A method for simulating agricultural non-point source pollution at a watershed scale

Abstract

The application relates to the technical field of non-point source pollution and discloses a method for simulating agricultural non-point source pollution at a river basin scale, which comprises the following steps: obtaining comprehensive pollution data of each pollution simulation area, constructing a pollution sequence of a to-be-simulated river basin area according to all the comprehensive pollution data, determining a pollution index value of each comprehensive pollution data on the pollution sequence based on a pollution model, splitting the pollution sequence based on the pollution index value, extracting different types of comprehensive pollution data on a pollution direct sequence and determining an association result of each pollution simulation area based on an association rule algorithm, constructing a pollution association area and a non-pollution association area based on the association result, constructing a precipitation prediction model based on historical meteorological data, determining a precipitation simulation trend according to a precipitation time sequence diagram, and respectively performing pollution evolution analysis on the pollution association area and the non-pollution association area based on the precipitation simulation trend. The application ensures the timeliness and accuracy of pollution evolution simulation.

Description

Technical Field

[0001] This invention relates to the field of non-point source pollution technology, and more specifically, to a watershed-scale agricultural non-point source pollution simulation method. Background Technology

[0002] With the rapid development of agriculture, agricultural non-point source pollution at the watershed scale has become increasingly prominent. Agricultural non-point source pollution is characterized by dispersed pollution sources, complex emission paths, and uneven spatial and temporal distribution. It mainly originates from the excessive application of chemical fertilizers and pesticides in farmland, rural domestic sewage, and disorderly treatment of solid waste. These pollutants enter water bodies through surface runoff and ground infiltration, leading to eutrophication and water quality deterioration in watersheds, thereby threatening the balance of the ecosystem.

[0003] Currently, existing simulation methods typically treat the watershed as a whole, making it difficult to accurately reflect the differences and correlations in pollution generation in different areas. This results in an inability to effectively simulate the complex environment of pollution diffusion. On the other hand, in terms of pollution treatment, they fail to effectively distinguish the impact of different factors on pollution, and lack dynamic prediction of precipitation trends. Consequently, the timeliness and accuracy of pollution evolution simulation are insufficient, making it difficult to meet the needs of precise prevention and control of watershed pollution and corresponding management decisions.

[0004] Therefore, it is necessary to design a watershed-scale agricultural non-point source pollution simulation method to address the problems existing in current technologies. Summary of the Invention

[0005] In view of this, the present invention proposes a watershed-scale agricultural non-point source pollution simulation method, which aims to solve the problems of insufficient timeliness and accuracy in pollution evolution simulation, such as difficulty in accurately reflecting the differences and correlations of pollution generation in different regions, inability to effectively simulate the complex environment of pollution diffusion, failure to effectively distinguish the impact of different factors on pollution, and lack of dynamic prediction of precipitation trends.

[0006] This invention proposes a watershed-scale agricultural non-point source pollution simulation method, comprising:

[0007] The watershed area to be simulated is determined, and the watershed area to be simulated is divided into several pollution simulation areas. The comprehensive pollution data of each pollution simulation area is obtained, and the pollution sequence of the watershed area to be simulated is constructed based on all the comprehensive pollution data.

[0008] Based on the pollution model, the pollution index value of each comprehensive pollution data in the pollution sequence is determined. Based on the pollution index value, the pollution sequence is split to determine the direct pollution sequence and the indirect pollution sequence. Different types of comprehensive pollution data are extracted from the direct pollution sequence and the association result of each pollution simulation area is determined based on the association rule algorithm. Based on the association result, pollution association area and non-pollution association area are constructed.

[0009] Historical meteorological data of the watershed area to be simulated are obtained, a precipitation prediction model is constructed based on the historical meteorological data, a precipitation time series map is constructed based on the precipitation prediction model, and the precipitation simulation trend is determined based on the precipitation time series map.

[0010] Based on the precipitation simulation trends, pollution evolution analysis was conducted in both pollution-associated and non-pollution-associated areas.

[0011] Furthermore, when acquiring comprehensive pollution data for each pollution simulation area, the following are included:

[0012] The data sources for each pollution simulation area are determined, including authoritative sources, third-party sources, and non-authoritative sources.

[0013] Obtain the original comprehensive pollution data from the data source, perform data tracing and error rate statistics, determine the original comprehensive pollution data that meets the conditions for data tracing and error rate statistics as the basic verification data, process the missing values ​​of the basic verification data, and determine the comprehensive pollution data based on the results of the missing value processing.

