A land pollution monitoring, forecasting and early warning system and method

Abstract

The application discloses a land pollution monitoring, predicting and early warning system and method, relates to the technical field of environmental monitoring and pollution early warning, and comprises the following steps: acquiring multi-source data of a target land; pre-processing, feature extraction and fusion of the multi-source data; inputting the multi-source data into a dynamic pollution diffusion model to obtain a pollutant concentration field of the target land; inputting the fused data and the pollutant concentration field into a land pollutant concentration prediction model for prediction to obtain a pollutant concentration prediction value of the target land in a future target time period; comparing the pollutant concentration prediction value of the target land in the future target time period with early warning threshold values of different levels, and generating early warning instructions of corresponding levels according to a comparison result; generating corresponding early warning information according to the early warning instructions; and executing corresponding early warning measures according to the early warning information; and the application can realize scientific evaluation and hierarchical control of pollution risks through multi-dimensional data integration and model optimization.

Description

Technical Field

[0001] This invention relates to the field of environmental monitoring and pollution early warning technology, specifically to a site pollution monitoring, prediction and early warning system and method. Background Technology

[0002] Soil and groundwater pollution is characterized by its concealment, delayed effects, and cumulative nature. Most monitoring methods rely on static data, making it difficult to dynamically capture pollution migration trends. Early warning systems suffer from the following shortcomings: First, insufficient data utilization, relying solely on monitoring data from single sites and ignoring surrounding environmental factors such as geological background and meteorological conditions. Second, crude early warning grading, with most systems providing only a binary "exceeding / not exceeding" judgment, lacking a multi-level early warning mechanism and failing to provide refined support for risk management. Third, weak dynamic prediction capabilities, failing to integrate pollution diffusion models with machine learning techniques, making it difficult to predict the long-term impact of pollution migration on sensitive targets (such as residential areas and water sources).

[0003] Therefore, there is an urgent need for a hierarchical early warning system that can integrate dynamic monitoring, spatiotemporal analysis and intelligent prediction, so as to achieve scientific assessment and hierarchical control of pollution risks through multi-dimensional data integration and model optimization. Summary of the Invention

[0004] In view of this, the present invention provides a site pollution monitoring, prediction and early warning system and method to solve the technical problems in related technologies.

[0005] In a first aspect, the present invention provides a method for monitoring, predicting, and warning of land pollution, comprising:

[0006] S1. Acquire multi-source data of the target site; the multi-source data includes satellite remote sensing data, environmental monitoring data, and meteorological and hydrological data;

[0007] S2. Preprocess the multi-source data;

[0008] S3. Perform feature extraction and fusion on the preprocessed multi-source data to obtain fused features;

[0009] S4. Input the multi-source data into the dynamic pollution diffusion model to obtain the pollutant concentration field of the target site;

[0010] S5. The fused data and pollutant concentration field are input into the pre-trained land pollutant concentration prediction model for prediction, to obtain the predicted pollutant concentration value of the target land within the target time period in the future; the land pollutant concentration prediction model is determined based on LSTM neural network and CNN neural network;

[0011] S6. Compare the predicted pollutant concentration of the target plot within the target time period with the warning thresholds at all levels, and generate warning instructions of the corresponding level based on the comparison results.

[0012] S7. Generate corresponding early warning information according to the early warning instruction; and execute corresponding early warning measures according to the early warning information.

[0013] In one optional implementation, S2 includes:

[0014] S21. Perform radiometric and geometric corrections on satellite remote sensing data;

[0015] S22. Perform missing value interpolation and outlier detection on environmental monitoring data, and adopt time-frequency domain joint denoising based on wavelet transform;

[0016] S23. Perform spatiotemporal alignment processing on meteorological and hydrological data and establish a spatiotemporal coordinate mapping matrix.

[0017] In one optional implementation, S3 includes:

[0018] S31. Construct a multi-scale spatiotemporal feature pyramid and use a 3D-CNN network to extract the spatial features of the preprocessed satellite remote sensing data.

