A method for predicting the amount of soil nutrient supply in response to physiological demand

Abstract

The present application belongs to the technical field of data processing and intelligent agriculture, and particularly relates to a soil nutrient supply amount prediction method responding to physiological demand, aiming to solve the problems of nutrient supply and demand imbalance, low fertilizer utilization rate and the like caused by ignoring plant physiological metabolism differences. The method comprises collecting environment, soil and plant physiological multi-source data; identifying citrus phenological stages by using a CNN-LSTM hybrid model; constructing a physiological demand index (PDI) model to quantify real-time physiological demand; establishing a soil nutrient bioavailability dynamic conversion model to simulate supply flux; integrating PDI and supply flux, predicting the optimal soil nutrient supply amount through a multi-objective optimization algorithm, and performing closed-loop adaptive correction based on plant physiological response. By using the above technical scheme, the present application realizes the change of nutrient prediction from yield driving to physiological response driving, and significantly improves the fertilizer utilization rate.

Description

Technical Field

[0001] This invention belongs to the field of data processing and smart agriculture technology, and in particular relates to a method for predicting soil nutrient supply in response to physiological needs. Background Technology

[0002] Citrus, as one of the most widely planted and economically valuable woody fruit trees globally, relies heavily on precise nutrient management systems for its sustainable development. In modern agricultural production, rational and efficient nutrient supply is not only crucial for ensuring citrus yield and fruit quality, but also a key means of reducing fertilizer use while increasing efficiency, mitigating agricultural non-point source pollution, and maintaining orchard ecological balance. With the increasing prevalence of precision agriculture technologies, how to make scientific decisions regarding fertilizer nutrient allocation based on data-driven approaches and physiological characteristic analysis has become a cutting-edge research area at the intersection of agronomy and information science.

[0003] Existing soil nutrient management and prediction technologies are mostly based on the traditional "soil testing and fertilizer recommendation" theory. This theoretical system establishes a relatively static nutrient replenishment logic by periodically sampling and testing the content of available nutrients in the soil, combined with a nutrient abundance / deficiency index system for specific varieties. For a long period of history, this empirical model, which uses soil test values ​​combined with target yields to estimate nutrient requirements, has played a significant supporting role in improving citrus yields and correcting the phenomenon of indiscriminate fertilization. The basic principle of this method is that it assumes there is a linear or quasi-linear balance between the supply capacity of the soil nutrient pool and the total amount absorbed by crops, maintaining soil fertility levels by compensating for nutrients removed by crops.

[0004] However, with the increasing sophistication of agricultural IoT technology in sensing crop growth environments and the deepening understanding of crop physiological and metabolic mechanisms, the aforementioned prediction logic based on empirical weights and static indicators is gradually revealing its inherent limitations when dealing with complex production environments. The root cause lies in the fact that citrus, as a perennial evergreen crop, exhibits distinct phenological phases in its growth cycle, including budding, flowering, physiological fruit drop, fruit enlargement, and ripening. During these different phenological successions, the intensity of physiological metabolism, nutrient transport flux, and root absorption preferences for specific ions within the citrus plant exhibit significant spatiotemporal heterogeneity. Existing prediction methods often overlook this dynamic "physiological feedback" that changes with phenological phases, making it difficult to accurately match nutrient supply with the actual peak physiological nutrient uptake of the crop on a temporal scale.

[0005] Traditional prediction models typically treat soil as a single, homogeneous nutrient storage medium, failing to adequately consider the complex dynamics of nutrient transformation from "chemical totality" to "bioavailable quantity." In actual farming environments, nutrient migration and transformation are influenced by the nonlinear coupling of multiple factors, including soil temperature and humidity, root exudate activity, soil microbial community evolution, and the physicochemical properties of the rhizosphere microenvironment. This influence leads to significant time lags and transformation losses in nutrient supply at the soil-crop interface. When prediction methods cannot capture and quantify the driving force of citrus plant physiological signals on nutrient availability in real time, an imbalance can easily occur, resulting in relative nutrient scarcity during physiological peaks and excessive accumulation during physiological dormancy. This imbalance not only reduces fertilizer utilization but may also induce root physiological disorders due to excessively high local nutrient concentrations, leading to root autotoxicity or secondary soil salinization, thus producing irreversible negative feedback on citrus growth.

[0006] If we only focus on the static analysis of soil data without deeply exploring the nutritional status and dynamic nutrient absorption patterns of citrus plants themselves, it will be difficult to build a truly meaningful "supply and demand synergy" prediction model. Current prediction technologies often lack efficient means to process nonlinear and non-stationary signals when dealing with the complex mapping relationships between multi-source environmental data and physiological and biochemical indicators, resulting in limited robustness and applicability of prediction results.

[0007] Therefore, how to overcome the static and one-sided nature of the nutrient prediction logic in existing technologies, deeply couple the physiological fertilizer requirements of citrus with the dynamic supply capacity of soil, and construct a precise prediction scheme for soil nutrient supply that can adaptively respond to changes in crop physiological state has become an urgent technical problem to be solved in the field of efficient utilization of agricultural resources. Summary of the Invention

[0008] The purpose of this invention is to provide a method for predicting soil nutrient supply in response to physiological needs, in order to solve the technical problem.

[0009] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: A method for predicting soil nutrient supply in response to physiological needs, the method being implemented based on an architecture coupling multidimensional data perception and physiological feedback, comprising the following steps: S1: Real-time acquisition of multi-source raw data including environmental meteorological data, soil physicochemical parameters, and plant physiological status data: First, a multi-source sensing array deployed in the citrus orchard is used to acquire environmental meteorological data, soil physicochemical parameters, and citrus plant physiological status data in real time. The environmental meteorological data includes photosynthetically active radiation (PAR), air temperature and humidity, rainfall, and wind speed; the soil physicochemical parameters include soil nitrogen, phosphorus, and potassium electrode potential values, soil moisture content, soil temperature, and soil electrical conductivity (EC) at different depths (20cm, 40cm, 60cm); the citrus plant physiological status data includes stem runoff rate and leaf chlorophyll relative content (SPAD value).

