Multi-factor adaptive precise regulation method for citrus clump root inoculation effect

Abstract

The application discloses a multi-factor adaptive precise regulation and control method for the inoculation effect of arbuscular mycorrhizal fungi of pome fruit trees, and relates to the technical field of intelligent agricultural fruit tree cultivation. The application collects multi-source data, quantifies the comprehensive importance of various factors on mycorrhizal infection rate, root growth, and plant growth promotion effect through a random forest model, and screens core regulation factors. A nested parameter optimization model is built, the inoculation effect is nonlinearly predicted through an XGBoost algorithm, and the initial parameters are obtained through multi-objective Pareto optimization of the inoculation dose, time, position, and method by an NSGA-II algorithm. The target parameters are dynamically corrected in combination with field environment and plant growth feedback. A pome fruit tree special compound microbial agent is prepared according to the parameters, and inoculation strategies are formulated for the three stages of seedling raising, young tree, and adult tree. The prediction model, optimization algorithm, and dynamic correction module are updated in layers and iteratively based on the field measured feedback data. The application realizes intelligent adaptive regulation and control of the whole process of arbuscular mycorrhizal fungi inoculation of pome fruit trees, and has wide application value.

Description

Technical Field

[0001] This invention relates to the field of smart agriculture fruit tree cultivation technology, specifically to a multi-factor adaptive method for precise control of the inoculation effect of arbuscular mycorrhizal fungi in pomelos. Background Technology

[0002] With the deepening of the concept of green and ecological cultivation of citrus fruits, arbuscular mycorrhizal fungi have been widely promoted and applied in the quality improvement cultivation of pomelo, orange, and tangerine trees due to their advantages such as forming a symbiotic system with the root system of fruit trees, enhancing the root system's nutrient absorption capacity, strengthening the plant's stress resistance, and improving the rhizosphere soil microenvironment. At present, the industry has carried out a large number of experimental studies on the selection of suitable symbiotic fungal strains for pomelos, the ratio of fungal inoculant carriers, and basic application methods. Some results have been transformed into standardized cultivation specifications and patented technologies, effectively solving basic problems such as insufficient strain compatibility and low field survival rate of fungal inoculants. However, in the actual implementation of large-scale orchards, there are significant spatial and temporal differences in the physicochemical properties of the rhizosphere soil of citrus, the field temperature, light and water environment, the physiological state of the tree at different growth stages, and cultivation management measures. Under the combined action of different factors, the efficiency of mycorrhizal infection and the plant growth promotion effect will be directly affected. It is difficult to adapt to complex field conditions by relying solely on fixed inoculation dosage and single application mode. This can easily lead to practical problems such as waste of inoculants, unstable mycorrhizal symbiosis rate, and large fluctuations in the control effect. Therefore, the industry has carried out a lot of related technical research on a complete set of control technologies for quantitative analysis of multiple influencing factors, intelligent optimization of inoculation parameters, and dynamic adaptive correction in the field.

[0003] Chinese patent (publication number CN110892845A) discloses a method for improving the quality of citrus fruit by inoculating arbuscular mycorrhizal fungi. This method involves preparing a mixed inoculum by combining three arbuscular mycorrhizal fungi strains and applying it directly to the root system of citrus plants. The method relies on the symbiotic effect of the fungi to improve the efficiency of nutrient absorption in the tree and thus improve the intrinsic quality of the fruit. However, this method limits the fixed ratio of fungi and the root application method, and does not conduct quantitative analysis of multi-dimensional influencing factors such as soil, plant, and management. It cannot make differentiated adjustments to the inoculation plan according to the different conditions in the field, and the parameter adaptability is poor.

[0004] Chinese patent (publication number CN121569695A) discloses a method for cultivating container mycorrhizal seedlings of citrus. This method combines arbuscular mycorrhizal fungi with plant hydrolyzed protein to complete the mycorrhizal inoculation and cultivation of seedlings in a seedling substrate, thereby improving the root development level and transplant survival rate of container seedlings. However, this method is only suitable for static cultivation of container seedlings in the seedling stage and cannot cover the entire growth cycle of saplings and mature trees. At the same time, it lacks optimization and screening of multiple parameters such as inoculation dosage, inoculation location, and application frequency, making it unsuitable for dynamic control in the field of mature pomelo trees.

[0005] Chinese patent (publication number CN114303811A) discloses a method for reducing phosphorus non-point source pollution in navel orange orchards based on AM fungi. This method reduces phosphorus loss from orchard soil by investigating the native arbuscular mycorrhizal fungal community in the orchard and combining it with a vegetation interception system. It achieves non-point source pollution control through the nutrient absorption function of mycorrhizal fungi. However, the core objective of this patent is water and soil environment management, and it cannot optimize inoculation parameters for multiple objectives, making it difficult to achieve precise control guided by inoculation effect.

[0006] In summary, existing arbuscular mycorrhizal inoculation techniques for pomelo and closely related citrus trees are insufficient to meet the needs of modern pomelo orchards for refined, adaptive, and long-term stable mycorrhizal symbiotic cultivation. Therefore, a multi-factor adaptive method for precise control of the inoculation effect of pomelo arbuscular mycorrhizal inoculation is proposed. Summary of the Invention

[0007] To address the aforementioned technical issues, this application discloses a multi-factor adaptive method for precisely regulating the inoculation effect of arbuscular mycorrhizal fungi in citrus, specifically including:

[0008] Collect multi-source heterogeneous data from pomelo planting areas;

[0009] A multi-factor importance analysis model was constructed, and the multi-source heterogeneous data was input into the multi-factor importance analysis model. By analyzing the influence relationship between different data factors and the inoculation effect of arbuscular mycorrhizal inoculation of citrus, the core regulatory factors affecting the inoculation effect of citrus arbuscular mycorrhizal inoculation were screened.

[0010] A nested inoculation parameter optimization model is constructed. The nested inoculation parameter optimization model takes in a core regulatory factor, predicts the arbuscular mycorrhizal formation effect under different combinations of inoculation parameters, and performs multi-objective optimization on inoculation dosage, inoculation time, inoculation location and inoculation method based on the prediction results, and outputs initial inoculation parameters that meet the preset optimization conditions.

[0011] Based on environmental change data and plant growth feedback data during the inoculation process, the initial inoculation parameters are dynamically adjusted to obtain the target inoculation parameters;

[0012] A compound arbuscular mycorrhizal fungal agent specifically for citrus was prepared according to the target inoculation parameters, and corresponding inoculation strategies were matched according to different growth stages of citrus.

[0013] After completing the arbuscular mycorrhizal inoculation, data on the growth status of pomelo plants, changes in the rhizosphere environment, and the effect of mycorrhizal formation were continuously collected. The nested inoculation parameter optimization model and the process parameters for dynamically adjusting the initial inoculation parameters were iteratively updated regularly.

