Method for reducing noon break proportion of citrus stomata based on moisture regulation and control

Abstract

The invention discloses a method for reducing the noon break proportion of citrus stomata based on moisture regulation and control, and particularly relates to the technical field of fruit tree physiological regulation and control. The method comprises the following steps: acquiring a citrus leaf stomatal opening time sequence G (t) and a leaf water potential parameter psi (t), and constructing a stomatal activity response model M1; setting a target noon time period, and calculating a pore closing proportion; establishing a moisture regulation response model M2 according to the soil moisture content and the transpiration rate, and predicting psi (t) changes under different irrigation conditions; inputting the psi (t) output by the M2 into the M1, predicting the influence of different moisture regulation and control strategies on the theta m, and selecting an optimal irrigation scheme to control the target irrigation time, water volume and frequency; dynamic iterative optimization is carried out through theta m actual measurement and a model self-learning mechanism, and accurate moisture management and pore behavior regulation and control are achieved; the method has high adaptability, intellectualization and self-evolution ability, the noon stomata closing proportion can be effectively reduced, and the photosynthetic efficiency and the fruit tree yield are improved.

Description

Technical Field

[0001] This invention relates to the field of fruit tree physiological regulation technology, specifically to a method for reducing the midday depression rate of citrus stomata based on water regulation. Background Technology

[0002] During the growing season, especially in the hot and sunny periods of summer, citrus trees exhibit a significant "midday depression" phenomenon in their stomatal behavior, meaning that stomata close or partially close at midday, leading to a marked decrease in the rate of photosynthesis. This phenomenon is particularly pronounced under hot and dry conditions and is a self-protective mechanism adopted by fruit trees to reduce water transpiration. However, it significantly inhibits carbon assimilation efficiency, thereby affecting yield and quality.

[0003] While some existing technologies attempt to alleviate midday depression through variety selection, shading, reflective film, or drip irrigation, these are mostly static management methods lacking dynamic regulation mechanisms based on real-time water status control. This makes it difficult to accurately address stomatal dynamic responses to environmental changes. Furthermore, the quantitative analysis and feedback mechanisms for stomatal dynamic behavior in citrus leaves are still incomplete, hindering the application of precision agriculture technologies in orchard management. Summary of the Invention

[0004] The purpose of this invention is to provide a method for reducing the midday depression ratio of citrus stomata based on water regulation, so as to overcome the shortcomings of the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for reducing the midday depression ratio of citrus stomatal tissues based on water regulation, comprising:

[0006] S100. Collect time series data of stomatal aperture G(t) and leaf water potential parameter Ψ(t) of the target citrus fruit tree leaves, and construct the corresponding stomatal activity response model M1 to reflect the dynamic relationship between stomatal opening and closing state and water state.

[0007] S200, set the target midday period Tm, compare the trend of stomatal opening changes within Tm, and calculate the midday stomatal closure ratio θm;

[0008] S300. Based on the changing trends of soil moisture content θs and transpiration rate E in the target area, a water regulation response model M2 is established to simulate the changes in leaf Ψ(t) under different irrigation levels.

[0009] S400. Input the Ψ(t) output by model M2 into model M1 to predict the impact of different water management schemes on θm, and select the optimal water management scheme θm. opt ;

[0010] S500, according to θ optControl irrigation parameters for the target period, including irrigation time, irrigation volume, and watering frequency;

[0011] S600, repeat θm measurements and model iteration updates to optimize the moisture control scheme and reduce the stomatal closure rate during high-temperature periods.

[0012] Preferably, in S100;

[0013] S101. Obtain real-time stomatal aperture image sequences of target citrus fruit tree leaves under different time periods and light conditions, and extract and construct the stomatal aperture time function G(t);

[0014] S102. For the area where the midrib and main vein of the target blade intersect, measure the local water potential Ψ(t) of the blade in real time, and perform multi-scale noise reduction and fusion processing on the Ψ(t) data;

[0015] S103. Using a neural fuzzy reasoning system, nonlinear mapping modeling is performed on the collected G(t) and Ψ(t) data to establish a stomatal activity response model M1.

[0016] Preferably, in S200;

[0017] S201. Determine the target midday time Tm based on changes in environmental spectral reflectance;

[0018] S202. Obtain the surface temperature Tleaf of citrus leaves during the target time period, and calculate the leaf temperature difference ΔT = Tleaf – Tair in combination with the ambient air temperature Tair. Select the closed-pore response region based on the trend of ΔT change.

