Environment regulation and control model algorithm and algorithm model platform fusing crop physiology

Abstract

The invention relates to the crossing field of agricultural intelligent algorithms and intelligent equipment technologies, in particular to an environment regulation and control model algorithm and algorithm model platform fusing crop physiology. According to the model algorithm, through four-step calculation of spatio-temporal heterogeneity adaptation, multi-source data dynamic coupling, stress self-adaptive adjustment and self-learning iterative optimization, a partition precise regulation and control decision value is generated; the algorithm model platform comprises a data acquisition layer (a partition sensor and a multi-source interface), a feature processing layer (a three-dimensional feature library and data preprocessing), a model algorithm layer (deploying the model algorithm) and a platform application layer (partition regulation and control and closed loop feedback), and full-link cooperation of algorithm calculation, platform execution, effect feedback and algorithm iteration is achieved. The crop yield is increased by 15-20%, water resources are saved by 25-30%, and the method is suitable for multi-crop facility agriculture scenes.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of intelligent algorithm and intelligent equipment technology in agriculture, in particular to the combination of environmental regulation model algorithm and algorithm model platform based on crop physiology. BACKGROUND

[0002] The core bottleneck of intelligentization of facility agriculture is the disconnection between "algorithm" and "platform". On the one hand, the existing environmental regulation algorithm is mostly "static threshold type", which does not integrate the spatio-temporal differences of crop physiology and multi-source data, resulting in low decision accuracy. On the other hand, the existing regulation platform is only a tool for "data collection + instruction issuing", without deep integration of the self-learning ability of the algorithm, and cannot realize the closed loop of "algorithm optimization - platform iteration". The specific defects are as follows: 1. Core defects of existing regulation algorithm First, the physiological spatio-temporal heterogeneity is not adapted. The existing algorithm assumes that the physiological state of crops in the whole greenhouse is consistent, for example, the same algorithm parameters are used in the center of the greenhouse (stable temperature, sufficient light) and the edge (large temperature fluctuation, weak light), resulting in a 15-20% difference in crop leaf water potential between the edge and the center, and a 10-15% difference in yield. At the same time, the algorithm does not consider the difference in physiological needs during the day and night (such as the high respiration consumption of crops at night, which requires accurate temperature control, while the existing algorithm still calculates according to the parameters during the day), resulting in a 30-40% waste of electric energy at night.

[0003] Second, the multi-source data integration is insufficient. The existing algorithm only relies on internal sensor data in the greenhouse, without integrating external weather forecasts (such as strong light and rain in the next 24 hours) and macro monitoring data (such as unmanned aerial vehicle canopy temperature): for example, the weather forecast predicts a heavy rain the next day, and the algorithm still controls the humidity according to the normal threshold, resulting in high humidity stress in the greenhouse the next day, and a 25-30% increase in the incidence of crop diseases; the unmanned aerial vehicle detects a high temperature in the local canopy (indicating excessive transpiration), but the algorithm cannot associate the soil moisture data in that area and still controls irrigation according to the whole area, resulting in a 20-25% waste of water resources.

[0004] Third, there is no self-learning and iterative ability. The weight coefficients (such as the priority of CO2 and temperature) of the existing algorithm are mostly manually preset and fixed, and cannot be optimized according to the actual planting data: for example, the soil fertility of a certain tomato greenhouse decreases, and the influence of SPAD value on yield increases from 0.2 to 0.3, but the algorithm still calculates according to the weight of 0.2, resulting in insufficient nutrient regulation and a 8-10% decrease in soluble sugar content. In addition, the physiological characteristics of different batches of crops (such as the improvement of varieties between years) cannot be adapted by the algorithm, and the code needs to be manually rewritten, which is high in operation cost.

[0005] 2. Core defects of existing regulation platform First, the algorithm is decoupled from the platform. Existing platforms only call algorithms as "black boxes" without providing interfaces for adjusting algorithm parameters or setting up a linkage mechanism between "algorithm optimization" and "platform execution": for example, an algorithm optimizes the CO2 weight through self-learning, but the platform still issues instructions based on the old weight, resulting in the inability of the algorithm optimization to take effect. At the same time, the data preprocessing capability of the platform is weak (such as only being able to remove obvious outliers and unable to complete spatio-temporal missing data), resulting in poor quality of algorithm input data and a decision value error of 20-25%.

