A distributed farmland soil nutrient dynamic prediction method and system for precision planting

Abstract

The application belongs to the technical field of agricultural information processing, and relates to a distributed farmland soil nutrient dynamic prediction method and system for precision planting, which comprises the following steps: obtaining an initial nutrient state value; obtaining a local coupling reaction source term; marking unknown neighbor node concentration of an adjacent grid as a space diffusion dependent boundary; generating a Dirichlet boundary condition; obtaining a next time nutrient prediction value; and generating a blocking type fertilization control instruction when a high concentration material moving speed exceeds an environmental bearing safety threshold point. The application solves the technical problem that in the prior art, it is difficult to synchronously fuse local biochemical reaction kinetics and neighborhood space physical diffusion effect in a distributed plot network, so that high-fidelity prediction cannot be achieved.

Description

Technical Field

[0001] This invention belongs to the technical field of agricultural information processing, and relates to a distributed method and system for dynamic prediction of farmland soil nutrients for precision planting. Background Technology

[0002] Precision farming is an important direction for the development of modern agriculture. Its core lies in implementing differentiated inputs and management based on the spatial heterogeneity and temporal variability within farmland, aiming to improve crop yield and quality while reducing resource consumption and environmental burden. Real-time dynamic prediction of soil nutrients is a key technological link in achieving this goal. In actual production scenarios, farmland typically consists of numerous scattered plots that are physically connected. The soil nutrient status of each plot is not only governed by internal biochemical processes such as local fertilization and crop absorption, but also significantly affected by the lateral migration of substances between adjacent plots due to water transport.

[0003] Currently, partial differential equations are commonly used for mechanistic modeling in farmland water and nutrient prediction. For example, Chinese patent application CN120689164A uses an improved Richards partial differential equation to establish a water transport model and relies on a central cloud platform for global matrix numerical simulation.

[0004] However, the centralized solution architecture of the above-mentioned fusion mechanism equations reveals significant shortcomings in collaborative solution when dealing with modern large-scale, physically segmented distributed farmland parcel networks: First, using a global centralized server to perform joint numerical simulations of partial differential equations on massive grids results in huge sparse matrix inversion and computational bottlenecks, making it difficult to meet the high-frequency, real-time dynamic prediction requirements of distributed massive parcels; Second, the data flow of this type of centralized system is mainly a one-way convergence from the bottom "end" to the top "cloud", lacking a low-level collaborative mechanism for real-time boundary state exchange between independent distributed computing nodes (i.e., parcel unit edge devices).

[0005] In actual contiguous farmland, dissolved nutrients undergo significant lateral infiltration with soil moisture, and nutrient changes in one plot are directly influenced by material transport from adjacent plots. Traditional centralized mechanistic models struggle to synchronize the physical lateral infiltration effects of adjacent plots in real time through asynchronous communication between underlying nodes when sudden disturbances occur (such as localized fertilization or localized heavy rainfall). The lack of collaborative computational capabilities, utilizing boundary exchange mechanisms to unlock equations in a distributed architecture, leads to the accumulation of prediction errors when dealing with complex farmland networks.

[0006] Based on the above problems, this invention aims to solve the technical problem that existing technologies struggle to simultaneously integrate local biochemical reaction dynamics and neighborhood spatial physical diffusion effects in distributed land parcel networks, thus hindering the achievement of high-fidelity predictions. Summary of the Invention

[0007] In a first aspect, the present invention provides a method for dynamic prediction of distributed farmland soil nutrients for precision planting, comprising the following steps: S1. Based on the multi-source farmland spatiotemporal monitoring data and spatial grid partitioning mapping table, construct a physical adjacency topology graph, integrate baseline variables and current soil nutrient concentration values ​​to obtain initial nutrient status values; S2. Map the sequence of external disturbance events to a standard impulse response signal sequence, extract the nonlinear feedback coefficients to modulate the standard impulse response signal sequence, and obtain the locally coupled reaction source term; S3. Construct a convection-diffusion physics term that includes the current soil nutrient concentration value and the concentration of unknown neighbor nodes. The convection-diffusion physics term and the local coupled reaction source term are merged into a partial differential equation with undetermined boundary conditions, and the concentration of unknown neighbor nodes in adjacent grids is marked as a spatial diffusion-dependent boundary. S4. Receive the adjacent nutrient concentration dataset based on the physical adjacency topology graph, fill the nutrient concentration feature values ​​of the adjacent nutrient concentration dataset into the memory address corresponding to the boundary, and generate Dirichlet boundary conditions. S5. Based on the Dirichlet boundary conditions, perform time integration on the partial differential equation with the given boundary conditions to obtain the nutrient prediction value at the next time step. S6. Upload the nutrient prediction values ​​so that the central server can perform spatial interpolation to reconstruct the continuous soil physical field from the uploaded prediction values ​​and extract the nutrient migration front parameters. When the movement rate of high-concentration substances exceeds the environmental carrying capacity safety threshold, an interception-type fertilization control command is generated.

[0008] A further aspect of the present invention, obtaining the initial nutrient state value, includes the following steps: Acquire the multi-source spatiotemporal monitoring data of farmland, which includes the current soil nutrient concentration value and the dictionary of basic attributes of the current plot; The dictionary of basic attributes of the current plot is parsed to read the matching baseline function of natural nutrient decay in order to extract the baseline variable; The physical adjacency topology graph is established based on the latitude and longitude coordinates of the nodes, and the first-level spatial neighbor node set of the current node is determined. Based on preset weighting coefficients, the baseline variable and the current soil nutrient concentration value are weighted and summed to generate the initial nutrient status value.

[0009] A further aspect of the present invention maps an external disturbance event sequence to a standard impulse response signal sequence, comprising the following steps: The discrete agricultural parameters contained in the multi-source farmland spatiotemporal monitoring data are extracted and combined with sudden meteorological data to generate the external disturbance event sequence; Analyze the content attributes of the external disturbance event sequence and map the external disturbance event sequence into the standard impulse response signal sequence with time-domain decay characteristics.

[0010] A further aspect of the present invention, obtaining locally coupled reaction source terms, includes the following steps: Calculate the current instantaneous amplitude and the first derivative of the nutrient natural decay baseline function associated with the initial nutrient state value; The current instantaneous amplitude and the first derivative of the baseline are used as the nonlinear feedback coefficients to modulate the peak parameters and signal attenuation slope of the standard impulse response signal sequence, and converge to generate the locally coupled response source term.

[0011] A further aspect of this invention involves constructing a convection-diffusion physics term that includes the current soil nutrient concentration value and the concentrations of unknown neighboring nodes, comprising the following steps: Declare the local nutrient concentration change rate variable over time, and define the local coupling reaction source term as the internal transformation driving term; Convection-diffusion physics term is constructed, which includes a hydrodynamic velocity field and a spatial Laplace operator. The convection-diffusion physics term encompasses the convection characteristics driven by water flow, as well as the spatial second derivative of the local node concentration value and the concentration of the unknown neighbor node.

[0012] A further aspect of this invention involves marking the concentrations of unknown neighbor nodes in adjacent grids as spatial diffusion-dependent boundaries, including the following steps: The convection-diffusion physics term and the internal transformation driving term are weighted and combined to generate the partial differential equation for the undetermined boundary conditions; The concentration of the unknown neighbor node in the convection-diffusion physics term is extracted and marked as a spatial diffusion-dependent boundary to lock the closure operation state of the partial differential equation with the undetermined boundary conditions.