[0014] Preferably, the multi-factor importance analysis model specifically involves: normalizing the collected multi-source heterogeneous data to obtain a standardized factor dataset;

[0015] The standardized factor dataset is input into the pre-trained multi-decision tree random forest sub-model, and the outputs are mycorrhizal infection rate, root growth status, and plant growth promotion effect.

[0016] The out-of-bag data average precision reduction method was used to calculate the independent perturbation error of each input data factor on each inoculation effect index. Based on the preset perturbation weights, the weighted fusion was performed to obtain the comprehensive importance of multiple targets.

[0017] Based on a preset importance threshold, data factors with importance scores higher than the threshold are identified as core regulatory factors, thus completing factor screening.

[0018] Preferably, the multi-decision tree random forest sub-model is a parallel multi-decision tree ensemble model constructed based on the idea of ​​ensemble learning. It adopts a multi-output parallel regression architecture and contains multiple classification and regression decision trees with independent structures, random feature sampling, and random sample sampling. It can simultaneously fit the correlation between input factors and mycorrhizal infection rate, root growth status, and plant growth promotion effect.

[0019] During pre-training, all standardized multi-source heterogeneous data factors are used as common input features. Bootstrap sampling with replacement is used to randomly sample the training set samples and random subspace sampling is performed on the input features. Simultaneously, a parallel regression training task is established to create three-dimensional output labels for mycorrhizal infection rate, root growth status score, and plant growth promotion effect score. The splitting threshold and branch weight of each decision tree node are iteratively optimized until the overall prediction error of the model converges to the preset accuracy.

[0020] Preferably, the nested vaccination parameter optimization model specifically includes a vaccination effect prediction layer and a multi-objective parameter optimization layer.

[0021] The inoculation effect prediction layer uses a gradient boosting tree prediction model to predict the arbuscular mycorrhizal formation effect under different combinations of inoculation parameters based on the input core regulatory factors.

[0022] The multi-objective parameter optimization layer performs multi-objective optimization on the inoculation dosage, inoculation time, inoculation location and inoculation method based on the prediction results. The optimization objectives are mycorrhizal infection rate, plant growth promotion effect, inoculant usage cost and environmental adaptability. The layer outputs initial inoculation parameters that meet the preset optimization conditions.

[0023] Preferably, the inoculation effect prediction layer specifically employs the XGBoost gradient boosting tree prediction algorithm, using the screened core regulatory factors as input features and mycorrhizal infection rate, root growth score, and plant growth promotion effect score as prediction outputs.

[0024] The prediction algorithm uses a second-order Taylor expansion to approximate the objective function, combined with L2 regularization to allocate leaf weights, as shown in the formula:

[0025]

[0026]

[0027] in, Let be the objective function. The first gradient of the loss function, The loss function is the second-order Hessian matrix. For the first Tree prediction output, For tree structure regularization terms; For the first Optimal prediction weights for each leaf node For the first The set of all training samples covered by each leaf node. The L2 regularization coefficient;

[0028] By performing parallel operations on multiple decision trees, accumulating the prediction results of each tree, and outputting multi-dimensional prediction values ​​of vaccination effect, the system can effectively achieve the desired results.

[0029] Preferably, the multi-objective parameter optimization layer specifically employs the NSGA-II non-dominated sorting genetic algorithm, with inoculation dosage, inoculation time, inoculation location, and inoculation method as optimization variables, and with the multi-objective optimization functions of maximizing mycorrhizal infection rate, maximizing plant growth promotion effect, minimizing inoculant usage cost, and maximizing environmental adaptability, to optimize parameters.

[0030] During parameter optimization, an initial population containing several sets of inoculation parameter combinations is randomly generated, and the individuals in the population are encoded. Based on four optimization objectives, the individuals in the population are stratified and non-dominatedly sorted to define the Pareto optimal level. The crowding distance between individuals in the same level is calculated to ensure population diversity. High-quality individuals are selected based on an elite retention strategy, and genetic crossover and mutation operations are performed to generate the offspring population. The parent and offspring populations are merged, and the non-dominated sorting and selection are repeated until a preset number of iterations is reached. Parameter combinations that meet a preset threshold are selected from the Pareto optimal solution set, and the initial inoculation parameters are output.

[0031] Preferably, the dynamic adjustment of the initial inoculation parameters specifically involves: acquiring environmental change data and plant growth feedback data during the inoculation process; and, based on the environmental change data and plant growth feedback data, employing a deviation accumulation adaptive correction algorithm, combined with deviation threshold damping constraints, time-series cumulative error compensation, and an adaptive weight update mechanism, dynamically correcting the inoculation parameters to obtain the target inoculation parameters, as shown in the formula:

[0032]

[0033] in, The corrected target inoculation parameters, These are the initial optimal inoculation parameters output by the NSGA-II algorithm; This represents the cumulative value of environmental deviation over time. This represents the cumulative value of plant growth feedback deviation over time. These are the sensitivity adjustment coefficients for environmental deviation and growth deviation, respectively. It is a hyperbolic tangent damping function; These are the adaptively adjusted weights that are dynamically updated with the iteration sequence, and the formula is:

[0034]

[0035] in, These are the initial baseline weights, This represents the number of model iterations. , The first Real-time environmental deviations and plant growth feedback deviations at each time-series sampling point.

[0036] Preferably, the preparation of the pomelo-specific compound arbuscular mycorrhizal fungal agent specifically involves: screening arbuscular mycorrhizal fungal strains suitable for the rhizosphere environment of pomelo according to the target inoculation parameters, including strain compatibility type, effective viable count, and agent ratio; mixing and fermenting these strains with an organic carrier, water-retaining agent, and rhizosphere growth promoter to prepare a specific compound agent suitable for different pomelo growth stages.

[0037] Preferably, the inoculation strategy matched according to different growth stages of pomelo is as follows: In the seedling stage: with the core goal of rapid root germination and formation, the shallow root inoculation location is determined, and a high-frequency, low-dose inoculation frequency is set to ensure that the seedling roots can quickly infect and symbiotically grow.

[0038] In the sapling stage, with the core objective of horizontal and vertical root expansion, the application method of fungicide is adjusted to combine ring application and trench application to adapt to the root growth range of saplings.

[0039] During the mature tree stage, the core objectives are to enhance the plant's resistance to adverse conditions and stabilize its vigor. Based on the soil environment and plant growth, the amount of inoculant supplementation is dynamically adjusted upwards or downwards, and the inoculant supplementation cycle is extended or shortened to maintain the long-term mycorrhizal symbiotic effect.

