A water and fertilizer integrated management method and system based on high-density planting and high-yield cultivation of soybeans

Abstract

The application relates to the field of intelligent agriculture and crop cultivation technology, in particular to a water and fertilizer integrated management method and system based on soybean high-density high-yield cultivation, which comprises a digital twin model and a machine learning high-density frame recommendation model, can fuse variety characteristics, soil data and meteorological conditions, and recommend the optimal seeding density for a specific plot, and breaks through the limitation of traditional experience or regional unified density. Based on the digital twin model and future weather forecast, the application can predict the crop growth state and soil water and fertilizer dynamics in advance, and accurately calculate the irrigation and fertilization demand accordingly, change the experience irrigation and fertilization into prediction decision fertilization, realize the on-demand accurate supply of water and fertilizer resources, and improve the utilization efficiency.

Description

Technical Field

[0001] This invention relates to the field of smart agriculture and crop cultivation technology, specifically to a water and fertilizer integrated management method and system based on high-density, high-yield soybean cultivation. Background Technology

[0002] As an important food and economic crop, soybeans have attracted much attention for their high-yield and high-efficiency cultivation techniques. Traditional soybean planting and management rely heavily on farmers' experience, using fixed sowing densities and water and fertilizer management patterns, lacking dynamic responses to weather, soil, and crop growth conditions, resulting in low water and fertilizer utilization rates and limited yield increases.

[0003] In recent years, precision agriculture technologies, such as fertigation, have been promoted. However, their decision-making is mostly based on static soil tests or simple growth period divisions, making it difficult to adapt to the dynamic needs of crop growth and real-time changes in environmental conditions. Therefore, applying fertilizer and irrigation according to fixed cycles or experience, without making dynamic adjustments based on the real-time needs of crops and the effective supply of soil, can easily lead to water and fertilizer waste or stress-induced yield reduction.

[0004] Existing technologies urgently need to develop a method for integrated water and fertilizer management of high-yield soybean dense planting, which deeply integrates crop growth mechanisms, real-time environmental perception, and intelligent decision control, to achieve digital, precise, and intelligent management of water and fertilizer throughout the entire process from sowing to harvest. Summary of the Invention

[0005] The purpose of this invention is to provide an integrated water and fertilizer management method based on high-density soybean cultivation, in order to solve the technical problem that existing technologies apply water and fertilizer according to fixed cycles or experience, failing to make dynamic adjustments based on real-time crop needs and effective soil supply, which easily leads to water and fertilizer waste or stress-induced yield reduction.

[0006] To solve the above-mentioned technical problems, the present invention specifically provides the following technical solution:

[0007] A water and fertilizer integration management method based on high-density, high-yield soybean cultivation includes the following steps:

[0008] Using historical meteorological, soil, variety parameters, and field trial data, a digital twin model was constructed to dynamically simulate the growth process of soybean populations.

[0009] Based on the growth simulation of the digital twin model, a high-yield soybean high-density planting structure recommendation model is constructed to integrate variety, soil, meteorological and yield data to recommend the optimal planting density for a specific plot.

[0010] The high-yield soybean dense planting population structure is recommended based on the dense planting structure model output, and the digital twin model is calibrated based on the canopy structure, soil moisture, and multispectral inversion vegetation index obtained by UAV or sensor monitoring.

[0011] By inputting future weather forecast data into a calibrated digital twin model for growth simulation, the predicted results of growth status, soil moisture and nutrient dynamics of high-yield soybean dense planting populations are obtained.

[0012] Based on the predicted growth status, soil moisture and nutrient dynamics of high-yield soybean dense planting populations, the water and fertilizer requirements are calculated, and the drip irrigation system is controlled based on the water and fertilizer requirements to carry out integrated irrigation and fertilization operations for soybean planting.

[0013] As a preferred embodiment of the present invention, the method for constructing the digital twin model includes:

[0014] Based on historical meteorological, soil, variety parameters and field test data, we constructed morphological growth rules, soil moisture balance rules and nitrogen, phosphorus and potassium nutrient dynamic rules for the soybean population growth process.

[0015] Based on historical meteorological, soil, variety parameters and field test data, the physical entities of soybean population, meteorological and soil environment are mapped to virtual entities using the visualization engine Unity, resulting in a 3D model of the soybean population growth process.

[0016] The digital twin model is formed by combining 3D modeling of soybean population growth process, morphological growth rules, soil moisture balance rules, and dynamic rules of nitrogen, phosphorus and potassium nutrients.

[0017] As a preferred embodiment of the present invention, the method for constructing the densely planted architecture recommendation model includes:

[0018] The morphological growth rules are based on the plant height model and leaf area index (LAI) model obtained by fitting the Logistic growth curve, wherein:

[0019] The plant height model In the formula, Let be the plant height at time t. For variety c and planting density The corresponding potential maximum plant height, The plant height growth rate, This is the time point when plant height increases the most;

[0020] The leaf area index (LAI) model In the formula, Let be the leaf area index at time t. Planting density and water and fertilizer supply The corresponding potential maximum LAI, For LAI growth rate, This is the time when plant height increases the most. This is a coercion correction term;

[0021] The soil moisture balance rules In the formula, , , , , These represent the soil volumetric water content at times t+1 and t, respectively. Let be the rainfall at time t. Let be the irrigation amount at time t. Let be the crop transpiration at time t. Let be the amount of soil seepage at time t. Let be the surface runoff at time t. For time step, For soil layer thickness, These are the basic crop coefficient, water stress coefficient, and soil evaporation coefficient. The reference evapotranspiration rate is calculated using the Penman-Monteith formula. Let be the field water holding capacity at time t. Let be the soil water storage at time t. CN represents the number of curves in the SCS curve number method;

[0022] The rules for nitrogen, phosphorus, and potassium nutrient requirements In the formula, For the first The amount of nutrients that need to be supplemented, For the first Total nutrient requirements For the first The percentage of each nutrient in total demand. For the first The current stock or existing level of the nutrient. For the first The conversion or efficiency coefficient of the current stock of nutrients. For the first The efficiency or output coefficient of nutrient supplementation.