[0010] S2: The multi-source raw data is preprocessed, and core environmental factors related to nutrient absorption rate are screened using the Pearson correlation coefficient method. The preprocessing includes using a Kalman filter algorithm to denoise the non-stationary electrical signals collected by the sensors, and using a local outlier detection algorithm to remove abnormal data points. For the processed data, this invention constructs a time-series feature space using the sliding window method, extracts frequency domain features reflecting the citrus growth status using discrete wavelet transform (DWT), and screens core environmental factors with a correlation coefficient greater than 0.75 with the nutrient absorption rate using the Pearson correlation coefficient method.

[0011] S3: Using a hybrid recognition model based on convolutional neural networks and long short-term memory networks, preprocessed physiological state data and environmental meteorological data are used as input vectors to automatically identify the current phenological state. W stage This study establishes a hybrid recognition model based on convolutional neural networks and long short-term memory networks (CNN-LSTM) to automatically identify the specific phenological stage of citrus trees. The phenological stages include budding, flowering, physiological fruit drop, fruit enlargement, and ripening / coloring. The hybrid recognition model uses physiological state data of the citrus plant and environmental meteorological data as input vectors. It extracts spatial features through a CNN layer and captures long-range dependencies in the time series through an LSTM layer, thereby determining the phenological state of the citrus tree at the current moment. W stage .

[0012] S4: Construct a dynamic nutrient demand model that responds to physiological needs, and utilize the phenological state quantities output by the hybrid recognition model. W stage The physiological fertilizer intensity index (PDI) is calculated at the current moment by combining photosynthetically active radiation, ambient temperature, and the geometric increment of the plant.

[0013] S5: Construct a dynamic transformation model of soil nutrient bioavailability to characterize the transformation flux of nutrients to bioavailable quantities in the soil and calculate the soil nutrient supply flux. Jsoil The model takes into account soil moisture content. i and soil temperature T soil The nonlinear effects of soil nutrient diffusion rate and microbial mineralization rate are investigated. Soil nutrient supply flux is defined. J soil for: ; This model can simulate in real time the soil’s ability to provide absorbable nutrients to citrus roots under complex physical conditions.

[0014] S6: Integrate the Physiological Nutrient Demand Index (PDI) with the Soil Nutrient Supply Flux J soil Calculate the optimal soil nutrient supply based on adaptive weight allocation. Q predict A multi-objective optimization algorithm was employed, with the constraints of minimizing the nutrient supply-demand gap and minimizing the risk of rhizosphere salinization. The objective function is expressed as: By solving this optimization problem, we can obtain the optimal nutrient dosage that the soil needs to be supplemented with while ensuring the physiological needs of citrus.

[0015] S7: Establish an error compensation mechanism based on feedback regulation. By comparing the deviation between the actual physiological response of the plant and the expected response under the predicted supply, the parameter bias in the prediction model is dynamically corrected. This error compensation mechanism dynamically corrects the parameter bias in the prediction model by comparing the deviation between the actual physiological response of the citrus plant (such as the instantaneous change in stem runoff and the shift in leaf reflectance spectrum) and the expected response under the predicted supply, using PID control logic or a fuzzy logic controller. When the actual physiological response is lower than expected, the correction module automatically increases the nutrient prediction supply coefficient for the next cycle; when a nutrient enrichment warning is detected in the rhizosphere environment, the predicted supply is forcibly decreased, thus achieving true closed-loop adaptive prediction.

[0016] Preferably, the preprocessing process in step S2 is as follows: S21: The Kalman filter algorithm is used to recursively denoise the non-stationary electrical signal output by the sensor on the multi-source raw data, establish the system state equation and observation equation, complete the prediction of state and error covariance through time update, calculate the Kalman gain through observation update and complete the state correction, and remove electromagnetic interference, random drift and vibration noise in the signal to obtain a smooth and stable time series signal. S22: The local outlier detection algorithm is used to identify and remove abnormal data points caused by hardware failures and drastic environmental fluctuations, retaining valid observation samples. The sliding window method is used to construct a multi-dimensional feature space for the cleaned time series. The window length is dynamically set according to the data sampling frequency, and the unstructured time series data is transformed into a fixed-dimensional feature vector that is suitable for subsequent models. S23: The processed data is decomposed into multi-scale frequency domain using discrete wavelet transform. The signal is decomposed into low-frequency approximation coefficients that reflect the overall growth trend and high-frequency detail coefficients that reflect environmental stress and physiological fluctuations. The mean, standard deviation, energy and information entropy statistics of each layer coefficient are extracted to construct frequency domain features that can characterize the growth status and sensitivity to environmental fluctuations. S24: The correlation between various environmental factors, soil factors and nutrient uptake rates was calculated using the Pearson correlation coefficient method, and core environmental factors with an absolute correlation coefficient greater than 0.75 were screened out.

[0017] Preferably, the specific process of step S24 is as follows: S241: Using the amount of nutrients absorbed by the plant per unit time as the dependent variable for correlation calculations: v nut =Δ M nut / Δ t ; in, v nut Nutrient absorption rate; Δ M nut Δ: Increase in nutrient absorption by the plant during the specified time period; t : Observation time interval; S242: For each candidate environmental factor X, the relationship between nutrient uptake rate and v nut Calculate the linear correlation coefficient: ; in, r X,v Environmental factor X and nutrient absorption rate v nut The Pearson correlation coefficient; X i : No. i Measured values ​​of environmental factors at each sample point; v i : No. i Measured values ​​of nutrient absorption rate at each sample point; : Sample mean of environmental factor X; Nutrient absorption rate v nut The sample mean; nTotal number of valid samples; S243: Perform core factor screening: ; Wherein, Ω: the set of core environmental factors obtained through screening; 0.75: the threshold for determining strong correlation.