[0013] A further aspect of the present invention generates Dirichlet boundary conditions, comprising the following steps: Broadcast the current soil nutrient concentration value to establish a boundary condition exchange request, and enable the underlying listening port to receive the adjacent nutrient concentration dataset sent by the first-level spatial neighbor node set; Verify the adjacent nutrient concentration dataset and perform spatial alignment operations based on the spatial coordinate vectors; Extract the nutrient concentration feature values ​​from the aligned adjacent nutrient concentration dataset and fill them into the memory address.

[0014] A further aspect of the present invention, obtaining the nutrient prediction value for the next time step, includes the following steps: Read the Dirichlet boundary conditions to replace the unknown neighbor node concentrations in the convection-diffusion physics term, and complete the external input parameters of the spatial Laplace operator; The completed convection-diffusion physics term is combined with the partial differential entity containing the locally coupled reaction source term, and a differential discretization transformation is performed on the spatial grid.

[0015] A further aspect of the present invention involves generating an interception-type fertilization control command when the movement rate of a high-concentration substance exceeds the environmental carrying capacity safety threshold, comprising the following steps: The central server aggregates the predicted values ​​of all nodes and performs three-dimensional spatial surface interpolation smoothing to render the continuous soil physical field. The nutrient migration front parameters are extracted by tracing the distribution density of contour lines, and the movement rate of the high-concentration substances is analyzed. When the movement rate of high-concentration substances exceeds the environmental carrying capacity safety threshold, a soil non-point source pollution early warning signal is triggered, and an interception-type fertilization control command is issued to the associated field management terminal.

[0016] Secondly, the present invention provides a distributed farmland soil nutrient dynamic prediction system for precision planting, comprising the following modules: The topology network and initial state generation module constructs a physical adjacency topology graph based on multi-source farmland spatiotemporal monitoring data and a spatial grid partitioning mapping table, and integrates baseline variables and current soil nutrient concentration values ​​to obtain initial nutrient state values; The local coupling reaction source term generation module maps the sequence of external disturbance events into a standard impulse response signal sequence, extracts nonlinear feedback coefficients to modulate the standard impulse response signal sequence, and obtains the local coupling reaction source term. The partial differential equation construction module for undetermined boundary conditions constructs a convection-diffusion physics term that includes the current soil nutrient concentration value and the concentration of unknown neighbor nodes. The convection-diffusion physics term and the local coupled reaction source term are merged into a partial differential equation for undetermined boundary conditions, and the concentration of unknown neighbor nodes in adjacent grids is marked as a spatial diffusion-dependent boundary. The boundary condition generation module receives the adjacent nutrient concentration dataset based on the physical adjacency topology graph, fills the nutrient concentration feature values ​​of the adjacent nutrient concentration dataset into the memory address corresponding to the boundary, and generates the Dirichlet boundary condition. The distributed parallel solution and prediction module performs time integration on the partial differential equation with given boundary conditions based on the Dirichlet boundary conditions to obtain the nutrient prediction value at the next time step. The global field reconstruction and decision intervention module uploads nutrient prediction values ​​so that the central server can perform spatial interpolation to reconstruct the continuous soil physical field from the uploaded prediction values ​​and extract nutrient migration front parameters. When the movement rate of high-concentration substances exceeds the environmental carrying capacity safety threshold, an interception-type fertilization control command is generated.

[0017] In summary, the present invention has the following beneficial technical effects: 1. A partial differential equation with undetermined boundary conditions, including locally coupled reaction source terms and physical diffusion terms, is constructed. Dirichlet boundary conditions are generated using a boundary exchange protocol between nodes, effectively integrating microscopic dynamics and macroscopic permeation effects within a distributed computing architecture. This approach utilizes the spatial Laplace operator to calculate cross-plot concentration gradients and unlocks the equations by listening to neighboring node data, transforming the originally complex global centralized solution into low-dimensional asynchronous parallel computations on each node. This alleviates the pressure of centralized computation while maintaining the physical correlation of substances during lateral migration with the water medium, thus solving the technical problem that existing data-driven models struggle to characterize the cross-boundary evolution of nutrients in distributed plot networks.

[0018] 2. Baseline variables associated with measured data and historical attributes of the land parcel are extracted and weighted for coordinate fusion, providing a physically consistent initial computational field for the model. Simultaneously, the instantaneous amplitude and first derivative of the baseline function are used to nonlinearly modulate the impulse response sequence containing external disturbances such as agricultural and meteorological events. This processing mechanism dynamically correlates external input signals with the real-time buffering capacity and consumption trends within the land parcel. Instantaneous background concentration is used to adjust the impulse peak value to reflect marginal effects, and the consumption rate is used to adjust the signal attenuation slope to reflect the land parcel's resilience. Compared to traditional empirical models that ignore environmental feedback, this mechanism objectively characterizes the time-varying response of a specific land parcel to external disturbances, providing realistic internal transformation driving terms for the partial differential equations.

[0019] 3. Three-dimensional spatial surface interpolation is performed on the local prediction results of distributed nodes to render a continuously evolving dynamic soil physics scene. Nutrient migration front parameters are then extracted to quantify the movement rate and topological distance of high-concentration substances. The numerical results of each discrete grid are transformed into an intuitive macroscopic non-point source pollution evolution trend, enabling proactive capture of the advancing trajectory of nutrient concentration boundaries. By logically comparing the migration rate with a preset environmental carrying capacity safety threshold, the system automatically triggers a non-point source pollution warning when the threshold is exceeded and issues intercepting fertilization control instructions to the field management terminals of related downstream plots. This achieves an extension from mechanism prediction to automated regulation to prevent derivative environmental risks. 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 schematic diagram of the overall steps in the embodiments of this application is disclosed.

[0022] Figure 2 A graph of the nonlinear dynamic modulation mechanism of external disturbance events for the environmental bearing state in the embodiments of this application is disclosed.

[0023] Figure 3 A one-dimensional visualization deconstruction curve of the convection translation and physical diffusion effects in the embodiments of this application is disclosed.

[0024] Figure 4 A schematic diagram of the module architecture in the embodiments of this application is disclosed. Detailed Implementation

[0025] The following is in conjunction with the appendix Figure 1 - Figure 4 A preferred description of the present invention is provided below.

[0026] See attached document Figure 1 This invention proposes a distributed method for predicting the dynamics of soil nutrients in farmland for precision planting, comprising the following steps: S1. Based on the multi-source farmland spatiotemporal monitoring data and spatial grid partitioning mapping table, construct a physical adjacency topology graph, integrate baseline variables and current soil nutrient concentration values ​​to obtain initial nutrient status values; S2. Map the sequence of external disturbance events to a standard impulse response signal sequence, extract the nonlinear feedback coefficients to modulate the standard impulse response signal sequence, and obtain the locally coupled reaction source term; S3. Construct a convection-diffusion physics term that includes the current soil nutrient concentration value and the concentration of unknown neighbor nodes. The convection-diffusion physics term and the local coupled reaction source term are merged into a partial differential equation with undetermined boundary conditions, and the concentration of unknown neighbor nodes in adjacent grids is marked as a spatial diffusion-dependent boundary. S4. Receive the adjacent nutrient concentration dataset based on the physical adjacency topology graph, fill the nutrient concentration feature values ​​of the adjacent nutrient concentration dataset into the memory address corresponding to the boundary, and generate Dirichlet boundary conditions. S5. Based on the Dirichlet boundary conditions, perform time integration on the partial differential equation with the given boundary conditions to obtain the nutrient prediction value at the next time step. S6. Upload the nutrient prediction values ​​so that the central server can perform spatial interpolation to reconstruct the continuous soil physical field from the uploaded prediction values ​​and extract the nutrient migration front parameters. When the movement rate of high-concentration substances exceeds the environmental carrying capacity safety threshold, an interception-type fertilization control command is generated.