[0014] Furthermore, when determining the pollution index value for each comprehensive pollution data point in the pollution sequence based on the pollution model, the process includes:

[0015] Obtain a historical dataset and divide it into a training set and a test set. Determine the model building parameters based on grid search and build a random forest model. Iterate and train the random forest model using the training set. Test the iteratively trained random forest model using the test set to determine the contaminated model.

[0016] Furthermore, when determining the pollution index value for each comprehensive pollution data point in the pollution sequence based on the pollution model, the method also includes:

[0017] If the mean squared error of the random forest model after the current iteration is greater than or equal to the mean squared error of the random forest model after the previous iteration, then reduce the depth of the tree and the number of leaf nodes, and continue iterative training until the mean squared error of the random forest model after the current iteration is less than the mean squared error of the random forest model after the previous iteration.

[0018] If the mean squared error of the random forest model after the current iteration is less than the mean squared error of the random forest model after the previous iteration, then stop the iteration training and determine the random forest model after the current iteration as the contaminated model.

[0019] Substitute each comprehensive pollution data point in the pollution sequence into the pollution model to determine the corresponding pollution index value.

[0020] Furthermore, when splitting the pollution sequence based on the pollution index value to determine the direct pollution sequence and the indirect pollution sequence, the process includes:

[0021] Compare each pollution index value in the pollution sequence with the pollution index threshold;

[0022] Extract all pollution index values ​​greater than or equal to the pollution index threshold from the pollution sequence, and generate the pollution direct sequence;

[0023] Extract all pollution index values ​​less than the pollution index threshold from the pollution sequence and generate the pollution indirect sequence.

[0024] Furthermore, when extracting comprehensive pollution data of different types from the direct pollution sequence and determining the association results for each pollution simulation region based on the association rule algorithm, and constructing pollution-associated regions and non-pollution-associated regions based on the association results, the process includes:

[0025] Several candidate item sets are generated according to the Eclat algorithm. Frequent itemsets are determined based on the support of the candidate item sets. The association results of each pollution simulation region with other pollution simulation regions are determined based on the frequent itemsets, and the non-association results are determined.

[0026] All pollution simulation regions with correlated results are defined as pollution-correlated regions, and all pollution simulation regions with uncorrelated results are defined as non-pollution-correlated regions.

[0027] Furthermore, when acquiring historical meteorological data for the watershed area to be simulated and constructing a precipitation prediction model based on the historical meteorological data, the process includes:

[0028] The historical meteorological data includes historical radar echo data, historical precipitation data, historical lightning data, and historical water vapor content. A precipitation prediction model with a convolutional neural network architecture is constructed based on the historical meteorological data.

[0029] Furthermore, when constructing a precipitation prediction model based on the historical meteorological data using a convolutional neural network architecture, the following steps are included:

[0030] The input layer of the precipitation prediction model receives the historical meteorological data, extracts the spatial distribution features of historical radar echo data and historical lightning data based on 3×3 convolution kernels, and extracts the numerical variation features of historical precipitation data and historical water vapor content based on 1×1 convolution kernels. The feature fusion layer integrates the outputs of several branch channels and forms multi-scale features based on BatchNorm and LeakyReLU activation. The multi-scale features are compressed based on a 2×2 max pooling layer.

[0031] The output layer of the precipitation prediction model outputs the simulated precipitation time and the corresponding simulated precipitation amount based on the compressed multi-scale features and a fully connected layer containing one neuron.

[0032] Furthermore, when constructing a precipitation time series map based on the precipitation prediction model and determining the precipitation simulation trend based on the precipitation time series map, the process includes:

[0033] The simulated precipitation time, simulated precipitation value, and historical precipitation data are used to construct a precipitation time series graph. The slope between the simulated precipitation time and the previous historical time is extracted from the precipitation time series graph, and the precipitation simulation trend is determined based on the slope.

[0034] Furthermore, when conducting pollution evolution analysis on pollution-associated and non-pollution-associated areas based on the aforementioned precipitation simulation trends, the analysis includes:

[0035] The pollution evolution was simulated based on the SWAT model to analyze the precipitation simulation trend and the pollution-related areas.

[0036] Based on the HSPF model, the precipitation simulation trend and pollution evolution of non-pollution-associated areas were simulated.

