Pesticide effect activity coefficient quantitative test method under multivariable coupling of pesticide adjuvants

Abstract

The invention relates to the technical field of testing, and discloses a quantitative testing method for a pharmacodynamic activity coefficient under multivariable coupling of pesticide adjuvants. The method comprises the following steps: defining nine core variables in three types of environment, application and biology; eight variables are accurately regulated and controlled through a multi-chamber high-flux experiment system; adopting Latin hypercube sampling to generate an orthogonal experiment matrix and automatically executing an experiment; constructing a gradient boosting decision tree model based on the high-dimensional pharmacodynamic data; decoupling the marginal contribution of the assistant by using a Sharpley value algorithm; finally, a multi-dimensional response curved surface of the pharmacodynamic activity coefficient is generated through Gaussian process regression fitting. Accurate, dynamic and high-reproducibility quantification of the synergistic effect of the auxiliaries is achieved, and a scientific basis is provided for pesticide formula optimization and accurate application.

Description

Technical Field

[0001] This invention belongs to the field of testing technology, specifically relating to a method for quantitative testing of the efficacy activity coefficient of pesticide adjuvants under multivariate coupling. Background Technology

[0002] Pesticide adjuvants, as key components for enhancing efficacy and improving the physicochemical properties of pesticide solutions, are widely used in modern agricultural plant protection systems. Their core function lies in amplifying the bioavailability of active ingredients through mechanisms such as regulating surface tension, enhancing wettability and spreadability, and promoting penetration and absorption. However, the actual synergistic effect of adjuvants is not isolated but highly dependent on the combined effects of the application environment and operational parameters. Current pesticide efficacy evaluation systems are mostly based on experiments controlling a single or few variables, making it difficult to truly reflect the dynamic response characteristics under complex field conditions.

[0003] Nine variables—including temperature, humidity, light intensity, wind speed, pesticide pH, spray pressure, adjuvant concentration, crop leaf characteristics, and active ingredient type—constitute the core parameter set affecting the efficacy coefficient of pesticides. These variables do not act independently during actual application but exhibit a strongly nonlinear coupling relationship. For example, under high temperature and humidity conditions, the volatilization rate of adjuvants and their residence time on the leaf surface are mutually restrictive, while spray pressure alters droplet size distribution, thus affecting the uniformity of adjuvant deposition on different leaf structures. This multidimensional interaction effect fundamentally limits traditional static calibration methods in quantifying the contribution of adjuvants.

[0004] Existing technologies typically employ fixed-variable methods or orthogonal experimental designs to evaluate the efficacy of adjuvants, which are essentially static decoupling approaches. While these methods are applicable when the coupling strength between variables is low, they fail to effectively separate the interactive interference between variables in highly dynamic coupling scenarios, leading to a systematic overestimation or underestimation of the adjuvant's synergistic effect.

[0005] Especially under conditions of concentrated variable coupling frequency bands and intense interaction effects, measurement errors are significantly amplified, easily leading to misjudgments of adjuvant effectiveness and thus misleading formulation optimization and field application decisions. Therefore, there is an urgent need for a quantitative testing method for pharmacodynamic activity coefficients that can identify the strength of variable coupling and implement selective dynamic compensation, in order to achieve accurate attribution and scientific evaluation of adjuvant efficacy. Summary of the Invention

[0006] To address the above problems, this invention provides a method for quantifying the efficacy activity coefficient of pesticide adjuvants under multivariate coupling, comprising: Define and parameterize several core variables that affect pesticide efficacy; Construct a high-dimensional experimental dataset; Based on the high-dimensional experimental dataset, a gradient boosting decision tree prediction model is constructed and trained. The gradient boosting decision tree prediction model takes the parameter values ​​of the multiple core variables as input and the scalarized drug efficacy evaluation value as output. A set of decision trees is generated through sequential iterative training, so that the model can learn and characterize the nonlinear mapping relationship and higher-order interaction effect between the multiple core variables and the drug efficacy evaluation value. The trained gradient boosting decision tree prediction model is applied to each data point in the high-dimensional experimental dataset. An attribution analysis algorithm based on Shapley additive interpretation is used to decompose the predicted efficacy value of each data point and calculate the marginal contribution value of the adjuvant mass concentration variable to the predicted efficacy value. The marginal contribution values ​​of the adjuvant mass concentration variable from all data points are collected, and combined with the parameter values ​​of the other core variables, a multidimensional response surface function is generated by fitting using the Gaussian process regression method. The multidimensional response surface function is the pharmacodynamic activity coefficient of the adjuvant under the coupling effect of nine-degree variables, which measures the synergistic contribution of the adjuvant as a continuous function of the state of the other variables.