[0040] Preferably, the iterative update of the nested inoculation parameter optimization model and the process parameters for dynamically adjusting the initial inoculation parameters specifically involves: using the latest field-measured core regulatory factors as input and the actual inoculation effect as a label, and employing an incremental iterative training method to update the inoculation effect prediction layer of the nested inoculation parameter optimization model; using the updated XGBoost prediction model as a fitness evaluation function, and updating the constraint threshold and weight preference of the multi-objective parameter optimization layer based on the latest effect feedback data; and updating the adaptive weight decay coefficient in the dynamic adjustment process round by round based on the time-accumulated environmental deviation and plant growth deviation, and correcting the hyperbolic tangent nonlinear damping interval threshold based on long-term feedback data.

[0041] Compared with the prior art, the technical solution of this application has the following technical effects:

[0042] This invention collects four types of multi-source heterogeneous data: soil, plant physiology, inoculant characteristics, and cultivation management. Based on a random forest model and an out-of-bag data error perturbation method, it quantifies the comprehensive contribution of each factor to the inoculation effect, objectively screens core regulatory factors, and provides a reliable data foundation for subsequent parameter optimization.

[0043] This invention constructs a nested two-layer parameter optimization model. It uses the XGBoost algorithm to mine the complex nonlinear mapping relationship between core factors and mycorrhizal infection effect to achieve effect prediction. Then, it uses the NSGA-II multi-objective genetic algorithm to find the Pareto optimal solution for the four inoculation parameters with the objectives of infection effect, growth promotion ability, inoculant cost, and environmental adaptability, which greatly improves the scientificity and economy of the inoculation program.

[0044] This invention utilizes a nonlinear adaptive correction model with hyperbolic tangent damping constraints and time-series cumulative deviation to dynamically correct initial inoculation parameters based on real-time field environmental fluctuations and pomelo tree growth feedback. This enables real-time adaptive fine-tuning of the inoculation strategy according to field conditions, adapting to the hysteresis response characteristics of pomelo trees at different growth stages.

[0045] This invention designs inoculation methods, application locations, and application frequencies differently according to the three growth cycles of seedlings, saplings, and mature trees, and customizes a special compound arbuscular mycorrhizal fungicide for citrus based on the optimized parameters, so as to achieve a staged and precise match between the fungicide product and the field application strategy.

[0046] This invention enables the entire control system to have continuous self-learning and self-optimization capabilities by using a stratified incremental update prediction model based on long-term field measurement feedback data, multi-objective optimization constraints, and nonlinear correction weight coefficients.

[0047] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the preferred embodiments of this application are described in detail below with reference to the accompanying drawings.

[0048] The above and other objects, advantages and features of this application will become more apparent to those skilled in the art from the following detailed description of specific embodiments in conjunction with the accompanying drawings. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In all drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.

[0050] Based on the description of the figures and their corresponding technical content in the document, the titles of the figures are as follows:

[0051] Figure 1 A flowchart illustrating the overall process of a multi-factor adaptive method for precisely controlling the inoculation effect of arbuscular mycorrhizal fungi in citrus.

[0052] Figure 2 This is a diagram of the architecture of a multi-factor importance analysis model.

[0053] Figure 3 Architecture diagram of the nested inoculation parameter optimization model;

[0054] Figure 4 Comparison of typical fields of view under a microscope for trypan blue chromosomes in the root system of pomelo trees in each treatment group;

[0055] Figure 5 A comparative chart of the core indicators of mycorrhizal colonization for each method;

[0056] Figure 6 A comparison chart of pomelo tree growth and physiological stress resistance indicators;

[0057] Figure 7 The indicators for improving the rhizosphere soil microenvironment for each method;

[0058] Figure 8 A comparison chart of the system's long-term adaptive stability indices for each method;

[0059] Figure 9 This is a comparison chart of the economic indicators of each method. Detailed Implementation

[0060] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. In the following description, specific details such as specific configurations and components are provided merely to help fully understand the embodiments of this application. Therefore, those skilled in the art should understand that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this application. In addition, for clarity and brevity, descriptions of known functions and structures are omitted in the embodiments.

[0061] It should be understood that the phrase "an embodiment" or "this embodiment" throughout the specification means that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of this application. Therefore, "an embodiment" or "this embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments.

[0062] Furthermore, reference numerals and / or letters may be repeated in different examples within this application. Such repetition is for the purpose of simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or settings discussed.

[0063] In this article, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can mean: A exists alone, B exists alone, and A and B exist simultaneously. The term " / and" in this article describes another type of relationship between related objects, indicating that two relationships can exist. For example, A / and B can mean: A exists alone, and A and B exist alone. In addition, the character " / " in this article generally indicates that the related objects before and after it are in an "or" relationship.

[0064] In this article, the term "at least one" is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, "at least one of A and B" can mean: A exists alone, A and B exist simultaneously, or B exists alone.

[0065] It should also 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 "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion.

[0066] Example 1 describes a multi-factor adaptive method for precisely controlling the inoculation effect of arbuscular mycorrhizal fungi in citrus, such as... Figure 1 As shown, it specifically includes:

[0067] Collect multi-source heterogeneous data from pomelo planting areas;

[0068] A multi-factor importance analysis model was constructed, and heterogeneous data from multiple sources were input into the model. By analyzing the influence relationship between different data factors and the inoculation effect of arbuscular mycorrhizal fungi in citrus, the core regulatory factors affecting the inoculation effect of citrus arbuscular mycorrhizal fungi were screened.

[0069] A nested inoculation parameter optimization model is constructed. The model takes core regulatory factors as input, predicts the arbuscular mycorrhizal formation effect under different combinations of inoculation parameters, and performs multi-objective optimization on inoculation dosage, inoculation time, inoculation location and inoculation method based on the prediction results, and outputs the initial inoculation parameters that meet the preset optimization conditions.

[0070] Based on environmental change data and plant growth feedback data during the inoculation process, the initial inoculation parameters are dynamically adjusted to obtain the target inoculation parameters;

[0071] A compound arbuscular mycorrhizal fungal agent specifically for citrus was prepared based on the target inoculation parameters, and corresponding inoculation strategies were matched according to different growth stages of citrus.

[0072] After completing the arbuscular mycorrhizal inoculation, data on the growth status of pomelo plants, changes in the rhizosphere environment, and the effect of mycorrhizal formation were continuously collected. The nested inoculation parameter optimization model and the process parameters for dynamically adjusting the initial inoculation parameters were iteratively updated regularly.

[0073] Furthermore, the multi-source heterogeneous data includes soil environmental data, pomelo plant physiological data, arbuscular mycorrhizal fungal inoculant characteristic data, and cultivation management data. Among them, the optional soil environmental data includes soil pH, soil organic matter content, soil available nitrogen, phosphorus and potassium content, soil moisture content, soil aeration, and rhizosphere microbial community abundance; the optional pomelo plant physiological data includes plant height, stem diameter, total root length, root surface area, leaf chlorophyll content, plant net photosynthetic rate, and root activity; the optional arbuscular mycorrhizal fungal inoculant characteristic data includes fungal species, spore activity, mycelial growth rate, effective viable count of inoculants, and fungal infection adaptability coefficient; and the cultivation management data includes field irrigation amount, fertilizer application amount, tillage frequency, field temperature and humidity, and light duration.

