Agricultural enzyme adaptive generation and optimization method based on multi-source data fusion

Abstract

This invention discloses an adaptive generation and optimization method for agricultural enzymes based on multi-source data fusion, relating to the field of multi-source data fusion technology. The method includes: collecting soil physicochemical index data, rhizosphere autotoxic substance concentration data, and component characteristic data of agricultural waste to be treated to obtain a first input matrix; importing the first input matrix into a first-level multi-objective optimization model, using the maximization of autotoxic substance degradation rate as the objective function, calculating the first-stage fermentation process parameters, and outputting execution instructions; collecting residual autotoxic substance concentration data and microbial community diversity data from the soil, calculating the micro-ecological pressure index; constructing a second input matrix, and importing the micro-ecological pressure index as a constraint boundary condition into a second-level multi-objective optimization model; dynamically adjusting the weight coefficients of the objective function according to the numerical range of the micro-ecological pressure index, calculating the second-stage fermentation process parameters, and outputting seedling enzyme preparation instructions.

Description

Technical Field

[0001] This invention relates to the field of multi-source data fusion technology, and in particular to an adaptive generation and optimization method for agricultural enzymes based on multi-source data fusion. Background Technology

[0002] As an important economic and medicinal crop, the artificial cultivation of Dictamnus dasycarpus faces severe challenges, particularly the problems of continuous cropping obstacles and high seedling mortality. During long-term cultivation, the roots of Dictamnus dasycarpus secrete phenolic acid autotoxic substances. These substances accumulate in the soil, not only inhibiting the growth of the plant itself but also becoming a breeding ground for root rot pathogens, leading to an imbalance in the soil's microecology.

[0003] Currently, while traditional agricultural waste enzyme production technologies can improve soil to some extent, they suffer from significant technical shortcomings. First, existing technologies mostly employ static, universal formulas, lacking targeted treatment capabilities for specific crop rotation obstacles and failing to precisely degrade specific autotoxic substances. Second, soil improvement and seedling cultivation are often disconnected, lacking a dynamic control mechanism based on data flow. That is, soil remediation data from the land preparation stage is not effectively transferred to the seedling stage, resulting in enzyme preparation during the seedling period being unable to adaptively adjust to residual microecological pressure in the soil. Furthermore, traditional fermentation processes are heavily influenced by environmental and meteorological factors, lacking a dynamic compensation mechanism, leading to unstable enzyme quality.

[0004] Therefore, there is an urgent need for an agricultural enzyme generation method that can integrate multi-source data, achieve synergistic optimization of soil and seedlings, and possess adaptive capabilities, in order to solve the problems of continuous cropping obstacles and low seedling survival rate in the cultivation of Dictamnus dasycarpus. Summary of the Invention

[0005] This invention provides an adaptive generation and optimization method for agricultural enzymes based on multi-source data fusion, which solves the problem that the traditional fermentation process in the prior art is greatly affected by environmental and meteorological factors and lacks a dynamic compensation mechanism, resulting in unstable enzyme quality, and achieves the technical effect of co-optimization between the land and seedlings.

[0006] The present invention provides a method for adaptive generation and optimization of agricultural enzymes based on multi-source data fusion, comprising:

[0007] Soil physicochemical index data, rhizosphere autotoxic substance concentration data, and component characteristic data of agricultural waste to be treated were collected from the planting plots of Dictamnus dasycarpus, and normalized to form the first input matrix;

[0008] The first input matrix is ​​imported into the first-level multi-objective optimization model. The objective function is to maximize the degradation rate of autotoxic substances. The first-stage fermentation process parameters are calculated, including the waste mixing ratio, the inoculum amount of degradation bacteria and the temperature control curve. The execution command is then output.

[0009] Within a preset time window after the first phase of formulation instructions is completed, data on the concentration of residual autotoxic substances and the diversity of microbial communities in the soil are collected, and a microecological pressure index characterizing the degree of soil remediation is obtained through weighted calculation.

[0010] A second input matrix containing physiological requirement parameters of Dictamnus dasycarpus seedlings is constructed, and the microecological pressure index is used as a constraint boundary condition to be imported into the second-level multi-objective optimization model. According to the numerical range of the microecological pressure index, the weight coefficient of the objective function is dynamically adjusted to calculate the second-stage fermentation process parameters, including the nutrient substrate ratio, functional microbial community combination and fermentation time, and output the seedling enzyme preparation instructions.

[0011] The process for acquiring and processing rhizosphere autotoxic substance concentration data includes:

[0012] Rhizosphere soil samples were collected from plots where Dictamnus dasycarpus was continuously cropped. Characteristic peak area data were extracted using liquid chromatography-mass spectrometry. The concentration values ​​of p-hydroxybenzoic acid, vanillic acid, and ferulic acid were identified and locked. The concentration values ​​were compared with the preset tolerance threshold of Dictamnus dasycarpus to calculate the multiple of each substance exceeding the standard. The multiple of exceeding the standard was used as the penalty factor weight in the first-level multi-objective optimization model.

[0013] Data on lignin content and carbon-to-nitrogen ratio of agricultural waste were collected to establish a database of waste degradation potential. The molecular structure characteristics of autotoxic substances were matched with the waste degradation potential data to screen out waste types that can induce the production of phenolic acid degrading enzymes. These waste types were marked as recommended raw material data and stored in the first input matrix.

[0014] The computational logic of the first-level multi-objective optimization model includes:

[0015] The decision variables are set as the mass ratio of different types of waste and the inoculation concentration of each single bacterium in the compound microbial agent; the objective function is constructed and defined as maximizing the predicted value of phenolic acid degrading enzyme activity in the fermentation product.

[0016] In the first-level multi-objective optimization model, the component characteristic data of waste are called in during the operation to fit the carbon source release curve during the fermentation process. When the fitting result shows that the carbon source release rate is lower than the enzyme protein synthesis requirement, the feeding ratio parameter of the fast-acting carbon source is increased. The non-dominated sorting genetic algorithm is used to optimize the objective function and output a set of Pareto optimal solutions. Based on the current ambient temperature data, an optimal solution is automatically selected and converted into a digital control command containing specific stirring frequency and ventilation volume.

[0017] The calculation method for the microecological stress index includes:

[0018] Obtain soil sample data after enzyme treatment during land preparation, including residual phenolic acid concentration, Fusarium gene copy number, and Trichoderma gene copy number;

[0019] The chemical stress component was obtained by dividing the residual phenolic acid concentration by the lethal concentration constant of Dictamnus dasycarpus seedlings.

[0020] Calculate the ratio of gene copy numbers of Fusarium to Trichoderma, and use the ratio to obtain the biological stress component through a logarithmic function mapping;

[0021] The fractal dimension data of soil aggregate structure were collected and normalized to obtain the physical structure components.

[0022] The entropy weights of the three components were calculated using the entropy weight method, and the weighted sum was defined as the dimensionless soil microecological pressure index. The higher the value, the greater the resistance of the soil to the survival of Dictamnus dasycarpus seedlings, and the more intensive intervention is needed in the next stage.

