A flat and low-lying land dwarfing and densification seeding decision support method fusing an agricultural knowledge graph

Abstract

The present application belongs to the technical field of data processing, and relates to a flat and depressed land dwarfing and densifying seeding decision support method fusing an agricultural knowledge graph, comprising the following steps: calculating and constructing a hydrological dynamics characteristic vector; extracting a hydrological vulnerability index of each grid independently; determining an operation mode of a current spatial grid; calculating a joint seeding probability weight coefficient; fusing and generating a composite variable seeding prescription matrix; and controlling an airborne seed metering and executing unit to execute entity seeding operation of matched density. The present application solves the defects of the prior art in the lack of comprehensive quantification of microtopography and dynamic meteorology, and provides dynamic differentiated seeding decision support that takes into account both high yield and risk hedging.

Description

Technical Field

[0001] This invention belongs to the technical field of data processing and relates to a decision support method for reducing plant density and increasing sowing density in flat and low-lying areas by integrating agricultural knowledge graphs. Background Technology

[0002] Variable seeding technology requires dynamic adjustment of seeding density based on the spatial heterogeneity of the field. Existing schemes mostly guide seeding by dividing management units based on historical yields or static soil properties. However, in flat and low-lying fields with undulating terrain, hydrological heterogeneity dominated by micro-topography is often the core factor leading to local waterlogging or drought stress.

[0003] As a specific application of the aforementioned technology, the Canadian Patent Office has disclosed a field-based risk-adjusted mixed seed selection and crop yield optimization method, patent publication number CA3112581C. This technology analyzes historical agricultural data and regional reference data, including grower yields and seed distribution, to establish a historical yield distribution model and define target yield ranges. Based on historical rankings, it then generates yield improvement suggestions for different fields by changing seed quantity or density and controls machinery to perform the sowing adjustments.

[0004] While this technology enables field-level density adjustment, its over-reliance on retrospective assessments of historical data fails to consider the interaction between micro-topography and dynamic rainfall, and lacks a mechanism for quantifying soil topography and weather as a complete hydrodynamic system. This makes it difficult to accurately assess the actual water stress risk posed by extreme weather events in hydrologically sensitive areas such as flatlands, thus hindering the development of dynamic intercropping planting strategies that balance high yield potential with risk avoidance.

[0005] In summary, how to overcome the shortcomings of existing technologies in comprehensively quantifying micro-topography and dynamic weather, and provide dynamic differentiated seeding decision support that balances high yield and risk hedging, is a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0006] To address the aforementioned issues, this invention provides a decision support method for denser sowing in flat and low-lying areas by integrating agricultural knowledge graphs.

[0007] A decision support method for reducing plant density and increasing sowing density in flat and low-lying areas by integrating agricultural knowledge graphs includes the following steps: S1. Obtain basic agricultural geographic parameters and meteorological forecast data streams, and calculate and construct hydrodynamic feature vectors; S2. Perform time series perturbation calculation on the hydrodynamic feature vector and extract the independent hydrological vulnerability index for each grid. S3. Query the agricultural knowledge graph to obtain the vulnerability threshold of crop varieties, compare the hydrological vulnerability index with the vulnerability threshold, and determine the current operation mode of the spatial grid. S4. When the hydrological vulnerability index is less than the vulnerability limit threshold, obtain the benchmark sowing density that meets the target yield threshold; when the hydrological vulnerability index is greater than or equal to the vulnerability limit threshold, obtain the high-yield benchmark density index and the risk bottom line density index from the agricultural knowledge graph, and calculate the joint sowing probability weighting coefficient. S5. Based on the benchmark sowing density that meets the target yield threshold, a deterministic sowing filling zone is generated, and the high-yield benchmark density index and the risk bottom line density index are intertwined and arranged according to the joint sowing probability weighting coefficient, and a composite variable sowing prescription matrix is ​​generated. S6. Compile the composite variable seeding prescription matrix into a low-level driving instruction file and send it to the front-end operation terminal through the variable frequency seeding operation bus to control the on-board seeding execution unit to perform the physical seeding operation with matching density.

[0008] A further aspect of this invention involves acquiring basic agricultural geographic parameters and meteorological forecast data streams, and calculating and constructing hydrodynamic feature vectors, including the following steps: Read the continuous digital elevation model (DEM) data of the target field and simultaneously acquire the surrounding soil physical parameters, including the soil saturated hydraulic conductivity. Combine the precipitation evolution trajectory contained in the short-term meteorological ensemble forecast data with the DEM data to perform spatial grid segmentation and form independent spatial grids. Calculate the estimated surface water flow path of each independent spatial grid based on the DEM data, and calculate the surface soil moisture evapotranspiration potential under precipitation-free conditions by combining the surrounding soil physical parameters. Segment the estimated surface water flow path and the surface soil moisture evapotranspiration potential to generate a hydrodynamic feature vector for the entire grid of the target field.

[0009] A further aspect of this invention involves splicing the estimated surface water flow path with the surface soil moisture evapotranspiration potential, and fusing them to generate a hydrodynamic feature vector for the entire target field grid, including the following steps: The predicted surface water outflow direction and cumulative flow volume corresponding to each independent spatial grid are spliced ​​with the surface soil moisture evapotranspiration potential and soil saturated hydraulic conductivity of the same grid in a preset order; normalization processing is performed, and the resulting data are fused to generate a hydrodynamic feature vector.

[0010] A further aspect of this invention involves performing time-series perturbation calculations on the hydrodynamic feature vectors to extract independent hydrological vulnerability indices for each grid cell, including the following steps: Multiple simulated precipitation fluctuation distribution trajectories were retrieved from short-term meteorological ensemble forecast data; multiple evolutionary simulations were initiated for each independent spatial grid using a Monte Carlo simulation model to calculate the dynamic bias data of soil moisture triggered by the effect of each simulated precipitation fluctuation distribution trajectory on the hydrodynamic feature vector; the dynamic bias data of soil moisture output from multiple evolutionary simulations were summarized, and the variance of water content fluctuation reflecting the hydrological sensitivity characteristics of a single point was calculated; the hydrological vulnerability index of each independent spatial grid was converted based on the variance of water content fluctuation, and the corresponding reference estimate of average water content was calculated.

[0011] A further aspect of this invention involves extracting and summarizing soil moisture dynamic bias data from multiple independent evolutionary simulations, and calculating the variance of water content fluctuations reflecting the hydrological sensitivity of a single point. This includes the following steps: Calculate the time average of each of the multiple soil moisture dynamic bias data; calculate the sample variance of the sample set composed of multiple time averages, and define the sample variance as the water content fluctuation variance value.

[0012] A further aspect of this invention involves querying an agricultural knowledge graph to obtain a vulnerability threshold for crop varieties, comparing the hydrological vulnerability index with the vulnerability threshold, and determining the current operational mode of the spatial grid, including the following steps: Access the agricultural knowledge graph that stores crop physiological parameters; retrieve and extract the vulnerability threshold from the agricultural knowledge graph; traverse all grid points and compare the hydrological vulnerability index with the vulnerability threshold; when the hydrological vulnerability index is less than the vulnerability threshold, determine the operation mode as a single-density operation mode; when the hydrological vulnerability index is greater than or equal to the vulnerability threshold, determine the operation mode as a dual-density interleaved operation mode.