[0037] Compared with existing technologies, the advantages of this invention are as follows: By dividing the watershed to be simulated into several pollution simulation areas and constructing pollution sequences, the differences in pollution in different areas are accurately captured, providing precise basic data for the simulation and improving the regional specificity and data accuracy of the simulation. Based on the pollution model, the pollution sequences are split and pollution-related and non-pollution-related areas are constructed using association rule algorithms, effectively distinguishing the direct and indirect effects of pollution and clarifying the role of different factors in pollution and the pollution correlation between regions. This helps to accurately identify pollution factors and propagation paths, ensuring the stability and reliability of the simulation process. A precipitation prediction model is constructed based on historical meteorological data, and the precipitation simulation trend is determined, ensuring the accuracy of dynamic precipitation prediction. Combined with regional division and correlation results, the impact of different precipitation on the pollution evolution of different pollution simulation areas is dynamically reflected, ensuring the timeliness and accuracy of pollution evolution simulation and improving the reliability of the simulation process in conforming to actual environmental changes. Attached Figure Description

[0038] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0039] Figure 1 A flowchart of a watershed-scale agricultural non-point source pollution simulation method provided in an embodiment of the present invention. Detailed Implementation

[0040] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the disclosure to those skilled in the art. It should be noted that, unless otherwise specified, embodiments and features in the embodiments of the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0041] In some embodiments of this application, see Figure 1 As shown, a watershed-scale agricultural non-point source pollution simulation method includes:

[0042] S100: Determine the watershed area to be simulated, and divide the watershed area into several pollution simulation areas. Obtain comprehensive pollution data for each pollution simulation area, and construct the pollution sequence of the watershed area to be simulated based on all comprehensive pollution data.

[0043] S200: Based on the pollution model, determine the pollution index value of each comprehensive pollution data in the pollution sequence, split the pollution sequence based on the pollution index value, determine the direct pollution sequence and the indirect pollution sequence, extract different types of comprehensive pollution data in the direct pollution sequence, and determine the association result of each pollution simulation area based on the association rule algorithm, and construct pollution association area and non-pollution association area based on the association result.

[0044] S300: Obtain historical meteorological data of the watershed area to be simulated, construct a precipitation prediction model based on the historical meteorological data, construct a precipitation time series map based on the precipitation prediction model, and determine the precipitation simulation trend based on the precipitation time series map.

[0045] S400: Based on precipitation simulation trends, pollution evolution analysis is performed on both pollution-associated and non-pollution-associated areas.

[0046] Specifically, the first step is to determine the watershed area to be simulated and divide it into several pollution simulation areas. The number and size of these areas can be adjusted based on the actual size of the watershed. Comprehensive pollution data for each simulation area is then acquired to construct a pollution sequence. Agricultural non-point source pollution is characterized by dispersed sources and uneven spatial and temporal distribution. Different simulation areas exhibit variations in agricultural fertilizer and pesticide use, rural domestic sewage treatment, and solid waste disposal, leading to different pollution conditions. Simulating the entire watershed as a single entity fails to capture these differences. However, by dividing the watershed into simulation areas and collecting comprehensive pollution data, the pollution characteristics of each area can be accurately captured, providing detailed foundational data for subsequent simulations and avoiding the limitations of holistic simulation. Based on the pollution model, the pollution index values ​​for each comprehensive pollution data point in the pollution sequence are determined, further separating the direct and indirect pollution sequences. The pollution sequence, comprehensively analyzed using topographic, land use, soil, and hydrological monitoring data (concentrations of nitrogen, phosphorus, dissolved phosphorus, and phosphate), is analyzed. Since pollution generation and propagation are influenced by multiple factors, including both direct and indirect pollution factors affecting the watershed, the pollution sequence is broken down to determine direct and indirect pollution sequences. Furthermore, pollution in different simulated areas may interact. By determining the direct pollution sequence and using association rule algorithms to identify the association results for each simulated area, the propagation path and scope of pollution can be clearly identified, ensuring the relevance and reliability of the pollution status simulation for the watershed area. Historical meteorological data for the simulated watershed area are obtained to construct precipitation prediction models and precipitation time-series maps to determine precipitation simulation trends. This is because precipitation plays a role in the diffusion and migration of agricultural non-point source pollution through surface runoff and groundwater infiltration. Ignoring the impact of precipitation on non-point source pollution would reduce the reliability and timeliness of the pollution evolution simulation.

