River water quality parameter supervision method and system based on deep learning

Abstract

The invention provides a river water quality parameter supervision method and system based on deep learning. The method comprises the steps of self-calibration multi-source data acquisition, diffusive water quality prediction, extreme water quality parameter simulation, reverse diffusion pollution positioning and water quality parameter intelligent supervision. The invention relates to the technical field of river water quality supervision, in particular to a river water quality parameter supervision method and system based on deep learning. By introducing a graph convolutional neural network and a physical diffusion constraint model, space-time diffusion trend modeling of pollutants in a river channel is realized; constructing an extreme pollution event simulation and attribution mechanism by combining a generative adversarial network and physical verification; further adopting a multi-modal Bayesian inversion model and a graph deconvolution structure to realize accurate source tracing of the pollution source; the system can dynamically sense hydrological changes, construct an adaptive threshold judgment mechanism, realize prediction, tracking and response to pollution risks, and provide efficient and intelligent technical support for river ecological safety management.

Description

Technical Field

[0001] The present invention relates to the technical field of river water quality supervision, and specifically refers to a method and system for river water quality parameter supervision based on deep learning. Background Art

[0002] The method and system for river water quality parameter supervision based on deep learning is a comprehensive supervision means that uses a deep neural network model to intelligently analyze and predict multi-source river water quality monitoring data, combines physical diffusion laws, extreme event simulation, and pollution source tracing technology to achieve dynamic perception, trend prediction, anomaly identification, and source tracing judgment of river water quality parameters. Its core role is to improve the accuracy and timeliness of water quality anomaly early warning, assist environmental management departments in scientifically formulating pollution prevention and control strategies, and realize the intelligent and systematic management of river ecological environments.

[0003] However, in the existing intelligent river water quality parameter supervision methods, there are technical problems that most of the existing methods stay at the level of "monitoring - alarm", fail to achieve early warning of pollution trends, and the supervision response often triggers based on a single indicator, ignoring the pollution diffusion chain and source factors, resulting in the disconnection of the pollution source tracing, diffusion prediction, and risk assessment processes, with a high model response delay and unable to achieve joint decision-making under complex hydrological conditions; in the existing water quality prediction methods, there is a problem of insufficient modeling of the river hydrodynamic diffusion process. During river rainstorm flood periods, due to the sharp increase in flow velocity, the diffusion distance of pollutants increases, but the model still limits the prediction results to the usual diffusion radius because it does not consider the change in flow velocity, resulting in a serious underestimation of the warning area, which further leads to the technical problems of separate modeling of spatial structure and temporal dynamics dependence, and the prediction results not conforming to physical laws; in the existing water quality parameter simulation methods, there is a technical problem of lacking the coverage ability for extreme event conditions, and the generated results cannot meet the physical consistency constraints, making it difficult to provide reliable simulation data for abnormal pollution risk drills and supervision plan deductions; in the existing pollution location methods, there are technical problems of static tracing or simple reverse simulation, lacking support for multi-source pollution data fusion and uncertainty modeling, and the accuracy of pollution path restoration being limited by data quality and kinetic noise. Summary of the Invention

[0004] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides a river water quality parameter supervision method and system based on deep learning. In view of the fact that most of the existing intelligent river water quality parameter supervision methods remain at the "monitoring-alarm" level, failing to achieve early warning of pollution trends, and the supervision response is often triggered based on a single indicator, ignoring the pollution diffusion chain and source factors, resulting in pollution tracing, diffusion prediction and risk assessment process separation, high model response delay, and inability to achieve linkage decision-making under complex hydrological conditions. Technical problems, this solution creatively adopts a comprehensive supervision method of integrated prediction, extreme simulation and reverse tracing, and realizes technical problems such as sudden floods, drastic changes in flow rate, etc. It realizes adaptive adjustment of thresholds driven by sensitive factors, so as to quickly identify and generate actionable governance instructions in the early stage of pollution, greatly improving the real-time and scientific nature of basin-level water quality warning and intervention; in view of the problem that the existing water quality prediction methods do not adequately model the hydrodynamic diffusion process of rivers, during the period of heavy rain and flooding in rivers, the flow velocity increases sharply and the diffusion distance of pollutants increases, but the model does not consider the change in flow velocity, and the prediction results are still limited to the usual diffusion radius, resulting in a serious underestimation of the warning area, which further leads to the technical problems of separate modeling of spatial structure and temporal dynamic dependence and the inconsistency of prediction results with physical laws. This scheme creatively adopts the space-time graph convolutional network improved by physical diffusion constraints to perform diffusive water quality prediction. The method not only ensures that the prediction results are physically reasonable, but also improves the generalization ability of the model in the case of data missing or sparse sampling, and significantly enhances the stability of pollution trend evolution modeling; in view of the technical problems that the existing water quality parameter simulation methods lack the ability to cover extreme event conditions, the generated results cannot meet the physical consistency constraints, and it is difficult to provide reliable simulation data for abnormal pollution risk drills and regulatory program deductions, this solution creatively adopts the conditional adversarial generation method combined with physical verification simulation to simulate extreme water quality parameters, realize the causal chain reconstruction of simulated pollution sources and the reasonable inversion of location intensity, thereby effectively supporting advanced regulatory tasks such as simulation inspection, pollution drills and sensitivity analysis; for the existing Some pollution location methods involve static tracing or simple reverse simulation, lack support for multi-source pollution data fusion and uncertainty modeling, and the accuracy of pollution path restoration is limited by technical issues such as data quality and dynamic noise. This solution creatively adopts the standard multimodal Bayesian reverse tracing method to perform reverse diffusion pollution location, achieving high-confidence inversion location of pollution sources under the premise of observing pollutant concentrations. At the same time, in conjunction with the constructed graph deconvolution neural network, it enhances the ability to restore the reverse path of pollution in time and space, and ultimately outputs a pollution source probability heat map. By superimposing simulated event probability reference data, it assists regulators in accurately locking in the responsible entity for emissions, thereby improving the targeted nature of water environment law enforcement and the scientific nature of traceability.

[0005] The technical solution adopted by the present invention is as follows: The present invention provides a method for supervising river water quality parameters based on deep learning, and this method includes the following steps:

[0006] Step S1: Self-calibrated multi-source data collection;

[0007] Step S2: Diffusive water quality prediction;

[0008] Step S3: Simulation of extreme water quality parameters;

[0009] Step S4: Inverse diffusion pollution location;

[0010] Step S5: Supervision of river water quality parameters.