[0013] A further aspect of this invention involves obtaining a benchmark sowing density that meets the target yield threshold when the hydrological vulnerability index is less than the vulnerability threshold, and obtaining high-yield calibration density indicators and risk baseline density indicators from an agricultural knowledge graph when the hydrological vulnerability index is greater than or equal to the vulnerability threshold, and calculating the joint sowing probability weighting coefficient. This includes the following steps: The average moisture content reference estimate is input into the benchmark stress calculation function set of the agricultural knowledge graph for calculation to generate the benchmark sowing density that meets the target yield threshold. The average moisture content reference estimate is input into the pre-configured high-yield potential prediction function and risk avoidance compensation function in the agricultural knowledge graph to calculate the high-yield calibration density index and the risk bottom line density index. The joint sowing probability weight coefficient is converted based on the hydrological vulnerability index. The joint sowing probability weight coefficient is used to determine the spatial coverage ratio of the high-yield calibration density index.

[0014] A further aspect of this invention involves generating deterministic seeding zones based on a benchmark seeding density that meets the target yield threshold, and interweaving high-yield benchmark density indicators and risk baseline density indicators according to a joint seeding probability weighting coefficient to generate a composite variable seeding prescription matrix, including the following steps: A first spatial raster table containing the benchmark sowing density that meets the target yield threshold is extracted, and a second spatial raster table containing the high-yield calibration density index, the risk baseline density index, and the co-sowing probability weight coefficient is extracted. A deterministic sowing filling zone is constructed based on the first spatial raster table. The co-sowing probability weight coefficient is transformed into a spatially interleaved nested hierarchical instruction sequence of the second spatial raster table through a spatially interleaved nested hierarchical instruction sequence. Following the spatially interleaved nested hierarchical instruction sequence, the high-yield calibration density index and the risk baseline density index are interleaved and injected to generate a composite variable sowing prescription matrix for the entire domain.

[0015] A further aspect of this invention involves using a spatial jitter mapping algorithm based on a controlled random process to transform the joint seeding probability weight coefficients into a spatially interleaved nested hierarchical instruction sequence for a second spatial grid table, comprising the following steps: The joint seeding probability weighting coefficients are input into a spatial jitter mapping algorithm based on a controlled random process, and a two-dimensional Boolean matrix is ​​output. The two-dimensional Boolean matrix is ​​then transformed into a spatially interleaved nested hierarchical instruction sequence, which is used to define the density distribution of sub-cells.

[0016] A further aspect of this invention involves compiling the composite variable seeding prescription matrix into a low-level driving instruction file and sending it to the front-end operation terminal via a variable frequency seeding operation bus to control the onboard seeding execution unit to perform a matching density physical seeding operation, including the following steps: The composite variable sowing prescription matrix is ​​compiled into a low-level driver instruction file adapted to the communication protocol; the low-level driver instruction file is sent to the seeding operation bus to control the release and rest cycle of the onboard seeding execution motor; the operation terminal is controlled to follow the onboard navigation movement trajectory and execute stepless adjustment sowing operations that match the benchmark sowing density that meets the target yield threshold, or switch between the high-yield calibration density index and the risk bottom line density index; the measured deviation value of crop field survival density calculated after sowing is obtained, and the measured deviation value is fed back to the agricultural knowledge graph to correct the vulnerability limit threshold.

[0017] In summary, the present invention has the following beneficial technical effects: 1. A hydrodynamic feature vector is constructed by acquiring basic geographic parameters and meteorological forecast data streams, and a time-series perturbation calculation is performed using Monte Carlo simulation to extract the hydrological vulnerability index. This process integrates static topographic data with dynamic weather forecasts, simulates soil moisture response under different rainfall scenarios, and transforms the complex waterlogging risk of flat and low-lying areas into a quantitative indicator. This overcomes the shortcomings of existing technologies in comprehensively quantifying the impact of micro-topography and dynamic weather, and improves the objectivity of decision-making.

[0018] 2. Vulnerability thresholds for crop varieties are obtained from an agricultural knowledge graph and compared with a hydrological vulnerability index to trigger a single-modal scheduling signal or a bimodal forced jump instruction. This method solidifies agronomic prior knowledge into objective criteria and forcibly switches to a dual-objective calculation mode that balances high-yield potential and risk avoidance when the risk exceeds the threshold, thus constructing an adaptive risk response mechanism.

[0019] 3. High-yield calibration and risk baseline density indices are derived in parallel under a dual-modal computing mode. The joint sowing probability weighting coefficient is calculated based on the hydrological vulnerability index, and a spatial jitter mapping algorithm is imported to generate a composite variable sowing prescription matrix. This process achieves a finely staggered layout of high and low densities at the sub-grid scale, improving crop ventilation and light penetration conditions and forming a physical risk buffer structure internally, thus overcoming the shortcomings of existing technologies in outputting spatially interwoven sowing strategies. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. The drawings are used to provide a further understanding of the present invention.

[0021] Figure 1 A flowchart illustrating an embodiment of this application is disclosed.

[0022] Figure 2 The present application discloses the curves showing the evolution of simulated soil volumetric water content over time for grid 5 under different precipitation scenarios in the embodiments of this application.

[0023] Figure 3 The probability weighting coefficient mapping curve of joint sowing in the embodiments of this application is disclosed.

[0024] Figure 4 A schematic diagram of the connection of the decision support system module integrating agricultural knowledge graph in the embodiments of this application is disclosed. Detailed Implementation

[0025] See attached document Figure 1 This invention proposes a decision support method for denser planting in flat and low-lying areas by integrating agricultural knowledge graphs, comprising the following steps: S1. Obtain basic agricultural geographic parameters and meteorological forecast data streams, and calculate and construct hydrodynamic feature vectors; S2. Perform time series perturbation calculation on the hydrodynamic feature vector and extract the independent hydrological vulnerability index for each grid. S3. Query the agricultural knowledge graph to obtain the vulnerability threshold of crop varieties, compare the hydrological vulnerability index with the vulnerability threshold, and determine the current operation mode of the spatial grid. S4. When the hydrological vulnerability index is less than the vulnerability limit threshold, obtain the benchmark sowing density that meets the target yield threshold; when the hydrological vulnerability index is greater than or equal to the vulnerability limit threshold, obtain the high-yield benchmark density index and the risk bottom line density index from the agricultural knowledge graph, and calculate the joint sowing probability weighting coefficient. S5. Based on the benchmark sowing density that meets the target yield threshold, a deterministic sowing filling zone is generated, and the high-yield benchmark density index and the risk bottom line density index are intertwined and arranged according to the joint sowing probability weighting coefficient, and a composite variable sowing prescription matrix is ​​generated. S6. Compile the composite variable seeding prescription matrix into a low-level driving instruction file and send it to the front-end operation terminal through the variable frequency seeding operation bus to control the on-board seeding execution unit to perform the physical seeding operation with matching density.

[0026] The process of acquiring basic agricultural geographic parameters and meteorological forecast data streams, and calculating and constructing hydrodynamic feature vectors, is performed by a central data server equipped with a geographic information processing module and a hydrological model. The server's storage unit retrieves and stores pre-mapped and stored continuous digital elevation model data covering the target field. This data is stored in a geotagged image file format with a spatial resolution of 1m × 1m, ensuring a detailed depiction of micro-topographic undulations. Simultaneously, the server synchronously reads a table of surrounding soil physical parameters matching the geographical location of the target field from an associated agricultural database. This parameter table includes at least soil saturated hydraulic conductivity, field capacity, and field wilting coefficient, collectively characterizing the basic hydraulic conductivity and water-holding capacity of the soil under different water conditions. The value range of its core parameter, soil saturated hydraulic conductivity, depends on soil texture; for example, the saturated hydraulic conductivity of sandy soil can be set at 1.0 × 10⁻⁶. -4 Up to 1.0×10 -2 The range is in the m / s range, while clay is in the range of 1.0 × 10 m / s. -9 Up to 1.0×10 -6 Within the range of m / s.

