An intelligent groundwater quality supervision method and system

Abstract

The present invention discloses an intelligent groundwater water quality supervision method and system. The method includes heterogeneous sensing acquisition, water quality benchmark modeling, sewage source tracing enhancement, water quality prediction and early warning, and groundwater water quality supervision. The present invention relates to the technical field of water quality supervision, specifically an intelligent groundwater water quality supervision method and system. The present invention adopts a comprehensive groundwater water quality supervision method combining water quality benchmark modeling, enhanced pollution source tracing, and dynamic threshold water quality prediction and early warning, effectively improving the perception ability, identification ability, and response ability of groundwater pollution incidents, and realizing the supervision transformation from "passive response" to "active prediction"; an improved spatio-temporal graph convolutional network method combined with a spatio-temporal dynamic baseline algorithm is used for water quality benchmark modeling; an improved adversarial generation network method for virtual and real source tracing modeling is used for pollution source tracing enhancement.

Description

Technical Field

[0001] The present invention relates to the technical field of water quality supervision, and specifically refers to an intelligent groundwater quality supervision method and system. Background Art

[0002] The intelligent groundwater quality supervision method is a modern environmental monitoring technology system for intelligent analysis. It collects multi-dimensional water quality data in real time by deploying an intelligent sensing network, constructs an adaptive evaluation benchmark using spatio-temporal dynamic modeling and machine learning algorithms, combines digital twin and pollution source tracing technologies to achieve precise positioning and diffusion prediction of pollution events, and finally forms a full-process closed-loop management of "monitoring - early warning - decision-making - disposal", so as to provide a scientific basis for water resource protection, pollution control and emergency decision-making. It is the core technical means to realize the sustainable management of groundwater resources and intelligent environmental protection.

[0003] However, in the existing groundwater quality supervision methods, there are basic problems such as scattered monitoring data, difficulty in timely tracking of pollution changes, and lagging early warning mechanisms, resulting in a lack of overall understanding of pollution evolution during the supervision process, making it difficult to take effective countermeasures in a timely manner. In the face of the actual situation of unclear pollution sources, complex groundwater flow paths and diverse pollutants, traditional methods often rely only on fixed-point monitoring, and are unable to achieve comprehensive tracking and trend prediction of groundwater pollution; in the existing water quality benchmark modeling methods, there is a technical problem that existing methods often use static standards or historical averages as the judgment basis, ignoring the natural volatility and complex impacts of human interference of groundwater quality at different regions and different time scales. This "averaging" modeling method is prone to misjudgment, either false reporting of pollution or missing abnormal reports; in the existing sewage source tracing enhancement methods, there is a technical problem that traditional source tracing methods mostly rely on the concentration inversion of pollutants and the inference of hydrogeological maps, which are both easily limited by sparse sample points and incomplete data, and are also difficult to cope with complex scenarios of multi-source superposition or hidden pollution, resulting in unclear source tracing results and inaccurate path judgment. Summary of the Invention

[0004] In view of the above situation, to overcome the defects of the existing technology, the present invention provides an intelligent groundwater quality supervision method and system. In the existing groundwater quality supervision methods, there are basic problems such as scattered monitoring data, difficulty in timely tracking of pollution changes, and lagging warning mechanisms, resulting in a lack of overall understanding of pollution evolution in the supervision process and difficulty in taking effective response measures in a timely manner. In the face of the actual situation of unclear pollution sources, complex groundwater flow paths, and diverse pollutants, traditional methods often rely only on fixed-point monitoring and cannot achieve comprehensive tracking and trend prediction of groundwater pollution. To solve this technical problem, this solution creatively adopts a comprehensive groundwater quality supervision method that combines water quality benchmark modeling, enhanced pollution source tracing, and dynamic threshold water quality prediction and warning, effectively improving the perception ability, identification ability, and response ability of groundwater pollution incidents, and realizing the supervision transformation from "passive response" to "active prediction"; in the existing water quality benchmark modeling methods, there is a technical problem that existing methods often use static standards or historical averages as the judgment basis, ignoring the natural volatility and complex impacts of human interference of groundwater quality at different regions and different time scales. This "averaging" modeling method is prone to judgment errors, either false alarms of pollution or missed reports of anomalies. To solve this problem, this solution creatively adopts an improved spatio-temporal graph convolutional network method combined with a spatio-temporal dynamic baseline algorithm for water quality benchmark modeling, which can more scientifically reflect which water quality changes belong to normal fluctuations and which belong to abnormal pollution, thus providing a more accurate judgment basis for subsequent source tracing and warning; in the existing sewage source tracing enhancement methods, there is a technical problem that traditional source tracing methods mostly rely on the concentration inversion of pollutants and the inference of hydrogeological maps, which are both easily limited by sparse sample points and incomplete data, and are difficult to handle complex scenarios of multi-source superposition or hidden pollution, resulting in unclear source tracing results and inaccurate path judgment. To solve this problem, this solution creatively adopts an improved generative adversarial network method for virtual and real source tracing modeling to enhance pollution source tracing, which not only improves the accuracy of pollution source location, but also reveals the diffusion trend of pollutants in groundwater, enabling the supervision unit to detect hidden pollution earlier and accurately lock the pollution source, significantly enhancing the forward-looking and pertinence of pollution prevention and control.

[0005] The technical solution adopted by the present invention is as follows: An intelligent groundwater quality supervision method provided by the present invention, the method includes the following steps:

[0006] Step S1: Heterogeneous sensor acquisition;

[0007] Step S2: Water quality benchmark modeling;

[0008] Step S3: Sewage source tracing enhancement;

[0009] Step S4: Water quality prediction and warning;

[0010] Step S5: Groundwater quality supervision.