[0047] Understandably, precipitation prediction models built from historical meteorological data can dynamically predict precipitation patterns, while precipitation time-series maps reflect precipitation changes in the simulated watershed area. Determining precipitation simulation trends based on these time-series maps provides crucial environmental parameters for simulating pollution evolution. Pollution evolution analysis is conducted separately for polluted and non-polluted areas based on these trends because the way and extent to which precipitation affects pollution diffusion differs between these regions. Pollution in polluted areas interacts with each other; precipitation washes away pollutants, exacerbating the diffusion caused by this interaction. Pollution in non-polluted areas, however, is relatively independent, and precipitation washes away pollutants to a certain extent. Therefore, accurately simulating the pollution evolution process under the influence of precipitation in both polluted and non-polluted areas ensures the timeliness and accuracy of the pollution evolution simulation. By dividing the area into several pollution simulation zones and determining the direct and indirect pollution sequences, the correlation results between pollution-associated and non-pollution-associated zones are dynamically quantified. This accurately reflects the differences and correlations in pollution generation in different pollution simulation zones, effectively simulating the complex environment of pollution diffusion. Furthermore, by determining the precipitation simulation trend based on a precipitation prediction model, the timeliness and accuracy of pollution evolution simulation are improved, providing a basis for precise prevention and control of watershed pollution.

[0048] In some embodiments of this application, when acquiring comprehensive pollution data for each pollution simulation area, the process includes: determining the data source for each pollution simulation area, including authoritative sources, third-party sources, and non-authoritative sources; acquiring the original comprehensive pollution data from the data sources; performing data tracing and error rate statistics; determining the original comprehensive pollution data that meets the conditions for data tracing and error rate statistics as basic verification data; processing the missing values ​​in the basic verification data; and determining the comprehensive pollution data based on the results of the missing value processing.

[0049] Specifically, when acquiring comprehensive pollution data for each simulated pollution area, raw comprehensive pollution data is first obtained from authoritative sources, third-party departments, and non-authoritative sources. Authoritative sources, such as environmental protection and water resources departments, provide data with a certain degree of officiality and systematicity. Third-party departments, such as research institutions, provide data with strong professional expertise. Non-authoritative sources include grassroots village-level self-governing organizations (rough records from village committees based on simple equipment), small agricultural cooperatives (scattered monitoring conducted independently), and individual monitoring enthusiasts (data collected using non-professional equipment). This data is usually recorded for non-professional purposes. After obtaining the raw aggregate pollution data, data tracing and error rate statistics are performed. Data from different sources varies in quality; directly using unfiltered data may lead to distorted simulation results due to errors or unreliable sources. Data tracing tracks the entire data generation process, clarifying the collection methods and monitoring equipment. Error rate statistics quantify the degree of data deviation. Combining these two methods allows for the selection of basic validation data that meets quality requirements. If the obtained raw aggregate pollution data cannot be traced, or if the allowable error range of the monitoring equipment cannot be guaranteed, then the conditions for data tracing or error rate statistics cannot be met. Basic validation data must be raw aggregate pollution data that simultaneously meets the conditions for data tracing and error rate statistics. The monitoring environment for agricultural non-point source pollution data is complex. Missing values ​​in the basic validation data are handled through interpolation and imputation to improve the data and avoid data deviations that could affect the stability of the simulation.

[0050] In some embodiments of this application, when determining the pollution index value of each comprehensive pollution data in the pollution sequence based on the pollution model, the process includes: obtaining a historical dataset and dividing the historical dataset into a training set and a test set; determining the model building parameters based on grid search and building a random forest model; iteratively training the random forest model based on the training set; and testing the iteratively trained random forest model based on the test set to determine the pollution model.

[0051] In some embodiments of this application, when determining the pollution index value of each comprehensive pollution data in the pollution sequence based on the pollution model, the method further includes: if the mean squared error of the random forest model after the current iteration training is greater than or equal to the mean squared error of the random forest model after the previous iteration training, then the depth of the tree and the number of leaf nodes are reduced, and iterative training continues until the mean squared error of the random forest model after the iterative training is less than the mean squared error of the random forest model after the previous iteration training; if the mean squared error of the random forest model after the current iteration training is less than the mean squared error of the random forest model after the previous iteration training, then iterative training is stopped, and the random forest model after the current iteration training is determined as the pollution model, and each comprehensive pollution data in the pollution sequence is substituted into the pollution model to determine the corresponding pollution index value.