[0027] The server connects to the application programming interface (API) of a meteorological data center or a commercial meteorological service provider to import future data in real time. For example, the sky Short-term weather ensemble forecast data stream. Short-term weather ensemble forecast data is a set of prediction results generated by meteorological models through multiple integration operations after applying small perturbations to initial conditions or physical process parameters. It is used to characterize the uncertainty of future weather forecasts. The data stream encapsulates multiple possible precipitation intensities and spatiotemporal evolution trajectories.

[0028] The processing module extracts the coordinate range and resolution of the continuous digital elevation model data. Based on this, it performs spatial raster segmentation on the entire target field, forming a basic coordinate system composed of independent spatial rasters. Each raster is assigned a unique two-dimensional coordinate identifier. Based on this coordinate system, the hydrological model calculates the surface water outflow direction of each independent spatial raster using a deterministic eight-direction algorithm, based on its elevation and the elevation difference between its eight adjacent rasters. It also recursively calculates the cumulative water flow into all upstream rasters, collectively forming the estimated surface water flow path.

[0029] While performing this calculation, another parallel computing unit combines the soil type from the surrounding soil physical parameters, as well as non-precipitation meteorological parameters such as average temperature, wind speed, and sunshine duration extracted from short-term meteorological ensemble forecast data, and uses the Penman Montes equation to estimate the baseline daily evapotranspiration under the condition of no precipitation event interference. This value is the surface soil moisture evapotranspiration potential.

[0030] The estimation of surface soil moisture evapotranspiration potential assumes that the crop is in a healthy growth state under full irrigation. The Penmanmont equation, upon which the calculation is based, is the standard method for calculating crop water requirements recommended by the Food and Agriculture Organization of the United Nations. The data integration module concatenates the estimated surface water outflow direction and cumulative flow volume from the predicted surface water flow path corresponding to each independent spatial grid with the surface soil moisture evapotranspiration potential and the saturated hydraulic conductivity value, which characterizes the basic hydraulic conductivity of the same grid, in a preset order to generate a multi-dimensional hydrodynamic feature vector that characterizes the hydrological dynamics of that grid point. This vector is used for each independent spatial grid. Generated hydrodynamic feature vectors It can be expressed by the following formula: ; In the formula, Indicates that the index is The hydrodynamic feature vector corresponding to an independent spatial grid is a data set used to describe the hydrological characteristics of that grid. Indicates that the index is The direction of surface water outflow for an independent spatial grid is calculated using a deterministic eight-direction algorithm based on the elevation comparison between the grid and its eight neighboring grids. It is usually encoded as an integer, for example, 1 represents due east, 2 represents southeast, 4 represents due south, and so on, for a total of eight directions. Indicates that the index is The cumulative water flow of an independent spatial grid represents the total number of units into which the water flow from all upstream grids eventually converges, based on the direction of surface water outflow. Its magnitude reflects the potential water collection capacity of the grid and is an indicator for determining whether it is a depression or a runoff channel. This value is set as a dimensionless count value, ranging from 0 to the total number of grids in the field. Indicates that the index is The soil saturated hydraulic conductivity corresponding to the independent spatial grid is obtained from the surrounding soil physical parameter table. Its value is determined based on the infiltration test of the soil sample on site or assigned by the existing high-precision soil type map database. It is a physical indicator for measuring the speed of water infiltration in the soil. Indicates that the index is The surface soil moisture evapotranspiration potential of an independent spatial grid is calculated using the Penmanmont equation based on multiple meteorological factors such as historical or predicted temperature, humidity, wind speed, and solar radiation in the region. It represents the maximum water loss rate of the surface and vegetation canopy under ideal conditions of no rainfall supplementation and sufficient water supply.

[0031] Among them, hydrodynamic eigenvectors It is a feature vector used as input to subsequent machine learning or simulation models, and its components are an ordered concatenation of four independent parameters with different physical meanings and dimensions. Specifically, The encoded value is dimensionless; This is a dimensionless raster count value; The dimension of is length divided by time, for example or ; The dimension of is also length divided by time, for example Before inputting hydrodynamic feature vectors into subsequent computational models, normalization or standardization is usually performed to scale all feature values ​​to a uniform numerical range, such as [0, 1] or [-1, 1], in order to eliminate the influence of different units on the model training weights.

[0032] For example, suppose this step is performed on a 3×3 micro-target field area, with its raster cells numbered 1 to 9 from left to right and top to bottom. The server first reads the continuous digital elevation model data of this area, with the elevation values ​​of each raster in meters, as shown in the matrix below: ; At the same time, the server retrieved the surrounding soil physical parameter table and learned that the area is mainly loam soil with a soil saturation hydraulic conductivity of... Unified settings The server also imported short-term weather ensemble forecast data and calculated the surface soil moisture evapotranspiration potential for the region. Unified as Next, the hydrological model applies the central grid, i.e. Calculations were performed on grid number 5. By comparing its own elevation of 9.8m with the elevations of its eight neighboring grids, it was found that the water flow would originate from all neighboring grids and flow towards grid number 5. Therefore, the direction of surface water outflow was determined. In this deterministic octant flow algorithm, it can be represented as pointing towards itself or having no definite single outlet, for example, encoded as 0; its cumulative flow volume The value is 8, because the water from all 8 neighboring areas flows into this point.

[0033] For grid cell 1 in the top left corner, with an elevation of 10.5m, the lowest elevation among its neighbors is grid cell 5. The water flow will flow diagonally towards grid cell 5. Assuming the deterministic oc-flow algorithm encodes the southeast direction as 2, then... The value is 2, and there are no grids upstream of it, therefore The value is 0. This process continues until the flow direction and accumulation of all grid cells are calculated. Finally, the feature vector for the central grid cell (cell #5) is constructed. The data integration module extracts... The value is 0. The value is 8. value ,as well as value The hydrodynamic feature vector generated for grid number 5. for This vector will be passed to the next processing step.

[0034] The process of performing time-series perturbation calculations on hydrodynamic feature vectors and extracting independent hydrological vulnerability indices for each grid cell is completed by a simulation computing unit deployed within the central data server. This simulation computing unit first retrieves the hydrodynamic feature vectors generated in the previous processing stage for each independent spatial grid cell in the target field. Simultaneously, this unit extracts from the short-term meteorological ensemble forecast data stream... One, for example, 20 to 50 independent, future Day (for example) to The simulated precipitation fluctuation distribution trajectory within a given time step (day) is used as the source of perturbation variables in the Monte Carlo simulation. This unit activates a soil water balance calculation model based on physical processes, which is constructed based on the law of water conservation, i.e., the change in soil moisture content within a certain time step equals precipitation minus surface runoff, evapotranspiration, and deep infiltration. The necessary parameters for model operation are initialized from the corresponding components of the hydrodynamic eigenvectors through the following specific mapping relationships: slope runoff coefficient... Based on the cumulative water flow mapped through a logarithmic function, the formula is as follows: ,in, , The maximum soil infiltration rate is an empirical constant determined based on field measurements. Then, by inverse normalization of the physical dimensions, it can be directly equivalent to the soil saturated hydraulic conductivity. ,Right now = The maximum daily evapotranspiration is directly taken as the evapotranspiration potential of the surface soil moisture. This ensures that the simulation process accurately reflects the local hydrophysical characteristics of each grid cell.

