Coupling model nitrogen and phosphorus simulation method based on agricultural season decomposition characteristics

Abstract

The invention discloses a coupling model nitrogen and phosphorus simulation method based on agricultural seasonal decomposition characteristics, provides an agricultural seasonal decomposition model SDA, and simulates the total nitrogen flux and the total phosphorus flux of a drainage basin in combination with a hydrological model and a multivariate deep learning model. The method not only can follow a watershed hydrological cycle process, considers migration and transformation processes of nitrogen and phosphorus along with hydrological processes such as rainfall, infiltration and evaporation, has a solid hydrological physical process, but also can deeply excavate a complex relationship between nitrogen and phosphorus and factors influencing migration and transformation of the nitrogen and phosphorus, so that the method has a good application prospect. According to the method, the features in the time series data can be efficiently extracted and fused, the interpretability of machine learning can be improved, the data requirement and parameter uncertainty of a traditional hydrological model can be reduced, and therefore the accuracy and robustness of total nitrogen flux and total phosphorus flux prediction are improved.

Description

Technical Field

[0001] The present invention belongs to the field of water environment, and particularly relates to a method for simulating nitrogen and phosphorus by a coupled model based on agricultural seasonal decomposition characteristics. Background Art

[0002] Water eutrophication refers to a phenomenon in which water bodies such as rivers, reservoirs, and lakes, under the influence of natural factors and human activities, receive excessive nutrient salts such as nitrogen and phosphorus, and gradually change from an oligotrophic state with a relatively low productivity level to a eutrophic state with a relatively high productivity level. In the process of industrial and agricultural development, along with human activities such as the application of pesticides and fertilizers, livestock and poultry breeding, and factory sewage discharge, pollutants such as nitrogen and phosphorus enter the hydrological cycle with precipitation and finally accumulate and enrich in rivers and lakes.

[0003] At present, the spatio-temporal simulation of nitrogen and phosphorus mainly adopts two methods: field measurement and model simulation. Although the field observation data has high accuracy, it is limited by the spatio-temporal scale and is difficult to perform basin-scale and long-time-series simulations; model simulations are mainly divided into two categories. One is the hydrological model based on physical processes, such as the SWAT hydrological model, the HSPF hydrological model, the NTT-Watershed hydrological model, etc. Among them, the SWAT hydrological model is more widely used. However, the simulation accuracy of the hydrological model and the accuracy of meteorological data are closely related to the sensitivity of hydrological parameters, etc., so it has high uncertainty. The other is the data-driven model, such as the linear regression model, the non-linear regression model, etc. In recent years, machine learning models have shown good applicability in nitrogen and phosphorus simulation, but because they belong to black-box models, when applied, it is necessary to balance model performance and interpretability, which often reduces the simulation accuracy. Summary of the Invention

[0004] In view of the problem of low accuracy in the existing technology for nitrogen and phosphorus simulation, the present invention proposes a method for simulating nitrogen and phosphorus by a coupled model based on agricultural seasonal decomposition characteristics.

[0005] The above technical problems of the present invention are mainly solved by the following technical solutions: A method for simulating nitrogen and phosphorus by a coupled model based on agricultural seasonal decomposition characteristics, comprising the following steps: Step 1, prepare a training set, which includes spatial data sequences, meteorological data sequences, and hydrographic and water quality data sequences corresponding to each region in the basin; the spatial data sequences include spatial data at different time points, and the spatial data includes land use information; the meteorological data sequences include meteorological data at different time points; the hydrographic and water quality data sequences include measured phosphorus fluxes and measured nitrogen fluxes at different time points; Step 2: Establish a hydrological model, calibrate the hydrological model using the training set to obtain the calibrated hydrological model; input the spatial data series and meteorological data series of the training set into the calibrated hydrological model to correspondingly obtain the preliminary simulated total nitrogen flux series and the preliminary simulated total phosphorus flux series of the basin; the preliminary simulated total nitrogen flux series includes the preliminary simulated total nitrogen fluxes at different time points, and the preliminary simulated total phosphorus flux series includes the preliminary simulated total phosphorus fluxes at different time points; Step 3: Take the meteorological data corresponding to a region of the entire basin in a training set at a time point as a regional meteorological data sample point; construct an agricultural seasonal decomposition model SDA. The agricultural seasonal decomposition model SDA classifies the regional meteorological data sample points, the preliminary simulated total nitrogen flux, and the preliminary simulated total phosphorus flux corresponding to the training set based on land use information, and each classification corresponds to a nitrogen impact period, a phosphorus impact period, and a weak nitrogen and phosphorus impact period respectively; Step 4: Build a multi - layer deep learning model, and use the data classified by the agricultural seasonal decomposition model SDA from the regional meteorological data sample points, the preliminary simulated total nitrogen flux, and the preliminary simulated total phosphorus flux corresponding to the training set to train the multi - layer deep learning model to obtain the trained multi - layer deep learning model; Step 5: Couple the calibrated hydrological model, the agricultural seasonal decomposition model SDA, and the trained multi - layer deep learning model to obtain a coupled model; input the spatial data series, meteorological data series, and hydro - water quality data series of the sample to be processed into the coupled model to obtain the corresponding final simulated total nitrogen flux series and final simulated total phosphorus flux series.

[0006] Step 3 as described above specifically includes the following steps: Step 3.1: Take the meteorological data corresponding to a region of the entire basin in a training set at a time point as a regional meteorological data sample point, and take the land use information corresponding to the time point of the regional meteorological data sample point as a basic attribute of the regional meteorological data sample point. The land use information includes the land type of the region at the corresponding time point, and the land types are divided into crop land types and non - crop land types. The crop land types are further divided into nitrogen crop land types, phosphorus crop land types, and non - nitrogen - phosphorus crop land types according to the types of crop fertilization and pesticide application; for the land use information corresponding to the crop land types, it also includes the cultivated land time characteristics, and the cultivated land time characteristics are the pesticide and fertilizer application period or the non - pesticide and fertilizer application period; Step 3.2: Classify the regional meteorological data sample points corresponding to the land type of nitrogen crops and the cultivated land time feature of the pesticide and fertilizer application period into the first characteristic sample sequence set, and also classify the preliminary simulated total nitrogen flux and preliminary simulated total phosphorus flux corresponding to the pesticide and fertilizer application period of the nitrogen crop land type into the first characteristic sample sequence set; regard the regional meteorological data sample points, preliminary simulated total nitrogen flux, and preliminary simulated total phosphorus flux corresponding to the same time point in the first characteristic sample sequence set as an agricultural season decomposition sample point in the first characteristic sample sequence set, and the agricultural season decomposition sample points with continuous time points in the first characteristic sample sequence set form an agricultural season decomposition sample sequence of the first characteristic sample sequence set, and the first characteristic sample sequence set corresponds to the nitrogen influence period; Classify the regional meteorological data sample points corresponding to the land type of phosphorus crops and the cultivated land time feature of the pesticide and fertilizer application period into the second characteristic sample sequence set, and also classify the preliminary simulated total nitrogen flux and preliminary simulated total phosphorus flux corresponding to the pesticide and fertilizer application period of the phosphorus crop land type into the second characteristic sample sequence set; regard the regional meteorological data sample points, preliminary simulated total nitrogen flux, and preliminary simulated total phosphorus flux corresponding to the same time point in the second characteristic sample sequence set as an agricultural season decomposition sample point in the second characteristic sample sequence set, and the agricultural season decomposition sample points with continuous time points in the first characteristic sample sequence set form an agricultural season decomposition sample sequence of the second characteristic sample sequence set, and the second characteristic sample sequence set corresponds to the phosphorus influence period; Classify the regional meteorological data sample points of the land type of non-crop land, the regional meteorological data sample points of the land type of non-nitrogen and phosphorus crops, and the regional meteorological data sample points with the cultivated land time feature of non-pesticide and fertilizer application period into the third characteristic sample sequence set, and also classify all the preliminary simulated total nitrogen flux and preliminary simulated total phosphorus flux into the third characteristic sample sequence set; regard the regional meteorological data sample points, preliminary simulated total nitrogen flux, and preliminary simulated total phosphorus flux corresponding to the same time point in the third characteristic sample sequence set as an agricultural season decomposition sample point in the third characteristic sample sequence set, and the agricultural season decomposition sample points with continuous time points in the third characteristic sample sequence set form an agricultural season decomposition sample sequence of the third characteristic sample sequence set, and the third characteristic sample sequence set corresponds to the weak nitrogen and phosphorus influence period.

