A tea garden fertility dynamic balance compensation method, system, device and medium

Abstract

The application relates to a tea garden fertility dynamic balance compensation method, system, device and medium. The method comprises the following steps: acquiring historical meteorological data, historical soil layering detection data, historical tea tree picking data, tea tree canopy multispectral image data and fertilizer attribute data of a tea garden and constructing a soil fertility time-varying model to calculate soil layer effective nutrient time sequence deduction values; based on the tea tree canopy multispectral image data, tea tree physiological state parameters are obtained through inversion; based on the soil layer effective nutrient time sequence deduction values and the tea tree physiological state parameters, theoretical fertility time sequence values are calculated; a nutrient demand period is extracted, and based on the nutrient demand period, the fertilizer attribute data and the theoretical fertility time sequence values, nutrient supply-demand matching is performed to generate a nutrient release scheme and generate a fertility compensation execution instruction; and model parameters of the soil fertility time-varying model are optimized according to an execution effect deviation. The method can realize accurate dynamic compensation of tea garden fertility.

Description

Technical Field

[0001] This invention belongs to the field of intelligent tea garden fertilization management, and in particular relates to a method, system, equipment and medium for dynamic balance compensation of fertility in tea gardens. Background Technology

[0002] With the development of intelligent tea garden fertilization management technology, precision fertilization technology based on soil testing and crop growth models has emerged. This type of technology attempts to formulate scientific fertilization plans by collecting soil nutrient data and tea tree growth indicators. In traditional tea garden fertilization management, compound fertilizers are usually applied regularly and in fixed quantities, or rough judgments are made based on farmers' experience.

[0003] Traditional techniques primarily involve manually collecting soil samples for laboratory analysis, and then developing relatively fixed fertilization plans based on the general growth patterns of tea trees. This method relies on discrete soil testing results and static agricultural experience, lacking the ability to continuously monitor and accurately predict the dynamic changes in soil nutrients.

[0004] However, current traditional methods or existing technologies have the following problems: they are difficult to fully consider the time-varying effects of meteorological conditions such as temperature and precipitation on soil nutrient mineralization and leaching processes, making it difficult to accurately grasp the natural ebb and flow of soil nutrients; they fail to effectively integrate real-time physiological state parameters retrieved from multispectral images of tea tree canopies, resulting in a significant deviation between fertilization decisions and the actual nutrient requirements of tea trees; and they lack a dynamic matching mechanism between fertilizer release characteristics and the stage-specific nutrient requirements of tea trees, which can easily lead to over- or under-fertilization, wasting fertilizer resources and potentially causing environmental pollution and a decline in tea quality. Summary of the Invention

[0005] Therefore, it is necessary to provide a method, system, equipment, and medium for dynamic balance compensation of fertility in tea gardens that can solve the above problems.

[0006] Firstly, this application provides a method for dynamic balance compensation of fertility in tea gardens, including:

[0007] Acquire historical meteorological data, historical soil stratification data, historical tea tree harvesting data, tea tree canopy multispectral image data, and fertilizer attribute data for the tea garden;

[0008] Based on historical meteorological data, historical soil stratification detection data, and historical tea tree harvesting data, a first soil fertility time-varying model was constructed, and based on the first soil fertility time-varying model, the temporal inference value of effective nutrients in the soil layer was calculated.

[0009] Based on multispectral image data of tea tree canopy, physiological state parameters of tea trees were obtained through inversion processing of tea tree physiological state.

[0010] Based on the temporal projection values ​​of effective nutrients in the soil layer and the physiological state parameters of tea trees, the theoretical fertility temporal values ​​are calculated through the preset physiological-soil demand mapping relationship of tea trees.

[0011] Based on historical tea tree harvesting data, the nutrient demand period is extracted. Based on the nutrient demand period, fertilizer attribute data and theoretical fertility time series value, a nutrient release plan is generated through nutrient supply and demand matching. Based on the nutrient release plan, a fertility compensation execution instruction is generated.

[0012] The execution effect deviation of the fertility compensation execution command is obtained, and the model parameters of the first soil fertility time-varying model are optimized based on the execution effect deviation to obtain the second soil fertility time-varying model. The second soil fertility time-varying model is used to generate the time series extrapolation value of effective nutrients in the soil layer.

[0013] Secondly, this application also provides a dynamic balance compensation system for tea garden fertility, comprising:

[0014] The tea garden data acquisition module is used to acquire historical meteorological data, historical soil stratification detection data, historical tea tree harvesting data, tea tree canopy multispectral image data, and fertilizer attribute data of the tea garden.

[0015] The extrapolation model construction module is used to construct the first soil fertility time-varying model based on historical meteorological data, historical soil stratification detection data and historical tea tree harvesting data, and to calculate the temporal extrapolation value of effective nutrients in the soil layer based on the first soil fertility time-varying model.

[0016] The physiological parameter inversion module is used to obtain the physiological state parameters of tea trees by inverting the physiological state of tea trees based on multispectral image data of the tea tree canopy.

[0017] The theoretical fertility calculation module is used to calculate the theoretical fertility time series value based on the temporal extrapolation value of effective nutrients in the soil layer and the physiological state parameters of tea trees, through the preset physiological-soil demand mapping relationship of tea trees;

[0018] The release plan generation module is used to extract the nutrient demand period based on historical tea tree harvesting data, and generate a nutrient release plan by matching nutrient supply and demand based on the nutrient demand period, fertilizer attribute data and theoretical fertility time series value, and generate fertility compensation execution instructions based on the nutrient release plan.

[0019] The model parameter optimization module is used to obtain the execution effect deviation of the fertility compensation execution command, and optimize the model parameters of the first soil fertility time-varying model based on the execution effect deviation to obtain the second soil fertility time-varying model. The second soil fertility time-varying model is used to generate the time series extrapolation value of effective nutrients in the soil layer.

[0020] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-described tea garden fertility dynamic balance compensation method.

[0021] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described tea garden fertility dynamic balance compensation method.

[0022] The aforementioned method, system, equipment, and medium for dynamic balance compensation of tea garden fertility comprehensively capture key factors affecting soil fertility, such as meteorological, soil, tea tree growth, and fertilizer characteristics, by acquiring multi-source heterogeneous data from tea gardens, laying a data foundation for subsequent modeling. Based on historical meteorological, soil stratification, and harvesting data, a first time-varying soil fertility model is constructed, and the temporal extrapolation value of effective nutrients in the soil layer is calculated, enabling dynamic simulation of the time-varying patterns of soil nutrients under the dual effects of natural growth and decline and harvest removal. Physiological state parameters of tea trees are obtained by inverting multispectral images of the tea tree canopy, acquiring the true nutrient demand signal of the tea trees. Combined with soil... The effective nutrient time series projection value of each layer is used to calculate the theoretical fertility time series value based on the physiological state parameters of tea trees, achieving a precise mapping between soil supply capacity and the actual needs of tea trees. Nutrient demand periods are extracted based on historical harvesting data and matched with fertilizer attribute data to generate nutrient release plans, achieving synchronous coordination between fertilizer release rhythm and the tea tree's stage absorption peaks, solving the problems of time mismatch and improper ratio in traditional fertilization. Based on the feedback of execution effect deviation, the parameters of the soil fertility time-varying model are optimized, forming a closed-loop mechanism of perception-decision-execution-optimization, enabling the model to continuously self-adapt and correct, and improving the accuracy and reliability of subsequent fertility compensation. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a flowchart of a tea garden fertility dynamic balance compensation method according to the present invention;

[0025] Figure 2 This is a diagram illustrating the steps involved in generating a nutrient release scheme for a tea garden fertility dynamic balance compensation method according to one embodiment of the present invention.

[0026] Figure 3 This is a structural diagram of a dynamic balance compensation system for tea garden fertility according to the present invention. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0028] In one embodiment, such as Figure 1 As shown, a dynamic balance compensation method for tea garden fertility is provided. This embodiment illustrates the application of this method to a compensation terminal. It is understood that this method can also be applied to a server, or to a system including both a compensation terminal and a server, and is implemented through the interaction between the compensation terminal and the server. Specifically, the compensation terminal can be an intelligent monitoring device or mobile management terminal deployed in the tea garden, with a built-in sensor module for collecting meteorological data, soil stratification temperature and humidity and nutrient content, multispectral images of the tea tree canopy, and historical harvesting records, etc. The server is responsible for storing the above-mentioned multi-source heterogeneous data and running a soil fertility time-varying model, nutrient demand analysis algorithm, and fertilization scheme generation engine. In practical applications, when tea garden managers need to formulate precise fertilization plans, the compensation terminal can upload real-time collected meteorological, soil, multispectral imagery, and harvesting data to the server. The server then calls a pre-trained model to calculate the temporal projection values ​​of effective nutrients in the soil layer and the physiological state parameters of the tea trees, thereby generating theoretical fertility temporal values ​​and matching fertilizer release schemes. The server can issue fertilization instructions to the compensation terminal, which can then connect to automatic fertilization equipment to perform the operation. After execution, the compensation terminal can collect effect data again and feed it back to the server for iterative optimization of model parameters, forming a closed-loop collaborative working mode from data perception to decision execution and then to feedback correction. This is suitable for the intelligent and automated fertilization management needs of large-scale tea gardens.

[0029] In this embodiment, the method includes the following steps:

[0030] S01, acquire historical meteorological data, historical soil stratification data, historical tea tree harvesting data, tea tree canopy multispectral image data, and fertilizer attribute data of the tea garden.