[0023] The processing logic of the second-level multi-objective optimization model based on the micro-ecological pressure index is as follows:

[0024] Set a critical threshold for the stress index. When the input microecological stress index is higher than the threshold, the model enters the defense priority calculation path: set the objective function to maximize the content of antibiotic secondary metabolites and chitinase in the fermentation product, and force the increase of the inoculation ratio parameter of antagonistic bacteria in the constraints, and output the formula instruction with high antibacterial activity.

[0025] When the input microecological pressure index is lower than the threshold, the growth-promoting priority calculation path is entered: the objective function is set to maximize the content of indoleacetic acid and amino acids in the fermentation product, and the inoculation ratio parameter of phosphorus-solubilizing and potassium-solubilizing bacteria is increased in the constraints, and the formula instruction with high nutritional activity is output.

[0026] To achieve a digital response of the functional properties of enzymes during the seedling stage to the effect of land preparation.

[0027] Dynamic compensation of fermentation parameters based on environmental meteorological data, including:

[0028] The system receives real-time weather forecast data for the next week for the location of the fermentation workshop and extracts the environmental temperature and humidity change sequences. The temperature change sequence is input into a thermodynamic conduction model to calculate the heat loss rate of the fermentation tank. When the predicted environmental temperature is lower than the optimal metabolic temperature range for microorganisms, the system increases the proportion of high-calorific-value waste in the formula and outputs compensation data to extend the fermentation cycle. When the predicted environmental humidity is too high and affects the moisture content of the waste, the system adjusts the water addition parameter in the formula.

[0029] Closed-loop correction based on seedling feedback data includes:

[0030] After the seedling stage, data on the survival rate, root development index, and root rot incidence of Dictamnus dasycarpus seedlings were collected to obtain posterior data. The posterior data were then combined with the previous soil microecological pressure index and the second-stage fermentation process parameters to form a training sample pair. A deep learning algorithm was used to train a formula efficacy prediction model, and the deviation between the actual growth data and the model prediction was calculated. If the deviation exceeded the preset range, the weight coefficient matrix in the second-level optimization model was adjusted in reverse.

[0031] The beneficial effects of this invention are:

[0032] 1. Achieving precise management and digital response through coordinated soil and seedling cultivation: This invention constructs a microecological pressure index as a key link between the land preparation and seedling stages, achieving strong coupling between the two phases. During land preparation, enzymes focus on maximizing the degradation of autotoxic substances, while during seedling cultivation, enzymes adaptively adjust the weights of defense and growth-promoting functions based on the residual soil pressure after land preparation. This data-flow-based closed-loop control ensures that enzymes during seedling cultivation can provide the most suitable biological support for the current soil conditions, thereby significantly improving seedling survival rates in complex soil environments.

[0033] 2. Overcoming environmental fluctuations and ensuring the stability of fermentation quality: This invention introduces a thermodynamic compensation mechanism based on environmental meteorological data, which monitors future weather changes in real time, calculates the heat loss rate of the fermenter through a thermodynamic conduction model, and dynamically adjusts the formula or extends the fermentation cycle accordingly. It effectively solves the problem of delayed microbial metabolism caused by temperature fluctuations in open or semi-open environments, especially in cold northern regions, and ensures the consistency and stability of the technical indicators of the final enzyme product.

[0034] 3. This invention endows the system with self-evolution capabilities, enabling precision agriculture. It utilizes deep learning algorithms to construct a formula efficacy prediction model and establishes a closed-loop correction mechanism based on seedling feedback data. After the seedling stage, the system collects actual growth data, compares it with predicted values, and then adjusts and optimizes the model's weight coefficients. This feedback training mechanism allows the system to continuously correct its understanding of the formula-effect mapping relationship as usage time and data accumulation increase, gradually approaching the optimal solution for a specific plot of land, thus achieving truly intelligent precision agriculture. Attached Figure Description

[0035] Figure 1 This is a flowchart illustrating the adaptive generation and optimization method for agricultural enzymes based on multi-source data fusion, as described in this invention.

[0036] Figure 2 This is a schematic diagram showing the comparison of total phenolic acid degradation rate data in the comparative experiment of this invention;

[0037] Figure 3This is a schematic diagram showing the data comparison of seedling survival rate in the comparative experiment of this invention. Detailed Implementation

[0038] The above technical solutions will be described in detail below with reference to the accompanying drawings and specific embodiments to better understand them. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments of the present invention. It should be understood that the present invention is not limited to the exemplary embodiments used only to explain the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. It should also be noted that, for ease of description, only the parts related to the present invention are shown in the drawings, not all of them.

[0039] Example 1:

[0040] This invention proposes an adaptive generation and optimization method for agricultural enzymes based on multi-source data fusion. The process of the method is as follows: Figure 1 As shown, it includes:

[0041] Soil physicochemical index data, rhizosphere autotoxic substance concentration data, and component characteristic data of agricultural waste to be treated were collected from the planting plots of Dictamnus dasycarpus, and normalized to form the first input matrix;

[0042] The first input matrix is ​​imported into the first-level multi-objective optimization model. The objective function is to maximize the degradation rate of autotoxic substances. The first-stage fermentation process parameters are calculated, including the waste mixing ratio, the inoculum amount of degradation bacteria and the temperature control curve. The execution command is then output.

[0043] Within a preset time window after the first phase of formulation instructions is completed, data on the concentration of residual autotoxic substances and the diversity of microbial communities in the soil are collected, and a microecological pressure index characterizing the degree of soil remediation is obtained through weighted calculation.

[0044] A second input matrix containing physiological requirement parameters of Dictamnus dasycarpus seedlings is constructed, and the microecological pressure index is used as a constraint boundary condition to be imported into the second-level multi-objective optimization model. According to the numerical range of the microecological pressure index, the weight coefficient of the objective function is dynamically adjusted to calculate the second-stage fermentation process parameters, including the nutrient substrate ratio, functional microbial community combination and fermentation time, and output the seedling enzyme preparation instructions.

[0045] In this embodiment, the multi-source data fusion acquisition step covers specific feature parameters in three dimensions:

[0046] Soil physicochemical index data mainly include pH value, which determines the microbial colonization environment; soil bulk density and total porosity, which characterize the root respiration environment and aggregate structure; soil organic matter content (SOM) and soil carbon-nitrogen ratio, which reflect the basic fertility and carbon-nitrogen cycling capacity of the soil; and electrical conductivity (EC value) and available nitrogen, phosphorus and potassium content, which assess salt stress and nutrient supply levels.

[0047] The data on rhizosphere autotoxic substances focus on the chemical causes of continuous cropping obstacles of Dictamnus dasycarpus. Specifically, it quantifies the concentration of key monomeric phenolic acids such as p-hydroxybenzoic acid, vanillic acid, ferulic acid and cinnamic acid in the rhizosphere soil, as well as the total phenolic acid content and the residual amount of Dictamnus dasycarpus alkaloid, in order to accurately assess the intensity of allelopathic effects.

