Method for determining soil leaching performance of nanometer fumigant

Abstract

This invention provides a method for determining the soil leaching performance of nano-nematicides, belonging to the technical field of soil leaching performance determination of nano-nematicides. This invention achieves dual-channel in-situ monitoring by embedding a high-frequency micro-pressure sensor and a flexible optical fiber surface-enhanced Raman scattering sensor in a soil column. It dynamically corrects the leaching flux using a seepage mechanics random walk and a lattice Boltzmann coupled transport algorithm. Effluent from various depths is collected, quantified by high-performance liquid chromatography, and Raman signals are recorded simultaneously. All data are input into a structure-adaptive graph attention spatiotemporal evolution network (artificial intelligence) to predict the longitudinal leaching curve and the constitutive relationship of the drug-loaded body stability. Finally, a quantitative evaluation system for soil physicochemical parameters and the mobility of nano-nematicides is established. This solves the technical problem that the abnormal leaching flux caused by the dynamic dissociation of non-covalent bonds in the self-assembled drug-loaded body during soil leaching, and the inability to achieve in-situ real-time quantitative characterization, is due to the nano-nematicides.

Description

Technical Field

[0001] This invention belongs to the technical field of soil leaching performance testing of nano-nematicides, specifically, it relates to a method for testing the soil leaching performance of nano-nematicides. Background Technology

[0002] Nano-nematicides are novel pesticide formulations that encapsulate active ingredients such as abamectin and fluopyram within nanocarriers using self-assembly technology. Their soil leaching performance directly determines the persistence of efficacy and the results of environmental risk assessment. Existing methods for measuring soil leaching mainly rely on column leaching experiments combined with offline high-performance liquid chromatography (HPLC). Leaching temperature curves are plotted by collecting eluent in segments and measuring the concentration of active ingredients. Some studies have also introduced Darcy's law to correct for percolation flow and used traditional diffusion models to simulate the transport behavior of nanoparticles.

[0003] However, the above methods face inherent limitations when processing self-assembled nanocarriers. When soil ionic strength changes gradient, pH stratifies vertically, or soil colloids compete with the drug carrier for adsorption, the non-covalently driven supramolecular structure may undergo sudden dissociation at a certain soil depth, leading to a cliff-like anomaly in leaching flux. Traditional Darcy's law and diffusion models are based on the assumption of a continuous homogeneous medium, which cannot describe the dynamic flow field redistribution of non-covalent aggregates during the alternating blockage and unblocking process, nor can they capture the dissociation state of the drug carrier in situ without damaging the soil column structure.

[0004] In existing technologies, the lack of in-situ sensing methods capable of simultaneously detecting transient changes in pore water pressure and the integrity of the self-assembled structure of the drug-loaded organism, coupled with the inability of traditional transport algorithms to couple the randomness of Brownian motion with the probability of non-covalent adhesion within the same computational framework, makes it impossible to capture abnormal cliffs and tail-hitting phenomena in leaching curves in real time, nor to provide quantitatively interpretable predictions based on physical mechanisms. In other words, existing technologies suffer from the technical problem of abnormal leaching flux caused by the dynamic dissociation of non-covalent bonds in the self-assembled drug-loaded organism during soil leaching of nano-nematicides, and the inability to achieve in-situ real-time quantitative characterization. Summary of the Invention

[0005] In view of this, the present invention provides a method for determining the soil leaching performance of nano-nematicides, which can solve the technical problem in the prior art that the leaching flux of nano-nematicides is abnormal due to the dynamic dissociation of non-covalent bonds of self-assembled drug bodies during soil leaching, and that in-situ real-time quantitative characterization cannot be achieved.

[0006] This invention is achieved as follows: This invention provides a method for determining the soil leaching performance of nano-nematicides, comprising the following steps:

[0007] A self-assembled drug-loaded nano-negative agent was prepared, and its particle size distribution, zeta potential, and Fourier transform infrared spectral matrix were measured. These parameters were used as the basic characterization data for subsequent model input.

[0008] A standard soil column leaching device was constructed, and a high-frequency micro-pressure sensor was embedded in the longitudinal direction of the soil column. At the same time, flexible fiber surface-enhanced Raman scattering sensors were deployed at the longitudinal gradient position to monitor the pore water pressure and the Raman signal of the drug-loaded self-assembled structure in situ.

[0009] The seepage mechanics random walk and lattice Boltzmann coupled transport algorithm is activated to discretize the soil pore structure into a three-dimensional grid space. Combined with the pore water pressure data collected in real time by the high-frequency micro-pressure sensor, the local permeability loss is dynamically corrected to achieve accurate compensation of leaching flux.

[0010] Effluent from different depths of the soil column was collected, and the peak area was determined by high performance liquid chromatography. The Raman signal intensity and redshift of the flexible fiber surface-enhanced Raman scattering sensor at each soil depth were recorded simultaneously to quantitatively evaluate the free release rate of the drug carrier in situ.

[0011] Particle size distribution, Zeta potential, Fourier transform infrared spectral matrix, peak area of ​​effluent at each depth and Raman signal data are input into the structure-adaptive graph attention spatiotemporal evolution network. The non-covalent bond dissociation critical point is identified through the adaptive jump connection mechanism, the longitudinal leaching curve is predicted and the constitutive relationship of drug-loaded stability at each soil depth is output.

[0012] Based on the leaching curves and stability constitutive relations output by the structural adaptive graph attention spatiotemporal evolution network, and combined with the measured concentration data of effluent from each soil layer, a quantitative evaluation system for soil physicochemical parameters and the mobility of nano-nematicides was established, and the comprehensive determination of leaching performance was completed.

[0013] The self-assembly loaded nano-nematicide refers to a supramolecular aggregate formed by the spontaneous assembly of active ingredients and nano-carrier molecules driven by non-covalent bonds. The non-covalent bonds include hydrogen bonds, hydrophobic interactions, and tannic acid-metal coordination bonds.

[0014] The particle size distribution was determined using a dynamic light scattering instrument, the Zeta potential reflected the electrostatic stability between particles, and the Fourier transform infrared spectral matrix was in the range of 400–4000. The data were collected within the wavenumber range, and the normal characterization intervals of the three parameters were determined by repeatedly preparing at least a batch of threshold batches, and then using the mean ± 2 times the standard deviation.

[0015] The high-frequency micro-pressure sensor is a piezoresistive micro-sensor with a response frequency not lower than the frequency threshold and a measurement range covering the pore water pressure threshold range. The embedding spacing is determined through pre-experimentation to identify the maximum spacing that can detect pressure change signals.

[0016] The flexible fiber surface-enhanced Raman scattering sensor embeds a flexible fiber probe with gold nanoparticles on its surface into a soil column, and amplifies the Raman scattering signal of the target molecule by utilizing the local plasmon resonance effect. The characteristic peak redshift threshold is determined by the average of no less than three independent experiments.

[0017] The seepage mechanics random walk and lattice Boltzmann coupled transport algorithm uses a random walk model to simulate Brownian motion random displacement at the micro level, and solves the Navier-Stokes equations through two steps of collision and migration at the macro level. The adhesion probability threshold is obtained by combining no less than 5 sets of penetration curve experiments with DLVO theoretical parameter inversion.

[0018] Wherein, the free release rate is The standard curve is calculated by substituting the Raman signal intensity attenuation rate at each soil depth into the standard curve. The linear regression correlation coefficient of the standard curve must be no less than the correlation coefficient threshold before it can be used for quantitative calculation.

[0019] The structure-adaptive graph attention spatiotemporal evolution network uses an improved graph attention network as its backbone, maps the pore network topology to spatial graph nodes, and maps the time step to temporal edges to form a spatiotemporal heterogeneous graph. The feature vector of each spatial node includes pore water pressure, Raman signal intensity, chromatographic peak area, Zeta potential projection value, and permeability correction coefficient.

[0020] The adaptive graph attention spatiotemporal evolution network has four graph attention layers. The attention coefficient calculation incorporates a rigid constraint factor composed of soil permeability and hydrodynamic shear force. The inter-layer feature transfer adopts a dual-channel structure with residual jump connections and adaptive jump connections in parallel.

[0021] The adaptive jump connection is activated when the Raman signal intensity is detected to have decayed beyond the jump decay threshold within three consecutive time steps, and the dissociation state information is bypassed from the intermediate layer and directly transmitted to the dissolution dynamic prediction layer.

[0022] The structure-adaptive graph attention spatiotemporal evolution network is configured with a seepage dynamic compensation function. Based on the rate of change of pore water pressure Raman signal attenuation rate and effluent concentration gradient Calculate the dynamic compensation index of seepage , After weighted summation, normalized using the sigmoid function to the nearest integer. Interval.