[0027] Step S1 is executed on an edge computing node deployed in a distributed farmland plot. The edge computing node subscribes to and acquires real-time multi-source spatiotemporal monitoring data of the farmland from the underlying sensor network deployed within the local plot via its integrated communication interface, following protocols such as Message Queuing Telemetry Transport (MQTT). This multi-source spatiotemporal monitoring data is a structured data stream collected by a distributed array of sensor nodes deployed in the field. The acquisition period for this data is typically set to 15-60 minutes. This range is chosen because a period shorter than 15 minutes would overload the edge node's communication power consumption and generate data redundancy, as nutrient transport in the soil is relatively slow. A period longer than 60 minutes could easily miss nutrient surges caused by sudden heavy rain or integrated fertigation. 15-60 minutes represents a balance between feature capture accuracy and system energy consumption. The multi-source spatiotemporal monitoring data is encapsulated into data packets, the core content of which includes two parts: One is the current soil nutrient concentration value representing the instantaneous chemical state of the soil, specifically a numerical vector composed of nitrogen, phosphorus, and potassium concentrations measured by an ion-selective electrode sensor or a spectrometer. For example, a vector containing the concentrations of multiple elements. The current soil nutrient concentration is expressed in mg / kg. The second part is a dictionary of basic attributes of the current land parcel, which describes the static physical and management attributes of the land parcel. It is encoded in JSON format and records information such as the unique identifier of the land parcel, soil type (e.g., clay loam, sandy loam), pH value, organic matter content, current crops planted, and the latitude and longitude coordinates of the land parcel center determined by the GPS module.

[0028] Meanwhile, the current plot's basic attribute dictionary, serving as a configuration file describing the plot's fixed or gradually changing characteristics, is typically entered by technicians or imported from a farmland geographic information system during system initialization. The processor within the node parses the current plot's basic attribute dictionary, using a set of predefined retrieval rules—a multidimensional lookup table—to map combinations of attributes such as soil type and crop type to a function identifier.

[0029] Using a function identifier, the processor queries and retrieves a matching nutrient natural decay baseline function from the locally stored SQLite database. This nutrient natural decay baseline function is derived from a large amount of historical agronomic experimental data and describes the decay pattern of nutrients in a specific plot over time due to crop absorption and natural loss without external intervention. The processor then retrieves the nutrient natural decay baseline function... Extract its key baseline variables:

[0030] In the formula, The attenuation coefficient; The decay rate; To stabilize the residual, the parameters of the baseline variable... The settings are based on the standard nutrient consumption curve under a specific soil-crop combination. The processor retrieves a preset spatial grid mapping table to establish the spatial relationships between nodes. The spatial grid mapping table stores the identifiers of all plot nodes within the region and their corresponding latitude and longitude coordinates. The processor reads the coordinates of this node. And iterate through all other nodes in the mapping table. The great circle distance between two points can be calculated using the Haversine formula below. :

[0031] in, The average radius of the Earth; , They are nodes and nodes Latitude; The difference in latitude between two points; The difference in longitude between the two points. If the calculated distance... Less than the preset adjacency distance threshold Then the node Add nodes to a set to generate a first-level spatial neighbor set representing the physical neighbors of the current node, and use this set to construct a local view of the physical adjacency topology. Adjacency distance threshold. The settings primarily consider the typical radius of influence for nutrient spatial migration via groundwater infiltration and surface runoff. The processor integrates the acquired current soil nutrient concentration values. Compared with baseline variables at the current time step Calculated baseline expectation The processor generates initial nutrient state values, assigning coordinate weights, according to the following formula through a weighted fusion operation. :

[0032] In the formula, This represents the initial nutrient state value of the final generated product; It is the measured value of the current soil nutrient concentration obtained from the sensor; It is the theoretical baseline value at the current moment calculated based on the selected nutrient natural decay baseline function; These are weighting coefficients used to balance the influence between measured data and theoretical models. As a coordinate-related confidence parameter, it is set to reflect the representativeness of sensor measurements in characterizing the state of the entire plot cell, thereby completing the generation of the initial state and providing a starting point for subsequent spatiotemporal evolution calculations.

[0033] Among them, for the baseline function of natural nutrient decay Parameter settings, such as the rate of nitrogen consumption decay for high-yield maize grown in sandy loam soil. The threshold will be set within the range of 0.04-0.06 based on real-world experimental settings. When defining the first-level spatial neighbor node set, in a plain irrigation area, the adjacency distance threshold... The threshold is typically set at 50-200 m. This threshold is based on the hydrological infiltration patterns in plain irrigation areas. Under these geological conditions, the effective water-salt lateral infiltration and surface runoff influence radius driven by a single agricultural activity or rainfall event is usually within tens to hundreds of meters on the scale of a single standard farmland. The 50-200 m setting can fully cover the effective physical coupling between adjacent plots while avoiding the introduction of irrelevant distant nodes that could cause divergence in the Laplace operator calculation.

[0034] In the calculation of the initial nutrient state value assigned to coordinate weights, the weighting coefficients... Defined as a dimensionless number ranging from 0 to 1, with typical values ​​set between 0.7 and 0.95. The weighting coefficient is set based on the representativeness of the sensor deployment location; if the sensor is located at the center of the plot and has been calibrated, then... A higher value is chosen to enhance the reliability of real-time measurement data; conversely, a lower value is chosen if the sensor is located at the edge of the plot or has low accuracy. It relies more on baseline values ​​derived from historical statistical models.

[0035] For example, suppose we are currently processing an edge computing node with plot ID "A-01". At any given time, the edge computing node obtains the current soil nutrient concentration value from the soil sensor attached to it, and the nitrogen concentration is: mg / kg. At the same time, the node reads the basic attribute dictionary of the current plot stored locally, which contains {"Plot ID":"A-01","Soil Type":"Clay Loam","Crop":"Winter Wheat","Latitude and Longitude":[116.40,39.90]}.

[0036] Based on the current land parcel's basic attribute dictionary, the node matches the nitrogen nutrient natural decay baseline function for "clay loam-winter wheat" using retrieval rules, and extracts its corresponding baseline expectation value. Assuming the baseline expectation value is... mg / kg. The node obtains a spatial grid mapping table, which includes the coordinates of the neighboring plot "A-02" as [116.401, 39.90]. The Earth radius is set. km, adjacency distance threshold m. Calculate the distance between the node and "A-02". , Substituting into the Haversine formula, the result is... m. Due to Node "A-02" was identified as a member of the first-level spatial neighbor node set.