[0048] The component characteristics data of agricultural waste to be treated cover the contents of lignin, cellulose and hemicellulose that determine the enzyme induction potential, the carbon-nitrogen ratio, total sugar / reducing sugar content, moisture content and pH value of the raw materials that regulate fermentation kinetics, and pathogen-carrying characteristics used to assess biosafety risks.

[0049] The process for acquiring and processing rhizosphere autotoxic substance concentration data includes:

[0050] Rhizosphere soil samples were collected from plots where Dictamnus dasycarpus was continuously cropped. Characteristic peak area data were extracted using liquid chromatography-mass spectrometry. The concentration values ​​of p-hydroxybenzoic acid, vanillic acid, and ferulic acid were identified and locked. The concentration values ​​were compared with the preset tolerance threshold of Dictamnus dasycarpus to calculate the multiple of each substance exceeding the standard. The multiple of exceeding the standard was used as the penalty factor weight in the first-level multi-objective optimization model.

[0051] During the implementation process, rhizosphere soil samples were first collected from the plots where Dictamnus dasycarpus was continuously cropped.

[0052] Chromatograms of the samples were acquired using liquid chromatography-mass spectrometry (LC-MS), and characteristic peak area data were extracted. The system has a built-in standard database that automatically identifies and pinpoints p-hydroxybenzoic acid, vanillic acid, and ferulic acid—the three main allelopathic autotoxic substances found in Dictamnus dasycarpus root bark. Assuming a detected p-hydroxybenzoic acid concentration of 50 mg / kg, while the preset seedling tolerance threshold is 10 mg / kg, the exceedance factor is calculated to be five times. This five-fold value is not directly output but is instead converted into penalty factor weights in the first-level multi-objective optimization model.

[0053] This step employs a threshold comparison method combined with a penalty function method. The principle is that the severity of continuous cropping obstacles is non-linearly positively correlated with the concentration of autotoxic substances. By mapping the excess multiple to a penalty weight in the optimization model, the algorithm is forced to prioritize the degradation ability of high-concentration excess substances when searching for the optimal formulation; otherwise, the fitness score of the formulation will be significantly reduced.

[0054] This technical solution enables the digital identification of the core causes of continuous cropping obstacles in Dictamnus dasycarpus; by converting chemical detection data into algorithm weight parameters, it ensures that the generated enzyme formula has strong targeting, solving the technical problem of blind fermentation and inability to accurately detoxify soil toxicity in traditional methods.

[0055] Data on lignin content and carbon-to-nitrogen ratio of agricultural waste were collected to establish a database of waste degradation potential. The molecular structure characteristics of autotoxic substances were matched with the waste degradation potential data to screen waste types that could induce the production of phenolic acid degrading enzymes. These waste types were marked as recommended raw material data and stored in the first input matrix.

[0056] In this embodiment, the logic for matching the molecular structural characteristics of autotoxic substances with waste degradation potential data is as follows: a method combining molecular fingerprint-based structural similarity calculation with enzyme induction efficacy weighting is employed. The specific steps are as follows:

[0057] Step 1: Molecular fingerprint extraction and initial similarity screening;

[0058] First, the chemical structures of the autotoxic substances and major components of the waste to be treated were converted into SMILES codes. Using the RDKit cheminformatics toolkit, corresponding MACCS bond molecular fingerprints were generated, covering 166 key molecular structural features. Subsequently, the Tanimoto coefficient was introduced as an evaluation index for similarity measurement, with a similarity threshold set at 0.65.

[0059] When the calculated Tanimoto coefficient is greater than the threshold, the waste component is determined to contain functional groups such as benzene rings, hydroxyl groups or methoxy groups that are highly similar to the autotoxic substance, and has the structural basis to induce microorganisms to synthesize specific degradation enzymes as a structural analog, and is included in the candidate list.

[0060] Step 2: Verification of enzyme induction efficacy database;

[0061] A pre-built enzyme induction potential database is invoked. This database, constructed from historical experimental data, records the induction folds of different waste substrates on the target enzyme system. Waste substrates in the candidate list are then subjected to secondary validation, and their corresponding enzyme induction folds are read. If a waste substrate, although structurally similar, has an induction fold below 1.5 times according to historical data (i.e., not significantly inducing), it is removed.

[0062] Step 3: The final determination of recommended raw materials will be based on the waste types that have undergone double screening. The waste types will be sorted according to the comprehensive score of (structural similarity × induction factor), and the top-ranked types will be marked as recommended raw material data.

[0063] Collect physicochemical data of agricultural waste to be treated, such as corn stalks and fruit tree branches, and establish a degradation potential database including lignin content and carbon-nitrogen ratio.

[0064] For example, when the main autotoxic substance is ferulic acid with a benzene ring structure, a molecular structure matching algorithm is used to search the database for raw materials that can induce microorganisms to secrete laccase or peroxidase. The screening results show that fruit tree branches with high lignin content have extremely strong enzyme induction ability, so they are marked as recommended raw materials and stored in the first input matrix.

[0065] This step is based on the theory of enzyme-induced synthesis and the principle of structural analogues. During metabolism, if specific substrates or their structural analogues are present in the environment, such as lignin and phenolic acid autotoxic substances, microorganisms will be induced to secrete corresponding degradation enzyme systems. The algorithm optimizes the substrate composition from the source by matching the biochemical characteristics of waste with the molecular structure of autotoxic substances.

[0066] This solution cleverly utilizes the biochemical characteristics of agricultural waste to transform it into a functional enzyme inducer. This not only improves the degradation efficiency of autotoxic substances but also realizes the high-value utilization of waste resources, embodying the ecological cycle concept of treating toxic substances with waste.

[0067] The computational logic of the first-level multi-objective optimization model includes:

[0068] The decision variables are set as the mass ratio of different types of waste and the inoculation concentration of each single bacterium in the compound microbial agent; the objective function is constructed and defined as maximizing the predicted value of phenolic acid degrading enzyme activity in the fermentation product.

[0069] In the first-level multi-objective optimization model, the component characteristic data of waste are called in during the operation to fit the carbon source release curve during the fermentation process. When the fitting result shows that the carbon source release rate is lower than the enzyme protein synthesis requirement, the feeding ratio parameter of the fast-acting carbon source is increased. The non-dominated sorting genetic algorithm is used to optimize the objective function and output a set of Pareto optimal solutions. Based on the current ambient temperature data, an optimal solution is automatically selected and converted into a digital control command containing specific stirring frequency and ventilation volume.

[0070] In the first-level multi-objective optimization model operation, the decision variables are set as the feed ratio of straw to soybean meal and the inoculum size of Bacillus subtilis. Data on the cellulose and protein content of the waste are used, and the carbon source release curve during fermentation is fitted using kinetic equations. If the simulation shows that during the rapid cell proliferation period on the third day of fermentation, the rate of carbon source release from straw is lower than the demand for cell-synthesized enzymes (i.e., carbon starvation), the model will automatically adjust the parameters to increase the proportion of readily available carbon sources (such as molasses). Finally, a non-dominated sorting genetic algorithm is used to solve the problem, outputting a Pareto optimal solution that balances activity and cost, which is then converted into specific aeration and stirring commands.