[0006] Second, the partition regulation and coordination are missing. Existing platforms mostly use the "unified regulation of the whole domain" mode, for example, when the CO2 concentration in a partition is insufficient, the CO2 generator in the whole greenhouse is started, resulting in an over-standard CO2 concentration in the central area (waste of 30-35% CO2); at the same time, the platform has no cross-partition coordination mechanism, and when multiple partitions trigger regulation at the same time (such as multiple partitions needing to be cooled during the high-temperature period in summer), the actuator is started at the same time, resulting in a 3-5 times increase in power grid load, and there is a risk of tripping.

[0007] Third, the closed-loop feedback is incomplete. The effect feedback of the existing platform is only based on whether the environmental indicators meet the standard (such as whether the temperature returns to 25-28℃), without combining whether the physiological indicators are improved (such as whether the leaf water potential returns to -0.6~-0.4MPa), resulting in invalid regulation of "environmental standard but physiological damage": for example, a platform detects that the soil humidity returns to 20-25% (environmental standard), but the crop leaf water potential is still -0.9MPa (physiological damage), and the platform stops regulation, ultimately resulting in a 8-10% reduction in crop yield.

[0008] 3. Overall consequences of algorithm decoupling from platform The separation of "algorithm" and "platform" in existing technologies leads to the dilemma of "precise algorithm but unable to land, platform landing but low precision" in facility agriculture: on the one hand, advanced regulation algorithms developed by universities (such as photosynthetic coupling algorithms based on machine learning) cannot be popularized in actual greenhouses due to the lack of suitable platform carriers; on the other hand, the regulation platforms produced by enterprises cannot meet the needs of precision planting due to the simplicity of the integrated algorithms. In addition, the decoupling of the two also leads to complex user operations; users need to manage both "algorithm parameter software" and "platform control software", and need to manually synchronize data, which has human error (such as data entry errors leading to algorithm decision deviation), and does not meet the needs of modern agriculture for "simplicity and integration".

[0009] Therefore, it is urgent to develop a solution of "deep integration of algorithm and platform": to provide a regulation model algorithm that considers physiological spatio-temporal heterogeneity and multi-source data, and to provide an algorithm model platform that can deeply integrate the algorithm, realize partition coordination and closed-loop feedback, and solve the core problems of "inaccurate algorithm, unintelligent platform, and decoupling of the two". SUMMARY

[0010] In view of the deficiencies of the prior art, the core purposes of the present application are: (1) to provide an environmental regulation model algorithm integrated with crop physiology, which improves the regulation decision accuracy through spatio-temporal heterogeneity adaptation, multi-source data coupling, stress self-adaptation and self-learning iteration; (2) to provide an environmental regulation algorithm model platform integrated with crop physiology, which is a carrier of the above algorithm, realizes the full-link closed loop of "data collection-algorithm calculation-instruction execution-effect feedback-algorithm optimization", and solves the problem of disconnection between algorithm and platform; and finally realizes the increase of crop yield by 15-20% and the reduction of resource consumption by 25-30%, and promotes the development of facility agriculture towards the intelligent direction of "algorithm driving and platform iteration".