[0035] The unit uses this model to initiate each individual spatial grid. Each evolutionary simulation is a separate iteration. In each iteration, the model takes a unique simulated precipitation fluctuation trajectory as input and uses parameters from the hydrodynamic eigenvectors to set boundary conditions for hydrological response processes such as runoff generation, soil infiltration, and evapotranspiration within the grid. For example, cumulative runoff is used to regulate lateral inflow, and soil saturated hydraulic conductivity is used to constrain infiltration rates. The model iterates daily, outputting a single line of length... The time series of soil moisture content in a single simulation is the dynamic bias data of soil moisture under the influence of the corresponding input precipitation trajectory.

[0036] After completing the entire process of a grid After the next evolutionary deduction, the simulation computing unit collected all... The data includes dynamic bias data of soil moisture, and statistical analysis is performed on this time series data. Specifically, the calculation of this... The time average of each time series is calculated, and then this... The sample variance of a sample set consisting of time averages is defined as the water content fluctuation variance that reflects the hydrological sensitivity of the grid point.

[0037] The variance of this water content fluctuation is either directly assigned or normalized to calculate a hydrological vulnerability index specific to that individual spatial grid. A higher hydrological vulnerability index indicates that the soil moisture status of that grid is more susceptible to the uncertainty of future rainfall. In the event of extreme rainfall events, the risk of its water content deviating from the suitable growth range for crops is higher, corresponding to a higher level of hydrological environmental vulnerability. Simultaneously, the simulation unit also calculates... The arithmetic mean of the time averages is used as the reference estimate of the average moisture content corresponding to that independent spatial raster. The above time series perturbation calculation and index extraction process is as follows: for any index... An independent spatial grid, with an average moisture content reference estimate. Hydrological Vulnerability Index The calculation process is as follows: First, through The Monte Carlo simulation yielded... The group of soil moisture dynamic bias data is denoted as ,in .

[0038] Each set of data is a time series. Calculate the first The time-averaged moisture content of the simulation ; Calculate the reference estimate of average moisture content ; Calculate the variance of water content fluctuation. : ; The hydrological vulnerability index was calculated. : ; In the formula, Indicates for grid In the Under the influence of the simulated precipitation fluctuation distribution trajectory, the first The simulated soil volumetric water content for 1 day is a dimensionless number between 0 and soil porosity, usually expressed as a percentage or decimal, such as 0.35. This value represents the healthy water content of typical soils near field capacity. The day index representing the extrapolated time series ( ); This indicates the number of days the simulation model can be projected, and its setting is based on the effective forecast duration of short-term meteorological ensemble forecast data; This represents the total number of independent simulated precipitation fluctuation distribution trajectories included in the short-term meteorological ensemble forecast data; Indicates the first In this simulation, the grid exist Average soil moisture content within a day; That is, the reference estimate of average moisture content, representing the grid under all possible precipitation scenarios. The expected average soil moisture content; That is, the variance of water content fluctuation, which describes the raster. The degree of dispersion of average soil moisture content under different precipitation scenarios; That is, the hydrological vulnerability index; function It is an identity mapping, that is Variance can be used directly as a vulnerability measure, or it can be a normalization function that maps the variance to... Within a given interval, comparisons across different regions are easier.

[0039] For example, this step of the calculation is performed for the central grid cell No. 5, and the simulation computing unit has obtained its hydrodynamic feature vector. Now set the simulation days. Days, and selected from short-term weather ensemble forecast data. Using three simulated precipitation fluctuation trajectories as examples, the future... The total rainfall for the day was divided into three scenarios: low, medium, and high. The simulation unit utilized a soil water balance calculation model and based on hydrodynamic characteristic vectors. The parameters, especially its high flow accumulation. The water balance model of the grid was configured to exhibit strong catchment and weak drainage characteristics. After inputting three precipitation trajectories, the model completed three independent evolutionary simulations, calculating three dynamic soil moisture bias data points with a length of 7.

[0040] After time averaging, the average soil moisture content obtained from the three simulations was obtained. They are respectively: , , The calculation unit calculates a reference estimate of the average moisture content based on these three average values. The calculation process is as follows: Calculate the variance of water content fluctuation. Its calculation is The variance value is directly assigned to the hydrological vulnerability index, resulting in... Therefore, for grid 5, the output of this step is a set of data pairs: an average moisture content reference estimate of 0.383 and a hydrological vulnerability index of 0.0233. This set of data will be used in subsequent decision-making processes.

[0041] See appendix Figure 2 The three solid lines in the figure represent the dynamic bias trajectories of soil moisture under simulated evolution scenarios of low precipitation, moderate precipitation, and high precipitation, respectively. The dashed lines in the figure correspond to the arithmetic mean of the time averages of each trajectory, i.e., the reference estimate of the average moisture content, which is 0.383 for grid number 5.

[0042] The process of querying the agricultural knowledge graph to obtain a benchmark threshold and triggering a mandatory operation mode switch based on the hydrological vulnerability index is executed by a knowledge reasoning engine within the central data server. This engine automatically constructs entity attribute query links based on the crop variety to be planted in the target field specified in the current task, such as "XXX958" corn, connecting to the pre-built agricultural knowledge graph containing detailed physiological parameters of that crop variety.

[0043] An agricultural knowledge graph is a structured knowledge network stored in the form of a graph database. Its construction involves integrating publicly available agronomic literature, crop breeding reports, regional planting guidelines, and historical yield data. Natural language processing techniques are used for entity recognition and relation extraction, and the data is then reviewed and verified by agronomic experts before being stored in a graph database such as NebulaGraph or Neo4j. This graph contains various entities, including crops, soil, climate, and management practices, as well as their interrelationships.

[0044] The inference engine performs a precise query within the agricultural knowledge graph through this query chain. The query target is a preset relationship node bound to the crop variety entity tag "XXX958". This node is a structured design that takes a specific crop variety as a core entity and links it to an attribute node containing multiple specific threshold parameters through predefined relationships, such as hasPhysiologicalThreshold, to store the attribute value named "vulnerability threshold".

[0045] The vulnerability threshold is a threshold pre-set and stored in the data set by agronomic experts based on physiological indicators such as the variety's waterlogging tolerance, root development characteristics, and lodging resistance, through extensive field trials and model calibration. Its setting is based on crop physiology principles, quantifying that when the expected fluctuation range of soil moisture content exceeds a certain limit, the probability of root respiration being inhibited, or lodging due to insufficient support at the base of the stem after the jointing stage, will increase. Therefore, the vulnerability threshold serves as a benchmark for risk management decisions. Assuming the scenario is to avoid yield losses exceeding a set proportion due to water stress, such as 5%, the vulnerability threshold can be set as the variance of moisture content fluctuation corresponding to this loss risk.

[0046] Upon successful query, the inference engine extracts and obtains the vulnerability threshold. The inference engine then initiates a traversal process, sequentially scanning every spatial raster point within the entire target field area and reading its hydrological vulnerability index calculated in the previous step. For each raster point, the inference engine performs a comparison operation, numerically comparing the point's hydrological vulnerability index with the vulnerability threshold obtained from the map.

