Aerial seeding seed microenvironment prediction system and method based on sand movement simulation

Abstract

The invention relates to an aerial seeding seed microenvironment prediction system based on sand grain motion simulation and a method thereof, belongs to the field of unmanned aerial vehicle aerial seeding technology and environment simulation prediction, and provides a method for acquiring environment data through a multi-source heterogeneous micro-scale sensing network, predicting flow field data by using a sand grain-airflow coupling simulation engine driven by a graph neural network, and predicting the flow field data by using a multi-source heterogeneous micro-scale sensing network. And calculating a seed motion track and an anti-displacement coefficient through a seed-sand multi-physics field coupling behavior prediction model, and generating a displacement risk probability distribution diagram. And the self-adaptive aerial seeding parameter intelligent optimization system optimizes the hovering height and the ejection parameters according to the parameters, and precise ejection is realized through a variable-rigidity ejection execution mechanism. According to the system, the anti-displacement coefficient of the seeds is remarkably improved to 2.8 times, the problem that the seeds are prone to displacement in the sand wind environment is effectively solved, the precision and success rate of aerial seeding of the unmanned aerial vehicle are improved, and efficient technical support is provided for desertification control.

Description

Technical Field

[0001] The present invention relates to a prediction system and method for the microenvironment of aerial seeding seeds based on sand grain movement simulation, belonging to the fields of unmanned aerial vehicle (UAV) aerial seeding technology and environmental simulation prediction. Background Art

[0002] Aerial seeding is an important means to restore vegetation and prevent desertification. Traditional aerial seeding technologies such as CN 118270236A disclose a dual-mode intelligent aerial seeding system for precision aerial seeding operations, including an intelligent UAV, a multi-agent control platform, a seeding end with seedlings, a cold launch system, a digging robot, and their collaborative operation mechanisms. This system improves the overall efficiency of aerial seeding operations through high-precision navigation and positioning, dual-mode operation methods, and intelligent control.

[0003] However, the existing technologies have the following obvious deficiencies: First, they lack the ability to predict the microenvironment, and cannot accurately predict the microenvironment conditions after the seeds land, especially in sandy environments where the seeds are easily blown away and displaced; second, the simulation of the seed landing trajectory is insufficient, lacking the real-time simulation ability of the movement trajectory of the seeds under the action of wind; third, the hovering height is fixed and cannot be adaptively adjusted to the optimal seeding height according to the real-time wind field environment; fourth, the seed stability is poor, and the traditional system mainly focuses on the delivery accuracy and ignores the stability guarantee after the seeds land; fifth, there is a problem of calculation delay, lacking the ability to respond quickly in complex wind field environments.

[0004] The above deficiencies lead to low efficiency of aerial seeding in sandy environments and low seed survival rate, especially in arid and semi-arid regions, where aerial seeding for afforestation faces huge challenges. Summary of the Invention

[0005] The purpose of the present invention is to provide a prediction system and method for the microenvironment of aerial seeding seeds based on sand grain movement simulation, and through multi-scale microenvironment perception and intelligent prediction, solve the technical problem of easy displacement of seeds in sandy environments, and achieve accurate seed delivery and stable landing.

[0006] The present invention proposes a prediction system for the microenvironment of aerial seeding seeds based on sand grain movement simulation, including:

[0007] A data acquisition module, used to collect three-dimensional point cloud data through multi-sensor fusion;

[0008] A preprocessing module, used to perform denoising and downsampling processing on the three-dimensional point cloud data;

[0009] A feature extraction module, used to extract features from the preprocessed point cloud data and generate semantic labels;

[0010] A reference plane acquisition module, used to calculate the probability optimal fitting plane based on the semantic labels;

[0011] The pixel feature conversion module is used to map the point cloud features into the pixel feature space to achieve multi-scale feature output.

[0012] Preferably, the data acquisition module includes a depth camera and a laser radar, wherein the depth camera adopts a binocular camera, includes a data acquisition unit, an image stitching unit, an image correction unit and a data storage unit, and the depth camera and the laser radar are installed on the drone for scanning and imaging the target working area.

[0013] Preferably, the preprocessing module comprises:

[0014] The geometric denoising submodule is used to remove flying spots generated during data acquisition and uses a straight-through filtering algorithm and a voxel grid filtering algorithm to downsample the data;

[0015] The outlier removal submodule is used to filter outliers outside the point cloud distribution area using a statistical filtering algorithm to reduce the number of subsequent calculations.

[0016] Preferably, the feature extraction module is implemented using a PointNet++ convolutional computing network, including point cloud feature extraction, feature upsampling, feature fusion and semantic label generation processes, wherein a local feature enhancement mechanism and a context-aware module are used to capture the hierarchical structure information of the point cloud.

[0017] Preferably, the feature extraction module further includes a generative adversarial network for receiving the semantic label vector and generating a lithology label, wherein the generative adversarial network includes:

[0018] A generator, used to convert semantic label vectors into lithology distribution features;

[0019] The discriminator is used to distinguish the generated labels from the real labels and continuously optimize the generated features through error feedback.

[0020] Preferably, the reference plane acquisition module is used for:

[0021] Calculate the ratio of the number of point clouds of each semantic category to the total number of point clouds;

[0022] When the ratio is not greater than a preset threshold, feature extraction is performed on each semantic category, including the centroid, normal vector, local surface normal vector, covariance matrix, point pair, principal component and distance from point to fitting surface of the point cloud;

[0023] The hybrid RANSAC algorithm is used to fit semantic categories and features, and multiple groups of different semantic planes and the confidence probability of each semantic plane are obtained;

[0024] A final reference surface is determined based on the confidence probability.

[0025] Preferably, the pixel feature conversion module is implemented using the Unet architecture, which includes an encoding part and a decoding part. The encoding part consists of convolution, pooling, and downsampling, and the decoding part is implemented using transposed convolution. A skip connection is used in the middle of the Unet architecture to fuse the high-level semantics of each scale in the encoding process with the detailed features of the lower layer in the decoding process.