[0011] Furthermore, in step S1, the self-calibrated multi-source data collection is used to deploy sensing nodes and collect multi-source water quality data. Specifically, a multi-source spatio-temporal dynamic calibration fusion sensing method is adopted, and data self-calibration is performed through raw data collection to obtain spatio-temporal calibration data, including the following steps:

[0012] Step S11: Deployment of a multi-source sensing network. Specifically, key collection nodes are selected along the river channel, a sensor sensing network is deployed to collect data, and raw river sensing parameter data is obtained;

[0013] The key collection nodes include upstream nodes, midstream nodes, downstream nodes, tributary confluence nodes, and outfall nodes; the sensor sensing network includes basic water quality sensors, pollutant sensors, and environmental sensing sensors; the raw river sensing parameter data includes basic water quality parameters, pollutant parameters, and environmental sensing auxiliary parameters;

[0014] Step S12: Dynamic calibration. Specifically, a standard reference segment matching method based on a sliding time window is adopted to perform data self-calibration to obtain spatio-temporal calibration data.

[0015] Furthermore, in step S2, the diffusive water quality prediction is used to predict the dynamic diffusion trend of pollutants in the river channel. Specifically, based on the spatio-temporal calibration data, a spatio-temporal graph convolutional network improved by physical diffusion constraints is adopted to perform diffusive water quality prediction to obtain diffusion prediction data, including the following steps:

[0016] Step S21: Construction of a watershed topological graph. Specifically, based on the spatio-temporal calibration data, the river channel is formalized and modeled as a graph structure, including graph node definition and directed edge definition, and by introducing a virtual river confluence point, normalization of river tributaries is performed to obtain watershed topological graph data;

[0017] The definition of the graph nodes specifically takes each key collection node as a monitoring point and represents the monitoring point with a graph node; the definition of the directed edges specifically defines the directed edges according to the flow direction of the river and sets the edge weight of the directed edges as the reciprocal of the channel hydraulic distance;

[0018] Step S22: Embedding of diffusivity physical coefficients. Specifically, by introducing pollutant concentration, diffusivity coefficient, and water body velocity parameters, a continuous physical equation is constructed for integrated physical constraint modeling. By taking the diffusivity coefficient and water body velocity parameters in the continuous physical equation as learnable parameters, discrete differentiable parameter training for physical coefficient embedding is carried out, and a physical residual regularization loss function is introduced to optimize the model training process for diffusivity water quality prediction, obtaining an embedded physical operator; the embedded physical operator specifically includes a continuous physical equation and a physical residual regularization loss function;

[0019] Step S23: Spatiotemporal prediction modeling. Specifically, a graph convolutional neural network that mixes spatiotemporal features is constructed as the basic prediction subnet for diffusivity water quality prediction, and model training is carried out based on the embedded physical operator to obtain a spatiotemporal prediction model;

[0020] The graph convolutional neural network that mixes spatiotemporal features includes a spatial feature layer and a temporal feature layer; the spatial feature layer specifically uses a standard graph convolutional structure for spatial diffusion prediction between river channel monitoring points; the temporal feature layer specifically uses a causal dilated convolutional structure for temporal evolution feature prediction between river channel detection points;

[0021] Step S24: Diffusivity water quality prediction. Specifically, through the construction of the basin topology graph, the embedding of diffusivity physical coefficients, and the spatiotemporal prediction modeling, based on the spatiotemporal calibration data, the spatiotemporal prediction model is used for diffusivity water quality prediction to obtain diffusion prediction data.

[0022] Furthermore, in step S3, the simulation of extreme water quality parameters is used to enhance the system's ability to identify sudden pollution. Specifically, a conditional adversarial generation method combined with physical verification simulation is adopted for extreme water quality parameter simulation to obtain extreme parameter simulation data, including the following steps:

[0023] Step S31: Construction of a generative adversarial network. Specifically, by successively constructing a standard generator structure and a discriminator structure and introducing conditional control labels, extreme polluted water quality parameters are generated to obtain a basic network for generating extreme water quality parameters;

[0024] The conditional control labels include extreme event types, spatiotemporal positions, and meteorological environmental conditions;

[0025] The extreme event types include industrial emissions, agricultural activities, urban activities, meteorological disasters, and human-caused accidents;

[0026] The spatio-temporal position includes the river channel monitoring point number, longitude, latitude, and extreme event time;

[0027] The meteorological environment conditions include rainfall intensity, water body temperature, ambient temperature, wind speed, and humidity parameters;

[0028] Step S32: Physical verification simulation. Specifically, the physical residual regularization loss function is used as the physical verification loss and embedded into the discriminator structure to physically correct the extreme water quality parameters generated by the basic network for the extreme water quality parameters, obtaining physically verified and optimized simulation parameters;

[0029] Step S33: Parameter anomaly attribution optimization. Specifically, based on the physically verified and optimized simulation parameters, by constructing a simulated extreme water quality parameter space and constructing a reverse attribution parameter optimization network, a parameter attribution vector is generated to obtain extreme water quality parameter attribution reference data, and based on the extreme water quality parameter attribution reference data, manual inspection and simulation verification of the extreme water quality parameters are carried out;

[0030] The simulated extreme water quality parameter space includes the extreme pollution source location, pollution occurrence time, emission intensity, and flow velocity perturbation parameters;

[0031] The reverse attribution parameter optimization network specifically adopts an encoder-decoder structure, uses a standard time series graph attention network as the encoder structure, and uses a multi-head regression output layer as the decoder structure;

[0032] The extreme water quality parameter attribution reference data includes pollution type, pollution source location, pollution occurrence time, predicted pollution emission intensity, predicted pollution flow velocity, and true pollution comparison value;

[0033] Step S34: Extreme water quality parameter simulation. Specifically, through the construction of the generative adversarial network, the physical verification simulation, and the parameter anomaly attribution optimization, extreme water quality parameter simulation is carried out to obtain extreme parameter simulation data;

[0034] The extreme parameter simulation data specifically includes a simulated pollution concentration parameter sequence, meteorological parameters, human control parameters, and a simulated physical verification score.