[0071] In this embodiment, the specific construction and calculation process of the first-level multi-objective optimization model is as follows:

[0072] First, the calculation benchmark for the penalty factor weight was determined; the tolerance thresholds of Dictamnus dasycarpus seedlings to three key autotoxic substances were set as follows: 20 μg / g of p-hydroxybenzoic acid, 15 μg / g of vanillic acid, and 10 μg / g of ferulic acid. These thresholds were determined through a pre-conducted plate bioassay experiment, i.e., the concentration at which the inhibition rate reached 10%; the excess multiple was calculated as the penalty weight.

[0073] Secondly, an objective function is constructed to maximize the activity of phenolic acid degrading enzymes. This embodiment uses a second-order polynomial regression model trained based on response surface methodology as the prediction kernel. Its mathematical expression is as follows:

[0074] ;

[0075] in, The response variable predicted by the model refers to the predicted activity of phenolic acid degrading enzymes in the fermentation products. , These are the independent variables that affect the fermentation outcome, i.e., the decision variables; For constant terms, the expected value of the response variable when all independent variables take the central level (i.e., the coded value is 0); The coefficient of the linear term represents the linear effect of the i-th factor on the response variable; The coefficient of the quadratic term represents the nonlinear effect of the i-th factor on the response variable; The interaction coefficient represents the interaction between factor i and factor j.

[0076] coefficient , , , The enzyme activity data was obtained using a Box-Behnken experimental design (BBD) combined with least squares regression. The specific steps were as follows: The lignin content, carbon-to-nitrogen ratio, and inoculum size of the waste were selected as independent variables, and the phenolic acid degrading enzyme activity was used as the response value to design a three-factor, three-level fermentation experiment. The experimentally measured enzyme activity data were imported into a multiple quadratic regression algorithm. By minimizing the sum of squared residuals, the regression coefficients were calculated, thereby constructing a predictive model that accurately describes the nonlinear relationship between process parameters and enzyme activity.

[0077] Microbial fermentation is a highly nonlinear biochemical process; the effect of a single factor on enzyme activity typically exhibits a bell-shaped curve rather than a nonlinear relationship. Therefore, a quadratic term is introduced to capture this parabolic characteristic. In agricultural waste fermentation, the factors are strongly coupled; for example, the degree of influence of the carbon-nitrogen ratio changes with the inoculum size. An interaction term is introduced to quantify this synergistic or antagonistic effect, which simple linear regression cannot achieve.

[0078] Meanwhile, the fitting of the carbon source release curve follows a modified first-order kinetic equation:

[0079] ;

[0080] in, The cumulative carbon mineralization at time t; The total potential mineralizable carbon in the waste is determined by the hemicellulose + cellulose content in the component characteristic data; k is the degradation rate constant, determined by the activity of the microbial agent. When the simulation calculation shows t=3 days... When the carbon source consumption rate is less than the critical carbon source consumption rate required for microbial growth, carbon starvation is determined, and a fast-acting carbon source replenishment command is automatically triggered.

[0081] In this embodiment, the first-level multi-objective optimization model is solved using the Non-Dominated Sorting Genetic Algorithm with Elite Strategy (NSGA-II). The specific algorithm parameters are set as follows:

[0082] Population size: set to 100 to ensure sufficient coverage of the initial solution space and avoid premature convergence due to an insufficiently small population.

[0083] Maximum number of generations: set to 200 generations. Preliminary experiments have verified that this number of iterations can ensure that the algorithm converges to a stable Pareto front.

[0084] Crossover probability: set to 0.9, using a simulated binary crossover operator, and the crossover distribution index is set to 20 to enhance the global search capability of the population;

[0085] Mutation probability: set to 0.1, using a multinomial mutation operator, and the mutation distribution index is set to 20 to maintain population diversity and prevent getting trapped in local optima;

[0086] Selection strategy: A binary tournament selection method is adopted; in each iteration, non-dominated solutions are screened based on the crowding distance sorting mechanism, and the final Pareto optimal solution set is the optimal set of process parameters that balances the maximization of phenolic acid degrading enzyme activity and waste resource utilization efficiency.

[0087] Enzyme fermentation is a complex and dynamic process. The real-time balance of the carbon-nitrogen ratio determines the enzyme yield. By fitting the release curve, the algorithm can predict nutrient bottlenecks and compensate in advance to ensure that microorganisms are always in the best metabolic state.

[0088] This solution addresses the problem that traditional static formulations cannot adapt to dynamic fermentation processes. Through simulation fitting and multi-objective optimization, it ensures that the generated process parameters are theoretically feasible and efficient, maximizing the activity of phenolic acid degrading enzymes while avoiding raw material waste.

[0089] The calculation method for the microecological stress index includes:

[0090] Obtain soil sample data after enzyme treatment during land preparation, including residual phenolic acid concentration, Fusarium gene copy number, and Trichoderma gene copy number;

[0091] The chemical stress component was obtained by dividing the residual phenolic acid concentration by the lethal concentration constant of Dictamnus dasycarpus seedlings.

[0092] Calculate the ratio of gene copy numbers of Fusarium to Trichoderma, and use the ratio to obtain the biological stress component through a logarithmic function mapping;

[0093] The fractal dimension data of soil aggregate structure were collected and normalized to obtain the physical structure components.

[0094] The entropy weights of the three components were calculated using the entropy weight method, and the weighted sum was defined as the dimensionless soil microecological pressure index. The higher the value, the greater the resistance of the soil to the survival of Dictamnus dasycarpus seedlings, and the more intensive intervention is needed in the next stage.

[0095] After the land preparation period, soil sample data were obtained. If the residual phenolic acid concentration was still slightly higher than the safe value (high chemical component), and the number of pathogenic Fusarium was greater than that of beneficial Trichoderma (high biological component), but the soil aggregate structure was good (low physical component), these three components were normalized to eliminate dimensional differences. Then, the entropy weight method was used to objectively assign weights based on the dispersion of each indicator data. If the calculated microecological pressure index was 0.8 (high value), it indicated that the soil environment still posed a huge threat to the survival of seedlings.

[0096] Among them, chemical stress component :

[0097] ;

[0098] in, The total residual concentration of phenolic acids was measured after land preparation. The lethal concentration constant for seedlings is set to 150 ug / g in this embodiment;

[0099] Biological stress component The ratio of Fusarium to Trichoderma was calculated using the logarithmic normalization method.

[0100] ;

[0101] in, This represents the copy number of the Fusarium gene. This represents the copy number of the Trichoderma gene. , These are the minimum and maximum statistical values ​​of this ratio recorded in the historical database, respectively.

[0102] Physical structural components :

[0103] ;

[0104] in, The fractal dimension of soil aggregate structure ranges from 2.0 to 3.0. A higher fractal dimension indicates a poorer structure, hence inverse normalization is used. The fractal dimension is a dimensionless value obtained by fitting the mass percentage of aggregates of different sizes.