[0019] S32. Utilize graph attention networks to extract spatial correlation features from environmental monitoring data;

[0020] S33. Capture the temporal dependence characteristics of meteorological and hydrological data using a Transformer encoder;

[0021] S34. A feature alignment network is used to fuse the spatial features of satellite remote sensing data, the spatial correlation features of environmental monitoring data, and the temporal dependence features of meteorological and hydrological data across modes. The fusion weights are dynamically adjusted through an adaptive gating mechanism.

[0022] In one optional implementation, the dynamic pollution diffusion model is:

[0023]

[0024] σ y = 0.1x(1+0.0001x) -0.5

[0025] σ z = 0.06x(1+0.0015x) -0.5

[0026] Where C(x,y,z) represents the pollutant concentration at the pollution source at a distance x downwind, y crosswind, and z vertically, in g / m³.3 Q represents the pollution source strength, in kg / s; u represents the average wind speed, in m / s; H represents the effective emission height of the pollution source; σ y σ is the horizontal diffusion parameter. z This is the vertical diffusion parameter.

[0027] In one optional implementation, the land parcel pollutant concentration prediction model is a spatiotemporal parallel dual-stream network structure, comprising an LSTM temporal encoding stream and a CNN spatial encoding stream; the LSTM temporal encoding stream adopts a structure combining gated recurrent units and an attention mechanism; the CNN spatial encoding stream uses a residual shrinking network to process high-dimensional spatial features.

[0028] Set up a cross-modal fusion layer for spatiotemporal feature interaction, and configure a transfer learning interface at the output end.

[0029] In one optional implementation, S6 includes:

[0030] S61. Establish a multi-level threshold system; the multi-level threshold system includes: Level IV blue warning, Level III yellow warning, Level II orange warning and Level I red warning, and the thresholds of each level correspond to θ4, θ3, θ2 and θ1 respectively;

[0031] S62. Compare the predicted pollutant concentration of the target plot at the target time with the threshold values ​​of each level of early warning.

[0032] S63. When the predicted pollutant concentration falls within the range of [θ4, θ3), a Level IV blue alert is triggered at the target time; when the predicted pollutant concentration falls within the range of [θ3, θ2), a Level III yellow alert is triggered at the target time; when the predicted pollutant concentration falls within the range of [θ2, θ1) and continues to exceed a preset time period, a Level II orange alert is triggered at the time corresponding to the preset time period at the target time; when the predicted pollutant concentration falls within the range of [θ1, +∞) and the diffusion range covers sensitive targets, a Level I red alert is triggered at the time corresponding to the preset time period at the target time.

[0033] S64. Generate a warning instruction corresponding to the determined warning level.

[0034] In an optional implementation, S63 further includes:

[0035] When the predicted pollutant concentration fluctuates between the thresholds of the two-level warning, the warning level is determined by combining the growth or decline trend of the predicted pollutant concentration of the target plot in the future target time period output by the plot pollutant concentration prediction model.

[0036] Secondly, the present invention provides a land pollution monitoring, prediction, and early warning system, comprising:

[0037] The data acquisition module is used to acquire multi-source data of the target site; the multi-source data includes satellite remote sensing data, environmental monitoring data, and meteorological and hydrological data.

[0038] The data preprocessing module is used to preprocess the multi-source data;

[0039] The feature fusion module is used to extract and fuse features from preprocessed multi-source data to obtain fused features.

[0040] The pollutant concentration calculation module is used to input the multi-source data into the dynamic pollution diffusion model to obtain the pollutant concentration field of the target site;

[0041] The pollutant concentration prediction module is used to input the fused data and the pollutant concentration field into a pre-trained land pollutant concentration prediction model for prediction, so as to obtain the predicted value of pollutant concentration of the target land within a future target time period; the land pollutant concentration prediction model is determined based on LSTM neural network and CNN neural network;

[0042] The warning level determination module is used to compare the predicted pollutant concentration of the target plot in the future target time period with the warning thresholds of each level, and generate warning instructions of the corresponding level based on the comparison results.

[0043] The early warning measure determination module is used to generate corresponding early warning information based on the early warning instruction, and to execute corresponding early warning measures based on the early warning information.