[0023] As a preferred embodiment of the present invention, the method for constructing the densely planted architecture recommendation model includes:

[0024] In the digital twin model, the soybean varieties, soil, and weather data of a fixed plot are used to simulate the growth of the plot at different planting densities, and the yield data of each planting density is recorded.

[0025] The planting density corresponding to the maximum yield data is taken as the optimal sowing density for this plot;

[0026] By combining variety, soil, and meteorological data with recommended planting density into a dataset, a convolutional neural network is trained to obtain a dense planting architecture recommendation model that predicts the optimal sowing density based on the input variety, soil, and meteorological data.

[0027] As a preferred embodiment of the present invention, the calibration method for the digital twin model includes:

[0028] Adjust the planting density in the digital twin model to the optimal sowing density for a specific plot;

[0029] The plant height, leaf area index (LAI), soil moisture, and multispectral inversion vegetation index corresponding to the canopy structure at the recommended planting density were obtained by monitoring with drones or sensors as observation values.

[0030] In the digital twin model, key uncertain parameters are selected as calibration parameters based on the mechanistic rules of the soybean population growth process.

[0031] The plant height, leaf area index (LAI), soil moisture, multispectral inversion vegetation index, and parameters to be calibrated are combined into a state vector based on the mechanistic rules of soybean population growth process in the digital twin model.

[0032] By using the ensemble Kalman filter algorithm EnKF, the state vector is updated through Kalman gain based on the observations and the state vector. This prompts the digital twin model to continuously assimilate new observations to update the model parameters and complete the calibration of the digital twin model, resulting in a digital twin model for dynamically simulating the population growth process under the dense planting and high-yield soybean population structure.

[0033] As a preferred embodiment of the present invention, the method for calculating water and fertilizer requirements based on the predicted growth status, soil moisture, and nutrient dynamics of a high-yield soybean densely planted population includes:

[0034] The future soil volumetric water content of a specific plot can be obtained using soil moisture balance rules. The target values ​​of soil volumetric water content for the corresponding crop development stage of a high-yield, densely planted soybean population were obtained. And calculate the future irrigation water volume. ;

[0035] The future amount of nitrogen fertilizer to be irrigated is calculated based on the nitrogen, phosphorus, and potassium nutrient requirements rules. Future irrigation phosphate fertilizer amount Future irrigation potassium fertilizer amount .

[0036] As a preferred embodiment of the present invention, the present invention provides an integrated water and fertilizer management system based on high-density soybean cultivation, which is applied to an integrated water and fertilizer management method for high-density soybean cultivation. The system includes:

[0037] The digital twin model building module is used to construct a digital twin model for dynamically simulating the growth process of soybean populations by utilizing historical meteorological, soil, variety parameters and field test data.

[0038] The dense planting structure recommendation module is used to construct a dense planting structure recommendation model for soybean dense planting high-yield population structure that integrates variety, soil, meteorological and yield data to recommend the optimal planting density for a specific plot based on the growth simulation of the digital twin model.

[0039] The model calibration module is used to recommend the soybean high-yield population structure based on the dense planting architecture, and to calibrate the digital twin model based on the canopy structure, soil moisture, and multispectral inversion vegetation indices obtained by UAV or sensor monitoring.

[0040] The growth prediction module is used to input future weather forecast data into a calibrated digital twin model to simulate growth and obtain prediction results of the growth status, soil moisture and nutrient dynamics of high-yield soybean dense planting populations.

[0041] The water and fertilizer decision control module is used to calculate the water and fertilizer requirements based on the growth status of high-yield soybean dense planting populations and the prediction results of soil moisture and nutrient dynamics, and to control the drip irrigation system to carry out integrated irrigation and fertilization operations for soybean planting based on the water and fertilizer requirements.

[0042] As a preferred embodiment of the present invention, the digital twin model construction module includes:

[0043] The rule-building unit is used to construct morphological growth rules, soil moisture balance rules, and nitrogen, phosphorus, and potassium nutrient dynamic rules for the soybean population growth process based on historical meteorological, soil, variety parameters, and field test data.

[0044] The 3D visualization unit is used to map the physical entities of soybean populations, weather, and soil environment into virtual entities using the Unity visualization engine, based on historical meteorological, soil, variety parameters, and field test data, to obtain a 3D model of the soybean population growth process.

[0045] The model integration unit is used to combine 3D modeling of soybean population growth process, morphological growth rules, soil moisture balance rules, and dynamic rules of nitrogen, phosphorus and potassium nutrients to form the digital twin model.

[0046] As a preferred embodiment of the present invention, the densely planted architecture recommendation module includes:

[0047] The simulation execution unit is used in the digital twin model to simulate the growth of soybean varieties, soil and meteorological data of a fixed plot of land at different planting densities, and to record the yield data of each planting density of the plot.

[0048] The optimal density determination unit is used to determine the planting density corresponding to the maximum yield data as the optimal sowing density for the plot.

[0049] The model training unit is used to combine variety, soil, meteorological data and recommended planting density into a dataset, train a convolutional neural network, and obtain a dense planting architecture recommendation model that predicts the optimal sowing density based on the input variety, soil and meteorological data.