[0037] Weighting coefficients are set based on the sensor deployment conditions of plot "A-01". Substituting the above values ​​into the fusion formula, the initial nutrient state values ​​assigned to the coordinate weights are calculated. mg / kg. Initial nutrient state value assigned to coordinate weights. mg / kg will be used as the value for plot "A-01" in The initial nitrogen state at time t is passed to subsequent steps for dynamic prediction calculation.

[0038] Step S2 is executed on the edge computing node. The processor parses the multi-source farmland spatiotemporal monitoring data packets acquired in the previous stage and extracts discrete agricultural parameters related to nutrient concentration. Discrete agricultural parameters are data instructions that guide agricultural production activities, such as variable fertilizer prescription maps generated by precision agriculture platforms. Discrete agricultural parameters are usually issued by farmland management information systems or entered through manual interfaces, such as structured data recorded as {"event type":"fertilization","fertilizer type":"urea","application amount":Q,"time":t_event}.

[0039] Meanwhile, the node calls the external weather service application programming interface (API) through the integrated network interface to obtain sudden meteorological data within the current calculation cycle. Sudden meteorological data includes events such as heavy rain and persistent drought that may cause drastic changes in nutrient concentration, such as {"event type":"heavy rainfall","rainfall amount":R,"duration":T_d}. The data usually comes from external meteorological data service interfaces or publicly available commercial meteorological databases, and the data refresh frequency can reach the minute level.

[0040] The processor assembles and sorts event records from different sources with similar timestamps, forming a specific sequence of external disturbance events arranged chronologically. The processor then analyzes the content attributes of each event in the sequence. For a fertilization event, it is mapped to a positive pulse signal representing the injection of nutrient concentration; for a heavy rainfall event, it is mapped to a negative pulse signal representing nutrient loss due to leaching. The processor implements this mapping by selecting a standard impulse response signal model from a pre-built function library. For example, it may use a gamma distribution function to simulate the release and transformation process after nutrient injection, or an exponential decay function to simulate the leaching process.

[0041] The above mapping process generates one or more standard impulse response signal sequences with standard waveform parameters. Waveform parameters of a standard impulse response signal sequence, such as the shape parameters in a gamma distribution. Parameters related to decay rate Pre-calibrated based on typical agronomic experimental data. The processor retrieves the nutrient natural decay baseline function matched to the initial nutrient state value in S1. Regarding the current calculation step size The processor calculates two key local state indicators: One is the current instantaneous amplitude of the baseline function. This represents the buffering capacity of soil nutrients; The second is the first derivative of the baseline function at that point. This represents the rate of natural nutrient depletion. The processor performs a nonlinear modulation process based on the current instantaneous amplitude and the first derivative of the baseline. The current instantaneous amplitude... First derivative with baseline As a nonlinear feedback coefficient, it is dynamically fed back to the event response model, thereby adjusting the parameters of the standard impulse response signal sequence.

[0042] Specifically, using Adjusting the peak value of the pulse signal, i.e., the peak parameter, reflects the impact of soil background fertility on the availability of newly added nutrients; simultaneously, utilizing... Adjusting the attenuation slope of the pulse signal, i.e., the signal attenuation slope, reflects the impact of the current crop absorption or soil retention rate on the persistence of newly added nutrients. In the nonlinear modulation process of this step, the specific modulation algorithm can be expressed as follows:

[0043] Among them, the modulated peak parameters Parameters related to signal attenuation It is calculated using the following formula:

[0044]

[0045] In the formula, It is a modulated pulse signal; It is a standard pulse signal; and These are the standard and modulated peak amplitudes, respectively. and These are the parameters related to the attenuation rate of the standard and modulated attenuation rates, respectively. This is the upper limit of nutrient capacity corresponding to the soil type of this plot; It is the historical maximum consumption rate, used as a normalization factor to ensure the correct dimensions and numerical stability of the modulation coefficient; This is a preset modulation sensitivity coefficient. It indicates the sensitivity when soil background nutrients... At higher levels, the apparent peak value of new nutrient events It will be inhibited; while the natural rate of nutrient consumption At a faster rate, the decay rate Increase.

[0046] Modulated pulse signal This reflects the response of a specific site to external disturbances. The modulated pulse signals generated by all events within the current calculation period are algebraically superimposed, ultimately converging to generate a single time-series function, namely the locally coupled reaction source term. Locally coupled reaction source term It will serve as the source term input for the partial differential equation, driving the overall spatiotemporal evolution of subsequent nutrient concentrations.

[0047] Among these, the waveform parameter calibration of the standard impulse response signal sequence, for example, the fitting of the nitrogen release curve of the urea hydrolysis-nitrification process in soil at 25°C, is as follows: Gamma distribution (in days). Locally coupled reaction source term. The unit is expressed as the rate of change in nutrient concentration, for example, mg / kg / day. In addition, the modulation sensitivity coefficient... As an empirical constant, its value ranges from 0.5 to 1.5. The basis for setting this empirical range is to ensure the numerical stability of the nonlinear feedback mechanism. If the coefficient is lower than 0.5, the source term will be too slow to respond to the soil background consumption rate and will not be able to reflect the marginal effect. If it is higher than 1.5, it will easily cause mathematical divergence distortion of the source term when nutrients are at the extreme point of consumption. 0.5-1.5 can achieve a balance between physical reality and numerical calculation stability.

[0048] For example, continuing with plot "A-01", its initial nitrogen nutrient status value mg / kg. At the time step The node received a discrete agricultural parameter instruction: apply urea to plot "A-01", equivalent to the amount of pure nitrogen used. kg / ha. Fertilization events constitute a sequence of external disturbance events.

[0049] Map this fertilization event to a standard impulse response signal, selecting shape parameters. Attenuation rate related parameters The gamma function, its standard peak parameter Set to 15 mg / kg / day. At this point, the processor retrieves the nutrient natural decay baseline function determined in S1, assuming it is... .exist The baseline instantaneous amplitude is calculated at the time of day 10. mg / kg and its first derivative mg / kg / day.

[0050] Assuming the upper limit of nitrogen capacity for this soil type mg / kg, historical maximum consumption rate mg / kg / day, modulation sensitivity coefficient Substitute the numerical values ​​into the modulation formula to calculate the modulated peak parameters: mg / kg / day. Calculate the relevant parameters of the modulated attenuation rate: Therefore, this fertilization event produces a modulated gamma pulse function with a shape parameter of 2 and a decay rate correlation parameter of 4.18. The modulated gamma pulse function is the source term of the locally coupled reaction. This information is then passed to the next step to describe the nonlinear growth and transformation process of nitrogen concentration within plot "A-01" after fertilization.

[0051] See appendix Figure 2 The horizontal dashed line in the figure represents the soil natural attenuation baseline. The light pink thick line represents the traditional standard pulse superimposed without environmental feedback, while the green thick line represents the modulated pulse introduced by the local coupling reaction source term in this invention. The shaded area with red diagonal lines formed between the light pink and green curves represents the ineffective nutrient predictions reduced by soil environmental capacity limitations. This embodiment extracts the instantaneous amplitude and the first derivative of the baseline as nonlinear feedback coefficients, dynamically suppressing the pulse peak and adjusting the attenuation slope through mathematical modulation. The removal of the aforementioned red diagonal shaded area demonstrates that the mechanism of this embodiment can restore the true biochemical buffering effect under soil capacity constraints, thereby significantly eliminating redundant predictions and improving the high fidelity of dynamic prediction.

