A pesticide application control method and system based on the coupling of three-dimensional pest distribution and aerodynamic field

Abstract

This invention relates to a pesticide application control method and system based on the coupling of three-dimensional pest distribution and aerodynamic field, belonging to the field of pest control technology. It includes collecting pest data within a grain pile through a sensor network deployed in a three-dimensional grid, generating a three-dimensional pest density distribution field based on the pest data, and combining this data with temporal information such as pressure, temperature, humidity, and pile height within the grain pile. This enables an airflow control scheme that utilizes the synergistic effects of first-order axial flow, second-order radial flow, and third-order tangential vortex to guide gaseous insecticides to act on the pest-affected area. This facilitates comprehensive, timely, and thorough detection and control of pests within the grain pile, thereby improving the pest control effect.

Description

Technical Field

[0001] This invention relates to the field of pest control technology, and in particular to a method and system for pesticide application control based on the coupling of three-dimensional pest distribution and aerodynamic field. Background Technology

[0002] During storage, grains are highly susceptible to damage from pests such as rice weevils, corn weevils, and grain borers. Pests not only cause weight loss and quality degradation (reduced germination rate and nutrient loss), but their excrement and carcasses also cause contamination and mold, seriously threatening national food security. Currently, pest detection and control mainly rely on manual sampling, but this method is inefficient and yields unsatisfactory results. Summary of the Invention

[0003] In view of the above-mentioned shortcomings and deficiencies of the prior art, the present invention provides a method and system for pesticide application control based on the coupling of three-dimensional distribution of pests and aerodynamic field, which solves the technical problem of unsatisfactory detection and control effects of pests in grain piles in the prior art.

[0004] To achieve the above objectives, the main technical solutions adopted by the present invention include:

[0005] The first aspect of this invention provides a pesticide application control method based on the coupling of three-dimensional pest distribution and aerodynamic field.

[0006] The pesticide application control method based on the coupling of three-dimensional pest distribution and aerodynamic field proposed in this embodiment of the invention includes:

[0007] Data on pests in grain piles are collected by a sensor network deployed in a three-dimensional grid, and a three-dimensional density distribution field of pests is generated based on the pest data in grain piles.

[0008] The temporal information of pressure, temperature, humidity and height inside the grain pile is input into a neural network to predict and generate a three-dimensional dynamic porosity field that reflects the pore structure of the grain pile.

[0009] The gap ratio field is converted into a permeability field, and the permeability field is input as a target physical parameter into a physical information neural network solver to obtain the three-dimensional airflow velocity field inside the grain pile.

[0010] The three-dimensional density distribution field of the pests is coupled with the three-dimensional airflow velocity field, and combined with the preset pest correction weight and critical flow velocity, a three-dimensional field for guiding the application of pesticides is determined.

[0011] Based on the numerical distribution of the three-dimensional field of control priority, an airflow control scheme consisting of the coordinated action of first-order axial flow, second-order radial flow, and third-order tangential vortex is determined to guide the gaseous insecticide to act on the pest-infested area.

[0012] In some instances, the step of inputting temporal information on pressure, temperature, humidity, and pile height inside the grain pile into a neural network to predict and generate a three-dimensional dynamic porosity field reflecting the pore structure of the grain pile includes:

[0013] The time-series information on pressure, temperature, humidity and height inside the grain pile is normalized to obtain normalized data.

[0014] The normalized data is input into the neural network to obtain the predicted gap rate of the key points.

[0015] Based on the predicted gap ratio values ​​of the key points, the three-dimensional dynamic gap ratio field is obtained through spatial interpolation reconstruction.

[0016] In some instances, converting the gap ratio field into a permeability field includes:

[0017] Determine the penetration conversion model;

[0018] Based on the gap ratio field and the permeability conversion model, the gap ratio field is converted into a permeability field; wherein, the permeability conversion model is:

[0019] Where ε is the interstitial ratio, K is the permeability, and d p The average diameter of the grain particles.

[0020] In some instances, the physical information neural network solver is constructed from a velocity field neural network and a pressure field neural network; wherein, during the construction of the physical information neural network solver, the velocity field neural network and the pressure field neural network are physically strongly coupled through Darcy's law to establish an initial correlation between permeability, the three-dimensional airflow velocity field and the pressure field;

[0021] Before inputting the permeability field as a target physical parameter into a physical information neural network solver to obtain the three-dimensional airflow velocity field inside the grain pile, the method includes:

[0022] Three-dimensional airflow velocity field sample data and pressure field sample data of the experimental grain silo were obtained, and the objective loss function of the optimized physical information neural network solver was established.

[0023] Based on the three-dimensional airflow velocity field sample data and pressure field sample data of the experimental grain silo, the physical information neural network solver is optimized with the goal of minimizing the target loss function, resulting in an optimized physical information neural network solver; wherein, the optimized physical information neural network solver includes optimized permeability, optimized correlation between the three-dimensional airflow velocity field and pressure field.

[0024] In some instances, the coupling of the three-dimensional pest density distribution field with the three-dimensional airflow velocity field, and combining preset pest correction weights and critical flow velocities, determines a three-dimensional field for guiding pesticide application control priorities, including:

[0025] Determine the coupling model between the three-dimensional density distribution field of the pests and the three-dimensional airflow velocity field;

[0026] Based on the coupling model of the three-dimensional density distribution field of the pests and the three-dimensional airflow velocity field, and combined with preset pest correction weights and critical flow velocities, a control priority three-dimensional field for guiding pesticide application is determined; wherein, the coupling model of the three-dimensional density distribution field of the pests and the three-dimensional airflow velocity field includes:

[0027] ;in, To control the priority three-dimensional field; For spatial location The actual pest density at the location; The critical flow velocity; This is the corrected three-dimensional airflow velocity field; To prevent the elimination of zero factors; This represents the maximum value in the pest density data set corresponding to the spatial coordinates.

[0028] In some instances, the modified three-dimensional airflow velocity field ,include:

[0029] ;in, ;

[0030] ;in, This represents the average value of the pest density data set corresponding to the spatial coordinates; The airflow velocity in the x-axis direction is the output of the three-dimensional airflow velocity field from the physical information neural network solver. The airflow velocity in the y-axis direction is the output of the three-dimensional airflow velocity field from the physical information neural network solver. The output of the physical information neural network solver is the airflow velocity in the z-axis direction of the three-dimensional airflow velocity field.