[0050] As a preferred embodiment of the present invention, the water and fertilizer decision control module includes:

[0051] The irrigation decision unit obtains the future soil volumetric water content of a specific plot through soil moisture balance rules. The target values ​​of soil volumetric water content for the corresponding crop development stage of a high-yield, densely planted soybean population were obtained. And calculate the future irrigation water volume. ;

[0052] The fertilization decision unit is used to calculate the future amount of nitrogen fertilizer to be applied during irrigation, based on the nitrogen, phosphorus, and potassium nutrient requirement rules. Future irrigation phosphate fertilizer amount Future irrigation potassium fertilizer amount .

[0053] Compared with the prior art, the present invention has the following advantages:

[0054] This invention, by constructing a digital twin model and a machine learning-based dense planting architecture recommendation model, can integrate variety characteristics, soil data, and meteorological conditions to recommend the optimal sowing density for a specific plot of land. This breaks through the limitations of traditional reliance on experience or uniform regional density and lays the foundation for a high-yield, densely planted population structure.

[0055] Based on digital twin models and future weather forecasts, this invention can predict crop growth status and soil water and fertilizer dynamics in advance, and accurately calculate irrigation and fertilization needs accordingly. It transforms experience-based irrigation and fertilization into predictive decision-making fertilization, realizing the precise supply of water and fertilizer resources on demand and improving utilization efficiency. Attached Figure Description

[0056] To more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary, and those skilled in the art can derive other embodiments based on the provided drawings without creative effort.

[0057] Figure 1 This is a flowchart of an integrated water and fertilizer management method for high-yield, densely planted soybean cultivation, provided in an embodiment of the present invention.

[0058] Figure 2 A block diagram of an integrated water and fertilizer management system for high-yield, dense-planting soybean cultivation provided in an embodiment of the present invention. Detailed Implementation

[0059] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0060] like Figure 1 As shown, this invention provides a water and fertilizer integration management method based on high-density, high-yield soybean cultivation, comprising the following steps:

[0061] Using historical meteorological, soil, variety parameters, and field trial data, a digital twin model was constructed to dynamically simulate the growth process of soybean populations.

[0062] Based on growth simulation using a digital twin model, a high-yield soybean dense planting population structure recommendation model is constructed to integrate variety, soil, meteorological, and yield data to recommend the optimal planting density for specific plots.

[0063] The high-yield soybean dense planting population structure is recommended based on the dense planting structure model output, and the digital twin model is calibrated based on the canopy structure, soil moisture, and multispectral inversion vegetation index obtained by UAV or sensor monitoring.

[0064] By inputting future weather forecast data into a calibrated digital twin model for growth simulation, the predicted results of growth status, soil moisture and nutrient dynamics of high-yield soybean dense planting populations are obtained.

[0065] Based on the predicted growth status, soil moisture and nutrient dynamics of high-yield soybean dense planting populations, the water and fertilizer requirements are calculated, and the drip irrigation system is controlled based on the water and fertilizer requirements to carry out integrated irrigation and fertilization operations for soybean planting.

[0066] Historical meteorological data, derived from daily observations of meteorological stations in the target area over the past 10 years, includes daily maximum / minimum temperatures, precipitation, sunshine hours, average wind speed, and relative humidity. The data is in CSV or NetCDF format and is used for model initialization and long-term analysis.

[0067] Soil parameters were measured through pre-sowing gridded sampling (e.g., 30m × 30m), including soil texture (percentage of sand, silt, and clay particles) and volumetric water content (θ) of the 0-20cm and 20-40cm soil layers.FC ), permanent wilting point (θ) WP The initial contents of organic matter, pH value, and available nitrogen, available phosphorus, and available potassium were determined, and a spatial distribution map of soil properties was constructed.

[0068] Variety parameters, derived from variety approval reports or specialized trials, include: growth stage group, photoperiod sensitivity, potential maximum plant height range, potential maximum leaf area index, and basic crop coefficient (K) for each growth stage. cb ), dry matter partition coefficient curve. Key growth parameters (such as r in the Logistic model) h , r l This can be obtained by fitting observation data from variety comparison trials.

[0069] Drip irrigation system parameters are inherent system attributes, including: dripper rated flow rate (e.g., 2 L / h), dripper spacing, capillary tube spacing, irrigated area, fertilizer applicator stock solution tank capacity, and maximum fertilizer injection rate. These parameters are used to convert the decision water volume (mm) and fertilizer amount (kg / ha) into specific irrigation duration (minutes) and stock solution pumping duration (minutes).

[0070] Methods for constructing digital twin models include:

[0071] Based on historical meteorological, soil, variety parameters and field test data, we constructed morphological growth rules, soil moisture balance rules and nitrogen, phosphorus and potassium nutrient dynamic rules for the soybean population growth process.

[0072] Based on historical meteorological, soil, variety parameters and field test data, the physical entities of soybean population, meteorological and soil environment are mapped to virtual entities using the visualization engine Unity, resulting in a 3D model of the soybean population growth process.

[0073] A digital twin model is formed by combining 3D modeling of soybean population growth process, morphological growth rules, soil moisture balance rules, and dynamic rules of nitrogen, phosphorus, and potassium nutrients.

[0074] This invention collects historical meteorological data (temperature, rainfall, radiation, etc.), soil physicochemical property data (texture, bulk density, field water holding capacity, initial nutrient content, etc.), soybean variety parameters (growth period, potential growth rate, etc.), and multi-year field trial data for the target area. Based on this, it first forms three core rule bases for constructing a digital twin model through mechanistic analysis:

[0075] First, morphological growth rules: A modified Logistic growth curve was used to construct plant height and leaf area index (LAI) models, respectively. The maximum potential value (H) in the model... max LAI maxThe formula was designed as a function of variety (c), planting density (ρ), and water and fertilizer supply (g), which more realistically reflects the impact of group competition and resource supply on individual growth.