[0031] Optionally, historical meteorological data refers to the time-series records of meteorological elements such as temperature, precipitation, and sunshine in the tea garden area over the past few years, which can be obtained through meteorological stations or remote sensing data; historical soil stratification detection data refers to structured datasets obtained by regularly sampling and analyzing the physicochemical properties of different soil layers, such as the surface and subsurface layers, including nitrogen, phosphorus, potassium content, and organic matter, to characterize the spatiotemporal distribution characteristics of soil nutrients; historical tea tree harvesting data refers to time-series data recording the time, batch, yield, and corresponding nutrient removal amount of each harvest, reflecting the total amount of nutrients removed during tea tree growth due to harvesting; tea tree canopy multispectral image data refers to tea tree canopy reflectance images obtained by multispectral sensors carried by drones or satellites, which can include multiple bands and are used to invert the physiological and biochemical parameters of tea trees; fertilizer attribute data refers to characteristic parameters such as the type, composition, nutrient release rate curve, release mode, and effective utilization rate of the applied fertilizer. The compensation terminal can acquire historical meteorological data, historical soil stratification detection data, historical tea tree harvesting data, tea tree canopy multispectral image data, and fertilizer attribute data of tea gardens, providing a complete and reliable data foundation for subsequent construction of time-varying soil fertility models, inversion of tea tree physiological state parameters, and generation of nutrient release schemes.

[0032] S02. Based on historical meteorological data, historical soil stratification detection data and historical tea tree harvesting data, a first soil fertility time-varying model was constructed, and based on the first soil fertility time-varying model, the temporal inference value of effective nutrients in the soil layer was calculated.

[0033] Optionally, the first soil fertility time-varying model is a mathematical model that can dynamically simulate the continuous change of soil nutrients over time under the dual effects of natural growth and decline and human harvesting and removal. This mathematical model breaks through the limitations of traditional static or semi-static soil nutrient assessment methods by coupling the time-varying driving effects of meteorological factors such as temperature and precipitation on soil nutrient mineralization and leaching processes, as well as the amount of nutrient removal caused by tea tree harvesting.

[0034] Specifically, the compensation terminal can use the first soil fertility time-varying model to decompose historical meteorological data into multiple scales to extract excitation components at different time scales. Combined with soil stratification detection data, it can use physicochemical mechanisms such as mineralization kinetic equations and convection diffusion equations to fit the nutrient mineralization rate parameters and leaching migration parameters of each soil layer under different conditions, thereby establishing a stratified multi-scale soil nutrient natural balance sub-model. At the same time, the compensation terminal can extract the time-series information of harvested nutrient removal from historical tea tree harvesting data as an external input and couple it to the above-mentioned soil nutrient natural balance sub-model to form a first soil fertility time-varying model that can comprehensively reflect the dynamic balance of soil nutrient supply and consumption.

[0035] Based on the first soil fertility time-varying model, the compensation terminal can, by inputting meteorological driving data and harvesting plans at corresponding time points, generate temporal variation curves of available nutrients in each soil layer for future periods, i.e., the temporal projection values ​​of available nutrients in the soil layer. These temporal projection values ​​represent the expected trend of changes in the content of available nutrients in the soil that can be absorbed and utilized by tea trees under the combined influence of natural conditions and agricultural activities. This step, by introducing meteorological time-varying driving and harvest removal feedback, achieves a characterization of the continuous evolution of soil fertility, improving the accuracy of nutrient prediction.

[0036] S03, based on multispectral image data of tea tree canopy, obtains the physiological state parameters of tea trees through inversion processing of tea tree physiological state.

[0037] Optionally, tea canopy multispectral image data refers to tea canopy reflectance images acquired simultaneously in multiple specific wavelength bands, such as blue, green, red, and near-infrared, using drones, satellites, or ground-based imaging equipment equipped with multispectral sensors. These images capture the absorption and reflection characteristics of leaves at different wavelengths of light, thus indirectly reflecting the biochemical composition and structural information of the leaves. Tea tree physiological state inversion processing refers to a process that uses algorithms such as radiative transfer theory, statistical regression models, or deep learning networks to convert the reflectance values ​​of each pixel in the multispectral image into biologically meaningful physiological parameters. Common tea tree physiological state parameters include leaf nitrogen content, chlorophyll content, and leaf area index. These parameters are directly related to the tea tree's nutritional level, photosynthetic capacity, and growth vigor.

[0038] In practice, the compensation terminal can extract reflectance data from each band of multispectral images, preprocess the images to eliminate environmental interference, and input the preprocessed reflectance data into a pre-trained inversion model. This model establishes a mapping relationship between reflectance and physiological parameters based on measured samples and can output a spatial distribution map of physiological parameters at the whole field scale. The compensation terminal can perform spatial smoothing and interpolation processing on the spatial distribution map of physiological parameters to obtain continuous physiological state parameters of tea trees, providing timely and accurate physiological feedback signals for subsequent matching of soil nutrient supply and actual needs of tea trees.

[0039] S04, based on the temporal projection of effective nutrients in the soil layer and the physiological state parameters of tea trees, calculates the theoretical fertility temporal value through the preset physiological-soil demand mapping relationship of tea trees.

[0040] Optionally, the temporal projection value of effective nutrients in the soil layer refers to the expected change curve of the effective nutrient content of each soil layer over time, output by the first soil fertility time-varying model, under the combined effects of natural growth and decline and harvest removal, reflecting the soil's own nutrient supply potential. The tea tree physiological-soil layer demand mapping relationship refers to a conversion rule or model that transforms the physiological demand signal of the tea tree's aboveground parts into the amount of nutrients that should be supplemented in each underground soil layer. This mapping relationship can be established based on various technical paths, such as: using the three-dimensional topology of the tea tree root system and the root vitality index of the soil layer for a weighted allocation of physiological state-root function, combining prior knowledge of phenological periods and the nutrient antagonism coefficient matrix, and learning the parameter set of the dynamic mapping through a time-series coding model or a multi-head self-attention mechanism. Optionally, the tea tree physiological-soil layer demand mapping relationship can also be obtained through reasoning using a mechanistic model based on radiative transfer theory, a statistical regression model based on a large amount of measured data, or by using empirical rules from an expert knowledge base. The theoretical fertility time series value is an idealized time series curve of the total amount of nutrients that need to be applied to the soil at each time point to enable the tea tree to reach its optimal growth state, obtained after processing the physiological-soil demand mapping relationship of tea trees.

[0041] The compensation terminal can couple the objective supply capacity of the soil with the subjective physiological needs of the tea trees in two directions, rather than simply supplementing according to the amount of soil deficiency. Through a dynamic mapping mechanism, it can achieve on-demand supply, so that the fertilization target matches the actual absorption rhythm of the tea trees, avoiding the problems of over- or under-fertilization caused by the mismatch between supply and demand in traditional methods.

[0042] S05. Based on historical tea tree harvesting data, the nutrient demand period is extracted. Based on the nutrient demand period, fertilizer attribute data and theoretical fertility time series value, a nutrient release plan is generated through nutrient supply and demand matching. Based on the nutrient release plan, a fertility compensation execution instruction is generated.

[0043] Optionally, historical tea tree harvesting data records information such as the time, batch, and yield of each harvest. This allows analysis of the peak and trough periods of nutrient absorption during the tea tree's growth cycle, thus defining the critical periods requiring additional nutrient supplementation, i.e., nutrient demand periods. Fertilizer attribute data refers to the inherent characteristics of the applied fertilizer, including nutrient release rate curves (describes the change in the rate of nutrient release in the soil over time), nutrient release patterns (attributes of nutrient release, such as fast-acting, slow-release, and controlled-release), and fertilizer nutrient utilization efficiency. These attributes determine the rhythm and total amount of nutrients that the fertilizer can provide to the soil after actual application. Nutrient supply and demand matching refers to the temporal alignment and quantitative comparison between the demand-side curve represented by theoretical fertility time-series values ​​and the supply-side curve determined by fertilizer attribute data. By adjusting the timing, amount, frequency, and combination of fertilizer types, the actual nutrient release distribution is made to match the peak demand of the tea tree in terms of time, generating a specific nutrient release plan.

[0044] The compensation terminal can divide the fertilization cycle into multiple cycles, analyze the nutrient release ratio and corresponding fertilizer demand for each cycle, generate a nutrient release plan, and generate a fertility compensation execution command based on the nutrient release plan. The fertility compensation execution command can be directly sent to the automatic fertilization equipment or provided to the management personnel as an operation guide to achieve precise compensation of soil nutrients in tea gardens.

[0045] S06, obtain the execution effect deviation of the fertility compensation execution command, and optimize the model parameters of the first soil fertility time-varying model based on the execution effect deviation to obtain the second soil fertility time-varying model. The second soil fertility time-varying model is used to generate the time series extrapolation value of effective nutrients in the soil layer.

[0046] Optionally, the execution effect deviation refers to the difference between the actual soil nutrient changes obtained through on-site sampling or remote sensing monitoring after fertilization operations are completed according to the fertility compensation execution instructions, and the theoretical values ​​predicted by the first soil fertility time-varying model. This deviation can be expressed as the difference in nutrient content at different soil layers and different time points, reflecting the degree of deviation between the model prediction and the actual state. Optimizing model parameters based on the execution effect deviation refers to using parameter correction methods in feedback control or machine learning to adjust and update key parameters in the first soil fertility time-varying model with the goal of reducing the error between predicted and measured values, thereby obtaining a second soil fertility time-varying model with higher accuracy.

[0047] Specifically, the compensation terminal can decompose the deviation of the execution effect according to soil depth and time scale, identify the sensitive parameters that contribute the most to the deviation, and iteratively correct the sensitive parameters through optimization strategies such as gradient descent, Kalman filtering, Bayesian inference, or genetic algorithms until the model output achieves satisfactory consistency with the actual observation data. The second soil fertility time-varying model is the updated model obtained after the above parameter optimization. It will have higher accuracy and adaptability in subsequent fertility predictions. The compensation terminal continuously performs adaptive calibration and optimization of the soil fertility time-varying model based on the actual fertilization effect, thereby continuously narrowing the gap between prediction and reality and improving the stability and accuracy of long-term operation.