[0105] Specifically, the following steps were taken: Uncirculated soil samples were collected from the 0-20cm soil layer of the plot to be tested. After air drying, the samples were broken into small pieces with a diameter of about 1cm along the natural structure. 50g of soil sample was weighed and placed on the top layer of a sieve with apertures of 2mm, 1mm, 0.5mm, and 0.25mm, respectively, thus separating the particles into five sizes: >2mm, 1-2mm, 0.5-1mm, 0.25-0.5mm, and <0.25mm. The sieve was then immersed in water and vibrated up and down at an amplitude of 3cm, a frequency of 30 times / minute, and a vibration time of 5 minutes. The aggregates on each sieve were collected, dried, and weighed to obtain the mass distribution of the aggregates in each size size. Based on a fractal model, a double logarithmic linear regression equation was established between the particle size distribution and the cumulative mass. The slope of the regression line was obtained, and the fractal dimension was obtained by subtracting the slope of the regression line from 3.

[0106] Finally, the entropy weight method is used to determine the weights, and the entropy weights of the three components are calculated to determine the final microecological pressure index:

[0107] ;

[0108] MPI stands for Microbial Stress Index, with a value range of [0,1]. The larger the value, the greater the environmental stress. , , These represent the corresponding weights.

[0109] Because the main problems of different plots are different, some plots are dominated by diseases, resulting in large dispersion of biological data; others are dominated by compaction, resulting in large dispersion of physical data. The entropy weight method automatically assigns weights based on the degree of data variation. The greater the fluctuation of an indicator, the more discriminative information it provides, and the higher its weight, thus achieving the adaptability of the evaluation system.

[0110] This invention utilizes a comprehensive evaluation method from systems engineering theory. A single indicator cannot fully reflect the soil health status. The entropy weight method uses the principle of information entropy to objectively determine the weight of each indicator in the comprehensive evaluation, avoiding the subjectivity of human scoring. The micro-ecological pressure index, as a dimensionless comprehensive indicator, has become a key link between the land preparation period and the seedling stage.

[0111] This index enables the digital quantification of soil remediation effectiveness. It not only serves as an assessment of the previous stage of work but also provides precise environmental pressure inputs for decision-making in the next stage, enabling agricultural production decisions to shift from experience-driven to data-driven.

[0112] The processing logic of the second-level multi-objective optimization model based on the micro-ecological pressure index is as follows:

[0113] Set a critical threshold for the stress index. When the input microecological stress index is higher than the threshold, the model enters the defense priority calculation path: set the objective function to maximize the content of antibiotic secondary metabolites and chitinase in the fermentation product, and force the increase of the inoculation ratio parameter of antagonistic bacteria in the constraints, and output the formula instruction with high antibacterial activity.

[0114] When the input microecological pressure index is lower than the threshold, the growth-promoting priority calculation path is entered: the objective function is set to maximize the content of indoleacetic acid and amino acids in the fermentation product, and the inoculation ratio parameter of phosphorus-solubilizing and potassium-solubilizing bacteria is increased in the constraints, and the formula instruction with high nutritional activity is output.

[0115] To achieve a digital response of the functional properties of enzymes during the seedling stage to the effect of land preparation.

[0116] The algorithm receives a microecological stress index. If this value exceeds a set critical threshold, it automatically determines that the soil environment is deteriorating and triggers a defense-priority calculation path. At this point, the algorithm adjusts its objective function to maximize the production of *Bacillus polymyxa* (antibiotic-producing bacteria) and chitinase (lysing bacteria). Under constraints, the inoculation ratio of antagonistic bacteria is forcibly increased to over 5%. The final output instruction is to prepare a medicinal enzyme with high antibacterial activity, rather than a regular nutrient solution.

[0117] Crops employ different survival strategies under varying environmental stresses: under stress, they prioritize resilience and survival, while under comfortable conditions, they prioritize growth. The algorithm dynamically adjusts its optimization objectives based on the stress index, simulating the survival intelligence of plants and enabling on-demand customization of enzyme functions.

[0118] In this embodiment, the second-level multi-objective optimization model is the core algorithm module for achieving seedling-land synergy. Its construction and operation process includes constructing a second input matrix, which contains the physiological requirement parameters and environmental constraints of Dictamnus dasycarpus seedlings, specifically defined as:

[0119] The current age of the seedling determines its basic nutritional needs;

[0120] The expected root biomass growth rate is used as a benchmark for growth promotion targets;

[0121] The predicted average temperature during the seedling stage will affect the fermentation heat compensation.

[0122] The upper limit of the cost budget per ton for enzyme preparation during the seedling stage.

[0123] The optimal process parameter vector sought by the decision variable setting model includes:

[0124] Inoculation ratio (%) of growth-promoting bacteria (such as Bacillus amyloliquefaciens);

[0125] Inoculation rate (%) of antagonistic functional bacteria (such as Bacillus polymyxa);

[0126] Nutrient substrate (such as soybean meal hydrolysate) addition amount (L / ton);

[0127] The amount of inducing substrate (such as shrimp and crab shell powder) added (kg / ton).

[0128] To balance defense and growth, a general objective function is constructed. :

[0129] ;

[0130] The two sub-objective functions are defined as follows:

[0131] Defend sub-targets The normalized weighted sum of chitinase activity and antibiotic potency in fermentation broth was predicted, and its value was mainly positively correlated with the amount of antagonistic bacteria inoculum and the inducing substrate.

[0132] fertility goals The normalized weighted sum of the plant growth hormone IAA content and free amino acid content in the fermentation broth was predicted, and its value was mainly positively correlated with the inoculum size of growth-promoting bacteria and the nutrient substrate.

[0133] Defense weight With growth-promoting weight The weights are determined based on a piecewise continuous function of the microecological pressure index; and the sum of the defense weights and the growth-promoting weights is 1.

[0134] Two critical thresholds are set for the microecological stress index; including a safety threshold. This indicates a good soil environment; and the warning threshold. This indicates a harsh soil environment;

[0135] When the microecological stress index is between 0 and the safe threshold During this period, the soil microenvironment pressure is low, and the system focuses on promoting growth while retaining a minimum level of defense capabilities; at this time, the weight for promoting growth is set to 0.9, and the weight for defense is set to 0.1.

[0136] When the microecological stress index is at a safe threshold With warning threshold During this period, soil pressure is at a moderate level, and the defense weight increases linearly with the increase of the pressure index.

[0137] ;

[0138] ;

[0139] When the microecological stress index exceeds the warning threshold At this time, the soil is severely hampered by continuous cropping, and the plant enters a survival mode, sacrificing growth rate to ensure survival; at this time, the growth promotion weight is set to 0.1, and the defense weight is set to 0.9.

[0140] This solution demonstrates a high level of intelligence, avoiding blind fertilization or pesticide application. Instead, it determines the medicinal or supplementary properties of enzymes during the seedling stage based on the actual health condition of the soil, thereby significantly improving the survival rate of Dictamnus dasycarpus seedlings in complex soil environments.