[0035] Further, in step S4, the inverse diffusion pollution localization is used to trace the pollution source location. Specifically, based on the extreme parameter simulation data and the diffusion prediction data, a standard multi-modal Bayesian reverse tracing method is used for inverse diffusion pollution localization to obtain pollution tracing data, including the following steps:

[0036] Step S41: Inverse diffusion modeling, specifically constructing a mathematical model for Bayesian inversion problem, taking the predicted diffusion concentration in the diffusion prediction data as input data, constructing an inverse diffusion model and performing inverse calculation of pollution source parameters to obtain inverse diffusion parameters of the pollution source;

[0037] The inverse diffusion model includes a forward diffusion sub-model and an inverse inversion model; the forward diffusion sub-model specifically uses the continuous physical equation for mathematical modeling; the inverse inversion model specifically uses the standard Bayesian inversion expression for mathematical modeling and is optimized by sampling using the standard Markov chain Monte Carlo method;

[0038] Step S42: Spatiotemporal deconvolution improvement, specifically constructing a graph deconvolution neural network for mixing spatiotemporal features based on the spatiotemporal prediction model as the basic prediction subnet for inverse diffusion modeling to obtain a spatiotemporal deconvolution inverse diffusion prediction model;

[0039] The graph deconvolution neural network for mixing spatiotemporal features includes a spatial inverse path layer and a temporal inverse path layer; the spatial inverse path layer specifically uses the standard graph deconvolution structure to predict the inverse pollution path features between river monitoring points; the temporal inverse path layer specifically uses the reverse dilated convolution structure to predict the reverse feature dependence of the pollution time series between river detection points;

[0040] Step S43: Visualization of pollution probability, specifically using the spatiotemporal deconvolution inverse diffusion prediction model based on the inverse diffusion parameters of the pollution source to predict the spatiotemporal data of the inverse diffusion pollution source, constructing a spatiotemporal distribution probability heat map of the pollution source to obtain visualization reference data for pollution probability;

[0041] Step S44: Inverse diffusion pollution location, specifically performing inverse diffusion pollution location based on the visualization reference data of pollution probability and the pollution source tracing prediction data output by the spatiotemporal deconvolution inverse diffusion prediction model to obtain pollution tracing data; the pollution tracing data specifically includes pollution source position parameters, pollution time parameters, emission intensity parameters, visualization reference data for pollution probability, and reference data for the probability of extreme pollution events.

[0042] Furthermore, in step S5, the supervision of river water quality parameters is used to integrate prediction, extreme simulation, and inverse tracing results and achieve monitoring and management. Specifically, a dynamic threshold intelligent supervision decision-making method is adopted to supervise the river water quality parameters to obtain reference data for river water quality supervision decision-making;

[0043] The dynamic threshold intelligent supervision decision-making method specifically calculates the dynamic water quality supervision threshold by introducing environmental characteristic factors to construct environmental sensitivity factors based on the pollution source tracing data and the diffusion prediction data, and classifies and supervises water quality abnormal events according to the dynamic water quality supervision threshold.

[0044] The reference data for river water quality supervision decision-making specifically includes pollution source tracing instructions, sewage discharge limit instructions, verification monitoring instructions, and a reference ranking of the priorities for treating polluted water quality.

[0045] The river water quality parameter supervision system based on deep learning provided by the present invention includes a self-calibrating multi-source data acquisition module, a diffusive water quality prediction module, an extreme water quality parameter simulation module, an inverse diffusion pollution location module, and a river water quality parameter supervision module.

[0046] The self-calibrating multi-source data acquisition module is used for self-calibrating multi-source data acquisition. Through self-calibrating multi-source data acquisition, spatio-temporal calibration data is obtained and sent to the diffusive water quality prediction module.

[0047] The diffusive water quality prediction module is used for diffusive water quality prediction. Through diffusive water quality prediction, diffusion prediction data is obtained and sent to the inverse diffusion pollution location module and the river water quality parameter supervision module.

[0048] The extreme water quality parameter simulation module is used for extreme water quality parameter simulation. Through extreme water quality parameter simulation, extreme parameter simulation data is obtained and sent to the river water quality parameter supervision module.

[0049] The inverse diffusion pollution location module is used for inverse diffusion pollution location. Through inverse diffusion pollution location, pollution source tracing data is obtained and sent to the river water quality parameter supervision module.

[0050] The river water quality parameter supervision module is used for river water quality parameter supervision. Through river water quality parameter supervision, reference data for river water quality supervision decision-making is obtained.

[0051] The beneficial effects achieved by the present invention by adopting the above solution are as follows:

[0052] (1)In view of the technical problems existing in the existing intelligent river water quality parameter supervision methods, where most of the existing methods stay at the level of "monitoring - alarm", fail to achieve early warning of pollution trends, and the supervision response often triggers based on a single indicator, ignoring the pollution diffusion chain and source factors, resulting in the fragmentation of the pollution tracing, diffusion prediction and risk assessment processes, with a high model response delay and inability to achieve linkage decision - making under complex hydrological conditions. This solution creatively adopts a comprehensive supervision method integrating prediction, extreme simulation and reverse tracing, realizing the adaptive adjustment of thresholds driven by sensitive factors under various hydrological disturbance situations (such as sudden floods and drastic changes in flow velocity), thus quickly identifying and generating operable treatment instructions in the early stage of pollution, greatly improving the timeliness and scientific nature of basin - level water quality early warning and intervention;

[0053] (2)In view of the problem existing in the existing water quality prediction methods that the modeling of the hydrodynamic diffusion process in rivers is insufficient. During the period of river rainstorm flood rise, due to the sharp increase in flow velocity, the diffusion distance of pollutants increases, but the model still confines the prediction result within the normal diffusion radius because it does not consider the change in flow velocity, resulting in a serious underestimation of the warning area, which further leads to the technical problems of separate modeling of spatial structure and temporal dynamics dependence, and the prediction result not conforming to physical laws. This solution creatively adopts a spatio - temporal graph convolutional network improved by physical diffusivity constraints for diffusivity water quality prediction, realizing that while ensuring the physical reasonableness of the prediction result, it can also improve the model generalization ability in the case of data loss or sparse sampling, significantly enhancing the stability of the pollution trend evolution modeling;

[0054] (3)In view of the technical problems existing in the existing water quality parameter simulation methods, such as the lack of coverage ability for extreme event conditions, the generated results not meeting the physical consistency constraints, and being difficult to provide reliable simulation data for abnormal pollution risk drills and supervision plan deduction. This solution creatively adopts a conditional adversarial generation method combined with physical verification simulation for extreme water quality parameter simulation, realizing the reasonable inversion of the causal chain and location intensity of the simulated pollution source, thus effectively supporting advanced supervision tasks such as simulation verification, pollution drills and sensitivity analysis;