[0047] Based on the comparison results, subsequent mode switching logic is triggered. The inference engine determines that the hydrological environment at a grid point is relatively stable only if the hydrological vulnerability index is less than the vulnerability threshold, and generates a single-density operation instruction. Subsequent modules using this instruction employ a conventional, single-objective optimization calculation mode. Conversely, if the inference engine determines that the hydrological vulnerability index at a grid point is greater than or equal to the vulnerability threshold, it means that the uncertainty of future soil moisture conditions at that point exceeds the range within which the crop variety can robustly adapt. In this case, the inference engine generates a dual-density interleaved operation instruction, which directs subsequent calculation modules to switch to a dual-objective, composite strategy solution mode that balances high-yield potential and risk avoidance.

[0048] Among them, single-density operation instructions and dual-density interleaved operation instructions are internal control signals generated by the inference engine. They exist in the form of binary flags or predefined strings and are used to guide the computation process into different code branches to achieve dynamic switching of algorithm modes.

[0049] For example, the knowledge reasoning engine has received input data for the central grid cell 5, namely its hydrological vulnerability index. Suppose the current task specifies the crop variety to be planted as "XX Waxy Corn No. 1". The inference engine first connects to the agricultural knowledge graph and locates the entity node for "XX Waxy Corn No. 1". Under this entity, along the hasPhysiologicalThreshold relation edge, a set of physiological parameters is retrieved, including the vulnerability threshold, whose value is... The vulnerability threshold is set based on the variety's relatively sensitive flood resistance characteristics.

[0050] The inference engine performs a comparison operation, comparing the hydrological vulnerability index of 0.0233 with the vulnerability threshold of 0.015. Since 0.0233 > 0.015, the comparison result is true, meaning the hydrological vulnerability index exceeds the vulnerability threshold. Based on preset logic rules, the inference engine determines that grid 5 belongs to a high-risk area and generates a dual-density interleaving operation instruction for this grid. This instruction, such as the string "DUAL_MODE", will be appended to the data packet of grid 5 and transmitted to the downstream processing module. Suppose that for another grid, such as grid 1, its calculated hydrological vulnerability index is 0.008. In the same comparison process, since 0.008 < 0.015, the inference engine will generate a single-density operation instruction for grid 1, such as the string "SINGLE_MODE". The system traverses all grids, assigning a calculation mode label to each grid, completing the instruction switching process for the calculation mode. Furthermore, to ensure the robustness of the system under complex real-world conditions, this solution is configured with an anomaly degradation handling strategy. During the traversal and judgment process, if the knowledge graph query fails due to the absence of the target crop variety, or if the confidence index of the short-term weather forecast data stream (defined as the probability consistency ratio of each independent evolutionary member in the short-term weather ensemble forecast for the prediction of future precipitation trends) is lower than the preset threshold (e.g., lower than 60%, which is a lower limit threshold determined by statistical backtracking model based on the false alarm rate of the historical weather model in the region and the system's tolerable risk of yield reduction due to sowing errors), the system will trigger a downgrade alarm, automatically block the comparison of the hydrological vulnerability index, and force the entire raster to adopt a static single-density fallback operation mode based on historical experience yields to avoid the risk of large-scale yield reduction caused by erroneous dynamic decision-making driven by severely insufficient confidence in the preceding meteorological input data.

[0051] The process of deriving differentiated sowing density combinations from the agricultural knowledge graph based on the defined operation mode instructions is executed by a sowing density solver coupled to the knowledge inference engine. This solver continuously monitors the raster data stream transmitted from upstream. When it receives a raster data packet with a single-density operation instruction, such as the string "SINGLE_MODE", it invokes a pre-mounted set of baseline stress calculation functions within the agricultural knowledge graph in response to this instruction. This function set contains mathematical models describing the relationship between yield and sowing density for this crop variety under different soil moisture conditions. The core of the baseline stress calculation function set is a yield-density model in the form of a quadratic or Gaussian function, with the following formula: ; In the formula, This represents the expected yield at a given seeding density; Indicates sowing density; Curvature coefficient, representing the sensitivity of yield to density; This represents the theoretical peak seeding density corresponding to the potential maximum yield under these moisture conditions, resulting in the expected yield. The target yield threshold is equal to the target yield threshold, and the quadratic equation is solved to obtain the benchmark sowing density that satisfies the target yield threshold. The target yield threshold refers to the expected economic minimum yield value set for the crop variety under the target moisture conditions, which is pre-fixed in the agricultural knowledge graph and combines historical statistical data and agronomic prior knowledge. This represents the potential maximum yield under these moisture conditions. The parameter is... , and These are all average moisture content reference estimates. The functions, whose relationships are obtained by fitting a large amount of historical field trial data, are stored in the agricultural knowledge graph.

[0052] The solver extracts the average moisture content reference estimate from the raster data packet and uses it as an input variable, pushing it to the baseline stress calculation function set to perform a routine simulation. This calculation process involves solving for constraints, specifically finding the planting density that enables the expected yield to reach the target yield threshold under given moisture conditions. The solution is the baseline planting density generated for the current spatial raster that satisfies the target yield threshold, and this value is the calculation result of the equation that satisfies the yield threshold constraint under given moisture conditions. Conversely, when the solver receives a data packet with a dual-density interleaving operation instruction, such as the string "DUAL_MODE", it responds to this instruction by simultaneously calling two other independent sets of functions in the agricultural knowledge graph: the high-yield potential prediction function and the risk aversion compensation function.

[0053] These two sets of functions are based on two independent, computable models fitted from agronomic experiments. Specifically, the high-yield potential prediction function uses an absolute value decay model, and its expression is: ,in, Ideal capping density; The optimal moisture content; The high-yield sensitivity coefficient is used; the risk aversion compensation function adopts a parabolic bottoming model, and its expression is: ,in, This represents the physiological survival limit density. This represents the risk compensation coefficient. Taking the variety in this embodiment as an example, the parameters issued by the knowledge graph are: When the solver extracts the reference estimate of the average moisture content... When the value is 0.383, the two functions mentioned above can be imported in parallel to calculate and output the high-yield calibration density index plants respectively. Plant / hm 2 and risk bottom line density index Plant / hm 2 .

[0054] The risk aversion compensation function, based on the crop variety's flood tolerance limit, calculates the minimum safe density required to ensure root survival and effective yield under extremely wet conditions, i.e., the risk baseline density index. The solver imports the average moisture content reference estimate into these two functions in parallel, calculating and outputting the high-yield calibration density index and the risk baseline density index, respectively. Simultaneously, the solver extracts a hydrological vulnerability index from this data package and, through a pre-defined nonlinear mapping function, such as an sigmametric function, converts it into a joint sowing probability weighting coefficient between 0 and 1. The formula for this conversion function is: ; In the formula, Indicates a grid The joint sowing probability weighting coefficient; Represents a grid The hydrological vulnerability index; Indicates the sensitivity parameter; This indicates the parameters for transforming the center point. and In agricultural knowledge graphs, the sensitivity of weights to changes in the hydrological vulnerability index and the transition center point are jointly determined. Here, the vulnerability threshold and the transition center point parameter are... The associated logic uses a vulnerability threshold as a triggering threshold for discrete states to determine whether to initiate a dual-density mode; while As an inflection point for continuous weight calculation, it is usually set slightly higher than the vulnerability threshold. For example, in this embodiment, the threshold is 0.015. The value is set to 0.018 to preserve a buffer zone when the risk threshold is just crossed, preventing a sudden drop in the weight distribution.