[0018] Preferably, the specific process of automatically identifying phenological periods using a hybrid recognition model in step S3 is as follows: S31: A high-dimensional spatiotemporal input feature vector is constructed based on plant physiological state data, environmental meteorological data, and frequency domain features obtained through feature enhancement. X input : X input =[ X env , X phy , X freq ]; in, X env This represents the environmental meteorological feature vector; X phy This represents the physiological characteristic vector of the plant. X freq This represents the frequency domain eigenvector of the DWT. S32: Extract spatial features from the feature vector using the CNN layer in the hybrid recognition model, wherein the CNN layer adopts a dilated convolution structure to increase the receptive field and capture large-scale environmental spatiotemporal features. S33: The LSTM layer in the hybrid recognition model is used to capture the long-term dependence of growth and development at different time scales, wherein the LSTM layer adopts a bidirectional structure and introduces an attention mechanism to adaptively allocate the weights of different environmental factors according to the input features. S34: Output the phenological state at the current moment through a fully connected layer and a Softmax classifier. W stage The phenological state quantity W stage This includes the budding stage, flowering stage, physiological fruit drop stage, fruit enlargement stage, and ripening and color change stage.

[0019] The hybrid recognition model employs CNN layers with dilated convolution structures to increase the receptive field and capture broader spatiotemporal environmental features. The LSTM layers utilize a bidirectional (Bi-LSTM) structure, considering not only the evolutionary patterns of historical data but also allocating weights for different environmental factors through an attention mechanism. For example, during the physiological fruit drop period, the algorithm automatically increases the weight allocation for potassium and calcium features to respond to the physiological need for citrus fruits to strengthen their stems.

[0020] Preferably, the specific process of step S4 is as follows: S41: Obtain the current phenological state quantity W stage Load the corresponding photosynthetic cumulative physiological weight coefficient α plant geometric increment weighting coefficient β ; S42: Based on photosynthetically active radiation PAR Calculate instantaneous photosynthetic strength P n : P n = or · PAR ; in, or Light energy utilization efficiency; S43: Constructing the ambient temperature regulation function f ( T air Influence factors on saturated vapor pressure deficit g ( VPD ): ; in, T air Ambient temperature; T min ,T opt ,T max : These represent the lower limit temperature, optimum temperature, and upper limit temperature for growth, respectively; λ1, λ2: shape factors; ; in, VPD : Saturated vapor pressure deficit; VPD 0: Stress threshold; c : Fit coefficient; S44: Calculate the geometric increment of the plant during the monitoring period ΔD / Δt, where ΔD is the real-time increment of the stem or fruit diameter during the monitoring period; and Δt is the monitoring time interval. S45: Perform integration and weighted summation to obtain the Physiological Digestive Intensity Index (PDI) at the current moment: .

[0021] Preferably, the specific process of step S5 is as follows: S51: Soil nutrient supply flux is defined as the vector sum of effective diffusion flux, water convection transport flux, and microbial competitive consumption flux, as shown in the following formula: ; S52: Calculate the effective nutrient diffusion coefficient D eff Soil volumetric moisture content i With soil temperature T soil The nonlinear correction is performed using the following formula: ; in, D eff The effective diffusion coefficient of nutrients; i This refers to the soil volumetric water content. T soil D0 represents soil temperature; D0 represents the nutrient diffusion coefficient under standard conditions. d This refers to the temperature response coefficient. e The response coefficient is the moisture content coefficient. S53: Determine the nutrient concentration gradient ∇C between the rhizosphere and the soil as the driving term for nutrient diffusion; S54: Calculate moisture convective transport terms v · C ,in v The average velocity of soil moisture infiltration or convection. C This refers to the concentration of available nutrients in the soil solution. S55: Excluding rhizosphere microbial competitive consumption flux R sink The uncorrected soil nutrient supply flux was obtained. J soil ; S56: Introduction of a rhizosphere pH dynamic correction factor x ( pH rhizo The availability of fixed phosphorus and trace elements was further corrected to obtain the final soil nutrient supply fluxes used in the model calculations. ; in, x ( pH rhizo ) is a rhizosphere pH correction factor; pH rhizo pH value of the rhizosphere microenvironment; This provides flux for the final soil nutrients after modification by the rhizosphere environment.

[0022] Furthermore, this invention addresses the unique characteristics of the citrus rhizosphere microenvironment by introducing a root exudate activity correction factor into the prediction model. Citrus trees secrete different types of organic acids (such as citric acid and malic acid) at different physiological stages, and these exudates significantly alter rhizosphere pH, thereby affecting the availability of fixed phosphorus and trace elements in the soil. This invention establishes a rhizosphere pH dynamic evolution sub-model, combining initial soil pH and estimated root biomass values, to predict soil nutrient supply fluxes. J soil Nonlinear corrections were applied. The corrected formula further considers the influence of ion exchange capacity (CEC), ensuring extremely high accuracy in predicting nutrient supply under different soil pH conditions.

[0023] More specifically, an integrated soil multi-parameter sensor array is employed, with its probe surface treated with a special coating to reduce polarization effects, enabling the simultaneous acquisition of high-precision electrochemical signals. For physiological signal acquisition, a non-contact infrared thermal imaging sensor is used to monitor citrus canopy temperature. By calculating the difference between canopy temperature and air temperature (CWSI, Crop Water Stress Index), the metabolic activity of the citrus is assessed, serving as an important correction parameter for the PDI index.

[0024] Preferably, the specific process of step S6 is as follows: S61: Construct a multi-objective optimization objective function, which weights and fuses the nutrient supply and demand balance error with the rhizosphere conductivity safety constraint. The formula is as follows: ; Where F is the value of the multi-objective optimization objective function; w 1. Nutrient supply and demand balance weighting coefficient w 2. Rhizosphere environmental safety weighting coefficient; PDI(t) is the physiological fertilizer intensity index at time t; J soil ( t (t) represents the soil nutrient supply flux at time t; EC soil (z) represents the real-time soil electrical conductivity at soil depth z; EC safe The threshold for root system safety tolerance to soil electrical conductivity; z is the vertical soil depth. The degree of soil electrical conductivity exceeding the standard within the root distribution depth range is integrated; S62: From phenological state quantities W stage Dynamically allocate adaptive weights w 1. w2. Adjust the priority of supply and demand matching and root security in different phenological stages; S63: Under the constraint that the soil electrical conductivity is within a safe range, solve the above multi-objective optimization function to make the objective function... F Take the minimum value and output the optimal soil nutrient supply for the next fertilization cycle. Q predict .