[0052] Specifically, historical datasets include historical comprehensive pollution data, planting area of ​​different crop types, farming methods, and agricultural irrigation methods. These datasets reflect the seasonal and cyclical changes in pollution sources. The historical dataset is divided into training and testing sets, typically with 70%-80% of the data allocated to the training set and the remainder to the testing set. This ensures the model's generalization ability. Grid search traverses parameter combinations to find the optimal parameters for the model, laying the foundation for building the random forest model. The training set is used to train the random forest model, while the testing set is used to evaluate the model's performance. Random forest models have multiple decision trees designed to capture complex relationships in the data for ensemble learning, thereby reducing the risk of overfitting and improving prediction stability. The random forest model is iteratively trained using data from the training set. In each iteration, the model attempts to learn patterns and relationships in the data to improve its prediction or classification ability. After each training iteration, the model is tested using data from the test set to determine the mean squared error (MSE). The MSE is used to measure the model's performance. During iterative training, the depth of the tree and the number of leaf nodes are dynamically adjusted based on the MSE. If the MSE after the current iteration is greater than or equal to the MSE after the previous iteration, it indicates that the model may be overfitting (the tree is too deep and there are too many leaf nodes, leading to excessive learning of noise). In this case, the tree depth and the number of leaf nodes need to be reduced, and iterative training continues. This helps the model stably approach the global optimum. If the MSE after the current iteration is less than the MSE after the previous iteration, it indicates that the model has reached optimal generalization ability. At this point, iterative training stops, and the random forest model trained in the current iteration is identified as a contaminated model. By continuously training the random forest model, the contamination index value for each comprehensive contaminated data can be accurately output, avoiding judgment errors and thus improving the efficiency of the simulation.

[0053] In some embodiments of this application, when splitting a contaminated sequence based on contamination index values ​​to determine direct and indirect contamination sequences, the process includes: comparing each contamination index value in the contamination sequence with a contamination index threshold, extracting all contamination index values ​​in the contamination sequence that are greater than or equal to the contamination index threshold, and generating a direct contamination sequence; and extracting all contamination index values ​​in the contamination sequence that are less than the contamination index threshold, and generating an indirect contamination sequence.

[0054] Specifically, based on the pollution model, the pollution index values ​​for each piece of comprehensive pollution data are output. This comprehensive pollution data includes topographic data, land use data, soil data, and hydrological monitoring data. Pollution sequences are then divided into direct and indirect pollution sequences using pollution index thresholds. The pollution index threshold serves as a critical value for judging the intensity of pollution impact. When a pollution index value is greater than or equal to the threshold, it indicates that the pollution impact corresponding to that data is relatively direct, thus forming a direct pollution sequence. Conversely, when the value is less than the threshold, the pollution contribution is weak and pollution may occur through indirect pathways, thus forming an indirect pollution sequence. Generally, topographic data (such as slope and altitude), land use data (such as the proportion of arable land and the distribution of residential areas), and soil data (such as soil texture and organic matter content) are not themselves direct pollutants, but rather influence the generation and migration of pollution indirectly. For example, steep terrain accelerates the carrying of pollutants by surface runoff, the proportion of arable land determines the scale of fertilizer and pesticide application, and soil adsorption capacity affects the infiltration rate of pollutants. However, these data need to be transformed into a quantitative impact on pollution through complex reasoning. In contrast, the concentrations of nitrogen, phosphorus, and other pollutants in hydrological monitoring data are direct manifestations of pollutants in water bodies, and their values ​​can directly reflect the degree of pollution. Therefore, based on pollution index thresholds, the difference between these values ​​and the pollution contribution can be accurately distinguished, providing a data foundation for targeted simulation analysis and ensuring the stability of the simulation.

[0055] In some embodiments of this application, when extracting comprehensive pollution data of different types from the direct pollution sequence and determining the association results of each pollution simulation region based on the association rule algorithm, and constructing pollution-associated regions and non-pollution-associated regions based on the association results, the process includes: generating several candidate itemsets according to the Eclat algorithm, determining frequent itemsets based on the support of the candidate itemsets, determining the association results of each pollution simulation region with other pollution simulation regions based on the frequent itemsets, and determining non-association results, identifying all pollution simulation regions with associated results as pollution-associated regions, and identifying all pollution simulation regions with non-association results as non-pollution-associated regions.

[0056] Specifically, the Eclat algorithm is used to uncover potential correlations between different types of integrated pollution data and corresponding pollution simulation areas in direct pollution sequences. For example, nitrogen data from hydrological monitoring in one pollution simulation area and phosphorus data from hydrological monitoring in another pollution simulation area may have a potential correlation. The Eclat algorithm stores integrated pollution data from direct pollution sequences in a vertical data format, generating candidate itemsets containing pollution features from multiple regions. It then calculates the support of these itemsets in the data (i.e., the frequency of co-occurrence), filtering out frequent itemsets with support higher than a support threshold. These frequent itemsets reflect the co-occurrence patterns of pollution features between regions, thereby determining the correlation results (frequent co-occurrence) and non-correlation results (no frequent co-occurrence) between regions, and further dividing pollution-related and non-pollution-related regions. The Eclat algorithm efficiently uncovers hidden correlations in the data, avoiding the blindness of complex and intertwined pollution extrapolation between pollution simulation areas, improving the accuracy of the simulation. Through correlation analysis, it objectively reveals the linkage patterns between various pollution simulation areas, providing direction for pollution source tracing and improving the reliability and timeliness of non-point source pollution simulation.