[0048] The aforementioned method for dynamic balance compensation of tea garden fertility comprehensively captures key factors affecting soil fertility, such as meteorological, soil, tea tree growth, and fertilizer characteristics, by acquiring multi-source heterogeneous data from tea gardens, laying a data foundation for subsequent accurate modeling. Based on historical meteorological, soil stratification, and harvesting data, a first time-varying soil fertility model is constructed, and the temporal inference value of effective nutrients in the soil layer is calculated. This dynamically simulates the time-varying pattern of soil nutrients under the dual effects of natural growth and decline and harvest removal, overcoming the deficiency of traditional static detection in reflecting continuous nutrient changes. Furthermore, by inverting multispectral images of the tea tree canopy, physiological state parameters of the tea trees are obtained, enabling real-time and non-destructive acquisition of the true nutritional needs of the tea trees, avoiding reliance on experience-based judgments. This addresses the supply-demand mismatch caused by soil nutrient availability. By combining the temporal projections of effective nutrients in the soil layer with the physiological parameters of tea trees to calculate theoretical fertility temporal values, a precise mapping between soil supply capacity and the actual needs of tea trees is achieved, making fertilization targets more targeted. Nutrient demand periods are extracted from historical harvesting data and matched with fertilizer attribute data to generate nutrient release plans. This enables the synchronization and coordination of fertilizer release rhythm with the tea tree's periodic absorption peaks, solving the problems of time mismatch and improper proportions in traditional fertilization. Based on feedback from execution effect deviations, the parameters of the soil fertility time-varying model are optimized, forming a closed-loop mechanism of perception-decision-execution-optimization, enabling the model to continuously self-adapt and correct, thereby continuously improving the accuracy and reliability of subsequent fertility compensation.

[0049] In one embodiment, a first time-varying soil fertility model is constructed based on historical meteorological data, historical soil stratification detection data, and historical tea tree harvesting data, including:

[0050] S11, perform multi-scale decomposition on historical meteorological data to obtain multi-scale excitation component sequences, which include temperature-related components and precipitation-related components.

[0051] S12, based on historical soil stratification detection data and temperature correlation components, stratification fitting is performed through first-order mineralization kinetic equation to obtain a multi-scale nutrient mineralization rate parameter set for each soil layer at different time scales.

[0052] S13, based on historical soil stratification detection data and precipitation correlation components, stratified fitting was performed through convection diffusion equation to obtain the multi-scale nutrient leaching and migration parameter set of each soil layer at different time scales;

[0053] S14, parameter coupling mapping is performed on the multi-scale nutrient mineralization rate parameter set and the multi-scale nutrient leaching and migration parameter set to construct a hierarchical multi-scale soil nutrient natural growth and decline sub-model;

[0054] S15, based on historical tea tree harvesting data, extract the time series data of nutrient removal during harvest;

[0055] S16. Input the time series data of harvested nutrient removal into the hierarchical multi-scale soil nutrient natural growth and decline sub-model, and couple to obtain the first soil fertility time-varying model.

[0056] Specifically, in S11, the compensation terminal can perform multi-scale decomposition on historical meteorological data to obtain multi-scale excitation component sequences, including temperature-related components and precipitation-related components. Multi-scale decomposition refers to using signal processing methods such as wavelet transform or empirical mode decomposition to break down raw meteorological time-series data, such as daily average temperature and daily precipitation, into subsequences at different frequencies or time scales. Among them, the temperature-related component can reflect the impact of different cycles, such as seasonal temperature differences and diurnal temperature differences, on soil biochemical processes, while the precipitation-related component can distinguish the different effects of heavy rain events and drizzle on water infiltration. Multi-scale decomposition can quantify how meteorological fluctuations at different time scales independently and superimposedly drive the transformation and migration of soil nutrients.

[0057] Optionally, in S12, the compensation terminal can perform stratified fitting based on historical soil stratification detection data and temperature correlation components, using the first-order mineralization kinetic equation, to obtain a multi-scale nutrient mineralization rate parameter set for each soil layer at different time scales. The first-order mineralization kinetic equation can be expressed as:

[0058] ;

[0059] in The amount of nutrients mineralized at time t. Here, k represents the potential mineralizable nutrients, and k is the mineralization rate constant. The compensation terminal uses temperature-related components at different scales as driving variables, substituting them into the equation to perform piecewise or parallel fitting of the mineralization process of each soil layer (such as the topsoil layer of 0-20cm and the subsoil layer of 20-40cm), obtaining a set of rate parameters that vary with the temperature scale. This is more effective than the traditional single rate parameter in characterizing the complex impact of temperature fluctuations on microbial activity.

[0060] Optionally, in S13, the compensation terminal can perform stratified fitting based on historical soil stratification detection data and precipitation-related components, using the convection dispersion equation to obtain a multi-scale nutrient leaching and migration parameter set for each soil layer at different time scales. The convection dispersion equation (CDE) describes the transport of solutes in porous media, and its simplified form can be:

[0061] ;

[0062] Where C is the nutrient concentration, D is the dispersion coefficient, v is the pore water velocity, and z is the depth. This step also introduces the concept of multi-scale, using precipitation-related components at different scales, such as the pulsed infiltration of short-duration heavy rainfall and the steady infiltration of long-duration light rain, as boundary conditions. These are then solved or fitted for each soil layer to obtain a set of dispersion coefficients and velocity parameters related to the precipitation scale. This allows for a more refined description of the risk and path of nutrient leaching downwards under different rainfall patterns.

[0063] Optionally, in S14, the compensation terminal can perform parameter coupling mapping on the multi-scale nutrient mineralization rate parameter set and the multi-scale nutrient leaching and migration parameter set to construct a hierarchical multi-scale soil nutrient natural ebb and flow sub-model. The compensation terminal can integrate the nutrient increment caused by mineralization and the nutrient reduction caused by leaching at the same time step and spatial point by constructing a joint probability distribution or a nonlinear mapping function. In this way, the hierarchical multi-scale soil nutrient natural ebb and flow sub-model can comprehensively consider the source term of mineralization and the sink term of leaching at different time scales within each soil layer, simulating the natural ebb and flow dynamics of soil nutrients with time and depth under the condition of no human intervention.

[0064] Optionally, in S15, the compensation terminal can extract time-series data on nutrient removal during harvest based on historical tea tree harvesting data. For example, the compensation terminal can obtain the fresh or dry weight of tea leaves recorded for each harvest, multiply it by the pre-determined average content coefficients of major nutrients such as nitrogen, phosphorus, and potassium in the tea leaves, and convert the harvest yield data into the total amount of nutrients removed from the tea garden ecosystem. The time-series data on nutrient removal during harvest indicates the instantaneous consumption of the soil nutrient pool caused by each harvesting event.

[0065] Optionally, in S16, the compensation terminal can use the harvested nutrient removal time-series data as an external forcing term and input it into the hierarchical multi-scale soil nutrient natural fluctuation sub-model to couple and obtain the first soil fertility time-varying model. The coupling method can be: adding a time-related negative source term to the original nutrient balance equation of the hierarchical multi-scale soil nutrient natural fluctuation sub-model; that is, at the moment of harvest, directly subtracting the corresponding removal amount from the simulated nutrient storage of the corresponding soil layer (assuming nutrients are mainly removed from the surface layer). Through this coupling, the first soil fertility time-varying model can simultaneously respond to meteorological-driven natural processes and agricultural activities, such as human disturbances caused by harvesting, achieving continuous and dynamic extrapolation of the changes in soil available nutrient content over time, providing reliable soil-side supply predictions for fertilization decisions.

[0066] In one embodiment, based on multispectral image data of the tea tree canopy, physiological state parameters of the tea tree are obtained through tea tree physiological state inversion processing, including:

[0067] S21. Based on the multispectral image data of tea tree canopy, the light reflectance of multiple bands is extracted, and the light-shadow classification of the multispectral image data of tea tree canopy is performed based on the light reflectance to obtain the direct light area and the shadow area.

[0068] S22, based on the input of light reflectance, direct light area and shadow area into the physical-guided inversion network, the optimization inversion is performed with the goal of minimizing the reflectance residual, and a preliminary spatial distribution map of physiological parameters is obtained. The physical-guided inversion network is constructed by embedding the positive physical mapping relationship between reflectance and tea tree physiological state parameters, with radiative transfer theory as the physical prior constraint.

[0069] S23, smooth the spatial distribution map of preliminary physiological parameters and output the physiological state parameters of tea trees, including leaf nitrogen content, chlorophyll content and leaf area index.

[0070] For example, in S21, the compensation terminal can extract the light reflectance of multiple bands based on the multispectral image data of the tea canopy, and perform light-shade classification on the tea canopy multispectral image data based on the light reflectance to obtain directly illuminated areas and shaded areas. Light reflectance refers to the surface reflectance value obtained after radiometric calibration and atmospheric correction of pixel brightness values ​​in multiple bands such as blue, green, red, and near-infrared, reflecting the reflection characteristics of the tea canopy to different wavelengths of light. Light-shade classification utilizes the difference in reflectance between the visible and near-infrared bands to divide each pixel in the image into directly illuminated areas or shaded areas formed by other objects, for example, by constructing the Normalized Difference Shading Index (NDSI) or a decision tree based on multi-band thresholds. The reflectance of shaded areas is affected by reduced light intensity and increased scattered light ratio. Direct inversion without differentiation will lead to systematic biases in the estimation of tea tree physiological parameters, such as underestimating the chlorophyll content in tea tree physiological parameters.