[0141] Dynamic compensation of fermentation parameters based on environmental meteorological data, including:

[0142] The system receives real-time weather forecast data for the next week for the location of the fermentation workshop and extracts the environmental temperature and humidity change sequences. The temperature change sequence is input into a thermodynamic conduction model to calculate the heat loss rate of the fermentation tank. When the predicted environmental temperature is lower than the optimal metabolic temperature range for microorganisms, the system increases the proportion of high-calorific-value waste in the formula and outputs compensation data to extend the fermentation cycle. When the predicted environmental humidity is too high and affects the moisture content of the waste, the system adjusts the water addition parameter in the formula.

[0143] The system connects in real time to the meteorological interface of the fermentation workshop to obtain a forecast of a sudden drop of 10 degrees Celsius in temperature over the next week. This temperature sequence is then input into a thermodynamic conduction model to calculate the heat loss rate of the fermentation tank under no-intervention conditions. The model predicts that the temperature of the fermentation liquid will be lower than the optimal metabolic range for microorganisms. The formula is then adjusted to increase the proportion of high-calorific-value waste such as corn flour (utilizing bio-fermentation for heat generation), or an instruction is output to extend the fermentation cycle by 48 hours to compensate for the metabolic stagnation caused by the low temperature.

[0144] Thermodynamic conduction models are used to calculate the heat loss rate of fermentation tanks. The calculation formula is as follows:

[0145] ;

[0146] Where K is the overall heat transfer coefficient of the fermentation device, and the fermentation tank used in this embodiment is made of double-layer stainless steel, which was determined experimentally; A is the heat dissipation surface area of ​​the fermentation tank. The optimal metabolic temperature for microorganisms is set at 30 degrees Celsius. This represents the average ambient temperature for the future time period predicted by the weather forecast. When the heat production rate exceeds the rate of microbial metabolic heat production, calculate the heat deficit and, based on the bio-oxidative calorific value of each gram of high-calorific-value waste, calculate the additional mass of high-calorific-value substrate to be added and write it into the formulation instruction.

[0147] Environmental temperature fluctuations are a key external factor affecting enzyme quality. Through predictive control algorithms, environmental changes can be detected in advance and feedforward compensation can be performed. By using biothermal or time-for-space methods, the stability of the fermentation process can be maintained.

[0148] This solution addresses the problem of unstable enzyme fermentation quality in open or semi-open environments. Especially in northern Dictamnus dasycarpus planting areas, it can effectively cope with the impact of extreme weather such as late spring frosts on fermentation, ensuring that the technical indicators of the final enzyme product remain consistent despite weather changes.

[0149] Closed-loop correction based on seedling feedback data includes:

[0150] After the seedling stage, data on the survival rate, root development index, and root rot incidence of Dictamnus dasycarpus seedlings were collected to obtain posterior data. The posterior data were then combined with the previous soil microecological pressure index and the second-stage fermentation process parameters to form a training sample pair. A deep learning algorithm was used to train a formula efficacy prediction model, and the deviation between the actual growth data and the model prediction was calculated. If the deviation exceeded the preset range, the weight coefficient matrix in the second-level optimization model was adjusted in reverse.

[0151] The root development index is a composite indicator reflecting the overall growth status of seedling roots, and it is quantitatively measured using the WinRHIZO root analysis system. The specific steps and calculation formulas are as follows:

[0152] Sample acquisition and processing: The roots of complete Dictamnus dasycarpus seedlings were collected by excavation, and a soil column with a radius of 20 cm and a depth of 30 cm centered on the base of the stem was retained. The roots were gently rinsed with slow-flowing water through a 0.5 mm mesh screen until the roots were completely exposed and free of soil. The surface moisture was then absorbed with filter paper.

[0153] Image scanning and analysis: After cleaning, the roots were placed in a transparent resin tray with a small amount of water to allow them to spread out naturally, avoiding overlap. An Epson Expression 12000XL professional scanner was used with WinRHIZO Pro image analysis software for scanning. Three core parameters were automatically extracted: total root length, total root surface area, and number of root tips.

[0154] Index Calculation: In order to eliminate the difference in dimensions, the three parameters above are first normalized by Min-Max, and then weighted and summed to obtain the root development index. The weights are all defaulted to 1 / 3.

[0155] Furthermore, the formulation efficacy prediction model is trained using a strategy that combines historical data warm-start with sample augmentation. The specific training data configuration and hyperparameter settings are as follows:

[0156] Construction of the training dataset:

[0157] Basic Database: The planting log data of Dictamnus dasycarpus accumulated in the planting base over the past 3 years was integrated and cleaned to obtain a total of 1,200 effective basic sample pairs, including growth data under different climate years and different fertilization schemes.

[0158] Data augmentation: To prevent model overfitting, data augmentation is performed on the basic samples. Specifically, while keeping the input feature covariance structure unchanged, Gaussian white noise following the N(0,0.05^2) distribution is introduced to expand the training sample size to 12,000 sets, which are used as the initial training set for the model.

[0159] Training cycle and hyperparameter settings:

[0160] Iteration rounds: The maximum number of iterations is set to 2,000; an early stopping mechanism is introduced, which terminates training early when the mean square error of the validation set decreases by less than 10^{-5} within 50 consecutive epochs to prevent overfitting.

[0161] Batch size: Set to 64 to balance memory usage and gradient descent stability;

[0162] Update frequency: An incremental learning mode is adopted. After each seedling cycle, the newly collected posterior data is added to the dataset, and the model weights are fine-tuned in short cycles of 500 epochs to enable the model to adapt to the latest climate change and soil evolution trends.

[0163] After the seedling stage, the survival rate of *Dictamnus dasycarpus* seedlings was 95%, but the root development index was slightly low. These posterior data were combined with the previous microecological stress index and the parameters of the second-stage formulation to form a training sample. A deep learning algorithm was used to train a formulation efficacy prediction model, revealing that the actual root growth was lower than the model's prediction. After calculating the deviation, the algorithm adjusted the coefficient matrix of phosphate-solubilizing bacteria weights in the second-level optimization model through backpropagation, automatically increasing the priority of root-promoting bacteria in the next calculation of similar formulations.

[0164] The formula efficacy prediction model adopts a BP neural network architecture, and the specific network structure configuration is as follows:

[0165] Input layer: It has 8 neurons, corresponding to the input feature vector, including microecological stress index, nitrogen, phosphorus and potassium ratio, inoculum amount, fermentation time, average temperature, etc.

[0166] Hidden layers: Two hidden layers are set up. The first layer contains 16 neurons and the second layer contains 8 neurons. The ReLU function is used for both layers to improve the nonlinear fitting ability.

[0167] Output layer: It has 3 neurons, which correspond to the predicted seedling survival rate, root development index and disease incidence rate respectively.

[0168] Training parameters: The loss function is mean squared error, the optimizer is the Adam algorithm, and the learning rate is set to 0.001.

[0169] After each seedling cycle, the actual process parameters are used as input and the actual growth data are used as labels to train the network through backpropagation, updating the weight matrix and thus continuously correcting the model's understanding of the formula-effect mapping relationship.