[0074] Furthermore, soil environmental data is used to characterize the rhizosphere environment of citrus, plant physiological data is used to characterize the growth status of citrus, inoculant characteristic data is used to characterize the inoculation capacity of arbuscular mycorrhizal fungi, and cultivation management data is used to characterize the inoculation implementation conditions.

[0075] Furthermore, such as Figure 2 The diagram shows the architecture of the multi-factor importance analysis model. Specifically, the multi-factor importance analysis model involves normalizing the collected heterogeneous data from multiple sources to obtain a standardized factor dataset.

[0076] The standardized factor dataset is input into the pre-trained multi-decision tree random forest sub-model, and the outputs are mycorrhizal infection rate, root growth status, and plant growth promotion effect.

[0077] The out-of-bag data mean precision reduction method was used to calculate the independent perturbation error of each input data factor to each inoculation effect index. Based on preset perturbation weights, a weighted fusion was performed to obtain the comprehensive importance of the multiple objectives, as shown in the formula:

[0078]

[0079]

[0080] in, Factors Error perturbation difference in mycorrhizal infection rate, root growth status, and plant growth promotion effect; These are the out-of-bag baseline errors of the original model for the three performance indicators; These are the random shuffling factors. After the data, the model's out-of-bag perturbation error for the three performance indicators; For the first The final comprehensive importance score of each data factor; These are the preset weighting coefficients for mycorrhizal infection rate, root growth status, and plant growth promotion effect, respectively.

[0081] Based on a preset importance threshold, data factors with importance scores higher than the threshold are identified as core regulatory factors, thus completing factor screening.

[0082] Furthermore, the multi-decision tree random forest sub-model is specifically a parallel multi-decision tree ensemble model built based on the idea of ​​ensemble learning. It adopts a multi-output parallel regression architecture and contains multiple classification and regression decision trees with independent structures, random feature sampling, and random sample sampling. It can simultaneously fit the correlation between input factors and mycorrhizal infection rate, root growth status, and plant growth promotion effect.

[0083] During pre-training, all standardized multi-source heterogeneous data factors are used as common input features. Bootstrap sampling with replacement is used to randomly sample the training set samples and random subspace sampling is performed on the input features. Simultaneously, a parallel regression training task is established to create three-dimensional output labels for mycorrhizal infection rate, root growth status score, and plant growth promotion effect score. The splitting threshold and branch weight of each decision tree node are iteratively optimized until the overall prediction error of the model converges to the preset accuracy.

[0084] Furthermore, the multi-decision tree random forest sub-model adopts a two-layer ensemble hierarchical structure, including a bottom-level basic decision tree layer and a top-level ensemble fusion layer. The bottom-level basic decision tree layer pre-sets the number of independent regression decision trees to be 100-200, the maximum depth of a single decision tree to be set to 8-12 layers, the minimum number of sample splits per node to be 4-6, the input feature sampling ratio of a single decision tree to be 0.7-0.8, and the sample bootstrap sampling ratio to be 0.75. The top-level ensemble fusion layer adopts an equal-weight fusion strategy, performing parallel weighted averaging of the three-dimensional output results of all basic decision trees to uniformly output the predicted values ​​of the three inoculation effects. At the same time, an out-of-bag data sampling mechanism is enabled, reserving 30% of the out-of-bag samples for subsequent factor importance error calculation to ensure the effectiveness and stability of the quantification results.

[0085] Furthermore, such as Figure 3 The diagram shows the architecture of the nested vaccination parameter optimization model, which specifically includes a vaccination effect prediction layer and a multi-objective parameter optimization layer.

[0086] Among them, the inoculation effect prediction layer adopts a gradient boosting tree prediction model to predict the arbuscular mycorrhizal formation effect under different combinations of inoculation parameters based on the input core regulatory factors.

[0087] Based on the prediction results, the multi-objective parameter optimization layer performs multi-objective optimization on the inoculation dosage, inoculation time, inoculation location and inoculation method, with mycorrhizal infection rate, plant growth promotion effect, inoculant usage cost and environmental adaptability as optimization objectives, and outputs initial inoculation parameters that meet the preset optimization conditions.

[0088] Furthermore, the inoculation effect prediction layer specifically employs the XGBoost gradient boosting tree prediction algorithm, using the screened core regulatory factors as input features and mycorrhizal infection rate, root growth score, and plant growth promotion effect score as prediction outputs.

[0089] The prediction algorithm uses a second-order Taylor expansion to approximate the objective function, combined with L2 regularization to assign leaf weights, as shown in the formula:

[0090]

[0091]

[0092] in, Let be the objective function. The first gradient of the loss function, The loss function is the second-order Hessian matrix. For the first Tree prediction output, For tree structure regularization terms; For the first Optimal prediction weights for each leaf node For the first The set of all training samples covered by each leaf node. The L2 regularization coefficient;

[0093] By performing parallel operations on multiple decision trees, accumulating the prediction results of each tree, and outputting multi-dimensional prediction values ​​of vaccination effect, the system can effectively achieve the desired results.

[0094] Furthermore, the XGBoost gradient boosting tree prediction algorithm adopts a multi-layer stacked progressive boosting architecture, including a bottom-level feature input layer, a middle-level iterative optimization layer with multiple gradient boosting decision trees, and a top-level multi-dimensional effect output layer. The bottom-level feature input layer receives normalized core regulatory factor data to achieve high-dimensional feature regularization input. The middle-level iterative boosting layer presets 80-120 basic decision trees, with a maximum depth of 6-10 layers per decision tree, a learning rate of 0.01-0.1, an L2 regularization coefficient of 3-5, a leaf splitting penalty coefficient of 0.1-0.5, and a minimum number of samples per node of 3-5. It continuously reduces prediction error by fitting residuals tree by tree and superimposing optimization. The top-level multi-dimensional effect output layer simultaneously outputs predicted values ​​for mycorrhizal infection rate, root growth status score, and plant growth promotion effect, completing multi-objective simultaneous prediction.

[0095] The model employs an early stopping mechanism, terminating training when the prediction error on the test set fails to decrease for 10 consecutive iterations. This locks in the optimal model structure and weight parameters, preventing overfitting and ensuring the accuracy and robustness of the nonlinear correlation prediction between the core regulatory factors and the inoculation effect of arbuscular mycorrhizal fungi in pomelos.

[0096] Furthermore, the multi-objective parameter optimization layer specifically employs the NSGA-II non-dominated sorting genetic algorithm, with inoculation dosage, inoculation time, inoculation location, and inoculation method as optimization variables, and multi-objective optimization functions of maximizing mycorrhizal infection rate, maximizing plant growth promotion effect, minimizing inoculant usage cost, and maximizing environmental adaptability, to optimize parameters.