[0019] S203. Calculate the midday stomatal closure ratio θm based on the ratio of the total area of ​​the closed area to the total area of ​​the monitored leaf area.

[0020] Preferably, in S300;

[0021] S301. Continuously monitor the soil moisture θs at different depths in the target area, and reconstruct the three-dimensional distribution map of soil moisture in the root zone of the target fruit trees by combining remote sensing vegetation index and micro-meteorological data through multi-dimensional data interpolation.

[0022] S302. Utilizing the dynamic relationship between leaf transpiration rate E and root zone θs distribution, a water flux tracking algorithm based on a spatiotemporal inversion mechanism is established. Under the premise that the irrigation boundary conditions are known, the expected response curve of Ψ(t) is calculated backward to form an input-oriented water potential inversion model.

[0023] S303. Construct a hydrodynamic plant response coupling model M2 that integrates the improved Penman-Monteith formula and the neural network structure. The input variables include meteorological parameters, multi-level θs data, estimated E values ​​and previous measured Ψ(t) values. The output is the predicted leaf Ψ(t) curve.

[0024] Preferably, in S400;

[0025] S401. Construct multiple sets of irrigation parameter combinations, each containing different irrigation start times, irrigation water volumes, and durations. Predict the corresponding Ψ for each set of parameters using model M2. i (t) curve;

[0026] S402, each group Ψ i (t) Input model M1 and simulate the corresponding midday stomatal closure ratio θ. i m, and use the fuzzy C-means clustering algorithm to construct Ψ(t)-θm response mapping clusters to identify a subset of irrigation strategies with similar stomatal response behaviors;

[0027] S403. In the response mapping cluster, based on the objective function J = αθm + βV, where θ m Given the stomatal closure ratio, V as the irrigation water volume, and α and β as weighting coefficients, a non-dominated sorting multi-objective genetic algorithm is used to search for the optimal water management scheme θ. opt .

[0028] Preferably, in S500;

[0029] S501, The optimal irrigation parameter θ opt The control variables, namely irrigation time Ts, irrigation water volume Vs, and irrigation frequency fs, are input to the orchard edge computing node, and the control signals are transmitted to the intelligent solenoid valve actuator group in the set area.

[0030] S502. The water application rate Q(t) is monitored in real time using a flow meter, and a small correction is made for each irrigation to ensure that the actual total water output Vreal is controlled within ±3%.

[0031] Preferably, in S600;

[0032] S601. Obtain midday leaf temperature data of the target citrus tree, combine it with real-time meteorological and soil information, and calculate the actual stomatal closure ratio θmobs in real time through a multimodal fusion inference algorithm.

[0033] S602. Compare θmobs with the predicted value θmpred output by model M1. If the difference exceeds the set threshold δ, trigger the local model drift detection mechanism.

[0034] S603. After the update condition is triggered, the stomatal response model M1 is iteratively corrected, and then the optimal moisture regulation scheme θ is updated. opt .

[0035] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0036] 1. This invention constructs a dual-model system coupling stomatal behavior and leaf water potential in citrus trees. By combining multimodal data acquisition, nonlinear response modeling, and multi-objective optimization strategies, it achieves dynamic prediction and precise control of stomatal closure behavior in citrus trees during midday high-temperature periods. This method forms a complete closed-loop control process from five stages: perception, modeling, prediction, control, and updating. This effectively prolongs the efficient photosynthetic period and improves the carbon assimilation efficiency and yield potential of fruit trees.

[0037] 2. Compared to traditional methods based on experience-based irrigation or static regulation, this invention integrates edge computing, remote sensing monitoring, reinforcement learning, and intelligent optimization algorithms to achieve precise matching between stomatal behavior and physiological drivers and irrigation execution strategies. The technical system possesses strong adaptability, a high level of automation, and a self-evolving model, significantly improving the response efficiency of citrus cultivation to climate change and physiological fluctuations, providing advanced technical support for the development of smart orchards and precision agriculture. Attached Figure Description

[0038] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0039] Figure 1 This is a mind map of the method of the present invention. Detailed Implementation

[0040] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0041] For examples, please refer to Figure 1 As shown in this embodiment, the method for reducing the midday depression ratio of citrus stomatal tissues based on water regulation includes:

[0042] S100. Collect time series data of stomatal aperture G(t) and leaf water potential parameter Ψ(t) of the target citrus fruit tree leaves, and construct the corresponding stomatal activity response model M1 to reflect the dynamic relationship between stomatal opening and closing state and water state.