[0025] Preferably, the specific process of step S7 is as follows: S71: Select the slope of the stem runoff rate change S sr The Crop Water Stress Index (CWSI) is used as a core physiological response indicator to calculate the relative deviation E between the actual and expected responses. ; in, R actual : Actual physiological response value of the plant; R predict The expected physiological response value output by the model; S72: Deviation Judgment and Grading Correction: When E<0 and |E|> E th At that time, the intensity of nutrient supply in the next cycle will be adjusted upwards. E th The threshold for determining deviation; Soil electrical conductivity EC soil ≥ EC safe At that time, a security-mandated correction was triggered, and the setting was lowered. Q predict ; S73: Calculate correction coefficients using a PID controller. K adj Dynamically optimize the PDI weight and supply flux coefficient; S74: Apply the correction coefficient to the prediction model for the next cycle to obtain the updated optimal soil nutrient supply.

[0026] Preferably, it also includes a regulation mechanism based on trace element sensitivity: collecting the reflectance of leaves in the characteristic bands of 550nm and 680nm, and using a spectral inversion algorithm to estimate the activity levels of key metabolic enzymes related to trace elements in the plant; adjusting the optimal soil nutrient supply based on the real-time status of enzyme activity. Q predict The weighting of each trace element in the spectrum is determined, and a warning and replenishment of a specific trace element is triggered based on the offset of the red edge position in the spectrum.

[0027] Furthermore, the prediction method described in this invention specifically introduces an adsorption-desorption isotherm model (Langmuir or Freundlich isotherm) for predicting phosphorus and potassium supply. Since phosphorus is readily fixed in soil, by monitoring the ion concentration in the soil liquid phase in real time, the pooling capacity of solid nutrients can be calculated backwards, thereby accurately determining the minimum exogenous replenishment required to achieve a specific physiological demand intensity. This prediction logic based on kinetic equilibrium greatly reduces the ineffective accumulation of phosphorus in the soil.

[0028] To verify the effectiveness of the prediction results, physiological health indicators were also evaluated. The maximum photochemical efficiency of photosystem II was assessed by real-time monitoring of fluorescence parameters in citrus leaves, such as the Fv / Fm value. If the Fv / Fm value stabilized within the ideal range of 0.80-0.85 after the predicted nutrient supply scheme was implemented, the current nutrient supply prediction scheme was deemed effective; if the value showed a downward trend, the prediction algorithm's self-calibration procedure was triggered, and the soil buffer capacity parameters were re-evaluated.

[0029] 1. Nutrient prediction through the Physiological Nutrient Intensity Index (PDI): This shifts the focus from traditional "yield-driven" to "plant physiological demand-driven" methods, accurately matching the real-time nutrient requirements of different phenological stages. This addresses the fundamental issue of misaligned nutrient supply and demand in time and space, significantly improving fertilizer utilization efficiency and reducing fertilizer input costs.

[0030] 2. Construct a dynamic transformation model of soil nutrient bioavailability, comprehensively considering the effects of soil temperature and humidity, concentration gradient, water convection, microbial consumption and rhizosphere pH changes on nutrient availability, to achieve dynamic quantification and accurate simulation of soil supply capacity, overcome the shortcomings of traditional methods that rely solely on static soil testing data, and significantly improve the scientific rigor and accuracy of predictions.

[0031] 3. The CNN-LSTM hybrid recognition model is used to automatically determine the phenological stage of citrus. By combining multi-source sensing data and frequency domain features, the phenological stage can be identified efficiently, objectively and continuously, avoiding errors and lags in manual observation, and providing a reliable stage basis for precise nutrient management throughout the entire cycle.

[0032] 4. Establish multi-objective optimization prediction to constrain the safe range of rhizosphere soil electrical conductivity while meeting the balance of nutrient supply and demand, taking into account both nutrient supply sufficiency and rhizosphere environment safety, effectively reducing the risks of soil salinization, nutrient leaching and agricultural non-point source pollution, and improving the ecological sustainability of orchard production.

[0033] 5. Construct a physiological feedback closed-loop adaptive correction mechanism to dynamically adjust model parameters and nutrient supply based on the actual physiological response of the plant. It has strong anti-interference and self-optimization capabilities and can adapt to changes in weather fluctuations, soil differences and plant growth status, making the system more robust.

[0034] 6. By precisely responding to the physiological needs of plants and optimizing nutrient supply, fertilizer utilization can be improved, which can significantly promote citrus growth, increase single fruit weight and soluble solids content, and steadily improve fruit quality and yield, resulting in outstanding synergistic economic and ecological benefits. Attached Figure Description

[0035] Figure 1 This is a flowchart illustrating the method for predicting soil nutrient supply in response to physiological needs according to the present invention.

[0036] Figure 2 This is a schematic diagram of the architecture of the hybrid recognition model of the present invention. Detailed Implementation

[0037] The following is in conjunction with the appendix Figure 1~Figure 2 The present invention will be further described in detail below: See appendix Figure 1 As shown, a method for predicting soil nutrient supply in response to physiological needs is described. This method is based on an architecture that couples multidimensional data perception with physiological feedback, and includes the following steps: S1: Real-time collection of multi-source raw data, including environmental meteorological data, soil physicochemical parameters, and physiological status data of citrus plants, is achieved through a multi-source sensing array deployed in the citrus orchard. The multi-source sensing array includes environmental meteorological sensors, soil parameter sensors, plant physiological sensors, and infrared thermal imaging sensors.