[0052] Step S3 is executed on the edge computing node, and the processor in memory is the current computing node (denoted as node). Declare a core state variable, namely the rate of change of local nutrient concentration over time. The concentration value at the next moment is calculated from the rate of change of local nutrient concentration over time using a time integration algorithm. It is a node Over time The changing nutrient concentration, its initial value Set as the initial nutrient state value generated in S1 The processor will compute the locally coupled reaction source terms generated in S2. Loaded into memory.

[0053] The locally coupled reaction source term, serving as the internal transformation driving term within the constructed partial differential equation system, describes the increase or decrease in local nutrient concentration caused by external events such as fertilization and rainfall, as well as the soil's own biochemical reactions. Based on the soil type in the current plot's basic attribute dictionary parsed in S1, the processor queries and retrieves the corresponding material diffusion coefficient from the local parameter database. This is used to construct the convective diffusion physics terms. The value of the mass diffusion coefficient is set based on factors such as soil porosity, water content, and temperature.

[0054] Processor constructs spatial Laplacian operator Discretization expression, the discretization expression is obtained by introducing nodes. Compared to its neighboring nodes in the physical adjacency topology graph, such as node This is achieved through concentration differences, specifically by summing the concentration differences between all neighboring nodes and the local node. Therefore, the convection-diffusion physics term encompasses the local node concentration value. And the currently unknown concentration of neighboring nodes. The processor then uses the spatial second derivative information. It further weights and merges this convection-diffusion physics term with the previously defined internal transformation driving term to generate a partial differential equation reflecting the undetermined boundary conditions of the soil microphysical processes:

[0055] In the partial differential equations with undetermined boundary conditions constructed above, For the hydrodynamically driven convection term, where The velocity field vector of water flow caused by farmland rainfall, surface runoff, or groundwater seepage; and These are dimensionless weighting coefficients that respectively adjust the relative importance of diffusion / convection effects and reaction effects; the spatial Laplace operator. In a two-dimensional discrete grid spacing of The common central difference approximation form is as follows:

[0056] in, The current node concentration is represented by , and the others by the concentrations of the four neighboring nodes (up, down, left, and right). The partial differential equation with undetermined boundary conditions describes the change in local nutrient concentration as a result of the combined effects of spatial diffusion and local reactions. However, because this equation lacks definite values ​​from neighboring nodes as external input, it is in a state of undetermined boundary conditions and cannot be solved independently.

[0057] The processor extracts all variables involving neighboring nodes from the convection-diffusion physics terms through syntax analysis, i.e., the unknown neighboring node concentrations for each term. The processor marks the memory addresses of these variables as external dependencies to be solved, and uniformly names this set of dependencies as the spatial diffusion dependency boundary. The spatial diffusion dependency boundary is a set of neighboring node state data on which solving the current node state depends; these data constitute the Dirichlet or Neumann boundary conditions required to solve the partial differential equation.

[0058] This marking action declares that the partial differential equation with undetermined boundary conditions cannot be solved independently due to the lack of boundary conditions, thus locking the node's own computational task and forcing it to wait for external data input before completing the subsequent integration operation.

[0059] For specific soil types, such as clay loam, the mass diffusion coefficient The effective diffusion coefficient of nitrogen is usually in Within the range. Weighting coefficients and The diffusion coefficient is usually set to 1; The unit is Laplace operator The unit is Therefore, the convective-diffusion physics term The unit is The local coupling reaction source terms need to be converted using a time conversion factor. The units are simultaneously converted to seconds, thus making the reaction terms and concentration change rates consistent. The dimensions are kept consistent.

[0060] For example, let's continue with the edge computing node of plot "A-01". At any given time, the processor constructs the nitrogen concentration for that node. The evolution equation. Its initial concentration value is the initial nutrient state value calculated in S1. mg / kg. The processor loads the locally coupled reaction source term generated for this node in S2. The processor reads the nitrogen diffusion coefficient based on the soil type "clay loam" for "A-01". Given that "A-02" is its only neighbor, and the distance between the plots is... m. Set weights Since the current grid only establishes a single valid adjacent node, the system uses a unidirectional mass transfer flux approximation to replace the complete spatial second-order Laplace operator. It is assumed that there are no heavy rainfall or irrigation events during the current computation period, and the surface and groundwater dynamic velocity fields are... ≈0, meaning the hydrodynamic convection term is temporarily zero. Under this baseline condition, the specific form of the partial differential equation for the undetermined boundary conditions generated for node "A-01" is:

[0061] The processor then analyzes the partial differential equations with undetermined boundary conditions and identifies the variables. That is, the nitrogen concentration of neighboring node "A-02" is an unknown variable. The processor will store this variable in memory. The associated pointer is marked as "unresolved_boundary_dependency", forming a spatially diffused dependency boundary. The equation solving process is suspended due to the detection of this unresolved dependency marker, and the node computation process enters a locked state, awaiting external acquisition. The specific value.

[0062] See appendix Figure 3 The figure shows four bell-shaped evolution curves (dark blue, purple, orange, and green) representing the spatial concentration distribution of the same locally high-concentration nutrient cluster at four consecutive progressive evolutionary moments, from the initial moment T0 to the deep evolution moment T3. It can be observed that as time progresses, the waveform representing the nutrient cluster becomes shorter and wider. This physical phenomenon of amplitude decay and penetration into the low-concentration regions on both sides is precisely due to the spatial Laplace operator in the partial differential equation, i.e., the physical diffusion term. The driving force truly reflects the spontaneous lateral diffusion effect of dissolved substances under the influence of the medium concentration gradient.

[0063] Furthermore, the red dotted lines connecting the peaks of each evolution curve in the figure represent the centroid of the high-concentration nutrient clusters, i.e., the horizontal displacement trajectory of the highest point of material accumulation over time. In the diffusion model, the wave crests only decay in place; however, the wave crests in the figure exhibit a significant and continuous displacement to the right. This horizontal displacement trajectory is precisely due to the hydrodynamic velocity field vector in the partial differential equation, i.e., the convection term. Driven by this, the macroscopic process of fertilizer substances being physically carried by water flow and pushed downstream under the influence of surface runoff or subsurface infiltration is quantified. This partial differential equation successfully achieves the mathematical decoupling and coupling of convective translation phenomena (displacement of the peak connection line) and physical diffusion phenomena (shortening and widening of multiple curves) under a unified time integral, providing physical mechanism support for subsequent capture of the movement rate of high-concentration substances and implementation of non-point source pollution early warning.

[0064] In step S4, the edge computing node activates the boundary exchange protocol in the underlying network communication module. Through active communication between nodes, it releases the computational lock-in state formed in the previous stage, providing the necessary external conditions for solving partial differential equations. The boundary exchange protocol defines the data packet format, exchange timing, and error handling mechanism, following a publish / subscribe model. Nodes share their current soil nutrient concentration value, i.e., ... The initial nutrient state value calculated by S1 at time S1 Together with the node identifier and the current timestamp, it is encapsulated into a standardized data packet.