[0026] Preferably, the adversarial generative network learns based on physical constraints, including:

[0027] Constructing adversarial learning between the semantic model and the physical model;

[0028] Designing a new semantic segmentation model label based on terrain label information and color map information;

[0029] Optimizing the semantic model through the cross-entropy loss function to enable it to learn physical characteristics;

[0030] Adopting a progressive constraint strategy, with a focus on feature learning in the initial stage of training and gradually increasing the physical constraint weight in the later stage.

[0031] Preferably, the formula for the reference plane acquisition module to calculate the plane confidence is:

[0032] Confidence = semantic consistency × geometric fitness × physical rationality,

[0033] where semantic consistency represents the degree of consistency of semantic labels of points within the plane, geometric fitness represents the residual error of plane fitting, and physical rationality represents the rationality of evaluating the plane based on physical rules.

[0034] The method based on the intelligent fitting system for 3D point cloud semantic segmentation and measurement reference plane includes:

[0035] The step of collecting 3D point cloud data, using a depth camera and a lidar to scan and image the target operation area to obtain 3D point cloud data;

[0036] The preprocessing step, denoising and downsampling the 3D point cloud data;

[0037] The feature extraction step, using the PointNet++ network to extract features from the preprocessed point cloud data and generate a semantic label vector;

[0038] The lithology label generation step, inputting the semantic label vector into the adversarial generative network to generate a lithology label;

[0039] The reference plane acquisition step, calculating the probability optimal fitting plane based on the lithology label, determining the confidence of the plane, and if the confidence is higher than the preset threshold, determining the plane as the measurement reference plane;

[0040] Online inference step: Input the three-dimensional point cloud data into the trained model for processing to obtain the measurement reference plane within the target area.

[0041] The beneficial effects of the present invention are as follows:

[0042] 1. Significantly improve the seed anti-displacement coefficient by 2.8 times, effectively solving the core problem of easy displacement of seeds in sandy and windy environments;

[0043] 2. The accurate seed landing rate is increased to 92%, an increase of 15% compared with traditional technologies;

[0044] 3. The seed survival rate is increased to 85%, an increase of 25% compared with traditional technologies;

[0045] 4. The wind speed range suitable for the system is expanded to 0 - 12 m / s, significantly extending the window period for aerial seeding operations;

[0046] 5. The seeding efficiency is increased by 30%, and the daily operation area of a single unmanned aerial vehicle is increased to 150 mu. Description of the Drawings

[0047] Figure 1 It is the overall architecture diagram of the aerial seeding seed microenvironment prediction system based on sand grain motion simulation of the present invention;

[0048] Figure 2 It is the structural schematic diagram of the multi-source heterogeneous micro-scale perception network of the present invention;

[0049] Figure 3 It is the workflow diagram of the sand grain-airflow coupling simulation engine driven by the graph neural network of the present invention;

[0050] Figure 4 It is the schematic diagram of the seed-sandy land multi-physical field coupling behavior prediction model of the present invention;

[0051] Figure 5 It is the structural diagram of the adaptive aerial seeding parameter intelligent optimization system of the present invention;

[0052] Figure 6 It is the structural diagram of the variable stiffness ejection actuator of the present invention;

[0053] Figure 7 It is the schematic diagram of the physics-informed graph neural network in the system of the present invention;

[0054] Figure 8 It is the visualization effect diagram of the sand grain motion field prediction in the system of the present invention;

[0055] Figure 9 It is the schematic diagram of the influencing factors of the seed anti-displacement coefficient in the system of the present invention;

[0056] Figure 10 It is the flowchart of the method of the present invention. Detailed implementation manners

[0057] Please refer to the attached Figure 1-10 , and the following further elaborates on the detailed implementation manners of the present invention in conjunction with the accompanying drawings.

[0058] Embodiment 1

[0059] As Figure 1 shown, the sand-seed microenvironment prediction system based on sand particle motion simulation of the present invention includes a multi-source heterogeneous micro-scale perception network 1, a sand-airflow coupling simulation engine 2 driven by a graph neural network, a seed-sand multi-physical field coupling behavior prediction model 3, an adaptive intelligent optimization system 4 for aerial seeding parameters, and a variable stiffness ejection actuator 5. These five modules form a closed-loop system of perception - prediction - optimization - execution, and work together to achieve precise seed placement and stable landing.

[0060] The multi-source heterogeneous micro-scale perception network 1 is responsible for collecting wind field data, sand particle motion data, and surface characteristic data, and generating a standard environmental state matrix E(t). The sand-airflow coupling simulation engine 2 driven by a graph neural network receives the standard environmental state matrix E(t), and calculates and generates flow field prediction data F(t+Δt) through a physics-informed graph neural network. The seed-sand multi-physical field coupling behavior prediction model 3 calculates the seed motion trajectory T(s) and the seed displacement resistance coefficient k(s,E) based on the flow field prediction data F(t+Δt), and generates a displacement risk probability distribution map P(x,y). The adaptive intelligent optimization system 4 for aerial seeding parameters calculates the optimal hovering height h* and ejection parameters (v,θ) based on the displacement risk probability distribution map P(x,y), the seed motion trajectory T(s), and the seed displacement resistance coefficient k(s,E), and generates a control instruction C(t). The variable stiffness ejection actuator 5 receives the control instruction C(t), adjusts the ejection stiffness k* and the seed wrapping parameters, executes seed ejection, and collects actual trajectory data Ta(s) as feedback data and sends it to the adaptive intelligent optimization system 4 for aerial seeding parameters.

[0061] The data transfer between the system modules uses a high-speed transmission of 100 Mbps, and the delay is controlled within 10 ms to ensure real-time performance. The working cycle of the entire system is controlled within 500 ms to achieve fast response and decision-making.

