data spatiotemporal inconsistency and inconsistent formats, and outputs a standardized dataset with unified structure, spatiotemporal synchronization, and consistent dimensions. This provides a high-quality data foundation for subsequent feature fusion and model calculation, and reduces the interference of data noise on monitoring results.

[0036] Preferably, a cross-modal feature fusion model for water pollution is constructed, and the standardized water quality dataset is subjected to feature fusion processing based on the cross-modal feature fusion model to obtain fused features. Specifically, this includes the following steps:

[0037] S2.1 Divide the standardized water quality dataset according to data modality type to obtain numerical modal data, image modal data and text modal data;

[0038] S2.2 Extract the numerical modal data using a fully connected neural network to obtain numerical features. , , The dimension of the numerical feature. Representing numerical features yes A real-valued vector of dimension; and the image-type modal data is extracted using a lightweight convolutional neural network to obtain image features. , , The dimension of image features. Representing image features yes A real-valued vector of dimension; and semantic features are obtained by extracting the textual modality data through a BERT-trained model. , , For the dimensions of text features, Representing semantic features yes A dimensional real vector;

[0039] S2.3, Numerical features Image features and semantic features Mapping to the same dimension yields features with unified dimensions. , and ,

[0040] ,

[0041] ,

[0042] ,

[0043] in, , , For each modality, the adaptation mapping matrix, , , For the corresponding bias term;

[0044] S2.4 Construct a modal attention mechanism module, and apply the modal attention mechanism module to the features after dimensional unification. , and The attention weights, and the expression for the modal attention mechanism module, are as follows:

[0045] ,

[0046] ,

[0047] in, Here is the attention weight parameter matrix. For attention bias, For the Sigmoid activation function, For feature splicing operations, For attention projection matrix, Numerical features Image features and semantic features The corresponding attention weights;

[0048] S2.5, based on various features , and Attention weights for each feature , and Weighted fusion is performed to obtain the final fusion features. , .

[0049] In the above embodiments, a cross-modal feature fusion model for water pollution is constructed. Through modal adaptation mapping and attention weight allocation, numerical, image, and text modal features are weighted and fused. The model adaptively mines complementary information from different modal data, strengthens the contribution of key features, avoids the limitations of single modal data, and improves the representation ability and pertinence of fused features.

[0050] Preferably, spatial topology information from multiple monitoring points is imported, and the fused features are concatenated with the spatial topology information to obtain comprehensive features. An adaptive pollution concentration estimation model is then constructed. Based on the adaptive pollution concentration estimation model, the comprehensive features are estimated, and the estimated pollution concentration values ​​for each monitoring point are output. This includes the following specific steps:

[0051] S3.1 Import spatial topology information from multiple monitoring points (including latitude and longitude coordinates of each monitoring point, water connectivity relationships, distances between monitoring points, and watershed boundaries, etc.). Using graph structure modeling, transform the spatial topology information into a relative distance matrix between monitoring points and a water connectivity adjacency matrix. Based on these matrixes, obtain a standardized spatial topology feature set. The relative distance matrix between monitoring points is represented as follows: The water body connectivity adjacency matrix is ​​represented as follows: ;

[0052] S3.2, Fusing Features Based on the one-to-one correspondence between monitoring point identifiers, the fused features of each monitoring point are concatenated with the corresponding row vector in the standardized spatial topology feature set (including the distance and connectivity information of the monitoring point to all other monitoring points) to obtain the comprehensive features of each monitoring point. It achieves information complementarity between multi-source data fusion features and standardized spatial topology features.

[0053] S3.3 Construct an adaptive pollution concentration estimation model, which includes an input layer, an adaptive feature adjustment layer, a temporal feature extraction layer, and an output layer.

[0054] The input layer receives integrated features. ,

[0055] The adaptive feature adjustment layer dynamically adjusts the weight ratio of standardized spatial topological features and fused features through a gating mechanism to obtain the adaptively adjusted feature vector. , Represented as:

[0056] ,

[0057] in, For adaptive weighting coefficients, This is to perform a normalization operation on each comprehensive feature.