[0011] Further, in step S1, the heterogeneous sensing acquisition is used to collect the original data set required for groundwater quality supervision. Specifically, the original groundwater quality data set is obtained through sensing array data acquisition, and the optimized preprocessing is performed on the original groundwater quality data set to obtain optimized groundwater sensing data;

[0012] The original groundwater quality data set includes physical and chemical parameter data, ion concentration data, and sensing metadata;

[0013] The steps of the optimized preprocessing include data cleaning, data alignment, noise filtering, data standardization, data normalization, feature optimization, and data labeling;

[0014] The optimized groundwater sensing data includes optimized physical and chemical parameter data, optimized ion concentration data, and optimized sensing metadata.

[0015] Further, in step S2, the water quality benchmark modeling is used to construct a dynamic water quality assessment benchmark model. Specifically, based on the optimized groundwater sensing data, an improved spatio-temporal graph convolutional network method combined with a spatio-temporal dynamic baseline algorithm is used for water quality benchmark modeling to obtain multi-dimensional reference data on water quality healthiness, including the following steps:

[0016] Step S21: Multi-scale feature extraction. Specifically, a one-dimensional convolutional long short-term neural network is constructed as a hybrid encoder to extract multi-scale feature data from the optimized groundwater sensing data; the multi-scale feature data includes local features and global features; the local features are used to represent hourly-level sensing time data, and the global features are used to represent seasonal-level sensing time data;

[0017] The one-dimensional convolutional long short-term neural network is improved by introducing dilated convolution to replace the standard one-dimensional convolutional layer for time series perception enhancement;

[0018] Step S22: Spatio-temporal dynamic benchmark modeling. Specifically, spatio-temporal dynamic baseline modeling is performed through spatial correlation modeling and time dynamics modeling to obtain time-geological feature data;

[0019] The spatial correlation modeling specifically constructs a groundwater spatial relationship graph structure through hydrogeological conductivity and pollutant diffusivity modeling, and calculates spatial correlation feature data through a standard graph convolutional network;

[0020] The time dynamics modeling is specifically carried out by adopting an improved spatio-temporal graph convolutional network for spatio-temporal attention extraction. The improved spatio-temporal graph convolutional network specifically introduces a gated spatio-temporal attention mechanism on the basis of the standard spatio-temporal graph convolutional network to process spatio-temporal features and obtain time dynamics modeling feature data;

[0021] The time geological feature data is specifically obtained by combining the spatial correlation feature data and the time dynamics modeling feature data into a spatio-temporal dynamic baseline, and is used as a reference benchmark for water quality health;

[0022] Step S23: Water quality health calculation. Specifically, the Mahalanobis distance between the optimized groundwater sensing data and the time geological feature data is calculated to obtain deviation data, and a multi-dimensional health fusion mechanism is constructed to calculate the water quality health and obtain a water quality health output;

[0023] Step S24: Water quality benchmark modeling training. Specifically, through the multi-scale feature extraction, the spatio-temporal dynamic baseline modeling, and the water quality health calculation, water quality benchmark modeling training is carried out to obtain a water quality benchmark model Model WHI ;

[0024] Step S25: Water quality benchmark modeling. Specifically, based on the optimized groundwater sensing data, the water quality benchmark model Model WHI is used to carry out water quality benchmark modeling to obtain multi-dimensional reference data of water quality health;

[0025] The multi-dimensional reference data of water quality health includes multi-dimensional quantitative output of water quality health and comprehensive health rating of water quality health;

[0026] The multi-dimensional quantitative output of water quality health includes chemical dimension, ecological dimension and toxicity dimension;

[0027] The comprehensive health rating of water quality health specifically includes excellent, good, medium and poor.

[0028] Furthermore, in step S3, the sewage source tracing enhancement is used to locate the pollution source and restore the pollution diffusion path. Specifically, based on the multi-dimensional reference data of water quality health and the optimized groundwater sensing data, an improved adversarial generation network method of virtual-real source tracing modeling is adopted to carry out pollution source tracing enhancement to obtain probabilistic underground water quality pollution source tracing analysis data, including the following steps:

[0029] Step S31: Virtual-real data reconstruction. Specifically, based on the multi-dimensional reference data of water quality health, the optimized groundwater sensing data corresponding to the multi-dimensional reference data of water quality health with medium and poor ratings in the comprehensive health rating of water quality health is used as the actual data input for the virtual-real data reconstruction. By constructing a groundwater hydrodynamics model, setting boundary conditions and initial pollution sources, and based on the actual data input, a virtual pollution diffusion data map is constructed to obtain virtual pollution diffusion data.

[0030] The steps for constructing the virtual pollution diffusion data map include calculating pollution intensity indicators, interpolating and generating a pollution heat map, and generating a binary pollution mask. The virtual pollution diffusion data specifically refers to the binary pollution high-risk mask map data.

[0031] Step S32: Construct a pollution path generation network. Specifically, by collecting the real observed pollution diffusion data corresponding to the virtual pollution diffusion data, a generator is constructed to obtain a pollution path generation network, and pollution path generation data is generated. The pollution path generation data includes generated predicted pollution path data and generated pollution source diffusion heat map data.

[0032] The construction of the generator specifically uses a dual-branch ResNet to separately process the data generation of the generated predicted pollution path data and the generated pollution source diffusion heat map data, and a feature fusion embedding layer is constructed for conditional feature embedding to obtain pollution path generation data.

[0033] Step S33: Construct a dual-path discriminator network. Specifically, a coarse-grained discriminator and a fine-grained discriminator are respectively constructed to construct a dual-path discriminator network to obtain a discriminator network.

[0034] The coarse-grained discriminator is used to evaluate the overall trend of the pollution path and the regional pollution probability. The fine-grained discriminator is used to analyze the path change characteristics of local pollution diffusion.

[0035] Step S34: Loss improvement design. Specifically, by constructing an adversarial loss, a path consistency loss, and a pollution map guidance loss, a loss improvement design for generative adversarial training is carried out, and through the weighted combination of the adversarial loss, the path consistency loss, and the pollution map guidance loss, a comprehensive loss function is obtained.