[0043] S200, set the target midday period Tm, compare the trend of stomatal opening changes within Tm, and calculate the midday stomatal closure ratio θm;

[0044] S300. Based on the changing trends of soil moisture content θs and transpiration rate E in the target area, a water regulation response model M2 is established to simulate the changes in leaf Ψ(t) under different irrigation levels.

[0045] S400. Input the Ψ(t) output by model M2 into model M1 to predict the impact of different water management schemes on θm, and select the optimal water management scheme θopt.

[0046] S500: Control irrigation parameters for the target time period based on θopt, including irrigation time, irrigation volume and watering frequency;

[0047] S600, repeat θm measurements and model iteration updates to optimize the moisture control scheme and reduce the stomatal closure rate during high-temperature periods.

[0048] In this embodiment, to address the need for dynamic monitoring and prediction of midday stomatal closure in citrus trees, a technical process for constructing a stomatal activity response model M1 is designed, mainly including the following three steps:

[0049] Three-year-old 'Ehime 38' citrus trees from an orchard were selected as target samples, and monitoring was conducted during the high-temperature period of the growing season. A combination system of a miniature stomatal meter (such as the SC-1 type) and a visible light microscopic imaging device was used, with sampling points set on the underside of leaves. Four sampling areas were set for each leaf, and the sampling frequency was set to 5Hz. Simultaneously, a deep learning image processing algorithm (such as YOLOv5) was used to segment the image sequence in real time, extracting the pixel area of ​​stomatal opening and constructing a continuous stomatal opening time series G(t). To adapt to image interference under strong light, the imaging system was equipped with a polarization filter module, and a light shield was used to stabilize the image quality.

[0050] A thin-film tension sensor array (sensitivity ≥ 0.01 MPa) is embedded in the corresponding part of the same leaf blade, arranged in a dot matrix pattern on the midrib and its extended mesophyll region. The leaf water potential Ψ(t) is measured in real time, and the data is aggregated through a custom signal acquisition module (STM32 platform). A wavelet threshold denoising algorithm is used to perform multi-scale processing on the Ψ(t) signal to eliminate high-frequency noise caused by physical disturbances such as wind speed and blade tumbling, while retaining its true response signal to the stomatal opening and closing process.

[0051] The synchronously acquired data pairs of G(t) and Ψ(t) are fed into an Adaptive Neural Fuzzy Inference System (ANFIS) for nonlinear mapping modeling. The model input consists of three variables (current value of Ψ, rate of change of Ψ, and time t), and the output variable is the predicted value of G(t). The model is iteratively trained using fuzzy rule initialization and backpropagation parameter tuning mechanisms, with a training set to validation set ratio of 8:2.

[0052] The final model M1 can dynamically predict the stomatal state within the next hour, with a validation error (RMSE) of less than 0.04 (unit normalized), which is significantly better than conventional linear regression models or empirical formula methods (RMSE≥0.12).

[0053] The M1 model constructed in this embodiment can dynamically predict stomatal behavior based on the current water potential, providing a reliable physiological basis for subsequent water regulation strategies. Especially during the midday high-temperature period, it can effectively identify stomatal regions that are about to close, laying the foundation for intelligent decision-making on the precise timing and amount of irrigation, thereby achieving the goal of reducing the midday rest period and extending the photosynthetic efficiency period.

[0054] In this embodiment, to accurately identify the dynamic closure trend of citrus stomata during midday and quantitatively calculate the midday stomatal closure ratio θm, the following technical process is adopted:

[0055] The test site was a citrus orchard during the peak fruiting season in summer. A drone equipped with a multispectral camera was used to conduct three flights between 8:00 am and 3:00 pm on a sunny day to collect spectral reflectance data of the orchard canopy in the 700–900 nm band.

[0056] By combining the collected photosynthetically active radiation (PAR) and temperature data, the slope and inflection point of the canopy reflectivity change are calculated to identify spectral abrupt change segments. The target midday stomatal inhibition peak period Tm is automatically defined with the point of maximum reflectivity change (i.e., the typical light suppression onset point) as the lower limit and the end of the high temperature plateau segment as the upper limit. The range is generally between 11:30 and 14:00.