[0031] In some instances, generating a three-dimensional pest density distribution field based on the pest data within the grain pile includes:

[0032] Based on the pest data in the grain pile, a three-dimensional density distribution field of pests is generated using an inverse distance weighted interpolation algorithm.

[0033] In some instances, the determination of an airflow control scheme based on the numerical distribution of the three-dimensional field of the control priority, comprising the combined action of first-order axial flow, second-order radial flow, and third-order tangential vortex, to guide gaseous insecticides to act on the pest-infested area includes:

[0034] Based on the degree of insect infestation in the grain pile, several priority airflow control schemes are determined; each airflow control scheme consists of at least two synergistic effects from first-order axial flow, second-order radial flow, and third-order tangential vortex.

[0035] Based on the aforementioned multiple prioritized airflow control schemes, gaseous insecticides are guided to act on the pest-infested areas.

[0036] A second aspect of the present invention provides a precision drug application control system, comprising:

[0037] A multi-sensor fusion monitoring network is used to collect time-series information on pests in grain piles, pressure, temperature, humidity, and pile height inside the grain piles.

[0038] The central processing unit is used to generate a three-dimensional density distribution field of pests based on the pest data within the grain pile, and to predict and generate a three-dimensional dynamic porosity field reflecting the pore structure of the grain pile; and

[0039] The gap ratio field is converted into a permeability field, and the permeability field is input as a target physical parameter into a physical information neural network solver to obtain the three-dimensional airflow velocity field inside the grain pile; and

[0040] By coupling the three-dimensional pest density distribution field with the three-dimensional airflow velocity field, and combining it with preset pest correction weights and critical flow velocities, a three-dimensional control priority field is determined to guide pesticide application; and

[0041] Based on the numerical distribution of the three-dimensional field of control priority, an airflow control scheme consisting of the coordinated action of first-order axial flow, second-order radial flow, and third-order tangential vortex is determined.

[0042] The coordinating actuator is used to coordinating the generation of first-order axial flow, second-order radial flow, and third-order tangential vortex according to instructions from the central processing unit.

[0043] A third aspect of the present invention provides a computer-readable storage medium storing a pesticide application control program based on the coupling of three-dimensional distribution of pests with aerodynamic field. When the pesticide application control program based on the coupling of three-dimensional distribution of pests with aerodynamic field is executed by a processor, it implements the pesticide application control method based on the coupling of three-dimensional distribution of pests with aerodynamic field described in the first aspect.

[0044] This invention discloses a pesticide application control method based on the coupling of three-dimensional pest distribution and aerodynamic field, comprising: collecting pest data within a grain pile through a sensor network deployed in a three-dimensional grid, and generating a three-dimensional pest density distribution field based on the pest data; inputting the temporal information of pressure, temperature, humidity, and pile height inside the grain pile into a neural network to predict and generate a three-dimensional dynamic gap rate field reflecting the pore structure of the grain pile; converting the gap rate field into a permeability field, and inputting the permeability field as a target physical parameter into a physical information neural network solver to obtain a three-dimensional airflow velocity field inside the grain pile; coupling the three-dimensional pest density distribution field and the three-dimensional airflow velocity field, and combining preset pest correction weights and critical flow velocities to determine a control priority three-dimensional field for guiding pesticide application; and determining an airflow control scheme based on the numerical distribution of the control priority three-dimensional field, consisting of the coordinated action of first-order axial flow, second-order radial flow, and third-order tangential vortex, to guide the gaseous pesticide to act on the pest-affected area. In this application, a sensor network deployed in a three-dimensional grid is used to collect pest data in the grain pile. Based on the pest data, a three-dimensional density distribution field of pests is generated. Combined with the temporal information of pressure, temperature, humidity and pile height inside the grain pile, an airflow control scheme with the coordinated action of first-order axial flow, second-order radial flow and third-order tangential vortex is realized to guide gaseous insecticide to act on the pest area. This is conducive to the comprehensive, timely and sufficient detection and control of pests in the grain pile, thereby improving the pest control effect in the grain pile. Attached Figure Description

[0045] Figure 1 A flowchart of a pesticide application control method based on the coupling of three-dimensional pest distribution and aerodynamic field is provided in an embodiment of the present invention;

[0046] Figure 2 A schematic diagram of the spatial velocity field provided for an embodiment of the present invention;

[0047] Figure 3 This is a schematic diagram of the training of the physical information neural network solver model provided in an embodiment of the present invention;

[0048] Figure 4 This is a schematic diagram comparing the velocity field prediction accuracy provided in an embodiment of the present invention;

[0049] Figure 5 This is a velocity distribution diagram at different depths of a grain pile provided in an embodiment of the present invention;

[0050] Figure 6 A graph showing the relationship between model convergence and the number of sampling points provided in an embodiment of the present invention;

[0051] Figure 7 A diagram illustrating the degree of satisfaction of mass conservation constraints provided in embodiments of the present invention;

[0052] Figure 8A comparison chart of training efficiency for different network sizes provided in embodiments of the present invention;

[0053] Figure 9 This is a schematic diagram of the predicted flow field profile provided in an embodiment of the present invention.

[0054] Figure 10 This is a schematic diagram illustrating the loss function optimization of the physical information neural network solver provided in an embodiment of the present invention.

[0055] Figure 11 This is a schematic diagram of the refined drug application control system provided in an embodiment of the present invention. Detailed Implementation

[0056] To better explain and facilitate understanding of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0057] The pesticide application control method based on the coupling of three-dimensional pest distribution and aerodynamic field proposed in this invention is used to solve the technical problem of unsatisfactory pest detection and control effects in grain piles. It collects pest data within the grain pile through a sensor network deployed in a three-dimensional grid, and generates a three-dimensional pest density distribution field based on this data. Combined with temporal information such as pressure, temperature, humidity, and pile height within the grain pile, it achieves an airflow control scheme involving the synergistic action of first-order axial flow, second-order radial flow, and third-order tangential vortex. This guides the gaseous pesticide to act on the pest-affected area, thus facilitating comprehensive, timely, and sufficient detection and control of pests within the grain pile, thereby improving the pest control effect.

[0058] To better understand the above technical solutions, exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that the present invention can be understood more clearly and thoroughly, and that the scope of the present invention can be fully conveyed to those skilled in the art.