[0036] Step S35: Training of the sewage source tracing model. Specifically, through the virtual-real data reconstruction, the construction of the pollution path generation network, the construction of the dual-path discriminator network, and the loss improvement design, using the generative adversarial method, the sewage source tracing model is trained to obtain the sewage source tracing model Model VRG ;

[0037] Step S36: Enhanced sewage source tracing, specifically, based on the multi-dimensional reference data of water quality health and the optimized groundwater sensing data, use the sewage source tracing model Model VRG , perform enhanced sewage source tracing to obtain probabilistic groundwater quality pollution source tracing analysis data;

[0038] The probabilistic groundwater quality pollution source tracing analysis data includes pollution source heat map data, pollution path map data, and groundwater quality pollution source tracing credibility score reference data.

[0039] Furthermore, in step S4, the water quality prediction and early warning is used to achieve predictive groundwater quality early warning. Specifically, based on the probabilistic groundwater quality pollution source tracing analysis data, and by setting intelligent sub-period early warning thresholds, perform water quality prediction and early warning to obtain groundwater quality prediction and early warning reference data.

[0040] Furthermore, in step S5, the groundwater quality supervision is used to comprehensively supervise and manage the groundwater quality by combining the water quality health, the results of pollution source tracing analysis, and the results of water quality prediction and early warning. Specifically, based on the multi-dimensional reference data of water quality health, perform multi-dimensional health degree modeling of groundwater, and through visual analysis, assist in the basic management of groundwater quality, and based on the probabilistic groundwater quality pollution source tracing analysis data and the groundwater quality prediction and early warning reference data, perform comprehensive groundwater quality supervision to obtain a groundwater quality supervision reference plan.

[0041] An intelligent groundwater quality supervision system provided by the present invention includes a heterogeneous sensing acquisition module, a water quality baseline modeling module, a sewage source tracing enhancement module, a water quality prediction and early warning module, and a groundwater quality supervision module;

[0042] The heterogeneous sensing acquisition module is used for heterogeneous sensing acquisition. Through heterogeneous sensing acquisition, obtain optimized groundwater sensing data, and send the optimized groundwater sensing data to the water quality baseline modeling module and the sewage source tracing enhancement module;

[0043] The water quality baseline modeling module is used for water quality baseline modeling. Through water quality baseline modeling, obtain multi-dimensional reference data of water quality health, and send the multi-dimensional reference data of water quality health to the sewage source tracing enhancement module and the groundwater quality supervision module;

[0044] The sewage source tracing enhancement module is used for sewage source tracing enhancement. Through sewage source tracing enhancement, obtain probabilistic groundwater quality pollution source tracing analysis data, and send the probabilistic groundwater quality pollution source tracing analysis data to the water quality prediction and early warning module;

[0045] The water quality prediction and early warning module is used for water quality prediction and early warning. Through water quality prediction and early warning, reference data for groundwater quality prediction and early warning are obtained, and the reference data for groundwater quality prediction and early warning are sent to the groundwater quality supervision module;

[0046] The groundwater quality supervision module is used for groundwater quality supervision. Through groundwater quality supervision, a reference plan for groundwater quality supervision is obtained.

[0047] The beneficial effects achieved by the present invention using the above solution are as follows:

[0048] (1) Aiming at the basic problems existing in the existing groundwater quality supervision methods, such as scattered monitoring data, difficult to track pollution changes in a timely manner, and lagging early warning mechanisms, which lead to a lack of overall understanding of pollution evolution during the supervision process and difficulty in taking effective response measures in a timely manner. In the face of the actual situation of unclear pollution sources, complex groundwater flow paths, and diverse pollutants, traditional methods often rely only on fixed-point monitoring and cannot achieve comprehensive tracking and trend prediction of groundwater pollution. This solution creatively adopts a comprehensive groundwater quality supervision method that combines water quality benchmark modeling, enhanced pollution source tracing, and dynamic threshold water quality prediction and early warning, effectively improving the perception ability, identification ability, and response ability of groundwater pollution incidents, and realizing the transformation of supervision from "passive response" to "active prediction";

[0049] (2) Aiming at the technical problems existing in the existing water quality benchmark modeling methods, such as the existing methods often using static standards or historical averages as the judgment basis, ignoring the natural volatility and complex influence of human interference of groundwater quality at different regions and different time scales. This "averaging" modeling method is prone to judgment errors, either false alarms of pollution or missed reports of abnormalities. This solution creatively adopts an improved spatio-temporal graph convolutional network method combined with a spatio-temporal dynamic baseline algorithm for water quality benchmark modeling, which can more scientifically reflect which water quality changes belong to normal fluctuations and which belong to abnormal pollution, thus providing a more accurate judgment basis for subsequent source tracing and early warning;

[0050] (3) Aiming at the technical problems existing in the existing sewage source tracing enhancement methods, such as traditional source tracing methods mostly relying on the concentration inversion of pollutants and the inference of hydrogeological maps, which are both easily limited by sparse sample points and incomplete data, and difficult to cope with complex scenarios of multi-source superposition or hidden pollution, resulting in unclear source tracing results and inaccurate path judgment. This solution creatively adopts an improved generative adversarial network method for virtual and real source tracing modeling to enhance pollution source tracing, not only improving the accuracy of pollution source location, but also revealing the diffusion trend of pollutants in groundwater, enabling the supervision unit to detect hidden pollution earlier and accurately lock the pollution source, significantly enhancing the forward-looking and pertinence of pollution prevention and control. Description of the Drawings

[0051] Figure 1 It is a schematic flow diagram of an intelligent groundwater quality supervision method provided by the present invention;

[0052] Figure 2 It is a schematic diagram of an intelligent groundwater quality supervision system provided by the present invention;

[0053] Figure 3 It is a schematic flow diagram for optimizing preprocessing in step S1;

[0054] Figure 4 It is a schematic flow diagram for building a water quality benchmark in step S2;

[0055] Figure 5 It is a schematic flow diagram for enhancing sewage source tracing in step S3.

[0056] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation to the present invention. Specific Embodiments

[0057] 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.