[0057] Three infrared thermal imagers were set up in the main canopy layer of the target fruit trees to collect images of the citrus leaf surface temperature (Teaf) every 5 minutes within a range of Tm. The air temperature (Tair) in the same area was recorded simultaneously, and the leaf temperature difference ΔT = Tleaf - Tair was calculated.

[0058] Based on the established ΔT threshold model (ΔT≥2.0℃ indicates possible stomatal closure), the closure response region in the image is marked, and the stomatal activity region is spatially identified and labeled in the full canopy image using an image segmentation algorithm (such as U-Net), generating a binarized leaf surface activity layer.

[0059] The area of ​​the aforementioned closed-pore response region is statistically analyzed, and its area ratio within the time period Tm is calculated, which is the pore closure ratio θm. If there are multiple frames of images within Tm (e.g., 18 frames), the instantaneous θm(t) of each frame is calculated separately, and the θm time series is constructed.

[0060] To improve the environmental robustness of θm, the Trend Stability Index (TSI) is introduced, and its calculation method is as follows: Where σ θ Let θm(t) be the standard deviation, and θˉ be its mean. The final value is: θm,final = θˉ·TSI; this method can effectively suppress instantaneous fluctuations caused by short-term wind speed, shading, or equipment errors, thereby improving the reliability of θm determination.

[0061] This embodiment aims to establish a response model M2 for predicting the water potential Ψ(t) of citrus leaves under different irrigation strategies, supporting irrigation decisions that precisely regulate midday stomatal behavior.

[0062] Within the target citrus orchard, three buried TDR soil moisture sensor arrays were deployed at depths (10cm, 30cm, and 60cm) × 4 azimuths, centered on each sample tree, and θ data was collected every 10 minutes. s value.

[0063] Meanwhile, a spatial model of plant evapotranspiration was established using the NDVI sensing system carried by the drone and data from ground weather stations (temperature, wind speed, solar radiation).

[0064] By performing Kriging interpolation and time-weighted regression on TDR monitoring data and NDVI spatial maps, high-precision θ values ​​are obtained. s The (x,y,z,t) four-dimensional water field is used as input for subsequent models.

[0065] The empirical coupling interval between transpiration rate E(t) and leaf Ψ(t) was obtained based on previous experimental data. The transpiration flux for the target time period was estimated using the Penman-Monteith formula. Where E is the transpiration rate per unit leaf area (mm·s) -1 ), which is the rate at which water evaporates through stomata per unit time; Δ is the slope (kPa·℃⁻¹) of saturated vapor pressure relative to temperature, representing the sensitivity of air vapor saturation to temperature; R n Net radiative flux density (W·m) -2 ), representing the total energy absorbed by the leaves; G is the soil heat flux (W·m). -2 The density of air (ρa) can be approximately ignored; ρa is the air density (kg·m³). -3 );c p Specific heat capacity of air (J·kg) -1·℃-1); e s The saturated vapor pressure (kPa); e a The actual water vapor pressure (kPa); r a aerodynamic drag (s·m) -1 This reflects the resistance to water vapor diffusion from the leaf surface to the air; r s Pore ​​resistance (s·m) -1 ), which is the main variable for plants to regulate water loss; γ is the psychological constant (kPa·℃-1), which represents the degree of influence of the wet-dry ratio on the evaporation process.

[0066] Introducing a moisture inversion module, utilizing θ s Given inputs of (x,y,z,t) and E(t), a reverse modeling framework (based on an unsupervised LSTM-AE network) is used to deduce the predicted response curve of Ψ(t) from changes in leaf transpiration.

[0067] Finally, a water potential regulation response model M2 is constructed, adopting the following improved structure:

[0068] Input layer: θs multi-layer moisture content, estimated E value, real-time meteorological variables, historical Ψ value;

[0069] The modeling entity consists of a physical driving term derived from Penman-Monteith connected in parallel with a two-channel neural network (static + dynamic).

[0070] Optimization mechanism: A Bayesian parameter optimizer is introduced, which triggers automatic parameter tuning of the model when the difference between the target Ψ(t) and the measured Ψ(t) exceeds 0.1 MPa.

[0071] The model can provide a predicted curve of leaf Ψ(t) for the next 2 hours within 30 minutes, providing a basis for the design of the next step of precise water replenishment scheme.