[0007] Step 3 as described above further includes the following steps: Step 3.3: The data in the first feature sample sequence set is grouped according to elements, and the data corresponding to each element is sequentially subjected to min-max normalization and standardization conversion to obtain the cleaned first feature sample sequence set; the data in the second feature sample sequence set is grouped according to elements, and the data corresponding to each element is sequentially subjected to min-max normalization and standardization conversion to obtain the cleaned second feature sample sequence set; the data in the third feature sample sequence set is grouped according to elements, and the data corresponding to each element is sequentially subjected to min-max normalization and standardization conversion to obtain the cleaned third feature sample sequence set. The elements corresponding to the data in the first feature sample sequence set, the second feature sample sequence set, and the third feature sample sequence set all include precipitation, maximum temperature, minimum temperature, solar radiation, relative humidity, wind speed, preliminary simulated total nitrogen flux, and preliminary simulated total phosphorus flux.

[0008] As described above, the training of the multi - variable deep learning model in step 4 includes the following steps: Based on the measured phosphorus flux and measured nitrogen flux at each time point in the hydrological and water quality data sequences corresponding to each region of the basin, the measured total phosphorus flux sequence and measured total nitrogen flux sequence of the entire basin are extracted. The measured total phosphorus flux sequence includes the measured total phosphorus flux at each time point, and the measured total nitrogen flux sequence includes the measured total nitrogen flux at each time point. Input the agricultural season decomposition sample sequence in the cleaned first feature sample sequence set into the multi - variable deep learning model, and use the measured total phosphorus flux and measured total nitrogen flux corresponding to the time point sequence of the agricultural season decomposition sample sequence in the cleaned first feature sample sequence set as the reference targets for model training to obtain the first trained multi - variable deep learning model; input the agricultural season decomposition sample sequence in the cleaned second feature sample sequence set into the multi - variable deep learning model, and use the measured total phosphorus flux and measured total nitrogen flux corresponding to the time point sequence of the agricultural season decomposition sample sequence in the cleaned second feature sample sequence set as the reference targets for model training to obtain the second trained multi - variable deep learning model; input the agricultural season decomposition sample sequence in the cleaned third feature sample sequence set into the multi - variable deep learning model, and use the measured total phosphorus flux and measured total nitrogen flux corresponding to the time point sequence of the agricultural season decomposition sample sequence in the cleaned third feature sample sequence set as the reference targets for model training to obtain the third trained multi - variable deep learning model.

[0009] As described above, in step 5, inputting the spatial data sequence, meteorological data sequence, and hydrological and water quality data sequence of the sample to be processed into the coupling model to obtain the corresponding final simulated total nitrogen flux sequence and final simulated total phosphorus flux sequence specifically includes the following steps: The spatial data sequence, meteorological data sequence, and hydrological water quality data sequence in the sample to be processed are input into the calibrated hydrological model to obtain the corresponding preliminary simulated total nitrogen flux sequence and preliminary simulated total phosphorus flux sequence. The meteorological data in the meteorological data sequence of the sample to be processed, the preliminary simulated total nitrogen flux in the preliminary simulated total nitrogen flux sequence, and the preliminary simulated total phosphorus flux in the preliminary simulated total phosphorus flux sequence are divided into the corresponding first characteristic sample sequence set, second characteristic sample sequence set, and third characteristic sample sequence set through the agricultural seasonal decomposition model SDA; The data in the first characteristic sample sequence set are grouped according to elements, and the data of each element are sequentially normalized by min-max and then standardized and converted, and then input into the first trained multi-layer deep learning model together to obtain the corresponding second simulated first total nitrogen flux sequence and second simulated first total phosphorus flux sequence; The data in the second characteristic sample sequence set are grouped according to elements, and the data of each element are sequentially normalized by min-max and then standardized and converted, and then input into the second trained multi-layer deep learning model together to obtain the corresponding second simulated second total nitrogen flux sequence and second simulated second total phosphorus flux sequence; The data in the third characteristic sample sequence set are grouped according to elements, and the data of each element are sequentially normalized by min-max and then standardized and converted, and then input into the third trained multi-layer deep learning model together to obtain the corresponding second simulated third total nitrogen flux sequence and second simulated third total phosphorus flux sequence; In the final simulated total nitrogen flux sequence, for each time point during the pesticide application and fertilization period of nitrogen crops, the corresponding final simulated total nitrogen flux is equal to the second simulated first total nitrogen flux at the same time point in the second simulated first total nitrogen flux sequence; In the final simulated total nitrogen flux sequence, for each time point not during the pesticide application and fertilization period of nitrogen crops, the corresponding final simulated total nitrogen flux is equal to the average of the second simulated second total nitrogen flux in the second simulated second total nitrogen flux sequence and the second simulated third total nitrogen flux in the second simulated third total nitrogen flux sequence at the same time point; In the final simulated total phosphorus flux sequence, for each time point during the pesticide application and fertilization period of phosphorus crops, the corresponding final simulated total phosphorus flux is equal to the second simulated second total phosphorus flux at the same time point in the second simulated second total phosphorus flux sequence; In the final simulated total phosphorus flux sequence, for each time point not during the pesticide application and fertilization period of phosphorus crops, the corresponding final simulated total phosphorus flux is equal to the average of the second simulated first total phosphorus flux in the second simulated first total phosphorus flux sequence and the second simulated third total phosphorus flux in the second simulated third total phosphorus flux sequence at the same time point.

[0010] As described above, in step 4, the multi - variable deep learning model is the RABA model. The RABA model sequentially includes a residual feature perception module RA and a bidirectional attention prediction module BA. The residual feature perception module RA sequentially includes a CNN input layer, a convolutional layer, a pooling layer, and a CNN output layer. The bidirectional attention prediction module BA sequentially includes a BILSTM input layer, an LSTM layer, an attention layer, an activation layer, and a BILSTM output layer. The input data of the CNN input layer is added to the output data of the CNN output layer through a skip connection and then input into the BILSTM input layer.

[0011] As described above, the attention layer adopts a channel attention mechanism module SE.

[0012] As described above, the hydrological model is the SWAT hydrological model.

[0013] A computer device includes a memory and a processor. The memory stores a computer program. When the processor executes the computer program, the steps 1 - 5 of any one of the nitrogen and phosphorus simulation methods described above are implemented.

[0014] A computer - readable storage medium stores a computer program. When the computer program is executed by a processor, the steps 1 - 5 of any one of the nitrogen and phosphorus simulation methods described above are implemented.

[0015] The present invention has the following beneficial effects compared with the prior art: Aiming at the problem of low accuracy in nitrogen and phosphorus simulation, the present invention proposes an agricultural seasonal decomposition model SDA, and combines the SWAT model and the RABA model for nitrogen and phosphorus simulation. The present invention can not only follow the basin hydrological cycle process, considering both the migration and transformation processes of nitrogen and phosphorus under hydrological processes such as precipitation, infiltration, and evaporation, with a solid hydrological physical process, but also deeply explore the complex relationship between nitrogen and phosphorus and the factors affecting their migration and transformation, and can efficiently extract and fuse the features in the time - series data. It can not only improve the interpretability of machine learning, but also reduce the data requirements and parameter uncertainties of traditional hydrological models, thereby improving the accuracy and robustness of total nitrogen flux and total phosphorus flux prediction. The agricultural seasonal decomposition model SDA can judge the correlation with nitrogen and phosphorus according to the characteristics of the input time - series data, and form a multi - feature input data set.

[0016] Compared with other methods, the Nash coefficient NSE of nitrogen and phosphorus simulation in the nitrogen and phosphorus simulation method proposed by the present invention exceeds 0.94, and the relative error between the simulated value and the measured value is less than 1%.

[0017] The SDA-SWAT-RABA coupling model proposed by the present invention has achieved better results in the simulation of total nitrogen flux. Compared with the SWAT hydrological model and the SWAT-RABA model, the Pearson correlation coefficient PCC of the SDA-SWAT-RABA coupling model in the verification period has increased by 33% and 4% respectively, the Nash coefficient NSE has increased by 34% and 3% respectively, and the percentage bias has decreased by 4.44% and 0.31%.

[0018] The SDA-SWAT-RABA coupling model proposed by the present invention has also achieved better results in the simulation of total phosphorus flux. Compared with the SWAT hydrological model and the SWAT-RABA model, the Pearson correlation coefficient PCC of the SDA-SWAT-RABA coupling model in the verification period has increased by 31% and 3% respectively, the Nash coefficient NSE has increased by 32% and 4% respectively, and the percentage bias has decreased by 6.55% and 4.54%. Brief Description of the Drawings

[0019] Figure 1 is a flowchart of the present invention.