[0097] During parameter optimization, an initial population containing several sets of inoculation parameter combinations is randomly generated, and the individuals in the population are encoded. Based on four optimization objectives, the individuals in the population are stratified and non-dominatedly sorted to define the Pareto optimal level. The crowding distance between individuals in the same level is calculated to ensure population diversity. High-quality individuals are selected based on an elite retention strategy, and genetic crossover and mutation operations are performed to generate the offspring population. The parent and offspring populations are merged, and the non-dominated sorting and selection are repeated until a preset number of iterations is reached. Parameter combinations that meet a preset threshold are selected from the Pareto optimal solution set, and the initial inoculation parameters are output.

[0098] Furthermore, the dynamic adjustment of the initial inoculation parameters specifically involves: acquiring environmental change data and plant growth feedback data during the inoculation process; based on this data, employing a deviation accumulation adaptive correction algorithm, combined with deviation threshold damping constraints, time-series cumulative error compensation, and an adaptive weight update mechanism, to dynamically correct the inoculation parameters and obtain the target inoculation parameters, as shown in the formula:

[0099]

[0100] in, The corrected target inoculation parameters, These are the initial optimal inoculation parameters output by the NSGA-II algorithm; This represents the cumulative value of environmental deviation over time. This represents the cumulative value of plant growth feedback deviation over time. These are the sensitivity adjustment coefficients for environmental deviation and growth deviation, respectively. It is a hyperbolic tangent damping function; These are the adaptively adjusted weights that are dynamically updated with the iteration sequence, and the formula is:

[0101]

[0102] in, These are the initial baseline weights, This represents the number of model iterations. , The first Real-time environmental deviations and plant growth feedback deviations at each time-series sampling point.

[0103] Furthermore, the preparation of a special compound arbuscular mycorrhizal fungal agent for pomelo is as follows: based on the species suitability type, effective viable number, and agent ratio parameters in the target inoculation parameters, arbuscular mycorrhizal fungal species suitable for the rhizosphere environment of pomelo are screened, and mixed with organic matter carrier, water-retaining agent, and rhizosphere growth promoter for fermentation to prepare a special compound agent suitable for different growth stages of pomelo.

[0104] Furthermore, the inoculation strategy is matched according to different growth stages of pomelo as follows: In the seedling stage: with the core goal of rapid root germination and formation, the shallow root inoculation site is determined, and a high-frequency, low-dose inoculation frequency is set to ensure that the seedling roots can quickly infect and symbiotically infect the seedlings.

[0105] In the sapling stage, with the core objective of horizontal and vertical root expansion, the application method of fungicide is adjusted to combine ring application and trench application to adapt to the root growth range of saplings.

[0106] During the mature tree stage, the core objectives are to enhance the plant's resistance to adverse conditions and stabilize its vigor. Based on the soil environment and plant growth, the amount of inoculant supplementation is dynamically adjusted upwards or downwards, and the inoculant supplementation cycle is extended or shortened to maintain the long-term mycorrhizal symbiotic effect.

[0107] Furthermore, the nested inoculation parameter optimization model and the process parameters for dynamically adjusting the initial inoculation parameters are iteratively updated. Specifically, the inoculation effect prediction layer of the nested inoculation parameter optimization model is updated using the latest field-measured core regulatory factors as input and the actual inoculation effect as a label, through incremental iterative training. The updated XGBoost prediction model is used as the fitness evaluation function, and the constraint threshold and weight preference of the multi-objective parameter optimization layer are updated based on the latest effect feedback data. Based on the time-accumulated environmental deviation and plant growth deviation, the adaptive weight decay coefficient in the dynamic adjustment process is updated round by round, and the hyperbolic tangent nonlinear damping interval threshold is corrected according to long-term feedback data.

[0108] This embodiment details a multi-factor adaptive method for precise regulation of arbuscular mycorrhizal inoculation effects in citrus trees. The method collects multi-source data, quantifies the comprehensive importance of each factor on mycorrhizal infection rate, root growth, and plant growth promotion effects using a random forest model, and screens core regulatory factors. A nested parameter optimization model is constructed, and the inoculation effect is nonlinearly predicted using the XGBoost algorithm. Initial parameters are obtained through multi-objective Pareto optimization of inoculation dosage, time, location, and method using the NSGA-II algorithm. Target parameters are dynamically corrected based on field environment and plant growth feedback. A citrus-specific compound microbial agent is prepared according to the parameters, and inoculation strategies are differentiated according to three stages: seedling, sapling, and mature tree. The prediction model, optimization algorithm, and dynamic correction module are iteratively updated hierarchically based on field measurement feedback data.

[0109] Example 2, based on Example 1, details a comparative experiment on the effects of this method and existing mainstream methods on arbuscular mycorrhizal inoculation of pomelos in a standardized orchard of Shatin pomelos, as follows:

[0110] The test materials in this orchard were three-year-old Shatang pomelo saplings with uniform growth and free from pests and diseases, and the basic soil type was red and yellow soil. All groups were subject to the same basic field management, such as basal fertilizer, irrigation cycle, and shading, to eliminate interference from basic environmental variables. The total test period was 150 days, with sample collection and index testing conducted every 30 days. Long-term stability data were recorded at 90 days, 120 days, and 150 days.

[0111] The comparative methods included a blank control group, a container seedling static mycorrhizal culture group (PH-AM), a fixed ratio quality improvement inoculation group (P-AM), an orchard environment-regulated mycorrhizal group (ENV-AM), and a fruit tree cold resistance nutrition regulation group (COL-AM).

[0112] The blank control group did not apply any arbuscular mycorrhizal fungi inoculants throughout the entire process, and only implemented the orchard's routine basic water and fertilizer management, adopting a completely extensive cultivation model.

[0113] The static mycorrhizal culture group for container seedlings (PH-AM) combines arbuscular mycorrhizal fungi with hydrolyzed plant protein to complete the mycorrhizal inoculation and cultivation of seedlings in the seedling substrate. The same fungal strains, substrate ratios, and temperature and humidity conditions were maintained throughout the seedling stage. After transplanting, the existing cultivation system was maintained, completing the entire cycle of field cultivation and sample testing. During the experiment, management was carried out according to the established cultivation procedures, and corresponding cultivation techniques were used to complete mycorrhizal culture and field transplanting.

[0114] A compound inoculum agent was prepared by combining a fixed ratio quality improvement inoculation group (P-AM) with a variety of arbuscular mycorrhizal fungi. The agent was applied directly to the roots. The symbiotic relationship between the fungi and the plant optimized the nutrient absorption process of the pomelo tree. The entire experimental management was completed with the matching field cultivation technology. Sample collection and index testing were carried out in accordance with the unified experimental standards.