[0055] Assuming the application scenario involves corn varieties sensitive to waterlogging, The threshold will be set relatively low, for example, to the critical variance value that historically led to stunted root development due to oxygen deficiency in this variety. This means that as soon as the hydrological vulnerability index exceeds this transition center point, the weighting coefficient drops rapidly, favoring a conservative planting strategy. This coefficient is configured to physically define the proportion of area where high-yield calibration density is used for planting. Within high-hydrological-vulnerability grid cells, it is used to balance potential gains and risks through spatial combinations of the two densities.

[0056] For example, the seeding density solver first processes grid 1, which receives the single-density operation instruction "SINGLE_MODE" and obtains its average moisture content reference estimate of 0.22. The solver calls the baseline stress calculation function set, substitutes the moisture content value of 0.22, and the internal model of the function set calculates that under this moisture condition, the set density is 75,000 plants / hm². 2 Therefore, for grid 1, the baseline seeding density output by the solver to meet the target yield threshold is 75,000 plants / hm². 2 The solver processed grid 5, received the dual-density interleaving operation command "DUAL_MODE", and read its data, obtaining the average moisture content reference estimate. Hydrological vulnerability index The instruction triggers parallel computation, and the solver will... The samples were fed into a high-yield potential prediction function and a risk aversion compensation function, respectively. The former calculated the high-yield target density index as 80,000 plants / hm². 2 The latter calculated the risk baseline density index to be 55,000 plants / hm². 2 .

[0057] At the same time, the solver will Substitute the conversion function for the combined sowing probability weighting coefficient. Assume the conversion function corresponding to this crop variety is... The calculation process is as follows: Therefore, for grid number 5, the solver outputs a density combination, including the high-yield calibration density index of 80,000 plants / hm². 2 Risk baseline density index: 55,000 plants / hm² 2 The result also includes a co-seeding probability weighting coefficient of 0.066. This indicates that within this high-risk grid, it is recommended that only about 6.6% of the area be sown at high density, while the remaining area should be sown at low density to mitigate risk.

[0058] See appendix Figure 3 As shown in the figure, the solid line is the mapping curve of the combined sowing probability weighting coefficient, which exhibits an S-shaped downward trend. The dots on the curve indicate the conversion center point parameter; when the hydrological vulnerability index equals this parameter, the weighting coefficient is 0.5. This curve reflects that as the hydrological vulnerability index increases, the spatial distribution probability of the high-yield calibration density index decreases rapidly, thus reflecting a conservative decision-making logic that tilts towards the risk baseline density in the calculation.

[0059] In one embodiment of the present invention, step S5 is as follows: The spatial layout planning module performs the process of integrating the baseline sowing density that meets the target yield threshold and the characteristic density set of dual-density interleaving operations, and outwardly planning to generate a composite variable sowing prescription matrix that presents a spatially interleaved arrangement. This module performs array segmentation and separation processing on the processing results of all upstream rasters. Through array segmentation and separation processing, rasters with different decision modes are logically grouped to prepare for subsequent differentiated space filling strategies.

[0060] The planning module traverses the raster data of the entire target field, extracting all raster data marked as "single-density operation" that contains only the benchmark planting density that meets the target yield threshold, and compiling them into an independent first spatial raster table. Simultaneously, all raster data marked as "dual-density interleaved operation" that contains a triplet of high-yield benchmark density index, risk baseline density index, and joint planting probability weighting coefficient are extracted and merged into a second spatial raster table.

[0061] After processing, the planning module first projects the first spatial grid table onto a two-dimensional physical coordinate system. The baseline sowing density within each grid cell that meets the target yield threshold is filled into the corresponding geographic coordinates, thus constructing a deterministic sowing zone with a constant sowing density at the front end. This module then calls a preset spatial jitter mapping algorithm to process the second spatial grid table. The spatial jitter mapping algorithm is a pattern generation technique based on controlled random processes, such as blue noise jitter or Poisson disk sampling. The input to this algorithm is the joint sowing probability weighting coefficient. The output is a two-dimensional Boolean matrix, where the proportion of values ​​"true" is equal to... Furthermore, the "true" values ​​are evenly distributed and avoid clustering, thus physically achieving a staggered arrangement of high and low density planting points at the microscale. For example, if a The grid needs to be The algorithm seeds at a high density based on probability, virtually dividing the cells into 10×10 sub-cells, and then... Select value from 100 sub-cells One location is marked as high density, and the rest are marked as low density.

[0062] The core task of the algorithm is to transform the joint seeding probability weight coefficients within each dual-density interleaved operation grid into a spatially directional, spatially interleaved, nested hierarchical instruction sequence. This spatially interleaved, nested hierarchical instruction sequence is the digital representation of this Boolean matrix, providing the seeding machine's controller with instructions for density switching at the sub-grid scale. For example, if a grid is subdivided into smaller sub-cell arrays, this instruction sequence defines which sub-cells are assigned high density and which are assigned low density.

[0063] Guided by the generated spatially nested hierarchical instruction sequence, the planning module injects high-yield calibration density indicators and risk baseline density indicators into the corresponding 3D coordinate empty points in the second spatial raster table in a cyclical and staggered manner—that is, geographic coordinates plus density values. After injection, the data from deterministic zones and staggered nested zones are integrated into a composite variable sowing prescription matrix that covers the entire region, has an internal structure, and can be directly executed by downstream machinery. The composite variable sowing prescription matrix not only defines the average density of each raster but also defines the micro-spatial structure of density within high-risk areas, enhancing the population's resistance to environmental fluctuations by forming a planting pattern of interspersed tall and short plants.

[0064] For example, continuing the previous example, the spatial layout planning module has completed data processing. The first spatial raster table containing raster 1 is projected, and thus, in the prescription matrix, the area corresponding to raster 1 is assigned a total value of 75,000 plants / hm². 2 The module processes the second spatial raster table containing raster number 5. For raster number 5, the data is the high-yield calibration density index of 80,000 plants / hm². 2 Risk baseline density index: 55,000 plants / hm² 2 The module imports a preset spatial jitter mapping algorithm and virtually subdivides grid number 5 into a 10×10 sub-cell array. The algorithm receives input... You need to select from 100 sub-cells. The algorithm selects specific locations for high-density planting. After execution, it generates a Boolean matrix as a spatially interleaved, nested hierarchical instruction sequence, containing seven randomly and uniformly distributed "true" values. Based on this instruction sequence, the planning module sets the planting density of these seven sub-cells to the high-yield benchmark density of 80,000 plants / hm². 2 The density values ​​for the remaining 93 sub-cells were set at a risk baseline density of 55,000 plants / hm². 2 .

[0065] Based on the actual sub-cell allocation ratio, the average density of grid cell 5 is (80000×7+55000×93) / 100=56750 plants / hm². 2 Its internal structure exhibits an interwoven pattern of two densities. All deterministic density values ​​and the internal structured density distribution map are spatially stitched together to form a complete composite variable seeding prescription matrix. This matrix is ​​saved in a standard geographic data format, awaiting downstream machine access.