[0065] Based on preset data distribution rules, nodes specify target topics or a direct list of target IP addresses and broadcast standardized data packets to their first-level neighbor nodes via wireless networks, such as LoRaWAN or NB-IoT. They also send boundary condition exchange requests to peripheral nodes via broadcast.

[0066] Meanwhile, the network interface of the edge computing node switches to listening mode, continuously listening to a preset communication port, such as UDP port 12345 or subscribing to a specific topic. Based on the physical adjacency topology established in S1, the node expects to receive data from members of the first-level spatial neighbor node set. When a data packet arrives at the listening port, the protocol stack delivers it to the processor.

[0067] After receiving the adjacent nutrient concentration dataset synchronized from neighboring nodes, the processor, which includes the neighbor node ID, timestamp, coordinates, and current nutrient concentration value, immediately executes a verification procedure. The processor checks the packet header to verify that the timestamp matches the current computing clock cycle and verifies the data structure header to ensure its format conforms to the protocol specification. After successful verification, the processor extracts the spatial coordinate vectors of the neighboring nodes from the packet and compares them with the coordinates of the local node, performing a spatial alignment operation. This spatial alignment operation ensures that data from different node coordinate systems can be unified into the same computational grid. Based on the relative orientation between the two nodes, the spatial alignment operation maps the received neighbor node concentration values ​​to the corresponding direction in the local discretized grid. For example, if a neighboring node is due east of the local node, its concentration value is used to calculate the eastward spatial gradient.

[0068] The processor extracts nutrient concentration feature values ​​from the aligned neighboring nutrient concentration dataset and fills these feature values ​​into the memory addresses marked as spatial diffusion dependency boundaries in the previous stage. This unlocks the unknown variables of the partial differential equation, forming Dirichlet boundary conditions that can be directly used for numerical computation. The Dirichlet boundary conditions directly specify the function values ​​on the solution domain boundary; that is, the boundary values ​​of the local computational unit are directly set by the concentration values ​​of neighboring nodes. This unlocks the equation and allows subsequent numerical integration operations to proceed.

[0069] Through the boundary exchange protocol in this step, edge computing nodes use communication ports to synchronize structured adjacent nutrient concentration datasets, perform spatial alignment operations, and construct Dirichlet boundary conditions, providing a basis for continuous numerical integration operations.

[0070] For example, continuing with node "A-01", its concentration value is calculated because it lacks the value of its neighbor "A-02". The node is in a locked state. At this time, node "A-01" initiates the boundary exchange protocol, sharing its own nitrogen concentration value. The mg / kg, ID "A-01", and timestamp are packaged and broadcast to neighbor "A-02".

[0071] Similarly, node "A-02" also broadcasts its own status to "A-01". The listening port of node "A-01" receives the data packet from "A-02", the processor parses and confirms the timestamp match, and extracts the core data: {ID:"A-02", N_concentration:34.5mg / kg, coordinate:[116.401, 39.90]}.

[0072] After performing spatial alignment, "A-02" was identified as being located on the "right side" of its one-dimensional diffusion model. The processor then extracted the concentration value. mg / kg is written to the memory address previously marked "unresolved_boundary_dependency", which is the variable in the partial differential equation. The location it points to. At this point, the equation... All variables in the equation have obtained definite values. The locked state of the equation is released, the complete Dirichlet boundary conditions are formed, and the numerical solution stage begins.

[0073] Step S5 is executed on the edge computing node. The processor reads the Dirichlet boundary conditions constructed in S4 through inter-node communication and solves the equations using the complete boundary conditions. The processor uses the nutrient concentration value sequence of neighboring nodes contained in the Dirichlet boundary conditions to replace the missing variables that are locked in the convection-diffusion physics term constructed in S3 one by one. By replacing the missing variables, the external input parameters of the Laplace operator in the computational space are completed, and the convection-diffusion physics term with undetermined boundary conditions is transformed into a closed mathematical expression.

[0074] The processor combines the parameter-completed convection-diffusion physics terms with the locally coupled reaction source terms generated in S2 to form a closed partial differential equation. The processor then performs a differential discretization transformation on the closed partial differential equation for the spatial grid. Specifically, using the finite difference method, the spatial derivatives in the equations are replaced with the algebraic difference of the concentration values ​​of adjacent grid nodes, transforming the partial differential equation into a system of ordinary differential equations or a system of algebraic equations.

[0075] Before time integration, the safety monitoring program is initiated. Based on the transformed difference matrix, the safety monitoring program calculates the spatial gradient distribution between grid nodes at the current time. The processor divides the concentration difference between adjacent nodes by the physical distance to obtain the local spatial gradient data, which is then compared with preset extreme points.

[0076] The preset extreme points are gradient thresholds set based on historical data statistics. If the absolute value of the local spatial gradient data exceeds the extreme point, the preset extreme point refers to the gradient threshold characterizing abnormal jumps in spatial concentration, such as 5 mg·kg⁻¹. -1 ·m -1 This value is set based on the effective physical diffusion limit of clayey soil media. Natural infiltration usually cannot generate a concentration difference exceeding this gradient within a short time and distance. If this physical limit is exceeded, it is likely due to sensor hardware failure or communication errors. Therefore, it can be used as a criterion for anomaly isolation. The corresponding grid nodes with anomalies are then marked as fault detection nodes and isolated from the main calculation process within this calculation cycle.

[0077] For grid nodes without anomalies, the processor invokes an explicit time integration algorithm, such as the forward Euler method, to perform a forward time step computation on the discretized system of equations. Using an explicit time integration algorithm (taking the forward Euler method as an example), the partial differential equations are discretized into a system of algebraic equations of the following form:

[0078] Among them, superscript and These represent the current and next time steps, respectively. It is the time step. It is the nutrient concentration to be solved at the next moment; It is the discrete form of the spatial Laplace operator at the current time step, including the local node concentration. and neighbor node concentration The explicit time integration algorithm can be used to directly calculate the nutrient levels based on the current known state. This algorithm uses algebraic operations to obtain the state at the next time step, generating a nutrient prediction value for the current node that includes the physical field infiltration effect of neighboring plots. This nutrient prediction value is used as the initial value for the next calculation cycle, or aggregated at the central platform to form a global situation map.

[0079] In explicit stepwise computation, since the local and neighboring node variables at the current time step have been completed to a known state through boundary conditions, the concentration at the next time step can be directly obtained algebraically without the need for complex sparse matrix inversion. When applying preset extreme points, such as for nitrogen in soil through natural diffusion, if its concentration gradient exceeds 5 mg·kg⁻¹ within 1 hour... -1 ·m -1 If so, it is considered an abnormal signal.

[0080] For example, the calculation process for plot "A-01" continues. The processor has already obtained the concentration of neighboring plot "A-02". mg / kg, the equation after unlocking is: The system performs gradient monitoring and calculates the spatial gradient as follows: mg·kg -1 ·m -1 The absolute value of the spatial gradient is much smaller than the preset value of 5 mg·kg⁻¹. -1 ·m -1 The security monitoring abnormal gradient threshold is met, and the node status is normal. The processor uses the forward Euler method for time-step calculation, and the time step size is set. h=3600 s.