[0170] This step utilizes a feedback training mechanism from deep learning. Agricultural ecosystems are extremely complex, and initial models cannot perfectly encompass all variables. By continuously introducing actual planting results as feedback signals, the system can self-evolve, correct model parameters, and gradually approach the optimal solution for a specific plot of land.

[0171] This solution endows the system with the ability to learn and continuously evolve. As usage time increases and data accumulates, the enzyme formulas generated by this system will become increasingly precise, adapting to the microenvironmental characteristics of specific farmers and plots, thus achieving true precision agriculture.

[0172] Example 2:

[0173] The present invention provides a method for adaptive generation and optimization of agricultural enzymes based on multi-source data fusion, comprising:

[0174] Soil physicochemical index data, rhizosphere autotoxic substance concentration data, and component characteristic data of agricultural waste to be treated were collected from the planting plots of Dictamnus dasycarpus, and normalized to form the first input matrix;

[0175] The first input matrix is ​​imported into the first-level multi-objective optimization model. The objective function is to maximize the degradation rate of autotoxic substances. The first-stage fermentation process parameters are calculated, including the waste mixing ratio, the inoculum amount of degradation bacteria and the temperature control curve. The execution command is then output.

[0176] Within a preset time window after the first phase of formulation instructions is completed, data on the concentration of residual autotoxic substances and the diversity of microbial communities in the soil are collected, and a microecological pressure index characterizing the degree of soil remediation is obtained through weighted calculation.

[0177] A second input matrix containing physiological requirement parameters of Dictamnus dasycarpus seedlings is constructed, and the microecological pressure index is used as a constraint boundary condition to be imported into the second-level multi-objective optimization model. According to the numerical range of the microecological pressure index, the weight coefficient of the objective function is dynamically adjusted to calculate the second-stage fermentation process parameters, including the nutrient substrate ratio, functional microbial community combination and fermentation time, and output the seedling enzyme preparation instructions.

[0178] To verify the actual effect of the adaptive generation and optimization method for agricultural enzymes based on multi-source data fusion of the present invention, a comparative experiment was conducted;

[0179] The experiment was conducted at a Dictamnus dasycarpus continuous cropping base. Three plots of land that had been continuously cropped with Dictamnus dasycarpus for five years were selected and numbered as Plot A, Plot B, and Plot C. Each plot had an area of ​​0.5 mu (approximately 333.3 m²). The basic soil conditions of the three plots were kept consistent, and the specific parameters are shown in Table 1.

[0180] Table 1 Basic Soil Condition Parameters

[0181]

[0182] During the trial, the field management of the three plots, including irrigation volume, weeding frequency, and light conditions, remained completely consistent, with the only difference being the preparation and application methods of the enzyme.

[0183] Experimental group: Enzymes were prepared using the adaptive generation and optimization method for agricultural enzymes based on multi-source data fusion proposed in this invention and applied to plot A;

[0184] Control group: The traditional general-purpose agricultural enzyme preparation method was used, with a static formula and no data fusion or dynamic optimization, and it was applied to plot B;

[0185] Blank group: No enzymes were applied, only routine management was carried out, serving as a baseline control, corresponding to plot C.

[0186] For the experimental group: During the first fermentation stage, i.e., the land preparation stage, soil physicochemical indicators, rhizosphere autotoxic substance concentrations, and agricultural waste component characteristics data of plot A were collected to form the first input matrix; preferably, the rhizosphere autotoxic substance concentrations include: p-hydroxybenzoic acid 42.3 μg / g, vanillic acid 28.7 μg / g, and ferulic acid 14.6 μg / g; agricultural wastes include corn stalks, fruit tree branches, and soybean meal;

[0187] The first-level multi-objective optimization model calculated the following process parameters: the mass ratio of corn stalks:fruit tree branches:soybean meal was 4:3:3; the inoculum amount of compound degrading microorganisms was 1.2%; and the temperature control curves were 28-32℃ (first 3 days), 30-33℃ (4-7 days), and 28-30℃ (8-15 days). Based on the weather forecast for the next week, the amount of high-calorific-value corn flour was increased by 5%, and the fermentation cycle was extended by 2 days. The compound degrading microorganisms included Bacillus subtilis and Aspergillus niger.

[0188] Microecological stress index calculation: 15 days after the first phase of enzyme application, soil samples were collected and the calculated microecological stress index (MPI) was 0.72, which is higher than the critical threshold of 0.5, indicating the entry into the defense priority path. Second phase fermentation, during the seedling stage, the second input matrix includes the physiological requirements parameters of *Dictamnus dasycarpus* seedlings, namely, a seedling age of 15 days and an expected root biomass growth rate of 1.8 g / plant / week. The second-level multi-objective optimization model outputs the following process parameters: 120 L / ton of soybean meal hydrolysate as the nutrient substrate; 6% inoculation ratio of the antagonistic bacterium *Bacillus polymyxa*; 3% inoculation ratio of functional bacteria; and a fermentation time of 8 days. The functional bacteria include phosphate-solubilizing bacteria and *Trichoderma*.

[0189] The control group used the traditional method: a conventional static formula was adopted, with a corn straw:soybean meal mass ratio of 7:3, a single microbial agent EM inoculation amount of 1.0%, a fermentation temperature controlled at 25-30℃, a fermentation cycle of 15 days, and no environmental compensation or dynamic adjustment; the same formula enzyme was directly applied during the seedling stage.

[0190] The experimental period was 60 days, from the application of enzymes during land preparation to the end of the seedling stage of Dictamnus dasycarpus. The detection time point was 15 days after the end of the land preparation period (T1): the soil micro-ecological restoration index was detected.

[0191] After the seedling stage (T2): Detect seedling growth and stress resistance indicators;

[0192] During fermentation (T0-T1): Record key quality indicators of the enzyme (enzyme activity, content of functional components).

[0193] The soil remediation effects at node T1 were compared, and data are shown in Table 2. The total phenolic acid degradation rates of the experimental group, control group, and blank group are as follows: Figure 2 As shown.

[0194] Table 2 Soil Remediation Effect Data

[0195]

[0196] After comparing the seedling growth at node T2 after the seedling period, data table 3 was obtained; the seedling survival rates of the experimental group, control group, and blank group are as follows: Figure 3 As shown.

[0197] Table 3 Seedling Generation Data Table

[0198]

[0199] Comparative experimental data clearly show that the soil remediation effect is significant: the total phenolic acid degradation rate in the experimental group reached 89.2%, which is 57.3% higher than that in the control group, and the microecological pressure index dropped to 0.32. The problem of soil microecological imbalance was effectively solved, which is far superior to the remediation efficiency of traditional methods.

[0200] Seedling growth and stress resistance were significantly improved: the survival rate of white peony root seedlings in the experimental group reached 92.5%, which was 35.4% higher than that in the control group. The incidence of root rot was only 3.2%, which was 82.7% lower than that in the control group. At the same time, the growth indicators such as fresh weight of single plants increased by more than 40%, which fully verified the technical effect of the synergistic optimization of the ground and seedlings.