[0062] Embodiment 2

[0063] As Figure 2 shown, the multi-source heterogeneous micro-scale perception network 1 includes a micro PIV particle image velocimetry system 11, a micro six-degree-of-freedom wind field sensor array 12, a surface characteristic rapid detector 13, and a data fusion processing unit 14.

[0064] The micro PIV particle image velocimetry system 11 consists of a 532 nm pulsed laser sheet light source 111 and a 4000 fps high-speed camera 112, with a spatial resolution of 0.5 mm, and is used to capture the movement trajectories and velocity fields of sand grains in the scale range of 10 - 500 μm. The micro PIV system adopts a double-pulse synchronization technology, and obtains the movement vectors of sand grains through the correlation analysis of consecutive frame images. The system sets the light sheet thickness to 1 mm, the pulse interval time to 50 μs, and the preferred laser power to 300 mW, which is sufficient to illuminate the tiny sand grains without affecting the measurement.

[0065] The micro six-degree-of-freedom wind field sensor array 12 consists of 8 micro hot-wire anemometers 121, which are distributed in a cubic array, with a sampling frequency of 200 Hz and a measurement accuracy of ±0.05 m / s, and is used to measure the three-dimensional wind field distribution in a 3 m 3 space. The hot-wire anemometer adopts a constant-temperature mode, and the working temperature differs from the ambient temperature by 50 °C to improve the sensitivity. The sensor array calculates the wind speed vector at any point in space through the Burman interpolation method, and the spatial resolution is 10 cm.

[0066] The surface characteristics rapid detector 13 includes a micro laser roughness meter 131 and a capacitive moisture content sensor 132, and is used to obtain the roughness, density and moisture content parameters of the surface sand layer. The laser roughness meter adopts the triangulation principle, with a measurement range of 0 - 10 mm and an accuracy of 0.01 mm. The moisture content sensor has a working frequency of 100 MHz, a measurement range of 0% - 20%, and an accuracy of 0.5%.

[0067] The data fusion processing unit 14 uses the Kalman filtering algorithm to perform spatio-temporal alignment and fusion on the multi-sensor data, and generates a standard environmental state matrix E(t). This matrix contains the wind field vector, the sand grain movement vector and the surface characteristics parameters, with a dimension of m×n, where m represents the number of spatial discrete points and n represents the characteristic dimension. The Kalman filtering algorithm can be expressed as:

[0068]

[0069] where: is the state estimation value at time k, F k is the state transition matrix, B k is the control matrix, u k is the control vector, K k is the Kalman gain, z k is the observation vector, H k is the observation matrix. The Kalman filtering algorithm realizes the optimal estimation of the environmental state through two steps of prediction and update. The filtering update frequency is 100 Hz, ensuring the real-time performance and accuracy of the environmental state matrix.

[0070] Example 3

[0071] As shown Figure 3 in Figure [not provided], the sand-gas flow coupling simulation engine 2 driven by a graph neural network includes a physical information enhanced graph neural network module 21, a multi-scale hybrid discrete element model 22, a recurrent convolutional GNN accelerator 23, and a flow field rapid prediction algorithm module 24.

[0072] The physical information enhanced graph neural network module 21 constructs a physical information enhanced graph network structure with sand grains as nodes and the flow field as edges. This structure integrates the N-S equation constraints in traditional CFD into the message passing mechanism of the graph network to achieve fast calculations under physical constraints. The graph network realizes information transmission through a node update function φ v and an edge update function φ e :

[0073]

[0074] Where: is the feature of the l-th layer node, is the feature of the l-th layer node, is the edge feature connecting nodes i and j, N(i) is the neighbor set of the node, and φ v and φ e are the update functions of the node and the edge respectively. The node update function is implemented using a multi-layer perceptron with a hidden layer dimension set to 128 and a LeakyReLU activation function. The edge update function also uses a multi-layer perceptron, but the hidden layer dimension is set to 64.

[0075] The multi-scale hybrid discrete element model 22 combines the discrete element method and computational fluid dynamics to construct a sand-seed-gas three-phase interaction mathematical model. This model uses a grid-based CFD method to solve the flow field at the macroscopic scale and the discrete element method to simulate the movement of sand grains at the microscopic scale, coupling the two scales through momentum exchange. The contact force between sand grains in the model is calculated using the Hertz-Mindlin contact model, with a damping coefficient set to 0.3, a static friction coefficient of 0.6, and a dynamic friction coefficient of 0.4.

[0076] The recurrent convolutional GNN accelerator 23, with the support of a pre-trained model, compresses traditional CFD calculations that take 1 - 2 minutes to be completed within 50 ms. The accelerator uses a 4-layer recurrent convolutional GNN network, with each layer containing 32 convolutional kernels, a stride of 1, and padding of SAME. The model is pre-trained on 50,000 sets of CFD simulation data, using the Adam optimizer with a learning rate of 0.001 and a batch size of 32. The model compression uses the Tucker decomposition technique, with a compression rate of 85% and a retained prediction accuracy of 98%.

[0077] The flow field rapid prediction algorithm module 24 uses dynamic adaptive grid partitioning technology to achieve 3m 3Prediction of the sand grain motion field within a spatial range with a spatial resolution of 2 cm 3 , with a time resolution of 50 ms. The algorithm automatically refines the grid in high-gradient regions and uses a coarse grid in regions with gentle gradients, optimizing the allocation of computing resources. The grid refinement threshold is set to 15% of the gradient value, the maximum refinement level is 3 levels, and the number of grids is dynamically controlled within 100,000.

[0078] Example 4

[0079] As Figure 4 shown, the seed-sand multi-physical field coupling behavior prediction model 3 includes a multi-physical field coupling model module 31, a seed motion trajectory prediction algorithm module 32, a sand grain migration risk assessment engine 33, and a real-time anti-displacement coefficient calculator 34.