[0058] The temporal feature extraction layer extracts adaptively adjusted feature vectors through a pre-trained ELSTM network. The temporal correlation is used to obtain temporal features;

[0059] Specifically, the hidden layer dimension of the ELSTM network is set to... Initialize the initial hidden state of the network. Compared with the initial cell state (Initially set to 0 for all vector values), the adaptively adjusted feature vectors The data is input into the ELSTM network in time series order, and then transformed through the dimension transformation module of the input layer. Adapt to the feature dimensions that the network can process.

[0060] Input features at each time step t The timing information is filtered and retained through the gating mechanism of the ELSTM network, specifically through the following steps:

[0061] Forget gate, based on pre-trained weight calculation Filter the historical time-series information that needs to be retained;

[0062] Input gate and candidate cell state, calculate and Update the feature information of the current time step;

[0063] Update cell state and hidden state, by Integrating historical and current information, and then... Output the hidden state at the current time step; This is the Sigmoid activation function.

[0064] Traversing the feature vectors of all time steps This yields the hidden state sequence for the entire time series;

[0065] The output layer uses a pre-trained BP network to perform regression fitting on the extracted time-series features, and outputs the estimated pollution concentration values ​​corresponding to each monitoring point.

[0066] In the above embodiments, by splicing and fusing features with standardized spatial topological features, an adaptive pollution concentration estimation model containing an adaptive feature adjustment layer and an ELSTM time-series extraction layer is constructed, which realizes accurate prediction of pollution concentration at monitoring points. By combining spatial correlation and temporal patterns, the accuracy and robustness of pollution concentration estimation are improved, providing accurate quantitative data support for subsequent pollution source tracing.

[0067] Preferably, a graph structure is constructed using multiple monitoring points as nodes and water connectivity as edges. The estimated pollution concentration values ​​corresponding to each monitoring point are embedded into the graph structure as graph node features to construct a graph neural network source tracing model. Based on the graph neural network source tracing model, the pollution propagation correlation between nodes in the graph structure is learned. Based on the pollution propagation correlation, the probability of each node being a pollution emission source is output through node classification. The location of the pollution emission source is located based on the probability of the pollution emission source. Specifically, the following steps are included:

[0068] S4.1. Define the graph structure based on the spatial distribution of multiple monitoring points and their connectivity with the water body. , where the set of nodes , For the number of monitoring points, each node Corresponding to one monitoring point,

[0069] edge set If monitoring point and If there is a water body connection between them, then construct an edge ( , The edge weights are calculated based on the water flow velocity and distance between the two monitoring points, and are expressed as follows:

[0070] ,

[0071] in, To measure the straight-line distance between monitoring points, The average flow velocity of the water body is represented by a weight value; a larger weight value indicates a stronger correlation with pollution transmission.

[0072] Estimated pollution concentrations at each monitoring point As a core feature, each node is constructed based on the latitude and longitude of the monitoring point and the watershed partition label. eigenvectors , and For longitude and latitude, The watershed is partitioned and encoded to form a node feature matrix. , For the first Estimated pollution concentration values ​​corresponding to each monitoring point;

[0073] S4.2. Construct a graph neural network source tracing model based on a graph attention network. The graph neural network source tracing model includes a feature aggregation layer, a causal propagation layer, and a classification output layer.

[0074] The feature aggregation layer uses an attention mechanism to weighted aggregate features of a node's neighborhood, calculating the node's... With neighboring nodes Attention coefficient ,

[0075] ,

[0076] in, For nodes The set of neighboring nodes, The characteristic transformation matrix, For feature splicing operations, Learnable attention weight vector transpose, For nodes The set of neighboring nodes Any node in the;

[0077] And update node features based on coefficients ,in, For activation function,

[0078] The causal propagation layer constructs causal constraints based on the water flow direction and pollutant diffusion rate, and corrects the neighborhood aggregation weights. If the water flows from a neighboring node... Flow to Node Then enhance the neighboring nodes For nodes The feature contribution weights are updated to update the features. ,in, For nodes The set of upstream neighboring nodes, The causal constraint coefficient has a range of values. ;

[0079] The classification output layer processes the final node features through a fully connected network. Classify and output nodes Probability of being a source of pollution emissions ,in, and This is the weight matrix. and For bias terms;

[0080] S4.3, structuring the graph Input the graph neural network source tracing model and output the probability that each node is a pollution emission source. If a node has a probability greater than or equal to the probability threshold, it is identified as a suspected pollution emission source, and the specific latitude and longitude coordinates of the emission source are output based on the geographical location information of the monitoring point corresponding to the node.