[0038] S2: The multi-source raw data is preprocessed using edge computing nodes, the electrical signals collected by the sensors are denoised using the Kalman filter algorithm, and the core environmental factors related to nutrient absorption rate are screened using the Pearson correlation coefficient method.

[0039] S3: Using a hybrid recognition model based on convolutional neural networks and long short-term memory networks (CNN-LSTM), preprocessed physiological state data and environmental meteorological data are used as input vectors to automatically identify the current phenological state of citrus trees. W stage .

[0040] S4: Construct a dynamic nutrient demand model that responds to the physiological needs of citrus, and utilize the phenological state quantities output by the hybrid recognition model. W stage By combining photosynthetically active radiation, ambient temperature, and geometric increment of citrus plants, the physiological fertilizer intensity index (PDI) at the current moment is calculated.

[0041] S5: Construct a dynamic transformation model of soil nutrient bioavailability to characterize the transformation flux of nutrients to bioavailable quantities in the soil and calculate the soil nutrient supply flux. J soil .

[0042] S6: Integrate the Physiological Nutrient Demand Index (PDI) with the Soil Nutrient Supply Flux J soil Calculate the optimal soil nutrient supply based on adaptive weight allocation. Q predict .

[0043] S7: Establish an error compensation mechanism based on feedback regulation, and dynamically correct the parameter bias in the prediction model by comparing the deviation between the actual physiological response and the expected response of citrus plants under the predicted supply.

[0044] The specific preprocessing steps in step S2 are as follows: S21: The Kalman filter algorithm is used to recursively denoise the non-stationary electrical signal output by the sensor on the multi-source raw data, establish the system state equation and observation equation, complete the prediction of state and error covariance through time update, calculate the Kalman gain through observation update and complete the state correction, and remove electromagnetic interference, random drift and vibration noise in the signal to obtain a smooth and stable time series signal.

[0045] S22: The Local Outlier (LOF) detection algorithm is used to identify and remove abnormal data points caused by hardware failures and drastic environmental fluctuations, retaining valid observation samples. A sliding window method is used to construct a multi-dimensional feature space for the cleaned time series, with the window length dynamically set according to the data sampling frequency, transforming unstructured time series data into fixed-dimensional feature vectors that are suitable for subsequent models.

[0046] S23: The processed data is decomposed into multiple-scale frequency domains using Discrete Wavelet Transform (DWT). The signal is decomposed into low-frequency approximation coefficients that reflect the overall growth trend and high-frequency detail coefficients that reflect environmental stress and physiological fluctuations. The mean, standard deviation, energy and information entropy statistics of each layer coefficient are extracted to construct frequency domain features that can characterize the growth status and environmental fluctuation sensitivity of citrus.

[0047] S24: The correlation between various environmental factors, soil factors and citrus nutrient uptake rate was calculated using the Pearson correlation coefficient method. Core environmental factors with an absolute correlation coefficient greater than 0.75 were selected as standard inputs for subsequent phenological period identification, physiological fertilizer demand intensity index calculation and soil nutrient supply flux simulation.

[0048] The specific process of step S24 is as follows: S241: Using the amount of nutrients absorbed by the citrus plant per unit time (nutrient absorption rate) as the dependent variable for correlation calculations: v nut =Δ M nut / Δ t ; in, v nut Citrus nutrient absorption rate, unit: mg / (plant·h); Δ M nut Δ represents the increase in nutrient absorption by the plant during the specified time period, expressed in mg. t Observation time interval, unit: h; S242: For each candidate environmental factor X, the relationship between nutrient uptake rate and v nut Calculate the linear correlation coefficient: ; in, r X,v Environmental factor X and nutrient absorption rate v nut The Pearson correlation coefficient, with a value range of [−1,1]; X i : No. i Measured values ​​of environmental factors (temperature, moisture content, PAR, EC, etc.) at each sample point; v i : No. i Measured values ​​of nutrient absorption rate at each sample point; : Sample mean of environmental factor X; Nutrient absorption rate v nut The sample mean; n : The total number of valid samples, i.e. the number of samples after noise reduction and anomaly removal; S243: Perform core factor screening: ; Where Ω: the set of core environmental factors obtained through screening; 0.75: the threshold for determining strong correlation; Core environmental factors were used as input variables for subsequent phenological stage identification, physiological fertilizer demand intensity index calculation, and soil nutrient supply flux simulation.

[0049] Example 2 Based on Example 1, see Figure 2 As shown, the specific process of automatically identifying the phenological period of citrus fruits using a hybrid recognition model in step S3 is as follows: S31: A high-dimensional spatiotemporal input feature vector is constructed based on the physiological state data of citrus plants, environmental meteorological data, and frequency domain features obtained through feature enhancement. X input : X input =[ X env ,X phy , X freq ]; in, X env The environmental meteorological characteristic vector includes photosynthetically active radiation (PAR), ambient temperature, relative humidity, saturated vapor pressure deficit (VPD), wind speed, and precipitation; X phy The plant physiological characteristic vector includes stem runoff rate, leaf SPAD value, canopy temperature, crop water stress index (CWSI), and stem / fruit diameter increment; X freq The DWT frequency domain feature vector includes the mean and energy of the low-frequency approximation coefficients; and the standard deviation and information entropy of the high-frequency detail coefficients.

[0050] S32: Extract spatial features from the feature vector using the CNN layer in the hybrid recognition model, wherein the CNN layer adopts a dilated convolution structure to increase the receptive field and capture large-scale environmental spatiotemporal features.

[0051] S33: The LSTM layer in the hybrid recognition model is used to capture the long-range dependence of citrus growth and development at different time scales. The LSTM layer adopts a bidirectional structure and introduces an attention mechanism to adaptively allocate the weights of different environmental factors according to the input features.

[0052] S34: Output the phenological state of the citrus fruit at the current moment through a fully connected layer and a Softmax classifier. W stage The phenological state quantity W stage This includes the budding stage, flowering stage, physiological fruit drop stage, fruit enlargement stage, and ripening and color change stage.