[0076] Second, soil moisture balance rules: The soil water balance equation based on physical processes is adopted, and the Penman-Monteith formula is integrated to calculate evapotranspiration and the SCS curve number method is used to calculate runoff, realizing dynamic simulation of multiple processes such as rainfall, irrigation, evapotranspiration, infiltration and runoff.

[0077] Third, dynamic rules for nitrogen, phosphorus, and potassium nutrients: Constructing the dynamic relationship between soil nutrient supply and crop absorption, and providing a core calculation formula F for water and fertilizer decision-making. i .

[0078] Subsequently, using visualization engines such as Unity, the aforementioned mechanism model was combined with 3D visualization modeling technology to create a virtual entity that can intuitively display the soybean population structure, canopy dynamics, and soil profile, forming an intuitive and interactive digital twin.

[0079] In this way, the mechanistic model is deeply integrated with 3D visualization to construct a computable and displayable digital twin, rather than a simple data panel or growth curve. Density and water / fertilizer are directly embedded as key input variables into the growth model, enhancing its interpretability and control interface.

[0080] Furthermore, this enables digital twin models not only to perform quantitative calculations but also to provide intuitive visualizations of population growth, helping agricultural technicians and farmers understand canopy development and competition under dense planting conditions. The mechanistic model provides a solid theoretical basis for core decision-making, ensuring the agronomical rationality of the decision-making logic.

[0081] Methods for constructing densely-planted architecture recommendation models include:

[0082] The morphological growth rules are based on the plant height model and leaf area index (LAI) model obtained by fitting the Logistic growth curve, where:

[0083] Plant height model In the formula, Let be the plant height at time t. For variety c and planting density The corresponding potential maximum plant height, The plant height growth rate, This is the time point when plant height increases the most;

[0084] Leaf Area Index (LAI) Model In the formula, Let be the leaf area index at time t. Planting density and water and fertilizer supply The corresponding potential maximum LAI, For LAI growth rate, This is the time when plant height increases the most. This is a coercion correction term;

[0085] Soil Moisture Balance Rules In the formula, , , , , These represent the soil volumetric water content at times t+1 and t, respectively. Let be the rainfall at time t. Let be the irrigation amount at time t. Let be the crop transpiration at time t. Let be the amount of soil seepage at time t. Let be the surface runoff at time t. For time step, For soil layer thickness, These are the basic crop coefficient, water stress coefficient, and soil evaporation coefficient. The reference evapotranspiration rate is calculated using the Penman-Monteith formula. Let be the field water holding capacity at time t. Let be the soil water storage at time t. CN represents the number of curves in the SCS curve number method;

[0086] Nitrogen, phosphorus, and potassium nutrient requirements rules In the formula, For the first The amount of nutrients that need to be supplemented, For the first Total nutrient requirements For the first The percentage of each nutrient in total demand. For the first The current stock or existing level of the nutrient. For the first The conversion or efficiency coefficient of the current stock of nutrients. For the first The efficiency or output coefficient of nutrient supplementation.

[0087] Methods for constructing densely-planted architecture recommendation models include:

[0088] In the digital twin model, the soybean varieties, soil, and weather data of a fixed plot are used to simulate the growth of the plot at different planting densities, and the yield data of each planting density is recorded.

[0089] The planting density corresponding to the maximum yield data is taken as the optimal sowing density for this plot;

[0090] By combining variety, soil, and meteorological data with recommended planting density into a dataset, a convolutional neural network is trained to obtain a dense planting architecture recommendation model that predicts the optimal sowing density based on the input variety, soil, and meteorological data.

[0091] In the constructed digital twin model, for a specific plot (with fixed soil and variety data), long-term historical meteorological data is input, and then multiple full-growth-cycle growth simulations are performed within a preset reasonable density range at different planting densities ρ. Each simulation runs to maturity, and the final yield is recorded. The density with the highest simulated yield is selected as the theoretically optimal planting density for that plot-variety combination. A large number of such "plot condition-optimal density" sample pairs are accumulated and used as a training dataset to train a convolutional neural network (CNN). This CNN learns to directly map the recommended optimal planting density from the input multidimensional data (variety, soil parameters, meteorological characteristics).

[0092] Furthermore, a two-stage hybrid modeling strategy is adopted, combining the batch generation of training data using mechanistic models with the learning and recommendation of machine learning models. This fully leverages the advantages of mechanistic models in conducting large-scale planting experiments in virtual space at low cost and without risk, overcoming the bottlenecks of long field trial cycles, high costs, and limited data.

[0093] This makes it possible to recommend customized optimal densities for each plot of land, achieving a leap from uniform density for a region to plot-specific density. This lays a scientific foundation for tapping the production potential of plots and achieving high yields through dense planting. This is the first and crucial step in achieving high yields.

[0094] This invention employs a convolutional neural network architecture to process variety-soil data and meteorological data separately, then fuses the features, and finally outputs the recommended density through a fully connected layer.

[0095] The convolutional neural network consists of three branches: a soil data branch, a meteorological data branch, and a variety feature branch, as well as a feature fusion and regression layer.