[0023] Among them, the seepage dynamic compensation index The number of activated attention heads and the number of time-series unfolding steps are dynamically adjusted based on the compensation index threshold range. Each weight coefficient is determined after cross-validation on no less than 10 sets of standard leaching experimental data of different soil types.

[0024] The training dataset of the structure-adaptive graph attention spatiotemporal evolution network includes no less than 50 sets of standard leaching experimental data covering sandy loam, clay loam, and soils with varying organic matter content. The measured leaching curves and actual free release rates at each depth are used as supervision labels. The training set, validation set, and test set are divided in an 8:1:1 ratio.

[0025] The structure-adaptive graph attention spatiotemporal evolution network is trained using the Adam optimizer. The loss function is a weighted sum of the mean square error loss and the graph structure regularization term. An early stopping strategy is adopted. After training, the model accuracy is evaluated by two indicators: root mean square error and Pearson correlation coefficient.

[0026] Wherein, the batch threshold is 5 batches; the frequency threshold is 100Hz; and the pore water pressure threshold range is 0–50 Hz. The correlation coefficient threshold is 0.99; the jump decay threshold is... The compensation index threshold range is: , , and The corresponding number of attention points activated are 8, 4, 2 and 1, respectively.

[0027] This invention describes the physical mechanism of dynamic clogging processes by simultaneously embedding a high-frequency micro-pressure sensor and a flexible optical fiber surface-enhanced Raman scattering sensor in a soil column, and then coupling the transient signal of pore water pressure with the redshift signal of the non-covalent characteristic peak of the drug carrier in real time into the seepage mechanics random walk and lattice Boltzmann coupled transport algorithm.

[0028] Traditional methods rely on offline chromatographic analysis, which cannot detect the real-time location and rate of non-covalent bond dissociation. This invention, however, amplifies Raman signals through the localized plasmon resonance effect of gold nanoparticles, quantifying the free release rate of the drug carrier in situ without damaging the soil column structure, thus overcoming the blind spots of traditional sensing methods in monitoring structural integrity. The structure-adaptive graph attention spatiotemporal evolution network maps the pore network topology to a spatiotemporal heterogeneous graph and incorporates physicochemical flow field constraints. This allows the critical abrupt change signal of non-covalent bond dissociation to bypass redundant time series layers and directly reach the prediction layer through adaptive jump connections, thereby ensuring both deep time series modeling capabilities and rapid response to sudden dissociation events.

[0029] In summary, this invention solves the technical problem mentioned in the background art, where the leaching flux of nano-nematicides is abnormal due to the dynamic dissociation of non-covalent bonds of self-assembled drug particles during soil leaching, and the inability to achieve in-situ real-time quantitative characterization. Attached Figure Description

[0030] Figure 1 This is a flowchart of the method of the present invention.

[0031] Figure 2 This is a time-series curve of pore water pressure variation at different depths of the soil column.

[0032] Figure 3 A comparison chart showing the leaching curves at various soil depths and the predicted values ​​from the structural adaptive attention spatiotemporal evolution network.

[0033] Figure 4 This is a distribution diagram showing the free release rate and redshift of Raman characteristic peaks at different soil depths. Detailed Implementation

[0034] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below.

[0035] like Figure 1 The diagram shown is a flowchart of a method for determining the soil leaching performance of a nano-nematicide provided by this invention. This method includes the following steps:

[0036] S01. Prepare self-assembled drug bodies of nano-nematicides, and determine their particle size distribution, Zeta potential and Fourier transform infrared spectral matrix. Use the above parameters as the basic characterization data for subsequent model input.

[0037] S02. Construct a standard soil column leaching device, embed high-frequency micro-pressure sensors every 1-5 cm in the longitudinal direction of the soil column, and simultaneously deploy flexible fiber surface-enhanced Raman scattering sensors at the longitudinal gradient position to monitor the pore water pressure and Raman signal of the drug-loaded self-assembly structure in situ.

[0038] S03. Activate the seepage mechanics random walk and lattice Boltzmann coupled transport algorithm to discretize the soil pore structure into a three-dimensional grid space. Combined with the pore water pressure data collected in real time by the high-frequency micro-pressure sensor, dynamically correct local permeability loss and achieve accurate compensation of leaching flux.

[0039] S04. Collect effluent from different depths of the soil column, determine the chromatographic peak areas of abamectin and fluopyram by high performance liquid chromatography, and simultaneously record the Raman signal intensity and redshift of the surface-enhanced Raman scattering sensor at each soil depth to quantitatively evaluate the free release rate of the drug carrier in situ.

[0040] S05. Input the particle size distribution, Zeta potential, Fourier transform infrared spectral matrix, peak area of ​​effluent at each depth and Raman signal data into the structure adaptive graph attention spatiotemporal evolution network. Through the adaptive jump connection mechanism, identify the non-covalent bond dissociation critical point, predict the longitudinal leaching curve and output the constitutive relationship of the drug-loaded body stability at each soil depth.

[0041] S06. Based on the leaching curves and stability constitutive relations output by the structural adaptive graph attention spatiotemporal evolution network, and combined with the measured concentration data of effluent from each soil layer, a quantitative evaluation system for soil physicochemical parameters and the mobility of nano-nematicides was established to complete the comprehensive determination of leaching performance.

[0042] Among them, the self-assembled drug body of nano-nematicides refers to the supramolecular aggregate formed by the spontaneous assembly of active ingredients such as abamectin or fluopyram with nanocarrier molecules by non-covalent bond driving forces (including hydrogen bonds, hydrophobic interactions, and tannic acid-metal coordination bonds). Its structural stability is affected by environmental ionic strength, pH and competitive adsorption of soil colloids.

[0043] Among them, particle size distribution refers to the frequency distribution of particle size in the nanocarrier suspension, which is measured by a dynamic light scattering instrument to characterize the uniformity of particle dispersion; Zeta potential refers to the potential at the sliding surface of the double electric layer on the particle surface, reflecting the electrostatic stability between particles; Fourier transform infrared spectral matrix refers to the range of 400–4000 nm. The absorbance data matrix collected within the wavenumber range was used to track the variation characteristics of non-covalent functional groups. The specific measurement range and threshold of the above three parameters were obtained through multiple batch preparation experiments combined with principal component analysis. The specific method was as follows: no less than 5 batches of drug-loaded materials were prepared repeatedly under the same formulation, and the mean and standard deviation of each parameter were statistically analyzed. The mean ± 2 times the standard deviation was used as the normal characterization interval.

[0044] Among them, high-frequency micro-pressure sensors refer to sensors with a response frequency of not less than 100Hz and a measurement range covering 0 to 50. The piezoresistive micro-sensor is used to capture transient fluctuations in pore water pressure in real time. The specific value of its embedding spacing of 1 to 5 cm is determined through preliminary experiments based on the total length of the soil column and the target resolution. The specific method is as follows: comparative leaching experiments are conducted by arranging sensors in the same soil column at spacings of 1 cm, 2 cm, 3 cm, and 5 cm, and the maximum spacing that can identify pressure change signals is taken as the final selection value.

[0045] Among them, the flexible fiber surface-enhanced Raman scattering sensor refers to embedding a flexible fiber probe with gold nanoparticles on its surface into a soil column. The Raman scattering signal of the target molecule is amplified by the local plasmon resonance effect generated by the gold nanoparticles, thereby detecting the intensity and redshift of the characteristic peak of tannic acid-metal coordination bond in situ without destroying the soil column structure. The method for determining the redshift threshold of the characteristic peak is as follows: standard solutions of drug carriers with different degrees of dissociation are prepared in advance, Raman spectra at each concentration are measured, and a standard curve of dissociation degree versus redshift is plotted. The redshift corresponding to a dissociation degree exceeding 20% ​​is taken as the dissociation critical threshold. This threshold is determined by the average of no less than three independent experiments.