[0053] The specific process of step S4 is as follows: S41: Obtain the current phenological state quantity W stage Load the corresponding photosynthetic cumulative physiological weight coefficient α plant geometric increment weighting coefficient β .

[0054] S42: Based on photosynthetically active radiation PAR Calculate instantaneous photosynthetic strength P n : P n = or · PAR ; in, or This refers to the efficiency of light energy utilization.

[0055] S43: Constructing the ambient temperature regulation function f ( T air Influence factors on saturated vapor pressure deficit g ( VPD ): ; in, T air Ambient temperature; T min ,T opt ,T max λ1 and λ2 are the lower limit temperature, optimum temperature, and upper limit temperature for citrus growth, respectively; λ1 and λ2 are shape coefficients, which are inherent physiological parameters of citrus varieties. They were obtained by fitting the temperature-physiological response curve using the nonlinear least squares method through field multi-gradient temperature experiments. ; in, VPD : Saturated vapor pressure deficit; VPD 0: Stress threshold; c : Fitting coefficient.

[0056] S44: Calculate the geometric increment of the plant ΔD / Δt during the monitoring period. ΔD is the real-time increment of the diameter of the citrus stem or fruit during the monitoring period, in mm; Δt is the monitoring time interval, in h. S45: Perform integration and weighted summation to obtain the Physiological Digestive Intensity Index (PDI) at the current moment: .

[0057] Example 3 Based on Example 1 or Example 2, the specific process of step S5 is as follows: S51: Soil nutrient supply flux is defined as the vector sum of effective diffusion flux, water convection transport flux, and microbial competitive consumption flux, as shown in the following formula: ; S52: Calculate the effective nutrient diffusion coefficient D eff Soil volumetric moisture content i With soil temperature T soil The nonlinear correction is performed using the following formula: ; in, D eff The effective nutrient diffusion coefficient is expressed in pseudo-cm.2 ⋅h −1 ; i This refers to the soil volumetric water content. T soil Soil temperature, in °C; D0 is the nutrient diffusion coefficient under standard conditions; d This refers to the temperature response coefficient. e The response coefficient is the moisture content coefficient. S53: Determine the nutrient concentration gradient ∇C between the rhizosphere and the soil as the driving term for nutrient diffusion; S54: Calculate moisture convective transport terms v · C ,in v The average velocity of soil moisture infiltration or convection (cm·h) −1 ), C The effective nutrient concentration in the soil solution is expressed in mg·cm³. −3 ; S55: Excluding rhizosphere microbial competitive consumption flux R sink The uncorrected soil nutrient supply flux was obtained. J soil ; S56: Introduction of a rhizosphere pH dynamic correction factor x ( pH rhizo The availability of fixed phosphorus and trace elements was further corrected to obtain the final soil nutrient supply fluxes used in the model calculations. ; in, x ( pH rhizo ) is a rhizosphere pH correction factor; pH rhizo pH value of the rhizosphere microenvironment; The final soil nutrient supply flux after rhizosphere environmental modification, in mg·cm³. −2 ·h −1 .

[0058] The specific process of step S6 is as follows: S61: Construct a multi-objective optimization objective function, which weights and fuses the nutrient supply and demand balance error with the rhizosphere conductivity safety constraint. The formula is as follows: ; Where F is the objective function value of the multi-objective optimization, used to measure the comprehensive cost of nutrient supply, and the optimization objective is to minimize the value; w 1. Nutrient supply and demand balance weighting coefficient w 2. The root environment safety weighting coefficient is derived from the phenological state quantity. Wstage Adaptive allocation w 1+ w 2=1; PDI(t) is the physiological fertilizer intensity index at time t; J soil ( t (t) represents the soil nutrient supply flux at time t; EC soil (z) represents the real-time soil electrical conductivity at soil depth z; EC safe The threshold for the safe tolerance of citrus roots to soil electrical conductivity can be set to 1.5–2.2 mS / cm; z represents the vertical soil depth, which is the depth of the root concentration layer, in cm. The degree of soil electrical conductivity exceeding the standard within the root distribution depth range is integrated to characterize the overall rhizosphere salinization risk; S62: Phenological state quantity output from S3 W stage Dynamically allocate adaptive weights w 1. w 2. Adjust the priority of supply and demand matching and root security in different phenological stages; S63: Under the constraint that the soil electrical conductivity is within a safe range, solve the above multi-objective optimization function to make the objective function... F Take the minimum value and output the optimal soil nutrient supply for the next fertilization cycle. Q predict .

[0059] The specific process of step S7 is as follows: S71: Calculation of Physiological Response Deviation: Selecting the Slope of Stem Runoff Rate Change S sr The Crop Water Stress Index (CWSI) is used as a core physiological response indicator to calculate the relative deviation E between the actual and expected responses. ; in, R actual Actual physiological response values ​​of citrus plants (stem runoff rate, CWSI, etc.) are from monitoring data after S2 pretreatment; R predict The expected physiological response value output by the model is determined by PDI and J soil Obtained through coupled calculations; S72: Deviation Judgment and Grading Correction: When E<0 and |E|> E th When this occurs, it indicates that the actual physiological response is insufficient, triggering a positive correction and increasing the intensity of nutrient supply in the next cycle. E thDeviation judgment threshold, which can be set to 0.10~0.15; correction will be initiated if the value exceeds this threshold. When soil electrical conductivity EC soil ≥ EC safe At that time, a security-mandated correction was triggered, and the setting was lowered. Q predict ; S73: Calculate correction coefficients using a PID controller. K adj Dynamically optimize the PDI weight and supply flux coefficient: ; in, K adj Model parameter correction coefficients, output by the PID controller, are used for global tuning; K p : Proportional coefficient K i Integral coefficient K d : Differential coefficients, PID controller parameters; S74: Apply the correction coefficient to the prediction model for the next cycle to obtain the updated optimal soil nutrient supply: ; when EC soil ≥ EC safe At that time, mandatory constraints: ; in, Q predict : The optimal soil nutrient supply before correction, from S6; The optimal soil nutrient supply for the new cycle, after feedback and correction.