[0007] Preferably, the multiple core variables include a set of environmental parameters, a set of application technology parameters, and a set of biological target parameters; The set of environmental parameters includes ambient temperature, relative humidity, light intensity, and carbon dioxide concentration; The set of application technical parameters includes the median diameter of the droplet size spectrum, the density of drug solution deposition per unit area, and the mass concentration of adjuvants in the drug solution; The set of biological target parameters includes the age of the target pest and the surface wetting characteristics of the target crop leaves.

[0008] Preferably, constructing a high-dimensional experimental dataset includes the following steps: A multi-chamber high-throughput experimental system comprising multiple independent environmental control units was constructed, and sensors and actuators were deployed in each of the independent environmental control units. The Latin hypercube sampling algorithm is used to generate a multidimensional orthogonal experimental matrix within the preset range of values ​​of the multiple core variables. Each row of the multidimensional orthogonal experimental matrix corresponds to a unique set of variable parameter combination settings. The multi-chamber high-throughput experimental system is controlled to automatically and sequentially execute all experiments according to the set values ​​of the multidimensional orthogonal experimental matrix. In each experiment, the drug efficacy response image data of the target organism is collected at regular intervals through the high-resolution imaging module, and the scalarized drug efficacy evaluation value corresponding to the set values ​​of each set of variable parameters is calculated by combining the image processing algorithm, thereby forming a high-dimensional experimental dataset.

[0009] Preferably, the precise, independent, and dynamic control of the remaining variables among the multiple core variables, excluding the mass concentration of the adjuvants, specifically includes: Within each of the independent environmental control units, a closed-loop temperature control system composed of a Peltier semiconductor cooling array and a resistance wire heating array controls the ambient temperature within a preset range of ±0.1 degrees Celsius. The closed-loop humidity control system, consisting of an ultrasonic atomizing humidifier and a molecular sieve dehumidifier, controls the relative humidity of the environment within ±1% of the preset value. The light intensity and spectral distribution are precisely controlled by a full-spectrum light-emitting diode surface light source array and a pulse width modulation dimming controller. High-purity carbon dioxide and air are precisely mixed using a mass flow controller, and the carbon dioxide concentration is stabilized using a non-dispersive infrared sensor for closed-loop feedback control. A microporous vibration atomizing nozzle driven by piezoelectric ceramics changes the median diameter of the droplet size spectrum by adjusting the driving voltage frequency. By using a high-precision injection pump linked to the atomizing nozzle, the total volume of the sprayed liquid is precisely controlled, and the liquid deposition density per unit area is determined by combining the pre-calibrated spray coverage area. Using biological microscopes and image analysis software, the morphological characteristics of target pests are identified and graded to determine their age. The surface wetting characteristics of target crop leaves were measured using a contact angle measuring instrument and quantified as static water contact angle values.

[0010] Preferably, the calculated standardized efficacy evaluation value specifically includes: Before drug administration, baseline images of the target organisms in their initial state are collected, and after drug administration, a series of response images are collected at preset time intervals. A semantic segmentation algorithm based on deep learning is used to automatically identify and count the number of surviving and dead target organisms in an image; Calculate the efficacy evaluation value according to the formula: ; This is the efficacy evaluation value. The number of target organisms that died. This represents the initial total number of target organisms.

[0011] Preferably, the training process of the gradient boosting decision tree prediction model includes: Initialize a base learner containing only constant values; In each iteration, the negative gradient, i.e. the residual, between the current model prediction value and the actual efficacy evaluation value is calculated; Using the residual as the target, train a new weak learner, namely a depth-limited decision tree; The newly trained decision tree is added to the existing model with a preset learning rate to update the overall model's predictive ability. Repeat the above iterative process until the preset number of iterations is reached or the model's performance on the validation set no longer improves, ultimately forming a strong learner composed of multiple weighted decision trees.