[0046] The principle of the seepage mechanics random walk coupled with lattice Boltzmann transport algorithm (S-LBMA) is as follows: the soil pore structure is discretized into a three-dimensional lattice space, and the concentration distribution of nanoparticles and the initial velocity field of the fluid are initialized; at the micro level, a random walk model is used to simulate the random displacement of nanoparticles caused by Brownian motion; at the macro level, the lattice Boltzmann dynamics equations are applied to solve the Navier-Stokes equations of fluid in capillary channels through two core steps of collision and migration; at each time step, the algorithm calculates the adhesion probability between particles and pore walls based on fluid shear force and the strength of non-covalent coordination bonds on the surface of nanoparticles. If adhesion occurs, the flow resistance coefficient of the lattice point is updated and the local flow field is dynamically corrected; when local pores are blocked, the algorithm triggers flow field redistribution calculation and automatically adjusts the flow velocity of surrounding unblocked channels; the above process is iterated alternately until the concentration of nanoparticles in the effluent reaches dynamic equilibrium, thereby realizing the numerical simulation of the longitudinal leaching curve. This algorithm couples the solution of stochastic Brownian motion and deterministic hydrodynamic equations within the same time step, allowing the migration behavior of nanoparticles in the porous network to be simultaneously constrained by both thermal and hydrodynamic forces. This overcomes the limitation of traditional Darcy's law in describing the dynamic clogging process of non-covalent nanoaggregates, providing a physical mechanism to support the correction of leaching flux rather than purely empirical fitting. Consequently, it significantly improves the interpretability and prediction accuracy of anomalous cliffs and tail-impact phenomena in the leaching curve. Specifically, the threshold for calculating the adhesion probability is obtained by conducting at least five sets of penetration curve experiments under different ionic strength conditions in a standard quartz sand column, combined with DLVO theory to calculate the particle-wall interaction energy barrier, and by minimizing the sum of squared residuals between the measured and simulated penetration curves through parameter inversion.

[0047] Among them, the free release rate refers to the mass of active ingredient that dissociates from the self-assembled drug body and enters the soil pore water in a free state per unit time. The standard curve is calculated by substituting the Raman signal intensity attenuation rate at each soil depth into a pre-established standard curve. The standard curve is established by preparing 0.1–50... A series of free avermectin or fluopyram standard solutions were used to determine the intensity of the surface-enhanced Raman scattering characteristic peak of the gold nanoparticle probe at each concentration. The intensity-concentration relationship was fitted by linear regression, and the correlation coefficient was not less than 0.99 before it could be used for quantitative calculation.

[0048] Among them, the stability constitutive relation refers to the quantitative functional relationship between soil physicochemical parameters (including pH, ionic strength, organic matter content, and clay mineral type) and the probability of maintaining the integrity of the self-assembled structure of the nano-drug carrier at the corresponding soil depth. It is expressed by the fitting equation output by the structure adaptive graph attention spatiotemporal evolution network after completing multiple batches of training.

[0049] The Structural Adaptive Graph Attention Spatiotemporal Evolution Network (SAGT-SEN) is based on deep learning. Its specific structure is as follows: the network uses an improved graph attention network as its backbone, mapping the pore network topology of the porous soil medium to spatial graph nodes and the time step to temporal edges, forming a spatiotemporal heterogeneous graph. Each spatial node has a feature vector dimension of 16, containing five physicochemical features: pore water pressure, Raman signal intensity, effluent chromatographic peak area, Zeta potential projection value, and permeability correction coefficient. The remaining dimensions are completed through learnable embedding layers. The graph attention layer consists of four layers, each with eight attention heads. The weight matrix between neurons is stored using sparse tensors, and the sparsity threshold is determined using a dynamic graph pruning algorithm. Specifically, in each forward propagation, the absolute value of the attention coefficient is calculated, and values ​​below the dynamic threshold are removed. The edge weights are reset to zero, thereby achieving dynamic sparsity of the connection weights between neurons; the feature transfer between layers adopts a dual-channel structure with residual skip connections and adaptive jump connections in parallel. The residual skip connections are used for regular temporal layer iterations, and the adaptive jump connections are used when a feature spectrum mutation is detected (i.e., the Raman signal intensity decays by more than 100 kJ / m² within 3 consecutive time steps). Activation occurs at specific times, directly transmitting dissociation state information to the back-end dynamic prediction layer, bypassing the intermediate layer. The calculation of the attention coefficient incorporates physicochemical flow field constraints, specifically by multiplying the standard attention weight calculation by a rigid constraint factor composed of soil permeability and hydrodynamic shear force, ensuring that weight allocation is simultaneously constrained by both data feature correlation and physical constraints. The temporal unrolling step count for data loops is set to 1–20 steps, with the specific number determined online by spectral clustering technology dynamically adjusting the graph structure boundary connections based on the heterogeneity of the input data. The clustering results are synchronously fed back to the temporal unrolling step count controller, preventing prediction accuracy drops under non-uniform soil media with a fixed graph structure. Memory allocation employs a gradient checkpoint strategy, retaining only the forward activation value of the graph attention layer, while the intermediate layer activation value is... Recalculation is performed during backpropagation to reduce peak GPU memory usage. In memory allocation, the node feature matrix and edge index matrix are loaded in blocks, with each loaded node block being 128 bytes. CUDA stream allocation distributes graph convolution computation and attention coefficient sparsification computation to two independent CUDA streams for parallel execution, with event synchronization ensuring computational order between the two streams. In the allocation of different CUDA kernel functions, the node aggregation kernel function and the edge weight update kernel function are allocated to different thread blocks, with a thread block size of 256 bytes. In terms of layer allocation, the first four graph attention layers are responsible for spatial feature extraction, the fifth layer is a temporal attention layer, and the last layer is a dynamic prediction layer that outputs the stability probability of the drug-loaded material and the predicted free release rate at each soil depth. The network also includes a seepage dynamic compensation function. This function is used to adjust the number of attention head activations in each layer during the network inference phase. It is based on the rate of change of pore water pressure. Raman signal attenuation rate and effluent concentration gradient Three data points are used to calculate the dynamic compensation index for seepage. ,when At that time, the number of attention heads activated remained at 8, and the network reasoned with full precision; when At this time, the number of activated attention heads in layers 2 and 3 is reduced to 4, freeing up CUDA kernel function resources to accelerate the parallel solution of the lattice Boltzmann dynamics equations; when Furthermore, the number of activated attention heads in layers 1 to 4 is uniformly reduced to 2, while the number of temporal unfolding steps is increased to a maximum of 20 steps to improve the temporal resolution of congestion prediction; when At that time, the adaptive jump connection is activated and the number of attention heads in all intermediate layers is reduced to 1, directly passing the current dissociation state information to the dynamic prediction layer to ensure real-time response speed; among which Depend on , , After weighted summation, normalized using the sigmoid function to the nearest integer. The intervals were obtained, and each weight coefficient was determined after cross-validation on no fewer than 10 sets of standard leaching experimental data of different soil types. The steps for establishing the training dataset for the structure-adaptive graph attention spatiotemporal evolution network specifically include: collecting no fewer than 50 sets of standard leaching experimental data covering sandy loam, clay loam, and soils with varying organic matter content; each set of data includes the entire particle size distribution, Zeta potential, Fourier transform infrared spectral matrix, pore water pressure time series at each depth, surface-enhanced Raman scattering signal time series, and effluent chromatographic peak area time series; using the measured leaching curve and the actual free release rate at each depth as supervision labels; normalizing the original data, and using min-max normalization to map each feature to... The training, validation, and test sets are divided into three parts: a training set, a validation set, and a test set, with a ratio of 8:1:1. The specific steps for training the structured adaptive graph attention spatiotemporal evolution network include: using the Adam optimizer, with an initial learning rate set to... The learning rate decays to 0.5 times its original value when the validation set loss does not decrease for five consecutive training epochs. The loss function is a weighted sum of mean squared error loss and graph structure regularization term, with the graph structure regularization weights obtained through grid search. to The training batch size is set to 32, and the maximum number of training epochs is set to 200. An early stopping strategy is adopted, and training is terminated when the validation set loss does not improve within 15 consecutive training epochs. After training, the model accuracy is evaluated by two indicators: root mean square error and Pearson correlation coefficient on the test set. Both indicators must reach the preset qualified threshold before deployment. The qualified threshold is determined by statistically analyzing the 25th percentile of each indicator after leaving one out of all 50 experimental data sets.

[0050] The structure-adaptive graph attention spatiotemporal evolution network maps the porosity network topology to a spatiotemporal heterogeneous graph and incorporates physicochemical flow field constraints. This ensures that the network's feature extraction process not only relies on statistical data patterns but is also rigidly constrained by the physical mechanisms of soil seepage. The adaptive jump connection mechanism allows the network to bypass redundant temporal hierarchical iterations when it detects critical abrupt changes in non-covalent bond dissociation, directly transmitting dissociation state information to the prediction layer. This ensures both the depth of temporal modeling and rapid response to sudden physicochemical events. The dynamic graph structure adjustment driven by spectral clustering enables the model to adapt online to the spatial heterogeneity of non-homogeneous soil media, eliminating the systematic bias in prediction results caused by fixed graph structures when facing soil media heterogeneity. The seepage dynamic compensation function directly couples the allocation of network inference resources with the physical state of the leaching process. It automatically tilts computing power towards prediction accuracy when blockage events occur and releases computing power to accelerate numerical simulation during steady-state phases, thereby improving the real-time performance and physical interpretability of the leaching performance measurement scheme as a whole.