[0020] Figure 2 is a comparison chart of the fitting effects between the total nitrogen flux simulated by the SDA-SWAT-RABA coupling model proposed by the present invention, the SWAT hydrological model, and the SWAT-RABA model and the measured total nitrogen flux. (a) is a comparison chart of the fitting effects between the total nitrogen flux simulated in the calibration period and the measured total nitrogen flux, and (b) is a comparison chart of the fitting effects between the total nitrogen flux simulated in the verification period and the measured total nitrogen flux; the colors represent different models, black is the fitting comparison between the total nitrogen flux simulated by the SWAT hydrological model and the measured total nitrogen flux, green is the fitting comparison between the total nitrogen flux simulated by the SWAT-RABA model and the measured total nitrogen flux, and red is the fitting comparison between the total nitrogen flux simulated by the SDA-SWAT-RABA coupling model and the measured total nitrogen flux, R 2 represents the coefficient of determination.

[0021] Figure 3It is a comparison chart of the fitting effects between the total phosphorus fluxes simulated by the SDA-SWAT-RABA coupling model, the SWAT hydrological model, and the SWAT-RABA model proposed by the present invention and the measured total phosphorus fluxes. (a) is the comparison chart of the fitting effects between the total phosphorus fluxes simulated during the calibration period and the measured total phosphorus fluxes. (b) is the comparison chart of the fitting effects between the total phosphorus fluxes simulated during the verification period and the measured total phosphorus fluxes. The colors represent different models. The black is the fitting comparison between the total phosphorus flux simulated by the SWAT hydrological model and the measured phosphorus flux. The green is the fitting comparison between the total phosphorus flux simulated by the SWAT-RABA model and the measured total phosphorus flux. The red is the fitting comparison between the total phosphorus flux simulated by the SDA-SWAT-RABA coupling model and the measured total phosphorus flux. R 2 represents the coefficient of determination.

[0022] Figure 4 It is a comparison chart of the monthly total nitrogen flux results from 2019 to 2022 simulated by the SDA-SWAT-RABA coupling model, the SWAT hydrological model, and the SWAT-RABA model proposed by the present invention. (a) represents the comparison chart of the monthly total nitrogen flux results during the calibration period. (b) represents the comparison chart of the monthly total nitrogen flux results during the verification period. The colors represent different models. The black is the measured total nitrogen flux. The green is the total nitrogen flux simulated by the SWAT hydrological model. The orange is the total nitrogen flux simulated by the SWAT-RABA model. The red is the total nitrogen flux simulated by the SDA-SWAT-RABA coupling model.

[0023] Figure 5 It is a comparison chart of the monthly total phosphorus flux results from 2019 to 2022 simulated by the SDA-SWAT-RABA coupling model, the SWAT hydrological model, and the SWAT-RABA model proposed by the present invention. (a) represents the comparison chart of the monthly total phosphorus flux results during the calibration period. (b) represents the comparison chart of the monthly total phosphorus flux results during the verification period. The colors represent different models. The black is the measured total phosphorus flux. The green is the total phosphorus flux simulated by the SWAT hydrological model. The orange is the total phosphorus flux simulated by the SWAT-RABA model. The red is the total phosphorus flux simulated by the SDA-SWAT-RABA coupling model.

[0024] Figure 6 It is the structural diagram of the RABA model. Among them, the RA module represents the residual feature perception module RA, the BA module represents the bidirectional attention prediction module BA, n represents the number of time points corresponding to the agricultural seasonal decomposition sample sequence. LSTM represents the long short-term memory network. Attention represents the attention layer. Sigmoid represents the Sigmoid activation function used in the activation layer.

[0025] Figure 7 It is the structural diagram of the residual module.

[0026] Figure 8 It is a schematic diagram of coupling a hydrological model, an agricultural seasonal decomposition model SDA, and a multi - variable deep learning model, where SDA represents the agricultural seasonal decomposition model SDA. Specific implementation manners

[0027] To facilitate the understanding and implementation of the present invention by those of ordinary skill in the art, the present invention will be further described in detail below in conjunction with implementation examples. It should be understood that the implementation examples described herein are only used to illustrate and explain the present invention and are not used to limit the present invention.

[0028] Example 1

[0029] A method for simulating nitrogen and phosphorus in a coupled model based on agricultural seasonal decomposition characteristics, which simulates the total nitrogen flux and total phosphorus flux of a basin based on coupling a hydrological model and a multi - variable deep learning model with agricultural seasonal decomposition. It includes establishing a hydrological model, constructing an agricultural seasonal decomposition model (SDA, Seasonal Decomposition of Agriculture), building and training a multi - variable deep learning model, and coupling the hydrological model, the agricultural seasonal decomposition model SDA, and the multi - variable deep learning model. The overall scheme is as Figure 1 shown, and the specific implementation steps are as follows: Step 1: Prepare a training set, which includes spatial data sequences, meteorological data sequences, and hydrological and water quality data sequences corresponding to each region in the basin.

[0030] The above - mentioned spatial data sequences, meteorological data sequences, and hydrological and water quality data sequences are all time - series data. The spatial data sequence includes spatial data at different time points, and each spatial data includes a digital elevation model, land use information, and soil type, etc. The spatial data needs to be processed by calling the ArcGis platform; the land use information includes the land types in the region at the corresponding time point, and the land types are divided into crop land types and non - crop land types. The crop land types are further divided into nitrogen - crop land types, phosphorus - crop land types, and non - nitrogen - phosphorus - crop land types according to the types of crop fertilization and pesticide application; for the land use information corresponding to the crop land types, it also includes the cultivated land time characteristics, which are the pesticide and fertilizer application period or non - pesticide and fertilizer application period, and the cultivated land time characteristics show whether the region is in the pesticide and fertilizer application period at the corresponding time point. The meteorological data sequence includes meteorological data at different time points, and the meteorological data mainly exists in the form of text and tables. The elements in the meteorological data include precipitation, maximum temperature, minimum temperature, solar radiation, relative humidity, and wind speed, and the meteorological data is stored in the Access database platform. The hydrological and water quality data sequence includes the measured phosphorus flux and the measured nitrogen flux at different time points.

[0031] The time points of the above - mentioned spatial data, meteorological data, and hydrological and water quality data are corresponding.

[0032] According to the temporal distribution range of the time points of spatial data, meteorological data, and hydrological and water quality data, the temporal distribution range is divided into a calibration period and a verification period. The spatial data, meteorological data, and hydrological and water quality data in the calibration period are classified as the training set and used to calibrate the hydrological model and train the coupled model obtained by coupling the seasonal decomposition model for agriculture (SDA) and the multi - deep - learning model. The spatial data, meteorological data, and hydrological and water quality data in the verification period are classified as the verification set to verify and evaluate the simulation accuracy of the hydrological model and the coupled model. In this embodiment, the first 75% of the time points in the temporal distribution range are classified as the calibration period, and the last 25% of the time points in the temporal distribution range are classified as the verification period.

[0033] Step 2: Establish a hydrological model, and calibrate the hydrological model using the spatial data sequence, meteorological data sequence, and hydrological and water quality data sequence in the training set to obtain the calibrated hydrological model; input the spatial data sequence and meteorological data sequence in the training set into the calibrated hydrological model to correspondingly obtain the preliminary simulated total nitrogen flux sequence and the preliminary simulated total phosphorus flux sequence of the basin. The preliminary simulated total nitrogen flux sequence includes the preliminary simulated total nitrogen fluxes at different time points, and the preliminary simulated total phosphorus flux sequence includes the preliminary simulated total phosphorus fluxes at different time points.

[0034] The hydrological model inputs the spatial data sequence, meteorological data sequence, and hydrological and water quality data sequence; outputs the preliminary simulated total nitrogen flux sequence and the preliminary simulated total phosphorus flux sequence; the hydrological model selected in this embodiment is the Soil and Water Assessment Tool (abbreviated as SWAT hydrological model). The SWAT hydrological model relies on Hydrological Response Units (HRUs) to estimate the preliminary simulated total nitrogen flux sequence and the preliminary simulated total phosphorus flux sequence of the basin. The hydrological response unit HRU is the smallest unit for SWAT hydrological model calculation. In this embodiment, the specific process of step 2 is as follows: Step 2.1: Establish a hydrological model. In this embodiment, a SWAT hydrological model is constructed.