[0012] Preferably, the calculation of the marginal contribution value using an attribution analysis algorithm based on Shapley additive interpretation specifically includes: For any specific experimental data point in the dataset, consider all subsets of variables that do not include the mass concentration of the adjuvant. For each subset of variables, calculate the predicted efficacy value of the model under the conditions of that subset, and the predicted efficacy value after adding the adjuvant mass concentration variable to the subset. The difference between the two predicted efficacy values, i.e. the predicted increment brought about by the introduction of the adjuvant mass concentration variable, is weighted and averaged. The weight is determined by the number of permutations and combinations of the subset of variables in all possible combinations of variables; The final weighted average increment is the Shapley value of the mass concentration variable of the adjuvant at that specific data point, which is its marginal contribution value.

[0013] Preferably, the multidimensional orthogonal experimental matrix generated by the Latin hypercube sampling algorithm ensures that there is exactly one sampling point in the equal probability interval of each variable, and that the sampling points between any two variables are uniformly distributed on the two-dimensional projection.

[0014] Preferably, the Gaussian process regression method uses the Marton kernel function as the covariance function and determines the optimal value of the length scale and signal standard deviation hyperparameter by maximizing the marginal likelihood function.

[0015] Preferably, the multi-chamber high-throughput experimental system comprises 16 mutually isolated environmental simulation chambers, and the inner wall of each environmental simulation chamber is coated with polytetrafluoroethylene to prevent drug residue contamination.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. It achieves precise decoupling and quantification of contribution. By combining the powerful nonlinear relationship fitting ability of the gradient boosting decision tree model with the strict attribution ability of the Shapley additive interpretation algorithm, this invention can accurately separate the net synergistic contribution of a single variable of pesticide adjuvant from complex, multi-variable interactively coupled efficacy data. Its quantification results exclude the synergistic or antagonistic effects brought about by changes in other variables, which greatly improves the accuracy and reliability of the test results.

[0017] 2. Testing efficiency and data throughput are fundamentally improved. The multi-chamber high-throughput automated experimental system adopted in this invention can execute a large number of experiments generated based on orthogonal experimental design in parallel or sequentially under unified control logic. Compared with traditional manual operation and single inefficient greenhouse or field experiments, its data acquisition speed and parameter space coverage are improved by orders of magnitude.

[0018] 3. It provides dynamic and contextualized activity evaluation. The final output of this invention is no longer a single, static activity coefficient value, but a multidimensional response surface function. This function reveals how the synergistic effect of adjuvants depends on specific combinations of environmental, application, and biological conditions, providing an unprecedented dynamic decision-making basis for pesticide formulation optimization and the development of precision application strategies.

[0019] 4. A standardized and highly reproducible testing paradigm has been established. This invention relies on a precisely controlled physical system and a deterministic algorithm process. The entire testing process is fully automated, eliminating the interference of human error and uncontrollable environmental factors, ensuring the high reproducibility of test results, and providing a technical foundation for establishing a unified industry standard for the evaluation of additives. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the overall technical solution architecture of the method for quantitative testing of pesticide activity coefficient under multivariate coupling of pesticide adjuvants proposed in this invention; Figure 2 This is a schematic diagram of the core principle framework of the adjuvant contribution attribution algorithm based on Shapley additive interpretation in this invention; Figure 3 This is a logical flowchart of the multidimensional orthogonal experimental design and high-throughput automated experimental execution in this invention; Figure 4 This is a flowchart illustrating the logical flow framework of gradient boosting decision tree prediction model construction and nonlinear coupling relationship learning in this invention. Figure 5 This is a schematic diagram of the multi-level interaction relationship and data flow between the multi-chamber high-throughput experimental system and the multi-dimensional environmental factors and application parameters collaborative control module in this invention; Figure 6 This is a logical flowchart of the response surface quantification and dynamic activity evaluation of the drug efficacy coefficient in this invention. Detailed Implementation

[0021] Please refer to Figures 1 to 6This invention provides a method for quantifying the efficacy activity coefficient of pesticide adjuvants under multivariate coupling. It aims to address the technical problem that traditional static calibration methods, under complex conditions of dynamic coupling and nonlinear interactions among nine key variables, cannot accurately separate and quantify the net contribution of pesticide adjuvants to the overall efficacy, leading to high measurement errors, poor reproducibility, and ultimately misjudgments of adjuvant effectiveness. The nine core variables include a set of environmental parameters, a set of application technique parameters, and a set of biological target parameters. The environmental parameter set includes ambient temperature, relative humidity, light intensity, and carbon dioxide concentration. The application technique parameter set includes the median diameter of the droplet size spectrum, the density of pesticide solution deposited per unit area, and the mass concentration of the adjuvant in the pesticide solution. The biological target parameter set includes the instar of the target pest and the surface wetting characteristics of the target crop leaves.