[0071] Optionally, in S22, the compensation terminal can input the light reflectance, the directly illuminated area, and the shaded area into the physical-guided inversion network to perform optimization inversion with the goal of minimizing the reflectance residual, thereby obtaining a preliminary spatial distribution map of physiological parameters. The physical-guided inversion network is a hybrid model integrating deep learning and radiative transfer theory. It can use a radiative transfer model (such as the PROSAIL model) as a physical prior constraint. This model describes the absorption, scattering, and transmission processes of incident light within the leaf and between the canopy, and can positively simulate the corresponding canopy reflectance spectrum based on given physiological state parameters (such as leaf nitrogen content, chlorophyll content, and leaf area index). The structure of the physical-guided inversion network can consist of an encoder-decoder. The encoder maps the input reflectance spectrum and light classification mask to the latent space of physiological parameters, and the decoder re-inputs the predicted physiological parameters into the embedded radiative transfer model to generate a reconstructed reflectance spectrum. The optimization objective is to minimize the residual between the reconstructed reflectance and the measured reflectance, i.e. ;

[0072] in, For the measured reflectance, For the radiative transfer forward model, P represents the physiological parameter to be determined. Through physical constraints, the physics-guided inversion network can not only learn the statistical regularities in the data but also ensure that the output parameters conform to the principles of optical physics. Even in shadow areas, it can obtain reasonable inversion results through the compensation mechanism of the physical model, thereby improving the generalization ability and robustness of the inversion. The compensation terminal can output the estimated physiological parameter value corresponding to each pixel through the physics-guided inversion network and fuse them to generate a preliminary spatial distribution map of physiological parameters.

[0073] Optionally, in S23, the compensation terminal can smooth the preliminary spatial distribution map of physiological parameters and output the physiological state parameters of the tea tree. The smoothing process can employ a low-pass filter in the spatial domain, such as a Gaussian filter or a median filter, to eliminate salt-and-pepper noise caused by sensor noise, individual abnormal pixels, or minor fluctuations during the inversion process, while preserving the main spatial variation trends. After smoothing, the physiological state parameters of the tea tree output by the compensation terminal can include leaf nitrogen content (mg / g), chlorophyll content (μg / cm²), and leaf area index. These physiological state parameters directly reflect the nutritional status, photosynthetic capacity, and canopy structure of the tea tree, providing real-time aboveground demand signals for subsequent calculations of theoretical fertility time-series values. Compared with traditional methods relying on empirical regression or purely data-driven approaches, this scheme solves the problems of shadow interference and sample extrapolation through a physically guided inversion network, achieving the conversion from multispectral images to physiological parameters.

[0074] In one embodiment, based on the temporal projection values ​​of effective nutrients in the soil layer and the physiological state parameters of the tea tree, the theoretical fertility temporal values ​​are calculated through a preset tea tree physiological-soil demand mapping relationship, including:

[0075] S31. Obtain stratified root system sample data of tea garden, construct a three-dimensional topological structure of tea tree root system based on the stratified root system sample data, and map the soil root vitality index based on the three-dimensional topological structure of tea tree root system.

[0076] S32, obtain the current phenological period data of the tea garden, and based on the current phenological period data, match the preset prior embedding vector of the corresponding phenological period through the preset phenological period template;

[0077] S33, based on historical soil stratification detection data, uses prior knowledge of nutrient interaction to invert and obtain the nutrient antagonism coefficient matrix and the upper limit threshold of soil nutrient supply capacity;

[0078] S34. Based on the physiological state parameters of tea trees, the three-dimensional topology of tea tree roots and the root vitality index of soil layers, the weighted physiological demand vector of roots in each soil layer is obtained through the weighted allocation of physiological state-root function-root vitality coupling.

[0079] S35, the root weighted physiological demand vector, the preset prior embedding vector and the temporal inference value of effective soil nutrients are input into the temporal coding model. The temporal dependency operation is performed through the multi-head self-attention mechanism to generate the dynamic mapping coefficient matrix. The dynamic mapping coefficient matrix and the root weighted physiological demand vector are weighted and summed to obtain the initial soil nutrient demand vector.

[0080] S36 takes the initial soil nutrient demand vector as the objective, the nutrient antagonism coefficient matrix as the inhibition rule, and the upper limit threshold of soil nutrient supply capacity as the physical boundary. By constructing and solving a system of differential equations including antagonism inhibition terms and supply upper limit constraints, it outputs the theoretical fertility time series value.

[0081] Specifically, in S31, the stratified root system sample data of the tea garden can be obtained by collecting tea tree root samples at different soil depths and measuring parameters such as biomass, length, and branching morphology. Based on the stratified root system sample data, the compensation terminal can use a 3D reconstruction algorithm to construct a 3D topological structure of the tea tree root system. This structure can digitally represent the distribution, orientation, and branching levels of the roots in the soil space. Based on the 3D topological structure, the compensation terminal can calculate indicators such as root length density, surface area density, and fine root ratio in each soil layer to map the soil layer root vitality index, for example: soil layer root vitality index. The biomass density of fine roots (<2mm in diameter) within this soil layer can be used as a metric. Total fine root biomass density The ratio is calculated to obtain, that is The soil root vitality index quantifies the active absorption capacity of roots in each soil layer for water and nutrients, and serves as the basis for nutrient distribution.

[0082] Optionally, in S32, the current phenological stage data of the tea garden may include budding stage data, leaf unfolding stage data, or harvesting stage data. Based on the current phenological stage data, the compensation terminal can query a preset phenological stage template to obtain the preset prior embedding vector for the corresponding phenological stage. The preset prior embedding vector is a high-dimensional numerical vector that encodes prior knowledge such as the inherent nutrient demand preferences, absorption kinetics characteristics, and growth and metabolic patterns of the tea tree under the corresponding phenological stage, providing a physiological benchmark for nutrient demand calculation.

[0083] Optionally, in S33, the compensation terminal can perform inversion calculations based on historical soil stratification data and nutrient interaction prior knowledge. This prior knowledge can include the synergistic and antagonistic mechanisms among major nutrient elements such as nitrogen, phosphorus, and potassium in the soil. The compensation terminal can obtain two key parameters through inversion calculations: a nutrient antagonism coefficient matrix, where each element represents the strength of the mutual inhibition between different nutrient elements (e.g., the inhibition coefficient of high potassium on magnesium absorption); and a soil nutrient supply capacity upper limit threshold, which defines the maximum effective nutrient concentration that each soil layer can stably supply without causing toxicity or loss under the current soil physicochemical properties. These two key parameters together constitute the physicochemical constraints for nutrient demand calculations.

[0084] Optionally, in S34, the compensation terminal can calculate the weighted physiological demand vector of the root system for each soil layer based on the physiological state parameters of the tea tree, the three-dimensional topology of the tea tree root system, and the root vitality index of the soil layer, using a weighted allocation algorithm that couples physiological state, root function, and root vitality.

[0085] For example, suppose a tea garden has L soil layers, numbered i=1,2,…,L, and define a vector of physiological state parameters of tea trees. Where M represents the number of physiological indicators, for example... Leaf nitrogen content (mg / g). Chlorophyll content (μg / cm²) Leaf area index.

[0086] Define the root system function allocation coefficient vector ;in, This represents the proportion of the root system in the i-th soil layer to the total root system length, calculated from the three-dimensional topological structure of the tea tree root system, and satisfies... ;

[0087] Define the soil root vitality index vector ,in Let be the root activity index of the i-th soil layer, reflecting the nutrient absorption activity of the roots in that soil layer. Based on this, the compensation terminal can construct a weighted allocation matrix that couples physiological state, root function, and root activity. Its elements This represents the contribution weight of the j-th physiological indicator to the nutrient demand of the i-th soil layer. This coupled weighted allocation matrix can be obtained by using historical measured data (such as the nutrient absorption of each soil layer under different physiological conditions) through multiple linear regression or neural network training, or it can be set based on prior knowledge of plant nutrition.

[0088] Then, the root weighted physiological demand of the i-th soil layer The formula for calculation is:

[0089] ;

[0090] In the formula, The basic physiological demand component generated in the i-th soil layer after the physiological state parameters are weighted and mapped, with the unit being the combined unit of physiological indicators. ); multiply the basic physiological needs component by This indicates the initial allocation of basic needs based on the spatial distribution ratio of the root system, reflecting the spatial differences in root system function, and then multiplied by... This indicates a second adjustment based on the actual root activity, assigning more absorption tasks to soil layers with high activity and less to those with low activity. This yields a weighted root physiological demand vector for each soil layer. This vector is used to subsequently calculate the initial soil nutrient demand vector together with the preset prior embedding vector and the time-series extrapolated values ​​of effective soil nutrients.

[0091] Specifically, the compensation terminal can decompose the overall physiological needs of tea trees according to the functional zones of the root system, and then perform secondary weighting by combining the root vitality index of each soil layer to form a multi-dimensional vector. Each dimension corresponds to the nutrient requirements of a soil layer, thus realizing the mapping from the above-ground needs to the needs of each underground layer.