[0081] In the above embodiments, a graph structure is constructed with monitoring points as nodes and water connectivity as edges. The estimated pollution concentration is embedded as node features. The graph neural network, inspired by causality, learns the pollution propagation relationship and outputs the emission source probability. This fits the spatial causal logic of water pollution propagation, accurately locates the pollution emission source and diffusion path, improves the scientificity and reliability of source tracing, and provides precise targeted basis for pollution control.

[0082] Example 2

[0083] like Figure 2 As shown, this embodiment of the invention also provides a water pollution monitoring device that integrates multi-source data. This device corresponds to the method described in Embodiment 1, and includes:

[0084] The data preprocessing module is used to import multi-source raw water quality data collected from multiple monitoring points, perform spatiotemporal alignment processing on the multi-source raw water quality data, and perform standardization processing on the aligned multi-source raw water quality data to obtain a standardized water quality dataset.

[0085] The cross-modal feature fusion module is used to construct a cross-modal feature fusion model for water pollution, and to perform feature fusion processing on the standardized water quality dataset based on the cross-modal feature fusion model for water pollution to obtain fused features;

[0086] The pollution concentration estimation module is used to import spatial topology information from multiple monitoring points, concatenate the fused features with the spatial topology information to obtain comprehensive features, construct an adaptive pollution concentration estimation model, estimate the comprehensive features based on the adaptive pollution concentration estimation model, and output the pollution concentration estimation value corresponding to each monitoring point.

[0087] The pollution source tracing and location module is used to construct a graph structure with multiple monitoring points as nodes and water body connectivity as edges. The estimated pollution concentration values ​​corresponding to each monitoring point are embedded into the graph structure as graph node features to construct a graph neural network source tracing model. Based on the graph neural network source tracing model, the module learns the pollution propagation correlation between nodes in the graph structure, and outputs the probability that each node is a pollution emission source based on the pollution propagation correlation through node classification. Based on the probability of pollution emission source, the module locates the pollution emission source.

[0088] Example 3

[0089] This invention also provides a water pollution monitoring device that integrates multi-source data, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it performs the water pollution monitoring method that integrates multi-source data described above.

[0090] Example 4

[0091] This invention also provides a computer-readable storage medium storing a computer program that, when executed, controls the device containing the computer-readable storage medium to perform the aforementioned water pollution monitoring method that integrates multi-source data.

[0092] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0093] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the above-described apparatus and unit can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0094] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed.

[0095] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of the embodiments of the present invention, depending on actual needs.

[0096] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0097] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory, random access memory, portable hard drives, magnetic disks, or optical disks.