[0057] In some embodiments of this application, when acquiring historical meteorological data of the watershed area to be simulated and constructing a precipitation prediction model based on the historical meteorological data, the following steps are taken: the historical meteorological data includes historical radar echo data, historical precipitation data, historical lightning data, and historical water vapor content, and a precipitation prediction model with a convolutional neural network architecture is constructed based on the historical meteorological data.

[0058] In some embodiments of this application, when constructing a precipitation prediction model based on historical meteorological data using a convolutional neural network architecture, the following steps are included: the input layer of the precipitation prediction model receives historical meteorological data, extracts the spatial distribution features of historical radar echo data and historical lightning data based on 3×3 convolutional kernels, and extracts the numerical variation features of historical precipitation data and historical water vapor content based on 1×1 convolutional kernels; the feature fusion layer integrates the outputs of several branch channels, and forms multi-scale features based on BatchNorm and LeakyReLU activation; the multi-scale features are compressed based on a 2×2 max pooling layer; and the output layer of the precipitation prediction model outputs the simulated precipitation time and the corresponding simulated precipitation value based on the compressed multi-scale features and a fully connected layer containing one neuron.

[0059] Specifically, precipitation formation is the result of the combined effects of spatial factors (such as cloud distribution) and numerical factors (such as humidity). A precipitation prediction model is constructed based on the spatial feature extraction capabilities of a convolutional neural network (CNN) architecture and by fusing multi-dimensional historical meteorological data, ensuring the stability of the simulation. The input layer of the precipitation prediction model receives historical meteorological data. The received historical radar echo data and historical lightning data have spatial distribution characteristics; a 3×3 convolutional kernel captures their regional correlations (such as the spatial gradient of radar echo intensity) through a sliding window. Historical precipitation data and historical water vapor content are sequences of specific values; a 1×1 convolutional kernel can efficiently extract their numerical change trends (such as the rate of humidity accumulation). The feature fusion layer enhances nonlinear fitting capabilities through BatchNorm normalization of data distribution and LeakyReLU activation, and integrates multi-branch features to form a multi-scale representation. A 2×2 max-pooling layer compresses redundant information to preserve features. Finally, a fully connected layer outputs simulated precipitation time and precipitation values, realizing the mapping from historical data to future precipitation. Multi-scale feature fusion improves the spatiotemporal accuracy of precipitation prediction and effectively captures the impact of short-duration heavy precipitation on pollution diffusion. The parallel computing capability based on the convolutional neural network architecture accelerates the training and prediction accuracy of the precipitation prediction model, thus meeting the needs of pollution simulation for dynamic updates of precipitation trends. This provides reliable simulation driving parameters for subsequent pollution evolution analysis and improves the reliability of pollution simulation.

[0060] In some embodiments of this application, when constructing a precipitation time series map based on a precipitation prediction model and determining the precipitation simulation trend based on the precipitation time series map, the method includes: constructing a precipitation time series map by combining the simulated precipitation time, the simulated precipitation value, and historical precipitation data, and extracting the slope between the simulated precipitation time and the previous historical time in the precipitation time series map, and determining the precipitation simulation trend based on the slope.

[0061] Specifically, by combining time-series visualization and slope quantification, the changing trends of precipitation were dynamically analyzed. Simulated precipitation times, simulated precipitation values, and historical precipitation data were integrated along a time axis into a precipitation time-series plot, which visually presents the historical evolution of precipitation and the continuous process of future simulations. The slope between the simulated precipitation time and the previous historical time was extracted, and the positive or negative (increasing / decreasing) slope was used to quantify the precipitation simulation trend. For example, a slope greater than zero indicates that the precipitation simulation trend is strengthening. The precipitation time-series plot clearly shows the continuous dynamic changes of precipitation, and the slope provides a quantitative description of the precipitation simulation trend, improving the temporal accuracy of pollution diffusion simulation. The impact of precipitation simulation trends (such as strengthening or abruptly decreasing) on ​​pollution diffusion is greater than the impact of single-point precipitation on pollution diffusion. If only the numerical value of precipitation is focused on, while its dynamic changes are ignored, it will lead to a disconnect between pollution migration simulation and actual hydrological processes. Quantifying the precipitation simulation trend through slope provides dynamic driving parameters for pollution evolution analysis, ensuring the accuracy and reliability of the simulation.