[0080] The multi-physical field coupling model module 31 integrates the mechanical force field, fluid force field, and surface energy field to construct a multi-field coupling mathematical model. This model considers the gravity, buoyancy, fluid drag force, lift force, and adhesion and friction forces between the seed and sand grains acting on the seed. The force balance equation can be expressed as:

[0081]

[0082] where: m is the mass of the seed, is the position vector of the seed, is the gravity, is the buoyancy, is the drag force, is the lift force, is the adhesion force, is the friction force. The drag force is calculated using a modified drag formula, considering the correction coefficient for the Reynolds number range between 10 - 1000. The adhesion force model is based on the JKR contact theory, considering the relationship between the surface energy of the seed and the contact area with the sand grains.

[0083] The seed motion trajectory prediction algorithm module 32 predicts the entire trajectory of the seed from release to landing based on the Lagrangian particle tracking method. The algorithm uses a fourth-order Runge-Kutta method with an adaptive time step to solve the motion equation. The initial time step is 10 ms, the minimum time step is 0.1 ms, and the relative error threshold is 10-5. Considering the influence of the seed geometry, the algorithm introduces a shape factor, and the drag coefficient and lift coefficient for different seed shapes are obtained by fitting the wind tunnel experimental data.

[0084] The sand grain migration risk assessment engine 33 establishes a risk assessment model based on Monte Carlo simulation to calculate the displacement probability distribution of seeds after landing under different wind speed conditions. The engine conducts 10,000 random sampling simulations, considering the random variations of factors such as wind speed, direction, surface roughness, and seed characteristics, to generate the displacement risk probability distribution map P(x,y). The risk levels are divided into three levels: low, medium, and high, corresponding to displacement probabilities <20%, 20% - 50%, and >50%.

[0085] The real-time calculator 34 of the anti-displacement coefficient constructs a prediction model for the anti-displacement coefficient of seeds, comprehensively considering seed shape, mass, surface characteristics, and surface parameters. The anti-displacement coefficient k(s,E) is defined as an index indicating the ability of seeds to resist displacement caused by wind force, ranging from 0 to 1, and the larger the value, the stronger the anti-displacement ability. The calculation formula is:

[0086]

[0087] Where: m s is the seed mass, A s is the windward area of the seed, F a is the adhesion force, F d is the wind drag force, R s is the surface roughness of the seed, M s is the moisture content of the seed, and α, β, γ, and δ are weight coefficients obtained by fitting experimental data, which are 0.3, 0.4, 0.2, and 0.1 respectively.

[0088] Example 5

[0089] As Figure 5 shown, the adaptive intelligent optimization system 4 for aerial seeding parameters includes a dynamic optimizer 41 for hovering height, a multi-objective Bayesian optimization framework 42, a self-learning module 43 for the air pressure - ejection relationship, and a state space feedback controller 44.

[0090] The dynamic optimizer 41 for hovering height uses a deep reinforcement learning algorithm to calculate the optimal hovering height in the range of 0.5 - 2.5m in real time. The optimizer adopts a deep Q-network (DQN) structure. The state space includes current wind field characteristics, surface characteristics, and seed parameters, and the action space is the discretized hovering height values (0.5m, 1.0m, 1.5m, 2.0m, 2.5m). The reward function comprehensively considers the seed landing accuracy and the anti-displacement coefficient. The network structure is a three-layer fully connected network, with the number of hidden layer nodes being 256, 128, and 64 respectively. The activation function is ReLU, the learning rate is 0.001, the discount factor is 0.95, and the initial value of ε in the ε-greedy strategy is 0.9, with a decay rate of 0.995.

[0091] The multi-objective Bayesian optimization framework 42 constructs a three-objective optimization framework for seed accuracy, anti-displacement, and efficiency, and dynamically calculates the optimal seeding parameters. The framework uses a Gaussian process regression model to construct a surrogate model and adopts the expected improvement (EI) as the acquisition function. The optimization variables include the ejection speed (1 - 10 m / s), ejection angle (30° - 60°), and seed wrapping parameters. The multi-objective optimization uses the Pareto front method and evaluates the solution set quality using the hypervolume metric. The kernel function of the Gaussian process selects the Matérn 5 / 2 kernel, with a length scale parameter of 0.1 and a signal variance of 1.0.

[0092] The air pressure - ejection relationship self-learning module 43 establishes the mapping relationship between air pressure and ejection speed and angle. The module adopts an online learning method and continuously updates the model parameters based on actual ejection data. The mapping relationship uses a quadratic polynomial model:

[0093] v = a1P 2 + b1P + c1,

[0094] θ = a2P 2 + b2P + c2,

[0095] where: v is the ejection speed, θ is the ejection angle, P is the air pressure value, and a1, b1, c1, a2, b2, c2 are model parameters. The initial parameters are obtained through pre-calibration. The online learning adopts the recursive least squares method with a forgetting factor set to 0.95 to ensure that the model can adapt to the dynamic changes of the air pressure - ejection relationship.

[0096] The state space feedback controller 44 realizes the real-time closed-loop control of parameters based on the microenvironment state space model. The controller adopts the linear quadratic regulator (LQR) design method. The state variables include the current position error, speed error, and cumulative error, and the control variable is the air pressure adjustment amount. The state space equation is:

[0097]

[0098] y = Cx,

[0099] where: x is the state vector, u is the control vector, y is the output vector, A is the system matrix, B is the input matrix, and C is the output matrix. The state feedback gain matrix K of the LQR controller is obtained by solving the Riccati equation. The state weight matrix Q is set as the diagonal matrix diag(10, 5, 1), the control weight matrix R is set to 1, and the control update frequency is 50 Hz.

[0100] Example 6

[0101] Such as Figure 6As shown, the variable stiffness ejection actuator 5 includes a magnetorheological fluid-based variable stiffness ejection mechanism 51, a seed surface bionic microstructure processing unit 52, an environmentally responsive hydrogel wrapping device 53, and an aerodynamic optimization processing unit 54.