[0081] exist At any given time, the local concentration mg / kg, reaction source term calculated in S2 instantaneous speed is mg·kg -1 ·day -1 After conversion mg·kg -1 ·s -1 Substitute the values ​​into the equation to perform the calculation: ; The concentration at the next moment is the concentration at the current moment plus two increments: The system calculated that the predicted nutrient value for plot "A-01" at the next moment is 36.0208 mg / kg.

[0082] In step S6, the master node control program deployed on the central cloud server sends a data aggregation command to the active edge computing nodes within the system at the end of each computing clock cycle, summarizing the local prediction results of all distributed nodes to generate a global physical field view. Upon receiving the command, each computing node packages and uploads the next-time nutrient prediction value calculated in S5, along with the plot identifier and physical coordinates. The central server aggregates the next-time nutrient prediction values ​​uploaded by all nodes, forming a discrete data point set containing [coordinates, prediction value].

[0083] The processor invokes three-dimensional spatial surface interpolation algorithms, such as Kriging interpolation, to generate a continuous soil physical scene reflecting the dynamic distribution of nutrient concentration in two-dimensional or three-dimensional space. This continuous dynamic soil physical scene is visualized and rendered as a pseudo-color cloud map or contour map. The process then proceeds to the feature extraction stage, tracking the density changes of the contour lines in the computational physical field. By analyzing the geometry and time-varying trajectory of dense contour line regions, nutrient migration front parameters characterizing the advancing boundary of high-concentration nutrient regions are extracted.

[0084] Nutrient migration front parameters quantify the movement characteristics of nutrient aggregates in the soil medium, including the location, shape, direction of movement, and velocity of the migration front. The processor analyzes these parameters and, in conjunction with information such as topographic slope and soil hydrological conductivity, assesses the geometric topological distance parameters traversed by the nutrient migration front per unit time for infiltration or runoff. The assessment of these geometric topological distance parameters takes into account Euclidean distance and the influence of topographic relief and ditch network on flow paths to quantify the extent of environmental impact.

[0085] The decision-making logic compares the calculated movement rate of high-concentration substances with a preset environmental carrying capacity safety threshold. Once the movement rate exceeds this threshold, a soil non-point source pollution warning signal is triggered. This threshold is set based on the trade-off between the actual topological buffer distance from the target farmland to sensitive water bodies (such as water source protection areas) and the soil's natural degradation rate. For example, a measured rate of 3 m / h means that the high-concentration pollution front will breach the physical buffer zone and reach the water body before soil microorganisms complete natural degradation, leading to irreversible eutrophication. Therefore, this rate must be used as the critical threshold for triggering physical interception.

[0086] Soil non-point source pollution early warning signals are triggered on the central monitoring platform. Simultaneously, the system automatically locates the device addresses of field management terminals linked to grids of downstream plots associated with the migrating front and proactively issues intercepting fertilization control commands. These commands proactively prevent potential environmental pollution events through adjustments, including actions such as "immediately suspending fertilization plans," "opening drainage ditch interception valves in specific areas," or "activating vegetation buffer zone protection mode at the edge of downstream plots." Based on the assessed nutrient migration front parameters and geometric topological distance parameters, and through spatial surface interpolation, feature extraction, and decision-making logic, intercepting fertilization control commands are issued to control the field management terminals.

[0087] For example, the central server is in The system aggregated the predicted nutrient values ​​for the next time step for 100 plot nodes, including "A-01". The value for "A-01" was 36.0208 mg / kg. Using Kriging interpolation, a nitrogen concentration distribution cloud map covering the entire farm was rendered. Analysis of the nitrogen concentration distribution cloud map revealed that in the fertilization area including "A-01", a sudden heavy rainfall or concentrated irrigation event activated the hydrodynamic convection term in the partial differential equation (i.e., the water flow velocity field vector v≠0 at this time), thus forming a high-concentration core area. The 50 mg / kg isopleth was rapidly moved downstream by approximately 5 m in the past calculation cycle under the influence of water flow and convection. Nutrient migration front parameters were extracted, and the high-concentration substance movement rate in the core area was calculated to be 5 m / h.

[0088] The processor assesses the geometric and topological distance parameters of the seepage boundary crossed by the migrating front and predicts that it will reach the drainage ditch adjacent to the secondary drinking water source protection area within the next 6 hours. The system's preset environmental carrying capacity safety threshold sets the upper limit for nitrogen migration rate in this area at 3 m / h. Since 5 m / h exceeds the environmental carrying capacity safety threshold, a soil non-point source pollution warning signal is triggered. The central control console marks the migrating front area in red and issues an alarm. Simultaneously, the field management terminal of plot "C-05" downstream of the drainage ditch is retrieved, and the following command is issued to it: {"command":"set_gate_status","gate_id":"G03","status":"closed","duration":"12h"}. This interception-type fertilization control command closes the smart drainage gate for 12 hours to intercept surface runoff carrying high concentrations of nitrogen.

[0089] See appendix Figure 4 This invention also proposes a distributed farmland soil nutrient dynamic prediction system for precision planting, comprising the following modules: The topology network and initial state generation module constructs a physical adjacency topology graph based on multi-source farmland spatiotemporal monitoring data and a spatial grid partitioning mapping table, and integrates baseline variables and current soil nutrient concentration values ​​to obtain initial nutrient state values; The local coupling reaction source term generation module maps the sequence of external disturbance events into a standard impulse response signal sequence, extracts nonlinear feedback coefficients to modulate the standard impulse response signal sequence, and obtains the local coupling reaction source term. The partial differential equation construction module for undetermined boundary conditions constructs a convection-diffusion physics term that includes the current soil nutrient concentration value and the concentration of unknown neighbor nodes. The convection-diffusion physics term and the local coupled reaction source term are merged into a partial differential equation for undetermined boundary conditions, and the concentration of unknown neighbor nodes in adjacent grids is marked as a spatial diffusion-dependent boundary. The boundary condition generation module receives the adjacent nutrient concentration dataset based on the physical adjacency topology graph, fills the nutrient concentration feature values ​​of the adjacent nutrient concentration dataset into the memory address corresponding to the boundary, and generates the Dirichlet boundary condition. The distributed parallel solution and prediction module performs time integration on the partial differential equation with given boundary conditions based on the Dirichlet boundary conditions to obtain the nutrient prediction value at the next time step. The global field reconstruction and decision intervention module uploads nutrient prediction values ​​so that the central server can perform spatial interpolation to reconstruct the continuous soil physical field from the uploaded prediction values ​​and extract nutrient migration front parameters. When the movement rate of high-concentration substances exceeds the environmental carrying capacity safety threshold, an interception-type fertilization control command is generated.