[0096] The encoding stage of the soil data branch includes two convolutional blocks, where:

[0097] The first convolutional block is used for local feature extraction, consisting of a 1D convolutional layer with 32 1×3 convolutional kernels, batch normalization, and ReLU activation function;

[0098] The second convolutional block is used for feature abstraction, consisting of a 1D convolutional layer with 64 1×3 convolutional kernels, batch normalization and ReLU activation function, and then a 64-dimensional feature vector is obtained through a global average pooling layer.

[0099] The encoding stage of the meteorological data branch includes three convolutional blocks, of which:

[0100] The first convolutional block is used for local feature extraction, consisting of a 1D convolutional layer with 32 1×3 convolutional kernels, batch normalization, and ReLU activation function;

[0101] The second convolutional block is used for feature abstraction, consisting of a 1D convolutional layer with 64 1×3 convolutional kernels, batch normalization, and ReLU activation function;

[0102] The third convolutional block is used for high-level feature extraction, consisting of a 1D convolutional layer with 128 1×3 convolutional kernels, batch normalization and ReLU activation function, and then a 128-dimensional feature vector is obtained through a global average pooling layer.

[0103] The variety feature branch includes a fully connected layer that maps the variety feature vector to 64 dimensions and uses the ReLU activation function.

[0104] Feature fusion and regression layers include:

[0105] The 64-dimensional features of the soil branch, the 128-dimensional features of the meteorological branch, and the 64-dimensional features of the variety branch are concatenated into a 256-dimensional feature vector.

[0106] Then, two fully connected layers (the first layer has 128 neurons, and the second layer has 64 neurons, both using the ReLU activation function).

[0107] Finally, the optimal seeding density prediction is obtained through an output layer (1 neuron, using the ReLU activation function).

[0108] The training loss of this convolutional neural network is the mean square error between the network's output and the true value of the optimal seeding density.

[0109] Training and deployment: The model is generated using a digital twin model, containing a large number of optimal density labels under different conditions. It utilizes the Adam optimizer and employs dynamic learning rate adjustment and early stopping strategies. The model features a lightweight design, supports online learning and uncertainty quantification, and is easy to deploy and apply in real-world production environments.

[0110] In other words, the variety feature branch: the input is a vector composed of variety one-hot encoding and numerical features (such as the number of days of the growing season), which is processed through a fully connected embedding layer.

[0111] Soil Feature Branch: The input consists of soil data with multiple soil layers and multiple indicators (such as N, P, K, OM, pH), which is treated as a two-dimensional matrix (indicator × soil layer). Features are extracted using two consecutive 1D convolutional layers (along the indicator dimension), followed by global pooling.

[0112] Meteorological Feature Branch: The input consists of daily meteorological data (temperature, precipitation, radiation) during the key growth period (e.g., from sowing to flowering), forming a two-dimensional matrix (meteorological elements × time). A one-dimensional temporal convolutional layer (kernel along the time dimension) is used to extract temporal patterns, followed by an attention layer to focus on key weather periods.

[0113] The feature vectors of the three branches are fused in the splicing layer, and then nonlinearly transformed through three fully connected layers. Finally, the recommended density (10,000 plants / hectare) is given by the linear output layer.

[0114] The calibration methods for digital twin models include:

[0115] Adjust the planting density in the digital twin model to the optimal sowing density for a specific plot;

[0116] The plant height, leaf area index (LAI), soil moisture, and multispectral inversion vegetation index corresponding to the canopy structure at the recommended planting density were obtained by monitoring with drones or sensors as observation values.

[0117] In the digital twin model, key uncertain parameters are selected as calibration parameters based on the mechanistic rules of the soybean population growth process.

[0118] The plant height, leaf area index (LAI), soil moisture, multispectral inversion vegetation index, and parameters to be calibrated are combined into a state vector based on the mechanistic rules of soybean population growth process in the digital twin model.

[0119] By using the ensemble Kalman filter algorithm EnKF, the state vector is updated through Kalman gain based on the observations and the state vector. This prompts the digital twin model to continuously assimilate new observations to update the model parameters and complete the calibration of the digital twin model, resulting in a digital twin model for dynamically simulating the population growth process under the dense planting and high-yield soybean population structure.

[0120] After sowing at the recommended optimal density in the field, a network of drones and soil sensors is deployed. Drones periodically collect canopy multispectral images, retrieving plant height, LAI (Lower Id), and vegetation indices (such as NDVI); sensors monitor soil volumetric water content in real time. These data are used as real-world "observations." In the digital twin model, key mechanistic parameters with significant initial uncertainty (such as growth rate r) are selected. h r l Water stress coefficient K s (e.g., plant height, LAI, soil moisture content) are used as parameters to be calibrated. The current simulated state of the model (plant height, LAI, soil moisture content, etc.) is combined with these parameters to form a state vector. Using the ensemble Kalman filter (EnKF) algorithm, the observed data is continuously assimilated into the model, and the state vector is dynamically adjusted to update the model parameters, so that the simulated trajectory of the model continuously approaches the actual observed values.

[0121] The parameters to be calibrated are selected from mechanistic parameters that are sensitive to model output and have high uncertainty, including: leaf area growth rate (r l ), water stress coefficient (K) s The response curve parameters and nitrogen use efficiency coefficient (β) of ) N ), soil available nitrogen content (N soil Soil moisture content (θ). A reasonable prior distribution (e.g., uniform distribution) should be set for each parameter.

[0122] State vector X t Defined as: X t = [Model state variables; parameters to be calibrated] = [LAI(t), θ(t), N] soil (t); r l , K s , β N ].