[0035] Call the ArcSWAT tool of the ArcGIS platform to complete the establishment of the SWAT hydrological model. The basin to be simulated is divided into 79 sub - basins and 1578 hydrological response units HRUs. Each hydrological response unit HRU includes corresponding spatial data, meteorological data, and hydrological and water quality data.

[0036] Step 2.2: Conduct sensitivity analysis and calibration on the parameters of the hydrological model, and input the spatial data and meteorological data in the training set into the calibrated hydrological model to correspondingly obtain the preliminary simulated total nitrogen flux sequence and the preliminary simulated total phosphorus flux sequence of the basin.

[0037] The SWAT hydrological model has numerous parameters, and it is difficult to manually adjust the range of each parameter of the SWAT hydrological model. Therefore, through parameter sensitivity analysis, the parameters that have a greater impact on the simulation results of the SWAT hydrological model are selected, and then these parameters of the SWAT hydrological model are adjusted focusly, which can improve work efficiency. In this embodiment, the SWAT-CUP (SWAT Calibration and Uncertainty Programs) software is called to perform parameter sensitivity analysis, calibration and verification work. The SWAT-CUP software incorporates five algorithms, and the SUFI-2 algorithm has high computational efficiency because it uses the Latin hypercube sampling method to obtain the main parameter values. In this embodiment, 16 parameters of the SWAT hydrological model with relatively high sensitivity are finally screened out, as shown in Table 1.

[0038] The specific process of performing parameter sensitivity analysis and calibration is as follows: First, set the default value range of the parameters of the SWAT hydrological model, and then call the SWAT-CUP software to perform multiple rounds of parameter sampling simulation iterations. After each round of parameter sampling simulation iteration, the SWAT-CUP software will give the recommended value range of the parameters of the SWAT hydrological model for the next round of parameter sampling simulation iteration. Adjust the parameters of the SWAT hydrological model according to the recommended value range until the simulation accuracy of the SWAT hydrological model can no longer be improved. In this embodiment, the number of parameter sampling simulations in each round of parameter sampling simulation iteration of the SWAT hydrological model is set to 1500, and the number of iteration rounds is 14 rounds.

[0039] Among them, the following method is used to evaluate the simulation accuracy of the SWAT hydrological model: This invention selects three evaluation indicators to evaluate the simulation accuracy of the SWAT hydrological model, namely the Pearson correlation coefficient PCC, the Nash coefficient NSE, and the percentage bias Pbias (unit: %), and the formulas are as follows: (1) In the formula is the measured flux in the i-th month, is the simulated flux in the i-th month; the serial number i of the month ∈ {1, 2,..., n}, and n is the total number of months; is the measured average value of the fluxes in n months; is the simulated average value of the fluxes in n months; among them, the fluxes are taken as the total nitrogen flux and the total phosphorus flux in turn, that is, the preliminary simulated total nitrogen flux and the preliminary simulated total phosphorus flux output by the SWAT hydrological model are respectively evaluated by the Pearson correlation coefficient PCC. The closer the evaluation index of the Pearson correlation coefficient PCC is to 1, the better.

[0040] (2) (3) It represents the Nash-Sutcliffe Efficiency (NSE). The closer the Nash-Sutcliffe Efficiency (NSE) is to 1, the better the performance of the SWAT hydrological model. It represents the Percentage Bias (Pbias). The Percentage Bias (Pbias) represents the error between the measured value and the simulated value, and usually takes values in the range of [-15, 15]%. The closer the Percentage Bias (Pbias) is to 0, the better.

[0041] Table 1 Parameters of the SWAT Hydrological Model (The T-score represents the sensitivity level. The larger the absolute value of the T-score, the more sensitive it is; the P-value represents the significance. The closer the P-value is to 0, the more significant the sensitivity.)

[0042] Step 3: Construct the Seasonal Decomposition of Agricultural Time Series (SDA) model.

[0043] Take the meteorological data corresponding to a region of the entire basin at a time point in the training set as a regional meteorological data sample point; construct the Seasonal Decomposition of Agricultural Time Series (SDA) model. The Seasonal Decomposition of Agricultural Time Series (SDA) model classifies the regional meteorological data sample points, the preliminary simulated total nitrogen flux, and the preliminary simulated total phosphorus flux based on land use information, and each classification corresponds to the nitrogen impact period, the phosphorus impact period, and the weak nitrogen and phosphorus impact period respectively.

[0044] To compare the preliminary simulated total nitrogen flux output by the hydrological model and the final simulated total nitrogen flux output by the subsequently constructed multi-layer deep learning model as much as possible, and to compare the preliminary simulated total phosphorus flux output by the hydrological model and the final simulated total phosphorus flux output by the subsequently constructed multi-layer deep learning model, the elements corresponding to the input meteorological data of the multi-layer deep learning model are the same as those corresponding to the input meteorological data of the hydrological model. Therefore, the elements corresponding to the input meteorological data of the multi-layer deep learning model include precipitation, maximum temperature, minimum temperature, solar radiation, relative humidity, wind speed, and the elements corresponding to the input data of the multi-layer deep learning model also include the preliminary simulated total nitrogen flux and the preliminary simulated total phosphorus flux.

[0045] The nitrogen and phosphorus loads mainly come from two forms: point source pollution and non-point source pollution. Point source pollution refers to, for example, factory sewage discharge, etc. At present, point source pollution has been effectively controlled. However, non-point source pollution is the main source of nitrogen and phosphorus loads at present due to its unclear sources and large diffusion area. Non-point source pollution mainly comes from agricultural economic activities (fertilizer application, pesticide application, etc.). Generally, the rules of agricultural activities are as follows: different types of crops have different time nodes for sowing, fertilizing, pesticide application, and harvesting, and the amounts of fertilizer and pesticide application are determined according to the crop planting area. Therefore, in order to better simulate the impact of non-point source pollution on the total nitrogen flux and total phosphorus flux, the present invention proposes a new agricultural seasonal decomposition model (SDA, Seasonal Decomposition of Agriculture) based on the rules of agricultural activities, so as to preprocess the data to be input into the RABA model first: Step 3.1: Take the meteorological data corresponding to a region in the entire basin at a time point in the training set as a regional meteorological data sample point, and take the land use information corresponding to the time point of the regional meteorological data sample point as a basic attribute of the regional meteorological data sample point. The land use information of each region in the basin obtained in Step 1 includes the land type of the region at the corresponding time point. The land types are divided into crop land types and non-crop land types. The crop land types are further divided into nitrogen crop land types, phosphorus crop land types, and non-nitrogen and phosphorus crop land types according to the types of fertilizer and pesticide application for the crops; for the land use information corresponding to the crop land types, it also includes the cultivated land time characteristics, and the cultivated land time characteristics are the pesticide and fertilizer application period or the non-pesticide and fertilizer application period. The cultivated land time characteristics indicate whether the region is in the pesticide and fertilizer application period at the corresponding time point. A region corresponds to a land type and a cultivated land time characteristic at a time point. The elements corresponding to the regional meteorological data sample point include: precipitation, maximum temperature, minimum temperature, solar radiation, relative humidity, and wind speed. In this embodiment, the time point is selected in units of days.