[0011] The technical scheme adopted by the present application to solve its technical problems is: an environmental regulation model algorithm integrated with crop physiology and an algorithm model platform, comprising the following steps: S1. Crop physiological spatio-temporal heterogeneity adaptation: obtain greenhouse subarea physiological data (relative chlorophyll content SPAD value, leaf water potential, stomatal conductance) and subarea environmental data (air temperature, humidity, CO2 concentration, light intensity), assign subarea basic weight Wi0 according to the distance between the subarea and the crop growth center (Wi0=1.0 when the distance is ≤2m, Wi0=0.9 when 2m<distance≤4m, and Wi0=0.8 when the distance>4m), and correct Wi0 according to the period (additional +0.05 to Wi0 of light / CO2 during the day photosynthesis vigorous period 9:00-15:00, and additional +0.05 to Wi0 of temperature during the night respiration period 22:00-6:00); S2. Multi-source data dynamic coupling: access future 24-hour weather forecast data (light duration, rainfall probability) and unmanned aerial vehicle canopy temperature data, calculate multi-source data correction weight Wi2 (humidity Wi2=1.2 when the rainfall probability is ≥60%, and temperature Wi2=1.2 when the canopy temperature is ≥30℃); combine crop growth period weight Wi1 (CO2 Wi1=0.3 and leaf water potential Wi1=0.25 during the flowering period, and CO2 Wi1=0.3 and temperature Wi1=0.2 during the fruiting period), calculate the comprehensive weight Wi=Wi0×Wi1×Wi2; standardize the physiological data and environmental data into deviation coefficients Pi (physiology) and Ei (environment), and generate subarea regulation decision value (E weather ×E canopy ) through the formula D=Σ(Wi×Pi×Ei×E weather is the meteorological deviation coefficient, and E canopy is the canopy temperature deviation coefficient), and trigger regulation when D<0.6; S3. Stress self-adaptive adjustment: preset subarea physiological stress threshold (leaf water potential ≤-0.8 MPa as drought stress, SPAD ≤40 as nutrient stress), combined with dynamic correction of the threshold (when the rainfall probability ≥60%, the high-humidity stress threshold is adjusted from ≤100 mmol / m 2 s to ≤80 mmol / m 2 s); when the threshold is triggered, the associated index Wi is increased to 1.2-1.5 times of the original weight, and a targeted regulation strategy is generated; S4. Self-learning iterative optimization: at 2:00 am every day, the "regulation instruction-physiological response" data pairs of the previous 7 days are extracted, and the gradient descent algorithm is used to optimize Wi0 and Wi1, and the optimization goal is that the average value of ΔPi ((Pi after regulation-Pi before regulation) / Pi before regulation) after regulation is ≥0.15; if a certain index has ΔPi <0.1 for 3 consecutive days, the index Wi1 is increased by 0.03-0.05 until ΔPi ≥0.1.

[0012] Specifically, the standardization processing method in step S2 is: if the measured value < the lower limit of the optimal range of the index, the deviation coefficient = 0; if the measured value > the upper limit, the deviation coefficient = 1; if it is within the range, the deviation coefficient = (measured value-lower limit) / (upper limit-lower limit); the growth period weight Wi1 satisfies that the sum of all indexes Wi1 in the same subarea is 1.

[0013] Specifically, the targeted regulation strategy in step S3 includes: starting subarea drip irrigation (frequency 2 hours / time, duration 30 minutes) + foliar spraying when drought stress occurs; when nutrient stress occurs, pushing a fertilization reminder to the user end, and synchronously increasing the Wi of the SPAD value associated environmental index (such as soil nutrients).

[0014] Specifically, the data pair screening in step S4 needs to exclude abnormal data (sensor failure, power failure period data), and the sample size ≥100 groups; the optimized Wi0 and Wi1 need to be verified in 1 test subarea for 3 days, and when ΔPi is increased by ≥10%, it is promoted to the whole greenhouse.

[0015] A kind of implementation described model algorithm's environmental regulation algorithm model platform of the integration of crop physiology, including data acquisition layer, feature processing layer, model algorithm layer and platform application layer, each layer cooperates and realizes algorithm deployment and closed loop regulation: The data acquisition layer includes subarea physiological sensor group, subarea environmental sensor group and multi-source auxiliary data interface; the subarea physiological sensor group is arranged according to every 50 m 2 1 subarea, including online SPAD sensor (accuracy ±1.0 SPAD unit), WP4-T leaf water potential meter (range-4.0~0 MPa), AP4 stomatal conductance meter (range 0~1000 mmol / m 2 The collection frequency is 1 time per 30 minutes in the seedling stage and 1 time per 15 minutes in the flowering and fruiting stage. The multi-source auxiliary data interface accesses meteorological data (update frequency 2 hours / time) through the HTTP protocol and receives unmanned aerial vehicle canopy temperature data (update frequency 6 hours / time) through the 4G module. The feature processing layer constructs a three-dimensional feature library of “crop type-growth period-zone” (stores the optimal range of physiological / environmental indicators of each zone, and the deviation of the same crop in the same growth period in different zones is ≤±5%), and performs outlier rejection (3σ criterion), spatial and temporal interpolation completion (Kriging interpolation, error ≤±3%) and standardization processing on the collected data, and outputs Pi, Ei, E weather , E canopy ; The model algorithm layer deploys environmental regulation model algorithms, executes steps S1-S4, and generates zone regulation decision values D and regulation instructions; the platform application layer includes a multi-source data visualization module, a zone regulation instruction issuing module, an effect feedback module, and a self-learning parameter management module; the instruction issuing module sends instructions to the zone execution mechanism (2 light supplement lamps, 1 drip irrigation electromagnetic valve, and 1 CO2 branch valve per zone) through the MQTT protocol (QoS level 2); the effect feedback module collects physiological data in the zone 1 hour after regulation, and calculates ΔPi; the self-learning parameter management module stores Wi0 / Wi1 optimization records, and supports parameter rollback and locking.