[0062] In some embodiments of this application, when performing pollution evolution analysis on pollution-associated areas and non-pollution-associated areas based on precipitation simulation trends, the methods include: simulating pollution evolution based on precipitation simulation trends and pollution-associated areas using the SWAT model, and simulating pollution evolution based on precipitation simulation trends and non-pollution-associated areas using the HSPF model.

[0063] Specifically, pollution propagation differs between polluted and non-polluted areas. Polluted areas are influenced by cross-regional hydrology, resulting in targeted pollution diffusion. In contrast, pollution processes in non-polluted areas are relatively closed and dominated by local environmental factors. The SWAT model (Soil and Water Assessment Tool) is a distributed hydrological model based on physical processes, while the HSPF model (Hydrological Simulation Program-Fortran) is a dynamic hydrological-water quality coupling model focusing on short- to medium-term hydrological processes within a watershed. The SWAT model excels at simulating material migration across multiple polluted areas in complex watersheds. It captures the interactions between polluted areas through spatial discretization (such as the impact of upstream pollution on downstream areas via runoff) and combines this with intensity changes in precipitation simulation trends to accurately output the spatiotemporal distribution of pollution in the associated areas. The HSPF model is suitable for refined simulations, focusing on independent pollution generation and transport processes in non-polluting areas (such as fertilizer runoff from local farmland). By combining the targeted processing of the two types of models and coupling the temporal characteristics in the precipitation simulation trend, the adaptability of pollution evolution simulation is improved, thereby outputting targeted pollution evolution results and ensuring the reliability and stability of the simulation.

[0064] In summary, the beneficial effects of this invention are as follows: By dividing the simulated watershed into several pollution simulation areas and constructing pollution sequences, the differences in pollution across different areas are accurately captured, providing precise basic data for the simulation and improving the regional specificity and data accuracy of the simulation. Based on the pollution model, pollution sequences are split and, combined with association rule algorithms, pollution-related and non-pollution-related areas are constructed, effectively distinguishing the direct and indirect impacts of pollution and clarifying the effects of different factors on pollution and the pollution correlations between areas. This helps to accurately identify pollution factors and propagation paths, ensuring the stability and reliability of the simulation process. Based on historical meteorological data, a precipitation prediction model is constructed and precipitation simulation trends are determined, ensuring the accuracy of dynamic precipitation prediction. Combined with regional division and correlation results, the impact of different precipitation events on the pollution evolution of different pollution simulation areas is dynamically reflected, guaranteeing the timeliness and accuracy of pollution evolution simulation and improving the reliability of the simulation process in reflecting actual environmental changes.

[0065] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program goods. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program goods embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0066] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program goods according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0067] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0068] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0069] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A watershed-scale agricultural non-point source pollution simulation method, characterized in that, include: The watershed area to be simulated is determined, and the watershed area to be simulated is divided into several pollution simulation areas. The comprehensive pollution data of each pollution simulation area is obtained, and the pollution sequence of the watershed area to be simulated is constructed based on all the comprehensive pollution data. Based on the pollution model, the pollution index value of each comprehensive pollution data in the pollution sequence is determined. Based on the pollution index value, the pollution sequence is split to determine the direct pollution sequence and the indirect pollution sequence. Different types of comprehensive pollution data are extracted from the direct pollution sequence and the association result of each pollution simulation area is determined based on the association rule algorithm. Based on the association result, pollution association area and non-pollution association area are constructed. Historical meteorological data of the watershed area to be simulated are obtained, a precipitation prediction model is constructed based on the historical meteorological data, a precipitation time series map is constructed based on the precipitation prediction model, and the precipitation simulation trend is determined based on the precipitation time series map. Based on the precipitation simulation trend, pollution evolution analysis was performed on both pollution-associated and non-pollution-associated areas. When extracting comprehensive pollution data of different types from the direct pollution sequence and determining the association results for each pollution simulation region based on the association rule algorithm, and constructing pollution-associated regions and non-pollution-associated regions based on the association results, the process includes: Several candidate item sets are generated according to the Eclat algorithm. Frequent itemsets are determined based on the support of the candidate item sets. The association results of each pollution simulation region with other pollution simulation regions are determined based on the frequent itemsets, and the non-association results are determined. All pollution simulation regions with correlated results are defined as pollution-correlated regions, and all pollution simulation regions with uncorrelated results are defined as non-pollution-correlated regions.