[0072] This embodiment aims to predict the stomatal response behavior of citrus leaves from multiple irrigation combinations, and then screen out the irrigation scheme θ that achieves the optimal balance between controlling the stomatal midday depression ratio and water conservation. opt The operation process is as follows:

[0073] Using the area where the target fruit trees are located as the experimental platform, three control variables are defined:

[0074] The irrigation start time (Tstart) ranges from 10:30 to 12:30.

[0075] Irrigation volume (V) ranges from 5 to 20 L per plant;

[0076] Irrigation duration (Tdur) ranges from 15 to 60 minutes.

[0077] An irrigation strategy set {P1,...,P} containing 30 parameter combinations was generated using the Latin hypercube sampling method (LHS). 30}, input them one by one into the water regulation response model M2 to obtain Ψ under each group of irrigation conditions. i (t) Predicted curve.

[0078] Each group Ψ i (t) Input the previously constructed stomatal response model M1 to simulate its corresponding midday stomatal closure ratio θ i m. A set of pore response maps is obtained: {P i →Ψ i (t)→θ i m}, i = 1, 2, ..., 30; then, for all Ψ i (t) and θ i Fuzzy C-means (FCM) clustering analysis was performed on m sample pairs to classify the schemes into three typical clusters (such as "fast response type", "lagging type" and "inefficient response type") based on stomatal response behavior, which facilitates focusing on the regions with good performance in multi-objective optimization.

[0079] Define a bi-objective optimization function: J = αθm + βVJ; where: θm represents the simulated midday stomatal closure ratio; V represents the irrigation water volume; α and β represent the weighting coefficients for controlling water use efficiency and stomatal regulation priority (e.g., α = 0.7, β = 0.3).

[0080] The non-dominated sorting multi-objective genetic algorithm NSGA-II was used for scheme evolution, iterating for 100 generations and retaining the Pareto optimal front in each generation. Finally, the optimal strategy θ that simultaneously has low θm and low V was selected from the solution set. opt , as recommended irrigation parameters.

[0081] In this embodiment, the optimal water regulation strategy θ obtained from the previous stage of optimization is used. opt Precision irrigation control was implemented for citrus trees to reduce the proportion of stomatal closure at midday and maintain stable soil water potential. The operation procedure is as follows:

[0082] θ opt The three corresponding control parameters—irrigation start time (Tstart), single irrigation volume (V), and irrigation duration (Tdur)—are uploaded to the orchard automatic irrigation control terminal, generating an irrigation schedule. For example, an optimization result might be: Tstart = 11:20; V = 12 L / tree; Tdur = 30 minutes; irrigation cycle = repeated every 3 days. These parameters are automatically converted into control commands and transmitted to the field solenoid valves and irrigation system, achieving automated control.

[0083] During irrigation, the flow rate of each plant is monitored in real time using a flow meter, and an irrigation log is generated based on the opening and closing status of the solenoid valve. If the measured water volume deviation exceeds ±5%, an automatic compensation control mechanism is triggered to adjust the flow rate. Simultaneously, after each irrigation, the system automatically records: soil moisture θs; the current measured value of Ψ(t); the change in stomatal closure area (Δθm); and the inputs for the next round of feedback adjustment models M2 and M1, improving the accuracy of future predictions.

[0084] Within a continuous sampling period (e.g., 7 days), if θm shows an abnormal upward trend (e.g., θm > 0.6 for two consecutive days), the system will automatically trigger the "emergency irrigation correction mode" without disrupting θ. opt Under the original strategy framework, Tstart is fine-tuned or Tdur is extended to restore the blade water potential balance.

[0085] In addition, the system has an embedded short-term weather forecast interface to address sudden climate changes (such as heat waves or strong winds), allowing strategies to be executed in advance and further enhancing the robustness of stomatal control.

[0086] In this embodiment, the selected optimal water management scheme θ opt For precise implementation in citrus orchard irrigation systems, the following execution process is proposed:

[0087] A typical citrus orchard (2.5 meters between trees, 4 meters between rows) was selected for the implementation of control strategies. The θ value output by the irrigation model optimization module was used. opt Including the following parameter combinations:

[0088] Irrigation start time Ts: 11:30;

[0089] Irrigation water volume Vs: 12L / plant;

[0090] Irrigation frequency fs: 1 time / day, for 3 consecutive days.