[0051] Optionally, the present invention also provides a method for forming a soil leaching performance testing system for nano-nematicides by means of a computer, wherein the computer is provided with a readable storage medium, the readable storage medium stores program instructions, and the program instructions are used to execute the above-described method when the computer is run.

[0052] The specific implementation of step S01 is as follows: Using tannic acid-metal coordination bonds as the main non-covalent driving force, abamectin or fluopyram is spontaneously assembled with nanocarrier molecules in an aqueous solution to obtain a suspension of self-assembled nano-nematicide. The particle size distribution, i.e., particle size distribution, in the suspension is measured using a dynamic light scattering instrument to characterize the uniformity of particle dispersion. The potential at the sliding surface of the double electric layer on the particle surface, i.e., the zeta potential, is measured using a potentiometer to reflect the electrostatic stability between particles. Fourier transform infrared spectroscopy is used in the range of 400–4000 nm. Absorbance data matrices were collected within the wavenumber range to track the variation characteristics of non-covalent functional groups. At least five batches of drug-loaded organisms were prepared repeatedly under the same formulation. The mean and standard deviation of each parameter were statistically analyzed. The mean ± 2 times the standard deviation was used as the normal characterization interval. Principal component analysis was combined to extract key feature dimensions, forming the basic characterization dataset for subsequent model input.

[0053] The specific implementation of step S02 is as follows: Select a standard soil column with representative particle size distribution, and embed a response frequency of not less than 100Hz and a measurement range covering 0-50 at intervals of 1-5cm along the longitudinal direction of the soil column. A piezoresistive high-frequency micro-pressure sensor is used to capture transient fluctuations in pore water pressure in real time. The specific value of the embedding spacing is determined through pre-experiments, namely, comparative leaching experiments are conducted by deploying sensors at spacings of 1cm, 2cm, 3cm, and 5cm in the same soil column, and the maximum spacing that can identify pressure change signals is selected as the final value. Flexible fiber surface-enhanced Raman scattering sensors are simultaneously deployed at the longitudinal gradient position. The sensor probe surface is modified with gold nanoparticles, and the Raman scattering signal of the characteristic peak of tannic acid-metal coordination bond is amplified by utilizing the local plasmon resonance effect. The method for determining the redshift threshold of the characteristic peak is as follows: standard solutions of drug-loaded materials with different degrees of dissociation are prepared in advance, Raman spectra at each concentration are measured, and a standard curve of dissociation degree versus redshift amount is plotted. The redshift amount corresponding to a dissociation degree exceeding 20% ​​is used as the dissociation critical threshold, which is determined by the average of no less than three independent experiments. The two types of sensors work synchronously to form a dual-channel in-situ monitoring system for pore water pressure and molecular structure.

[0054] The specific implementation of step S03 is as follows: The seepage mechanics random walk coupled with lattice Boltzmann transport algorithm (S-LBMA) discretizes the soil pore structure into a three-dimensional lattice space, initializing the concentration distribution of nanoparticles and the initial velocity field of the fluid. At the microscopic level, a random walk model is used to simulate the random displacement of nanoparticles caused by Brownian motion; at the macroscopic level, the lattice Boltzmann dynamics equations are applied to solve the Navier-Stokes equations for fluid in capillary channels through two core steps: collision and migration. At each time step, the algorithm calculates the adhesion probability between particles and pore walls based on the fluid shear force and the strength of non-covalent coordination bonds on the nanoparticle surface. If adhesion occurs, the flow resistance coefficient of that lattice point is updated and the local flow field is dynamically corrected. When local pores become blocked, the algorithm triggers flow field redistribution calculations and automatically adjusts the flow velocity in the surrounding unblocked channels. The threshold for calculating adhesion probability was determined by conducting at least five sets of penetration curve experiments under different ionic strength conditions in a standard quartz sand column. The particle-wall interaction energy barrier was calculated using DLVO theory, and parameter inversion was performed with the objective of minimizing the sum of squared residuals between the measured and simulated penetration curves. This process was iterated alternately until the concentration of the nanoparticles in the effluent reached a dynamic equilibrium, achieving accurate compensation and numerical simulation of the leaching flux.

[0055] The specific implementation of step S04 is as follows: During the leaching experiment, effluent samples from different depths of the soil column are collected at preset time intervals. High-performance liquid chromatography (HPLC) is used to quantitatively determine the peak areas of avermectin and fluopyram in each sample. The quantitative relationship between the peak area and the concentration of the active ingredient is pre-established using a standard curve. Simultaneously, the intensity and redshift of the characteristic wavelength Raman signal collected at each soil depth by a flexible fiber optic surface-enhanced Raman scattering sensor are recorded in real time. The free release rate is... The standard curve is calculated by substituting the Raman signal intensity attenuation rate at each soil depth into a pre-established standard curve. The standard curve is established by preparing 0.1–50... A series of free active ingredient standard solutions were prepared, and the intensity of the surface-enhanced Raman scattering characteristic peak of the gold nanoparticle probe at each concentration was measured. The intensity-concentration relationship was fitted by linear regression, and the correlation coefficient was not lower than 0.99 before it could be used for quantitative calculation. The above dual-channel data were acquired simultaneously to provide complete multi-dimensional data for subsequent model input.

[0056] The specific implementation of step S05 is as follows: The particle size distribution, Zeta potential, Fourier transform infrared spectral matrix, pore water pressure time series at each depth, Raman signal time series, and chromatographic peak area time series collected in steps S01 to S04 are mapped to a minimum-maximum normalized value. For missing data intervals, linear interpolation based on existing time-series data is used for completion. The training, validation, and test sets are then divided in an 8:1:1 ratio before being input into the Adaptive Graph Attention Spatiotemporal Evolution Network (SAGT-SEN). The network uses an improved graph attention network as its backbone, mapping the pore network topology to spatial graph nodes and the time step to temporal edges, forming a spatiotemporally heterogeneous graph. Four graph attention layers are set, and the attention coefficient calculation incorporates rigid constraint factors composed of soil permeability and hydrodynamic shear force. Adaptive jump connections are established when the Raman signal intensity decays beyond a jump attenuation threshold within three consecutive time steps. Activation occurs when dissociation state information is directly transmitted to the dynamic prediction layer for resolution. Linear clustering technology dynamically adjusts the boundary connections of the graph structure online, and a dynamic compensation function for seepage is applied. According to the seepage dynamic compensation index The number of attention heads and the number of temporal unfolding steps are dynamically adjusted. Training uses the Adam optimizer with an initial learning rate of... The loss function is a weighted sum of the mean squared error loss and the graph structure regularization term. An early stopping strategy is adopted, and the model accuracy is evaluated using both root mean square error and Pearson correlation coefficient. The pass / fail threshold is determined by statistically analyzing the 25th percentile of each index after leave-one-out cross-validation. The network outputs the constitutive relation of the drug-loaded body stability and the predicted values ​​of the longitudinal leaching curves at each soil depth.

[0057] The specific implementation of step S06 is as follows: Based on the longitudinal leaching curve and stability constitutive relation output in step S05, and combined with the measured concentration data of effluent from each soil layer collected in step S04, a systematic analysis is conducted on the quantitative relationship between soil physicochemical parameters such as pH, ionic strength, organic matter content, and clay mineral type, and the mobility of nano-nematicides at each soil layer depth. Using the fitted equation output from the stability constitutive relation as a bridge, the influence of each physicochemical parameter on the probability of the integrity of the self-assembled structure of the pesticide carrier is quantified into a specific functional relationship. The model prediction results are then verified using the measured leaching curves, ultimately forming a quantitative evaluation system for soil physicochemical parameters and nano-nematicide mobility covering the entire longitudinal depth of the soil column, completing the comprehensive determination of leaching performance. The output results of this evaluation system can be directly used for environmental risk assessment and application strategy optimization of nano-pesticides.