[0044] The present invention has the following beneficial effects:

[0045] (1) Data is fully utilized, no longer relying on monitoring data of a single plot, and considering the correlation with the surrounding environment, such as geological background and meteorological conditions, to integrate multi-dimensional data.

[0046] (2) Add a multi-level early warning mechanism to provide refined support for risk management.

[0047] (3) It has strong dynamic prediction capabilities and can combine pollution diffusion models with machine learning technology to predict pollutant concentrations more accurately.

[0048] In summary, the graded early warning system of this invention can integrate dynamic monitoring, spatiotemporal analysis and intelligent prediction, and achieve scientific assessment and graded control of pollution risks through multi-dimensional data integration and model optimization. Attached Figure Description

[0049] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0050] Figure 1 This is a flowchart illustrating a method for monitoring, predicting, and issuing early warnings of land pollution according to an embodiment of the present invention.

[0051] Figure 2 This is a schematic diagram of a land pollution monitoring, prediction and early warning system according to an embodiment of the present invention. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0053] Figure 1 This is a method for monitoring, predicting and warning of land pollution according to an embodiment of the present invention. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Also, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0054] like Figure 1 As shown, a method for monitoring, predicting, and warning of pollution in a land parcel includes the following steps:

[0055] S1. Acquire multi-source data; multi-source data includes satellite remote sensing data, environmental monitoring data, and meteorological and hydrological data;

[0056] Among them, satellite remote sensing data can be obtained through high-resolution satellites (such as Landsat and Sentinel) to acquire data on land use types (such as industrial land and agricultural land), water distribution, surface reflectance, vegetation index, and surface temperature of the plot and its surroundings.

[0057] Environmental monitoring data can be collected in real time by deploying an Internet of Things (IoT) sensor network to collect data on soil pollutant concentrations (heavy metal pollutant concentrations, organic pollutant concentrations, etc.), groundwater quality (soil pH, heavy metal ions), and air pollutants (PM2.5, VOCs).

[0058] Meteorological and hydrological data can be obtained by accessing the meteorological bureau's API to obtain meteorological parameters such as wind speed, wind direction, rainfall, temperature, humidity, and groundwater flow direction, and combined with hydrological models (such as SWAT) to analyze the impact of surface runoff on pollutant migration.

[0059] S2. Preprocess the multi-source data.

[0060] In one optional implementation, step S2 includes:

[0061] S21. Perform radiometric and geometric corrections on satellite remote sensing data.

[0062] Specifically, satellite remote sensing data is affected by factors such as sensor response characteristics, atmospheric absorption / scattering, and solar altitude angle during acquisition, leading to radiance distortion (radiative aberration) and image geometric deformation. Radiometric correction quantifies atmospheric effects by establishing a radiative transfer model (such as the MODTRAN model), converting the original digital quantized values ​​(DN values) into surface reflectance or radiance. Geometric correction, on the other hand, constructs a sensor imaging geometric model (such as the RPC rational function model) and combines it with ground control points (GCPs) or digital elevation models (DEMs) to eliminate geometric distortions caused by terrain undulations and satellite attitude changes, achieving accurate registration between the image and the geographic coordinate system.

[0063] S22. Perform missing value interpolation and outlier detection on environmental monitoring data, and adopt time-frequency domain joint denoising based on wavelet transform.

[0064] Specifically, environmental monitoring data (such as soil heavy metal concentration and groundwater pH) are often missing or abnormal due to equipment malfunctions and sampling errors. Missing value imputation must preserve the data distribution characteristics, and outlier detection must distinguish between genuine abrupt changes and noise. Wavelet transform, through time-frequency domain decomposition, suppresses high-frequency noise while preserving the effective signal.

[0065] S23. Perform spatiotemporal alignment processing on meteorological and hydrological data and establish a spatiotemporal coordinate mapping matrix.

[0066] Specifically, meteorological (wind speed, precipitation) and hydrological (flow velocity, water level) data typically come from different monitoring networks with inconsistent spatiotemporal resolutions. Spatiotemporal alignment requires unifying the time reference (e.g., interpolation to an hourly scale) and the spatial reference (e.g., unification to the target plot grid), and data fusion is achieved by constructing a spatiotemporal coordinate mapping matrix.