[0016] Specifically, the zone environment sensor group of the data collection layer includes an SHT31 temperature and humidity sensor (accuracy ±0.3℃ / ±2%RH), an SCD30 CO2 sensor (accuracy ±(30ppm+3%)), a TSL2591 light sensor (range 0.1~188000lux), and a TDR-350 soil moisture sensor (buried depth 20cm, accuracy ±1%vol).

[0017] Specifically, in the three-dimensional feature library of the feature processing layer, the optimal range of CO2 in the central zone of tomato in the flowering period is 800-1000ppm, and the optimal range of CO2 in the edge zone is 780-980ppm; the optimal range of temperature in the central zone of cucumber in the fruiting period is 23-26℃, and the optimal range of temperature in the edge zone is 24-27℃.

[0018] Specifically, the zone regulation instruction issuing module of the platform application layer supports cross-zone collaboration: when ≥3 adjacent zones trigger the same stress, a “master-slave” collaboration mode is started (the central zone is the master zone, and the instructions are executed first, and the slave zones are delayed for 10 minutes to execute), which avoids high power grid load.

[0019] Specifically, the secondary regulation logic of the effect feedback module is: if a certain partition ΔPi≤0.1, the associated index Wi of the partition is increased by 0.08-0.1, and D is recalculated; if the secondary regulation is invalid for three times in succession, a "partition sensor failure" warning is pushed to the user end (PC+mobile phone APP).

[0020] Specifically, the multi-source data visualization module supports "partition map-index curve-time axis" linkage, can display the correlation curve of single partition SPAD value and canopy temperature, a contrast chart of CO2 concentration of multiple partitions at the same time, and supports exporting a "regulation effect report" (including ΔPi, resource consumption, and yield contribution) according to the growth period.

[0021] The beneficial effects of the present application are: Algorithm accuracy improvement: spatiotemporal heterogeneity adaptation reduces the difference between partition physiological indexes from 15-20% to 3-5%, multi-source data fusion makes the regulation decision 6-12 hours in advance (such as preventing and controlling high humidity stress in advance), self-learning iteration increases ΔPi from 0.1 to 0.15-0.2, and the soluble sugar content of crops is increased by 12-15%.

[0022] Platform efficiency improvement: partition regulation saves 25-30% of water resources (avoiding global irrigation), cross-partition coordination reduces power grid load fluctuation from 3-5 times to 1.5-2 times, full automation closed loop reduces the frequency of manual intervention from 3 times a day to 1 time a week, and the operation cost is reduced by 60-70%.

[0023] Strong adaptability: the three-dimensional feature library supports the expansion of multiple crops (new crops only need to supplement the feature library, without reconstructing the algorithm), the parameter management module supports user-defined weights, and is suitable for facility agriculture in different climate zones such as North China and South China.

[0024] High reliability: multi-source data cross-validation (such as sensor data+unmanned aerial vehicle data) reduces data error to within ±3%, secondary regulation and fault warning mechanism increase the regulation efficiency from 70% to more than 95%, and the flower abscission rate of crops is reduced from 18% to 2-3%. BRIEF DESCRIPTION OF DRAWINGS

[0025] The present application will be further described below in conjunction with the drawings and examples.

[0026] Fig. 1 The step flow chart of the environmental regulation model algorithm integrating crop physiology provided by the present application; Fig. 2 The architecture diagram of the environmental regulation algorithm model platform integrating crop physiology provided by the present application. DETAILED DESCRIPTION

[0027] In order to make the technical means, creative features, purposes and effects of the present application easy to understand, the present application is further described below in combination with specific embodiments.

[0028] As shown in Figs. 1-2 The environmental regulation model algorithm combined with crop physiology and the algorithm model platform of the present application are divided into: core algorithm (step calculation method) and algorithm platform (hardware + software carrier), and the two achieve deep cooperation through "algorithm deployment on platform, platform feedback supporting algorithm optimization".