[0058] It should be noted that the key technologies of this invention include: First, dual-channel in-situ coupled monitoring of a high-frequency micro-pressure sensor and a flexible fiber surface-enhanced Raman scattering sensor. The former captures pressure surges caused by blockage events from a fluid dynamics perspective, while the latter quantifies the degree of non-covalent bond dissociation in real time from a molecular structure perspective. The two form complementary sensing dimensions, enabling abnormal events in the leaching process to have both macroscopic flow field evidence and microscopic structural evidence, breaking through the dual limitations of traditional offline detection in terms of temporal and spatial resolution. Second, the seepage mechanics random walk and lattice Boltzmann coupled transport algorithm couples the randomness of Brownian motion with deterministic Navier-Stokes fluid dynamics within the same time step, and dynamically calculates the adhesion probability and triggers flow field redistribution based on the non-covalent bond strength in each calculation step. This provides the numerical simulation with physical mechanism support for the dynamic blockage process, overcoming the limitations of Darcy's law and the continuous homogeneous medium assumption of traditional diffusion models. Third, the structure-adaptive graph attention spatiotemporal evolution network deeply integrates the data-driven capabilities of deep learning with the physical laws of soil seepage by constraining attention weights through physicochemical flow fields, responding to abrupt events through adaptive jump connections, driving dynamic graph structures through spectral clustering, and allocating computing power through a seepage dynamic compensation function. These three key technologies work synergistically to create a closed-loop feedback between sensing signals, physical simulation, and intelligent prediction, resolving the fundamental contradiction that a single technical approach cannot simultaneously achieve real-time perception, physical interpretability, and high-precision prediction.

[0059] It should be noted that when nano-nematicides are actually applied to non-uniform layered soils with significant differences in organic matter content and when the ionic strength of the leaching solution changes with depth, the non-covalent structure of the self-assembled drug-loaded body will dissociate at different rates at different soil depths. This results in a highly nonlinear spatial distribution of the effective drug loading concentration and the release rate of free active ingredients in each soil layer. Furthermore, the non-uniform layered structure also causes orders of magnitude differences in permeability between soil layers, further exacerbating the spatial heterogeneity of the leaching curve. The aforementioned technical problems arise because, in layered, non-uniform soils, the competitive adsorption intensity of organic matter on the drug carrier varies with depth, resulting in different environmental disturbances affecting the tannic acid-metal coordination bonds in different soil layers. Simultaneously, the ionic intensity gradient compresses the thickness of the electric double layer on the particle surface, reducing the absolute value of the zeta potential and weakening the electrostatic repulsion between particles with depth, potentially triggering aggregation and blockage at a certain depth. After blockage occurs, the local permeability drops sharply, leading to a redistribution of pore water pressure, which in turn alters the flow velocity field of adjacent soil layers, forming a dynamic cycle of blockage-unblocking-re-blockage. The entire process exhibits strong nonlinear and spatiotemporal coupling characteristics. The common solution to the aforementioned technical problems is to use segmented offline sampling combined with traditional solute transport models (such as convection-dispersion equations) for fitting. However, traditional convection-dispersion equations, based on homogeneous media and linear adsorption assumptions, cannot describe the time-varying particle size effects caused by the dynamic dissociation of non-covalent bonds. Some studies have introduced dual-porosity models to distinguish between preferential flow and matrix flow, but the parameters of dual-porosity models need to be independently calibrated layer by layer in layered heterogeneous media. As the number of soil layers increases, the calibration workload increases exponentially, and there is a lack of physical constraints between the parameters of each layer. Other studies have attempted to use machine learning models to directly fit leaching curves, but conventional neural networks use fixed graph structures or fully connected structures, which cannot adapt online to the topological changes caused by soil spatial heterogeneity, resulting in systematic prediction biases in layered heterogeneous media. This invention effectively solves this technical problem. The dynamic graph structure adjustment mechanism driven by phylogenetic clustering can perceive the heterogeneity of input data online and automatically adjust the boundary connections of the spatiotemporal heterogeneous graph, so that the graph topology matches the spatial heterogeneity of the current soil layer in real time, without the need to pre-calibrate parameters independently for each soil layer. The seepage mechanics random walk and lattice Boltzmann coupled transport algorithm dynamically corrects local permeability at each time step based on measured pore water pressure data, embedding the physical constraints of seepage between soil layers into the computational framework, thus ensuring the physical consistency of the velocity field redistribution process between adjacent soil layers. The structural adaptive graph attention spatiotemporal evolution network incorporates rigid constraint factors composed of permeability and shear force into the attention weights, enabling the network to automatically tilt attention allocation towards soil layer boundaries with significant permeability differences when processing layered non-uniform soil data, capturing cross-layer coupling effects; the adaptive jump connection is activated when a sudden change in cross-layer Raman signal is detected, sending dissociated state information across intermediate time series layers directly to the prediction layer, eliminating the response lag introduced by multi-layer iteration under sudden blockage events.The synergistic effect of the above three mechanisms enables the present invention to retain the interpretability of the physical model and possess the online adaptability of the data-driven model to complex spatial heterogeneity in the nonlinear leaching prediction scenario of layered non-uniform soil.

[0060] Specifically, the principle of this invention is:

[0061] The fundamental reason why this invention can solve the above-mentioned technical problems is that it constructs a closed-loop measurement system that is coupled from three levels: sensing, algorithm and model. Each level has designed a response mechanism for the core physicochemical event of dynamic dissociation of non-covalent bonds.

[0062] At the sensing level, a high-frequency micro-pressure sensor captures transient fluctuations in pore water pressure in real time with a response frequency of no less than 100Hz. When the self-assembled drug-loaded material agglomerates and clogs a certain soil layer, the pore water pressure at that layer will change abruptly within a millisecond timescale, allowing the sensor to accurately locate the longitudinal position of the blockage. Simultaneously, a flexible fiber optic surface-enhanced Raman scattering sensor utilizes the localized plasmon resonance effect generated by gold nanoparticles to amplify the characteristic Raman scattering signal of tannic acid-metal coordination bonds to a detectable level. By monitoring the redshift of characteristic peaks and the intensity decay rate, it directly reflects the real-time degree of dissociation of non-covalent bonds without interrupting the leaching experiment or damaging the soil column structure. The dual-channel in-situ signals provided by these two types of sensors simultaneously characterize the dynamic behavior of the drug-loaded material from both pressure field and molecular structure dimensions, overcoming the dual deficiencies of traditional offline detection in terms of temporal and spatial resolution.

[0063] At the algorithmic level, the seepage mechanics random walk and lattice Boltzmann coupled transport algorithm couples the Brownian motion random displacement with the Navier-Stokes equations for solution within the same time step. In each calculation step, the adhesion probability between particles and pore walls is calculated based on fluid shear force and non-covalent bond strength. Once adhesion occurs, the flow resistance coefficient of that lattice point is updated, triggering the redistribution calculation of the surrounding flow field. This mechanism physically corresponds to the real process of automatic switching of pore channels after non-covalent aggregate blockage, providing a clear physical mechanism to support leaching flux correction. This overcomes the fundamental deficiency of Darcy's law, which is based on the assumption of a continuous homogeneous medium and cannot describe dynamic blockage.

[0064] At the model level, the structure-adaptive graph attention spatiotemporal evolution network maps the pore network topology to a spatiotemporally heterogeneous graph. The attention weight calculation incorporates rigid constraint factors composed of soil permeability and hydrodynamic shear force, subjecting network feature extraction to both statistical data and physical constraints. Adaptive jump connections activate when the Raman signal continuously decays beyond a jump threshold, directly transmitting dissociation state information to the dynamic prediction layer, avoiding the response lag introduced by conventional residual connections under sudden events due to multi-level iterations. The spectral clustering-driven dynamic graph structure adjustment enables the model to adapt online to soil spatial heterogeneity, eliminating the systematic biases caused by fixed graph structures in non-uniform media. The seepage dynamic compensation function calculates the seepage dynamic compensation index based on the pore water pressure change rate, Raman signal decay rate, and effluent concentration gradient, dynamically allocating the number of attention heads and the number of temporal unfolding steps. During clogging events, computational power is prioritized for prediction accuracy, while in steady-state phases, computational power is released to accelerate numerical simulation, achieving real-time coupling of computational resources and leaching physical states. The synergistic effect of the above three levels enables the non-covalent bond dynamic dissociation process to be quantitatively characterized for the first time across the entire chain, from sensor signal acquisition to physical mechanism modeling and deep learning prediction, thus solving the core problem of the inability to quantitatively predict abnormal leaching flux in real time in principle.

[0065] The following provides a specific embodiment 1 of the present invention, and the specific implementation of each step in this embodiment 1 is described in detail below.