[0055] (4)In view of the technical problems existing in the existing pollution location methods, such as static traceability or simple reverse simulation, lack of support for multi-source pollution data fusion and uncertainty modeling, and the restoration accuracy of pollution paths being limited by data quality and kinetic noise, this solution creatively adopts the standard multi-modal Bayesian reverse tracing method for inverse diffusion pollution location, realizes high-confidence inversion and location of pollution sources under the premise of observing pollutant concentrations, and at the same time cooperates with the constructed graph deconvolution neural network to strengthen the ability to restore the reverse paths of pollution in time and space. Finally, a pollution source probability heat map is output, and by superimposing simulated event probability reference data, it assists supervisors in accurately locking the emission responsible entity, improving the pertinence of water environment law enforcement and the scientific nature of traceability. Description of the Drawings

[0056] Figure 1 It is a schematic flow chart of the river water quality parameter supervision method based on deep learning provided by the present invention;

[0057] Figure 2 It is a schematic diagram of the river water quality parameter supervision system based on deep learning provided by the present invention;

[0058] Figure 3 It is a schematic flow chart of the diffusive water quality prediction in step S2;

[0059] Figure 4 It is a schematic flow chart of the extreme water quality parameter simulation in step S3;

[0060] Figure 5 It is a schematic flow chart of the inverse diffusion pollution location in step S4.

[0061] The drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the present invention, and do not constitute a limitation to the present invention. Detailed Embodiments

[0062] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the 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 work shall fall within the protection scope of the present invention.

[0063] 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 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 therefore should not be construed as a limitation to the present invention.

[0064] Example 1. Refer to Figure 1 , the method for supervising river water quality parameters based on deep learning provided by the present invention includes the following steps:

[0065] Step S1: Self-calibrated multi-source data acquisition;

[0066] Step S2: Diffusive water quality prediction;

[0067] Step S3: Extreme water quality parameter simulation;

[0068] Step S4: Inverse diffusion pollution location;

[0069] Step S5: Supervision of river water quality parameters.

[0070] By performing the above operations, in the existing intelligent method for supervising river water quality parameters, most of the existing methods stay at the level of "monitoring - alarm", and fail to achieve early warning of pollution trends. The supervision response often triggers based on a single indicator, ignoring the pollution diffusion chain and source factors, resulting in the disconnection of the pollution tracing, diffusion prediction, and risk assessment processes, with a high model response delay and inability to achieve joint decision-making under complex hydrological conditions. The present solution creatively adopts a comprehensive supervision method integrating prediction, extreme simulation, and reverse tracing, achieving threshold adaptive adjustment driven by sensitive factors under various hydrological disturbance situations (such as sudden floods and drastic changes in flow velocity), thereby quickly identifying and generating operable treatment instructions in the early stage of pollution, greatly improving the real-time performance and scientific nature of basin-level water quality early warning and intervention.

[0071] Example 2. Refer to Figure 1 and Figure 2 , based on the above example, in step S1, the self-calibrated multi-source data acquisition is used to deploy sensing nodes and collect multi-source water quality data. Specifically, a multi-source spatio-temporal dynamic calibration fusion perception method is adopted to perform data self-calibration through raw data acquisition to obtain spatio-temporal calibrated data, including the following steps:

[0072] Step S11: Deployment of multi-source perception network. Specifically, key acquisition nodes are selected along the river channel, and a sensor perception network is deployed to collect data, obtaining the original river sensing parameter data;

[0073] The acquisition key nodes include upstream nodes, midstream nodes, downstream nodes, tributary confluence nodes, and outfall nodes; the sensor perception network includes basic water quality sensors, pollutant sensors, and environmental perception sensors; the original river sensing parameter data includes basic water quality parameters, pollutant parameters, and environmental perception auxiliary parameters;

[0074] The basic water quality parameters include dissolved oxygen, temperature, pH value, turbidity, conductivity, and redox potential parameters; the pollutant parameters include ammonia nitrogen, total phosphorus, total nitrogen, COD, and heavy metal pollution index parameters; the environmental perception auxiliary parameters include rainfall, wind speed, and river flow velocity parameters;

[0075] Step S12: Dynamic calibration, specifically, a standard reference segment matching method based on a sliding time window is used for data self-calibration to obtain spatio-temporal calibration data.

[0076] Example 3, refer to Figure 1 、 Figure 2 and Figure 3 Based on the above example, in step S2, the diffusive water quality prediction is used to predict the dynamic diffusion trend of pollutants in the river channel. Specifically, based on the spatio-temporal calibration data, a spatio-temporal graph convolutional network improved by physical diffusivity constraints is used for diffusive water quality prediction to obtain diffusion prediction data, including the following steps:

[0077] Step S21: Basin topology graph construction, specifically, based on the spatio-temporal calibration data, the river channel is formalized and modeled as a graph structure, including graph node definition and directed edge definition, and by introducing virtual river channel confluence points, the normalization of river channel tributaries is carried out to obtain basin topology graph data;

[0078] The graph node definition specifically takes each acquisition key node as a monitoring point and represents the monitoring point with a graph node; the directed edge definition specifically defines the directed edge according to the flow direction of the river and sets the edge weight of the directed edge to the reciprocal of the river channel hydraulic distance;

[0079] Step S22: Diffusive physical coefficient embedding, specifically, by introducing pollutant concentration, diffusivity coefficient, and water body flow velocity parameters, a continuous physical equation is constructed for integrated physical constraint modeling, and by taking the diffusivity coefficient and water body flow velocity parameters in the continuous physical equation as learnable parameters, discrete differentiable parameter training for physical coefficient embedding is carried out, and a physical residual regularization loss function is introduced to optimize the model training process of diffusive water quality prediction to obtain an embedded physical operator; the embedded physical operator specifically includes a continuous physical equation and a physical residual regularization loss function;

[0080] The calculation formula of the continuous physical equation is:

[0081] ;

[0082] In the formula, is the partial derivative term of concentration with respect to time, which is used to represent the output of the continuous physical equation. C is the pollutant concentration parameter, t is the time index, D is the diffusion coefficient, and v is the water body flow velocity parameter. is the water body flow velocity vector. is the concentration gradient parameter, and R(·) is the chemical reaction identification term.

[0083] The calculation formula of the physical residual regularization loss function is:

[0084] ;

[0085] In the formula, is the physical residual regularization loss function. is the regularization coefficient. is the predicted pollutant concentration. is the diffusion law correction term, which is used to represent the part where the model output conforms to the diffusion law in space. is the river channel flow velocity correction term, which is used to represent the part where the propagation trend of the model output in space is affected by the river convection.

[0086] Step S23: Spatiotemporal prediction modeling, specifically constructing a graph convolutional neural network that mixes spatiotemporal features as the basic prediction subnet for diffusive water quality prediction, and training the model according to the embedded physical operator to obtain a spatiotemporal prediction model.