[0066] In one embodiment of the present invention, step S6 is as follows: The process of compiling a composite variable seeding prescription matrix into a mechanical operation file that can be called by the underlying hardware, and then distributing it via the bus to control the front-end operation terminal to execute the actual seeding operation, is completed collaboratively by the onboard controller and the actuator. The compilation module within the onboard controller reads the composite variable seeding prescription matrix file and disassembles its geographic coordinate-indexed encoding structure line by line. For each coordinate point and its corresponding density value, the compiler rewrites and constructs it into a data frame adapted to the international tractor bus communication protocol, namely the ISO 11783 standard. These data frames constitute the underlying drive instruction file, which contains a list of correspondences between geographic locations and desired seeding rates. As a task control file, it defines the execution rate, on / off status, and other operational parameters for each geographic grid or path point.

[0067] The onboard controller feeds the underlying driver instruction file as a real-time data stream to the variable frequency seed metering bus connected between the tractor and the seeder via a hardware port. The variable frequency seed metering bus is the physical link for implementing variable seeding; the central controller sends control commands to each independent seeding unit through this bus. For each instruction, the seeding controller adjusts the rotation, release, and rest cycles of the motors of each onboard seed metering unit connected to it. For example, when the instruction requires high-density seeding, the controller increases the motor speed or decreases the rest interval; conversely, it decreases the speed or increases the rest interval, thereby controlling the seed drop rate.

[0068] During this process, the agricultural operation terminal, namely the combination of tractor and seeder, is guided by the onboard global navigation satellite system module to perform on-site cruising operations following a preset movement trajectory. While cruising, the seed metering device continuously executes the benchmark seeding density that meets the target yield threshold according to bus instructions, or quickly switches between high-yield benchmark density indicators and risk baseline density indicators between different sub-regions to perform stepless adjustment of plant spacing for seeding operations.

[0069] After crop emergence, the actual crop survival density in the field is obtained through UAV remote sensing and compared with the expected density in the original prescription matrix to calculate the measured deviation value of crop survival density in the field. This deviation value includes the influence of environmental factors such as sowing deviation, emergence rate, and early pests and diseases. Through a reverse communication link, the measured deviation value data packet, containing geographical location and deviation value, is pushed into the feedback data storage area associated with the corresponding crop variety in the agricultural knowledge graph. Through these feedback data, the model-based prior knowledge is continuously corrected by the posterior results of field measurements. In the next decision cycle, the system performs reverse correction and optimization on key parameters such as vulnerability threshold.

[0070] For example, the vehicle controller reads that its current geographical location is in Within the specified grid area, the corresponding density in the composite variable sowing prescription matrix is ​​the baseline sowing density of 75,000 plants / hm², which meets the target yield threshold. 2 The controller compiles and generates instructions conforming to the international tractor bus communication protocol. The instructions roughly state "Target rate: 75,000 plants / hm²". 2 And send it to the frequency conversion seeding operation bus.

[0071] After receiving the instruction, the seeder adjusts the speed of the seed metering motor to achieve the desired seeding rate and begins sowing. When the agricultural terminal enters grid area 5, the onboard controller begins reading the fine-structured density data within that grid. Assuming the navigation module reports the current vehicle position as a sub-cell within grid 5, and this sub-cell is marked as using the risk baseline density index in a spatially nested hierarchical instruction sequence, the controller then generates the instruction: "Target rate: 55,000 plants / hm²". 2 "and issue the command. When the vehicle moves to the next sub-cell, if that location is marked as using the high-yield calibration density index, the controller updates the command to "Target rate: 80,000 plants / hm²". 2 ".

[0072] The seeder achieved rapid, staggered density switching within grid 5. Weeks later, a drone conducted aerial photography and image analysis of the plot, revealing that the actual average survival density in grid 5 was 3% higher than expected. This set of measured field survival density deviation values, i.e., "grid 5, deviation +3%", was packaged and sent back to the server via a 4G network. Upon receiving this data, the server-side knowledge graph maintenance module analyzed it and concluded that the current vulnerability threshold set for this crop variety was too conservative, leading to an underestimation of its growth potential in risk areas. Therefore, the system automatically fine-tuned the vulnerability threshold, with the adjustment magnitude positively compensating for the measured deviation using a preset step size, for example, increasing it from 0.015 to 0.016. This ensures that future planting decisions can utilize the production potential of high-risk areas while maintaining safety, completing the entire operation and optimization loop.

[0073] See appendix Figure 4 This invention also proposes a decision support system for denser planting in flat and low-lying areas that integrates agricultural knowledge graphs, comprising the following modules: The feature vector construction module acquires basic agricultural geographic parameters and meteorological forecast data streams, and calculates and constructs hydrodynamic feature vectors. The hydrological vulnerability index calculation module performs time series perturbation calculations on the hydrodynamic feature vectors and extracts the independent hydrological vulnerability index for each grid. The spatial raster operation module queries the agricultural knowledge graph to obtain the vulnerability threshold of crop varieties, compares the hydrological vulnerability index with the vulnerability threshold, and determines the current operation mode of the spatial raster. The differentiated sowing density calculation module obtains the benchmark sowing density that meets the target yield threshold when the hydrological vulnerability index is less than the vulnerability limit threshold; when the hydrological vulnerability index is greater than or equal to the vulnerability limit threshold, it obtains the high-yield calibration density index and the risk bottom line density index from the agricultural knowledge graph and calculates the joint sowing probability weighting coefficient. The prescription matrix generation module generates deterministic planting filling zones based on the benchmark planting density that meets the target yield threshold, and interweaves and arranges the high-yield benchmark density index and the risk bottom line density index according to the joint planting probability weighting coefficient to generate a composite variable planting prescription matrix. The instruction compilation and feedback module compiles the composite variable seeding prescription matrix into a low-level driver instruction file, and sends it to the front-end operation terminal through the variable frequency seeding operation bus to control the on-board seeding execution unit to perform the physical seeding operation with matching density.

[0074] Those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A decision support method for reducing plant density and increasing sowing density in flat and low-lying areas by integrating agricultural knowledge graphs, characterized in that, Includes the following steps: S1. Obtain basic agricultural geographic parameters and meteorological forecast data streams, and calculate and construct hydrodynamic feature vectors; S2. Perform time series perturbation calculation on the hydrodynamic feature vector and extract the independent hydrological vulnerability index for each grid. S3. Query the agricultural knowledge graph to obtain the vulnerability threshold of crop varieties, compare the hydrological vulnerability index with the vulnerability threshold, and determine the current operation mode of the spatial grid. S4. When the hydrological vulnerability index is less than the vulnerability limit threshold, obtain the benchmark sowing density that meets the target yield threshold; when the hydrological vulnerability index is greater than or equal to the vulnerability limit threshold, obtain the high-yield benchmark density index and the risk bottom line density index from the agricultural knowledge graph, and calculate the joint sowing probability weighting coefficient. S5. Based on the benchmark sowing density that meets the target yield threshold, a deterministic sowing filling zone is generated, and the high-yield benchmark density index and the risk bottom line density index are intertwined and arranged according to the joint sowing probability weighting coefficient, and a composite variable sowing prescription matrix is ​​generated. S6. Compile the composite variable seeding prescription matrix into a low-level driving instruction file and send it to the front-end operation terminal through the variable frequency seeding operation bus to control the on-board seeding execution unit to perform the physical seeding operation with matching density.