[0059] Figure 1 This is a flowchart illustrating a pesticide application control method based on the coupling of three-dimensional pest distribution and aerodynamic field, provided as an embodiment of the present invention. Figure 1 As shown in the embodiment of the present invention, the pesticide application control method based on the coupling of three-dimensional pest distribution and aerodynamic field includes:

[0060] Step 100: Collect pest data in the grain pile through a sensor network deployed in a three-dimensional grid, and generate a three-dimensional density distribution field of pests based on the pest data in the grain pile.

[0061] Step 110: Input the temporal information of pressure, temperature, humidity and height inside the grain pile into the neural network to predict and generate a three-dimensional dynamic porosity field that reflects the pore structure of the grain pile.

[0062] Step 120: Convert the gap ratio field into a permeability field, and input the permeability field as a target physical parameter into the physical information neural network solver to obtain the three-dimensional airflow velocity field inside the grain pile;

[0063] Step 130: Couple the three-dimensional density distribution field of the pests with the three-dimensional airflow velocity field, and combine the preset pest correction weights and critical flow velocities to determine the three-dimensional control priority field used to guide pesticide application;

[0064] Step 140: Based on the numerical distribution of the three-dimensional field of the control priority, determine the airflow control scheme of the coordinated action of first-order axial flow, second-order radial flow and third-order tangential vortex to guide the gaseous insecticide to act on the pest area.

[0065] In this exemplary embodiment, when collecting pest data in the grain pile, traps (such as probe traps) can be arranged in a three-dimensional grid inside the grain pile, with a spacing of 3 meters between each layer and a horizontal spacing of 2 meters in 8 directions, forming a three-dimensional monitoring array.

[0066] The sensor data is converted into pest density values ​​with coordinates, and inverse distance weighted interpolation is used to form 0.1-meter resolution three-dimensional coordinate data of pest density C(x,y,z) (x / y / z are the three-dimensional coordinates inside the grain warehouse).

[0067] In this exemplary embodiment, dynamic gap ratio prediction (performed by an LSTM-GAP network) generates a three-dimensional gap ratio field ε(x,y,z).

[0068] Input: Time series s(t) of sensor data (pressure, temperature, humidity, stack height);

[0069] Output: Gap rate field ε.

[0070] Among them, the pressure P data is obtained by using a differential pressure sensor of the ventilation system to collect physical sensing data;

[0071] The grain condition monitoring system uses standard probes to monitor vertical lines inside the grain pile, collecting temperature (T) and relative humidity (RH). The sensors can be deployed at the center of the R-wall edge, the inner R / 3 transition zone, and the outer R / 3 transition zone.

[0072] For height layer selection, choose 0.2H, 0.5H (center position), and avoid using the height edge layer (08H).

[0073] Among them, the height H of the grain pile is collected using a laser rangefinder on the top of the silo.

[0074] In this exemplary embodiment, the input and output time series data are processed, historical data from the sensors are collected, and data on changes in grain pile height, pressure, temperature, and relative humidity in a three-dimensional field matrix are compiled and summarized. Sample data can be collected at 5-minute intervals, with 12 sample data points at each location.

[0075] The value of vector X is calculated, and its four dimensions correspond to four key indicators of the physical state of the grain pile: pressure difference reflects the distribution of airflow resistance, temperature gradient indicates the location of pest activity, humidity affects the grain respiration intensity, and changes in pile height are directly related to the evolution of the gap ratio. These data, fused through an LSTM-GAP network, can decouple the three-dimensional gap field, which traditional sensors cannot directly measure.

[0076] In this application, a sensor network deployed in a three-dimensional grid collects pest data within the grain pile, and generates a three-dimensional pest density distribution field based on this data. The temporal information of pressure, temperature, humidity, and pile height within the grain pile is input into a neural network to predict and generate a three-dimensional dynamic gap rate field reflecting the pore structure of the grain pile. This gap rate field is converted into a permeability field, which is then input as a target physical parameter into a physical information neural network solver to obtain a three-dimensional airflow velocity field within the grain pile. The three-dimensional pest density distribution field and the three-dimensional airflow velocity field are coupled, and combined with preset pest correction weights and critical flow velocities, a three-dimensional control priority field is determined to guide pesticide application. Based on the numerical distribution of the control priority three-dimensional field, an airflow control scheme involving the coordinated action of first-order axial flow, second-order radial flow, and third-order tangential vortex is determined to guide the gaseous insecticide to act on the pest-affected area. In this application, a sensor network deployed in a three-dimensional grid is used to collect pest data in the grain pile. Based on the pest data, a three-dimensional density distribution field of pests is generated. Combined with the temporal information of pressure, temperature, humidity and pile height inside the grain pile, an airflow control scheme with the coordinated action of first-order axial flow, second-order radial flow and third-order tangential vortex is realized to guide gaseous insecticide to act on the pest area. This is conducive to the comprehensive, timely and sufficient detection and control of pests in the grain pile, thereby improving the pest control effect in the grain pile.

[0077] In some instances, the step of inputting temporal information on pressure, temperature, humidity, and pile height inside the grain pile into a neural network to predict and generate a three-dimensional dynamic porosity field reflecting the pore structure of the grain pile includes:

[0078] The time-series information on pressure, temperature, humidity and height inside the grain pile is normalized to obtain normalized data.

[0079] The normalized data is input into the neural network to obtain the predicted gap rate of the key points.

[0080] Based on the predicted gap ratio values ​​of the key points, the three-dimensional dynamic gap ratio field is obtained through spatial interpolation reconstruction.

[0081] In this exemplary embodiment, the neural network can be an LSTM-GAP (Long Short-Term Memory with Global Average Pooling) network model. LSTM-GAP transforms sparse temporal signals into feature vectors containing global physical states, and then reconstructs the entire field using an interpretable 3D decoder, performing spatial interpolation reconstruction. This solves the industry challenge of traditional methods being unable to generate continuous physical fields from limited sensors and perform root cause analysis. Specifically, the resolution of the 3D gap length is referenced to the 3D field planning of pest distribution, and the radial direction uses the Gaussian function: an interpolation method adapted to the cylindrical coordinate system of the grain silo. Linear interpolation is used for the height to maintain continuity. The gap ratio range (0.03-0.5) needs to conform to the physical characteristics of the grain pile to avoid data discrepancies. Kalman filtering is used to smooth out outliers. Finally, the latest 3D dynamic gap ratio field is output.