[0066] The specific implementation of step S01 is as follows: A self-assembled drug-loaded nano-nematicide is prepared; the particle size distribution is determined by dynamic light scattering; the zeta potential is determined by electrophoresis; and the particle size distribution is measured by Fourier transform infrared spectroscopy in the range of 400–4000 nm. Absorbance data matrices were collected within the wavenumber range. The specific method for establishing the normal ranges for the characterization parameters was as follows: at least five batches of drug delivery systems were prepared repeatedly under the same formulation, and the mean value for each parameter was calculated. and standard deviation The normal characterization interval is Particle size distribution is represented by a frequency distribution function, Zeta potential is represented numerically, and the Fourier transform infrared spectral matrix is ​​represented in matrix form. In the formula, The number of wavenumber sampling points. This represents the number of sampling batches. For the first The wave number at the th wave number The absorbance values ​​of the batch samples are dimensionless.

[0067] The specific implementation of step S02 is as follows: Construct a standard soil column leaching device, and embed a response frequency of not less than 100Hz and a measurement range covering 0-50 at intervals of 1-5cm along the longitudinal direction of the soil column. A high-frequency micro-pressure sensor was developed, and flexible fiber optic surface-enhanced Raman scattering (SERS) sensors were simultaneously deployed at longitudinal gradient locations. The specific method for determining the embedding spacing of the high-frequency micro-pressure sensor was as follows: comparative leaching experiments were conducted in the same soil column using sensors at spacings of 1 cm, 2 cm, 3 cm, and 5 cm. The maximum spacing capable of identifying pressure surge signals was selected as the final value. The method for determining the characteristic peak redshift threshold of the flexible fiber optic SERS was as follows: at least three sets of independently prepared standard solutions of the drug-loaded material with dissociation degrees of 0%, 10%, 20%, and 30% were prepared in advance. Raman spectra at each dissociation degree were measured, and a standard curve of dissociation degree versus redshift was plotted. The redshift corresponding to a dissociation degree exceeding 20% ​​was taken as the critical dissociation threshold. The unit is .

[0068] The specific implementation of step S03 is as follows: The seepage mechanics random walk coupled with lattice Boltzmann transport algorithm (S-LBMA) is initiated. The soil pore structure is discretized into a three-dimensional lattice space, and the nanoparticle concentration distribution and initial fluid velocity field are initialized. At the microscopic level, a random walk model is used to simulate the random displacement of nanoparticles, and at the macroscopic level, the lattice Boltzmann dynamics equations are applied to solve the Navier-Stokes equations. The iterative equations of the S-LBMA algorithm are specifically expressed as follows:

[0069] ;

[0070] In the formula, For a moment Location Direction The velocity distribution function on the surface is dimensionless. For the first A discrete velocity direction, dimensionless. For time steps, the unit is , Relaxation time, in units of , Let be the equilibrium distribution function, which is dimensionless. The evolution equation for nanoparticle concentration is:

[0071] ;

[0072] In the formula, For a moment Location Nanoparticle concentration, in units of , This is the partial derivative of concentration with respect to time, in units of... , For fluid velocity field, the unit is . , For convection terms, the unit is... , This is the diffusion coefficient, in units of... , For diffusion terms, the unit is... , For adhesion sediment source terms, the unit is... .

[0073] Each time step is executed According to fluid shear force and the strength of non-covalent coordination bonds on the surface of nanoparticles Calculate the adhesion probability between particles and pore walls. Its calculation expression is:

[0074] ;

[0075] In the formula, This is the adhesion threshold coefficient, in units of... Obtained through parameter inversion. Fluid shear force, unit: , The strength of non-covalent coordination bonds on the surface of nanoparticles, in units of , The damping factor is dimensionless and obtained through parameter inversion. The threshold for calculating the adhesion probability. and The breakthrough curves were obtained by conducting at least five sets of experiments under different ionic strength conditions in a standard quartz sand column. The specific steps were as follows: Ionic strengths of 0.01... 0.05 0.1 0.5 1.0 Nanoparticle suspensions were passed through quartz sand columns, and the relationship between nanoparticle concentration and pore volume in the effluent was measured as measured breakthrough curves. The particle-wall interaction energy barrier under various conditions was calculated using DLVO theory. Parameter inversion was performed with the objective function of minimizing the sum of squared residuals between the measured breakthrough curves and the lattice Boltzmann simulation curves. The calculated DLVO interaction energy barrier is:

[0076] ;

[0077] In the formula, The total interaction energy barrier, in units of , The distance between the particle and the wall surface, in units of , The van der Waals interaction energy is expressed in units of 1000 kJ / m². , It is the electrostatic interaction energy, with units of 1000 kJ / m². If adhesion occurs, update the local pore flow resistance coefficient for that grid point. The local flow field is dynamically corrected, and the expression is:

[0078] ;

[0079] In the formula, For a moment No. The pore flow resistance coefficient at each grid point, in units of , The increase in flow resistance caused by a single adhesion event, in units of When local pores become blocked, a flow field redistribution calculation is triggered, automatically adjusting the flow velocity in surrounding unblocked channels. This process iterates until the concentration of the nano-agent in the effluent reaches dynamic equilibrium.

[0080] The specific implementation method of step S04 is as follows: collect effluent from different depths of the soil column, and determine the chromatographic peak area of ​​the active ingredient by high performance liquid chromatography. The unit is The intensity of characteristic wavelength Raman signals from surface-enhanced Raman scattering sensors at various soil depths was recorded simultaneously. The unit is and redshift The unit is ,in The depth of the soil layer is expressed in units of 1000 m. , For time, the unit is In-situ quantitative assessment of the free release rate of the drug delivery vehicle. The calculation equation is as follows:

[0081] ;

[0082] In the formula, For a moment depth The free release rate at the location, in units of , Here, represents the partial derivative of the Raman signal intensity with respect to time, in units of . , This is the conversion factor, in units of... The standard curve was established by preparing concentrations of 0.1, 0.5, 1, 2, 5, 10, 20, and 50 μL. Eight concentration gradients of the free active ingredient standard solution were used. The intensity of the surface-enhanced Raman scattering characteristic peak of the gold nanoparticle probe was measured at each concentration, and the intensity-concentration relationship was fitted by linear regression.

[0083] ;

[0084] In the formula, The intensity of the Raman characteristic peak, in units of , The slope is expressed in units of... , This represents the concentration of the standard solution, in units of... , This is the intercept, in units of A correlation coefficient of at least 0.99 is required for quantitative calculations. (Conversion coefficient) The slope of the standard curve It is calculated by combining experimental parameters such as soil density.

[0085] The specific implementation of step S05 is as follows: Particle size distribution, Zeta potential, Fourier transform infrared spectral matrix, peak area of ​​effluent at each depth, and Raman signal data are used as network inputs to a Structure Adaptive Graph Attention Spatiotemporal Evolution Network (SAGT-SEN). Through an adaptive jump connection mechanism, the network identifies the non-covalent bond dissociation critical point, predicts the longitudinal leaching curve, and outputs the constitutive relationship of the drug-loaded body stability at each soil depth. In the structural design of the SAGT-SEN network, the eigenvector dimension of each spatial node is 16, including pore water pressure. The unit is Raman signal strength The unit is elution peak area The unit is Zeta potential projection value The unit is and permeability correction factor The graph is dimensionless and has five physicochemical features. The remaining dimensions are completed using learnable embedding layers. There are four graph attention layers, each with eight attention heads. The weight matrix between neurons is stored using sparse tensors. The sparsity threshold is determined using a dynamic pruning algorithm for graph structures, where attention coefficients with absolute values ​​below the dynamic threshold are selected. The edge weight is reset to zero. It is a dimensionless threshold, typically ranging from 0.01 to 0.1, which is dynamically adjusted during training.