2. The decision support method for reducing plant density and increasing sowing density in flat and low-lying areas, which integrates agricultural knowledge graphs, as described in claim 1, is characterized in that... Acquire basic agricultural geographic parameters and meteorological forecast data streams, and calculate and construct hydrodynamic feature vectors, including the following steps: Read the continuous digital elevation model data of the target field and simultaneously acquire the surrounding soil physical parameters, including the soil saturated hydraulic conductivity. By combining precipitation evolution trajectories contained in short-term meteorological ensemble forecast data with continuous digital elevation model data, spatial raster segmentation is performed to form independent spatial raster. The predicted surface water flow path of each independent spatial grid is calculated based on the continuous digital elevation model data, and the surface soil moisture evapotranspiration potential under no precipitation disturbance is calculated in combination with the surrounding soil physical parameters. By splicing together the predicted surface water flow path and the surface soil moisture evapotranspiration potential, a hydrodynamic feature vector for the entire raster of the target field is generated.

3. The decision support method for reducing plant density and increasing sowing density in flat and low-lying areas, which integrates agricultural knowledge graphs according to claim 2, is characterized in that... The predicted surface water flow path and surface soil moisture evapotranspiration potential are spliced ​​together and fused to generate a hydrodynamic feature vector for the entire raster of the target field, including the following steps: The predicted surface water flow path corresponding to each independent spatial grid includes the surface water outflow direction and water flow accumulation, which are spliced ​​with the surface soil moisture evapotranspiration potential and soil saturated hydraulic conductivity of the same grid in a preset order. Normalization is performed, and the resulting data are fused to generate hydrodynamic feature vectors.

4. The decision support method for reducing plant density and increasing sowing density in flat and low-lying areas, which integrates agricultural knowledge graphs, as described in claim 1, is characterized in that... Perform time-series perturbation calculations on the hydrodynamic feature vectors and extract independent hydrological vulnerability indices for each grid cell, including the following steps: Retrieve multiple simulated precipitation fluctuation distribution trajectories from short-term meteorological ensemble forecast data; The Monte Carlo simulation model was used to initiate multiple evolutionary simulations for each independent spatial grid, and the dynamic bias data of soil moisture triggered by the action of each simulated precipitation fluctuation distribution trajectory on the hydrodynamic feature vector was calculated. Summarize the dynamic bias data of soil moisture output from multiple evolution simulations, and calculate the variance of water content fluctuation that reflects the hydrological sensitivity of a single point. The hydrological vulnerability index of each independent spatial grid is calculated based on the variance of water content fluctuation, and the corresponding average water content reference estimate is calculated.

5. The decision support method for reducing plant density and increasing sowing density in flat and low-lying areas, which integrates agricultural knowledge graphs according to claim 4, is characterized in that... Extract and summarize the soil moisture dynamic bias data from multiple independent evolutionary simulations, and calculate the variance of water content fluctuations reflecting the hydrological sensitivity of a single point. This includes the following steps: Calculate the time average of multiple soil moisture dynamic bias data points; Calculate the sample variance of a sample set consisting of multiple time averages, and define the sample variance as the water content fluctuation variance value.

6. The decision support method for reducing plant density and increasing sowing density in flat and low-lying areas, which integrates agricultural knowledge graphs according to claim 1, is characterized in that... The vulnerability threshold for crop varieties is obtained by querying the agricultural knowledge graph. The hydrological vulnerability index is compared with the vulnerability threshold to determine the current operation mode of the spatial raster, including the following steps: Access agricultural knowledge graphs that store crop physiological parameters; Vulnerability thresholds are retrieved and extracted from the agricultural knowledge graph. Traverse all grid points in the region and compare the hydrological vulnerability index with the vulnerability threshold. When the hydrological vulnerability index is less than the vulnerability limit threshold, the operation mode is determined to be a single-density operation mode; when the hydrological vulnerability index is greater than or equal to the vulnerability limit threshold, the operation mode is determined to be a dual-density interleaved operation mode.

7. The decision support method for reducing plant density and increasing sowing density in flat and low-lying areas, which integrates agricultural knowledge graphs according to claim 4, is characterized in that... When the hydrological vulnerability index is less than the vulnerability threshold, the baseline sowing density that meets the target yield threshold is obtained. When the hydrological vulnerability index is greater than or equal to the vulnerability threshold, the high-yield calibration density index and the risk baseline density index are obtained from the agricultural knowledge graph, and the combined sowing probability weighting coefficient is calculated, including the following steps: The average moisture content reference estimate is input into the benchmark stress calculation function set of the agricultural knowledge graph for calculation, and a benchmark sowing density that meets the target yield threshold is generated. The average moisture content reference estimate is input into the pre-configured high-yield potential estimation function and risk aversion compensation function in the agricultural knowledge graph to calculate the high-yield calibration density index and the risk bottom line density index. The probability weighting coefficient for joint sowing is calculated based on the hydrological vulnerability index. This coefficient is used to determine the spatial coverage ratio of the high-yield target density index.

8. The decision support method for reducing plant density and increasing sowing density in flat and low-lying areas, which integrates agricultural knowledge graphs according to claim 1, is characterized in that... Based on the benchmark sowing density that meets the target yield threshold, deterministic sowing filling zones are generated. Then, according to the joint sowing probability weighting coefficient, the high-yield benchmark density index and the risk baseline density index are interwoven and merged to generate a composite variable sowing prescription matrix, including the following steps: Extract a first spatial raster table containing the benchmark sowing density that meets the target yield threshold, and a second spatial raster table containing the high-yield calibration density index, the risk baseline density index, and the combined sowing probability weighting coefficient. Deterministic seeding filling zones are constructed based on the first spatial raster table; The joint seeding probability weight coefficients are transformed into a spatially interleaved nested hierarchical instruction sequence of the second spatial grid table through a spatial jitter mapping algorithm based on controlled random processes. Following the spatially interwoven nested hierarchical instruction sequence, high-yield calibration density indicators and risk bottom-line density indicators are interwoven and injected to generate a composite variable seeding prescription matrix for the entire domain.

9. The decision support method for reducing plant density and increasing sowing density in flat and low-lying areas, which integrates agricultural knowledge graphs according to claim 8, is characterized in that... The joint seeding probability weight coefficients are transformed into a spatially interleaved nested hierarchical instruction sequence for the second spatial grid table using a spatial dithering mapping algorithm based on controlled random processes, including the following steps: Input the joint seeding probability weighting coefficients into the spatial jitter mapping algorithm based on controlled random processes, and output a two-dimensional Boolean matrix; The two-dimensional Boolean matrix is ​​transformed into a spatially interleaved nested hierarchical instruction sequence, which is used to define the density distribution of sub-cells.

10. The decision support method for reducing plant density and increasing sowing density in flat and low-lying areas, which integrates agricultural knowledge graphs according to claim 1, is characterized in that... The composite variable seeding prescription matrix is ​​compiled into a low-level driver instruction file and sent to the front-end operation terminal via the variable frequency seeding operation bus to control the onboard seeding execution unit to perform the matching density physical seeding operation, including the following steps: Compile the composite variable seeding prescription matrix into a low-level driver instruction file adapted to the communication protocol; Send the underlying driver instruction file to the seeding operation bus to control the release and rest cycle of the onboard seeding execution motor; The control terminal follows the onboard navigation trajectory and performs stepless adjustment sowing operations that match the benchmark sowing density that meets the target yield threshold, or switch between the high-yield benchmark density index and the risk bottom line density index. The measured deviation values ​​of crop field survival density calculated after sowing are obtained, and the measured deviation values ​​are fed back to the agricultural knowledge graph to correct the vulnerability limit threshold.

Images