[0082] In some instances, converting the gap ratio field into a permeability field includes:

[0083] Determine the penetration conversion model;

[0084] Based on the gap ratio field and the permeability conversion model, the gap ratio field is converted into a permeability field; wherein, the permeability conversion model is:

[0085] Where ε is the porosity and K is the permeability.

[0086] d p The average diameter (m) of the grain particles is parameterized based on the grain stored in the grain depot.

[0087] In this exemplary embodiment, permeability K is a physical quantity that directly measures how easily a fluid passes through a porous medium. Regions with high porosity (more porous) have relatively high permeability (fluid flows through more easily); regions with low porosity (more dense) have drastically reduced permeability (fluid flows through more difficultly).

[0088] In some instances, the physical information neural network solver is constructed from a velocity field neural network and a pressure field neural network; wherein, during the construction of the physical information neural network solver, the velocity field neural network and the pressure field neural network are physically strongly coupled through Darcy's law to establish an initial correlation between permeability, the three-dimensional airflow velocity field and the pressure field;

[0089] Before inputting the permeability field as a target physical parameter into a physical information neural network solver to obtain the three-dimensional airflow velocity field inside the grain pile, the method includes:

[0090] Three-dimensional airflow velocity field sample data and pressure field sample data of the experimental grain silo were obtained, and the objective loss function of the optimized physical information neural network solver was established.

[0091] Based on the three-dimensional airflow velocity field sample data and pressure field sample data of the experimental grain silo, the physical information neural network solver is optimized with the goal of minimizing the target loss function, resulting in an optimized physical information neural network solver; wherein, the optimized physical information neural network solver includes optimized permeability, optimized correlation between the three-dimensional airflow velocity field and pressure field;

[0092] The step of inputting the permeability field as a target physical parameter into a physical information neural network solver to obtain the three-dimensional airflow velocity field inside the grain pile includes:

[0093] The permeability field is input as the target physical parameter into the optimized physical information neural network solver to obtain the three-dimensional airflow velocity field inside the grain pile.

[0094] In this exemplary embodiment, the experimental grain warehouse is one with relatively mild pest infestation or newly introduced grain; the three-dimensional airflow velocity field sample data and pressure field sample data are real spatial pressure and velocity data.

[0095] Darcy's Law is Where K is the penetration rate; It is a three-dimensional airflow velocity field; the The pressure gradient is the difference between the previous and current coordinates, reflecting the pressure differences and trends between different locations inside the grain pile. It is the core driving force for air to flow in the grain pile. The dynamic viscosity of air varies little in the range of 5-40℃, and the value is taken as... .

[0096] In this exemplary embodiment, both the velocity field neural network and the pressure field neural network can be MLP (Multi-Layer Perceptron) neural networks.

[0097] In this exemplary embodiment, the gap ratio ε(x,y,z) is a value in a certain space. The spatial velocity field is V(x,y,z). Figure 2 A schematic diagram of the spatial velocity field provided for an embodiment of the present invention. (See diagram below.) Figure 2 As shown, x, y, and z are the three three-dimensional spatial directions of the velocity field.

[0098] in, It is the rate of change of the pressure field in space, the difference between the previous coordinate and the current coordinate, reflecting the pressure differences and trends between different locations inside the grain pile, and is the core driving force for air to flow in the grain pile.

[0099] In this exemplary embodiment, a physical information neural network is first used to solve the physical field. The solution (three-directional vector data of the flow field) is approximated by the neural network, and the network is trained using the governing equations as constraints. The residual estimate is calculated, and the loss function includes continuity equation loss, data loss, and physical equation loss.

[0100] Data loss: Compare the network output with the measured value at a known sensor location.

[0101] Physical loss: Sampling points within the computational domain minimize the residuals of the control equations. By incorporating the residuals of the physical equations as part of the loss function, the network is made to satisfy the control equations after training. For residuals exceeding a loss threshold, the keypoint values ​​are recalculated using a denser mesh. The three vectors of the 3D airflow velocity field are converted into flow velocities in m / s (to simplify subsequent control processes, direction is temporarily disregarded; theoretically, the overall directional trend is from top to bottom).

[0102] In this exemplary embodiment, the neural network is internally formalized (multilayer perceptron MLP), and the neural network includes one input layer, L hidden layers, and one output layer.

[0103] The formulas for each layer are as follows:

[0104] (1) The input is a ternary variable (x, y, z), which can be directly represented as an input vector:

[0105] ;

[0106] Where a(0) is the activation value of the input layer, and the superscript (0) indicates "layer 0" (input layer).

[0107] (2) Hidden layer (layer l, 1≤l≤L)

[0108] Each hidden layer consists of two steps: "linear transformation" and "non-linear activation".

[0109] Linear transformation: The input activation value is multiplied by the weight matrix, and a bias term is added;

[0110] Nonlinear activation: Introducing nonlinearity through activation functions enhances fitting ability.

[0111] ;

[0112] The weight matrix of the l-th layer (n ln is the number of neurons in the l-th layer. l−1 (Number of neurons in layer l-1).

[0113] Let l be the bias vector of the l-th layer;

[0114] The result of the linear transformation of the l-th layer (pre-activation value);

[0115] Let l be the nonlinear activation function of the l-th layer;

[0116] , which is the output activation value of the l-th layer (used as the input of the next layer).

[0117] (3) Output layer (L+1th layer)

[0118] The output layer needs to match the type of the original function p(x,y,z) (the algorithm belongs to regression analysis and uses identity activation), and the final output is the prediction value of the neural network:

[0119] ;

[0120] p(x,y,z) is a scalar function, representing the number of neurons in the output layer. ,at this time ;

[0121] Parameter set That is, the weights and biases of all layers, by minimizing the loss function. Sure.

[0122] (4) Target loss function for:

[0123] ;

[0124] Loss due to Darcy's Law; Loss due to continuity equations; For data loss; among which,

[0125] Darcy's law loss characterizes the velocity v predicted by the neural network at coordinate point i. i It must be as close as possible to the theoretical speed calculated from Darcy's law. The deviation between the predicted flow velocity and the theoretical value of Darcy's law at the time sample point;

[0126] The continuity equation loss representation forces the neural network to output a velocity field v(x,y,z) with a divergence close to zero, thereby satisfying the vector direction pressure difference generated when mass is conserved;

[0127] Data loss characterizes the error between the predicted value and the actual measured value of the neural network at the measured point. It is used to "correct" the solution of pure physical constraints and ensure that the model prediction conforms to physical laws and is close to the actual observation data.