[0058] In the description of the present invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc. indicating the orientation or positional relationship are based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus cannot be understood as a limitation to the present invention.

[0059] Embodiment 1, refer to Figure 1 , An intelligent groundwater quality supervision method provided by the present invention, the method includes the following steps:

[0060] Step S1: Heterogeneous sensing acquisition;

[0061] Step S2: Water quality benchmark modeling;

[0062] Step S3: Sewage source tracing enhancement;

[0063] Step S4: Water quality prediction and early warning;

[0064] Step S5: Groundwater quality supervision.

[0065] By performing the above operations, in the existing groundwater quality supervision methods, there are basic problems such as scattered monitoring data, difficulty in timely tracking of pollution changes, and lagging warning mechanisms, resulting in a lack of overall understanding of pollution evolution during the supervision process, making it difficult to take effective response measures in a timely manner. In the face of the actual situation of unclear pollution sources, complex groundwater flow paths, and diverse pollutants, traditional methods often rely only on fixed-point monitoring and cannot achieve comprehensive tracking and trend prediction of groundwater pollution. To solve this technical problem, this solution creatively adopts a comprehensive groundwater quality supervision method that combines water quality benchmark modeling, enhanced pollution source tracing, and dynamic threshold water quality prediction and warning, effectively improving the perception ability, identification ability, and response ability of groundwater pollution incidents, and realizing the supervision transformation from "passive response" to "active prediction".

[0066] Example 2, refer to Figure 1 、 Figure 2 and Figure 3 In step S1, the heterogeneous sensing acquisition is used to collect the original data set required for groundwater quality supervision. Specifically, through the sensing array data acquisition, the original groundwater quality data set is obtained, and through the optimization and preprocessing of the original groundwater quality data set, the optimized groundwater sensing data is obtained;

[0067] The original groundwater quality data set includes physical and chemical parameter data, ion concentration data, and sensing metadata; the physical and chemical parameter data includes pH value, conductivity, dissolved oxygen, redox potential, temperature, turbidity, total dissolved solids, and chemical oxygen demand data; the ion concentration data includes sodium ion, potassium ion, calcium ion, magnesium ion, chloride ion, sulfate ion, lead (Pb), arsenic (As), mercury (Hg), cadmium (Cd), chromium (Cr), and nutrient ion concentration data; the sensing metadata includes sampling timestamp, sampling location information, sensing number information, sensor type information, sampling depth, and external air pressure information;

[0068] The steps of the optimization and preprocessing include data cleaning, data alignment, noise filtering, data standardization, data normalization, feature optimization, and data labeling;

[0069] The feature optimization includes the construction of water quality index derivative features, ion ratio features, and abnormal fluctuation indicators; the data labeling is specifically carried out through manual judgment of pollution concentration, and through data labeling, background water samples and polluted water samples are distinguished to obtain the optimized groundwater sensing data;

[0070] The optimized groundwater sensing data includes optimized physical and chemical parameter data, optimized ion concentration data, and optimized sensing metadata.

[0071] Example 3. Refer to Figure 1 、 Figure 2 and Figure 4 . Based on the above example, in step S2, the water quality benchmark modeling is used to construct a dynamic water quality assessment benchmark model. Specifically, according to the optimized groundwater sensing data, an improved spatio-temporal graph convolutional network method combined with a spatio-temporal dynamic baseline algorithm is used for water quality benchmark modeling to obtain multi-dimensional reference data on water quality health, including the following steps:

[0072] Step S21: Multi-scale feature extraction. Specifically, a one-dimensional convolutional long short-term neural network is constructed as a hybrid encoder to extract multi-scale feature data from the optimized groundwater sensing data; the multi-scale feature data includes local features and global features; the local features are used to represent hourly sensing time data, and the global features are used to represent seasonal sensing time data;

[0073] The one-dimensional convolutional long short-term neural network is improved by introducing dilated convolution to replace the standard one-dimensional convolutional layer for time series perception enhancement to obtain multi-scale feature data. The calculation formula is:

[0074] ;

[0075] In the formula, H is the multi-scale feature data, LSTM(·) is the long short-term neural network operation function, CNN(·) is the dilated convolution operation function, and X is the optimized groundwater sensing data;

[0076] Step S22: Spatio-temporal dynamic baseline modeling. Specifically, spatio-temporal dynamic baseline modeling is carried out through spatial correlation modeling and time dynamic modeling to obtain time-geological feature data;

[0077] The spatial correlation modeling specifically constructs a groundwater spatial relationship graph structure through hydrogeological conductivity and pollutant diffusivity modeling, and calculates spatial correlation feature data through a standard graph convolutional network. The calculation formula is:

[0078] ;

[0079] In the formula, w ij is the edge weight of the spatial correlation feature data, exp(·) is the natural exponential function, K ij is the hydrogeological conductivity, i is the index of the first monitoring point, j is the index of the second monitoring point, is the average value of the hydrogeological conductivity, is the variance of the hydrogeological conductivity, sim(·) is the pollutant diffusion similarity calculation function, h i is the multi-scale feature data corresponding to the first monitoring point, h jis the multi-scale feature data corresponding to the second monitoring point;

[0080] For the time dynamics modeling, specifically, an improved spatio-temporal graph convolutional network is adopted for spatio-temporal attention extraction. The improved spatio-temporal graph convolutional network specifically introduces a gated spatio-temporal attention mechanism on the basis of the standard spatio-temporal graph convolutional network to process spatio-temporal features and obtain time dynamics modeling feature data. The calculation formula is:

[0081] ;

[0082] In the formula, is the time dynamics modeling feature data of the (l + 1)-th layer of the improved spatio-temporal graph convolutional network layer. SIG(·) is the sigmoid activation function, A is the spatial adjacency matrix, is the time dynamics modeling feature data of the l-th layer of the improved spatio-temporal graph convolutional network layer. l is the layer index of the improved spatio-temporal graph convolutional network layer, is the weight of the standard spatio-temporal graph convolutional network, is the gated attention weight, which is specifically calculated by the temporal attention generated by the one-dimensional convolutional long short-term neural network;