[0022] The method includes the following steps: S1 defines and parameterizes nine core variables that affect pesticide efficacy; S2, Construct a multi-chamber high-throughput experimental system containing multiple independent environmental control units, and deploy sensors and actuators in each of the independent environmental control units to achieve precise, independent and dynamic control of the remaining 8 variables among the 9 core variables, excluding the mass concentration of the adjuvant; S3, using the Latin hypercube sampling algorithm, generates a multidimensional orthogonal experimental matrix within the preset value range of the nine core variables. Each row of the multidimensional orthogonal experimental matrix corresponds to a unique set of variable parameter combination settings. S4, control the multi-chamber high-throughput experimental system to automatically and sequentially execute all experiments according to the set value of the multidimensional orthogonal experimental matrix. In each experiment, the high-resolution imaging module periodically collects drug efficacy response image data of the target organism, and combines the image processing algorithm to calculate the scalarized drug efficacy evaluation value corresponding to the set value of each set of variable parameters, thereby forming a high-dimensional experimental dataset. S5. Based on the high-dimensional experimental dataset, a gradient boosting decision tree prediction model is constructed and trained. The gradient boosting decision tree prediction model takes the parameter values ​​of the nine core variables as input and the scalarized drug efficacy evaluation value as output. A set of decision trees is generated through sequential iterative training, so that the model can learn and characterize the nonlinear mapping relationship and higher-order interaction effect between the nine core variables and the drug efficacy evaluation value. S6, the trained gradient boosting decision tree prediction model is applied to each data point in the high-dimensional experimental dataset. The attribution analysis algorithm based on Shapley additive interpretation is used to decompose the predicted efficacy value of each data point and calculate the marginal contribution value of the adjuvant mass concentration variable to the predicted efficacy value. S7. Collect the marginal contribution values ​​of the adjuvant mass concentration variable from all data points, and combine them with the parameter values ​​of the other 8 core variables. Use Gaussian process regression to fit and generate a multidimensional response surface function. The multidimensional response surface function is the pharmacodynamic activity coefficient of the adjuvant under the coupling effect of nine variables. It measures the synergistic contribution of the adjuvant as a continuous function of the states of the other 8 variables.

[0023] In step S1, nine core variables affecting pesticide efficacy are defined and parameterized. The ambient temperature range is set to 0°C to 60°C, with a step size of 0.5°C; the relative humidity range is set to 10% to 100%, with a step size of 2%; the light intensity range is set to 0 lux to 100,000 lux, with a step size of 500 lux; the carbon dioxide concentration range is set to 50 ppm to 1500 ppm, with a step size of 50 ppm; and the median diameter of the droplet size spectrum ranges to 1 micrometer to 300 micrometers, with a step size of 1 micrometer. The length is 5 micrometers; the density of pesticide deposition per unit area is set to range from 0.01 ml to 50 ml per square meter, with a step size of 0.05 ml per square meter; the mass concentration of adjuvants in the pesticide solution is set to range from 0% to 90%, with a step size of 0.1%; the instars of target pests are divided into five discrete levels: first instar, second instar, third instar, fourth instar, and fifth instar; the surface wetting characteristics of target crop leaves are quantified by static water contact angle, with a value range from 0 degrees to 180 degrees, with a step size of 1 degree. The above parameter ranges cover most field and greenhouse application scenarios and ensure that the coupling effect between variables can be fully stimulated and observed under a wide range of operating conditions.

[0024] In step S2, a multi-chamber high-throughput experimental system is constructed. This system comprises 16 isolated environmental simulation chambers, each with a volume of 1.5 cubic meters. The inner walls of each chamber are coated with polytetrafluoroethylene (PTFE) to prevent contamination from residual chemicals. Each environmental simulation chamber integrates a closed-loop temperature control subsystem, a closed-loop humidity control subsystem, a programmable full-spectrum illumination subsystem, and a closed-loop carbon dioxide concentration control subsystem. The closed-loop temperature control subsystem consists of a Peltier thermoelectric cooler array and a resistance wire heating array, working in conjunction with a platinum resistance temperature sensor to achieve stable control of the ambient temperature within a preset range of ±0.1 degrees Celsius.