[0067] S3. Perform feature extraction and fusion on the preprocessed multi-source data to obtain fused features.

[0068] In one optional implementation, step S3 includes:

[0069] S31. Construct a multi-scale spatiotemporal feature pyramid and use a 3D-CNN network to extract the spatial features of the preprocessed satellite remote sensing data.

[0070] Specifically, the Multi-Scale Spatiotemporal Feature Pyramid (STFPN) captures spatiotemporal patterns of different granularities in satellite remote sensing data through hierarchical feature extraction. 3D-CNN employs an improved ResNeXt structure, dividing the input channels into multiple groups, each undergoing independent convolution operations. This reduces the number of parameters while enhancing feature diversity. Skip connections are used in residual connections to mitigate gradient vanishing. Furthermore, a Squeeze-and-Excitation (SE) module is embedded after the convolutional layers to capture channel-level importance through global average pooling.

[0071] S32. Use graph attention networks to extract spatial correlation features from environmental monitoring data.

[0072] Specifically, Graph Attention Network (GAT) captures the spatial dependencies between environmental monitoring stations by dynamically learning the weights between nodes. GAT constructs a graph of monitoring station relationships based on geographically weighted regression.

[0073] S33. Capture the time-dependent characteristics of meteorological and hydrological data through the Transformer encoder.

[0074] S34. A feature alignment network is used to fuse the spatial features of satellite remote sensing data, the spatial correlation features of environmental monitoring data, and the temporal dependence features of meteorological and hydrological data across modes. The fusion weights are dynamically adjusted through an adaptive gating mechanism.

[0075] The core objective of feature alignment networks is to eliminate the semantic gap between features of different modalities and map them into a shared latent space.

[0076] Specifically, for spatial feature alignment of satellite remote sensing data, networks such as 3D-CNN are used to extract spatial features from remote sensing images. These features may include information such as the shape, texture, and spatial distribution of ground features. Through feature alignment networks, these spatial features are mapped to a shared space, thereby learning the mapping relationship between spatial features of satellite remote sensing data and other modal features, ensuring semantic consistency of spatial features from different modalities.

[0077] For spatial correlation alignment of environmental monitoring data: Models such as Graph Attention Networks (GAT) are used to capture the spatial correlations between environmental monitoring stations. These correlations may reflect the spatial diffusion and transport patterns of pollutants. Feature alignment networks are applied to the features extracted by GAT, mapping these spatial correlation features to a shared space identical to the spatial features of satellite remote sensing data. In this way, spatial correlation features from different modalities can be compared and fused within the same semantic space.

[0078] For temporal dependency feature alignment of meteorological and hydrological data, models such as the Transformer encoder are used to capture these features. These features may include the temporal variation patterns of meteorological conditions and the impact of these variations on pollutant dispersion. A feature alignment network is applied to the features extracted by the Transformer, mapping these temporal dependency features into a shared space. In this way, temporal dependency features from different modalities can be fused within the same semantic space.

[0079] Secondly, based on the aforementioned feature alignment, it is necessary to further fuse these heterogeneous features. However, the importance of different modal features may dynamically change with time and scenario. An adaptive gating mechanism can dynamically generate fusion weights based on the input features, achieving intelligent feature fusion. Specifically, the spatial features of the feature-aligned satellite remote sensing data, the temporal dependency features of the meteorological and hydrological data, and the temporal dependency features of the meteorological and hydrological data are input into the adaptive gating mechanism to calculate the fusion weight of each modal feature; then, the fusion weights of each modal feature are weighted and summed to obtain the fused feature.

[0080] S4. Input the multi-source data into the dynamic pollution diffusion model to obtain the pollutant concentration field of the target site.

[0081] In one alternative implementation, the dynamic pollution diffusion model is as follows:

[0082]

[0083] σ y = 0.1x(1+0.0001x) -0.5

[0084] σ z = 0.06x(1+0.0015x) -0.5

[0085] Where C(x,y,z) represents the pollutant concentration at the pollution source at a distance x downwind, y crosswind, and z vertically, in g / m³. 3 Q represents the pollution source strength, in kg / s; u represents the average wind speed, in m / s; H represents the effective emission height of the pollution source; σy σ is the horizontal diffusion parameter. z This is the vertical diffusion parameter.