[0092] Optionally, in S35, the compensation terminal can input the root-weighted physiological demand vector, the preset prior embedding vector, and the temporal inference value of effective soil nutrients into a pre-trained time-series coding model. This time-series coding model consists of three parts: a sequence of root-weighted physiological demand vectors. (Dimension L, time step t), pre-set prior embedding vector sequence , dimension ) and the time-series extrapolated values ​​of effective nutrients in the soil layer The dimension is L. The compensation terminal can concatenate these three input sequences along the feature dimension to form a fused feature vector. ,right Perform a linear projection to map the model to its hidden dimensions. The model incorporates positional encoding, which can employ sine / cosine functions or learnable positional codes to preserve temporal sequence information. The main body of the model consists of N identical encoder layers stacked together, each containing two sub-layers: a multi-head self-attention sub-layer and a feedforward neural network sub-layer. Each sub-layer is followed by a residual connection and layer normalization. The multi-head self-attention sub-layer projects the input into query, key, and value matrices, and divides it into h heads. Each head independently computes the scaled dot product attention:

[0093] ;

[0094] in, The outputs of all heads are concatenated and then linearly transformed to obtain the output of this sublayer. The feedforward neural network sublayer consists of two fully connected layers, using the ReLU activation function in between; the hidden layer dimension can be set to 4. At the model output, the output of the last encoder layer is taken, passed through a linear transformation and a Softmax or Sigmoid activation function, to generate a dynamic mapping coefficient matrix. Each of its elements This represents the dynamic mapping weight of the root demand of the i-th soil layer to the j-th soil layer at time t. Mt is then compared with the root weighted physiological demand vector. By performing a weighted summation, the initial soil nutrient demand vector is obtained. .

[0095] The model can be pre-trained using supervised learning. The training dataset consists of historical field observation data, containing multiple sets. As input, the corresponding true optimal nutrient requirements (which can be labeled by experts or calculated back through field trials) are used as labels. The loss function uses mean squared error (MSE) to measure the difference between the predicted initial soil nutrient requirement vector and the labels. The optimizer is Adam, with an initial learning rate of 10⁻⁴, and a cosine annealing strategy is used to gradually decrease the learning rate. The batch size is set to 64, the training epochs are 200, and an early stopping method (patience=10) is used to prevent overfitting.

[0096] The model hyperparameters are configured as follows: encoder layers N=6, multi-head attention heads h=8, hidden dimensions... =256, feedforward network hidden layer dimension 1024, Dropout ratio 0.1. Input sequence length (time steps) is fixed at 30 days, sliding window sampling. Optionally, to improve generalization ability, random masking enhancement (masking ratio 0.15) is introduced during training to simulate missing data scenarios.

[0097] Optionally, in S36, the compensation terminal can use the initial soil nutrient demand vector as the optimization objective, the nutrient antagonism coefficient matrix as the inhibition rule, and the upper limit threshold of soil nutrient supply capacity as the physical boundary to construct a system of differential equations containing antagonistic inhibition terms and supply upper limit constraints, for example:

[0098] Let the total number of soil layers be L, and the time range be [0, T]. Let... This represents the time-series vector of theoretical fertility values, i.e., the nutrient concentration (unit: mg / kg) that should be replenished at time t for each soil layer to be solved. Let... Let represent the initial soil nutrient demand vector, which serves as the optimization objective. This is a matrix of nutrient antagonism coefficients, whose elements... Let represent the antagonistic inhibitory strength of nutrient j on nutrient i. This is a threshold vector representing the upper limit of soil nutrient supply capacity, indicating the maximum effective nutrient concentration allowed in each soil layer without toxicity or excessive loss.

[0099] The constructed system of differential equations can consist of two parts: a master equation describing the dynamic balance of nutrients, and auxiliary equations describing antagonistic inhibition and supply ceiling constraints. The master equation is: for each soil layer i, its theoretical fertility time-series value... The rate of change with time is described by the following first-order ordinary differential equation:

[0100] ;

[0101] in: >0 represents the response rate constant of the i-th soil layer (unit: 1 / day), which controls the tracking speed of nutrient replenishment to demand deviation in this soil layer. It can be calibrated using historical experimental data or set to the default value of 0.1. ≥0 represents the antagonistic inhibition coefficient (unit: 1 / day) of the i-th soil layer, adjusting the inhibitory strength of the antagonistic effect on nutrient accumulation; its value can range from 0.01 to 0.05. (First term) Indicating driving force: when the actual replenishment amount Less than demand When this term is positive, it prompts... An increase; conversely, a decrease leads to a negative result, causing a reduction. (Second term) The term "antagonistic inhibition" indicates that the replenishment of nutrients from other soil layers (or other nutrients) will inhibit the effective accumulation of nutrients in the native soil layer through the antagonistic coefficient, reflecting the competitive relationship between nutrients.

[0102] Supply ceiling constraint: To ensure that the solution does not exceed the physically feasible range, a hard constraint condition is introduced:

[0103] ;

[0104] This constraint is implemented numerically using a projection method: after each time step iteration, the... Cut to the interval [0, cmax, i]. Initial conditions: Assume that no fertilizer has been applied to any soil layer at the initial time, i.e. The above system of ordinary differential equations can be numerically integrated using the fourth-order Runge-Kutta method (RK4). A time step Δt is set to 0.1 days to ensure numerical stability. At each time step, the right-hand side is calculated first, and then updated. Perform projection clipping. Repeat the iteration until t=T (usually a complete tea tree growth cycle or a fertilization schedule cycle), and output the results. This refers to the theoretical fertility time-series value, which is the optimal nutrient replenishment curve that minimizes the deviation from the initial demand target under the dual constraints of antagonistic inhibition and supply ceiling. This set of equations describes the dynamic balance process of nutrients in the soil. By solving this set of equations, an optimal solution that satisfies physicochemical constraints in both time and space is obtained, namely the theoretical fertility time-series value. This value is a set of curves that change over time, indicating the ideal total amount of nutrients required to compensate for the gap between natural supply and the actual needs of tea trees at each time point and in each soil layer, thus providing a basis for fertilization decisions.

[0105] In one embodiment, such as Figure 2 As shown, based on nutrient demand periods, fertilizer attribute data, and theoretical fertility time-series values, a nutrient release plan is generated through nutrient supply and demand matching, including:

[0106] S41. Based on historical tea tree harvesting data, extract the time nodes of multiple harvesting batches, calculate the time interval of each time node, and set the nutrient demand period according to the time interval.

[0107] S42, based on fertilizer attribute data, extract the nutrient release rate curve and nutrient release pattern;

[0108] S43, align the nutrient demand period and nutrient release rate curve in time sequence to generate a time-rate matching mapping. Based on the time-rate matching mapping, divide the nutrient demand period into multiple fertilization cycles. Based on the nutrient release pattern, set the nutrient release ratio of the fertilization cycle to generate a nutrient release ratio sequence.

[0109] S44: Based on the nutrient release ratio sequence, theoretical fertility time series value and fertilization cycle, the fertilizer requirement is calculated, and a nutrient release plan is generated based on the fertilizer requirement.

[0110] For example, in S41, the compensation terminal can extract the time nodes of multiple harvesting batches based on historical tea tree harvesting data, calculate the time interval of each time node, and set the nutrient requirement period according to the time interval. Specifically, the historical tea tree harvesting data includes the specific date and corresponding yield of each harvest. By analyzing this data, multiple peak harvesting periods of tea trees throughout the year can be identified, and these time nodes can be denoted as... The time interval between two adjacent harvesting batches This reflects the growth cycle of tea tree shoots. Because tea trees undergo a rapid vegetative growth phase between harvests, their demand for nutrients is particularly high; therefore, these time intervals can be used to... This period is defined as the critical nutrient demand period. It is the core time window for subsequent fertilization decisions, ensuring that fertilization operations are highly synchronized with the tea plant's own biological rhythms for nutrient absorption.

[0111] Optionally, in S42, the compensation terminal can extract the nutrient release rate curve and nutrient release pattern based on fertilizer attribute data. The fertilizer attribute data contains detailed characteristics of the applied fertilizer. The nutrient release rate curve R(t) is a proportional function describing the cumulative release of available nutrients over time after the fertilizer is applied to the soil, which can be determined experimentally. The nutrient release pattern is a qualitative or quantitative description of the curve's characteristics; for example, it can be categorized as fast-acting (rapid release, steep curve), slow-release (stable and sustained release, flat curve), or controlled-release (release rate controlled by specific environmental factors, such as temperature and moisture). These two data points are the core basis for determining when and at what rate the fertilizer provides nutrients to the soil, and are a prerequisite for achieving supply-demand matching.

[0112] Optionally, in S43, the compensation terminal can time-series align the nutrient demand period and the nutrient release rate curve to generate a time-rate matching map. Based on the time-rate matching map, the nutrient demand period is divided into multiple fertilization cycles, and based on the nutrient release pattern, the nutrient release ratio of the fertilization cycle is set to generate a nutrient release ratio sequence. Time-series alignment involves aligning multiple nutrient demand periods... The nutrient release rate curve R(t) was compared and matched on the time axis, with the goal of aligning the main fertilizer release period with the peak nutrient demand of the tea plant. Based on this matching mapping, the entire nutrient demand period was mapped. Divided into K fertilization cycles Where k=1,...,K. The duration and start and end times of each fertilization cycle are determined by the matching mapping.

[0113] Optionally, the compensation terminal can set a nutrient release ratio Pk for each fertilization cycle based on the fertilizer's nutrient release pattern. This ratio represents the percentage of nutrients planned to be released in that cycle relative to the total demand for the entire fertilization cycle. All Pk values ​​constitute a nutrient release ratio sequence {P1, P2, ..., PK}, which ensures dynamic coordination between the fertilizer release rhythm and the tea tree's periodic absorption peaks.

[0114] Optionally, in S44, the compensation terminal can calculate the fertilizer requirement based on the nutrient release ratio sequence, theoretical fertility time series value, and fertilization cycle. Based on the fertilizer requirement, the fertilization time t of each cycle, and the corresponding fertilizer type, a nutrient release plan is generated. This plan specifies the time, amount, and fertilizer type for each fertilization, thereby achieving precise compensation for the fertility of the tea garden.