2. The watershed-scale agricultural non-point source pollution simulation method according to claim 1, characterized in that, When acquiring comprehensive pollution data for each pollution simulation area, the following is included: The data sources for each pollution simulation area are determined, including authoritative sources, third-party sources, and non-authoritative sources. Obtain the original comprehensive pollution data from the data source, perform data tracing and error rate statistics, determine the original comprehensive pollution data that meets the conditions for data tracing and error rate statistics as the basic verification data, process the missing values ​​of the basic verification data, and determine the comprehensive pollution data based on the results of the missing value processing.

3. The watershed-scale agricultural non-point source pollution simulation method according to claim 2, characterized in that, When determining the pollution index value for each comprehensive pollution data point in the pollution sequence based on the pollution model, the following steps are included: Obtain a historical dataset and divide it into a training set and a test set. Determine the model building parameters based on grid search and build a random forest model. Iterate and train the random forest model using the training set. Test the iteratively trained random forest model using the test set to determine the contaminated model.

4. The watershed-scale agricultural non-point source pollution simulation method according to claim 3, characterized in that, When determining the pollution index value for each comprehensive pollution data point in the pollution sequence based on the pollution model, the following steps are also included: If the mean squared error of the random forest model after the current iteration is greater than or equal to the mean squared error of the random forest model after the previous iteration, then reduce the depth of the tree and the number of leaf nodes, and continue iterative training until the mean squared error of the random forest model after the current iteration is less than the mean squared error of the random forest model after the previous iteration. If the mean squared error of the random forest model after the current iteration is less than the mean squared error of the random forest model after the previous iteration, then stop the iteration training and determine the random forest model after the current iteration as the contaminated model. Substitute each comprehensive pollution data point in the pollution sequence into the pollution model to determine the corresponding pollution index value.

5. The watershed-scale agricultural non-point source pollution simulation method according to claim 4, characterized in that, When splitting the pollution sequence based on the pollution index value to determine the direct pollution sequence and the indirect pollution sequence, the process includes: Compare each pollution index value in the pollution sequence with the pollution index threshold; Extract all pollution index values ​​greater than or equal to the pollution index threshold from the pollution sequence, and generate the pollution direct sequence; Extract all pollution index values ​​less than the pollution index threshold from the pollution sequence and generate the pollution indirect sequence.

6. The watershed-scale agricultural non-point source pollution simulation method according to claim 5, characterized in that, When acquiring historical meteorological data for the watershed area to be simulated and constructing a precipitation prediction model based on the historical meteorological data, the following steps are included: The historical meteorological data includes historical radar echo data, historical precipitation data, historical lightning data, and historical water vapor content. A precipitation prediction model with a convolutional neural network architecture is constructed based on the historical meteorological data.

7. The watershed-scale agricultural non-point source pollution simulation method according to claim 6, characterized in that, When constructing a precipitation prediction model based on the historical meteorological data using a convolutional neural network architecture, the following steps are included: The input layer of the precipitation prediction model receives the historical meteorological data, extracts the spatial distribution features of historical radar echo data and historical lightning data based on 3×3 convolution kernels, and extracts the numerical variation features of historical precipitation data and historical water vapor content based on 1×1 convolution kernels. The feature fusion layer integrates the outputs of several branch channels and forms multi-scale features based on BatchNorm and LeakyReLU activation. The multi-scale features are compressed based on a 2×2 max pooling layer. The output layer of the precipitation prediction model outputs the simulated precipitation time and the corresponding simulated precipitation amount based on the compressed multi-scale features and a fully connected layer containing one neuron.

8. The watershed-scale agricultural non-point source pollution simulation method according to claim 7, characterized in that, When constructing a precipitation time series map based on the precipitation prediction model and determining the precipitation simulation trend based on the precipitation time series map, the process includes: The simulated precipitation time, simulated precipitation value, and historical precipitation data are used to construct a precipitation time series graph. The slope between the simulated precipitation time and the previous historical time is extracted from the precipitation time series graph, and the precipitation simulation trend is determined based on the slope.

9. The watershed-scale agricultural non-point source pollution simulation method according to claim 8, characterized in that, When conducting pollution evolution analysis on pollution-associated and non-pollution-associated areas based on the aforementioned precipitation simulation trends, the analysis includes: The pollution evolution was simulated based on the SWAT model to analyze the precipitation simulation trend and the pollution-related areas. Based on the HSPF model, the precipitation simulation trend and pollution evolution of non-pollution-associated areas were simulated.

Images