[0086] Among these, the pollution source strength Q can be obtained from the Internet of Things sensor network (i.e., from environmental monitoring data); the wind speed u is obtained by accessing the meteorological bureau's API (i.e., from meteorological and hydrological data); and the effective emission height H of the pollution source refers to the total height at which pollutants affect atmospheric diffusion after actual emission. This height is composed of the chimney's own geometric height (physical height) and the rise height of the flue gas due to dynamic and thermal factors, and can be obtained through actual measurement.

[0087] S5. The fused data and pollutant concentration field are input into the pre-trained land pollutant concentration prediction model for prediction, so as to obtain the predicted value of pollutant concentration of the target land in the future target time period; the land pollutant concentration prediction model is determined based on LSTM neural network and CNN neural network.

[0088] In one optional implementation, the land parcel pollutant concentration prediction model is a spatiotemporal parallel dual-stream network structure, comprising an LSTM temporal encoding stream and a CNN spatial encoding stream; the LSTM temporal encoding stream adopts a structure combining gated recurrent units and an attention mechanism; the CNN spatial encoding stream uses a residual shrinking network to process high-dimensional spatial features.

[0089] Set up a cross-modal fusion layer for spatiotemporal feature interaction, and configure a transfer learning interface at the output end.

[0090] Specifically, the LSTM time-series coding stream employs a structure combining gated recurrent units (GRUs) and an attention mechanism. As a variant of LSTM, the GRU effectively captures long-term dependencies in time-series data through its gating mechanism, which is crucial for time-series tasks such as pollutant concentration prediction. Simultaneously, the introduction of the attention mechanism allows the model to focus on information at key time points in the time series, ignoring redundant information, further improving the accuracy of time-series coding and thus enhancing prediction accuracy.

[0091] Secondly, the CNN spatial encoding stream uses a residual shrinking network (RSN) to process high-dimensional spatial features and extract more refined spatial patterns, which is crucial for spatial tasks such as pollutant concentration prediction. The residual shrinking network effectively addresses the gradient vanishing and overfitting problems in deep neural networks by introducing residual connections and a shrinking mechanism, thereby improving the model's generalization ability.

[0092] Subsequently, a cross-modal fusion layer is set up to facilitate spatiotemporal feature interaction. This layer is responsible for fusing the spatiotemporal features extracted from the LSTM temporal encoding stream and the CNN spatial encoding stream, achieving complementarity and enhancement of spatiotemporal information, thereby improving the accuracy of pollutant concentration prediction. Simultaneously, through the interaction and fusion of spatiotemporal features, the model can capture more complex pollutant diffusion patterns, further improving prediction performance.

[0093] Finally, the output end is configured with a transfer learning interface, which can support the transfer application of the model across different plots or scenarios. By fine-tuning the pre-trained model, it can quickly adapt to new prediction tasks, thereby improving the model's generalization ability and further reducing deployment costs and time.

[0094] S6. Compare the predicted pollutant concentration of the target plot within the target time period with the warning thresholds at all levels, and generate corresponding warning instructions based on the comparison results.

[0095] In one optional implementation, step S6 includes:

[0096] S61. Establish a multi-level threshold system; the multi-level threshold system includes: Level IV blue warning, Level III yellow warning, Level II orange warning and Level I red warning, and the thresholds of each level correspond to θ4, θ3, θ2 and θ1 respectively.

[0097] The thresholds for warnings at all levels can be determined by combining historical data clustering analysis with expert experience, or they can be dynamically adjusted based on meteorological and hydrological data (wind speed, precipitation, river flow velocity). No specific limitations are made here.

[0098] S62. Compare the predicted pollutant concentration of the target plot at the target time with the threshold values ​​of each level of early warning.