[0090] The above embodiments are only used to illustrate the technical solutions of the present invention. 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. 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 method for dynamic prediction of distributed farmland soil nutrients for precision planting, characterized in that, Includes the following steps: S1. Based on the multi-source farmland spatiotemporal monitoring data and spatial grid partitioning mapping table, construct a physical adjacency topology graph, integrate baseline variables and current soil nutrient concentration values ​​to obtain initial nutrient status values; S2. Map the sequence of external disturbance events to a standard impulse response signal sequence, extract the nonlinear feedback coefficients to modulate the standard impulse response signal sequence, and obtain the locally coupled reaction source term; S3. Construct a convection-diffusion physics term that includes the current soil nutrient concentration value and the concentration of unknown neighbor nodes. The convection-diffusion physics term and the local coupled reaction source term are merged into a partial differential equation with undetermined boundary conditions, and the concentration of unknown neighbor nodes in adjacent grids is marked as a spatial diffusion-dependent boundary. S4. Receive the adjacent nutrient concentration dataset based on the physical adjacency topology graph, fill the nutrient concentration feature values ​​of the adjacent nutrient concentration dataset into the memory address corresponding to the boundary, and generate Dirichlet boundary conditions. S5. Based on the Dirichlet boundary conditions, perform time integration on the partial differential equation with the given boundary conditions to obtain the nutrient prediction value at the next time step. S6. Upload the nutrient prediction values ​​so that the central server can perform spatial interpolation to reconstruct the continuous soil physical field from the uploaded prediction values ​​and extract the nutrient migration front parameters. When the movement rate of high-concentration substances exceeds the environmental carrying capacity safety threshold, an interception-type fertilization control command is generated.

2. The method for dynamic prediction of distributed farmland soil nutrients for precision planting according to claim 1, characterized in that, Obtaining the initial nutrient state value includes the following steps: Acquire the multi-source spatiotemporal monitoring data of farmland, which includes the current soil nutrient concentration value and the dictionary of basic attributes of the current plot; The dictionary of basic attributes of the current plot is parsed to read the matching baseline function of natural nutrient decay in order to extract the baseline variable; The physical adjacency topology graph is established based on the latitude and longitude coordinates of the nodes, and the first-level spatial neighbor node set of the current node is determined. Based on preset weighting coefficients, the baseline variable and the current soil nutrient concentration value are weighted and summed to generate the initial nutrient status value.

3. The method for dynamic prediction of distributed farmland soil nutrients for precision planting according to claim 1, characterized in that, Mapping a sequence of external disturbance events to a standard impulse response signal sequence includes the following steps: The discrete agricultural parameters contained in the multi-source farmland spatiotemporal monitoring data are extracted and combined with sudden meteorological data to generate the external disturbance event sequence; Analyze the content attributes of the external disturbance event sequence and map the external disturbance event sequence into the standard impulse response signal sequence with time-domain decay characteristics.

4. The method for dynamic prediction of distributed farmland soil nutrients for precision planting according to claim 2, characterized in that, To obtain the locally coupled reaction source term, the following steps are included: Calculate the current instantaneous amplitude and the first derivative of the nutrient natural decay baseline function associated with the initial nutrient state value; The current instantaneous amplitude and the first derivative of the baseline are used as the nonlinear feedback coefficients to modulate the peak parameters and signal attenuation slope of the standard impulse response signal sequence, and converge to generate the locally coupled response source term.

5. The method for dynamic prediction of distributed farmland soil nutrients for precision planting according to claim 1, characterized in that, Constructing a convection-diffusion physics term that includes the current soil nutrient concentration value and the concentrations of unknown neighboring nodes includes the following steps: Declare the local nutrient concentration change rate variable over time, and define the local coupling reaction source term as the internal transformation driving term; Convection-diffusion physics term is constructed, which includes a hydrodynamic velocity field and a spatial Laplace operator. The convection-diffusion physics term encompasses the convection characteristics driven by water flow, as well as the spatial second derivative of the local node concentration value and the concentration of the unknown neighbor node.

6. The method for dynamic prediction of distributed farmland soil nutrients for precision planting according to claim 5, characterized in that, Marking the concentrations of unknown neighbor nodes in adjacent grids as spatial diffusion-dependent boundaries includes the following steps: The convection-diffusion physics term and the internal transformation driving term are weighted and combined to generate the partial differential equation for the undetermined boundary conditions; The concentration of the unknown neighbor node in the convection-diffusion physics term is extracted and marked as a spatial diffusion-dependent boundary to lock the closure operation state of the partial differential equation with the undetermined boundary conditions.

7. The method for dynamic prediction of distributed farmland soil nutrients for precision planting according to claim 2, characterized in that, Generating Dirichlet boundary conditions includes the following steps: Broadcast the current soil nutrient concentration value to establish a boundary condition exchange request, and enable the underlying listening port to receive the adjacent nutrient concentration dataset sent by the first-level spatial neighbor node set; Verify the adjacent nutrient concentration dataset and perform spatial alignment operations based on the spatial coordinate vectors; Extract the nutrient concentration feature values ​​from the aligned adjacent nutrient concentration dataset and fill them into the memory address.

8. The method for dynamic prediction of distributed farmland soil nutrients for precision planting according to claim 5, characterized in that, To obtain the nutrient prediction value for the next time step, the following steps are included: Read the Dirichlet boundary conditions to replace the unknown neighbor node concentrations in the convection-diffusion physics term, and complete the external input parameters of the spatial Laplace operator; The completed convection-diffusion physics term is combined with the partial differential entity containing the locally coupled reaction source term, and a differential discretization transformation is performed on the spatial grid.

9. The method for dynamic prediction of distributed farmland soil nutrients for precision planting according to claim 1, characterized in that, When the movement rate of a high-concentration substance exceeds the environmental carrying capacity safety threshold, a blocking fertilization control instruction is generated, including the following steps: The central server aggregates the predicted values ​​of all nodes and performs three-dimensional spatial surface interpolation smoothing to render the continuous soil physical field. The nutrient migration front parameters are extracted by tracing the distribution density of contour lines, and the movement rate of the high-concentration substances is analyzed. When the movement rate of high-concentration substances exceeds the environmental carrying capacity safety threshold, a soil non-point source pollution early warning signal is triggered, and an interception-type fertilization control command is issued to the associated field management terminal.

10. A distributed farmland soil nutrient dynamic prediction system for precision planting, characterized in that, Includes the following modules: The topology network and initial state generation module constructs a physical adjacency topology graph based on multi-source farmland spatiotemporal monitoring data and a spatial grid partitioning mapping table, and integrates baseline variables and current soil nutrient concentration values ​​to obtain initial nutrient state values; The local coupling reaction source term generation module maps the sequence of external disturbance events into a standard impulse response signal sequence, extracts nonlinear feedback coefficients to modulate the standard impulse response signal sequence, and obtains the local coupling reaction source term. The partial differential equation construction module for undetermined boundary conditions constructs a convection-diffusion physics term that includes the current soil nutrient concentration value and the concentration of unknown neighbor nodes. The convection-diffusion physics term and the local coupled reaction source term are merged into a partial differential equation for undetermined boundary conditions, and the concentration of unknown neighbor nodes in adjacent grids is marked as a spatial diffusion-dependent boundary. The boundary condition generation module receives the adjacent nutrient concentration dataset based on the physical adjacency topology graph, fills the nutrient concentration feature values ​​of the adjacent nutrient concentration dataset into the memory address corresponding to the boundary, and generates the Dirichlet boundary condition. The distributed parallel solution and prediction module performs time integration on the partial differential equation with given boundary conditions based on the Dirichlet boundary conditions to obtain the nutrient prediction value at the next time step. The global field reconstruction and decision intervention module uploads nutrient prediction values ​​so that the central server can perform spatial interpolation to reconstruct the continuous soil physical field from the uploaded prediction values ​​and extract nutrient migration front parameters. When the movement rate of high-concentration substances exceeds the environmental carrying capacity safety threshold, an interception-type fertilization control command is generated.

Images