[0025] The closed-loop humidity control subsystem consists of an ultrasonic atomizing humidifier and a molecular sieve dehumidifier, working in conjunction with a capacitive humidity sensor to precisely maintain the ambient relative humidity within a preset range of ±1%. The programmable full-spectrum lighting subsystem comprises a surface light source array of red, blue, green, and white light-emitting diode chips. A pulse-width modulation dimming controller adjusts the proportion of each color and the total light intensity to simulate natural sunlight or specific photoperiod conditions.

[0026] The carbon dioxide concentration closed-loop control subsystem uses a mass flow controller to proportionally mix high-purity carbon dioxide with compressed air, and a non-dispersive infrared sensor monitors the carbon dioxide concentration in the chamber in real time, providing feedback to adjust the gas injection rate and ensure that concentration fluctuations do not exceed ±10 ppm of the preset value. In addition, the system is equipped with a six-degree-of-freedom multi-axis robotic arm, whose end effector is fitted with a piezoelectric ceramic-driven microporous vibrating atomizing nozzle and a linked high-precision injection pump.

[0027] The piezoelectric ceramic atomizing nozzle continuously changes the median diameter of the droplet size spectrum by adjusting the driving voltage frequency within the range of 20 kHz to 200 kHz; the high-precision injection pump has a flow control accuracy of 0.1 microliters per second, and combined with the pre-calibrated spray coverage area, the density of drug deposition per unit area can be accurately set.

[0028] The instars of the target pests were automatically captured using a high-magnification biological microscope deployed at the bottom of the environmental simulation chamber. The instars were then classified and identified by a pre-trained convolutional neural network model, with a classification accuracy of no less than 98%. The surface wetting characteristics of the target crop leaves were automatically measured before the experiment using a miniature contact angle measuring instrument integrated on a robotic arm. The measurement was repeated three times, and the average value was taken as the final static water contact angle value.

[0029] In step S3, a multidimensional orthogonal experimental matrix is ​​generated using the Latin hypercube sampling algorithm. First, the value ranges of the nine core variables are divided into N equally probable intervals, where N is the preset total number of experiments, with a value of 2560. The Latin hypercube sampling algorithm ensures that there is exactly one sampling point in each interval for each variable, and that the sampling points between any two variables are uniformly distributed on the two-dimensional projection, thereby maximizing the coverage efficiency of the parameter space and minimizing spurious correlations between variables. The generated multidimensional orthogonal experimental matrix is ​​a 2560-row, 9-column numerical matrix, with each row corresponding to a unique set of variable parameter combinations, serving as the instruction sequence for subsequent automated experiments.

[0030] In step S4, the multi-chamber high-throughput experimental system is controlled to execute all experiments. Based on the set values ​​of the multidimensional orthogonal experimental matrix, the system automatically schedules the robotic arm to sequentially enter each environmental simulation chamber, adjusts the environmental parameters to the target values, and stabilizes for 30 minutes before drug administration begins. During drug administration, the piezoelectric ceramic atomizing nozzle sprays a drug solution containing a specified mass concentration of adjuvants according to the set median diameter of the droplet size spectrum and the drug deposition density per unit area. Before drug administration, the high-resolution imaging module acquires the initial baseline image of the target organism; after drug administration, response images are automatically acquired at three time points: 24 hours, 48 ​​hours, and 72 hours.

[0031] The image processing algorithm uses a semantic segmentation model based on the U-Net architecture to perform pixel-level segmentation and counting of surviving and dead target organisms in the image.

[0032] Calculate the efficacy evaluation value according to the formula: ; This is the efficacy evaluation value. The number of target organisms that died. This represents the initial total number of target organisms.

[0033] The efficacy evaluation value, combined with the corresponding 9 variable parameters, constitutes a high-dimensional experimental dataset containing a total of 2560 data points. Each data point contains 9 input features and one output label.

[0034] In step S5, a gradient boosting decision tree prediction model is constructed and trained. The model is initialized as a base learner containing only a constant value, which is the mean of all drug efficacy evaluation values. In each iteration, the negative gradient, i.e., the residual vector, between the current model's predicted values ​​for all samples in the training set and the actual drug efficacy evaluation values ​​is calculated. Using this residual vector as the target, a new weak learner is trained, which is a decision tree with a maximum depth of 6. The prediction result of the new decision tree is multiplied by a preset learning rate of 0.1 and added to the existing model to update the overall model's predictive ability.