[0201] Outstanding comprehensive benefits: The method of this invention achieves precise matching between soil remediation and seedling cultivation through multi-source data fusion and dynamic optimization. Compared with traditional methods, it has made breakthrough improvements in core objectives such as the treatment of continuous cropping obstacles of Dictamnus dasycarpus and the protection of seedling survival, demonstrating significant technical advantages.

[0202] It should be understood that the embodiments disclosed in this invention and the above description enable those skilled in the art to implement this invention. However, this invention is not limited to the embodiments mentioned above. It should be understood that those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this invention, and should all be included within the protection scope of this invention.

Claims

1. A method for adaptive generation and optimization of agricultural enzymes based on multi-source data fusion, characterized in that, The method includes: Soil physicochemical index data, rhizosphere autotoxic substance concentration data, and component characteristic data of agricultural waste to be treated were collected from the planting plots of Dictamnus dasycarpus, and normalized to form the first input matrix; The first input matrix is ​​imported into the first-level multi-objective optimization model. The objective function is to maximize the degradation rate of autotoxic substances. The first-stage fermentation process parameters are calculated, including the waste mixing ratio, the inoculum amount of degradation bacteria and the temperature control curve. The execution command is then output. Within a preset time window after the first phase of formulation instructions is completed, data on the concentration of residual autotoxic substances and the diversity of microbial communities in the soil are collected, and a microecological pressure index characterizing the degree of soil remediation is obtained through weighted calculation. The calculation method for the microecological stress index includes: Obtain soil sample data after enzyme treatment during land preparation, including residual phenolic acid concentration, Fusarium gene copy number, and Trichoderma gene copy number; The chemical stress component was obtained by dividing the residual phenolic acid concentration by the lethal concentration constant of Dictamnus dasycarpus seedlings. Calculate the ratio of gene copy numbers of Fusarium to Trichoderma, and use the ratio to obtain the biological stress component through a logarithmic function mapping; The fractal dimension data of soil aggregate structure were collected and normalized to obtain the physical structure components. The entropy weights of the three components were calculated using the entropy weight method, and the weighted sum was defined as the dimensionless soil microecological pressure index. The higher the value, the greater the resistance of the soil to the survival of Dictamnus dasycarpus seedlings, and the more intensive intervention is needed in the next stage. A second input matrix containing physiological requirement parameters of Dictamnus dasycarpus seedlings is constructed, and the microecological pressure index is used as a constraint boundary condition to be imported into the second-level multi-objective optimization model. According to the numerical range of the microecological pressure index, the weight coefficient of the objective function is dynamically adjusted to calculate the second-stage fermentation process parameters, including the nutrient substrate ratio, functional microbial community combination and fermentation time, and output the seedling enzyme preparation instructions.

2. The method for adaptive generation and optimization of agricultural enzymes based on multi-source data fusion as described in claim 1, characterized in that, The process for acquiring and processing rhizosphere autotoxic substance concentration data includes: Rhizosphere soil samples were collected from plots where Dictamnus dasycarpus was continuously cropped. Characteristic peak area data were extracted using liquid chromatography-mass spectrometry. The concentration values ​​of p-hydroxybenzoic acid, vanillic acid, and ferulic acid were identified and locked. The concentration values ​​were compared with the preset tolerance threshold of Dictamnus dasycarpus to calculate the multiple of each substance exceeding the standard. The multiple of exceeding the standard was used as the penalty factor weight in the first-level multi-objective optimization model.

3. The method for adaptive generation and optimization of agricultural enzymes based on multi-source data fusion as described in claim 2, characterized in that, include: Collect data on lignin content and carbon-to-nitrogen ratio of agricultural waste to establish a database of waste degradation potential. The molecular structure characteristics of autotoxic substances are matched with the degradation potential data of waste to screen out the types of waste that can induce the production of phenolic acid degrading enzymes. These types of waste are marked as recommended raw material data and stored in the first input matrix.

4. The method for adaptive generation and optimization of agricultural enzymes based on multi-source data fusion as described in claim 3, characterized in that, The computational logic of the first-level multi-objective optimization model includes: The decision variables are set as the mass ratio of different types of waste and the inoculation concentration of each single bacterium in the compound microbial agent; the objective function is constructed and defined as maximizing the predicted value of phenolic acid degrading enzyme activity in the fermentation product. In the first-level multi-objective optimization model, the component characteristic data of waste are called in during the operation to fit the carbon source release curve during the fermentation process. When the fitting result shows that the carbon source release rate is lower than the enzyme protein synthesis requirement, the feeding ratio parameter of the fast-acting carbon source is increased. The non-dominated sorting genetic algorithm is used to optimize the objective function and output a set of Pareto optimal solutions. Based on the current ambient temperature data, an optimal solution is automatically selected and converted into a digital control command containing specific stirring frequency and ventilation volume.

5. The method for adaptive generation and optimization of agricultural enzymes based on multi-source data fusion as described in claim 1, characterized in that, The processing logic of the second-level multi-objective optimization model based on the micro-ecological pressure index is as follows: Set a critical threshold for the stress index. When the input microecological stress index is higher than the threshold, the model enters the defense priority calculation path: set the objective function to maximize the content of antibiotic secondary metabolites and chitinase in the fermentation product, and force the increase of the inoculation ratio parameter of antagonistic bacteria in the constraints, and output the formula instruction with high antibacterial activity. When the input microecological pressure index is lower than the threshold, the growth-promoting priority calculation path is entered: the objective function is set to maximize the content of indoleacetic acid and amino acids in the fermentation product, and the inoculation ratio parameter of phosphorus-solubilizing and potassium-solubilizing bacteria is increased in the constraints, and the formula instruction with high nutritional activity is output. To achieve a digital response of the functional properties of enzymes during the seedling stage to the effect of land preparation.

6. The method for adaptive generation and optimization of agricultural enzymes based on multi-source data fusion as described in claim 1, characterized in that, Dynamic compensation of fermentation parameters based on environmental meteorological data, including: The system receives real-time weather forecast data for the next week for the location of the fermentation workshop and extracts the environmental temperature and humidity change sequences. The temperature change sequence is input into a thermodynamic conduction model to calculate the heat loss rate of the fermentation tank. When the predicted environmental temperature is lower than the optimal metabolic temperature range for microorganisms, the system increases the proportion of high-calorific-value waste in the formula and outputs compensation data to extend the fermentation cycle. When the predicted environmental humidity is too high and affects the moisture content of the waste, the system adjusts the water addition parameter in the formula.

7. The method for adaptive generation and optimization of agricultural enzymes based on multi-source data fusion as described in claim 1, characterized in that, Closed-loop correction based on seedling feedback data includes: After the seedling stage, data on the survival rate, root development index, and root rot incidence of Dictamnus dasycarpus seedlings were collected to obtain posterior data. The posterior data were then combined with the previous soil microecological pressure index and the second-stage fermentation process parameters to form a training sample pair. A deep learning algorithm was used to train a formula efficacy prediction model, and the deviation between the actual growth data and the model prediction was calculated. If the deviation exceeded the preset range, the weight coefficient matrix in the second-level optimization model was adjusted in reverse.

Images