[0087] The graph convolutional neural network that mixes spatiotemporal features includes a spatial feature layer and a temporal feature layer. The spatial feature layer specifically uses a standard graph convolutional structure to predict the spatial diffusion between river channel monitoring points. The temporal feature layer specifically uses a causal dilated convolutional structure to predict the temporal evolution features between river channel detection points.

[0088] The modeling calculation formula of the spatial feature layer is:

[0089] ;

[0090] In the formula, H (l+1) is the output feature matrix of the (l + 1)-th layer of graph convolution, which is used to represent the spatial diffusion prediction features between river channel monitoring points. is the non-linear activation function. is the degree matrix of the adjacency matrix. is the self-connected adjacency matrix, and H (l) is the output feature matrix of the l-th layer of graph convolution, and W (l) is the weight of the l-th layer of the spatial feature layer.

[0091] The modeling calculation formula of the time feature layer is as follows:

[0092] ;

[0093] In the formula, y t is the output feature vector at the current time t, used to represent the predicted value of the time evolution feature between river channel detection points. K is the total scale of the causal dilation convolution kernel, used to represent the total number of time periods. k is the convolution kernel index, d is the dilation factor parameter, and w k is the convolution kernel weight corresponding to the k-th convolution kernel. is the input sequence feature vector, specifically used to represent the basin topology map data.

[0094] Preferably, Table 1 is the parameter example table of the graph convolutional neural network for spatio-temporal feature mixing. As shown in the table, the model type of the spatial feature layer is the standard graph convolutional structure, and the parameter types include the number of graph convolutional layers, activation function, feature dimension, and dropout ratio; the model type of the time feature layer is the causal dilation convolution structure, and the parameter types include the number of causal dilation convolution layers, convolution kernel size, dilation factor, activation function, and feature dimension; based on the basic training parameters for model training according to the embedded physical operator, the parameter types include optimizer, learning rate, training batch, and maximum number of training epochs.

[0095] Table 1 Parameter Example Table of Graph Convolutional Neural Network for Spatio-Temporal Feature Mixing

[0096]

[0097] Step S24: Diffusive water quality prediction, specifically, through the construction of the basin topology map, the embedding of the diffusive physical coefficient, and the spatio-temporal prediction modeling, according to the spatio-temporal calibration data, using the spatio-temporal prediction model, perform diffusive water quality prediction to obtain diffusion prediction data.

[0098] By performing the above operations, in the existing water quality prediction methods, there is a problem of insufficient modeling of the river hydrodynamic diffusion process. During the period of river storm flood rise, due to the sharp increase in flow velocity, the diffusion distance of pollutants increases, but the model does not consider the change in flow velocity, and the prediction result is still limited to the usual diffusion radius, resulting in a serious underestimation of the warning area, which further leads to the technical problems of separate modeling of spatial structure and time dynamics dependence, and the prediction result does not conform to the physical law. This solution creatively uses a spatio-temporal graph convolutional network improved by physical diffusivity constraints to perform diffusive water quality prediction, realizing that while ensuring the physical reasonableness of the prediction result, it can also improve the model generalization ability in the case of data loss or sparse sampling, and significantly enhance the stability of the pollution trend evolution modeling.

[0099] Example 4, refer to Figure 1 、Figure 2 and Figure 4 This embodiment is based on the above - mentioned embodiment. In step S3, the extreme water quality parameter simulation is used to enhance the system's ability to identify sudden pollution. Specifically, a conditional adversarial generation method combined with physical verification simulation is adopted to perform extreme water quality parameter simulation and obtain extreme parameter simulation data, including the following steps:

[0100] Step S31: Construction of the generative adversarial network. Specifically, by sequentially constructing a standard generator structure and a discriminator structure and introducing conditional control labels, extreme polluted water quality parameters are generated to obtain a basic network for generating extreme water quality parameters;

[0101] The standard generator structure adopts a 5 - layer transposed convolutional network and generates a sequence of simulated pollution parameters;

[0102] The discriminator structure adopts a standard three - dimensional convolutional network and outputs a data authenticity probability value;

[0103] The conditional control labels include extreme event types, spatio - temporal locations, and meteorological environmental conditions;

[0104] The extreme event types include industrial emissions, agricultural activities, urban activities, meteorological disasters, and human - caused accidents;

[0105] The spatio - temporal locations include river monitoring point numbers, longitudes, latitudes, and extreme event times;

[0106] The meteorological environmental conditions include rainfall intensity, water body temperature, ambient temperature, wind speed, and humidity parameters;

[0107] Preferably, the industrial emissions specifically include industrial leaks, illegal discharges, and sudden increases in sewage load; the agricultural activities include chemical fertilizer leakage, pesticide seepage, and livestock and poultry breeding wastewater leakage; the urban activities include urban drainage network pollution, rain - sewage mixing pollution, and urban sewage pipe leakage; the meteorological disasters include heavy rain, drought, and flood backflow; the human - caused accidents include transport vehicle accidents, illegal mud entering the river, and poisoning pollution;

[0108] Step S32: Physical verification simulation. Specifically, the physical residual regular loss function is used as the physical verification loss and embedded into the discriminator structure to perform physical consistency correction on the extreme water quality parameters generated by the basic network for generating extreme water quality parameters, and obtain physically verified and optimized simulation parameters;

[0109] Step S33: Parameter anomaly attribution optimization, specifically, optimizing simulation parameters based on the physical verification, generating a parameter attribution vector by constructing a simulated extreme water quality parameter space and a reverse attribution parameter optimization network, obtaining reference data for extreme water quality parameter attribution, and performing manual inspection and simulation verification of extreme water quality parameters based on the reference data for extreme water quality parameter attribution;

[0110] The simulated extreme water quality parameter space includes extreme pollution source locations, pollution occurrence times, emission intensities, and flow velocity disturbance parameters;

[0111] The reverse attribution parameter optimization network specifically adopts an encoder-decoder structure, uses a standard time series graph attention network as the encoder structure, and uses a multi-head regression output layer as the decoder structure;

[0112] The reference data for extreme water quality parameter attribution includes pollution types, pollution source locations, pollution occurrence times, predicted pollution emission intensity values, predicted pollution flow velocity values, and true pollution comparison values;

[0113] Step S34: Extreme water quality parameter simulation, specifically, performing extreme water quality parameter simulation through the construction of the generative adversarial network, the physical verification simulation, and the parameter anomaly attribution optimization to obtain extreme parameter simulation data;

[0114] The extreme parameter simulation data specifically includes a simulated pollution concentration parameter sequence, meteorological parameters, human control parameters, and a simulated physical verification score.