[0046] Step 3.2: Classify the regional meteorological data sample point, the preliminary simulated total nitrogen flux, and the preliminary simulated total phosphorus flux according to the land type and the cultivated land time characteristics in the land use information corresponding to the regional meteorological data sample point: Classify the regional meteorological data sample points corresponding to the land type of nitrogen crop land type and the cultivated land time characteristic of the pesticide and fertilizer application period into the first characteristic sample sequence set, and also classify the preliminary simulated total nitrogen flux and the preliminary simulated total phosphorus flux corresponding to the pesticide and fertilizer application period of the nitrogen crop land type into the first characteristic sample sequence set; take the regional meteorological data sample point, the preliminary simulated total nitrogen flux, and the preliminary simulated total phosphorus flux corresponding to the same time point in the first characteristic sample sequence set as an agricultural seasonal decomposition sample point in the first characteristic sample sequence set. The agricultural seasonal decomposition sample points with continuous time points in the first characteristic sample sequence set form an agricultural seasonal decomposition sample sequence of the first characteristic sample sequence set, and the first characteristic sample sequence set corresponds to the nitrogen influence period; The regional meteorological data sample points with the land type being the phosphorus crop land type and the cultivated land time feature corresponding to the pesticide application and fertilization period are classified into the second characteristic sample sequence set, and the preliminary simulated total nitrogen flux and the preliminary simulated total phosphorus flux corresponding to the pesticide application and fertilization period of the phosphorus crop land type are also classified into the second characteristic sample sequence set; the regional meteorological data sample points, the preliminary simulated total nitrogen flux, and the preliminary simulated total phosphorus flux corresponding to the same time point in the second characteristic sample sequence set are used as an agricultural season decomposition sample point in the second characteristic sample sequence set, and the agricultural season decomposition sample points with continuous time points in the first characteristic sample sequence set constitute an agricultural season decomposition sample sequence of the second characteristic sample sequence set, and the second characteristic sample sequence set corresponds to the phosphorus influence period; The regional meteorological data sample points with the land type being non-crop land type, the regional meteorological data sample points with the land type being non-nitrogen and phosphorus crop land type, and the regional meteorological data sample points with the cultivated land time feature being non-pesticide application and fertilization period are classified into the third characteristic sample sequence set, and all the preliminary simulated total nitrogen flux and the preliminary simulated total phosphorus flux are also classified into the third characteristic sample sequence set. The regional meteorological data sample points, the preliminary simulated total nitrogen flux, and the preliminary simulated total phosphorus flux corresponding to the same time point in the third characteristic sample sequence set are used as an agricultural season decomposition sample point in the third characteristic sample sequence set, and the agricultural season decomposition sample points with continuous time points in the third characteristic sample sequence set constitute an agricultural season decomposition sample sequence of the third characteristic sample sequence set, and the third characteristic sample sequence set corresponds to the weak nitrogen and phosphorus influence period.

[0047] In this embodiment, the division process of the regional meteorological data sample points is as Figure 8 shown, specifically: If the land type is non-crop land type, the corresponding regional meteorological data sample points are classified into the third characteristic sample sequence set; if the land type is crop land type, further judge whether the crop land type in this area belongs to nitrogen crop land type based on the crop fertilization and pesticide application types. If it is not nitrogen crop land type, continue to judge whether it belongs to phosphorus crop land type. If the crop land type is non-nitrogen and phosphorus crop land type, that is, the crop land type belongs to neither nitrogen crop land type nor phosphorus crop land type, then the corresponding regional meteorological data sample points are classified into the third characteristic sample sequence set; if the crop land type belongs to nitrogen crop land type (such as the corresponding crops are rice and wheat), then judge the cultivated land time feature of the regional meteorological data sample points again. If the cultivated land time feature is the pesticide application and fertilization period (such as rice is generally fertilized at the seedling stage and tillering stage), then the corresponding regional meteorological data sample points are classified into the first characteristic sample sequence set. Similarly, if the crop land type belongs to phosphorus crop land type and the cultivated land time feature is the pesticide application and fertilization period, then the corresponding regional meteorological data sample points are classified into the second characteristic sample sequence set; if the crop land type belongs to nitrogen crop land type but the cultivated land time feature is non-pesticide application and fertilization period, then the corresponding regional meteorological data sample points are classified into the third characteristic sample sequence set; if the crop land type belongs to phosphorus crop land type but the cultivated land time feature is non-pesticide application and fertilization period, then the corresponding regional meteorological data sample points are classified into the third characteristic sample sequence set.

[0048] The agricultural seasonal decomposition model SDA decomposes the seasonal components of regional meteorological data sample points, the preliminary simulated total nitrogen flux, and the preliminary simulated total phosphorus flux according to land use information and the characteristics of cultivated land time, so as to obtain time series data that is more relevant to the total nitrogen flux and total phosphorus flux to be simulated, with more prominent characteristics and easier to be captured by the subsequent constructed multi - variable deep learning model.

[0049] Furthermore, since there are obvious differences in the magnitude of the data corresponding to each element in the regional meteorological data sample points, step 3 also includes step 3.3 for data cleaning, and the specific operations are as follows: Before the data in the first feature sample sequence set, the second feature sample sequence set, and the third feature sample sequence set are input into the subsequent constructed multi - variable deep learning model, the data in the first feature sample sequence set is grouped by element, and the data corresponding to each element is successively subjected to min - max normalization and standardization conversion to obtain the cleaned first feature sample sequence set; the data in the second feature sample sequence set is grouped by element, and the data corresponding to each element is successively subjected to min - max normalization and standardization conversion to obtain the cleaned second feature sample sequence set; the data in the third feature sample sequence set is grouped by element, and the data corresponding to each element is successively subjected to min - max normalization and standardization conversion to obtain the cleaned third feature sample sequence set. The elements corresponding to the data in the first feature sample sequence set, the second feature sample sequence set, and the third feature sample sequence set all include precipitation, maximum temperature, minimum temperature, solar radiation, relative humidity, wind speed, the preliminary simulated total nitrogen flux, and the preliminary simulated total phosphorus flux, where precipitation, maximum temperature, minimum temperature, solar radiation, relative humidity, and wind speed all belong to meteorological data.

[0050] During the data cleaning process, min - max normalization is performed to make the values of the data corresponding to each element fall within the interval [0, 1]; however, because the data change ranges of different elements are different, for example, the changes in maximum temperature, minimum temperature, and wind speed are smaller than those in solar radiation, the preliminary simulated total nitrogen flux, and the preliminary simulated total phosphorus flux, the distributions of maximum temperature, minimum temperature, and wind speed in the interval [0, 1] may be more uniform after min - max normalization. This difference in distribution requires further standardization conversion to make the data distributions corresponding to each element more conform to the normal distribution. Since there are special 0 values (representing no rain and no wind) in precipitation and wind speed data, the data corresponding to precipitation and wind speed are respectively standardized using the mean value, and the other elements (i.e., maximum temperature, minimum temperature, solar radiation, relative humidity, and the preliminary simulated total nitrogen flux and the preliminary simulated total phosphorus flux) are standardized using the median.

[0051] Step 4: Build and train a multi - deep learning model.

[0052] Step 4.1: Build a multi - deep learning model: The multi - deep learning model built in this embodiment is the RABA (Residual - Aware Bidirectional Attention) model proposed based on the CNN - BILSTM model. As Figure 6 shown, the RABA model sequentially includes a residual feature perception module RA and a bidirectional attention prediction module BA; the residual feature perception module RA sequentially includes a CNN input layer, a convolutional layer, a pooling layer, and a CNN output layer; the bidirectional attention prediction module BA sequentially includes a BILSTM input layer, an LSTM layer, an attention layer, an activation layer, and a BILSTM output layer; the input data of the CNN input layer is added to the output data of the CNN output layer after skip connection and then input to the BILSTM input layer.

[0053] In this embodiment, the input data of the CNN input layer is the agricultural seasonal decomposition sample sequence in the first cleaned feature sample sequence set or the agricultural seasonal decomposition sample sequence in the second cleaned feature sample sequence set or the agricultural seasonal decomposition sample sequence in the third cleaned feature sample sequence set; when the input data of the CNN input layer is the agricultural seasonal decomposition sample sequence in the first cleaned feature sample sequence set, the output of the RABA model is the corresponding second - simulated first total nitrogen flux sequence and the second - simulated first total phosphorus flux sequence; when the input data of the CNN input layer is the agricultural seasonal decomposition sample sequence in the second cleaned feature sample sequence set, the output of the RABA model is the corresponding second - simulated second total nitrogen flux sequence and the second - simulated second total phosphorus flux sequence; when the input data of the CNN input layer includes the agricultural seasonal decomposition sample sequence in the third cleaned feature sample sequence set, the output of the RABA model is the corresponding second - simulated third total nitrogen flux sequence and the second - simulated third total phosphorus flux sequence.

[0054] In the RABA model, the residual feature perception module RA is based on the CNN model, and the bidirectional attention prediction module BA is based on the BILSTM model, that is, the LSTM layer uses the BILSTM model. The CNN model can extract local features of the input data and input the local features into the BILSTM model. However, to avoid problems such as network degradation when the depth of the CNN model increases, a residual module is added between the CNN model and the BILSTM model (as Figure 7As shown in [Figure 0], that is, the input data of the input layer of the CNN is added to the output data of the output layer of the CNN through a skip connection and then input into the input layer of the BILSTM. Although the BILSTM model can consider the forward and backward relationships of the input data (time series data) at the same time, thus better capturing the time dependence, too many invalid features are added during the skip connection of the features. Therefore, the bidirectional attention prediction module BA of the present invention embeds an attention layer on the basis of the BILSTM model, and the attention layer adopts the channel attention mechanism module SE, so that the final prediction result pays more attention to the important part of the features reconstructed by the CNN model. The channel attention mechanism module SE calculates the influence weights of different feature factors on the classification result, and screens out the important feature information with higher weights to improve the prediction ability of the model.