[0083] The time geological feature data is specifically obtained by combining the spatial correlation feature data and the time dynamics modeling feature data into a spatio-temporal dynamic baseline, and is used as a reference benchmark for water quality health;

[0084] Step S23: Water quality health calculation. Specifically, by calculating the Mahalanobis distance between the optimized groundwater sensing data and the time geological feature data, deviation data is obtained, and a multi-dimensional health fusion mechanism is constructed to calculate the water quality health and obtain the water quality health output;

[0085] The calculation formula of the multi-dimensional health fusion mechanism is:

[0086] ;

[0087] In the formula, WHI i,t is the water quality health output of the first monitoring point i at time t. k is the slope parameter of the sigmoid activation function, is the deviation data of the first monitoring point i at time t, is the deviation threshold parameter, and its specific value is 90% of the historical deviation data;

[0088] By constructing a water quality health grading standard, the water quality health is quantified and rated according to the water quality health output;

[0089] Preferably, Table 1 is the reference table for the water quality health grading standard. As shown in the table, it is output based on the water quality health degree to quantify and rate the water quality health degree;

[0090] Table 1 Reference Table for Water Quality Health Grading Standard

[0091]

[0092] Step S24: Water quality benchmark modeling training, specifically, through the multi-scale feature extraction, the spatio-temporal dynamic benchmark modeling, and the water quality health degree calculation, conduct water quality benchmark modeling training to obtain the water quality benchmark model Model WHI ;

[0093] Step S25: Water quality benchmark modeling, specifically, based on the optimized groundwater sensing data, use the water quality benchmark model Model WHI , conduct water quality benchmark modeling to obtain multi-dimensional reference data of water quality health degree;

[0094] The multi-dimensional reference data of water quality health degree includes multi-dimensional quantitative output of water quality health degree and comprehensive health rating of water quality health degree;

[0095] The multi-dimensional quantitative output of water quality health degree includes chemical dimension, ecological dimension, and toxicity dimension;

[0096] The comprehensive health rating of water quality health degree specifically includes excellent, good, medium, and poor.

[0097] By performing the above operations, in the existing water quality benchmark modeling methods, there are technical problems that existing methods often use static standards or historical averages as the judgment basis, ignoring the natural volatility and complex influence of human interference of groundwater quality at different regions and different time scales. This "averaging" modeling method is prone to misjudgment, either false reporting of pollution or missed reporting of anomalies. The present solution creatively uses an improved spatio-temporal graph convolutional network method combined with a spatio-temporal dynamic baseline algorithm to conduct water quality benchmark modeling, which more scientifically reflects which water quality changes belong to normal fluctuations and which belong to abnormal pollution, thereby providing a more accurate judgment basis for subsequent traceability and early warning.

[0098] Example 4, refer to Figure 1 、 Figure 2 and Figure 5 Based on the above example, in step S3, the sewage traceability enhancement is used to locate the pollution source and restore the pollution diffusion path. Specifically, based on the multi-dimensional reference data of water quality health degree and the optimized groundwater sensing data, an improved adversarial generative network method of virtual-real traceability modeling is used to conduct pollution traceability enhancement to obtain probabilistic underground water quality pollution traceability analysis data, including the following steps:

[0099] Step S31: Reconstruction of virtual and real data. Specifically, based on the multi-dimensional reference data of water quality health, the optimized groundwater sensing data corresponding to the multi-dimensional reference data of water quality health with medium and poor ratings in the comprehensive health rating of water quality health is used as the actual data input for the reconstruction of virtual and real data. By constructing a groundwater hydrodynamics model, setting boundary conditions and initial pollution sources, and based on the actual data input, a virtual pollution diffusion data map is constructed to obtain virtual pollution diffusion data;

[0100] The steps for constructing the virtual pollution diffusion data map include calculating pollution intensity indicators, interpolating and generating a pollution heat map, and generating a binary pollution mask. The virtual pollution diffusion data specifically refers to the binary pollution high-risk mask map data;

[0101] The calculation formula for generating the binary pollution mask is:

[0102] ;

[0103] In the formula, M(x, y) is the binary pollution high-risk mask map data, where x is the pixel index in the horizontal direction and y is the pixel index in the vertical direction. PI(·) is the pollution intensity index function, which is specifically obtained through the calculation of the pollution intensity indicators. is the pollution intensity threshold;

[0104] Preferably, the calculation formula for calculating the pollution intensity indicators is:

[0105] ;

[0106] In the formula, PI(·) is the pollution intensity index function. is the pollution index sample index, C is the total number of pollutants, c is the pollutant index, w c is the pollution factor weight. is the pollution index value of the i-th pollution index sample corresponding to the c-th pollutant, and S c is the water quality standard limit;

[0107] The pollution intensity threshold is set to 95%;

[0108] Step S32: Construct a pollution path generation network. Specifically, by collecting the real observed pollution diffusion data corresponding to the virtual pollution diffusion data, a generator is constructed to obtain a pollution path generation network, and pollution path generation data is generated. The pollution path generation data includes generated predicted pollution path data and generated heat maps of pollution source diffusion;

[0109] The construction of the generator specifically uses a dual-branch ResNet to separately process the data generation of generating predicted pollution path data and generating heat diffusion maps of pollution sources, and constructs a feature fusion embedding layer for conditional feature embedding to obtain pollution path generation data;

[0110] Step S33: Construct a dual-path discriminator network, specifically by separately constructing a coarse-grained discriminator and a fine-grained discriminator to construct a dual-path discriminator network and obtain a discriminator network;

[0111] The coarse-grained discriminator is used to evaluate the overall trend of the pollution path and the regional pollution probability; the fine-grained discriminator is used to analyze the path change characteristics of local pollution diffusion;