[0102] The magnetorheological fluid-based variable stiffness ejection mechanism 51 realizes dynamic adjustment of the spring stiffness by adjusting the magnetic field strength. The mechanism adopts a flexible cavity structure filled with magnetorheological fluid, and the magnetic field is generated by an electromagnetic coil. The number of turns of the coil is 500, the maximum current is 2A, and the generated magnetic field strength ranges from 0 to 1.2T. The magnetorheological fluid is prepared by suspending carbonyl iron powder (average particle size 5μm, volume fraction 30%) in silicone oil (viscosity 0.1Pa·s). The stiffness adjustment range is 5 - 50N / m, the response time is less than 50ms, and the relationship between stiffness and magnetic field strength is non-linear, and fast mapping is achieved through the look-up table method. The volume of the compressed air storage tank is 500mL, the maximum working pressure is 1.5MPa, the air pressure is precisely controlled by a proportional pressure regulating valve, the pressure adjustment range is 0.1 - 1.0MPa, and the accuracy is 0.01MPa.

[0103] The seed surface bionic microstructure processing unit 52 constructs an anti-wind and sand microstructure array on the seed surface. The unit is based on the bionic design of the microstructure on the surface of desert plant seeds, and a 3D micro-nano printing technology is used to manufacture the microstructure template, and the feature size of the template is 50 - 200μm. The processing process includes: cleaning the seed surface, coating with photocurable resin, imprinting the microstructure, UV curing, and post-treatment. The microstructure morphology includes conical arrays, groove arrays, and hybrid structures, with a density of 100 - 400 per mm 2 , and a height of 30 - 120μm. The microstructure increases the contact area between the seed and the sand grains and improves the adhesion force.

[0104] The environmentally responsive hydrogel wrapping device 53 provides a composite hydrogel wrapping layer containing silicate and alginate for the seeds. The hydrogel formulation includes sodium alginate (2% w / v), sodium silicate (1% w / v), calcium chloride (1% w / v), and glycerol (5% w / v). The wrapping process adopts a fluidized bed technology to control the thickness of the wrapping layer within the range of 50 - 200μm. The hydrogel has strong water absorption and swelling characteristics, and can quickly absorb soil moisture to form a stable layer after landing. The water absorption rate can reach 20 times its own weight, and the swelling time is 10 - 30s. Nutrient elements and moisturizers are also added to the wrapping layer to improve the seed survival rate.

[0105] The aerodynamic optimization processing unit 54 optimizes the aerodynamic shape of the seeds, reduces the wind load ratio, and improves stability. The unit constructs a database of the aerodynamic characteristics of the seeds based on the previous wind tunnel experiment data, including the drag coefficients and lift coefficients of 50 typical shapes. The optimization uses shape modification technology to adjust the center of gravity position and aerodynamic characteristics of the seeds by adding aerodynamic attachments. The attachment material is biodegradable plant fiber, and the attachment shapes include stabilizing wings, vortex generators, and drag rings, with the weight controlled within 5% of the seed weight. The optimization goal is to minimize the wind load ratio (wind force / gravity) while ensuring a stable falling attitude.

[0106] Example 7

[0107] In this embodiment, the key parameters of the system are set as follows:

[0108] The dimension of the standard environmental state matrix E(t) is 1000×15, including 1000 spatial discretization points and 15 characteristic dimensions (3D wind speed, 3D sand grain speed, 3D position coordinates, 3D surface characteristic parameters, and 3D time stamps). The update frequency is 100Hz, and the data transmission rate is 100Mbps.

[0109] The spatial resolution of the flow field prediction data F(t+Δt) is 2cm, the time prediction step is 50ms, and the calculation time is controlled within 100ms. The prediction accuracy is verified by comparing with high-precision CFD simulations, and the relative error is less than 5%.

[0110] The spatial coverage of the displacement risk probability distribution map P(x,y) is 10m×10m, the resolution is 10cm×10cm, and the calculation is based on 10,000 Monte Carlo simulations. The risk levels are divided into three levels: low risk (green, displacement probability <20%), medium risk (yellow, displacement probability 20% - 50%), and high risk (red, displacement probability >50%).

[0111] The search range of the optimal hovering height h* is 0.5 - 2.5m, the step size is 0.1m, and the optimization response time is less than 150ms. The ejection parameters include the speed v (range 1 - 10m / s) and the angle θ (range 30° - 60°), and the parameter resolution is 0.1m / s and 1°.

[0112] The control instruction C(t) includes the hovering height, ejection speed, ejection angle, stiffness parameter, and wrapping parameter, encoded as a 32-byte instruction packet, and the transmission delay is less than 5ms.

[0113] The working cycle of the system is 500 ms, including four stages: perception (100 ms), prediction (100 ms), optimization (100 ms), and execution (200 ms). The system can work stably within the wind speed range of 0 - 12 m / s, with an operating temperature range of -20°C to 55°C, and a relative humidity range of 10% - 90%.

[0114] Example 8

[0115] As Figure 7 shown, the system also includes a closed-loop feedback and continuous optimization module 6, which is used to compare the deviation between the theoretical trajectory and the actual trajectory, calculate the correction factor, update the model parameters, and continuously optimize the decision-making strategy based on the reinforcement learning method.

[0116] The closed-loop feedback and continuous optimization module 6 includes a trajectory deviation analysis unit 61, a model adaptive adjustment unit 62, a reinforcement learning optimization unit 63, and a performance monitoring and evaluation unit 64.

[0117] The trajectory deviation analysis unit 61 compares the deviation between the theoretical prediction trajectory T(s) and the actual measured trajectory Ta(s), and calculates the correction factor c. The deviation is quantified by the Euclidean distance and the dynamic time warping (DTW) algorithm, and the correction factor is calculated by the exponentially weighted moving average method with a smoothing coefficient of 0.3. The deviation analysis considers three dimensions: position, velocity, and angle, and generates a comprehensive deviation index.