[0055] The RABA model is built by calling the MATLAB 2024b deep network designer. The convolution kernel size in the residual feature perception module RA is (3, 1), the maximum pooling window is set to (448, 1), and the number of channels is 64; the number of attention heads in the channel attention mechanism module SE is 2 (two-dimensional channels of time and space), and the activation layer uses the Sigmoid activation function for output.

[0056] Step 4.2: Train the multi - deep learning model: According to the measured phosphorus flux and measured nitrogen flux at each time point in the hydrological and water quality data sequences corresponding to each region of the basin, the measured total phosphorus flux sequence and measured total nitrogen flux sequence of the entire basin are extracted. The measured total phosphorus flux sequence includes the measured total phosphorus flux at each time point, and the measured total nitrogen flux sequence includes the measured total nitrogen flux at each time point.

[0057] Input the agricultural season decomposition sample sequences in the first cleaned feature sample sequence set into the multi - deep learning model, and use the measured total phosphorus flux and the measured total nitrogen flux corresponding to the time - point sequence of the agricultural season decomposition sample sequences in the first cleaned feature sample sequence set as the reference targets for model training to obtain the first trained multi - deep learning model; input the agricultural season decomposition sample sequences in the second cleaned feature sample sequence set into the multi - deep learning model, and use the measured total phosphorus flux and the measured total nitrogen flux corresponding to the time - point sequence of the agricultural season decomposition sample sequences in the second cleaned feature sample sequence set as the reference targets for model training to obtain the second trained multi - deep learning model; input the agricultural season decomposition sample sequences in the third cleaned feature sample sequence set into the multi - deep learning model, and use the measured total phosphorus flux and the measured total nitrogen flux corresponding to the time - point sequence of the agricultural season decomposition sample sequences in the third cleaned feature sample sequence set as the reference targets for model training to obtain the third trained multi - deep learning model; that is, the trained multi - deep learning model includes the first trained multi - deep learning model, the second trained multi - deep learning model, and the third trained multi - deep learning model.

[0058] In this embodiment, input the agricultural season decomposition sample sequences in the first cleaned feature sample sequence set into the RABA model, and use the measured total phosphorus flux and the measured total nitrogen flux corresponding to the time - point sequence of the agricultural season decomposition sample sequences as the reference targets for model training to obtain the first trained RABA model; input the agricultural season decomposition sample sequences in the second cleaned feature sample sequence set into the RABA model, and use the measured total phosphorus flux and the measured total nitrogen flux corresponding to the time - point sequence of the agricultural season decomposition sample sequences as the reference targets for model training to obtain the second trained RABA model; input the agricultural season decomposition sample sequences in the third cleaned feature sample sequence set into the RABA model, and use the measured total phosphorus flux and the measured total nitrogen flux corresponding to the time - point sequence of the agricultural season decomposition sample sequences as the reference targets for model training to obtain the third trained RABA model.

[0059] During the model training process, use the adaptive model optimizer Adam function for training, the loss function is the mean square error loss function MSE Loss, the learning rate is set to 0.0001, and the number of iteration rounds is set to 600.

[0060] Step 5: Couple the calibrated hydrological model, the agricultural season decomposition model SDA, and the trained multi - deep learning model to obtain a coupled model. Input the spatial data sequence, meteorological data sequence, and hydrological water quality data sequence in the sample to be processed into the coupled model to obtain the corresponding final simulated total nitrogen flux sequence and final simulated total phosphorus flux sequence.

[0061] The spatial data sequence, meteorological data sequence, and hydrological water quality data sequence in the sample to be processed are input into the calibrated hydrological model to obtain the corresponding preliminary simulated total nitrogen flux sequence and preliminary simulated total phosphorus flux sequence. The meteorological data in the meteorological data sequence of the sample to be processed, the preliminary simulated total nitrogen flux in the preliminary simulated total nitrogen flux sequence, and the preliminary simulated total phosphorus flux in the preliminary simulated total phosphorus flux sequence are divided into the corresponding first characteristic sample sequence set, second characteristic sample sequence set, and third characteristic sample sequence set through the agricultural seasonal decomposition model SDA; The data in the first characteristic sample sequence set are grouped according to elements, and the data of each element are successively subjected to min-max normalization and standardized conversion and then input into the first trained multi-layer deep learning model together to obtain the corresponding second simulated first total nitrogen flux sequence and second simulated first total phosphorus flux sequence; The data in the second characteristic sample sequence set are grouped according to elements, and the data of each element are successively subjected to min-max normalization and standardized conversion and then input into the second trained multi-layer deep learning model together to obtain the corresponding second simulated second total nitrogen flux sequence and second simulated second total phosphorus flux sequence; The data in the third characteristic sample sequence set are grouped according to elements, and the data of each element are successively subjected to min-max normalization and standardized conversion and then input into the third trained multi-layer deep learning model together to obtain the corresponding second simulated third total nitrogen flux sequence and second simulated third total phosphorus flux sequence; In the final simulated total nitrogen flux sequence, for each time point during the pesticide application and fertilization period of nitrogen crops, the corresponding final simulated total nitrogen flux is equal to the second simulated first total nitrogen flux at the same time point in the second simulated first total nitrogen flux sequence; In the final simulated total nitrogen flux sequence, for each time point not during the pesticide application and fertilization period of nitrogen crops, the corresponding final simulated total nitrogen flux is equal to the average of the second simulated second total nitrogen flux in the second simulated second total nitrogen flux sequence and the second simulated third total nitrogen flux in the second simulated third total nitrogen flux sequence at the same time point; In the final simulated total phosphorus flux sequence, for each time point during the pesticide application and fertilization period of phosphorus crops, the corresponding final simulated total phosphorus flux is equal to the second simulated second total phosphorus flux at the same time point in the second simulated second total phosphorus flux sequence; In the final simulated total phosphorus flux sequence, for each time point not during the pesticide application and fertilization period of phosphorus crops, the corresponding final simulated total phosphorus flux is equal to the average of the second simulated first total phosphorus flux in the second simulated first total phosphorus flux sequence and the second simulated third total phosphorus flux in the second simulated third total phosphorus flux sequence at the same time point.

[0062] In this embodiment, the coupled model obtained by coupling the calibrated SWAT hydrological model, the agricultural seasonal decomposition model SDA, and the trained RABA model is the SDA-SWAT-RABA coupled model. The trained RABA model includes the first trained RABA model, the second trained RABA model, and the third trained RABA model. The to-be-processed spatial data sequence, meteorological data sequence, and hydrological and water quality data sequence are input into the calibrated SWAT hydrological model to obtain the corresponding preliminary simulated total nitrogen flux sequence and preliminary simulated total phosphorus flux sequence. The meteorological data in the meteorological data sequence in the to-be-processed sample, the preliminary simulated total nitrogen flux in the preliminary simulated total nitrogen flux sequence, and the preliminary simulated total phosphorus flux in the preliminary simulated total phosphorus flux sequence are divided into the corresponding first characteristic sample sequence set, second characteristic sample sequence set, and third characteristic sample sequence set by the agricultural seasonal decomposition model SDA. The data of each element in the first characteristic sample sequence set are input into the first trained RABA model after being normalized by min-max and standardized and converted together to obtain the corresponding secondarily simulated first total nitrogen flux sequence and secondarily simulated first total phosphorus flux sequence; the data of each element in the second characteristic sample sequence set are input into the second trained RABA model after being normalized by min-max and standardized and converted to obtain the corresponding secondarily simulated second total nitrogen flux sequence and secondarily simulated second total phosphorus flux sequence; the data of each element in the third characteristic sample sequence set are input into the third trained RABA model after being normalized by min-max and standardized and converted to obtain the corresponding secondarily simulated third total nitrogen flux sequence and secondarily simulated third total phosphorus flux sequence.

[0063] Step 6: Evaluate the coupled model.

[0064] The evaluation indicators selected for the evaluation of the coupled model are the same as those for the evaluation of the SWAT hydrological model, namely, the Pearson correlation coefficient PCC, the Nash coefficient NSE, and the percentage bias Pbias (the unit of the percentage bias Pbias is: %).