[0115] In one embodiment, the formula for calculating the fertilizer requirement is:

[0116] ;

[0117] Where F represents the fertilizer requirement, and K represents the total number of fertilization cycles. Let be the start time of the kth fertilization cycle. Let k be the end time of the fertilization cycle. It is a function of the theoretical fertility time-series value changing with time. This is a function of the temporal variation of the effective nutrient content of the soil layer over time. Let be the nutrient release ratio for the k-th fertilization cycle, which is taken from the nutrient release ratio sequence. The effective utilization rate of fertilizer nutrients in the kth fertilization cycle is determined by fertilizer attribute data.

[0118] Specifically, in the formula, F represents the fertilizer requirement, with the dimension being mass units; K represents the total number of fertilization cycles. and These are the start and end times of the kth fertilization cycle, respectively; Let be a function of the theoretical fertility time series value as a function of time, representing the curve of the ideal nutrient concentration required to meet the optimal growth state of tea trees as a function of time, and its integral . This represents the total amount of nutrients theoretically required within this cycle. Let be a function of the temporal projection of available nutrients in the soil layer, representing the change of the actual available nutrient concentration over time under the combined effects of natural growth and decline and harvesting / relocation. Its integral... This represents the total amount of nutrients supplied by the soil itself during this cycle; the difference between the two is the net nutrient gap that needs to be replenished through fertilization during this cycle. The nutrient release ratio for the kth fertilization cycle is taken from the nutrient release ratio sequence. Its function is to allocate the net deficit of the cycle to the actual fertilization amount according to the planned ratio based on the fertilizer release pattern, so as to ensure that the fertilizer release rhythm is coordinated with the tea tree's periodic absorption peak. Let be the fertilizer nutrient utilization rate for the k-th fertilization cycle. When the same fertilizer is used in multiple fertilization cycles... It is a constant value; when different fertilizers are used in different cycles, Based on the attribute data of the fertilizer applied during this cycle, the proportion of fertilizer actually absorbed and utilized by the tea trees after being applied to the soil can be determined, thus correcting for efficiency losses caused by volatilization, fixation, leaching, etc. The compensation terminal can use this formula to transform the abstract nutrient demand curve into specific, executable fertilizer dosage instructions.

[0119] In one embodiment, based on the deviation in execution performance, the model parameters of the first time-varying soil fertility model are optimized to obtain a second time-varying soil fertility model, including:

[0120] S51, the deviation of the execution effect is broken down according to the preset soil layer depth and time scale to obtain a multi-layer nutrient deviation time sequence;

[0121] S52, the partial derivatives of each parameter in the multi-scale nutrient mineralization rate parameter set and the multi-scale nutrient leaching and migration parameter set in the first soil fertility time-varying model are calculated using the multi-layer nutrient deviation time series, respectively, to obtain the parameter partial derivatives, and a two-dimensional deviation sensitivity matrix is ​​constructed based on the parameter partial derivatives.

[0122] S53, take the absolute value of each element in the two-dimensional deviation sensitivity matrix, and normalize the absolute value of the sensitivity to obtain the proportion of deviation contribution;

[0123] S54. Based on the multi-scale nutrient mineralization rate parameter set and the multi-scale nutrient leaching and migration parameter set, parameters whose deviation contribution ratio is greater than the preset threshold are marked as parameters to be corrected, forming a set of parameters to be corrected.

[0124] S55, with the goal of minimizing the sum of squared residuals of the nutrient deviation time series, uses gradient descent to find the parameter correction values ​​from the set of parameters to be corrected.

[0125] S56, replace the corresponding values ​​of the parameters to be corrected in the first soil fertility time-varying model with the parameter correction values ​​to obtain the second soil fertility time-varying model.

[0126] For example, in S51, the compensation terminal can decompose the execution effect deviation according to preset soil depth and time scale to obtain a multi-layer nutrient deviation time series. Specifically, the execution effect deviation is a multi-dimensional overall error value. In order to trace the source of the error, it needs to be decomposed according to spatial and temporal dimensions. In the spatial dimension, the compensation terminal can extract the corresponding deviation values ​​according to preset soil depth classification standards, such as the topsoil layer of 0-20cm, the subsoil layer of 20-40cm, etc. In the temporal dimension, the compensation terminal can allocate the deviation value to various time nodes according to the time step predicted by the model or a specific time window, such as by week or by month. After this decomposition, the compensation terminal can transform the originally single overall deviation into a set of ordered deviation sequences with soil layer labels and timestamps, that is, a multi-layer nutrient deviation time series.

[0127] Optionally, in S52, the compensation terminal can calculate the partial derivatives of each parameter in the multi-scale nutrient mineralization rate parameter set and the multi-scale nutrient leaching and migration parameter set of the first soil fertility time-varying model using the multi-layer nutrient deviation time series, respectively, to obtain the parameter partial derivatives. Based on the parameter partial derivatives, a two-dimensional deviation sensitivity matrix is ​​constructed. The parameter partial derivative refers to the derivative of the model output (i.e., the predicted nutrient content) with respect to a specific model parameter, mathematically expressed as... Here, y represents the model output, and θ represents the parameter to be analyzed, quantifying the magnitude of the change in model output caused by a small change in this parameter. In this step, the compensation terminal can calculate the degree of influence of each parameter in the multi-scale nutrient mineralization rate parameter set and each parameter in the multi-scale nutrient leaching and migration parameter set on the bias at each soil layer and at each time scale. These partial derivatives are arranged according to two dimensions: soil layer depth and time scale. That is, the row direction represents different soil layers, and the column direction represents different time scales or parameter categories, thus constructing a two-dimensional bias sensitivity matrix. Each element value of this two-dimensional bias sensitivity matrix reflects the sensitivity of a specific parameter at a specific spatiotemporal location, providing a quantitative basis for subsequent screening.

[0128] Optionally, in S53, the compensation terminal can take the absolute value of each element in the two-dimensional bias sensitivity matrix and normalize the absolute value of the sensitivity to obtain the bias contribution ratio. Since the partial derivatives may be positive or negative, representing positive or negative correlation respectively, in order to uniformly measure the influence of the parameters, the compensation terminal can take the absolute value of all elements to eliminate the influence of the sign, and normalize the matrix after taking the absolute value. Max-min value normalization or sum-normalization methods can be used. The value after normalization is the bias contribution ratio, which represents the relative contribution of each parameter to the overall performance bias among all parameters. Parameters with a larger ratio mean that their impact on the model prediction error is more significant and should be given priority for attention and adjustment.

[0129] Optionally, in S54, the compensation terminal can, based on a multi-scale nutrient mineralization rate parameter set and a multi-scale nutrient leaching and migration parameter set, mark parameters whose deviation contribution ratio exceeds a preset threshold as parameters to be corrected, forming a set of parameters to be corrected. This preset threshold is an empirical or cross-validated critical value, typically set to 0.1. Only parameters whose deviation contribution ratio exceeds this threshold are considered major factors causing model deviation. Parameters that do not reach the threshold are considered to have a small contribution to the current deviation and their original values ​​are temporarily maintained. This screening mechanism allows for focused optimization of resources, avoiding blind adjustments to all parameters and improving the efficiency of model correction. The selected parameters constitute the set of parameters to be corrected, which includes the key parameters that most urgently need calibration.

[0130] Optionally, in S55, the compensation terminal can use gradient descent to optimize the set of parameters to be corrected, with the goal of minimizing the sum of squared residuals of the nutrient deviation time series, to obtain the corrected parameter values. The sum of squared residuals refers to the sum of the squares of the differences between the model's predicted values ​​and the actual observed values, mathematically expressed as: The smaller the value, the higher the model fitting accuracy. Gradient descent optimization is an iterative optimization algorithm. Its basic principle is to gradually adjust the parameter values ​​along the negative gradient direction of the objective function (i.e., the sum of squared residuals) relative to the parameters to be corrected. The step size of each adjustment is controlled by the learning rate until the objective function converges to the minimum value or reaches the preset number of iterations. In this embodiment, let the set of parameters to be corrected be... Initialize all parameters to their values ​​in the current model. In each iteration, calculate the loss function L with respect to each parameter using all samples (i.e., all time points and soil layers). partial derivatives This partial derivative can be approximated using the chain rule or numerical difference. Update the parameters, where α is the learning rate, which can be a fixed value such as 0.01 or 0.001. Repeat the above iterative process until the change in the loss function L is less than the preset tolerance. The iteration count may reach a maximum of 500. The parameter values ​​obtained at the end of the iteration are the parameter correction values. In each iteration, the sum of squared residuals and its gradient under the current parameters are calculated, and then the parameters are updated according to the gradient descent formula to obtain a set of parameter values ​​that minimize the sum of squared residuals, i.e., the parameter correction values.

[0131] Optionally, in S56, the compensation terminal can replace the corresponding values ​​of the parameters to be corrected in the first soil fertility time-varying model with the parameter correction values ​​to obtain the second soil fertility time-varying model. Specifically, the compensation terminal can assign the parameter correction values ​​obtained in step S55 to the corresponding parameters to be corrected in the first soil fertility time-varying model one by one, while leaving other parameters not selected into the set of parameters to be corrected unchanged. After this replacement operation, the original model is updated to the second soil fertility time-varying model, which will have higher accuracy and adaptability in subsequent fertility prediction, forming a closed-loop mechanism of perception-decision-execution-optimization, enabling the model to continuously adaptively correct and continuously improve the accuracy and reliability of fertility compensation.