[0123] Observed value Y t For: Y t = [LAI obs (UAV multispectral inversion), θ obs (Deep sensor monitoring values)

[0124] The observation operator H takes LAI(t) from the state vector for LAI and θ(t) from the corresponding soil layer directly for θ. During calibration, the parameters of each instance are randomly sampled from the prior distribution. After each assimilation of the observation data, the Kalman gain is calculated to update the state and parameters of all instances, so that the mean of the model set converges to the observation values.

[0125] Introducing the advanced technology of data assimilation into field crop management enables online, dynamic calibration of digital twin models. The models are no longer static but possess learning capabilities, adapting to the actual weather and crop performance of specific plots and years.

[0126] This significantly improves the model's prediction accuracy during the current growing season. The calibrated model can more accurately reflect the actual growth process and stress status of crops in the plot, providing a reliable guarantee for subsequent prediction-based water and fertilizer decisions and reducing decision-making biases caused by initial model errors or weather anomalies.

[0127] Based on the predicted growth status, soil moisture, and nutrient dynamics of high-yield soybean dense planting populations, the methods for calculating water and fertilizer requirements include:

[0128] The future soil volumetric water content of a specific plot can be obtained using soil moisture balance rules. The target values ​​of soil volumetric water content for the corresponding crop development stage of a high-yield, densely planted soybean population were obtained. And calculate the future irrigation water volume. , The field water holding capacity θ was set according to the water requirement characteristics of soybeans at different growth stages. FC Specific percentages: 60-65% during seedling stage (VE-VC), 70-75% during branching to flowering (V1-R2), 75-80% during pod formation and pod filling (R3-R6), and 60% during maturity (R7-R8).

[0129] The future amount of nitrogen fertilizer to be irrigated is calculated based on the nitrogen, phosphorus, and potassium nutrient requirements rules. Future irrigation phosphate fertilizer amount Future irrigation potassium fertilizer amount .

[0130] This invention inputs weather forecast data for the next 7-15 days into a calibrated digital twin model to perform advanced growth simulation. The predicted output includes the crop growth status and soil water and fertilizer dynamics for each day in the future.

[0131] Irrigation Decision: System Analysis and Prediction of Soil Moisture Content Sequence When the predicted value will fall below the lower limit threshold of suitable water for the current growth stage of the crop on a future day. When the irrigation decision is triggered, the required replenishment water volume is calculated. Multiplying by 10 converts the unit from m (water depth) to mm (common irrigation unit).

[0132] Fertilization decisions: based on model predictions of available nitrogen, phosphorus, and potassium stocks in the soil. Combined with the nutrient requirements of the crop in the next critical growth stage and stage absorption ratio Using the nutrient requirement formula, the topdressing requirements for nitrogen, phosphorus, and potassium are calculated respectively. , , .

[0133] During the execution phase, the calculated irrigation volume and nutrient requirements are converted into control commands for the drip irrigation system (such as irrigation duration, mother liquor pumping ratio and time), so that the precise supply of water and nutrients can be completed simultaneously during a single irrigation process.

[0134] This enables predictive preventative management. Decisions are not based on current or past water and fertilizer deficiencies, but on predictions of future conditions, proactively intervening before stress occurs to keep crops near their optimal growing environment.

[0135] It minimizes the irreversible impact of water or nutrient stress on yield. It achieves precise and synchronized supply of water and fertilizer in terms of time, quantity, and ratio, significantly improving water and fertilizer utilization efficiency, saving resources, and reducing environmental pollution risks. It is a core element in achieving high yield, high efficiency, and environmental protection.

[0136] like Figure 2 As shown, this invention provides an integrated water and fertilizer management system for high-density, high-yield soybean cultivation, applicable to an integrated water and fertilizer management method for high-density, high-yield soybean cultivation. The system includes:

[0137] The digital twin model building module is used to construct a digital twin model for dynamically simulating the growth process of soybean populations by utilizing historical meteorological, soil, variety parameters and field test data.

[0138] The dense planting architecture recommendation module is used for growth simulation based on digital twin models to construct a dense planting architecture recommendation model for soybean dense planting high-yield population architecture that integrates variety, soil, meteorological and yield data to recommend the optimal planting density for a specific plot.

[0139] The model calibration module is used to recommend the soybean high-yield population structure based on the dense planting architecture, and to calibrate the digital twin model based on the canopy structure, soil moisture, and multispectral inversion vegetation indices obtained by UAV or sensor monitoring.

[0140] The growth prediction module is used to input future weather forecast data into a calibrated digital twin model to simulate growth and obtain prediction results of the growth status, soil moisture and nutrient dynamics of high-yield soybean dense planting populations.

[0141] The water and fertilizer decision control module is used to calculate the water and fertilizer requirements based on the growth status of high-yield soybean dense planting populations and the prediction results of soil moisture and nutrient dynamics, and to control the drip irrigation system to carry out integrated irrigation and fertilization operations for soybean planting based on the water and fertilizer requirements.

[0142] The digital twin model building module includes:

[0143] The rule-building unit is used to construct morphological growth rules, soil moisture balance rules, and nitrogen, phosphorus, and potassium nutrient dynamic rules for the soybean population growth process based on historical meteorological, soil, variety parameters, and field test data.

[0144] The 3D visualization unit is used to map the physical entities of soybean populations, weather, and soil environment into virtual entities using the Unity visualization engine, based on historical meteorological, soil, variety parameters, and field test data, to obtain a 3D model of the soybean population growth process.

[0145] The model integration unit is used to combine 3D modeling of soybean population growth process, morphological growth rules, soil moisture balance rules, and dynamic rules of nitrogen, phosphorus and potassium nutrients to form a digital twin model.

[0146] Recommended modules for densely planted architecture include:

[0147] The simulation execution unit is used in the digital twin model to simulate the growth of soybean varieties, soil and meteorological data of a fixed plot of land at different planting densities, and to record the yield data of each planting density of the plot.