[0128] Where N is the number of sample points; Let i be the three-dimensional velocity predicted by the neural network at coordinate point i.

[0129] Let be the pressure gradient at coordinate point i. In the physical information neural network solver, this gradient is obtained by using automatic differentiation on the pressure pi output by the neural network.

[0130] Where N represents the internal sample point data; The airflow velocity in the x-axis direction is predicted by the physical information neural network solver at the i-th sampling point; The airflow velocity in the y-axis direction is predicted by the physical information neural network solver at the i-th sampling point; The airflow velocity in the z-axis direction is predicted by the physical information neural network solver at the i-th sampling point;

[0131] for The partial derivative with respect to x represents the rate of change of velocity along the x-axis in the x-direction; for The partial derivative with respect to y represents the rate of change of the velocity along the y-axis in the y-direction; for The partial derivative with respect to z represents the rate of change of the velocity along the z-axis in the z-direction.

[0132] ; This represents the three-dimensional velocity predicted by the neural network at the coordinates of the j-th sample point. M represents the three-dimensional velocity of the corresponding sample point, and M is the number of sample points.

[0133] In some instances, the coupling of the three-dimensional pest density distribution field with the three-dimensional airflow velocity field, and combining preset pest correction weights and critical flow velocities, determines a three-dimensional field for guiding pesticide application control priorities, including:

[0134] Determine the coupling model between the three-dimensional density distribution field of the pests and the three-dimensional airflow velocity field;

[0135] Based on the coupling model of the three-dimensional density distribution field of the pests and the three-dimensional airflow velocity field, and combined with preset pest correction weights and critical flow velocities, a control priority three-dimensional field for guiding pesticide application is determined; wherein, the coupling model of the three-dimensional density distribution field of the pests and the three-dimensional airflow velocity field includes:

[0136] ;in, To control the priority three-dimensional field; For spatial location The actual pest density at the location; The critical flow velocity; This is the corrected three-dimensional airflow velocity field; To prevent the elimination of zero factors; This represents the maximum value in the pest density data set corresponding to the spatial coordinates.

[0137] In this exemplary embodiment, For pest control weight, express .in, (Protection against zero factor), (Critical flow rate).

[0138] In this exemplary embodiment, the corrected three-dimensional airflow velocity field ,include:

[0139] ;in, ;

[0140] ;in, This represents the average value of the pest density data set corresponding to the spatial coordinates; The airflow velocity in the x-axis direction is the output of the three-dimensional airflow velocity field from the physical information neural network solver. The airflow velocity in the y-axis direction is the output of the three-dimensional airflow velocity field from the physical information neural network solver. The output of the physical information neural network solver is the airflow velocity in the z-axis direction of the three-dimensional airflow velocity field.

[0141] In some instances, generating a three-dimensional pest density distribution field based on the pest data within the grain pile includes:

[0142] Based on the pest data in the grain pile, a three-dimensional density distribution field of pests is generated using an inverse distance weighted interpolation algorithm.

[0143] In some instances, the determination of an airflow control scheme based on the numerical distribution of the three-dimensional field of the control priority, comprising the combined action of first-order axial flow, second-order radial flow, and third-order tangential vortex, to guide gaseous insecticides to act on the pest-infested area includes:

[0144] Based on the degree of insect infestation in the grain pile, several priority airflow control schemes are determined; each airflow control scheme consists of at least two synergistic effects from first-order axial flow, second-order radial flow, and third-order tangential vortex.

[0145] Based on the aforementioned multiple prioritized airflow control schemes, gaseous insecticides are guided to act on the pest-infested areas.

[0146] In this exemplary embodiment, the first-order airflow (axial flow) penetrates vertically downwards through the grain layer, dominating the overall penetration of the agent, with a wind speed range of 0.4-0.6 m / s;

[0147] Second-order airflow (radial flow): diffuses along the radial direction, promoting uniform lateral distribution, with a wind speed range of 0.3-0.4 m / s;

[0148] The third-order airflow (tangential vortex) is a vortex flow that rotates around the bin wall, enhancing local mixing, with a wind speed range of 0.2-0.3 m / s.

[0149] Among them, when deploying the centrifugal fan array on the top of the airflow equipment, a rotatable lifting truss (covering radius R=0.8D, where D is the diameter of the silo) is installed at the center of the top of the silo. The height of the swivel arm is adjustable from 2 to 8m. The swivel arm truss is made of carbon fiber, weighs less than 200kg, and has a drive tower, a servo motor with a power of 1.5kW, integrated slip ring power supply for anti-winding, and a variable frequency axial flow fan group at the end.

[0150] The central fan vertically downwards to achieve first-order airflow; the edge fans tilt inwards at 45 degrees to achieve second-order airflow; and the edge fans tilt outwards at 30 degrees, with the rotating arm swaying to achieve third-order airflow.

[0151] Among them, the priority three-dimensional field decision-making third-order airflow collaborative control scheme after half an hour includes:

[0152] Priority level 1, with the maximum value of the Pc matrix (0.8~0.1.0) or the maximum value of Pc (0.5~0.8) in the lower half of the grain pile height H of 0.5, indicating dense pest infestation and airflow dead zones. The control decision is a third-order coordinated strong intervention.

[0153] Priority level 2, Pc matrix maximum value (0.5~0.8) or 0.5 grain pile height H lower half layer has Pc maximum value (0.2~0.5), first order main control, second order auxiliary;

[0154] Priority level 3, maximum Pc matrix (0.2~0.5), mild pest infestation + normal airflow, second-order auxiliary operation is sufficient;

[0155] Priority level 4, maximum Pc matrix value (0~0.2), no pests, fan stops running, saving energy.

[0156] Figure 3 This is a schematic diagram illustrating the training of a physical information neural network solver model provided in an embodiment of the present invention. Figure 3 As shown, the overall training process of the physical information neural network solver includes:

[0157] Step 300: Neural network model construction, including establishing model space coordinates, establishing pressure field network (i.e., pressure field neural network) and velocity field network (i.e., velocity field neural network), constructing physical loss based on Darcy's law and continuity equation constraints, and constructing total loss function (i.e., target loss function) based on physical loss and data loss.