[0098] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for monitoring water pollution by integrating multi-source data, characterized in that, Includes the following steps: S1: Import multi-source raw water quality data collected from multiple monitoring points, perform spatiotemporal alignment processing on the multi-source raw water quality data, and perform standardization processing on the aligned multi-source raw water quality data to obtain a standardized water quality dataset. S2: Construct a cross-modal feature fusion model for water pollution, and perform feature fusion processing on the standardized water quality dataset based on the cross-modal feature fusion model for water pollution to obtain fused features; S3: Import spatial topology information from multiple monitoring points, combine the fused features with the spatial topology information to obtain comprehensive features, and construct an adaptive pollution concentration estimation model. Estimate the comprehensive features based on the adaptive pollution concentration estimation model and output the pollution concentration estimation value corresponding to each monitoring point. S4: Construct a graph structure with multiple monitoring points as nodes and water connectivity as edges. Embed the estimated pollution concentration values ​​corresponding to each monitoring point as graph node features into the graph structure to construct a graph neural network source tracing model. Based on the graph neural network source tracing model, learn the pollution propagation correlation between nodes in the graph structure. Based on the pollution propagation correlation, output the probability that each node is a pollution emission source through node classification. And locate the pollution emission source location based on the probability of the pollution emission source. Step S4 further includes the following steps: S4.1: Define the graph structure based on the spatial distribution of multiple monitoring points and the connectivity of water bodies. , where the set of nodes , For the number of monitoring points, each node Corresponding to one monitoring point, edge set If monitoring point and If there is a water body connection between them, then construct an edge ( , The edge weights are calculated based on the water flow velocity and distance between the two monitoring points, and are expressed as follows: ， in, To measure the straight-line distance between monitoring points, The average flow velocity of the water body is represented by a weight value; a larger weight value indicates a stronger correlation with pollution transmission. Estimated pollution concentration values ​​at each monitoring point As a core feature, each node is constructed based on the latitude and longitude of the monitoring point and the watershed partition label. eigenvectors , and For longitude and latitude, The watershed is partitioned and encoded to form a node feature matrix. , For the first Estimated pollution concentration values ​​corresponding to each monitoring point; S4.2: Construct a graph neural network source tracing model based on a graph attention network. The graph neural network source tracing model includes a feature aggregation layer, a causal propagation layer, and a classification output layer. The feature aggregation layer uses an attention mechanism to weighted aggregate features of a node's neighborhood, calculating the node's... With neighboring nodes Attention coefficient , ， in, For nodes The set of neighboring nodes, The characteristic transformation matrix, For feature splicing operations, Learnable attention weight vector transpose, For nodes The set of neighboring nodes Any node in the; And update node features based on coefficients ,in, For activation function, The causal propagation layer constructs causal constraints based on the water flow direction and pollutant diffusion rate, and corrects the neighborhood aggregation weights. If the water flows from a neighboring node... Flow to Node Then enhance the neighboring nodes For nodes The feature contribution weights are updated to update the features. ,in, For nodes The set of upstream neighboring nodes, The causal constraint coefficient has a range of values. ; The classification output layer processes the final node features through a fully connected network. Classify and output nodes Probability of being a source of pollution emissions ,in, and This is the weight matrix. and For bias terms; S4.3: Graph Structure Input the graph neural network source tracing model and output the probability that each node is a pollution emission source. If a node has a probability greater than or equal to the probability threshold, it is identified as a suspected pollution emission source, and the specific latitude and longitude coordinates of the emission source are output based on the geographical location information of the monitoring point corresponding to the node.

2. The water pollution monitoring method integrating multi-source data according to claim 1, characterized in that, Step S1 further includes the following steps: S1.1: Import multi-source raw water quality data collected from multiple monitoring points, classify and label the imported multi-source raw water quality data according to data type and monitoring point affiliation, and obtain a classified and labeled data set; among them, data types include real-time sensor monitoring data, satellite remote sensing data, hydrological and meteorological data, and historical pollution event data, and each data entry is associated with a unique monitoring point identifier, data type label, and collection time and space stamp; S1.2: Extract time series of different types of data from each monitoring point based on the classification and labeling data set, and use the set time as the unified time granularity to adaptively interpolate and supplement the data with missing time nodes in the classification and labeling data set using the temporal correlation of the time series, so as to obtain a multi-type data subset of time synchronization for each monitoring point. S1.3: Using the latitude and longitude of each monitoring point as a reference, the spatial nearest neighbor interpolation method is used to correct the low spatial resolution positional deviation of each data in the multi-type data subset, so as to obtain a multi-source data set after spatiotemporal two-dimensional alignment; S1.4: An adaptive multi-source data format conversion model is used to uniformly convert the various data in different formats in the multi-source data set into a structured data table format to obtain a standardized water quality dataset; wherein, the structured data table format includes monitoring point identifiers, collection time, spatial coordinates, and data type labels.