[0148] The optimal density determination unit is used to determine the planting density corresponding to the maximum yield data as the optimal sowing density for the plot.

[0149] The model training unit is used to combine variety, soil, meteorological data and recommended planting density into a dataset, train a convolutional neural network, and obtain a dense planting architecture recommendation model that predicts the optimal sowing density based on the input variety, soil and meteorological data.

[0150] The water and fertilizer decision control module includes:

[0151] The irrigation decision unit obtains the future soil volumetric water content of a specific plot through soil moisture balance rules. The target values ​​of soil volumetric water content for the corresponding crop development stage of a high-yield, densely planted soybean population were obtained. And calculate the future irrigation water volume. ;

[0152] The fertilization decision unit is used to calculate the future amount of nitrogen fertilizer to be applied during irrigation, based on the nitrogen, phosphorus, and potassium nutrient requirement rules. Future irrigation phosphate fertilizer amount Future irrigation potassium fertilizer amount .

[0153] This invention, by constructing a digital twin model and a machine learning-based dense planting architecture recommendation model, can integrate variety characteristics, soil data, and meteorological conditions to recommend the optimal sowing density for a specific plot of land. This breaks through the limitations of traditional reliance on experience or uniform regional density and lays the foundation for a high-yield, densely planted population structure.

[0154] Based on digital twin models and future weather forecasts, this invention can predict crop growth status and soil water and fertilizer dynamics in advance, and accurately calculate irrigation and fertilization needs accordingly. It transforms experience-based irrigation and fertilization into predictive decision-making fertilization, realizing the precise supply of water and fertilizer resources on demand and improving utilization efficiency.

[0155] The above embodiments are merely exemplary embodiments of this application and are not intended to limit this application. The scope of protection of this application is defined by the claims. Those skilled in the art can make various modifications or equivalent substitutions to this application within its substance and scope of protection, and such modifications or equivalent substitutions should also be considered to fall within the scope of protection of this application.

Claims

1. A method for integrated water and fertilizer management based on high-density, high-yield soybean cultivation, characterized in that, Includes the following steps: Using historical meteorological, soil, variety parameters, and field trial data, a digital twin model was constructed to dynamically simulate the growth process of soybean populations. Based on the growth simulation of the digital twin model, a high-yield soybean high-density planting structure recommendation model is constructed to integrate variety, soil, meteorological and yield data to recommend the optimal planting density for a specific plot. The high-yield soybean dense planting population structure is recommended based on the dense planting structure model output, and the digital twin model is calibrated based on the canopy structure, soil moisture, and multispectral inversion vegetation index obtained by UAV or sensor monitoring. By inputting future weather forecast data into a calibrated digital twin model for growth simulation, the predicted results of growth status, soil moisture and nutrient dynamics of high-yield soybean dense planting populations are obtained. Based on the predicted growth status, soil moisture and nutrient dynamics of high-yield soybean dense planting populations, the water and fertilizer requirements are calculated, and the drip irrigation system is controlled based on the water and fertilizer requirements to carry out integrated irrigation and fertilization operations for soybean planting. Methods for constructing densely-planted architecture recommendation models include: The morphological growth rules are based on the plant height model and leaf area index (LAI) model obtained by fitting the Logistic growth curve, where: Plant height model In the formula, Let be the plant height at time t. For variety c and planting density The corresponding potential maximum plant height, The plant height growth rate, This is the time point when plant height increases the most; Leaf Area Index (LAI) Model In the formula, Let be the leaf area index at time t. Planting density and water and fertilizer supply The corresponding potential maximum LAI, For LAI growth rate, This is the time when plant height increases the most. This is a coercion correction term; Soil Moisture Balance Rules In the formula, , , , , These represent the soil volumetric water content at times t+1 and t, respectively. Let be the rainfall at time t. Let be the irrigation amount at time t. Let be the crop transpiration at time t. Let be the amount of soil seepage at time t. Let be the surface runoff at time t. For time step, For soil layer thickness, These are the basic crop coefficient, water stress coefficient, and soil evaporation coefficient. The reference evapotranspiration rate is calculated using the Penman-Monteith formula. Let be the field water holding capacity at time t. Let be the soil water storage at time t. CN represents the number of curves in the SCS curve number method; Nitrogen, phosphorus, and potassium nutrient requirements rules In the formula, For the first The amount of nutrients that need to be supplemented, For the first Total nutrient requirements For the first The percentage of each nutrient in total demand. For the first The current stock or existing level of the nutrient. For the first The conversion or efficiency coefficient of the current stock of nutrients. For the first The efficiency or output coefficient of nutrient supplementation.

2. The water and fertilizer integration management method based on high-density soybean cultivation according to claim 1, characterized in that: The method for constructing the digital twin model includes: Based on historical meteorological, soil, variety parameters and field test data, we constructed morphological growth rules, soil moisture balance rules and nitrogen, phosphorus and potassium nutrient dynamic rules for the soybean population growth process. Based on historical meteorological, soil, variety parameters and field test data, the physical entities of soybean population, meteorological and soil environment are mapped to virtual entities using the visualization engine Unity, resulting in a 3D model of the soybean population growth process. The digital twin model is formed by combining 3D modeling of soybean population growth process, morphological growth rules, soil moisture balance rules, and dynamic rules of nitrogen, phosphorus and potassium nutrients.