[0115] By performing the above operations, in the existing water quality parameter simulation methods, there are technical problems such as the lack of coverage ability for extreme event conditions, the generated results cannot meet the physical consistency constraints, and it is difficult to provide reliable simulation data for abnormal pollution risk drills and regulatory plan deductions. This solution creatively uses a conditional adversarial generation method combined with physical verification simulation to perform extreme water quality parameter simulation, realizing the causal chain reconstruction of simulated pollution sources and the reasonable inversion of location intensities, thereby effectively supporting advanced regulatory tasks such as simulation verification, pollution drills, and sensitivity analysis.

[0116] Example Five, refer to Figure 1 、 Figure 2 and Figure 5 Based on the above example, in step S4, the inverse diffusion pollution positioning is used to trace the pollution source location. Specifically, based on the extreme parameter simulation data and the diffusion prediction data, a standard multi-modal Bayesian reverse tracing method is used to perform inverse diffusion pollution positioning to obtain pollution tracing data, including the following steps:

[0117] Step S41: Inverse diffusion modeling, specifically, constructing a mathematical model for Bayesian inversion problem, taking the predicted diffusion concentration in the diffusion prediction data as input data, constructing an inverse diffusion model and performing inverse calculation of pollution source parameters to obtain inverse diffusion parameters of the pollution source;

[0118] The inverse diffusion model includes a forward diffusion sub-model and an inverse inversion model; for the forward diffusion sub-model, specifically, mathematical modeling is carried out using the continuous physical equation; for the inverse inversion model, specifically, mathematical modeling is carried out using the standard Bayesian inversion expression, and sampling optimization is carried out through the standard Markov chain Monte Carlo method;

[0119] The calculation formula of the standard Bayesian inversion expression is:

[0120] ;

[0121] In the formula, The overall is the posterior probability distribution term, which is used to represent the probability distribution of the pollution source parameters obs on the premise of the known predicted diffusion concentration C , where are the pollution source parameters, C obs is the predicted diffusion concentration in the diffusion prediction data, is the proportional symbol, The overall is the likelihood function term, which is used to represent the probability of obtaining the predicted diffusion concentration C based on the pollution source parameters obs , is the prior probability distribution term, which is used to represent the prior probability value of the pollution source parameters without introducing the predicted diffusion concentration;

[0122] Step S42: Spatiotemporal deconvolution improvement, specifically, based on the spatiotemporal prediction model, constructing a graph deconvolution neural network for mixing spatiotemporal features as the basic prediction subnet for inverse diffusion modeling to obtain a spatiotemporal deconvolution inverse diffusion prediction model;

[0123] The graph deconvolution neural network for mixing spatiotemporal features includes a spatial inverse path layer and a temporal inverse path layer; for the spatial inverse path layer, specifically, a standard graph deconvolution structure is used to predict the inverse pollution path features between river channel monitoring points; for the temporal inverse path layer, specifically, a reverse dilated convolution structure is used to predict the reverse feature dependence of the pollution time series between river channel detection points;

[0124] Step S43: Pollution probability visualization, specifically, based on the inverse diffusion parameters of the pollution source, using the spatiotemporal deconvolution inverse diffusion prediction model to predict the spatiotemporal data of the inverse diffusion pollution source, constructing a spatiotemporal distribution probability heat map of the pollution source to obtain pollution probability visualization reference data;

[0125] Step S44: Inverse diffusion pollution localization. Specifically, based on the pollution probability visualization reference data and the pollution source tracing prediction data output by the spatio-temporal deconvolution inverse diffusion prediction model, inverse diffusion pollution localization is performed to obtain pollution tracing data. The pollution tracing data specifically includes pollution source location parameters, pollution time parameters, emission intensity parameters, pollution probability visualization reference data, and extreme pollution event probability reference data.

[0126] By performing the above operations, in view of the technical problems existing in the existing pollution localization methods, such as static tracing or simple reverse simulation, lack of support for multi-source pollution data fusion and uncertainty modeling, and the restoration accuracy of the pollution path being limited by data quality and kinetic noise, this solution creatively adopts the standard multi-modal Bayesian inverse tracing method to perform inverse diffusion pollution localization, achieving a high-confidence inverse localization of the pollution source under the premise of observing pollutant concentrations. At the same time, in cooperation with the constructed graph deconvolution neural network, the ability to restore the reverse path of pollution in time and space is strengthened. Finally, a pollution source probability heat map is output, and by superimposing the simulated event probability reference data, it assists the supervisors to accurately lock in the emission responsible entity, improving the pertinence and tracing scientificity of water environment law enforcement.

[0127] Example 6, refer to Figure 1 and Figure 2 , based on the above example, in step S5, the river water quality parameter supervision is used to integrate the prediction, extreme simulation, and inverse tracing results and achieve monitoring and management. Specifically, the dynamic threshold intelligent supervision decision-making method is adopted to perform river water quality parameter supervision to obtain river water quality supervision decision reference data.

[0128] The dynamic threshold intelligent supervision decision-making method specifically calculates the dynamic threshold by introducing environmental characteristic factors to construct an environmental sensitivity factor based on the pollution tracing data and the diffusion prediction data, and obtains a dynamic water quality supervision threshold. And based on the dynamic water quality supervision threshold, water quality abnormal event classification and supervision are carried out.

[0129] The calculation formula for the dynamic threshold calculation is:

[0130] ;

[0131] In the formula, is the dynamic water quality supervision threshold, is the moving average of the i-th water quality parameter in the historical data of the current basin, is the standard deviation of the i-th water quality parameter in the historical data of the current basin, where i is the water quality parameter index, used to represent the water quality parameters in the pollution tracing data and the diffusion prediction data. is an environmental sensitivity factor, which is specifically adjusted and trained by introducing temperature, flow rate, and rainfall parameters and constructing a lightweight multi-layer perceptron. The specific calculation formula of the environmental sensitivity factor is , where MLP(·) is a lightweight multi-layer perceptron representation function, which adopts a three-layer fully connected network. The input is set as the environmental parameter vector, and the output is set as the scalar adjustment factor. Rain(t) is the rainfall parameter, Flow(t) is the flow rate parameter, and Temperature(t) is the temperature parameter;

[0132] The reference data for river water quality supervision and decision-making specifically includes pollution source tracing instructions, sewage discharge limit instructions, verification and monitoring instructions, and a reference ranking for the priority of polluted water quality treatment.