[0065] In this embodiment, the simulation accuracies of the total nitrogen flux and total phosphorus flux of the SDA-SWAT-RABA coupled model are compared with those of the SWAT hydrological model and the SWAT-RABA model. Figure 2 and Figure 3 are the fitting effect diagrams of the SDA-SWAT-RABA coupled model proposed by the present invention, the SWAT hydrological model, and the SWAT-RABA model in the simulation of the total nitrogen flux and total phosphorus flux. Tables 2 and 3 are respectively the comparison results of each evaluation indicator for the simulation of the total nitrogen flux and total phosphorus flux. Figure 4 and Figure 5 are the comparison diagrams of the simulation results of the SDA-SWAT-RABA coupled model proposed by the present invention, the SWAT hydrological model, and the SWAT-RABA model in the simulation of the total nitrogen flux and total phosphorus flux.

[0066] For the SWAT-RABA model, the spatial data sequence, meteorological data sequence, and hydrological and water quality data sequence in the sample to be processed are input into the calibrated SWAT hydrological model to obtain the preliminary simulated total nitrogen flux sequence and the preliminary simulated total phosphorus flux sequence; the meteorological data sequence, the preliminary simulated total nitrogen flux sequence, and the preliminary simulated total phosphorus flux sequence in the sample to be processed are input into the trained RABA model after min-max normalization and standard conversion to obtain the corresponding final simulated total nitrogen flux sequence and the final simulated total phosphorus flux sequence; the training method of the RABA model in the SWAT-RABA model is as follows: the meteorological data sequence, the preliminary simulated total nitrogen flux sequence, and the preliminary simulated total phosphorus flux sequence in the training set in step 1 are input into the RABA model, and the model is trained with the corresponding measured total phosphorus flux sequence and the measured total nitrogen flux sequence as the targets.

[0067] Table 2 Comparison of simulation results of the model on total nitrogen flux

[0068] From the data in Table 2, it can be seen that the SDA-SWAT-RABA coupling model proposed by the present invention has achieved better results in the simulation of total nitrogen flux. Compared with the SWAT hydrological model and the SWAT-RABA model, the Pearson correlation coefficient PCC of the SDA-SWAT-RABA coupling model has increased by 33% and 4% respectively during the verification period, the Nash coefficient NSE has increased by 34% and 3% respectively, and the percentage bias Pbias has decreased by 4.44% and 0.31%.

[0069] Table 3 Comparison of simulation results of the model on total phosphorus flux

[0070] From the data in Table 3, it can be seen that the SDA-SWAT-RABA coupling model proposed by the present invention has also achieved better results in the simulation of total phosphorus flux. Compared with the SWAT hydrological model and the SWAT-RABA model, the Pearson correlation coefficient PCC of the SDA-SWAT-RABA coupling model has increased by 31% and 3% respectively during the verification period, the Nash coefficient NSE has increased by 32% and 4% respectively, and the percentage bias Pbias has decreased by 6.55% and 4.54%.

[0071] Generally speaking, compared with other methods, the nitrogen and phosphorus simulation method proposed by the present invention has a Nash coefficient NSE for nitrogen and phosphorus simulation both exceeding 0.94, and the relative error between the simulated value and the measured value is less than 1%.

[0072] Example 2

[0073] A nitrogen and phosphorus simulation device for a coupling model based on agricultural season decomposition characteristics, which implements the nitrogen and phosphorus simulation method for a coupling model based on agricultural season decomposition characteristics described in Embodiment 1, includes: A hydrological model construction module for implementing Step 2 described in Embodiment 1; A construction module for the agricultural season decomposition model SDA for implementing Step 3 described in Embodiment 1; A construction and training module for a multi - element deep learning model for implementing Step 4 described in Embodiment 1; A model coupling module for implementing Step 5 described in Embodiment 1.

[0074] Embodiment 3

[0075] A computer device includes a memory and a processor. A computer program is stored in the memory, and when the processor executes the computer program, Steps 1 - 5 in the above - mentioned Embodiment 1 are implemented.

[0076] Embodiment 4

[0077] A computer - readable storage medium has a computer program stored thereon. When the computer program is executed by a processor, Steps 1 - 5 in the above - mentioned Embodiment 1 are implemented.

[0078] Embodiment 5

[0079] A computer program product includes a computer program. When the computer program is executed by a processor, Steps 1 - 5 in the above - mentioned Embodiment 1 are implemented.

[0080] The specific embodiments described herein are merely illustrative of the spirit of the present invention. Those skilled in the art to which the present invention pertains can make various modifications or supplements to the described specific embodiments or use similar ways to replace them, but will not deviate from the spirit of the present invention or exceed the scope defined by the appended claims.

Claims

1. A method for simulating nitrogen and phosphorus in a coupling model based on agricultural seasonal decomposition characteristics, characterized in that It includes the following steps: Step 1: Prepare a training set, which includes spatial data sequences, meteorological data sequences, and hydrographic and water quality data sequences corresponding to each region in the basin; the spatial data sequences include spatial data at different time points, and the spatial data includes land use information; the meteorological data sequences include meteorological data at different time points; the hydrographic and water quality data sequences include the measured phosphorus flux and the measured nitrogen flux at different time points; Step 2: Establish a hydrological model, calibrate the hydrological model using the training set to obtain the calibrated hydrological model; input the spatial data sequences and meteorological data sequences of the training set into the calibrated hydrological model to correspondingly obtain the preliminary simulated total nitrogen flux sequence and the preliminary simulated total phosphorus flux sequence of the basin; the preliminary simulated total nitrogen flux sequence includes the preliminary simulated total nitrogen flux at different time points, and the preliminary simulated total phosphorus flux sequence includes the preliminary simulated total phosphorus flux at different time points; Step 3: Take the meteorological data corresponding to a region in the entire basin at a time point in the training set as a regional meteorological data sample point; construct an agricultural seasonal decomposition model SDA, and the agricultural seasonal decomposition model SDA classifies the regional meteorological data sample points, the preliminary simulated total nitrogen flux, and the preliminary simulated total phosphorus flux corresponding to the training set based on the land use information, and each classification corresponds to a nitrogen impact period, a phosphorus impact period, and a weak nitrogen and phosphorus impact period; Step 4: Build a multi - layer deep learning model, and use the data classified by the agricultural seasonal decomposition model SDA from the regional meteorological data sample points, the preliminary simulated total nitrogen flux, and the preliminary simulated total phosphorus flux corresponding to the training set to train the multi - layer deep learning model to obtain the trained multi - layer deep learning model; Step 5: Couple the calibrated hydrological model, the agricultural seasonal decomposition model SDA, and the trained multi - layer deep learning model to obtain a coupled model; input the spatial data sequences, meteorological data sequences, and hydrographic and water quality data sequences of the sample to be processed into the coupled model to obtain the corresponding final simulated total nitrogen flux sequence and the final simulated total phosphorus flux sequence.

2. The nitrogen and phosphorus simulation method of the coupling model based on the agricultural season decomposition characteristics according to claim 1, wherein The specific steps of Step 3 are as follows: Step 3.1: Take the meteorological data corresponding to a region in the entire basin at a time point in the training set as a regional meteorological data sample point, and take the land use information corresponding to the time point of the regional meteorological data sample point as a basic attribute of the regional meteorological data sample point. The land use information includes the land type of the region at the corresponding time point, and the land type is divided into crop land types and non - crop land types. The crop land types are further divided into nitrogen crop land types, phosphorus crop land types, and non - nitrogen - phosphorus crop land types according to the type of crop fertilization and pesticide application; for the land use information corresponding to the crop land types, it also includes the cultivated land time characteristics, and the cultivated land time characteristics are the pesticide and fertilizer application period or the non - pesticide and fertilizer application period; Step 3.2: Classify the regional meteorological data sample points corresponding to the land type of nitrogen crops and the tillage time feature of the pesticide and fertilizer application period into the first characteristic sample sequence set, and also classify the preliminary simulated total nitrogen flux and preliminary simulated total phosphorus flux corresponding to the pesticide and fertilizer application period of the nitrogen crop land type into the first characteristic sample sequence set; Take the regional meteorological data sample points, preliminary simulated total nitrogen flux, and preliminary simulated total phosphorus flux corresponding to the same time point in the first characteristic sample sequence set as an agricultural season decomposition sample point in the first characteristic sample sequence set. The agricultural season decomposition sample points with continuous time points in the first characteristic sample sequence set form an agricultural season decomposition sample sequence of the first characteristic sample sequence set, and the first characteristic sample sequence set corresponds to the nitrogen influence period; Classify the regional meteorological data sample points corresponding to the land type of phosphorus crops and the tillage time feature of the pesticide and fertilizer application period into the second characteristic sample sequence set, and also classify the preliminary simulated total nitrogen flux and preliminary simulated total phosphorus flux corresponding to the pesticide and fertilizer application period of the phosphorus crop land type into the second characteristic sample sequence set; Take the regional meteorological data sample points, preliminary simulated total nitrogen flux, and preliminary simulated total phosphorus flux corresponding to the same time point in the second characteristic sample sequence set as an agricultural season decomposition sample point in the second characteristic sample sequence set. The agricultural season decomposition sample points with continuous time points in the first characteristic sample sequence set form an agricultural season decomposition sample sequence of the second characteristic sample sequence set, and the second characteristic sample sequence set corresponds to the phosphorus influence period; Classify the regional meteorological data sample points of the land type of non-crop land, the regional meteorological data sample points of the land type of non-nitrogen and phosphorus crops, and the regional meteorological data sample points with the tillage time feature of non-pesticide and fertilizer application period into the third characteristic sample sequence set, and also classify all the preliminary simulated total nitrogen flux and preliminary simulated total phosphorus flux into the third characteristic sample sequence set. Take the regional meteorological data sample points, preliminary simulated total nitrogen flux, and preliminary simulated total phosphorus flux corresponding to the same time point in the third characteristic sample sequence set as an agricultural season decomposition sample point in the third characteristic sample sequence set. The agricultural season decomposition sample points with continuous time points in the third characteristic sample sequence set form an agricultural season decomposition sample sequence of the third characteristic sample sequence set, and the third characteristic sample sequence set corresponds to the weak nitrogen and phosphorus influence period.