[0035] The iterative process continues until the preset number of iterations (2000) is reached, or the mean squared error on the validation set fails to decrease for 50 consecutive iterations. The resulting strong learner, composed of a weighted average of 2000 decision trees, accurately fits the nonlinear mapping relationship and higher-order interaction effects between the nine core variables and the drug efficacy evaluation values. The model's coefficient of determination on the test set... A square value of not less than 0.95 indicates that it has excellent generalization ability.

[0036] In step S6, the marginal contribution value of the adjuvant mass concentration variable is calculated using an attribution analysis algorithm based on Shapley additive interpretation. This is applied to any specific data point in the high-dimensional experimental dataset. Let F be the trained gradient boosting decision tree prediction model. For the mass concentration of the auxiliary agent, Not included An arbitrary subset of variables. For every possible The computational model is in Predicted values ​​under the given conditions and in Predicted values ​​under the given conditions The difference between the two Indicates that in the known In the case of introducing The resulting predicted increment. This increment The weight is ,in To select from the remaining 8 variables The number of combinations of variables. Additive mass concentration variable. At data points Shapley value at the location For all possibilities corresponding The sum of the products of its weights is expressed mathematically as follows: ; The Shapley value This refers to the marginal contribution value of the adjuvant mass concentration variable at this data point. Its physical meaning is the average marginal contribution of the adjuvant mass concentration to the final predicted efficacy value, taking into account the order of all variable combinations.

[0037] In step S7, the marginal contribution values ​​of the adjuvants from all 2560 data points are aggregated. The dataset is divided into eight input-one output datasets, with i = 1 to 2560. The parameter values ​​of the remaining eight core variables corresponding to each data point are extracted to form a new eight-dimensional input-one-dimensional output dataset. Gaussian process regression is used to fit this dataset. The covariance function for Gaussian process regression is the Marting kernel function, which has the following form: ; in An eight-dimensional input vector and The Euclidean distance between them, where l is the length scale hyperparameter. Let be the signal standard deviation hyperparameter. The optimal value of the hyperparameter is determined by maximizing the marginal likelihood function. After fitting, a continuous multidimensional response surface function is obtained. ,in ,to These represent ambient temperature, relative humidity, light intensity, carbon dioxide concentration, median diameter of droplet size spectrum, pesticide deposition density per unit area, target pest instar, and static water contact angle of target crop leaves, respectively. The function G is the efficacy activity coefficient of the adjuvant under the coupling effect of nine variables, and its output value directly quantifies the net synergistic contribution provided by the adjuvant under given eight-dimensional environmental-application-biological conditions.

[0038] The pesticide adjuvant efficacy activity coefficient quantification testing system under nine-dimensional variables includes a multi-dimensional environmental factor and application parameter synergistic regulation module, an orthogonal experimental matrix generation module, a high-throughput automated experimental execution and data acquisition module, a dynamic efficacy nonlinear coupling prediction model construction module, an adjuvant contribution decomposition module based on Shapley additive interpretation, and a efficacy activity coefficient response surface quantification module.

[0039] The multidimensional environmental factors and application parameter coordinated control module consists of 16 environmental simulation chambers, a closed-loop temperature control subsystem, a closed-loop humidity control subsystem, a programmable full-spectrum illumination subsystem, a carbon dioxide concentration closed-loop control subsystem, a six-degree-of-freedom multi-axis robotic arm, a piezoelectric ceramic atomizing nozzle, a high-precision injection pump, a biological microscope, a contact angle measuring instrument, and corresponding sensors and controllers, used to accurately set and stably maintain 8 core variables.

[0040] The orthogonalization experiment matrix generation module is deployed in the central control computer. Its software program implements the Latin hypercube sampling algorithm, which automatically generates a multidimensional orthogonal experiment matrix based on the range of values ​​of nine variables and the number of experiments input by the user, and converts it into an instruction sequence that can be executed by the device.

[0041] The high-throughput automated experiment execution and data acquisition module integrates an industrial programmable logic controller, motion control card, image acquisition card, and large-capacity solid-state storage array. After receiving the instruction sequence, it coordinates each hardware unit to automatically execute the experimental process and synchronously stores the original image data and variable parameters.

[0042] The dynamic pharmacodynamic nonlinear coupling prediction model construction module, the adjuvant contribution decomposition module based on Shapley additive interpretation, and the pharmacodynamic activity coefficient response surface quantification module are all deployed in the central data processing server. This server is equipped with dual CPUs, 128 gigabytes of memory, and four GPUs, running a customized machine learning framework. The server communicates bidirectionally with the high-throughput automated experimental execution and data acquisition module via industrial Ethernet. After receiving raw experimental data, it sequentially performs model training, contribution decomposition, and response surface fitting, and outputs the final pharmacodynamic activity coefficient function to the user interface in the form of a mathematical expression or a high-dimensional lookup table.