[0112] Step S34: Loss improvement design, specifically by constructing adversarial loss, path consistency loss, and pollution map guidance loss to perform loss improvement design for generative adversarial training, and obtaining a comprehensive loss function through the weighted combination of the adversarial loss, path consistency loss, and pollution map guidance loss;

[0113] The adversarial loss uses the standard generative adversarial loss function as the adversarial loss;

[0114] The path consistency loss is specifically calculated by taking the L2 norm difference between the generated predicted pollution path data and the actual pollution path as the path consistency loss;

[0115] The pollution map guidance loss is specifically calculated by combining the generated heat diffusion map data of pollution sources with the binary pollution high-risk mask map data for guidance calculation, which is used to make the generation model focus on the core area of pollution as the pollution map guidance loss;

[0116] The calculation formula of the comprehensive loss function is:

[0117] ;

[0118] In the formula, is the comprehensive loss function, is the adversarial loss weight, is the standard generative adversarial loss function, is the path consistency loss weight, is the path consistency loss function, is the pollution map guidance loss weight, is the pollution map guidance loss function;

[0119] Preferably, the value of the adversarial loss weight is 0.1, the value of the path consistency loss weight is 0.4, and the value of the pollution map guidance loss weight The value of is 0.5;

[0120] Step S35: Training the sewage source tracing model, specifically, through the virtual-real data reconstruction, the construction of the pollution path generation network, the construction of the dual-path discriminator network, and the loss improvement design, using the generative adversarial method to train the sewage source tracing model to obtain the sewage source tracing model Model VRG ;

[0121] Step S36: Strengthening the sewage source tracing, specifically, based on the multi-dimensional reference data of water quality health and the optimized data of groundwater sensing, using the sewage source tracing model Model VRG , to strengthen the sewage source tracing and obtain the probabilistic groundwater quality pollution source tracing analysis data;

[0122] The probabilistic groundwater quality pollution source tracing analysis data includes pollution source heat map data, pollution path map data, and groundwater quality pollution source tracing credibility score reference data.

[0123] By performing the above operations, in the existing sewage source tracing enhancement methods, there are technical problems that traditional tracing methods mostly rely on the concentration inversion of pollutants and the inference of hydrogeological maps, which are both easily limited by sparse sample points and incomplete data, and are also difficult to handle complex scenarios of multi-source superposition or hidden pollution, resulting in unclear tracing results and inaccurate path judgment. This solution creatively uses an improved adversarial generation network method for virtual-real source tracing modeling to enhance pollution source tracing. It not only improves the accuracy of pollution source location but also reveals the diffusion trend of pollutants in groundwater, enabling regulatory units to detect hidden pollution earlier and accurately lock the pollution source, significantly enhancing the forward-looking and pertinence of pollution prevention and control.

[0124] Example 5, refer to Figure 1 and Figure 2 , in step S4, the water quality prediction and early warning are used to achieve predictive groundwater quality early warning, specifically, based on the probabilistic groundwater quality pollution source tracing analysis data, and by setting intelligent sub-period early warning thresholds, to perform water quality prediction and early warning to obtain groundwater quality prediction and early warning reference data;

[0125] The intelligent sub-period early warning thresholds specifically include a short-term period (1 to 7 days), a medium-term period (7 to 30 days), and a long-term period (more than 30 days);

[0126] By constructing a dynamic adjustment threshold fitting mechanism, to perform early warning classification based on the risk level to obtain groundwater quality prediction and early warning reference data;

[0127] The calculation formula of the dynamic adjustment threshold fitting mechanism is:

[0128] ;

[0129] In the formula, is the periodic dynamic warning threshold, is the mean value of the water quality factor, and the water quality factor is used to represent the reference data of the groundwater quality pollution traceability credibility score in the probabilistic groundwater quality pollution traceability analysis data. p is the water quality factor index, is the pollution sensitivity coefficient, is the standard deviation of the water quality factor;

[0130] The warning levels include low risk, medium risk, and high risk;

[0131] Preferably, Table 2 is the reference table for the warning determination conditions of the warning levels. As shown in the table, x t is the credibility score of the groundwater quality pollution traceability, which is used to represent the pollution prediction value at time t. The warning levels are used for water quality prediction and warning.

[0132] Table 2 Reference Table for Warning Determination Conditions of Warning Levels

[0133]

[0134] Example 6, refer to Figure 1 and Figure 2 Based on the above example, in step S5, the groundwater quality supervision is used to comprehensively supervise and manage the groundwater quality by combining the water quality health degree, the pollution traceability analysis result, and the result of the water quality prediction and warning. Specifically, according to the multi-dimensional reference data of the water quality health degree, a multi-dimensional health degree model of the groundwater is established, and through visual analysis, the basic management of the groundwater quality is assisted. And according to the probabilistic groundwater quality pollution traceability analysis data and the groundwater quality prediction and warning reference data, the comprehensive groundwater quality supervision is carried out to obtain a reference plan for groundwater quality supervision.

[0135] Example 7, refer to Figure 1 and Figure 2 Based on the above example, an intelligent groundwater quality supervision system provided by the present invention includes a heterogeneous sensing acquisition module, a water quality benchmark modeling module, a sewage traceability enhancement module, a water quality prediction and warning module, and a groundwater quality supervision module;

[0136] The heterogeneous sensing acquisition module is used for heterogeneous sensing acquisition. Through heterogeneous sensing acquisition, the optimized groundwater sensing data is obtained, and the optimized groundwater sensing data is sent to the water quality benchmark modeling module and the sewage traceability enhancement module;

[0137] The water quality criteria modeling module is used for water quality criteria modeling. Through water quality criteria modeling, multi-dimensional reference data on water quality health is obtained, and the multi-dimensional reference data on water quality health is sent to the sewage source tracing enhancement module and the groundwater quality supervision module;

[0138] The sewage source tracing enhancement module is used for sewage source tracing enhancement. Through sewage source tracing enhancement, probabilistic analysis data on underground water quality pollution source tracing is obtained, and the probabilistic analysis data on underground water quality pollution source tracing is sent to the water quality prediction and early warning module;

[0139] The water quality prediction and early warning module is used for water quality prediction and early warning. Through water quality prediction and early warning, reference data on groundwater quality prediction and early warning is obtained, and the reference data on groundwater quality prediction and early warning is sent to the groundwater quality supervision module;

[0140] The groundwater quality supervision module is used for groundwater quality supervision. Through groundwater quality supervision, a reference plan for groundwater quality supervision is obtained.