[0060] In another embodiment of this invention, a regulation mechanism based on trace element sensitivity is also included: the reflectance of citrus leaves in the 550nm and 680nm characteristic bands is collected using the sensing array, and the activity levels of key metabolic enzymes related to trace elements in the plant are estimated using a spectral inversion algorithm; the optimal soil nutrient supply is adjusted according to the real-time status of enzyme activity. Q predict The weighting of each trace element in the spectrum is determined, and a warning and replenishment of a specific trace element is triggered based on the offset of the red edge position in the spectrum.

[0061] In summary, the soil nutrient supply prediction method responsive to physiological needs provided by this invention transforms complex citrus cultivation experience into a quantifiable, executable, and iterative process by constructing a closed-loop technical chain encompassing physiological perception, phenological identification, nutrient requirement modeling, transformation simulation, and feedback correction. This method overcomes the limitations of traditional nutrient management logic and demonstrates significant advantages in improving citrus quality, protecting the agricultural environment, and reducing production costs.

Claims

1. A method for predicting soil nutrient supply in response to physiological needs, characterized in that, The method is based on an architecture that couples multidimensional data perception with physiological feedback, and includes the following steps: S1: Real-time collection of multi-source raw data, including environmental meteorological data, soil physicochemical parameters, and plant physiological status data; S2: Preprocess the multi-source raw data and use the Pearson correlation coefficient method to screen out the core environmental factors related to nutrient absorption rate; S3: Using a hybrid recognition model based on convolutional neural networks and long short-term memory networks, preprocessed physiological state data and environmental meteorological data are used as input vectors to automatically identify the current phenological state. W stage ; S4: Construct a dynamic nutrient demand model that responds to physiological needs, and utilize the phenological state quantities output by the hybrid recognition model. W stage The physiological fertilizer intensity index (PDI) at the current moment is calculated by combining photosynthetically active radiation, ambient temperature, and geometric increment of the plant. S5: Construct a dynamic transformation model of soil nutrient bioavailability to characterize the transformation flux of nutrients to bioavailable quantities in the soil and calculate the soil nutrient supply flux. J soil ; S6: Integrate the Physiological Nutrient Demand Index (PDI) with the Soil Nutrient Supply Flux J soil Calculate the optimal soil nutrient supply based on adaptive weight allocation. Q predict ; S7: Establish an error compensation mechanism based on feedback regulation, and dynamically correct the parameter bias in the prediction model by comparing the deviation between the actual physiological response of the plant and the expected response under the predicted supply.

2. The method for predicting soil nutrient supply in response to physiological needs according to claim 1, characterized in that, The specific preprocessing steps in step S2 are as follows: S21: The Kalman filter algorithm is used to recursively denoise the non-stationary electrical signal output by the sensor on the multi-source raw data, establish the system state equation and observation equation, complete the prediction of state and error covariance through time update, calculate the Kalman gain through observation update and complete the state correction, and remove electromagnetic interference, random drift and vibration noise in the signal to obtain a smooth and stable time series signal. S22: The local outlier detection algorithm is used to identify and remove abnormal data points caused by hardware failures and drastic environmental fluctuations, retaining valid observation samples. The sliding window method is used to construct a multi-dimensional feature space for the cleaned time series. The window length is dynamically set according to the data sampling frequency, and the unstructured time series data is transformed into a fixed-dimensional feature vector that is suitable for subsequent models. S23: The processed data is decomposed into multi-scale frequency domain using discrete wavelet transform. The signal is decomposed into low-frequency approximation coefficients that reflect the overall growth trend and high-frequency detail coefficients that reflect environmental stress and physiological fluctuations. The mean, standard deviation, energy and information entropy statistics of each layer coefficient are extracted to construct frequency domain features that can characterize the growth status and sensitivity to environmental fluctuations. S24: The correlation between various environmental factors, soil factors and nutrient uptake rates was calculated using the Pearson correlation coefficient method, and core environmental factors with an absolute correlation coefficient greater than 0.75 were screened out.

3. The method for predicting soil nutrient supply in response to physiological needs according to claim 2, characterized in that, The specific process of step S24 is as follows: S241: Using the amount of nutrients absorbed by the plant per unit time as the dependent variable for correlation calculations: v nut =D M nut / D t ; in, v nut Nutrient absorption rate; Δ M nut Δ: Increase in nutrient absorption by the plant during the specified time period; t : Observation time interval; S242: For each candidate environmental factor X, the relationship between nutrient uptake rate and v nut Calculate the linear correlation coefficient: ； in, r X,v Environmental factor X and nutrient absorption rate v nut The Pearson correlation coefficient; X i : No. i Measured values ​​of environmental factors at each sample point; v i : No. i Measured values ​​of nutrient absorption rate at each sample point; : Sample mean of environmental factor X; Nutrient absorption rate v nut The sample mean; n Total number of valid samples; S243: Perform core factor screening: ； Wherein, Ω: the set of core environmental factors obtained through screening; 0.75: the threshold for determining strong correlation.

4. The method for predicting soil nutrient supply in response to physiological needs according to claim 3, characterized in that, The specific process of automatically identifying phenological periods using a hybrid recognition model in step S3 is as follows: S31: A high-dimensional spatiotemporal input feature vector is constructed based on plant physiological state data, environmental meteorological data, and frequency domain features obtained through feature enhancement. X input : X input =[ X env , X phy , X freq ]； in, X env This represents the environmental meteorological feature vector; X phy This represents the physiological characteristic vector of the plant. X freq This represents the frequency domain eigenvector of the DWT. S32: Extract spatial features from the feature vector using the CNN layer in the hybrid recognition model, wherein the CNN layer adopts a dilated convolution structure to increase the receptive field and capture large-scale environmental spatiotemporal features. S33: The LSTM layer in the hybrid recognition model is used to capture the long-term dependence of growth and development at different time scales, wherein the LSTM layer adopts a bidirectional structure and introduces an attention mechanism to adaptively allocate the weights of different environmental factors according to the input features. S34: Output the phenological state at the current moment through a fully connected layer and a Softmax classifier. W stage The phenological state quantity W stage This includes the budding stage, flowering stage, physiological fruit drop stage, fruit enlargement stage, and ripening and color change stage.