[0118] The model adaptive adjustment unit 62 updates the parameters of the prediction model according to the correction factor c. The adjustment uses the gradient descent method, and the learning rate is dynamically adjusted according to the deviation size, with a range of 0.001 - 0.1. The parameter update frequency is once every 10 flights to prevent overfitting. The key parameters to be adjusted include the turbulence intensity parameter in the flow field prediction model, the adhesion coefficient and the wind force coefficient in the seed-sand coupling model.

[0119] The reinforcement learning optimization unit 63 continuously optimizes the decision-making strategy based on the actual effect. The unit uses the PPO (Proximal Policy Optimization) algorithm. Both the policy network and the value network use a 3-layer fully connected network, with 128 hidden nodes and the activation function being tanh. The training is carried out online, with every 500 flight experiences as a batch, and 10 epochs are trained. The reward function comprehensively considers the seed landing accuracy, the anti-displacement effect, and the operation efficiency, with weights of 0.4, 0.4, and 0.2 respectively.

[0120] The performance monitoring and evaluation unit 64 monitors various indicators of the system in real time and generates a performance evaluation report. The monitoring indicators include the seed landing accuracy (target point error), anti-displacement effect (displacement distance / wind speed), prediction accuracy (deviation between the predicted trajectory and the actual trajectory), and system response time. The evaluation report is generated every 100 aerial seeding operations and includes statistical analysis of the indicators and improvement suggestions.

[0121] The iteration period of the closed-loop feedback and continuous optimization module 6 is 500 ms, and the model optimization speed per single iteration is not less than 15% per 100 aerial seeding operations. Through continuous learning and optimization, the system performance is continuously improved during long-term use to adapt to different environmental conditions.

[0122] Example 9

[0123] This system also has a variety of extended functions to adapt to different application scenarios and environmental conditions.

[0124] First of all, the system supports parametric configuration of different seed types. The built-in seed database contains the physical property parameters of 50 common afforestation and grass seeds, including geometric dimensions, mass, shape factor, surface roughness, and water absorption characteristics. Users can select the seed type through the interface, and the system automatically loads the corresponding parameters and optimizes the aerial seeding strategy. For new seed types not included in the database, the system provides a rapid parameter measurement process, including image recognition measurement and physical property testing, which can complete parameterization within 15 minutes.

[0125] Secondly, the system supports multi-UAV collaborative operation. A single control center can simultaneously manage 5 - 10 UAVs equipped with this system, allocate different operation areas, and monitor the operation progress. The UAVs share environmental data and aerial seeding experience through a self-organizing network to improve the overall operation efficiency. The collaborative strategy adopts an auction-based task allocation mechanism, considering the UAV battery power, position, and loading capacity to achieve task load balancing.

[0126] Thirdly, the system integrates the post-aerial seeding evaluation function. By recording the seed landing positions with the on-board high-resolution camera and combining with GPS positioning information, a seeding distribution map is generated. 30 days after seeding, the system can conduct a flight review, evaluate the seed germination and growth status through multi-spectral imaging technology, calculate the survival rate and coverage rate, and provide a basis for optimizing subsequent aerial seeding strategies.

[0127] In addition, the system also supports an emergency mode under extreme weather conditions. When the wind speed exceeds the threshold (12 m / s) or a sudden strong wind occurs, the system automatically switches to a conservative mode and adjusts the aerial seeding parameters to ensure safe operation. In the emergency mode, the system gives priority to the seed anti-displacement property, sacrifices the seeding accuracy appropriately, and adopts a lower hovering height and a stronger ejection force to ensure that the seeds land quickly and stably.

[0128] Example 10

[0129] As Figure 10 shown, the method for predicting the microenvironment of aerial seeding seeds based on sand grain movement simulation includes the following steps:

[0130] Step S1: Collect wind field data, sand grain movement data, and surface characteristic data through a multi-source heterogeneous micro-scale sensing network, and generate a standard environmental state matrix E(t).

[0131] Specifically, capture the sand grain movement trajectory and velocity field through a micro PIV particle image velocimetry system, measure the three-dimensional wind field distribution through a micro six-degree-of-freedom wind field sensor array, obtain the physical parameters of the surface sand layer through a surface characteristic rapid detector, and use the Kalman filter algorithm to perform spatio-temporal alignment and fusion of multi-sensor data to generate a standard environmental state matrix.

[0132] Step S2: Input the standard environmental state matrix E(t) into the sand grain-airflow coupling simulation engine driven by a graph neural network, and calculate and generate flow field prediction data F(t+Δt) through a physics-informed graph neural network.

[0133] Specifically, construct a physics-informed graph network structure with sand grains as nodes and the flow field as edges, combine the discrete element method and computational fluid dynamics, construct a sand grain-seed-airflow three-phase interaction mathematical model, perform rapid inference through a pre-trained recurrent convolutional GNN network, and use a dynamic adaptive grid division technology to achieve the prediction of the sand grain movement field within a 3m 3 spatial range.

[0134] Step S3: Based on the flow field prediction data F(t+Δt), calculate the seed movement trajectory T(s) and the seed anti-displacement coefficient k(s,E) through a seed-sandy land multi-physical field coupling behavior prediction model, and generate a displacement risk probability distribution map P(x,y).

[0135] Specifically, integrate a multi-field coupling mathematical model of a mechanical force field, a fluid force field, and a surface energy field, predict the entire process trajectory of the seed from release to landing based on the Lagrangian particle tracking method, establish a risk assessment model based on Monte Carlo simulation, calculate the displacement probability distribution of the seed after landing under different wind speed conditions, construct a seed anti-displacement coefficient prediction model, and comprehensively consider the seed shape, mass, surface characteristics, and surface parameters.

[0136] Step S4: Input the displacement risk probability distribution map P(x,y), the seed movement trajectory T(s), and the seed anti-displacement coefficient k(s,E) into an adaptive aerial seeding parameter intelligent optimization system, calculate the optimal hovering height h* and ejection parameters (v,θ), and generate a control instruction C(t).