3. The water and fertilizer integration management method based on high-density soybean cultivation according to claim 2, characterized in that: The method for constructing the densely-planted architecture recommendation model includes: In the digital twin model, the soybean varieties, soil, and weather data of a fixed plot are used to simulate the growth of the plot at different planting densities, and the yield data of each planting density is recorded. The planting density corresponding to the maximum yield data is taken as the optimal sowing density for this plot; By combining variety, soil, and meteorological data with recommended planting density into a dataset, a convolutional neural network is trained to obtain a dense planting architecture recommendation model that predicts the optimal sowing density based on the input variety, soil, and meteorological data.

4. The water and fertilizer integration management method based on high-density soybean cultivation according to claim 3, characterized in that: The calibration method for the digital twin model includes: Adjust the planting density in the digital twin model to the optimal sowing density for a specific plot; The plant height, leaf area index (LAI), soil moisture, and multispectral inversion vegetation index corresponding to the canopy structure at the recommended planting density were obtained by monitoring with drones or sensors as observation values. In the digital twin model, key uncertain parameters are selected as calibration parameters based on the mechanistic rules of the soybean population growth process. The plant height, leaf area index (LAI), soil moisture, multispectral inversion vegetation index, and parameters to be calibrated are combined into a state vector based on the mechanistic rules of soybean population growth process in the digital twin model. By using the ensemble Kalman filter algorithm EnKF, the state vector is updated through Kalman gain based on the observations and the state vector. This prompts the digital twin model to continuously assimilate new observations to update the model parameters and complete the calibration of the digital twin model, resulting in a digital twin model for dynamically simulating the population growth process under the dense planting and high-yield soybean population structure.

5. The water and fertilizer integration management method based on high-density soybean cultivation according to claim 4, characterized in that: Based on the predicted growth status, soil moisture, and nutrient dynamics of high-yield soybean dense planting populations, the methods for calculating water and fertilizer requirements include: The future soil volumetric water content of a specific plot can be obtained using soil moisture balance rules. The target values ​​of soil volumetric water content for the corresponding crop development stage of a high-yield, densely planted soybean population were obtained. And calculate the future irrigation water volume. ; The future amount of nitrogen fertilizer to be irrigated is calculated based on the nitrogen, phosphorus, and potassium nutrient requirements rules. Future irrigation phosphate fertilizer amount Future irrigation potassium fertilizer amount .

6. A water and fertilizer integrated management system based on high-density soybean cultivation, characterized in that, The water and fertilizer integration management method for high-density, high-yield soybean cultivation, as described in any one of claims 1-5, comprises: The digital twin model building module is used to construct a digital twin model for dynamically simulating the growth process of soybean populations by utilizing historical meteorological, soil, variety parameters and field test data. The dense planting structure recommendation module is used to construct a dense planting structure recommendation model for soybean dense planting high-yield population structure that integrates variety, soil, meteorological and yield data to recommend the optimal planting density for a specific plot based on the growth simulation of the digital twin model. The model calibration module is used to recommend the soybean high-yield population structure based on the dense planting architecture, and to calibrate the digital twin model based on the canopy structure, soil moisture, and multispectral inversion vegetation indices obtained by UAV or sensor monitoring. The growth prediction module is used to input future weather forecast data into a calibrated digital twin model to simulate growth and obtain prediction results of the growth status, soil moisture and nutrient dynamics of high-yield soybean dense planting populations. The water and fertilizer decision control module is used to calculate the water and fertilizer requirements based on the growth status of high-yield soybean dense planting populations and the prediction results of soil moisture and nutrient dynamics, and to control the drip irrigation system to carry out integrated irrigation and fertilization operations for soybean planting based on the water and fertilizer requirements.

7. The integrated water and fertilizer management system based on high-density soybean cultivation according to claim 6, characterized in that: The digital twin model construction module includes: The rule-building unit is used to construct morphological growth rules, soil moisture balance rules, and nitrogen, phosphorus, and potassium nutrient dynamic rules for the soybean population growth process based on historical meteorological, soil, variety parameters, and field test data. The 3D visualization unit is used to map the physical entities of soybean populations, weather, and soil environment into virtual entities using the Unity visualization engine, based on historical meteorological, soil, variety parameters, and field test data, to obtain a 3D model of the soybean population growth process. The model integration unit is used to combine 3D modeling of soybean population growth process, morphological growth rules, soil moisture balance rules, and dynamic rules of nitrogen, phosphorus and potassium nutrients to form the digital twin model.

8. The integrated water and fertilizer management system based on high-density soybean cultivation according to claim 7, characterized in that: The densely embedded architecture recommendation module includes: The simulation execution unit is used in the digital twin model to simulate the growth of soybean varieties, soil and meteorological data of a fixed plot of land at different planting densities, and to record the yield data of each planting density of the plot. The optimal density determination unit is used to determine the planting density corresponding to the maximum yield data as the optimal sowing density for the plot. The model training unit is used to combine variety, soil, meteorological data and recommended planting density into a dataset, train a convolutional neural network, and obtain a dense planting architecture recommendation model that predicts the optimal sowing density based on the input variety, soil and meteorological data.

9. The integrated water and fertilizer management system based on high-density soybean cultivation according to claim 8, characterized in that: The water and fertilizer decision control module includes: The irrigation decision unit obtains the future soil volumetric water content of a specific plot through soil moisture balance rules. The target values ​​of soil volumetric water content for the corresponding crop development stage of a high-yield, densely planted soybean population were obtained. And calculate the future irrigation water volume. ; The fertilization decision unit is used to calculate the future amount of nitrogen fertilizer to be applied during irrigation, based on the nitrogen, phosphorus, and potassium nutrient requirement rules. Future irrigation phosphate fertilizer amount Future irrigation potassium fertilizer amount .

Images