[0091] The above parameters are standardized and encoded before being input into the edge computing gateway node (Raspberry Pi + LoRa module). Control commands are then sent via multicast through the LoRaWAN network to 32 sets of solenoid valves in the orchard, controlling the irrigation start time, watering intensity, and valve holding time within the target area. The system communication latency is less than 200ms, supporting precise multi-point linkage within a range greater than 1000 meters.

[0092] Each valve assembly is equipped with a magnetic induction flow meter (accuracy ±1%), which transmits the applied water flow velocity Q(t) back to the control node in real time via RS485 protocol. Based on the outlet pipe inner diameter and the preset Vs value, the system adopts the following correction strategy: Error = Vreal - Vs; if the error exceeds the set threshold (±3%), the system will automatically adjust Ts or increase the Q(t) control setting in the next irrigation to ensure that the actual water volume approaches the target irrigation curve day by day. This method overcomes the static control problem of "no feedback" in traditional irrigation systems.

[0093] The system activates its prediction module 30 minutes before irrigation, capturing real-time micro-meteorological data (such as wind speed, temperature, and light intensity) and displaying the θm(t) trend chart. If the model determines that a strong increase in light intensity or temperature fluctuations will occur during the target irrigation period, it activates a time-varying scheduling mechanism, advancing Ts to 11:15 or delaying it to 11:45 (automatically graded according to the disturbance level) to maximize synchronization between the irrigation response and the time before stomatal closure. This mechanism, triggered by the introduction of a "stomatal inhibition frontier response window before irrigation," represents a breakthrough from the fixed irrigation period mechanism and effectively improves the synchronization of regulatory behavior with actual physiological needs.

[0094] In this embodiment, to improve the long-term adaptability and accuracy of the water regulation model M1, a dynamic iterative mechanism based on measured feedback and edge reinforcement learning is proposed, and its operation steps are as follows:

[0095] Thermal imaging infrared cameras (such as FLIRA655sc) were deployed in the target citrus orchard. Combined with air temperature, wind speed, and soil moisture information, a spatiotemporal map of leaf temperature was constructed based on thermal image data within a time period Tm. The distribution of the leaf temperature and air temperature difference ΔT in the thermal images was used to determine the fusion process according to the following rules: Where S(ΔT>2.0℃) represents the total area of ​​pixels inferred to be the closed-hole region, S total The total canopy leaf area is θm. By introducing micro-meteorological adjustment factors (such as wind speed compensation terms), a multimodal inference framework is constructed to accurately evaluate the measured value of θm based on multi-source data, which serves as the benchmark for model evaluation.

[0096] The system compares the difference Δθ between θmobs and the θmobs output by model M1 in real time: Δθ=|θmobs-θmpred|; if Δθ is greater than the threshold δ for two consecutive days (e.g. δ=0.1), the local model drift detection mechanism is triggered. The system automatically analyzes the scene features that may cause the model to fail (e.g., sudden weather changes, abnormal soil moisture, etc.) and judges that the model response parameters are no longer suitable for the current period conditions.

[0097] After model failure is determined, an edge computing device (such as Jetson Nano) is scheduled to enter the local update process. A lightweight DQN network structure is built using the current M1 model as the initial weights, and the θmobs, Ψ(t), and meteorological data of the past 5 days are used as the training dataset to construct a reinforcement learning state-behavior space.

[0098] A knowledge transfer strategy is introduced to transfer the response rules of the old model under "steady-state conditions" to the new model, thereby avoiding extreme biases in the new model in the early stages.

[0099] After training, the new model QM1 replaces the old model for subsequent θm prediction, and the updated water management strategy is regenerated based on the output of the new model to achieve closed-loop feedback optimization.

[0100] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method for reducing the midday depression rate of citrus stomata based on water regulation, characterized by: include: S100. Collect time series data of stomatal aperture G(t) and leaf water potential parameter Ψ(t) of the target citrus fruit tree leaves, and construct the corresponding stomatal activity response model M1 to reflect the dynamic relationship between stomatal opening and closing state and water state. S200, set the target midday period Tm, compare the trend of stomatal opening changes within Tm, and calculate the midday stomatal closure ratio θm; S300. Based on the changing trends of soil moisture content θs and transpiration rate E in the target area, a water regulation response model M2 is established to simulate the changes in leaf Ψ(t) under different irrigation levels. S400. Input the Ψ(t) output by model M2 into model M1 to predict the impact of different water management schemes on θm, and select the optimal water management scheme θm. opt ; S500, according to θ opt Control irrigation parameters for the target period, including irrigation time, irrigation volume, and watering frequency; S600, repeat θm measurements and model iteration updates to optimize the moisture control scheme and reduce the stomatal closure rate during high-temperature periods.