[0133] Example Seven, refer to Figure 1 and Figure 2 , based on the above embodiments, the river water quality parameter supervision system based on deep learning provided by the present invention includes a self-calibrating multi-source data acquisition module, a diffusive water quality prediction module, an extreme water quality parameter simulation module, an inverse diffusion pollution location module, and a river water quality parameter supervision module;

[0134] The self-calibrating multi-source data acquisition module is used for self-calibrating multi-source data acquisition. Through self-calibrating multi-source data acquisition, spatio-temporal calibration data is obtained and sent to the diffusive water quality prediction module;

[0135] The diffusive water quality prediction module is used for diffusive water quality prediction. Through diffusive water quality prediction, diffusion prediction data is obtained and sent to the inverse diffusion pollution location module and the river water quality parameter supervision module;

[0136] The extreme water quality parameter simulation module is used for extreme water quality parameter simulation. Through extreme water quality parameter simulation, extreme parameter simulation data is obtained and sent to the river water quality parameter supervision module;

[0137] The inverse diffusion pollution location module is used for inverse diffusion pollution location. Through inverse diffusion pollution location, pollution source tracing data is obtained and sent to the river water quality parameter supervision module;

[0138] The river water quality parameter supervision module is used for river water quality parameter supervision. Through river water quality parameter supervision, reference data for river water quality supervision and decision-making is obtained.

[0139] It should be noted that, in this document, 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 includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article or device.

[0140] Although embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the invention.

[0141] The present invention and its embodiments have been described above. 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. In general, 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. A method for monitoring river water quality parameters based on deep learning, characterized in that: The method includes the following steps: Step S1: Self-calibrated multi-source data acquisition to obtain spatio-temporal calibration data; Step S2: Diffusive water quality prediction. Using a spatio-temporal graph convolutional network improved by physical diffusivity constraints to perform diffusive water quality prediction and obtain diffusion prediction data, including the following steps: Step S21: Basin topology graph construction; Step S22: Embedding of diffusive physical coefficients. By introducing pollutant concentration, diffusivity coefficient, and water body flow velocity parameters, a continuous physical equation is constructed for integrated physical constraint modeling; Step S23: Spatio-temporal prediction modeling; Step S24: Diffusive water quality prediction; Step S3: Extreme water quality parameter simulation. Using a conditional adversarial generation method combined with physical verification simulation to perform extreme water quality parameter simulation and obtain extreme parameter simulation data; Step S4: Inverse diffusion pollution location. Using a standard multi-modal Bayesian reverse tracing method to perform inverse diffusion pollution location and obtain pollution tracing data, including the following steps: Step S41: Reverse diffusion modeling; Step S42: Spatio-temporal deconvolution improvement; Step S43: Visualization of pollution probability; Step S44: Inverse diffusion pollution location; Step S5: River water quality parameter supervision. Using a dynamic threshold intelligent supervision decision-making method to perform river water quality parameter supervision and obtain river water quality supervision decision reference data.

2. The river water quality parameter supervision method based on deep learning according to claim 1, characterized in that: In step S1, the self-calibrated multi-source data acquisition is used to deploy sensing nodes and collect multi-source water quality data. Specifically, a multi-source spatio-temporal dynamic calibration fusion sensing method is adopted to perform data self-calibration through raw data acquisition to obtain spatio-temporal calibration data, including the following steps: Step S11: Deployment of a multi-source sensing network. Specifically, key acquisition nodes are selected along the river channel to deploy a sensor sensing network for data acquisition to obtain raw river sensing parameter data; The key acquisition nodes include upstream nodes, midstream nodes, downstream nodes, tributary confluence nodes, and outfall nodes; the sensor sensing network includes basic water quality sensors, pollutant sensors, and environmental sensing sensors; the raw river sensing parameter data includes basic water quality parameters, pollutant parameters, and environmental sensing auxiliary parameters; Step S12: Dynamic calibration. Specifically, a standard benchmark segment matching method based on a sliding time window is adopted to perform data self-calibration to obtain spatio-temporal calibration data.

3. The method for monitoring river water quality parameters based on deep learning according to claim 2, characterized in that: In step S2, the diffusive water quality prediction is used to predict the dynamic diffusion trend of pollutants in the river channel. Specifically, based on the spatio-temporal calibration data, a spatio-temporal graph convolutional network improved by physical diffusivity constraints is adopted to perform diffusive water quality prediction and obtain diffusion prediction data, including the following steps: Step S21: Basin topology graph construction. Specifically, based on the spatio-temporal calibration data, the river channel is formalized and modeled as a graph structure, including graph node definition and directed edge definition, and by introducing a virtual river channel confluence point, normalization of river tributaries is performed to obtain basin topology graph data; Step S22: Embedding of diffusivity physical coefficients. Specifically, by introducing pollutant concentration, diffusivity coefficient, and water body flow velocity parameters, a continuous physical equation is constructed, integrated physical constraint modeling is carried out, and the diffusivity coefficient and water body flow velocity parameters in the continuous physical equation are used as learnable parameters to perform discrete differentiable parameter training for physical coefficient embedding. A physical residual regularization loss function is introduced to optimize the model training process for diffusivity water quality prediction, and an embedded physical operator is obtained. The embedded physical operator specifically includes a continuous physical equation and a physical residual regularization loss function. Step S23: Spatiotemporal prediction modeling. Specifically, a graph convolutional neural network that mixes spatiotemporal features is constructed as the basic prediction subnet for diffusivity water quality prediction, and the model is trained based on the embedded physical operator to obtain a spatiotemporal prediction model. The graph convolutional neural network that mixes spatiotemporal features includes a spatial feature layer and a temporal feature layer. The spatial feature layer specifically uses a standard graph convolutional structure to predict spatial diffusion between river monitoring points. The temporal feature layer specifically uses a causal dilated convolutional structure to predict temporal evolution features between river detection points. Step S24: Diffusivity water quality prediction. Specifically, through the construction of the watershed topology graph, the embedding of diffusivity physical coefficients, and the spatiotemporal prediction modeling, based on the spatiotemporal calibration data, the spatiotemporal prediction model is used to perform diffusivity water quality prediction to obtain diffusion prediction data.