3. The nitrogen and phosphorus simulation method of the coupling model based on the decomposition characteristics of agricultural seasons according to claim 2, characterized in that, Step 3 also includes the following steps: Step 3.3: Group the data in the first characteristic sample sequence set according to elements, and perform min-max normalization and standardization conversion on the data corresponding to each element in turn to obtain the cleaned first characteristic sample sequence set; Group the data in the second characteristic sample sequence set according to elements, and perform min-max normalization and standardization conversion on the data corresponding to each element in turn to obtain the cleaned second characteristic sample sequence set; Group the data in the third characteristic sample sequence set according to elements, and perform min-max normalization and standardization conversion on the data corresponding to each element in turn to obtain the cleaned third characteristic sample sequence set; The elements corresponding to the data in the first characteristic sample sequence set, the second characteristic sample sequence set, and the third characteristic sample sequence set all include precipitation, maximum temperature, minimum temperature, solar radiation, relative humidity, wind speed, preliminary simulated total nitrogen flux, and preliminary simulated total phosphorus flux.

4. The nitrogen and phosphorus simulation method of the coupling model based on the agricultural season decomposition characteristics according to claim 3, characterized in that The training of the multi - variable deep learning model in step 4 includes the following steps: According to the measured phosphorus flux and measured nitrogen flux at each time point in the hydrological and water quality data sequences corresponding to each region of the basin, the measured total phosphorus flux sequence and the measured total nitrogen flux sequence of the entire basin are extracted. The measured total phosphorus flux sequence includes the measured total phosphorus flux at each time point, and the measured total nitrogen flux sequence includes the measured total nitrogen flux at each time point; Input the agricultural season decomposition sample sequence in the first characteristic sample sequence set after cleaning into the multi - variable deep learning model, and use the measured total phosphorus flux and measured total nitrogen flux corresponding to the time point sequence of the agricultural season decomposition sample sequence in the first characteristic sample sequence set after cleaning as the reference target for model training to obtain the first trained multi - variable deep learning model; Input the agricultural season decomposition sample sequence in the second characteristic sample sequence set after cleaning into the multi - variable deep learning model, and use the measured total phosphorus flux and measured total nitrogen flux corresponding to the time point sequence of the agricultural season decomposition sample sequence in the second characteristic sample sequence set after cleaning as the reference target for model training to obtain the second trained multi - variable deep learning model; Input the agricultural season decomposition sample sequence in the third characteristic sample sequence set after cleaning into the multi - variable deep learning model, and use the measured total phosphorus flux and measured total nitrogen flux corresponding to the time point sequence of the agricultural season decomposition sample sequence in the third characteristic sample sequence set after cleaning as the reference target for model training to obtain the third trained multi - variable deep learning model.

5. The nitrogen and phosphorus simulation method of the coupling model based on the decomposition characteristics of agricultural seasons according to claim 4, characterized in that In step 5, inputting the spatial data sequence, meteorological data sequence, and hydrological and water quality data sequence of the sample to be processed into the coupling model to obtain the corresponding final simulated total nitrogen flux sequence and final simulated total phosphorus flux sequence specifically includes the following steps: The spatial data sequence, meteorological data sequence, and hydrological and water quality data sequence in the sample to be processed are input into the calibrated hydrological model to obtain the corresponding preliminary simulated total nitrogen flux sequence and preliminary simulated total phosphorus flux sequence. The meteorological data in the meteorological data sequence of the sample to be processed, the preliminary simulated total nitrogen flux in the preliminary simulated total nitrogen flux sequence, and the preliminary simulated total phosphorus flux in the preliminary simulated total phosphorus flux sequence are divided into the corresponding first characteristic sample sequence set, second characteristic sample sequence set, and third characteristic sample sequence set through the agricultural season decomposition model SDA; The data in the first characteristic sample sequence set are grouped according to elements, and the data of each element are successively input into the first trained multi - variable deep learning model after min - max normalization and standard conversion to obtain the corresponding secondary simulated first total nitrogen flux sequence and secondary simulated first total phosphorus flux sequence; The data in the second characteristic sample sequence set are grouped according to elements, and the data of each element are successively subjected to min-max normalization and standardized conversion and then input into the second trained multi-source deep learning model together, to obtain the corresponding secondary simulated second total nitrogen flux sequence and secondary simulated second total phosphorus flux sequence; The data in the third characteristic sample sequence set are grouped according to elements, and the data of each element are successively subjected to min-max normalization and standardized conversion and then input into the third trained multi-source deep learning model together, to obtain the corresponding secondary simulated third total nitrogen flux sequence and secondary simulated third total phosphorus flux sequence; In the final simulated total nitrogen flux sequence, for each time point in the pesticide application and fertilization period of nitrogen crops, the corresponding final simulated total nitrogen flux is equal to the secondary simulated first total nitrogen flux at the same time point in the secondary simulated first total nitrogen flux sequence; In the final simulated total nitrogen flux sequence, for each time point not in the pesticide application and fertilization period of nitrogen crops, the corresponding final simulated total nitrogen flux is equal to the average of the secondary simulated second total nitrogen flux in the secondary simulated second total nitrogen flux sequence and the secondary simulated third total nitrogen flux in the secondary simulated third total nitrogen flux sequence at the same time point; In the final simulated total phosphorus flux sequence, for each time point in the pesticide application and fertilization period of phosphorus crops, the corresponding final simulated total phosphorus flux is equal to the secondary simulated second total phosphorus flux at the same time point in the secondary simulated second total phosphorus flux sequence; In the final simulated total phosphorus flux sequence, for each time point not in the pesticide application and fertilization period of phosphorus crops, the corresponding final simulated total phosphorus flux is equal to the average of the secondary simulated first total phosphorus flux in the secondary simulated first total phosphorus flux sequence and the secondary simulated third total phosphorus flux in the secondary simulated third total phosphorus flux sequence at the same time point.

6. The nitrogen and phosphorus simulation method of the coupling model based on the decomposition characteristics of agricultural seasons according to claim 5, characterized in that, The multi-source deep learning model in step 4 is the RABA model. The RABA model successively includes a residual feature perception module RA and a bidirectional attention prediction module BA; the residual feature perception module RA successively includes a CNN input layer, a convolutional layer, a pooling layer, and a CNN output layer; the bidirectional attention prediction module BA successively includes a BILSTM input layer, an LSTM layer, an attention layer, an activation layer, and a BILSTM output layer; the input data of the CNN input layer are added to the output data of the CNN output layer through a skip connection and then input into the BILSTM input layer.

7. The nitrogen and phosphorus simulation method of the coupling model based on the decomposition characteristics of agricultural seasons according to claim 6, wherein The attention layer adopts a channel attention mechanism module SE.

8. A method for simulating nitrogen and phosphorus in a coupled model based on agricultural season decomposition characteristics according to any one of claims 1 to 7, characterized in that The hydrological model is the SWAT hydrological model.

9. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that, When the processor executes the computer program, it implements steps 1-5 of the nitrogen and phosphorus simulation method described in any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements steps 1-5 of the nitrogen and phosphorus simulation method described in any one of claims 1 to 8.

Images