[0158] Step 310: Data acquisition for experimental chamber training, including acquiring three-dimensional airflow velocity field data through a flow velocity measurement device and acquiring pressure field data through a pressure sensor array;

[0159] Step 320: Training data processing, including cleaning the three-dimensional airflow velocity field data and pressure field data, then performing spatiotemporal alignment processing based on data fluid simulation, and then performing normalization standard processing to obtain the training dataset.

[0160] Step 330, Model Training, including training on sampled data based on the training dataset, forward propagation, loss calculation based on the total loss function, backpropagation and optimization, and model parameter updating;

[0161] Step 340, Model Output and Application, includes collecting business warehouse data, calculating the gap ratio field, calculating the airflow velocity field through the gap ratio spatial field, and analyzing the output of the airflow velocity field.

[0162] Figure 4 This is a schematic diagram comparing the velocity field prediction accuracy provided in an embodiment of the present invention. Figure 4 As shown in the scatter plot, the model's predicted speed and the experimental chamber's measured values ​​are visually compared in three directions (V). x V y V z Consistency on the curve, ideal fitting line (black dashed line): indicates that the predicted value is completely consistent with the actual value. The fact that each colored scatter point in the graph is close to the ideal curve indicates high prediction accuracy.

[0163] Figure 5 This is a velocity distribution diagram at different depths of a grain pile provided in an embodiment of the present invention. Figure 5 As shown, the distribution curves of velocity magnitude along the radial position are displayed on the cross-sections at different heights (depths z=0.2H, 0.5H, 0.8H) of the shallow circular silo. The smooth shape of the curves conforms to the physical expectations of fluid flow in a granular porous medium, without any non-physical oscillations.

[0164] Figure 6This is a graph showing the relationship between model convergence and the number of sampling points provided in an embodiment of the present invention. Figure 6 As shown, the number of spatial sampling points used to calculate physical loss affects the final accuracy of the model. As the number of sampling points (x-axis) increases, the final loss value (y-axis) decreases. The curve shows a downward trend on logarithmic scales, indicating that increasing the number of physical constraint sampling points improves model accuracy.

[0165] Figure 7 This is a diagram illustrating the degree of satisfaction of the mass conservation constraints provided in an embodiment of the present invention. Figure 7 As shown, the satisfaction of the continuity equations for the physical constraints of the model is verified. Figure 7 Each point represents a calculated velocity divergence value at a location in space. Ideally, all points should be concentrated near zero. Training results show that most points are concentrated near zero (0-0.02), indicating that the velocity field predicted by the model satisfies mass conservation well across the entire computational domain.

[0166] Figure 8 This is a comparison chart of training efficiency for different network sizes provided in embodiments of the present invention. Figure 8 As shown, the training time and final accuracy of neural network architectures with different complexities (small / medium / large) are compared. The training results show that medium-sized networks achieve the best balance between accuracy and efficiency, making them a cost-effective choice.

[0167] Figure 9 This is a schematic diagram of the predicted flow field profile provided in an embodiment of the present invention. Figure 9 As shown, the final output of the model training is visualized. A streamline diagram illustrates the predicted airflow pattern on the longitudinal section (or a section at a certain depth) of the shallow circular chamber. Arrows indicate the direction and path of the airflow. The density and curvature of the streamlines visually reflect the airflow organization, vortex zones, and dead zones within the chamber. The cross-sectional view accurately reflects the gas flow in physical space.

[0168] Figure 10 A trend analysis chart of model training loss provided in an embodiment of the present invention. For example... Figure 10 As shown, the horizontal axis represents the number of training epochs (0-2000), signifying the complete training process from start to finish; the vertical axis represents the loss value (Loss), using a logarithmic scale for easy observation of changes in magnitude. The Total Loss plot shows the trend of the total loss. The trend is from approximately 10... 1 Rapidly dropped to 10 -2 After approximately 500 rounds, the performance stabilized. The training status was assessed as follows: training successfully converged, and the total loss decreased by three orders of magnitude, indicating that the model parameters had found a relatively optimal solution.

[0169] The Darcy Loss chart shows the trend of the Darcy loss. The trend is a rapid initial decrease, followed by slight fluctuations, and finally stabilization. Training state analysis shows a slight rebound between 100-200 epochs, reflecting the model's dynamic adjustment process as it learns the Darcy law constraint, eventually stabilizing at 10. -2 The magnitude indicates that the velocity field and pressure gradient predicted by the neural network satisfy Darcy's law well.

[0170] The Continuity Loss chart shows the trend of continuous loss. The trend is initially upward and then stabilizes, starting from 10. -3 Up to 10 -2 And remain at a high level, maintaining a level of 10. -2 The initial training focused more on Darcy's law and data fitting, as well as the competition and trade-offs among different loss terms. Although the results were high, they were still acceptable, and the weights could be adjusted for further optimization.

[0171] The data loss chart shows the trend of data loss. The trend is a steady but rapid decline, starting from 10... 0 Reduced to 10 -2 .

[0172] The training curve shows the smoothest convergence curve, indicating that the model can effectively fit the measured data.

[0173] The practical significance is that the error between the prediction speed and the measured data is continuously reduced, and the practicality of the model is enhanced.

[0174] Training results show that the model can simultaneously learn physical laws and fit measured data, achieve a reasonable balance among various losses, avoid a single loss dominating the process, and achieve stable convergence without violent oscillations.

[0175] This invention provides a precision pesticide application control system. Figure 11 This is a schematic diagram of the refined pesticide application control system provided in an embodiment of the present invention. Figure 11 As shown, the precision pesticide application control system includes:

[0176] The multi-sensor fusion monitoring network 30 is used to collect time-series information on pest data, pressure, temperature, humidity and height inside the grain pile.

[0177] Central processing unit 31 is used to generate a three-dimensional density distribution field of pests based on the pest data in the grain pile, and to predict and generate a three-dimensional dynamic porosity field reflecting the pore structure of the grain pile; and

[0178] The gap ratio field is converted into a permeability field, and the permeability field is input as a target physical parameter into a physical information neural network solver to obtain the three-dimensional airflow velocity field inside the grain pile; and

[0179] By coupling the three-dimensional pest density distribution field with the three-dimensional airflow velocity field, and combining it with preset pest correction weights and critical flow velocities, a three-dimensional control priority field is determined to guide pesticide application; and

[0180] Based on the numerical distribution of the three-dimensional field of control priority, an airflow control scheme consisting of the coordinated action of first-order axial flow, second-order radial flow, and third-order tangential vortex is determined.