[0137] Specifically, using the deep reinforcement learning algorithm, the optimal hovering height within the range of 0.5 - 2.5 m is calculated in real time, a three-objective optimization framework for seed accuracy, anti-displacement, and efficiency is constructed, the optimal seeding parameters are dynamically calculated, the mapping relationship between air pressure and ejection speed and angle is established, and a state-space feedback controller is designed to achieve real-time closed-loop control of the parameters.

[0138] Step S5: Receive the control command C(t) through the variable stiffness ejection actuator, adjust the ejection stiffness k* and the seed wrapping parameters, execute the seed ejection, and collect the actual trajectory data Ta(s) as feedback data to input into the adaptive aerial seeding parameter intelligent optimization system for continuously optimizing the decision-making strategy.

[0139] Specifically, the dynamic adjustment of the spring stiffness is realized by adjusting the magnetic field intensity, an anti-wind and sand microstructure array is constructed on the seed surface, an environmentally responsive hydrogel wrapping layer is provided for the seed, the aerodynamic shape of the seed is optimized, the seed ejection is executed, the actual trajectory data is collected, the deviation between the theoretical trajectory and the actual trajectory is compared, the correction factor is calculated, the model parameters are updated, and the decision-making strategy is continuously optimized based on the reinforcement learning method.

[0140] Example 11

[0141] The system of the present invention was verified in the aerial seeding afforestation experiment on the eastern edge of Kubuqi Desert in Inner Mongolia. The experimental area is 500 mu, the main vegetation is Artemisia desertorum and Salix psammophila, the average wind speed is 8.5 m / s, and the ground surface is mainly composed of mobile sand dunes and semi-fixed sand dunes. Two typical afforestation seeds, Pinus sylvestris var. mongolica and Caragana korshinskii, were selected for testing, and the number of seeds sown each time is 5000.

[0142] The comparison results between the conventional aerial seeding system (the system of Comparative Document 1) and the system of the present invention are as follows:

[0143] 1. Seed anti-displacement effect: Under the wind speed condition of 8.5 m / s, within 24 hours after sowing by the conventional system, the average displacement distance of the seeds is 3.6 m, and 47% of the seeds are blown more than 5 m away from the original position; within 24 hours after sowing by the system of the present invention, the average displacement distance of the seeds is only 1.3 m, and only 16% of the seeds are displaced more than 5 m, and the anti-displacement coefficient is increased by 2.8 times.

[0144] 2. Seed accurate landing rate: The average deviation between the landing position of the seeds of the conventional system and the target position is 2.4 m, and only 77% of the seeds land within the predetermined 3 m × 3 m grid; the average deviation of the system of the present invention is 0.9 m, and 92% of the seeds land within the predetermined grid, and the accuracy is increased by 15%.

[0145] 3. Seed survival rate: After 30 days of sowing, the average survival rate of the conventional system is 60%, while that of the system of the present invention reaches 85%, an increase of 25%. The survival rate of Pinus sylvestris var. mongolica increases from 55% to 82%, and the survival rate of Caragana korshinskii increases from 65% to 88%.

[0146] 4. Operation efficiency: The conventional system cannot operate stably when the wind speed is above 6 m / s, and the daily operation area is about 110 mu; the system of the present invention can operate stably when the wind speed is below 12 m / s, and the daily operation area reaches 150 mu, an increase of 36%.

[0147] 5. System adaptability: Tested under various terrain conditions such as the edge of the desert, hilly terrain, and river valleys, the system of the present invention shows good adaptability. Especially in areas with complex terrain, the system can automatically adjust the aerial seeding parameters to maintain high seeding accuracy and seed stability.

[0148] Through 16 months of tracking and monitoring, the vegetation coverage rate of the aerial seeding area of the system of the present invention reaches 65%, 18 percentage points higher than that of the conventional system. The vegetation height is uniform, the distribution is more reasonable, and the anti-wind and sand ability is significantly enhanced. The system operates stably, has a low failure rate, and the maintenance cost is reduced by 30% compared with the conventional system.

[0149] The aerial seeding seed microenvironment prediction system and method based on sand grain movement simulation of the present invention provide an efficient and reliable technical means for vegetation restoration and ecological construction in arid and semi-arid regions by solving the technical problem of easy displacement of seeds in the sandy environment. The system has been applied in aerial seeding afforestation projects in areas such as Kubuqi Desert and Mu Us Sandy Land, and has achieved remarkable ecological and economic benefits.

[0150] The present invention is not only applicable to the field of desertification control, but also can be popularized and applied to multiple ecological restoration fields such as grassland restoration, mine reclamation, and soil and water conservation. The modular design and standardized interface of the system make it easy to integrate with existing aerial seeding equipment, reducing the technical threshold and cost of popularization and application.

[0151] In addition, the multi-scale microenvironment perception and intelligent prediction technology and the fast flow field simulation method driven by graph neural network in the present invention can also be extended and applied to fields such as meteorological monitoring, environmental assessment, and disaster warning, and have broad application prospects and industrialization potential.

[0152] The above are only the preferred embodiments of the present invention and are not used to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A prediction system for the microenvironment of aerial seeding seeds based on sand grain movement simulation, characterized in that, include: A multi-source heterogeneous micro-scale sensing network is used to collect wind field data, sand movement data, and surface characteristics data, and generate a standard environmental state matrix; A sand-airflow coupling simulation engine driven by a graph neural network is connected to the multi-source heterogeneous microscale sensing network, and is used to receive the standard environmental state matrix and generate flow field prediction data by enhancing the graph neural network calculation through physical information; A seed-sand multi-physics field coupling behavior prediction model is connected to the sand-airflow coupling simulation engine driven by the graph neural network, and is used to calculate the seed motion trajectory and seed anti-displacement coefficient based on the flow field prediction data, and generate a displacement risk probability distribution map; An adaptive flying seeding parameter intelligent optimization system is connected to the seed-sand multi-physical field coupling behavior prediction model, and is used to calculate the optimal hovering height and ejection parameters based on the displacement risk probability distribution diagram, the seed motion trajectory and the seed anti-displacement coefficient, and generate control instructions; The variable stiffness ejection actuator is connected to the adaptive flying sowing parameter intelligent optimization system, and is used to receive the control instruction, adjust the ejection stiffness and seed wrapping parameters, execute seed ejection, and collect actual trajectory data as feedback data to be sent to the adaptive flying sowing parameter intelligent optimization system.