[0099] S63. When the predicted pollutant concentration falls within the range of [θ4, θ3), a Level IV blue alert is triggered at the target time; when the predicted pollutant concentration falls within the range of [θ3, θ2), a Level III yellow alert is triggered at the target time; when the predicted pollutant concentration falls within the range of [θ2, θ1) and continues to exceed a preset time period, a Level II orange alert is triggered at the time corresponding to the preset time period at the target time; when the predicted pollutant concentration falls within the range of [θ1, +∞) and the diffusion range covers sensitive targets, a Level I red alert is triggered at the time corresponding to the preset time period at the target time.

[0100] S64. Generate a warning instruction corresponding to the determined warning level.

[0101] In an optional implementation, step S63 further includes:

[0102] When the predicted pollutant concentration fluctuates between the thresholds of the two-level warning, the warning level is determined by combining the growth or decline trend of the predicted pollutant concentration of the target plot in the future target time period output by the plot pollutant concentration prediction model.

[0103] S7. Generate corresponding early warning information based on the early warning instruction; and execute corresponding early warning measures based on the early warning information.

[0104] Examples are shown in Table 1:

[0105] Table 1

[0106]

[0107] like Figure 2 As shown, the present invention also provides a land pollution monitoring, prediction and early warning system, comprising:

[0108] Data acquisition module 201 is used to acquire multi-source data of the target site; the multi-source data includes satellite remote sensing data, environmental monitoring data, and meteorological and hydrological data;

[0109] Data preprocessing module 202 is used to preprocess the multi-source data;

[0110] The feature fusion module 203 is used to extract and fuse features from the preprocessed multi-source data to obtain fused features.

[0111] The pollutant concentration calculation module 204 is used to input the multi-source data into the dynamic pollution diffusion model to obtain the pollutant concentration field of the target site;

[0112] The pollutant concentration prediction module 205 is used to input the fused data and the pollutant concentration field into the pre-trained land pollutant concentration prediction model for prediction, so as to obtain the predicted value of pollutant concentration of the target land in the future target time period; the land pollutant concentration prediction model is determined based on LSTM neural network and CNN neural network;

[0113] The early warning level determination module 206 is used to compare the predicted pollutant concentration of the target plot within a future target time period with the early warning thresholds at various levels, and to generate corresponding early warning instructions based on the comparison results; 201

[0114] The early warning measure determination module 207 is used to generate corresponding early warning information according to the early warning instruction; and to execute corresponding early warning measures according to the early warning information.

[0115] In summary, the site pollution monitoring, prediction, and early warning system and method of this invention integrates multi-source data from the site and its surrounding environment to construct a dynamic pollution diffusion model, a site pollutant concentration prediction model, and a multi-level early warning mechanism, thereby achieving scientific classification and precise control of pollution risks. Covering data acquisition, spatiotemporal analysis, intelligent prediction, and decision support, it can significantly improve the timeliness and accuracy of pollution early warning, providing key technical support for environmental management departments.

[0116] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A method for monitoring, predicting, and warning of pollution in a land parcel, characterized in that, include: S1. Acquire multi-source data of the target site; the multi-source data includes satellite remote sensing data, environmental monitoring data, and meteorological and hydrological data; S2. Preprocess the multi-source data; S3. Perform feature extraction and fusion on the preprocessed multi-source data to obtain fused features; S4. Input the preprocessed multi-source data into the dynamic pollution diffusion model to obtain the pollutant concentration field of the target site; S5. The fused features and pollutant concentration field are input into the pre-trained land plot pollutant concentration prediction model for prediction, to obtain the predicted pollutant concentration value of the target land plot in the future target time period; the land plot pollutant concentration prediction model is determined based on LSTM neural network and CNN neural network; S6. Compare the predicted pollutant concentration of the target plot within the target time period with the warning thresholds at all levels, and generate warning instructions of the corresponding level based on the comparison results. S7. Generate corresponding warning information according to the warning instruction; And based on the aforementioned warning information, execute the corresponding warning measures.

2. The method according to claim 1, characterized in that, S2 includes: S21. Perform radiometric and geometric corrections on satellite remote sensing data; S22. Perform missing value interpolation and outlier detection on environmental monitoring data, and adopt time-frequency domain joint denoising based on wavelet transform; S23. Perform spatiotemporal alignment processing on meteorological and hydrological data and establish a spatiotemporal coordinate mapping matrix.