[0043] The entire system's workflow is as follows: the user sets the value range and experimental scale of nine variables on the operating terminal; the orthogonalization experimental matrix generation module generates the instruction sequence; the high-throughput automated experimental execution and data acquisition module drives the multi-dimensional environmental factor and application parameter collaborative regulation module to complete all experiments and collect data; after receiving the data, the central data processing server sequentially calls the dynamic pharmacodynamic nonlinear coupling prediction model construction module, the adjuvant contribution decomposition module based on Shapley additive interpretation, and the pharmacodynamic activity coefficient response surface quantification module, finally outputting the adjuvant's pharmacodynamic activity coefficient response surface. This system achieves full automation from experimental design, execution, data acquisition to intelligent analysis, ensuring high accuracy, high throughput, and high reproducibility of the test results.

[0044] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "including," "comprise," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus.

[0045] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for quantitatively testing the efficacy activity coefficient of pesticide adjuvants under multivariate coupling, characterized in that, include: Define and parameterize several core variables that affect pesticide efficacy; Construct a high-dimensional experimental dataset; Based on the high-dimensional experimental dataset, a gradient boosting decision tree prediction model is constructed and trained. The gradient boosting decision tree prediction model takes the parameter values ​​of the multiple core variables as input and the scalarized drug efficacy evaluation value as output. A set of decision trees is generated through sequential iterative training, so that the model can learn and characterize the nonlinear mapping relationship and higher-order interaction effect between the multiple core variables and the drug efficacy evaluation value. The trained gradient boosting decision tree prediction model is applied to each data point in the high-dimensional experimental dataset. An attribution analysis algorithm based on Shapley additive interpretation is used to decompose the predicted efficacy value of each data point and calculate the marginal contribution value of the adjuvant mass concentration variable to the predicted efficacy value. The marginal contribution values ​​of the adjuvant mass concentration variable from all data points are collected, and combined with the parameter values ​​of the other core variables, a multidimensional response surface function is generated by fitting using the Gaussian process regression method. The multidimensional response surface function is the pharmacodynamic activity coefficient of the adjuvant under the coupling effect of nine-degree variables, which measures the synergistic contribution of the adjuvant as a continuous function of the state of the other variables.

2. The method for quantitative testing of pesticide activity coefficient under multivariate coupling of pesticide adjuvants according to claim 1, characterized in that, The multiple core variables include a set of environmental parameters, a set of application technology parameters, and a set of biological target parameters. The set of environmental parameters includes ambient temperature, relative humidity, light intensity, and carbon dioxide concentration. The set of application technology parameters includes the median diameter of the droplet size spectrum, the density of pesticide deposited per unit area, and the mass concentration of adjuvants in the pesticide solution. The set of biological target parameters includes the instar of the target pest and the surface wetting characteristics of the target crop leaves.

3. The method for quantitative testing of pesticide activity coefficient under multivariate coupling of pesticide adjuvants according to claim 2, characterized in that, Constructing a high-dimensional experimental dataset involves the following steps: A multi-chamber high-throughput experimental system comprising multiple independent environmental control units was constructed, and sensors and actuators were deployed in each of the independent environmental control units. The Latin hypercube sampling algorithm is used to generate a multidimensional orthogonal experimental matrix within the preset range of values ​​of the multiple core variables. Each row of the multidimensional orthogonal experimental matrix corresponds to a unique set of variable parameter combination settings. The multi-chamber high-throughput experimental system is controlled to automatically and sequentially execute all experiments according to the set values ​​of the multidimensional orthogonal experimental matrix. In each experiment, the drug efficacy response image data of the target organism is collected at regular intervals through the high-resolution imaging module, and the scalarized drug efficacy evaluation value corresponding to the set values ​​of each set of variable parameters is calculated by combining the image processing algorithm, thereby forming a high-dimensional experimental dataset.