3. The water pollution monitoring method integrating multi-source data according to claim 1, characterized in that, Step S2 further includes the following steps: S2.1: The standardized water quality dataset is divided according to the data modality type to obtain numerical modal data, image modal data and text modal data; S2.2: Extract the numerical modal data using a fully connected neural network to obtain numerical features. , , The dimension of the numerical feature. Representing numerical features yes A real-valued vector of dimension; the image-type modal data is extracted using a lightweight convolutional neural network to obtain image features. , , The dimension of image features. Representing image features yes A real-valued vector of dimension; semantic features are obtained by extracting the textual modal data using a BERT-trained model. , , For the dimensions of text features, Representing semantic features yes A dimensional real vector; S2.3: Numerical characteristics Image features and semantic features Mapping to the same dimension yields features with unified dimensions. , and , ， ， ， in, , , For each modality, the adaptation mapping matrix, , , For the corresponding bias term; S2.4: Construct a modal attention mechanism module, and apply the modal attention mechanism module to the features after dimensionality unification. , and The attention weights, and the expression for the modal attention mechanism module, are as follows: ， ， in, Here is the attention weight parameter matrix. For attention bias, It is the Sigmoid activation function. For feature splicing operations, For attention projection matrix, Numerical features Image features and semantic features The corresponding attention weights; S2.5: Based on various features , and Attention weights for each feature , and Weighted fusion is performed to obtain the final fusion features. , .

4. The water pollution monitoring method integrating multi-source data according to claim 1, characterized in that, Step S3 includes the following steps: S3.1: Import spatial topology information from multiple monitoring points. Through graph structure modeling, transform the spatial topology information into a relative distance matrix between monitoring points and a water body connectivity adjacency matrix. Based on the relative distance matrix between monitoring points and the water body connectivity adjacency matrix, obtain a standardized spatial topology feature set; wherein, the relative distance matrix between monitoring points is represented as... The water body connectivity adjacency matrix is ​​represented as follows: ; S3.2: Map the fused features one-to-one with the monitoring point identifiers, and concatenate the fused features of each monitoring point with the corresponding row vector in the standardized spatial topological feature set to obtain the comprehensive features of each monitoring point. ; S3.3: Construct an adaptive pollution concentration estimation model, which includes an input layer, an adaptive feature adjustment layer, a temporal feature extraction layer, and an output layer; The input layer receives integrated features ; The adaptive feature adjustment layer dynamically adjusts the weight ratio of standardized spatial topological features and fused features through a gating mechanism to obtain the adaptively adjusted feature vector. , Represented as: ， in, For adaptive weighting coefficients, To perform normalization operations on each comprehensive feature; The temporal feature extraction layer extracts adaptively adjusted feature vectors through a pre-trained ELSTM network. The temporal correlation is used to obtain temporal features; The output layer uses a pre-trained BP network to perform regression fitting on the extracted time-series features, and outputs the estimated pollution concentration values ​​corresponding to each monitoring point.

5. A water pollution monitoring device that integrates multi-source data, employing the water pollution monitoring method that integrates multi-source data as described in any one of claims 1 to 4, characterized in that, include: The data preprocessing module is used to import multi-source raw water quality data collected from multiple monitoring points, perform spatiotemporal alignment processing on the multi-source raw water quality data, and perform standardization processing on the aligned multi-source raw water quality data to obtain a standardized water quality dataset. The cross-modal feature fusion module is used to construct a cross-modal feature fusion model for water pollution, and to perform feature fusion processing on the standardized water quality dataset based on the cross-modal feature fusion model for water pollution to obtain fused features; The pollution concentration estimation module is used to import spatial topology information from multiple monitoring points, concatenate the fused features with the spatial topology information to obtain comprehensive features, construct an adaptive pollution concentration estimation model, estimate the comprehensive features based on the adaptive pollution concentration estimation model, and output the pollution concentration estimation value corresponding to each monitoring point. The pollution source tracing and location module is used to construct a graph structure with multiple monitoring points as nodes and water body connectivity as edges. The estimated pollution concentration values ​​corresponding to each monitoring point are embedded into the graph structure as graph node features to construct a graph neural network source tracing model. Based on the graph neural network source tracing model, the module learns the pollution propagation correlation between nodes in the graph structure, and outputs the probability that each node is a pollution emission source based on the pollution propagation correlation through node classification. Based on the probability of pollution emission source, the module locates the pollution emission source.

6. A water pollution monitoring device that integrates multi-source data, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it performs the water pollution monitoring method that integrates multi-source data as described in any one of claims 1 to 4.

7. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when the computer program is executed, controls the device containing the computer-readable storage medium to perform the water pollution monitoring method that integrates multi-source data as described in any one of claims 1 to 4.

Images