[0029] (1) Environmental regulation model algorithm combined with crop physiology This algorithm is the "brain" of the platform, which generates precise regulation decisions through four-step core calculation. The specific steps and technical details are as follows: Step S1: Adaptation of crop physiological spatio-temporal heterogeneity (solve the problem of global parameter unification) Input data: (1) Subzone physiological data: collected by online sensors, including SPAD value (resolution 0.1, response time <2s), leaf water potential (resolution 0.001 MPa, measurement range -4.0~0 MPa), stomatal conductance (resolution 1 mmol / m 2 s, range 0~1000 mmol / m 2 s); (2) Subzone division basis: according to greenhouse spatial heterogeneity, divide each 50m 2 into 1 subzone (less than 50m 2 into 1 subzone), and mark the distance of each subzone from the "crop growth center" (usually the geometric center of the greenhouse, with the most uniform light and the most stable temperature); Weight distribution logic: Subzone basic weight Wi0: according to the distance setting, ensure that the edge subzone weight adapts to its physiological needs; Distance ≤2m (central subzone): Wi0=1.0 (optimal physiological state, weight does not need to be adjusted); 2m<distance≤4m (transition subzone): Wi0=0.9 (moderate physiological state, weight fine-tuning); Distance >4m (edge subzone): Wi0=0.8 (poor physiological state, weight reduction of non-critical indicators, priority to critical indicators); Time correction: adjust Wi0according to the diurnal physiological characteristics of crops; Day (6:00-18:00, photosynthetic peak period): light, CO2are critical indicators, Wi0additional +0.05 (such as central subzone light Wi0from 1.0 to 1.05); Night (18:00-6:00, breathing period): temperature is the key indicator, Wi0 extra +0.05 (if the central zone temperature Wi0 changes from 1.0 to 1.05); Output result: the modified basic weight Wi0 of each zone (e.g. the edge zone night temperature Wi0=0.8+0.05=0.85).

[0030] Step S2: Dynamic coupling of multi-source data (solve the problem of single input algorithm) Input data: (1) Basic data: Wi0 output in step S1, physiological / environmental data of the zone; (2) Multi-source auxiliary data: 24-hour weather forecast (from the national meteorological data platform API, including light intensity, rainfall probability, accuracy ±500 lux / ±5%), unmanned aerial vehicle canopy temperature data (multi-rotor unmanned aerial vehicle DJIMavic3T collection, resolution 640x512, temperature accuracy ±0.5℃); Core calculation process: Growth period weight Wi1 setting: according to the priority of crop growth period demand, ensure that the algorithm fits the physiological purpose of different stages

[0031] Multi-source data correction weight Wi2: adjust according to external data to realize forward-looking decision; Weather data correction: rainfall probability ≥60% (high humidity risk) → humidity Wi2=1.2; light forecast >30000 lux (strong light risk) → light Wi2=0.8; Remote sensing data correction: canopy temperature ≥30℃ (high temperature risk) → temperature Wi2=1.2; canopy temperature <20℃ (low temperature risk) → temperature Wi2=1.1; Comprehensive weight Wi calculation: Wi=Wi0×Wi1×Wi2 (the product of the three ensures the coordination of time and space, growth period, and multi-source data), and the sum of all indicators Wi in the same zone =1; Example: tomato flowering period edge zone (Wi0=0.8), rainfall probability 70% (humidity Wi2=1.2), then humidity Wi=0.8×0.2×1.2=0.192; Bias coefficient standardization: convert all input data to coefficients in the range of 0-1 to ensure uniformity of algorithm calculation dimension; Physiological bias coefficient Pi: if the measured value of SPAD=48 (optimal range 50-55) → Pi=(48-50) / (55-50)=-0.4 → standardized to 0; if SPAD=52 → Pi=(52-50) / (55-50)=0.4; Environmental bias coefficient Ei: If CO2 measured value = 900 ppm (optimal range 800-1000) → Ei = (900-800) / (1000-800) = 0.5; Multi-source data bias coefficient: E weather (Meteorology), E canopy (Canopy) Calculation method same as Pi / Ei; Control decision value D calculation: D = Σ(Wi × Pi × Ei × E weather × E canopy ), The smaller the D value, the greater the deviation from the optimal state, and the need to start control (D < 0.6 is the control threshold); Example: Tomato flowering period edge partition Wi (CO2 = 0.8 × 0.3 × 1.0 = 0.24), Pi (leaf water potential = 0), Ei (CO2 = 0) → D = 0.24 × 0 × 0 × 1.0 × 1.0 +... ≈ 0.05 < 0.6 → Trigger CO2 control; Output results: Each partition control decision value D and priority control index (sorted by Wi from large to small).