4. The method for quantitative testing of pesticide activity coefficient under multivariate coupling of pesticide adjuvants according to claim 3, characterized in that, The precise, independent, and dynamic control of the remaining variables among the multiple core variables, excluding the mass concentration of the adjuvants, specifically includes: Within each of the independent environmental control units, a closed-loop temperature control system composed of a Peltier semiconductor cooling array and a resistance wire heating array controls the ambient temperature within a preset range of ±0.1 degrees Celsius. The closed-loop humidity control system, consisting of an ultrasonic atomizing humidifier and a molecular sieve dehumidifier, controls the relative humidity of the environment within ±1% of the preset value. The light intensity and spectral distribution are precisely controlled by a full-spectrum light-emitting diode surface light source array and a pulse width modulation dimming controller. High-purity carbon dioxide and air are precisely mixed using a mass flow controller, and the carbon dioxide concentration is stabilized using a non-dispersive infrared sensor for closed-loop feedback control. A microporous vibration atomizing nozzle driven by piezoelectric ceramics changes the median diameter of the droplet size spectrum by adjusting the driving voltage frequency. By using a high-precision injection pump linked to the atomizing nozzle, the total volume of the sprayed liquid is precisely controlled, and the liquid deposition density per unit area is determined by combining the pre-calibrated spray coverage area. Using biological microscopes and image analysis software, the morphological characteristics of target pests are identified and graded to determine their age. The surface wetting characteristics of target crop leaves were measured using a contact angle measuring instrument and quantified as static water contact angle values.

5. The method for quantitative testing of pesticide activity coefficient under multivariate coupling of pesticide adjuvants according to claim 4, characterized in that, The calculated standardized efficacy evaluation values ​​specifically include: Before drug administration, baseline images of the target organisms in their initial state are collected, and after drug administration, a series of response images are collected at preset time intervals. A semantic segmentation algorithm based on deep learning is used to automatically identify and count the number of surviving and dead target organisms in an image; Calculate the efficacy evaluation value according to the formula: ; This is the efficacy evaluation value. The number of target organisms that died. This represents the initial total number of target organisms.

6. The method for quantitative testing of pesticide activity coefficient under multivariate coupling of pesticide adjuvants according to claim 5, characterized in that, The training process of the gradient boosting decision tree prediction model includes: initializing a base learner that contains only constant values; In each iteration, the negative gradient, i.e. the residual, between the current model prediction value and the actual efficacy evaluation value is calculated; Using the residual as the target, train a new weak learner, namely a depth-limited decision tree; The newly trained decision tree is added to the existing model with a preset learning rate to update the overall model's predictive ability. Repeat the above iterative process until the preset number of iterations is reached or the model's performance on the validation set no longer improves, ultimately forming a strong learner composed of multiple weighted decision trees.

7. The method for quantitative testing of pesticide activity coefficient under multivariate coupling of pesticide adjuvants according to claim 6, characterized in that, The calculation of marginal contribution values ​​using the attribution analysis algorithm based on Shapley additive interpretation specifically includes: For any specific experimental data point in the dataset, consider all subsets of variables that do not include the mass concentration of the adjuvant. For each subset of variables, calculate the predicted efficacy value of the model under the conditions of that subset, and the predicted efficacy value after adding the adjuvant mass concentration variable to the subset. The difference between the two predicted efficacy values, i.e. the predicted increment brought about by the introduction of the adjuvant mass concentration variable, is weighted and averaged. The weight is determined by the number of permutations and combinations of the subset of variables in all possible combinations of variables; The final weighted average increment is the Shapley value of the mass concentration variable of the adjuvant at that specific data point, which is its marginal contribution value.

8. The method for quantitative testing of pesticide activity coefficient under multivariate coupling of pesticide adjuvants according to claim 7, characterized in that, The multidimensional orthogonal experimental matrix generated by the Latin hypercube sampling algorithm ensures that there is exactly one sampling point in the equal probability interval of each variable, and that the sampling points between any two variables are uniformly distributed on the two-dimensional projection.

9. The method for quantitative testing of pesticide activity coefficient under multivariate coupling of pesticide adjuvants according to claim 8, characterized in that, The Gaussian process regression method uses the Marton kernel function as the covariance function and determines the optimal values ​​of the length scale and signal standard deviation hyperparameters by maximizing the marginal likelihood function.

10. The method for quantitative testing of pesticide activity coefficient under multivariate coupling of pesticide adjuvants according to claim 9, characterized in that, The multi-chamber high-throughput experimental system comprises 16 mutually isolated environmental simulation chambers, each with a polytetrafluoroethylene coating on its inner wall to prevent contamination from residual drug solutions.

Images