5. The method for predicting soil nutrient supply in response to physiological needs according to claim 1, characterized in that, The specific process of step S4 is as follows: S41: Obtain the current phenological state quantity W stage Load the corresponding photosynthetic cumulative physiological weight coefficient α plant geometric increment weighting coefficient β ; S42: Based on photosynthetically active radiation PAR Calculate instantaneous photosynthetic strength P n : P n = η · PAR ； in, η Light energy utilization efficiency; S43: Construct an ambient temperature regulation function f ( T air Influence factors of saturated vapor pressure deficit g ( VPD ): ； in, T air Ambient temperature; T min ,T opt ,T max : These represent the lower limit temperature, optimum temperature, and upper limit temperature for growth, respectively; λ1, λ2: shape factors; ； in, VPD : Saturated vapor pressure deficit; VPD 0: Stress threshold; γ : Fit coefficient; S44: Calculate the geometric increment of the plant during the monitoring period ΔD / Δt, where ΔD is the real-time increment of the stem or fruit diameter during the monitoring period; and Δt is the monitoring time interval. S45: Perform integration and weighted summation to obtain the Physiological Digestive Intensity Index (PDI) at the current moment: 。 6. The method for predicting soil nutrient supply in response to physiological needs according to claim 5, characterized in that, The specific process of step S5 is as follows: S51: Soil nutrient supply flux is defined as the vector sum of effective diffusion flux, water convection transport flux, and microbial competitive consumption flux, as shown in the following formula: ； S52: Calculate the effective nutrient diffusion coefficient D eff Soil volumetric moisture content θ With soil temperature T soil The nonlinear correction is performed using the following formula: ； in, D eff The effective diffusion coefficient of nutrients; θ This refers to the soil volumetric water content. T soil D0 represents soil temperature; D0 represents the nutrient diffusion coefficient under standard conditions. δ This refers to the temperature response coefficient. ε The response coefficient is the moisture content coefficient. S53: Determine the nutrient concentration gradient ∇C between the rhizosphere and the soil as the driving term for nutrient diffusion; S54: Calculate moisture convective transport terms v · C ,in v The average velocity of soil moisture infiltration or convection. C This refers to the concentration of available nutrients in the soil solution. S55: Excluding rhizosphere microbial competitive consumption flux R sink The uncorrected soil nutrient supply flux was obtained. J soil ; S56: Introduction of a rhizosphere pH dynamic correction factor ξ ( pH rhizo The availability of fixed phosphorus and trace elements was further corrected to obtain the final soil nutrient supply fluxes used in the model calculations. ； in, ξ ( pH rhizo ) is a rhizosphere pH correction factor; pH rhizo pH value of the rhizosphere microenvironment; This provides flux for the final soil nutrients after modification by the rhizosphere environment.

7. The method for predicting soil nutrient supply in response to physiological needs according to claim 6, characterized in that, The specific process of step S6 is as follows: S61: Construct a multi-objective optimization objective function, which weights and fuses the nutrient supply and demand balance error with the rhizosphere conductivity safety constraint. The formula is as follows: ； Where F is the value of the multi-objective optimization objective function; w 1. Nutrient supply and demand balance weighting coefficient w 2. Rhizosphere environmental safety weighting coefficient; PDI(t) is the physiological fertilizer intensity index at time t; J soil ( t (t) represents the soil nutrient supply flux at time t; EC soil (z) represents the real-time soil electrical conductivity at soil depth z; EC safe The threshold for root system safety tolerance to soil electrical conductivity; z is the vertical soil depth. The degree of soil electrical conductivity exceeding the standard within the root distribution depth range is integrated; S62: From phenological state quantities W stage Dynamically allocate adaptive weights w 1. w 2. Adjust the priority of supply and demand matching and root security in different phenological stages; S63: Under the constraint that the soil electrical conductivity is within a safe range, solve the above multi-objective optimization function to make the objective function... F Take the minimum value and output the optimal soil nutrient supply for the next fertilization cycle. Q predict .

8. The method for predicting soil nutrient supply in response to physiological needs according to claim 7, characterized in that, The specific process of step S7 is as follows: S71: Select the slope of the stem runoff rate change S sr The Crop Water Stress Index (CWSI) is used as a core physiological response indicator to calculate the relative deviation E between the actual and expected responses. ； in, R actual : Actual physiological response value of the plant; R predict The expected physiological response value output by the model; S72: Deviation Judgment and Grading Correction: When E<0 and |E|> E th At that time, the intensity of nutrient supply in the next cycle will be adjusted upwards. E th The threshold for determining deviation; Soil electrical conductivity EC soil ≥ EC safe At that time, a security-mandated correction was triggered, and the setting was lowered. Q predict ; S73: Calculate correction coefficients using a PID controller. K adj Dynamically optimize the PDI weight and supply flux coefficient; S74: Apply the correction coefficient to the prediction model for the next cycle to obtain the updated optimal soil nutrient supply.

9. The method for predicting soil nutrient supply in response to physiological needs according to claim 1, characterized in that, It also includes a regulatory mechanism based on micronutrient sensitivity: collecting leaf reflectance in the 550nm and 680nm characteristic bands, using a spectral inversion algorithm to estimate the activity levels of key metabolic enzymes related to micronutrients in the plant; and adjusting the optimal soil nutrient supply based on the real-time status of enzyme activity. Q predict The weighting of each trace element in the spectrum is determined, and a warning and replenishment of a specific trace element is triggered based on the offset of the red edge position in the spectrum.

Images