2. The system according to claim 1, wherein The multi-source heterogeneous micro-scale sensing network includes: Micro PIV particle image velocimetry system, used to capture the motion trajectory and velocity field of sand particles with a scale of 10-500μm; Miniature six-degree-of-freedom wind field sensor array for measuring the three-dimensional wind field distribution in a 3m 3 space; Rapid surface property detector, used to obtain the roughness, density and water content parameters of the surface sand layer; The data fusion processing unit is connected to the micro PIV particle image velocimetry system, the micro six-DOF wind field sensor array and the surface characteristic rapid detector, and is used to perform spatiotemporal alignment and fusion of multi-sensor data using a Kalman filter algorithm to generate the standard environmental state matrix.

3. The system according to claim 1, wherein The graph neural network driven sand-airflow coupling simulation engine includes: Physical information enhanced graph neural network module, which is used to construct a physical information enhanced graph network structure by taking sand grains as nodes and flow fields as edges; Multiscale hybrid discrete element model, which is used to combine discrete element method with computational fluid dynamics to construct a mathematical model of sand-seed-airflow three-phase interaction; Circular convolutional GNN accelerator, which is used to compress traditional CFD calculations to within 50ms through the support of pre-trained models; Flow field rapid prediction algorithm module, which is used to implement the prediction of the sand grain motion field within a 3m 3 spatial range by using dynamic adaptive grid division technology.

4. The system according to claim 1, characterized in that, The seed-sand multi-physics field coupling behavior prediction model includes: Multi-physics field coupling model module, used to integrate mechanical force field, fluid force field and surface energy field to construct a multi-field coupling mathematical model; Seed motion trajectory prediction algorithm module, used to predict the trajectory of seeds from release to landing based on the Lagrangian particle tracking method; Sand migration risk assessment engine, which is used to establish a risk assessment model based on Monte Carlo simulation and calculate the displacement probability distribution of seeds after landing under different wind speed conditions; The real-time calculator of anti-displacement coefficient is used to construct a prediction model for seed anti-displacement coefficient, taking into account seed shape, quality, surface characteristics and ground surface parameters.

5. The system according to claim 1, characterized in that, The adaptive flying seeding parameter intelligent optimization system comprises: Hover height dynamic optimizer, which uses a deep reinforcement learning algorithm to calculate the optimal hover height in the range of 0.5 - 2.5 m in real time; Multi-objective Bayesian optimization framework, which is used to construct an optimization framework for three objectives of seed accuracy, anti-displacement ability and efficiency, and dynamically calculate the optimal seeding parameters; Air pressure - ejection relationship self-learning module, which is used to establish the mapping relationship between air pressure and ejection speed and angle; State space feedback controller, which is used to realize the real-time closed-loop control of parameters based on the microenvironment state space model.

6. The system according to claim 1, characterized in that The variable stiffness ejection actuator includes: Magnetorheological fluid-based variable stiffness ejection mechanism, which is used to dynamically adjust the spring stiffness by adjusting the magnetic field strength; Bionic micro-structure processing unit on the seed surface, which is used to construct an anti-wind and sand micro-structure array on the seed surface; Environment-responsive hydrogel wrapping device, which is used to provide a composite hydrogel wrapping layer containing silicate and alginate for the seeds; Aerodynamic optimization processing unit, which is used to optimize the aerodynamic shape of the seeds, reduce the wind load ratio and improve the stability.

7. The system according to claim 1, wherein The update frequency of the standard environmental state matrix E(t) is not less than 100 Hz, the calculation time of the flow field prediction data F(t+Δt) is less than 100 ms, and the spatial resolution of the displacement risk probability distribution map P(x,y) is not less than 2 cm 3 .

8. The system according to claim 1, characterized in that, The system also includes a closed-loop feedback and continuous optimization module, which is used to compare the deviation between the theoretical trajectory and the actual trajectory, calculate the correction factor, update the model parameters, and continuously optimize the decision-making strategy based on the reinforcement learning method.

9. The system according to claim 8, wherein The iteration period of the closed-loop feedback and continuous optimization module is less than 500 ms, and the optimization speed of the single-iteration model is not less than 15% per 100 aerial seeding operations.

10. A method for predicting the microenvironment of aerial seeding seeds based on sand grain movement simulation, using the method according to any one of claims 1-9, characterized in that, It includes the following steps: Collect wind field data, sand grain movement data and surface characteristic data through a multi-source heterogeneous micro-scale perception network, and generate a standard environment state matrix; Input the standard environment state matrix into a sand grain - air flow coupling simulation engine driven by a graph neural network, and calculate and generate flow field prediction data through a physics-informed graph neural network; Based on the flow field prediction data, calculate the seed movement trajectory and the seed anti-displacement coefficient through a seed - sandy land multi-physical field coupling behavior prediction model, and generate a displacement risk probability distribution map; Input the displacement risk probability distribution map, the seed movement trajectory and the seed anti-displacement coefficient into an adaptive aerial seeding parameter intelligent optimization system, calculate the optimal hover height and ejection parameters, and generate a control command; Receive the control command through the variable stiffness ejection actuator, adjust the ejection stiffness and the seed wrapping parameters, execute the seed ejection, and collect the actual trajectory data as feedback data and input it into the adaptive aerial seeding parameter intelligent optimization system for continuously optimizing the decision-making strategy.

Images