2. The method for reducing the stomatal depression ratio of citrus fruits based on water regulation according to claim 1, characterized in that: In S100; S101. Obtain real-time stomatal aperture image sequences of target citrus fruit tree leaves under different time periods and light conditions, and extract and construct the stomatal aperture time function G(t); S102. For the area where the midrib and main vein of the target blade intersect, measure the local water potential Ψ(t) of the blade in real time, and perform multi-scale noise reduction and fusion processing on the Ψ(t) data; S103. Using a neural fuzzy reasoning system, nonlinear mapping modeling is performed on the collected G(t) and Ψ(t) data to establish a stomatal activity response model M1.

3. The method for reducing the midday depression ratio of citrus stomata based on water regulation according to claim 1, characterized in that: In S200; S201. Determine the target midday time Tm based on changes in environmental spectral reflectance; S202. Obtain the surface temperature Tleaf of citrus leaves during the target time period, and calculate the leaf temperature difference ΔT = Tleaf – Tair in combination with the ambient air temperature Tair. Select the closed-pore response region based on the trend of ΔT change. S203. Calculate the midday stomatal closure ratio θm based on the ratio of the total area of ​​the closed area to the total area of ​​the monitored leaf area.

4. The method for reducing the midday depression ratio of citrus stomata based on water regulation according to claim 1, characterized in that: In S300; S301. Continuously monitor the soil moisture θs at different depths in the target area, and reconstruct the three-dimensional distribution map of soil moisture in the root zone of the target fruit trees by combining remote sensing vegetation index and micro-meteorological data through multi-dimensional data interpolation. S302. Utilizing the dynamic relationship between leaf transpiration rate E and root zone θs distribution, a water flux tracking algorithm based on a spatiotemporal inversion mechanism is established. Under the premise that the irrigation boundary conditions are known, the expected response curve of Ψ(t) is calculated backward to form an input-oriented water potential inversion model. S303. Construct a hydrodynamic plant response coupling model M2 that integrates the improved Penman-Monteith formula and the neural network structure. The input variables include meteorological parameters, multi-level θs data, estimated E values ​​and previous measured Ψ(t) values. The output is the predicted leaf Ψ(t) curve.

5. The method for reducing the midday depression ratio of citrus stomata based on water regulation according to claim 1, characterized in that: In S400; S401. Construct multiple sets of irrigation parameter combinations, each containing different irrigation start times, irrigation water volumes, and durations. Predict the corresponding Ψ for each set of parameters using model M2. i (t) curve; S402, each group Ψ i (t) Input model M1 and simulate the corresponding midday stomatal closure ratio θ. i m, and use the fuzzy C-means clustering algorithm to construct Ψ(t)-θm response mapping clusters to identify a subset of irrigation strategies with similar stomatal response behaviors; S403. In the response mapping cluster, based on the objective function J = αθm + βV, where θ m Given the stomatal closure ratio, V as the irrigation water volume, and α and β as weighting coefficients, a non-dominated sorting multi-objective genetic algorithm is used to search for the optimal water management scheme θ. opt .

6. The method for reducing the midday depression ratio of citrus stomata based on water regulation according to claim 1, characterized in that: In S500; S501, The optimal irrigation parameter θ opt The control variables, namely irrigation time Ts, irrigation water volume Vs, and irrigation frequency fs, are input to the orchard edge computing node, and the control signals are transmitted to the intelligent solenoid valve actuator group in the set area. S502. The water application rate Q(t) is monitored in real time using a flow meter, and a small correction is made for each irrigation to ensure that the actual total water output Vreal is controlled within ±3%.

7. The method for reducing the midday depression ratio of citrus stomatal tissues based on water regulation according to claim 1, characterized in that: In S600; S601. Obtain midday leaf temperature data of the target citrus tree, combine it with real-time meteorological and soil information, and calculate the actual stomatal closure ratio θmobs in real time through a multimodal fusion inference algorithm. S602. Compare θmobs with the predicted value θmpred output by model M1. If the difference exceeds the set threshold δ, trigger the local model drift detection mechanism. S603. After the update condition is triggered, the stomatal response model M1 is iteratively corrected, and then the optimal moisture regulation scheme θ is updated. opt .

Images