[0086] Feature transfer between layers employs a dual-channel structure combining residual skip connections and adaptive jump connections. Residual skip connections are used for regular temporal layer iterations, while adaptive jump connections are used when the Raman signal intensity decays by more than [amount missing] within three consecutive time steps. Activated at time, among which The attenuation threshold is expressed in units of 1000 ppm. Typically, this value is taken as 0.5 times the standard deviation of the feature, directly transmitting the dissociation state information to the back-end dynamic prediction layer. The calculation of the attention coefficient incorporates physicochemical flow field constraints, and the standard attention weight is calculated by multiplying it by the rigid constraint factor. This factor is composed of soil permeability and hydrodynamic shear force, and its expression is:

[0087] ;

[0088] In the formula, The rigid constraint factor is dimensionless. For depth Soil permeability at the location, in units of , For reference penetration rate, the unit is... The value is the average of the permeability of different soil types, usually 100%. , For a moment depth The hydrodynamic shear force at the point, in units of , For reference shear force, the unit is... The value is taken as the average shear force measured in the experiment. The number of time-series unfolding steps is dynamically adjusted by spectral clustering technology based on the heterogeneity of the input data, ranging from 1 to 20 steps. The seepage dynamic compensation function is:

[0089] ;

[0090] In the formula, The compensation index is dimensionless and has a range of [missing information]. , The sigmoid function has the following expression: , , , These are weighting coefficients, dimensionless. The normalized rate of change of pore water pressure The normalized Raman signal attenuation rate, This represents the normalized effluent concentration gradient. Specifically, In the formula The change rate of pore water pressure is expressed in units of... , The reference pressure change rate is expressed in units of... The value is the average of the maximum pressure change rate in historical experiments. In the formula The Raman signal attenuation rate is expressed in units of 1000 m / s. , Reference decay rate, in units of The value is the average of the maximum decay rate in historical experiments. In the formula This represents the effluent concentration gradient, in units of... , Reference concentration gradient, in units of The value is taken as the average of the maximum concentration gradients in historical experiments. Weighting coefficients. , , The determination was made by cross-validation on no fewer than 10 sets of standard leaching experimental data from different soil types. The determination method was as follows: grid search was used. Under the constraints, we seek the weight combination that maximizes the model's prediction accuracy. When 8 attention heads are activated, the network performs full-precision inference; At this point, the number of activated attention heads in layers 2 and 3 is reduced to 4, freeing up CUDA kernel function resources for accelerating the parallel solution of the lattice Boltzmann dynamics equations; when At that time, the number of attention heads activated in layers 1-4 is uniformly reduced to 2, while the number of time sequence steps is increased to 20; when The adaptive jump connection is activated and the number of attention heads in all intermediate layers is reduced to 1.

[0091] The specific implementation of step S06 is as follows: Based on the leaching curves and stability constitutive relations output by the structure-adaptive graph attention spatiotemporal evolution network, and combined with the measured concentration data of effluent from each soil layer, a quantitative evaluation system for soil physicochemical parameters and the mobility of nano-nematicides is established. The stability constitutive relation is a quantitative functional relationship describing the relationship between soil physicochemical parameters and the probability of the nano-carrier maintaining the integrity of its self-assembled structure at the corresponding soil depth, expressed as:

[0092] ;

[0093] In the formula, For depth The structural integrity probability at a given location is dimensionless and ranges from 1 to 1. , For depth The soil pH value at that location is dimensionless. For depth Ionic strength at a given location, in units of , For depth Organic matter content at the location, in units of , For depth The clay mineral type parameters at the location are dimensionless. This is a nonlinear function fitted after training the SAGT-SEN network.

[0094] Free release rate refers to the mass of active ingredient dissociated from a self-assembled drug substance per unit time, measured in units of... Network training dataset establishment: At least 50 sets of standard leaching experimental data covering sandy loam, clay loam, and soils with varying organic matter content were collected. Each set of data included particle size distribution, Zeta potential, Fourier transform infrared spectral matrix, pore water pressure time series at each depth, surface-enhanced Raman scattering signal time series, and effluent chromatographic peak area time series. Using the measured leaching curves and actual free release rates at each depth as supervision labels, the raw data underwent min-max normalization, mapping each feature to... For each interval, missing data is imputed using linear interpolation based on existing time-series data. The training, validation, and test sets are divided in an 8:1:1 ratio. Network training: The Adam optimizer is used, with an initial learning rate set to... The learning rate decays to 0.5 times its original value when the validation set loss does not decrease for five consecutive training epochs. The loss function is a weighted sum of the mean squared error loss and the graph structure regularization term, expressed as:

[0095] ;

[0096] In the formula, The total loss function is dimensionless. The mean square error loss is dimensionless. For the sample size, For the first The network prediction value for each sample is dimensionless. For the first The true value of a sample, dimensionless. For regularization weights, which are dimensionless, a grid search is used. to Determined within the scope, The graph structure regularization term is dimensionless and is used to promote the network to learn sparse graph structures. The training batch size is set to 32, and the maximum training epochs are set to 200. An early stopping strategy is adopted, and training is terminated when the validation set loss does not improve within 15 consecutive training epochs. After training is completed, the model accuracy is evaluated by two indicators: root mean square error on the test set and Pearson correlation coefficient. Both indicators must reach the preset qualified threshold.

[0097] To better understand and implement this invention, the following is a specific application scenario of this invention, Example 2:

[0098] Using typical red soil from southern China as the experimental medium, an organic matter content of 23.6% was selected. Standard soil columns were prepared from clay loam soil, with an inner diameter of 5 cm and a filling height of 40 cm. The soil bulk density was stabilized at 1.32. Using tannic acid-metal coordination bonds as the main driving force, a bamectin-polylactic acid-glycolic acid copolymer nanocarrier was spontaneously assembled in deionized water to prepare a self-assembled drug suspension of nano-nematicides.

[0099] The particle size distribution of the drug-loaded material was determined using dynamic light scattering, with an average particle size of 182 nm and a polydispersity index of 0.18, indicating uniform particle dispersion. The measured Zeta potential was -31.2 mV, indicating that the drug-loaded material was in an electrostatically stable state. The Fourier transform infrared spectral matrix was in the range of 400–4000. Data collected within the range, at 1620 Characteristic absorption peaks of tannic acid phenolic hydroxyl groups and metal coordination bonds were detected nearby, confirming the integrity of the non-covalent structure. Five batches of drug delivery systems were prepared repeatedly under the same formulation, and the mean and standard deviation of each parameter were statistically analyzed. The normal characterization range was determined by the mean ± 2 times the standard deviation, as shown in Table 1.

[0100] Table 1. Statistical table of basic characterization parameters of self-assembled drug particles of nano-nematicides

[0101]

[0102] Nineteen high-frequency micro-pressure sensors were embedded longitudinally in the soil column at 2cm intervals, with a response frequency of 200Hz and a measurement range covering 0–50. Preliminary experiments confirmed that a 2cm spacing could identify pressure surge signals. Flexible fiber optic surface-enhanced Raman scattering sensors were simultaneously deployed at five longitudinal depth locations, with the probe surfaces modified with gold nanoparticles of average size 35nm. The characteristic peak redshift threshold was determined as follows: standard solutions of the drug-loaded material with dissociation degrees of 0%, 10%, 20%, 30%, and 50% were prepared, and Raman spectra at each concentration were measured. A standard curve was plotted comparing dissociation degree to redshift, with the redshift corresponding to a dissociation degree exceeding 20% ​​(measured at 8.3%). The value of 1 / 3 was used as the dissociation threshold and was confirmed by the average of three independent experiments.

[0103] The leaching experiment used distilled water at a stable flow rate of 1.2. Inject from the top of the soil column, with an initial concentration of 50 for the drug-carrying suspension. Continuous sample injection was performed over a period of 240 minutes. A coupled random walk and lattice Boltzmann transport algorithm was used to discretize the soil pore structure into a three-dimensional grid space. Pore water pressure data acquired in real-time by a high-frequency micro-pressure sensor was input into the algorithm to dynamically correct for local permeability loss at each grid point. At approximately 90 minutes into the experiment, the sensor located at a soil column depth of 18–22 cm detected a sudden increase in pore water pressure of approximately 6.8 kPa. The algorithm then triggers flow field redistribution calculations, increasing the flow velocity in channels surrounding the clogging layer to compensate for the leaching flux, such as... Figure 2The curves shown are the time-series variations of pore water pressure at different depths of the soil column.

[0104] Effluent from different depths was collected in segments every 10 minutes. The peak area of ​​avermectin was determined by high-performance liquid chromatography (HPLC), and the concentration was calculated based on a standard curve. The free release rate standard curve was prepared by preparing 0.1–50 μL solutions. A series of avermectin standard solutions were used to determine the intensity of the surface-enhanced Raman scattering characteristic peaks of gold nanoparticle probes at various concentrations. The linear regression correlation coefficient reached 0.997, meeting the quantitative requirements. The free release rate at each soil depth was calculated by substituting the real-time Raman signal attenuation rate into the standard curve, and the results are shown in Table 2.