3. The method according to claim 2, characterized in that, S3 includes: S31. Construct a multi-scale spatiotemporal feature pyramid and use a 3D-CNN network to extract the spatial features of the preprocessed satellite remote sensing data. S32. Utilize graph attention networks to extract spatial correlation features from environmental monitoring data; S33. Capture the temporal dependence characteristics of meteorological and hydrological data using a Transformer encoder; S34. A feature alignment network is used to fuse the spatial features of satellite remote sensing data, the spatial correlation features of environmental monitoring data, and the temporal dependence features of meteorological and hydrological data across modes. The fusion weights are dynamically adjusted through an adaptive gating mechanism.

4. The method according to claim 1, characterized in that, The dynamic pollution diffusion model is as follows: in, The distance of the pollution source downwind x Crosswind distance y Vertical height z Pollutant concentration at the location, in g / m³ 3 ; The pollution source strength is expressed in kg / s. u The wind speed is the average wind speed, expressed in m / s. H The effective emission height of the pollution source; For horizontal diffusion parameters; This is the vertical diffusion parameter.

5. The method according to claim 1, characterized in that, The land plot pollutant concentration prediction model is a spatiotemporal parallel dual-stream network structure, which includes an LSTM temporal coding stream and a CNN spatial coding stream. The LSTM temporal coding stream adopts a structure that combines gated recurrent units with an attention mechanism. The CNN spatial coding stream uses a residual shrinking network to process high-dimensional spatial features. Set up a cross-modal fusion layer for spatiotemporal feature interaction, and configure a transfer learning interface at the output end.

6. The method according to claim 1, characterized in that, S6 includes: S61. Establish a multi-level threshold system; the multi-level threshold system includes: Level IV blue alert, Level III yellow alert, Level II orange alert, and Level I red alert, with each level's threshold corresponding to... θ 4. θ 3. θ 2 and θ 1; S62. Compare the predicted pollutant concentration of the target plot at the target time with the threshold values ​​of each level of early warning. S63, When the predicted pollutant concentration value belongs to When the target time is determined within the specified range, a Level IV blue alert is triggered; when the predicted pollutant concentration falls within the specified range... When the target time is determined within the specified range, a Level III yellow alert will be triggered; when the predicted pollutant concentration falls within the specified range... When the predicted pollutant concentration falls within the range and continues to exceed the preset time period, a Level II orange alert is triggered at the time corresponding to the preset time period after the target time has elapsed; when the predicted pollutant concentration falls within the range... When the scope and diffusion range cover sensitive targets, a Level I red alert is triggered at the time corresponding to the time when the target's time has passed through a preset time period. S64. Generate a warning instruction corresponding to the determined warning level.

7. The method according to claim 6, characterized in that, S63 further includes: When the predicted pollutant concentration fluctuates between the thresholds of the two-level warning, the warning level is determined by combining the growth or decline trend of the predicted pollutant concentration of the target plot in the future target time period output by the plot pollutant concentration prediction model.

8. A site pollution monitoring, prediction, and early warning system, characterized in that, include: The data acquisition module is used to acquire multi-source data of the target site; the multi-source data includes satellite remote sensing data, environmental monitoring data, and meteorological and hydrological data. The data preprocessing module is used to preprocess the multi-source data; The feature fusion module is used to extract and fuse features from preprocessed multi-source data to obtain fused features. The pollutant concentration calculation module is used to input preprocessed multi-source data into the dynamic pollution diffusion model to obtain the pollutant concentration field of the target site. The pollutant concentration prediction module is used to input the fused features and the pollutant concentration field into a pre-trained land pollutant concentration prediction model to make predictions and obtain the predicted pollutant concentration values ​​of the target land within a future target time period; the land pollutant concentration prediction model is determined based on LSTM neural network and CNN neural network; The warning level determination module is used to compare the predicted pollutant concentration of the target plot in the future target time period with the warning thresholds of each level, and generate warning instructions of the corresponding level based on the comparison results. The early warning measure determination module is used to generate corresponding early warning information based on the early warning instruction; And based on the aforementioned warning information, execute the corresponding warning measures.

Images