[0115] The Orchard Environmental Regulation Mycorrhizal Group (ENV-AM) is based on the native arbuscular mycorrhizal fungal community in the orchard, combined with the field vegetation interception system. It regulates the survival environment of rhizosphere microorganisms by optimizing the field fertilization mode, improves the soil nutrient status in the orchard, reduces field non-point source pollution, and completes the whole cycle test with a standardized field management process. Simultaneously, sample collection and various indicator detection and statistical work are carried out.

[0116] The Fruit Tree Cold Resistance Nutrition Regulation Group (COL-AM) carried out soil pre-improvement treatment based on the available phosphorus index in the soil, applied arbuscular mycorrhizal fungi in the radial trench method at the drip line of the tree canopy, improved the winter cold resistance of fruit trees through supporting cultivation techniques, implemented standardized field management throughout the process, and completed sample collection and testing of various indicators according to a unified experimental cycle.

[0117] In this methodology group (INV), four types of multi-source heterogeneous data were collected synchronously every 7 days. Soil indicators were obtained using a soil nutrient rapid analyzer and high-throughput sequencing of soil microorganisms; plant physiological indicators were measured in situ in the field using a portable photosynthesis meter, root scanning system, and chlorophyll meter; inoculant indicators were measured in the laboratory through spore counting and live bacteria plate culture; cultivation management data were automatically recorded by field IoT devices, and all data were imported into a database for normalization preprocessing.

[0118] The standardized dataset was input into the pre-trained two-layer multi-output random forest model. The comprehensive importance of each factor was calculated using the out-of-bag data accuracy reduction formula. A threshold of 0.6 was set to screen out five core regulatory factors: available phosphorus in soil, soil moisture content, root activity, effective viable bacteria count of microbial agents, and light duration.

[0119] The core factor input XGBoost gradient boost prediction layer is used to predict three inoculation effects using the second-order Taylor loss function and the optimal leaf weight formula. With infection rate, growth promotion effect, inoculant cost, and environmental adaptability as four objectives, the population is iteratively optimized through NSGA-II to output the initial inoculation parameters.

[0120] Every 7 days, environmental time-series deviation ΔEk and plant growth time-series deviation ΔGk are collected. These are then substituted into a nonlinear correction formula containing tanh damping, time-series cumulative deviation, and adaptive decay weight to finely adjust the inoculation dose, application location, and application frequency in real time, thereby obtaining the target inoculation parameters.

[0121] Based on the target parameters, a special compound AM microbial agent for citrus was prepared, and young trees were applied using a combination of ring trench and radial trench.

[0122] Data on actual vaccination effectiveness are collected every 30 days, and the XGBoost tree weights, NSGA-II population constraint thresholds, and nonlinear correction modules are updated incrementally in a hierarchical manner. The attenuation coefficient is used to continuously optimize the entire control model.

[0123] Each group has 3 biological replicate plots, with 12 pomelo trees planted in each plot. During the test, 5 trees are randomly selected from each plot for sampling. All indicators are measured in parallel 3 times and the average value is taken to reduce experimental error.

[0124] At day 90 of the experiment, five intact rootlets from each pomelo tree were randomly dug up from each plot. The roots were rinsed with water to remove soil from the root zone, and 1cm long fresh root segments were cut. The root samples were then treated with KOH alkaline dissociation and trypan blue staining to achieve a transparent staining effect, as shown below. Figure 4 The images shown are typical field-of-view comparisons of trypan blue chromosomes in the root systems of pomelo trees in each treatment group under a microscope. Figure 4It can be seen that the INV (this invention) group has a continuous and high-density mycelial network in the fine roots, with a large number and wide distribution of arbuscular and vesicular structures; the abundance of fungal symbiotic structures in the roots of the other control groups decreases step by step, and the CK blank group has only a very small number of scattered mycelia, and complete arbuscular structures are almost invisible.

[0125] Fifty fields of view were randomly selected under a stereomicroscope for manual counting and statistical analysis. The mycelial colonization rate, total mycorrhizal infection rate, root mycelial density, and arbuscular abundance were calculated. The core indicators of mycorrhizal colonization for each method were obtained by root staining and microscopic counting. The data are shown in Table 1 below.

[0126] Table 1 Core Indicators of Mycorrhizal Symbiotic Colonization

[0127]

[0128] According to Table 1 and Figure 5 The comparison chart of the core indicators of mycorrhizal colonization for each method shows that PH-AM relies solely on static treatment with fixed strains and exogenous proteins, and cannot adjust the inoculation scheme according to dynamic changes in soil and plants, resulting in a much lower mycorrhizal infection index than this method; LIN-ADJ linear correction has no damping constraints, and parameters frequently overshoot, limiting the symbiotic colonization effect; RF-SING optimizes only a single objective, ignoring root and environmental constraints, and its infection capacity is still inferior to this method; this method, through precise screening of core factors, optimization of multi-objective parameters, and nonlinear time-series dynamic fine-tuning, can sustainably maintain a high level of mycorrhizal symbiosis.

[0129] Simultaneously, plant growth and physiological indicators were tested. The pomelo trees were divided into aboveground and underground root systems. After blanching at 105℃, the roots were dried at 75℃ to constant weight, and the aboveground dry weight was obtained by weighing with an electronic balance. A root scanning system was used to scan complete root images, and the total root length was calculated using corresponding root analysis software. Functional leaves were collected, cryogenically ground, and extracted to prepare supernatants. The proline and malondialdehyde (MDA) contents of the leaves were determined spectrophotometrically. The growth and physiological stress resistance indicators of the pomelo trees were obtained through drying and weighing, field photosynthetic activity measurements, and spectrophotometric biochemical detection. The data are shown in Table 2 below.

[0130] Table 2. Growth and physiological stress resistance indicators of pomelo trees

[0131]

[0132] According to Table 2 and Figure 6As shown in the comparison chart of pomelo tree growth and physiological stress resistance indicators, in terms of growth volume, the aboveground dry weight and total root length of this method are the highest among all groups. Static PH-AM and artificial SF-TR, due to their fixed parameters, cannot adapt to the growth needs of the tree, and the improvement in root system and biomass is limited. In terms of photosynthesis and stress resistance, the photosynthetic efficiency of the plant is significantly improved under the regulation of this method, the accumulation of proline osmotic regulators is higher, and malondialdehyde oxidative damage is the lowest. Linear correction and single-objective optimization schemes cannot balance multiple growth indicators, and the improvement in stress resistance is weaker than that of this invention.

[0133] Sampling was conducted simultaneously with the collection of rhizosphere soil samples from the drip line (0–20 cm) of the tree canopy. The samples were then transported back to the laboratory under low-temperature refrigeration and sieved through a 2 mm sieve to remove root debris and gravel. Corresponding substrate reaction systems were prepared, and the catalytic activities of sucrase, urease, neutral phosphatase, and neutral protease in the soil were determined using spectrophotometry. The rhizosphere soil microenvironment improvement indicators for each method were obtained through spectrophotometric detection of soil enzymatic reactions. The data are shown in Table 3 below.