[0105] Table 2. Statistical table of free release rates at different soil depths

[0106]

[0107] All the above data were input into a structure-adaptive graph attention spatiotemporal evolution network. The network maps the pore network topology to a spatiotemporal heterogeneous graph. The feature vector of each spatial node includes pore water pressure, Raman signal intensity, chromatographic peak area, Zeta potential projection value, and permeability correction coefficient. At 90 min of the experiment, the Raman signal intensity at depths of 18–22 cm decayed beyond the jump decay threshold in three consecutive time steps. The adaptive jump connection is then activated, directly transmitting the dissociation state information of that layer to the dynamic prediction layer, avoiding the response lag introduced by multi-layer time-series iteration. (Seepage dynamic compensation index) During congestion events, the hashrate increased to 0.78, triggering a hashrate redistribution strategy that reduced the number of activated attention heads from layers 1 to 4 to 2 and increased the number of temporal unfolding steps to 20, thereby improving the temporal resolution of congestion prediction. For example... Figure 3 As shown in the comparison (leaching curves at various depths and model predictions), the longitudinal leaching curves output by the network agree well with the measured concentration data, with a root mean square error of 0.043. The Pearson correlation coefficient was 0.986, both of which are higher than the pass threshold determined by leave-one-out cross-validation.

[0108] The stability constitutive relation results indicate that the organic matter content at a soil depth of 20 cm is 23.6%. The ionic strength of the solution is 0.08. Under these conditions, the probability of the drug-loaded organism maintaining the integrity of its self-assembled structure decreased to 0.31, providing a quantitative basis for predicting sudden dissociation in this soil layer. Combining the measured concentration data of the effluent from each soil layer with the model output results, a quantitative evaluation system for soil pH, ionic strength, organic matter content, and the mobility of the nano-nematicide was established under these clay loam conditions.

[0109] This invention represents a significant technological advancement over traditional offline column leaching assays. Traditional methods rely on segmented offline sampling and post-analysis, failing to detect the real-time location and rate of non-covalent bond dissociation. The results are temporal and spatial averages, leading to the inability to capture cliff-like and tail-impact phenomena in the leaching curve in real time, and lacking physical explanation. This invention, through dual-channel in-situ coupling of a high-frequency micro-pressure sensor and a flexible fiber optic surface-enhanced Raman scattering sensor, simultaneously quantifies the pressure characteristics of clogging events and the molecular structural characteristics of non-covalent bond dissociation. For the first time, it achieves millisecond-level in-situ characterization of the dynamic behavior of the drug-loaded substance without damaging the soil column structure, fundamentally overcoming the limitations of traditional methods in both temporal and spatial resolution. The seepage mechanics random walk and lattice Boltzmann coupled transport algorithm couples the randomness of Brownian motion with the fluid dynamics equations within the same time step, dynamically calculating adhesion probability and triggering flow field redistribution based on non-covalent bond strength. This provides a physical mechanism to support leaching flux correction, overcoming the fundamental deficiency of Darcy's law, which is based on the assumption of a continuous homogeneous medium and cannot describe the dynamic clogging process. The structure-adaptive graph attention spatiotemporal evolution network embeds the physical laws of soil seepage into the feature extraction process of deep learning by constraining attention weights and adjusting dynamic graph structure through physicochemical flow field constraints. This prevents the model prediction from producing systematic biases when facing soil spatial heterogeneity, and fundamentally solves the inherent limitations of conventional neural networks in predicting non-uniform media.

[0110] It should be noted that the variables involved in this invention are explained in detail in Tables 3 and 4.

[0111] Table 3. Variable Explanation Table (Part 1)

[0112]

[0113] Table 4. Variable Explanation Table (Part Two)

[0114]

[0115] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for determining the soil leaching performance of a nano-nematicide, characterized in that, Includes the following steps: A self-assembled drug-loaded nano-negative agent was prepared, and its particle size distribution, zeta potential, and Fourier transform infrared spectral matrix were measured. These parameters were used as the basic characterization data for subsequent model input. A standard soil column leaching device was constructed, and a high-frequency micro-pressure sensor was embedded in the longitudinal direction of the soil column. At the same time, flexible fiber surface-enhanced Raman scattering sensors were deployed at the longitudinal gradient position to monitor the pore water pressure and the Raman signal of the drug-loaded self-assembled structure in situ. The seepage mechanics random walk and lattice Boltzmann coupled transport algorithm is activated to discretize the soil pore structure into a three-dimensional grid space. Combined with the pore water pressure data collected in real time by the high-frequency micro-pressure sensor, the local permeability loss is dynamically corrected to achieve accurate compensation of leaching flux. Effluent from different depths of the soil column was collected, and the peak area was determined by high performance liquid chromatography. The Raman signal intensity and redshift of the flexible fiber surface-enhanced Raman scattering sensor at each soil depth were recorded simultaneously to quantitatively evaluate the free release rate of the drug carrier in situ. Particle size distribution, Zeta potential, Fourier transform infrared spectral matrix, peak area of ​​effluent at each depth and Raman signal data are input into the structure-adaptive graph attention spatiotemporal evolution network. The non-covalent bond dissociation critical point is identified through the adaptive jump connection mechanism, the longitudinal leaching curve is predicted and the constitutive relationship of drug-loaded stability at each soil depth is output. Based on the leaching curves and stability constitutive relations output by the structural adaptive graph attention spatiotemporal evolution network, and combined with the measured concentration data of effluent from each soil layer, a quantitative evaluation system for soil physicochemical parameters and the mobility of nano-nematicides was established, and the comprehensive determination of leaching performance was completed.

2. The method for determining the soil leaching performance of nano-nematicides according to claim 1, characterized in that, The self-assembling drug body of the nano-negative agent refers to a supramolecular aggregate formed by the spontaneous assembly of active ingredients and nanocarrier molecules driven by non-covalent bonds. The non-covalent bond driving forces include hydrogen bonds, hydrophobic interactions, and tannic acid-metal coordination bonds.

3. The method for determining the soil leaching performance of nano-nematicides according to claim 2, characterized in that, The particle size distribution was determined using a dynamic light scattering instrument; the Zeta potential reflects the electrostatic stability between particles; and the Fourier transform infrared spectral matrix is ​​in the range of 400–4000. The data were collected within the wavenumber range, and the normal characterization intervals of the three parameters were determined by repeatedly preparing at least a batch of threshold batches, and then using the mean ± 2 times the standard deviation.

4. The method for determining the soil leaching performance of nano-nematicides according to claim 3, characterized in that, The high-frequency micro-pressure sensor is a piezoresistive micro-sensor with a response frequency not lower than the frequency threshold and a measurement range covering the pore water pressure threshold range. The embedding spacing is determined through pre-experimentation to identify the maximum spacing that can detect pressure change signals.

5. The method for determining the soil leaching performance of nano-nematicides according to claim 4, characterized in that, The flexible fiber surface-enhanced Raman scattering sensor embeds a flexible fiber probe with gold nanoparticles on its surface into a soil column. It amplifies the Raman scattering signal of the target molecule by utilizing the local plasmon resonance effect. The characteristic peak redshift threshold is determined by the average of no less than three independent experiments.

6. The method for determining the soil leaching performance of nano-nematicides according to claim 5, characterized in that, The seepage mechanics random walk and lattice Boltzmann coupled transport algorithm uses a random walk model to simulate Brownian motion random displacement at the micro level, and solves the Navier-Stokes equations through two steps of collision and migration at the macro level. The adhesion probability threshold is obtained by combining no less than 5 sets of penetration curve experiments with DLVO theoretical parameter inversion.

7. The method for determining the soil leaching performance of nano-nematicides according to claim 6, characterized in that, The free release rate is The standard curve is calculated by substituting the Raman signal intensity attenuation rate at each soil depth into the standard curve. The linear regression correlation coefficient of the standard curve must be no less than the correlation coefficient threshold before it can be used for quantitative calculation.

8. The method for determining the soil leaching performance of nano-nematicides according to claim 7, characterized in that, The structure-adaptive graph attention spatiotemporal evolution network uses an improved graph attention network as its backbone, maps the pore network topology to spatial graph nodes, and maps the time step to temporal edges, forming a spatiotemporal heterogeneous graph. The feature vector of each spatial node includes pore water pressure, Raman signal intensity, chromatographic peak area, Zeta potential projection value, and permeability correction coefficient.

9. The method for determining the soil leaching performance of nano-nematicides according to claim 8, characterized in that, The structure-adaptive graph attention spatiotemporal evolution network has a total of 4 graph attention layers. The attention coefficient calculation incorporates a rigid constraint factor composed of soil permeability and hydrodynamic shear force. Interlayer feature transfer employs a dual-channel structure that combines residual jump connections and adaptive jump connections.

10. The method for determining the soil leaching performance of the nano-nematicide according to claim 9, characterized in that, The adaptive jump connection is activated when the Raman signal intensity is detected to have decayed beyond the jump attenuation threshold within three consecutive time steps, and the dissociation state information is bypassed from the intermediate layer and directly transmitted to the dissolution dynamic prediction layer.

Images