[0141] 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 comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or device.

[0142] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention.

[0143] The above describes the present invention and its embodiments. Such description is not restrictive. What is shown in the drawings is only one of the embodiments of the present invention, and the actual structure is not limited thereto. Generally speaking, if those of ordinary skill in the art are inspired by it and design similar structural modes and embodiments without creative efforts without departing from the purpose of the present invention, they shall fall within the protection scope of the present invention.

Claims

1. An intelligent groundwater quality supervision method, characterized in that: The method includes the following steps: Step S1: Heterogeneous sensing acquisition to obtain optimized groundwater sensing data; Step S2: Water quality benchmark modeling. An improved spatio-temporal graph convolutional network method combined with a spatio-temporal dynamic baseline algorithm is used for water quality benchmark modeling to obtain multi-dimensional reference data on water quality health, including the following steps: Step S21: Multi-scale feature extraction. By introducing dilated convolution to replace the standard one-dimensional convolutional layer, temporal perception enhancement improvement is carried out; Step S22: Spatio-temporal dynamic baseline modeling; Step S23: Water quality health calculation. By calculating the Mahalanobis distance in the optimized groundwater sensing data and time-geological feature data, and constructing a multi-dimensional health fusion mechanism, the water quality health output is obtained; Step S24: Training of the water quality benchmark modeling; Step S25: Water quality benchmark modeling; Step S3: Sewage source tracing enhancement. An improved generative adversarial network method for virtual and real source tracing modeling is used for pollution source tracing enhancement to obtain probabilistic groundwater quality pollution source tracing analysis data, including the following steps: Step S31: Reconstruction of virtual and real data. By constructing a groundwater hydrodynamics model and setting boundary conditions and initial pollution sources, a virtual pollution diffusion data map is constructed to obtain virtual pollution diffusion data; Step S32: Construction of a pollution path generation network; Step S33: Construction of a dual-path discriminator network; Step S34: Loss improvement design; Step S35: Training of the sewage source tracing model; Step S36: Sewage source tracing enhancement; Step S4: Water quality prediction and early warning. By setting intelligent sub-period early warning thresholds, water quality prediction and early warning are carried out to obtain reference data for groundwater quality prediction and early warning; Step S5: Groundwater quality supervision to obtain a reference plan for groundwater quality supervision.

2. The intelligent groundwater quality supervision method according to claim 1, wherein: In step S1, the heterogeneous sensing acquisition is used to collect the original data set required for groundwater quality supervision. Specifically, the original data set of groundwater quality is obtained through sensing array data acquisition, and the optimized groundwater sensing data is obtained through optimized preprocessing of the original data set of groundwater quality; The original data set of groundwater quality includes physical and chemical parameter data, ion concentration data, and sensor metadata; the steps of the optimized preprocessing include data cleaning, data alignment, noise filtering, data standardization, data normalization, feature optimization, and data labeling; the optimized groundwater sensing data includes optimized physical and chemical parameter data, optimized ion concentration data, and optimized sensor metadata.

3. The intelligent groundwater quality supervision method according to claim 2, characterized in that: In step S2, the water quality benchmark modeling is used to construct a dynamic water quality assessment benchmark model. Specifically, based on the optimized groundwater sensing data, an improved spatio-temporal graph convolutional network method combined with a spatio-temporal dynamic baseline algorithm is used for water quality benchmark modeling to obtain multi-dimensional reference data on water quality health, including the following steps: Step S21: Multi-scale feature extraction. Specifically, a one-dimensional convolutional long short-term neural network is constructed as a hybrid encoder to extract multi-scale feature data from the optimized groundwater sensing data; the multi-scale feature data includes local features and global features; the local features are used to represent hourly-level sensing time data, and the global features are used to represent seasonal-level sensing time data; The one-dimensional convolutional long short-term neural network is improved in temporal perception enhancement by introducing dilated convolution to replace the standard one-dimensional convolutional layer; Step S22: Spatiotemporal dynamic baseline modeling, specifically, through spatial correlation modeling and temporal dynamics modeling, spatiotemporal dynamic baseline modeling is carried out to obtain temporal geological feature data; The spatial correlation modeling specifically constructs the structure of the groundwater spatial relationship map through the modeling of the hydrogeological conductivity coefficient and the pollutant diffusion degree, and calculates the spatial correlation feature data through the standard graph convolutional network; The temporal dynamics modeling specifically extracts spatiotemporal attention by using an improved spatiotemporal graph convolutional network. The improved spatiotemporal graph convolutional network specifically processes spatiotemporal features by additionally introducing a gated spatiotemporal attention mechanism on the basis of the standard spatiotemporal graph convolutional network to obtain temporal dynamics modeling feature data; The temporal geological feature data specifically combines the spatial correlation feature data and the temporal dynamics modeling feature data and outputs them as a spatiotemporal dynamic baseline to obtain the temporal geological feature data, which is used as a reference baseline for water quality health; Step S23: Water quality health calculation, specifically, by calculating the Mahalanobis distance between the optimized groundwater sensing data and the temporal geological feature data, deviation data is obtained, and a multi-dimensional health fusion mechanism is constructed to calculate the water quality health to obtain the water quality health output; Step S24: Water quality criteria modeling training, specifically, through the multi-scale feature extraction, the spatio-temporal dynamic criteria modeling, and the water quality health calculation, conduct water quality criteria modeling training to obtain a water quality criteria model Model WHI ; Step S25: Water quality criteria modeling, specifically, based on the optimized groundwater sensing data, use the water quality criteria model Model WHI , perform water quality criteria modeling to obtain multi-dimensional reference data on water quality health.