[0134] Table 3. Indicators for Improving the Rhizosphere Soil Microenvironment

[0135]

[0136] According to Table 3 and Figure 7 As shown in the comparison chart of the rhizosphere soil microenvironment improvement indicators of the various methods, it can be seen that the comparative schemes could not dynamically adjust the supply of microbial agents according to the soil nutrients, and the improvement of various enzyme activities in the soil was relatively low. The present invention continuously and dynamically matches the soil environment requirements, and the rhizosphere microbial metabolic activity is significantly better than all control groups, thus achieving long-term soil improvement.

[0137] During the experimental period, the total mycorrhizal infection rate of each group was uniformly determined every 10 days using root staining microscopy. The difference between the maximum and minimum infection rates at the three time points was calculated as the infection rate fluctuation range. The larger the range, the more severe the decay of the inoculation effect over time and the worse the system stability. Through multiple root infection tests in stages and range conversion, the long-term adaptive stability index of each method was obtained, as shown in Table 4 below:

[0138] Table 4 System Long-Term Adaptive Stability Indicators

[0139]

[0140] According to Table 4 and Figure 8The comparison chart of the system's long-term adaptive stability index for each method shows that PH-AM and SF-TR lack a model iterative update mechanism, and their inoculation effect continues to decline and fluctuates greatly as the orchard environment and tree growth change; LIN-ADJ and RF-SING lack a closed-loop self-updating system throughout the entire process, resulting in insufficient long-term stability; this invention can periodically update the prediction, optimization, and correction modules based on field measured data in a hierarchical manner, and the mycorrhizal infection effect remains almost unchanged throughout the entire cycle, demonstrating long-term adaptive regulation capability.

[0141] The total weight of the actual inoculant applied in each group was recorded throughout the entire experimental period. At the end of the experiment, the total fresh fruit yield of each plot was calculated, and the inoculant purchase and usage cost per unit of fresh fruit yield was calculated. At the same time, the total increase in aboveground and belowground dry matter of each group of plants was recorded, and the corresponding increase in dry matter per gram of inoculant input was calculated. The economic indicators of each method were obtained by converting field material usage records, fruit harvesting and weighing, and plant dry weight statistics. The data are shown in Table 5 below.

[0142] Table 5 Economic Indicators for Regulation

[0143]

[0144] According to Table 5 and Figure 9 The comparison chart of the economic indicators of the various methods shows that PH-AM and SF-TR use fixed large doses of bacterial agents, resulting in serious waste of bacterial agents and high production costs; LIN-ADJ and RF-SING only optimize parameters locally, and their cost control capabilities are limited; the present invention NSGA-II uses multi-objective optimization to simultaneously constrain infection effect and bacterial agent cost, and combines nonlinear dynamic fine-tuning to avoid over-application, resulting in the lowest production cost under the same growth-promoting effect, and has outstanding advantages for industrialization and promotion.

[0145] This embodiment details a comparative experiment on the effects of mycorrhizal inoculation of pomelos using this method and existing mainstream methods in a standardized pomelos orchard. The experiment included the present invention group, a blank control group, and comparative experimental groups corresponding to four existing patents. A 150-day full-cycle cultivation experiment and index testing were conducted according to standardized procedures. The results show that existing technologies all suffer from a gradual decline in regulatory efficacy with prolonged cultivation time. In contrast, the present invention, based on multi-factor screening, nested model optimization, dynamic correction, and self-learning iterative update mechanisms, can effectively suppress long-term efficacy decline. It exhibits significant advantages in mycorrhizal colonization, plant growth and stress resistance, soil improvement, application cost, and long-term stability, fully verifying the scientific validity and application value of the present invention.

[0146] The above are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. For those skilled in the art, the present invention can have various modifications and variations. Any changes, modifications, substitutions, integrations, and parameter changes made to these embodiments within the spirit and principles of the present invention, without departing from the principles and spirit of the present invention, through conventional substitutions or to achieve the same function, fall within the scope of protection of the present invention.

Claims

1. A multi-factor adaptive method for precise control of arbuscular mycorrhizal inoculation effect in citrus, characterized in that, include: Collect multi-source heterogeneous data from pomelo planting areas; A multi-factor importance analysis model was constructed, and the multi-source heterogeneous data was input into the multi-factor importance analysis model. By analyzing the influence relationship between different data factors and the inoculation effect of arbuscular mycorrhizal inoculation of citrus, the core regulatory factors affecting the inoculation effect of citrus arbuscular mycorrhizal inoculation were screened. A nested inoculation parameter optimization model is constructed. The nested inoculation parameter optimization model takes in a core regulatory factor, predicts the arbuscular mycorrhizal formation effect under different combinations of inoculation parameters, and performs multi-objective optimization on inoculation dosage, inoculation time, inoculation location and inoculation method based on the prediction results, and outputs initial inoculation parameters that meet the preset optimization conditions. Based on environmental change data and plant growth feedback data during the inoculation process, the initial inoculation parameters are dynamically adjusted to obtain the target inoculation parameters; A compound arbuscular mycorrhizal fungal agent specifically for citrus was prepared according to the target inoculation parameters, and corresponding inoculation strategies were matched according to different growth stages of citrus. After completing the arbuscular mycorrhizal inoculation, data on the growth status of pomelo plants, changes in the rhizosphere environment, and the effect of mycorrhizal formation were continuously collected. The nested inoculation parameter optimization model and the process parameters for dynamically adjusting the initial inoculation parameters were iteratively updated regularly.

2. The method for precise control of multi-factor adaptive arbuscular mycorrhizal inoculation effect in citrus trees according to claim 1, characterized in that, The multi-factor importance analysis model specifically involves normalizing the collected multi-source heterogeneous data to obtain a standardized factor dataset. The standardized factor dataset is input into the pre-trained multi-decision tree random forest sub-model, and the outputs are mycorrhizal infection rate, root growth status, and plant growth promotion effect. The out-of-bag data average precision reduction method was used to calculate the independent perturbation error of each input data factor on each inoculation effect index. Based on the preset perturbation weights, the weighted fusion was performed to obtain the comprehensive importance of multiple targets. Based on a preset importance threshold, data factors with importance scores higher than the threshold are identified as core regulatory factors, thus completing factor screening.

3. The method for precise control of multi-factor adaptive arbuscular mycorrhizal inoculation effect in citrus trees according to claim 2, characterized in that, The multi-decision tree random forest sub-model is specifically a parallel multi-decision tree ensemble model built based on the idea of ​​ensemble learning. It adopts a multi-output parallel regression architecture and contains multiple classification and regression decision trees with independent structures, random feature sampling, and random sample sampling. It can simultaneously fit the correlation between input factors and mycorrhizal infection rate, root growth status, and plant growth promotion effect. During pre-training, all standardized multi-source heterogeneous data factors are used as common input features. Bootstrap sampling with replacement is used to randomly sample the training set samples and random subspace sampling is performed on the input features. Simultaneously, a parallel regression training task is established to create three-dimensional output labels for mycorrhizal infection rate, root growth status score, and plant growth promotion effect score. The splitting threshold and branch weight of each decision tree node are iteratively optimized until the overall prediction error of the model converges to the preset accuracy.