[0132] The aforementioned method for dynamic balance compensation of tea garden fertility comprehensively captures key factors affecting soil fertility, including meteorological, soil, tea tree growth, and fertilizer properties, by acquiring historical meteorological data, historical soil stratification data, historical tea tree harvesting data, multispectral imagery of the tea tree canopy, and fertilizer attribute data. This lays a data foundation for subsequent modeling. Based on historical meteorological, soil stratification, and harvesting data, a first time-varying soil fertility model is constructed, and the temporal inference value of effective nutrients in the soil layer is calculated. This model dynamically simulates the time-varying pattern of soil nutrients under the dual effects of natural growth and decline and harvest removal, overcoming the shortcomings of traditional static detection methods that cannot reflect continuous nutrient changes. Furthermore, by inverting tea tree canopy multispectral images, physiological state parameters of tea trees can be obtained, enabling real-time monitoring of tea tree fertility. The system uses accurate nutrient demand signals to avoid supply-demand mismatches caused by relying solely on experience. By combining the temporal projection values ​​of effective nutrients in the soil layer with the physiological parameters of tea trees to calculate theoretical fertility temporal values, it can accurately map soil supply capacity with the actual needs of tea trees, making fertilization targets more targeted. By extracting nutrient demand periods from historical harvesting data and matching them with fertilizer attribute data to generate nutrient release plans, it can synchronize and coordinate fertilizer release rhythm with the tea tree's periodic absorption peaks, solving the problems of time mismatch and improper proportions in traditional fertilization. Based on feedback from execution effect deviations, the system optimizes the parameters of the time-varying soil fertility model, forming a closed-loop mechanism of perception-decision-execution-optimization, enabling the model to continuously self-adapt and correct, and constantly improving the accuracy and reliability of subsequent fertility compensation.

[0133] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0134] Based on the same inventive concept, this application also provides a tea garden fertility dynamic balance compensation system for implementing the aforementioned tea garden fertility dynamic balance compensation method. The solution provided by this system is similar to the solution described in the above method. Therefore, the specific limitations of one or more embodiments of the tea garden fertility dynamic balance compensation system provided below can be found in the above-described limitations of the tea garden fertility dynamic balance compensation method, and will not be repeated here.

[0135] In one exemplary embodiment, such as Figure 3 As shown, a dynamic balance compensation system for tea garden fertility is provided, comprising:

[0136] The tea garden data acquisition module 101 is used to acquire historical meteorological data, historical soil stratification detection data, historical tea tree harvesting data, tea tree canopy multispectral image data, and fertilizer attribute data of the tea garden.

[0137] The extrapolation model construction module 102 is used to construct a first soil fertility time-varying model based on historical meteorological data, historical soil stratification detection data and historical tea tree harvesting data, and to calculate the temporal extrapolation value of effective nutrients in the soil layer based on the first soil fertility time-varying model.

[0138] The physiological parameter inversion module 103 is used to obtain the physiological state parameters of tea trees by inverting the physiological state of tea trees based on the multispectral image data of the tea tree canopy.

[0139] The theoretical fertility calculation module 104 is used to calculate the theoretical fertility time series value based on the temporal extrapolation value of effective nutrients in the soil layer and the physiological state parameters of tea trees, through the preset physiological-soil demand mapping relationship of tea trees;

[0140] The release plan generation module 105 is used to extract the nutrient demand period based on historical tea tree harvesting data, and generate a nutrient release plan by matching nutrient supply and demand based on the nutrient demand period, fertilizer attribute data and theoretical fertility time series value, and generate a fertility compensation execution instruction based on the nutrient release plan.

[0141] The model parameter optimization module 106 is used to obtain the execution effect deviation of the fertility compensation execution command, and optimize the model parameters of the first soil fertility time-varying model based on the execution effect deviation to obtain the second soil fertility time-varying model. The second soil fertility time-varying model is used to generate the time series extrapolation value of effective nutrients in the soil layer.

[0142] In one embodiment, the inference model building module 102 can also be used for:

[0143] Historical meteorological data are decomposed into multi-scale components to obtain multi-scale excitation component sequences, which include temperature-related components and precipitation-related components.

[0144] Based on historical soil stratification detection data and temperature correlation components, stratification fitting was performed using the first-order mineralization kinetic equation to obtain a set of multi-scale nutrient mineralization rate parameters for each soil layer at different time scales.

[0145] Based on historical soil stratification data and precipitation-related components, the stratification fitting was performed using the convection diffusion equation to obtain a set of multi-scale nutrient leaching and migration parameters for each soil layer at different time scales.

[0146] A hierarchical multi-scale soil nutrient natural growth and decline sub-model was constructed by performing parameter coupling mapping on the multi-scale nutrient mineralization rate parameter set and the multi-scale nutrient leaching and migration parameter set.

[0147] Based on historical tea tree harvesting data, time-series data on the amount of nutrients removed during harvest were extracted.

[0148] The time-series data of nutrient removal from harvest are input into a hierarchical multi-scale soil nutrient natural fluctuation sub-model, and coupled to obtain the first time-varying model of soil fertility.

[0149] In one embodiment, the physiological parameter inversion module 103 can also be used for:

[0150] Based on tea tree canopy multispectral image data, the light reflectance of multiple bands was extracted, and the light-shade classification of tea tree canopy multispectral image data was performed based on the light reflectance to obtain the directly lit area and the shaded area.

[0151] The light reflectance, direct light area, and shadow area are input into the physical-guided inversion network. The optimization inversion is performed with the goal of minimizing the reflectance residual, and a preliminary spatial distribution map of physiological parameters is obtained. The physical-guided inversion network is constructed by embedding the positive physical mapping relationship between reflectance and tea tree physiological state parameters, with radiative transfer theory as the physical prior constraint.

[0152] The preliminary spatial distribution map of physiological parameters is smoothed to output the physiological state parameters of tea trees, which include leaf nitrogen content, chlorophyll content and leaf area index.

[0153] In one embodiment, the theoretical fertility calculation module 104 can also be used for:

[0154] Stratified root system sample data of tea garden was obtained. Based on the stratified root system sample data, a three-dimensional topological structure of tea tree root system was constructed. Based on the three-dimensional topological structure of tea tree root system, the root vitality index of soil layer was mapped.

[0155] Obtain the current phenological period data of the tea garden, and based on the current phenological period data, match the corresponding phenological period's preset prior embedding vector through a preset phenological period template;

[0156] Based on historical soil stratification data, the nutrient antagonism coefficient matrix and the upper limit threshold of soil nutrient supply capacity are obtained by inversion through prior knowledge of nutrient interaction.

[0157] Based on the physiological state parameters of tea trees, the three-dimensional topology of tea tree roots, and the root vitality index of soil layers, the weighted physiological demand vector of roots in each soil layer is obtained through the weighted allocation of physiological state-root function-root vitality coupling.

[0158] The root system weighted physiological demand vector, the preset prior embedding vector, and the temporal inference value of effective soil nutrients are input into the temporal coding model. The temporal dependency operation is performed through the multi-head self-attention mechanism to generate a dynamic mapping coefficient matrix. The dynamic mapping coefficient matrix and the root system weighted physiological demand vector are weighted and summed to obtain the initial soil nutrient demand vector.

[0159] Using the initial soil nutrient demand vector as the objective, the nutrient antagonism coefficient matrix as the inhibition rule, and the upper limit threshold of soil nutrient supply capacity as the physical boundary, a system of differential equations including antagonism inhibition terms and supply upper limit constraints is constructed and solved to output the theoretical fertility time series value.

[0160] In one embodiment, the release scheme generation module 105 can also be used for:

[0161] Based on historical tea tree harvesting data, the time nodes of multiple harvesting batches were extracted, the time intervals of each time node were calculated, and the nutrient demand period was set according to the time intervals.

[0162] Based on fertilizer attribute data, nutrient release rate curves and nutrient release patterns were extracted.

[0163] The nutrient demand period and nutrient release rate curve are aligned in time sequence to generate a time-rate matching mapping. Based on the time-rate matching mapping, the nutrient demand period is divided into multiple fertilization cycles. Based on the nutrient release pattern, the nutrient release ratio of the fertilization cycle is set to generate a nutrient release ratio sequence.

[0164] Based on the nutrient release ratio sequence, theoretical fertility time series value, and fertilization cycle, the fertilizer requirement is calculated, and a nutrient release plan is generated based on the fertilizer requirement.

[0165] In one embodiment, the model parameter optimization module 106 can also be used for:

[0166] The deviations in the execution results are broken down according to the preset soil layer depth and time scale to obtain a multi-layer nutrient deviation time series;

[0167] The partial derivatives of each parameter in the multi-scale nutrient mineralization rate parameter set and the multi-scale nutrient leaching and migration parameter set in the first soil fertility time-varying model were calculated using the multi-layer nutrient deviation time series, respectively. The parameter partial derivatives were obtained, and a two-dimensional deviation sensitivity matrix was constructed based on the parameter partial derivatives.

[0168] Take the absolute value of each element in the two-dimensional bias sensitivity matrix and normalize the absolute value of the sensitivity to obtain the bias contribution ratio.

[0169] Based on the multi-scale nutrient mineralization rate parameter set and the multi-scale nutrient leaching and migration parameter set, parameters whose deviation contribution ratio is greater than a preset threshold are marked as parameters to be corrected, forming a set of parameters to be corrected.

[0170] With the goal of minimizing the sum of squared residuals of the nutrient deviation time series, gradient descent is used to find the parameter correction values ​​from the set of parameters to be corrected.

[0171] The parameter correction value is replaced with the corresponding value of the parameter to be corrected in the first soil fertility time-varying model to obtain the second soil fertility time-varying model.

[0172] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of the tea garden fertility dynamic balance compensation method as described above.