4. The river water quality parameter supervision method based on deep learning according to claim 3, characterized in that: In step S3, the extreme water quality parameter simulation is used to enhance the system's ability to identify sudden pollution. Specifically, a conditional adversarial generation method combined with physical verification simulation is used to perform extreme water quality parameter simulation to obtain extreme parameter simulation data, including the following steps: Step S31: Construction of a generative adversarial network. Specifically, by sequentially constructing a standard generator structure and a discriminator structure and introducing conditional control labels, extreme polluted water quality parameters are generated to obtain a basic network for generating extreme water quality parameters. The conditional control labels include extreme event types, spatiotemporal positions, and meteorological environmental conditions. The extreme event types include industrial emissions, agricultural activities, urban activities, meteorological disasters, and human-caused accidents. Step S32: Physical verification simulation. Specifically, the physical residual regularization loss function is embedded into the discriminator structure as the physical verification loss to correct the physical consistency of the extreme water quality parameters generated by the basic network for generating extreme water quality parameters, and physically verified optimized simulation parameters are obtained. Step S33: Parameter anomaly attribution optimization. Specifically, based on the physically verified optimized simulation parameters, by constructing a simulated extreme water quality parameter space and a reverse attribution parameter optimization network, parameter attribution vectors are generated to obtain extreme water quality parameter attribution reference data, and the extreme water quality parameters are manually checked and simulated verified based on the extreme water quality parameter attribution reference data. The simulated extreme water quality parameter space includes extreme pollution source locations, pollution occurrence times, emission intensities, and flow velocity perturbation parameters. The reverse attribution parameter optimization network specifically adopts an encoder-decoder structure, uses a standard time series graph attention network as the encoder structure, and uses a multi-head regression output layer as the decoder structure; The extreme water quality parameter attribution reference data includes pollution type, pollution source location, pollution occurrence time, predicted pollution emission intensity, predicted pollution flow rate, and true pollution comparison value; Step S34: Extreme water quality parameter simulation, specifically, through the construction of the generative adversarial network, the physical verification simulation, and the parameter anomaly attribution optimization, perform extreme water quality parameter simulation to obtain extreme parameter simulation data.

5. The river water quality parameter supervision method based on deep learning according to claim 4, characterized in that: In step S34, the extreme parameter simulation data specifically includes a simulated pollution concentration parameter sequence, meteorological parameters, human control parameters, and a simulated physical verification score.

6. The method for supervising river water quality parameters based on deep learning according to claim 5, characterized in that: In step S4, the inverse diffusion pollution location is used to trace the pollution source location. Specifically, based on the extreme parameter simulation data and the diffusion prediction data, use the standard multi-modal Bayesian reverse tracing method to perform inverse diffusion pollution location to obtain pollution tracing data, including the following steps: Step S41: Inverse diffusion modeling, specifically, construct a mathematical model of the Bayesian inversion problem, use the predicted diffusion concentration in the diffusion prediction data as the input data, construct an inverse diffusion model and perform inverse calculation of the pollution source parameters to obtain the inverse diffusion parameters of the pollution source; The inverse diffusion model includes a forward diffusion sub-model and an inverse inversion model; the forward diffusion sub-model specifically uses the continuous physical equation for mathematical modeling; the inverse inversion model specifically uses the standard Bayesian inversion expression for mathematical modeling and performs sampling optimization through the standard Markov chain Monte Carlo method; Step S42: Spatiotemporal deconvolution improvement, specifically, based on the spatiotemporal prediction model, construct a graph deconvolution neural network that mixes spatiotemporal features as the basic prediction subnet for inverse diffusion modeling to obtain a spatiotemporal deconvolution inverse diffusion prediction model; The graph deconvolution neural network that mixes spatiotemporal features includes a spatial inverse path layer and a temporal inverse path layer; the spatial inverse path layer specifically uses the standard graph deconvolution structure to predict the inverse pollution path features between river channel monitoring points; the temporal inverse path layer specifically uses the reverse dilated convolution structure to predict the reverse feature dependence of the pollution time series between river channel detection points; Step S43: Pollution probability visualization, specifically, based on the inverse diffusion parameters of the pollution source, use the spatiotemporal deconvolution inverse diffusion prediction model to perform inverse diffusion pollution source spatiotemporal data prediction, construct a spatiotemporal distribution probability heat map of the pollution source to obtain pollution probability visualization reference data; Step S44: Inverse diffusion pollution location, specifically, based on the pollution probability visualization reference data and the pollution source tracing prediction data output by the spatiotemporal deconvolution inverse diffusion prediction model, perform inverse diffusion pollution location to obtain pollution tracing data; the pollution tracing data specifically includes pollution source location parameters, pollution time parameters, emission intensity parameters, pollution probability visualization reference data, and extreme pollution event probability reference data.

7. The method for supervising river water quality parameters based on deep learning according to claim 6, characterized in that: In step S5, the supervision of river water quality parameters is used to integrate the results of prediction, extreme simulation, and reverse tracing and to achieve monitoring and management. Specifically, a dynamic threshold intelligent supervision decision-making method is adopted to supervise the river water quality parameters, and reference data for river water quality supervision decisions is obtained. The dynamic threshold intelligent supervision decision-making method specifically calculates the dynamic water quality supervision threshold by constructing an environmental sensitivity factor by introducing environmental characteristic factors based on the pollution tracing data and the diffusion prediction data, and classifies and supervises water quality abnormal events according to the dynamic water quality supervision threshold. The reference data for river water quality supervision decisions specifically includes pollution tracing instructions, sewage discharge restriction instructions, verification monitoring instructions, and a reference ranking of the priorities for treating polluted water quality.

8. A river water quality parameter supervision system based on deep learning, which is used to implement the river water quality parameter supervision method based on deep learning according to any one of claims 1-7, and is characterized in that: It includes a self-calibrating multi-source data acquisition module, a diffusive water quality prediction module, an extreme water quality parameter simulation module, an inverse diffusion pollution location module, and a river water quality parameter supervision module.

9. The river water quality parameter supervision system based on deep learning according to claim 8, characterized in that: The self-calibrating multi-source data acquisition module is used for self-calibrating multi-source data acquisition. Through self-calibrating multi-source data acquisition, spatio-temporal calibration data is obtained and sent to the diffusive water quality prediction module. The diffusive water quality prediction module is used for diffusive water quality prediction. Through diffusive water quality prediction, diffusion prediction data is obtained and sent to the inverse diffusion pollution location module and the river water quality parameter supervision module. The extreme water quality parameter simulation module is used for extreme water quality parameter simulation. Through extreme water quality parameter simulation, extreme parameter simulation data is obtained and sent to the river water quality parameter supervision module. The inverse diffusion pollution location module is used for inverse diffusion pollution location. Through inverse diffusion pollution location, pollution tracing data is obtained and sent to the river water quality parameter supervision module. The river water quality parameter supervision module is used for river water quality parameter supervision. Through river water quality parameter supervision, reference data for river water quality supervision decisions is obtained.

Images