4. The method for precise control of multi-factor adaptive arbuscular mycorrhizal inoculation effect in citrus trees according to claim 1, characterized in that, The nested vaccination parameter optimization model specifically includes a vaccination effect prediction layer and a multi-objective parameter optimization layer. The inoculation effect prediction layer uses a gradient boosting tree prediction model to predict the arbuscular mycorrhizal formation effect under different combinations of inoculation parameters based on the input core regulatory factors. The multi-objective parameter optimization layer performs multi-objective optimization on the inoculation dosage, inoculation time, inoculation location and inoculation method based on the prediction results. The optimization objectives are mycorrhizal infection rate, plant growth promotion effect, inoculant usage cost and environmental adaptability. The layer outputs initial inoculation parameters that meet the preset optimization conditions.

5. The method for precise control of multi-factor adaptive arbuscular mycorrhizal inoculation effect in citrus trees according to claim 4, characterized in that, The inoculation effect prediction layer specifically employs the XGBoost gradient boosting tree prediction algorithm, using the screened core regulatory factors as input features and mycorrhizal infection rate, root growth score, and plant growth promotion effect score as prediction outputs. The prediction algorithm uses a second-order Taylor expansion to approximate the objective function, combined with L2 regularization to allocate leaf weights, as shown in the formula: in, Let be the objective function. The first gradient of the loss function, The loss function is the second-order Hessian matrix. For the first Tree prediction output, For tree structure regularization terms; For the first Optimal prediction weights for each leaf node For the first The set of all training samples covered by each leaf node. The L2 regularization coefficient; By performing parallel operations on multiple decision trees, accumulating the prediction results of each tree, and outputting multi-dimensional prediction values ​​of vaccination effect, the system can effectively achieve the desired results.

6. The method for precise control of multi-factor adaptive arbuscular mycorrhizal inoculation effect in citrus trees according to claim 5, characterized in that, The multi-objective parameter optimization layer specifically employs the NSGA-II non-dominated sorting genetic algorithm, with inoculation dosage, inoculation time, inoculation location, and inoculation method as optimization variables, and multi-objective optimization functions such as maximizing mycorrhizal infection rate, maximizing plant growth promotion effect, minimizing inoculant usage cost, and maximizing environmental adaptability to optimize parameters. During parameter optimization, an initial population containing several sets of inoculation parameter combinations is randomly generated, and the individuals in the population are encoded. Based on four optimization objectives, the individuals in the population are stratified and non-dominatedly sorted to define the Pareto optimal level. The crowding distance between individuals in the same level is calculated to ensure population diversity. High-quality individuals are selected based on an elite retention strategy, and genetic crossover and mutation operations are performed to generate the offspring population. The parent and offspring populations are merged, and the non-dominated sorting and selection are repeated until a preset number of iterations is reached. Parameter combinations that meet a preset threshold are selected from the Pareto optimal solution set, and the initial inoculation parameters are output.

7. The method for precise control of multi-factor adaptive arbuscular mycorrhizal inoculation effect in citrus trees according to claim 6, characterized in that, The dynamic adjustment of the initial inoculation parameters specifically involves: acquiring environmental change data and plant growth feedback data during the inoculation process; and, based on this data, employing a deviation accumulation adaptive correction algorithm, combined with deviation threshold damping constraints, time-series cumulative error compensation, and an adaptive weight update mechanism, dynamically correcting the inoculation parameters to obtain the target inoculation parameters, as shown in the formula: in, The corrected target inoculation parameters, These are the initial optimal inoculation parameters output by the NSGA-II algorithm; This represents the cumulative value of environmental deviation over time. This represents the cumulative value of plant growth feedback deviation over time. These are the sensitivity adjustment coefficients for environmental deviation and growth deviation, respectively; It is a hyperbolic tangent damping function; These are the adaptively adjusted weights that are dynamically updated with the iteration sequence, and the formula is: in, These are the initial baseline weights, This represents the number of model iterations. , The first Real-time environmental deviations and plant growth feedback deviations at each time-series sampling point.

8. The method for precise control of multi-factor adaptive arbuscular mycorrhizal inoculation effect in citrus trees according to claim 1, characterized in that, The preparation of the special compound arbuscular mycorrhizal fungal agent for pomelo is specifically as follows: based on the fungal strain suitability type, effective viable count, and fungal agent ratio parameters in the target inoculation parameters, arbuscular mycorrhizal fungal strains suitable for the rhizosphere environment of pomelo are screened, and mixed with organic matter carrier, water-retaining agent, and rhizosphere growth promoter for fermentation to prepare a special compound fungal agent suitable for different growth stages of pomelo.

9. The method for precise control of multi-factor adaptive arbuscular mycorrhizal inoculation effect in citrus trees according to claim 8, characterized in that, The inoculation strategy matched according to different growth stages of pomelo is as follows: In the seedling stage: with the core goal of rapid root germination and formation, the shallow root inoculation site is determined, and a high-frequency, low-dose inoculation frequency is set to ensure that the seedling roots can quickly infect and symbiotically infect the seedlings. During the sapling stage, the core objective is to expand the root system both horizontally and vertically. The application method of fungicide is adjusted to combine ring application and trench application to suit the root growth range of saplings. During the mature tree stage, the core objectives are to enhance the plant's resistance to adverse conditions and stabilize its vigor. Based on the soil environment and plant growth, the amount of inoculant supplementation is dynamically adjusted upwards or downwards, and the inoculant supplementation cycle is extended or shortened to maintain the long-term mycorrhizal symbiotic effect.

10. The method for precise control of multi-factor adaptive arbuscular mycorrhizal inoculation effect in citrus trees according to claim 1, characterized in that, The iterative update of the nested inoculation parameter optimization model and the process parameters for dynamically adjusting the initial inoculation parameters involves the following steps: using the latest field-measured core regulatory factors as input and the actual inoculation effect as a label, an incremental iterative training method is adopted to update the inoculation effect prediction layer of the nested inoculation parameter optimization model; using the updated XGBoost prediction model as the fitness evaluation function, the constraint threshold and weight preference of the multi-objective parameter optimization layer are updated based on the latest effect feedback data; based on the time-accumulated environmental deviation and plant growth deviation, the adaptive weight decay coefficient in the dynamic adjustment process is updated round by round, and the hyperbolic tangent nonlinear damping interval threshold is corrected according to long-term feedback data.

Images