[0173] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0174] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0175] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A method for dynamic balance compensation of fertility in tea gardens, characterized in that, The method includes: Acquire historical meteorological data, historical soil stratification data, historical tea tree harvesting data, tea tree canopy multispectral image data, and fertilizer attribute data for the tea garden; Based on the historical meteorological data, the historical soil stratification detection data, and the historical tea tree harvesting data, a first soil fertility time-varying model is constructed, and based on the first soil fertility time-varying model, the temporal inference value of effective nutrients in the soil layer is calculated. Based on the multispectral image data of the tea tree canopy, the physiological state parameters of the tea tree are obtained through the physiological state inversion processing of the tea tree. Based on the estimated effective nutrient time series values ​​of the soil layer and the physiological state parameters of the tea tree, the theoretical fertility time series values ​​are calculated through the preset tea tree physiological-soil layer demand mapping relationship. Based on the historical tea tree harvesting data, the nutrient demand period is extracted, and based on the nutrient demand period, the fertilizer attribute data, and the theoretical fertility time series value, a nutrient release plan is generated through nutrient supply and demand matching, and a fertility compensation execution instruction is generated based on the nutrient release plan. The execution effect deviation of the fertility compensation execution instruction is obtained, and the model parameters of the first soil fertility time-varying model are optimized based on the execution effect deviation to obtain the second soil fertility time-varying model. The second soil fertility time-varying model is used to generate the time-series extrapolation value of the effective nutrients in the soil layer.

2. The method according to claim 1, characterized in that, The first soil fertility time-varying model is constructed based on the historical meteorological data, the historical soil stratification detection data, and the historical tea tree harvesting data, including: The historical meteorological data is decomposed into multi-scale components to obtain a multi-scale excitation component sequence, which includes temperature-related components and precipitation-related components. Based on the historical soil stratification detection data and the temperature correlation component, the stratification fitting is performed through the first-order mineralization kinetic equation to obtain the multi-scale nutrient mineralization rate parameter set of each soil layer at different time scales. Based on the historical soil stratification detection data and the precipitation correlation components, the stratification fitting was performed by convection diffusion equation to obtain the multi-scale nutrient leaching and migration parameter set of each soil layer at different time scales. By performing parameter coupling mapping on the multi-scale nutrient mineralization rate parameter set and the multi-scale nutrient leaching and migration parameter set, a hierarchical multi-scale soil nutrient natural ebb and flow sub-model is constructed. Based on the historical tea tree harvesting data, time-series data on the amount of nutrients removed during harvest were extracted. The harvested nutrient removal time series data are input into the hierarchical multi-scale soil nutrient natural ebb and flow sub-model, and coupled to obtain the first soil fertility time-varying model.

3. The method according to claim 1, characterized in that, The physiological state parameters of the tea trees are obtained by inverting the physiological state of the tea trees based on the multispectral image data of the tea tree canopy, including: Based on the multispectral image data of the tea tree canopy, the light reflectance of multiple bands is extracted, and based on the light reflectance, the multispectral image data of the tea tree canopy is classified into light and shadow to obtain the directly lit area and the shadow area. The light reflectance, the directly illuminated area, and the shaded area are input into the physical-guided inversion network. The optimization inversion is performed with the goal of minimizing the reflectance residual, and a preliminary spatial distribution map of physiological parameters is obtained. The physical-guided inversion network is constructed by embedding a positive physical mapping relationship between reflectance and tea tree physiological state parameters, with radiative transfer theory as the physical prior constraint. The preliminary physiological parameter spatial distribution map is smoothed to output the tea tree physiological state parameters, which include leaf nitrogen content, chlorophyll content, and leaf area index.

4. The method according to claim 3, characterized in that, The theoretical fertility time series value is calculated based on the temporal extrapolation value of the effective nutrients in the soil layer and the physiological state parameters of the tea tree, through a preset tea tree physiological-soil demand mapping relationship, including: Stratified root system sample data of tea garden is obtained. Based on the stratified root system sample data, a three-dimensional topological structure of tea tree root system is constructed. Based on the three-dimensional topological structure of tea tree root system, the soil root vitality index is mapped. Obtain the current phenological period data of the tea garden, and based on the current phenological period data, match the preset prior embedding vector of the corresponding phenological period through a preset phenological period template; Based on the historical soil stratification detection data, the nutrient antagonism coefficient matrix and the upper limit threshold of soil nutrient supply capacity are obtained by inversion through nutrient interaction prior knowledge. Based on the physiological state parameters of the tea tree, the three-dimensional topology of the tea tree root system, and the root vitality index of the soil layer, the weighted physiological demand vector of the root system for each soil layer is obtained through the weighted allocation of physiological state-root function-root vitality coupling. The root weighted physiological demand vector, the preset prior embedding vector, and the temporal inference value of effective soil nutrients are input into the temporal coding model. Temporal dependency operation is performed through a multi-head self-attention mechanism to generate a dynamic mapping coefficient matrix. The dynamic mapping coefficient matrix and the root weighted physiological demand vector are weighted and summed to obtain the initial soil nutrient demand vector. Using the initial soil nutrient demand vector as the target, the nutrient antagonism coefficient matrix as the inhibition rule, and the upper limit threshold of soil nutrient supply capacity as the physical boundary, the theoretical fertility time series value is output by constructing and solving a system of differential equations including antagonism inhibition terms and supply upper limit constraints.

5. The method according to claim 1, characterized in that, The process of generating a nutrient release plan based on the nutrient demand period, fertilizer attribute data, and theoretical fertility time series value through nutrient supply and demand matching includes: Based on the historical tea tree harvesting data, the time nodes of multiple harvesting batches are extracted, the time interval of each time node is calculated, and the nutrient requirement period is set according to the time interval. Based on the fertilizer attribute data, the nutrient release rate curve and nutrient release pattern are extracted. The nutrient demand period and the nutrient release rate curve are aligned in time sequence to generate a time-rate matching mapping. Based on the time-rate matching mapping, the nutrient demand period is divided into multiple fertilization cycles. Based on the nutrient release pattern, the nutrient release ratio of the fertilization cycle is set to generate a nutrient release ratio sequence. Based on the nutrient release ratio sequence, the theoretical fertility time series value, and the fertilization cycle, the fertilizer requirement is calculated, and based on the fertilizer requirement, the nutrient release scheme is generated.

6. The method according to claim 5, characterized in that, The formula for calculating the fertilizer requirement is: ； Where F represents the fertilizer requirement, and K represents the total number of fertilization cycles. This represents the start time of the kth fertilization cycle. Let k be the end time of the fertilization cycle. The theoretical fertility time-series value is a function of the change over time. The effective nutrient content of the soil layer is a function of the temporal evolution of its estimated value over time. The nutrient release ratio is the nutrient release ratio for the kth fertilization cycle, and the nutrient release ratio is taken from the nutrient release ratio sequence. The effective utilization rate of fertilizer nutrients in the kth fertilization cycle is determined by the fertilizer attribute data.

7. The method according to claim 2, characterized in that, The step of optimizing the model parameters of the first soil fertility time-varying model based on the execution effect deviation to obtain the second soil fertility time-varying model includes: The deviation in the execution effect is broken down according to a preset soil layer depth and time scale to obtain a multi-layer nutrient deviation time sequence; The partial derivatives of the multi-scale nutrient mineralization rate parameter set and the multi-scale nutrient leaching and migration parameter set in the first soil fertility time-varying model are calculated respectively to obtain the parameter partial derivatives, and a two-dimensional deviation sensitivity matrix is ​​constructed based on the parameter partial derivatives. Take the absolute value of each element in the two-dimensional deviation sensitivity matrix and normalize the absolute value of the sensitivity to obtain the deviation contribution ratio. Based on the multi-scale nutrient mineralization rate parameter set and the multi-scale nutrient leaching and migration parameter set, the parameters whose deviation contribution ratio is greater than a preset threshold are marked as parameters to be corrected, forming a set of parameters to be corrected. With the goal of minimizing the sum of squared residuals of the nutrient deviation time series, gradient descent is used to optimize the set of parameters to be corrected to obtain the parameter correction values. The second soil fertility time-varying model is obtained by replacing the corresponding value of the parameter to be corrected in the first soil fertility time-varying model with the parameter correction value.

8. A dynamic balance compensation system for tea garden fertility, characterized in that, The system includes: The tea garden data acquisition module is used to acquire historical meteorological data, historical soil stratification detection data, historical tea tree harvesting data, tea tree canopy multispectral image data, and fertilizer attribute data of the tea garden. The extrapolation model construction module is used to construct a first soil fertility time-varying model based on the historical meteorological data, the historical soil stratification detection data and the historical tea tree harvesting data, and to calculate the temporal extrapolation value of effective nutrients in the soil layer based on the first soil fertility time-varying model. The physiological parameter inversion module is used to obtain the physiological state parameters of tea trees by inverting the physiological state of tea trees based on the multispectral image data of the tea tree canopy. The theoretical fertility calculation module is used to calculate the theoretical fertility time series value based on the effective nutrient time series extrapolation value of the soil layer and the physiological state parameters of the tea tree, through a preset tea tree physiological-soil layer demand mapping relationship; The release plan generation module is used to extract the nutrient demand period based on the historical tea tree harvesting data, and generate a nutrient release plan based on the nutrient demand period, the fertilizer attribute data and the theoretical fertility time series value through nutrient supply and demand matching, and generate a fertility compensation execution instruction based on the nutrient release plan. The model parameter optimization module is used to obtain the execution effect deviation of the fertility compensation execution command, and optimize the model parameters of the first soil fertility time-varying model based on the execution effect deviation to obtain the second soil fertility time-varying model. The second soil fertility time-varying model is used to generate the time-series extrapolation value of the effective nutrients in the soil layer.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images