4. The intelligent groundwater quality supervision method according to claim 3, characterized in that: In step S2, the multi-dimensional reference data of water quality health includes the multi-dimensional quantitative output of water quality health and the comprehensive health rating of water quality health; The multi-dimensional quantitative output of water quality health includes chemical dimension, ecological dimension and toxicity dimension; The comprehensive health rating of water quality health specifically includes excellent, good, medium and poor ratings.

5. The intelligent groundwater quality supervision method according to claim 4, wherein: In step S3, the sewage source tracing enhancement is used to locate the pollution source and restore the pollution diffusion path. Specifically, based on the multi-dimensional reference data of water quality health and the optimized groundwater sensing data, an improved adversarial generation network method of virtual-real source tracing modeling is used to perform pollution source tracing enhancement to obtain probabilistic underground water quality pollution source tracing analysis data, including the following steps: Step S31: Virtual-real data reconstruction, specifically, based on the multi-dimensional reference data of water quality health, the optimized groundwater sensing data corresponding to the multi-dimensional reference data of water quality health with medium and poor ratings in the comprehensive health rating of water quality health is used as the actual data input for the virtual-real data reconstruction, and a groundwater hydrodynamics model is constructed. By setting boundary conditions and initial pollution sources, based on the actual data input, a virtual pollution diffusion data map is constructed to obtain virtual pollution diffusion data; The steps of constructing the virtual pollution diffusion data map include pollution intensity index calculation, pollution heat map interpolation generation and pollution mask binarization generation; the virtual pollution diffusion data specifically refers to the binary pollution high-risk mask map data; Step S32: Construct a pollution path generation network. Specifically, by collecting the real observed pollution diffusion data corresponding to the virtual pollution diffusion data, construct a generator to obtain a pollution path generation network, and generate pollution path generation data, where the pollution path generation data includes generated predicted pollution path data and generated heat maps of pollution source diffusion data; Step S33: Construct a dual-path discriminator network. Specifically, construct a coarse-grained discriminator and a fine-grained discriminator respectively, and construct a dual-path discriminator network to obtain a discriminator network; The coarse-grained discriminator is used to evaluate the overall trend of the pollution path and the regional pollution probability; the fine-grained discriminator is used to analyze the path change characteristics of local pollution diffusion; Step S34: Loss improvement design. Specifically, by constructing adversarial loss, path consistency loss, and pollution map guidance loss, conduct loss improvement design for generative adversarial training, and obtain a comprehensive loss function through the weighted combination of the adversarial loss, path consistency loss, and pollution map guidance loss; Step S35: Training the sewage source tracing model, specifically, through the virtual-real data reconstruction, the construction of the pollution path generation network, the construction of the dual-path discriminator network, and the loss improvement design, using the generative adversarial method to train the sewage source tracing model and obtain the sewage source tracing model Model VRG ; Step S36: Enhanced sewage source tracing, specifically, based on the multi-dimensional reference data of water quality health and the optimized groundwater sensing data, using the sewage source tracing model Model VRG , perform enhanced sewage source tracing to obtain probabilistic groundwater pollution source tracing analysis data.

6. The intelligent groundwater quality supervision method according to claim 5, characterized in that: In step S3, the probabilistic groundwater pollution source tracing analysis data includes heat map data of pollution sources, pollution path map data, and reference data for the credibility score of groundwater pollution source tracing.

7. The intelligent groundwater quality supervision method according to claim 6, characterized in that: In step S4, the water quality prediction and early warning is used to achieve predictive groundwater water quality early warning. Specifically, based on the probabilistic groundwater pollution source tracing analysis data, and by setting an intelligent sub-period early warning threshold, conduct water quality prediction and early warning to obtain reference data for groundwater water quality prediction and early warning; In step S5, the groundwater water quality supervision is used to conduct comprehensive groundwater water quality supervision and management by combining the water quality health degree, the results of pollution source tracing analysis, and the results of water quality prediction and early warning. Specifically, based on the multi-dimensional reference data of water quality health degree, conduct multi-dimensional health degree modeling of groundwater, and through visual analysis, assist in the basic management of groundwater water quality, and based on the probabilistic groundwater pollution source tracing analysis data and the reference data for groundwater water quality prediction and early warning, conduct comprehensive groundwater water quality supervision to obtain a reference plan for groundwater water quality supervision.

8. An intelligent groundwater quality supervision system for implementing an intelligent groundwater quality supervision method as described in any one of claims 1-7, characterized in that: It includes a heterogeneous sensing acquisition module, a water quality benchmark modeling module, a sewage source tracing enhancement module, a water quality prediction and early warning module, and a groundwater water quality supervision module; The heterogeneous sensing acquisition module is used for heterogeneous sensing acquisition. Through heterogeneous sensing acquisition, obtain optimized groundwater sensing data, and send the optimized groundwater sensing data to the water quality benchmark modeling module and the sewage source tracing enhancement module; The water quality benchmark modeling module is used for water quality benchmark modeling. Through water quality benchmark modeling, obtain multi-dimensional reference data of water quality health degree, and send the multi-dimensional reference data of water quality health degree to the sewage source tracing enhancement module and the groundwater water quality supervision module; The sewage source tracing enhancement module is used for sewage source tracing enhancement. Through sewage source tracing enhancement, obtain probabilistic groundwater pollution source tracing analysis data, and send the probabilistic groundwater pollution source tracing analysis data to the water quality prediction and early warning module; The water quality prediction and early warning module is used for water quality prediction and early warning. Through water quality prediction and early warning, reference data for groundwater quality prediction and early warning is obtained, and the reference data for groundwater quality prediction and early warning is sent to the groundwater quality supervision module; The groundwater quality supervision module is used for groundwater quality supervision. Through groundwater quality supervision, a reference plan for groundwater quality supervision is obtained.

Images