[0181] The coordinating actuator 32 is used to coordinating the generation of first-order axial flow, second-order radial flow and third-order tangential vortex according to the instructions of the central processing unit.

[0182] In this exemplary embodiment, when collecting pest data in the grain pile, traps (such as probe traps) can be arranged in a three-dimensional grid inside the grain pile, with a spacing of 3 meters between each layer and a horizontal spacing of 2 meters in 8 directions, forming a three-dimensional monitoring array.

[0183] The sensor data is converted into pest density values ​​with coordinates, and inverse distance weighted interpolation is used to form 0.1-meter resolution three-dimensional coordinate data of pest density C(x,y,z) (x / y / z are the three-dimensional coordinates inside the grain warehouse).

[0184] In this exemplary embodiment, dynamic gap ratio prediction (performed by an LSTM-GAP network) generates a three-dimensional gap ratio field ε(x,y,z).

[0185] Input: Time series s(t) of sensor data (pressure, temperature, humidity, stack height);

[0186] Output: Gap rate field ε.

[0187] Among them, the pressure P data is obtained by using a differential pressure sensor of the ventilation system to collect physical sensing data;

[0188] The grain condition monitoring system uses standard probes to monitor vertical lines inside the grain pile, collecting temperature (T) and relative humidity (RH). The sensors can be deployed at the center of the R-wall edge, the inner R / 3 transition zone, and the outer R / 3 transition zone.

[0189] For height layer selection, choose 0.2H, 0.5H (center position), and avoid using the height edge layer (08H).

[0190] The height H of the grain pile was collected using a laser rangefinder on the top of the silo.

[0191] In this exemplary embodiment, the input and output time series data are processed, historical data from the sensors are collected, and data on changes in grain pile height, pressure, temperature, and relative humidity in a three-dimensional field matrix are compiled and summarized. Sample data can be collected at 5-minute intervals, with 12 sample data points at each location.

[0192] The value of vector X is calculated, and its four dimensions correspond to four key indicators of the physical state of the grain pile: pressure difference reflects the distribution of airflow resistance, temperature gradient indicates the location of pest activity, humidity affects the grain respiration intensity, and changes in pile height are directly related to the evolution of the gap ratio. These data, fused through an LSTM-GAP network, can decouple the three-dimensional gap field, which traditional sensors cannot directly measure.

[0193] In this application, a sensor network deployed in a three-dimensional grid collects pest data within the grain pile, and generates a three-dimensional pest density distribution field based on this data. The temporal information of pressure, temperature, humidity, and pile height within the grain pile is input into a neural network to predict and generate a three-dimensional dynamic gap rate field reflecting the pore structure of the grain pile. This gap rate field is converted into a permeability field, which is then input as a target physical parameter into a physical information neural network solver to obtain a three-dimensional airflow velocity field within the grain pile. The three-dimensional pest density distribution field and the three-dimensional airflow velocity field are coupled, and combined with preset pest correction weights and critical flow velocities, a three-dimensional control priority field is determined to guide pesticide application. Based on the numerical distribution of the control priority three-dimensional field, an airflow control scheme involving the coordinated action of first-order axial flow, second-order radial flow, and third-order tangential vortex is determined to guide the gaseous insecticide to act on the pest-affected area. In this application, a sensor network deployed in a three-dimensional grid is used to collect pest data in the grain pile. Based on the pest data, a three-dimensional density distribution field of pests is generated. Combined with the temporal information of pressure, temperature, humidity and pile height inside the grain pile, an airflow control scheme with the coordinated action of first-order axial flow, second-order radial flow and third-order tangential vortex is realized to guide gaseous insecticide to act on the pest area. This is conducive to the comprehensive, timely and sufficient detection and control of pests in the grain pile, thereby improving the pest control effect in the grain pile.

[0194] Since the systems / devices described in the above embodiments of the present invention are systems / devices used to implement the methods of the above embodiments of the present invention, those skilled in the art can understand the specific structure and modifications of the systems / devices based on the methods described in the above embodiments of the present invention, and therefore will not be repeated here. All systems / devices used in the methods of the above embodiments of the present invention fall within the scope of protection of the present invention.

[0195] This invention provides a computer-readable storage medium storing a pesticide application control program based on the coupling of three-dimensional distribution of pests with aerodynamic field. When the pesticide application control program based on the coupling of three-dimensional distribution of pests with aerodynamic field is executed by a processor, it implements the pesticide application control method based on the coupling of three-dimensional distribution of pests with aerodynamic field described in the above embodiments.

[0196] In the description of this invention, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0197] In this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0198] In this invention, unless otherwise explicitly specified and limited, "above" or "below" the second feature can mean that the first and second features are in direct contact, or that they are in indirect contact through an intermediate medium. Furthermore, "above," "over," or "on top" the second feature can mean that the first feature is directly above or diagonally above the second feature, or simply indicates that the first feature is at a higher horizontal level than the second feature. "Below," "below," or "beneath" the second feature can mean that the first feature is directly below or diagonally below the second feature, or simply indicates that the first feature is at a lower horizontal level than the second feature.

[0199] In the description of this specification, the terms "one embodiment," "some embodiments," "embodiment," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0200] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make modifications, alterations, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A pesticide application control method based on the coupling of three-dimensional pest distribution and aerodynamic field, characterized in that, include: Data on pests in grain piles are collected by a sensor network deployed in a three-dimensional grid, and a three-dimensional density distribution field of pests is generated based on the pest data in grain piles. The temporal information of pressure, temperature, humidity and height inside the grain pile is input into a neural network to predict and generate a three-dimensional dynamic porosity field that reflects the pore structure of the grain pile. The gap ratio field is converted into a permeability field, and the permeability field is input as a target physical parameter into a physical information neural network solver to obtain the three-dimensional airflow velocity field inside the grain pile. The three-dimensional density distribution field of the pests is coupled with the three-dimensional airflow velocity field, and combined with the preset pest correction weight and critical flow velocity, a three-dimensional field for guiding the application of pesticides is determined. Based on the numerical distribution of the three-dimensional field of control priority, an airflow control scheme consisting of the coordinated action of first-order axial flow, second-order radial flow, and third-order tangential vortex is determined to guide the gaseous insecticide to act on the pest-infested area.

2. The pesticide application control method based on the coupling of three-dimensional pest distribution and aerodynamic field according to claim 1, characterized in that, The process of inputting temporal information on pressure, temperature, humidity, and pile height inside the grain pile into a neural network to predict and generate a three-dimensional dynamic porosity field reflecting the pore structure of the grain pile includes: The time-series information on pressure, temperature, humidity and height inside the grain pile is normalized to obtain normalized data. The normalized data is input into the neural network to obtain the predicted gap rate of the key points. Based on the predicted gap ratio values ​​of the key points, the three-dimensional dynamic gap ratio field is obtained through spatial interpolation reconstruction.

3. The pesticide application control method based on the coupling of three-dimensional pest distribution and aerodynamic field according to claim 1, characterized in that, The process of converting the gap ratio field into a permeability field includes: Determine the penetration conversion model; Based on the gap ratio field and the permeability conversion model, the gap ratio field is converted into a permeability field; wherein, the permeability conversion model is: Where ε is the interstitial ratio, K is the permeability, and d p The average diameter of the grain grain.

4. The pesticide application control method based on the coupling of three-dimensional pest distribution and aerodynamic field according to claim 1, characterized in that, The physical information neural network solver is constructed from a velocity field neural network and a pressure field neural network. In the process of constructing the physical information neural network solver, the velocity field neural network and the pressure field neural network are physically strongly coupled through Darcy's law to establish the initial correlation between permeability, three-dimensional airflow velocity field and pressure field. Before inputting the permeability field as a target physical parameter into a physical information neural network solver to obtain the three-dimensional airflow velocity field inside the grain pile, the method includes: Three-dimensional airflow velocity field sample data and pressure field sample data of the experimental grain silo were obtained, and the objective loss function of the optimized physical information neural network solver was established. Based on the three-dimensional airflow velocity field sample data and pressure field sample data of the experimental grain silo, the physical information neural network solver is optimized with the goal of minimizing the target loss function, resulting in an optimized physical information neural network solver; wherein, the optimized physical information neural network solver includes optimized permeability, optimized correlation between the three-dimensional airflow velocity field and pressure field.

5. The pesticide application control method based on the coupling of three-dimensional pest distribution and aerodynamic field according to claim 1, characterized in that, The coupling of the pest three-dimensional density distribution field and the three-dimensional airflow velocity field, combined with preset pest correction weights and critical flow velocities, determines a three-dimensional field for guiding pesticide application control priorities, including: Determine the coupling model between the three-dimensional density distribution field of the pests and the three-dimensional airflow velocity field; Based on the coupling model of the three-dimensional density distribution field of the pests and the three-dimensional airflow velocity field, and combined with preset pest correction weights and critical flow velocities, a control priority three-dimensional field for guiding pesticide application is determined; wherein, the coupling model of the three-dimensional density distribution field of the pests and the three-dimensional airflow velocity field includes: ;in, To control the priority three-dimensional field; Spatial location The actual pest density at the location; The critical flow velocity; This is the corrected three-dimensional airflow velocity field; To prevent the elimination of zero factors; This represents the maximum value in the pest density data set corresponding to the spatial coordinates.

6. The pesticide application control method based on the coupling of three-dimensional pest distribution and aerodynamic field according to claim 5, characterized in that, The corrected three-dimensional airflow velocity field ,include: ;in, ; ;in, This represents the average value of the pest density data set corresponding to the spatial coordinates; The airflow velocity in the x-axis direction is the output of the three-dimensional airflow velocity field from the physical information neural network solver. The airflow velocity in the y-axis direction is the output of the three-dimensional airflow velocity field from the physical information neural network solver. The output of the physical information neural network solver is the airflow velocity in the z-axis direction of the three-dimensional airflow velocity field.

7. The pesticide application control method based on the coupling of three-dimensional pest distribution and aerodynamic field according to claim 5, wherein generating a three-dimensional pest density distribution field based on the pest data in the grain pile includes: Based on the pest data in the grain pile, a three-dimensional density distribution field of pests is generated using an inverse distance weighted interpolation algorithm.

8. The pesticide application control method based on the coupling of three-dimensional pest distribution and aerodynamic field according to claim 1, wherein determining an airflow control scheme based on the numerical distribution of the control priority three-dimensional field, consisting of the coordinated action of first-order axial flow, second-order radial flow, and third-order tangential vortex, to guide the gaseous insecticide to act on the pest-affected area, includes: Based on the degree of insect infestation in the grain pile, several priority airflow control schemes are determined; each airflow control scheme consists of at least two synergistic effects from first-order axial flow, second-order radial flow, and third-order tangential vortex. Based on the aforementioned multiple prioritized airflow control schemes, gaseous insecticides are guided to act on the pest-infested areas.

9. A precision pesticide application control system, characterized in that, include: A multi-sensor fusion monitoring network is used to collect time-series information on pests in grain piles, pressure, temperature, humidity, and pile height inside the grain piles. The central processing unit is used to generate a three-dimensional density distribution field of pests based on the pest data within the grain pile, and to predict and generate a three-dimensional dynamic porosity field reflecting the pore structure of the grain pile; and The gap ratio field is converted into a permeability field, and the permeability field is input as a target physical parameter into a physical information neural network solver to obtain the three-dimensional airflow velocity field inside the grain pile; and By coupling the three-dimensional pest density distribution field with the three-dimensional airflow velocity field, and combining it with preset pest correction weights and critical flow velocities, a three-dimensional control priority field is determined to guide pesticide application; and Based on the numerical distribution of the three-dimensional field of control priority, an airflow control scheme consisting of the coordinated action of first-order axial flow, second-order radial flow, and third-order tangential vortex is determined. The coordinating actuator is used to coordinating the generation of first-order axial flow, second-order radial flow, and third-order tangential vortex according to instructions from the central processing unit.

10. A computer-readable storage medium, characterized in that, It stores a pesticide application control program based on the coupling of three-dimensional distribution of pests and aerodynamic field. When the processor executes the pesticide application control program based on the coupling of three-dimensional distribution of pests and aerodynamic field, it implements the pesticide application control method based on the coupling of three-dimensional distribution of pests and aerodynamic field as described in any one of claims 1-8.

Images