[0032] Step S3: Stress self-adaptive adjustment (solve algorithm response lag problem) Input data: D value output by step S2, partition physiological data, multi-source auxiliary data; Core logic: Basic stress threshold preset: Set physiological trigger conditions for different stress types (based on crop physiology research conclusions);

[0033] Threshold dynamic correction: Adjust the threshold in combination with multi-source data to achieve early prevention and control; If the weather forecast rainfall probability ≥ 60% (future high humidity): High humidity stress threshold from "stomatal conductance ≤ 100" to "≤ 80", start ventilation in advance; If the unmanned aerial vehicle canopy temperature ≥ 30℃ (current high temperature): Drought stress threshold from "leaf water potential ≤ -0.8" to "≤ -0.7", start irrigation in advance; Stress response strategy generation: After triggering the threshold, adjust the weight and execute the action; Mild stress (physiological indicators are in the "threshold-optimal range lower limit"): Only adjust the associated environmental indicators (such as mild drought → drip irrigation frequency from 4 hours / time to 2 hours / time); Severe stress (physiological indicators ≤ threshold): Adjust environmental indicators and nutrient supply at the same time (such as severe drought → drip irrigation + foliar spraying, simultaneous suspension of ventilation); Output result: stress type, revised threshold, targeted regulation strategy (e.g., "edge partition severe drought → start drip irrigation for 30 minutes + leaf spraying for 10 minutes").

[0034] Step S4: Self-learning iterative optimization (to solve the problem of algorithm staticization) Input data: "regulation instruction-physiological response" data pairs for the previous 7 days (each set of data contains Pi before regulation, Pi after regulation, Wi, and D value), excluding abnormal data (sensor failure, power outage period); Core optimization process: Optimization cycle and target: optimization is performed at 2:00 am every day (non-regulation peak, to avoid affecting crop growth), with the goal of "average ΔPi after regulation ≥ 0.15" (ΔPi = (Pi after regulation - Pi before regulation) / Pi before regulation, the larger the value, the more obvious the physiological improvement); Optimization algorithm selection: gradient descent algorithm (learning rate 0.01, iteration number 100 times) is used, with "maximum average ΔPi" as the objective function, to optimize Wi0 (partition basic weight) and Wi1 (growth period weight); If a certain indicator has ΔPi < 0.1 for 3 consecutive days (poor regulation effect): increase Wi1 of that indicator by 0.03-0.05 (e.g., CO2 Wi1 from 0.3 to 0.33); If a certain indicator has ΔPi > 0.2 for 3 consecutive days (over-regulation): decrease Wi1 of that indicator by 0.02-0.04 (e.g., temperature Wi1 from 0.2 to 0.18); Parameter verification and promotion: the optimized Wi0 / Wi1 is first verified for 3 days in a test partition (selecting a central partition with stable physiological state); If the test partition ΔPi improves by ≥ 10% (e.g., from 0.12 to 0.132): promote to the entire greenhouse; If the improvement is < 10%: adjust the learning rate again and optimize again; Parameter locking and rollback: users can lock key parameters (e.g., result period CO2 Wi1, to avoid abnormal parameters caused by optimization) through the platform, and can also roll back to the historical optimal version (e.g., roll back to the weight combination with the highest ΔPi); Output result: optimized Wi0, Wi1, test partition verification report, and parameter promotion suggestion.

[0035] (II) Integrated crop physiology environmental regulation algorithm model platform This platform is the "body" of the algorithm, which realizes the deployment, execution, and feedback of the algorithm through a four-layer architecture. The specific structure and technical details are as follows: 1. Data acquisition layer (algorithm's "sensory organ") Function positioning: Collect all input data required by the algorithm (partition physiological / environmental data, multi-source auxiliary data), ensure data real-time and accuracy; Hardware composition and technical parameters: (1) Partition physiological sensor set: 1 set per partition, installed at the functional leaf height of crops (tomato 1.2 m, cucumber 0.8 m); Online SPAD sensor: Model SRS-SPAD, measurement range 0-99.9, accuracy ±1.0 SPAD unit, power supply 12VDC; WP4-T leaf water potential meter: measurement range -4.0~0MPa, resolution 0.001MPa, data output mode RS485; AP4 stomatal conductance meter: measurement range 0~1000mmol / m 2 s, accuracy ±5%, response time <1 second; (2) Partition environmental sensor set: 1 set per partition, installed at a position avoiding crop shading; SHT31 temperature and humidity sensor: measurement range -40~125℃ / 0~100%RH, accuracy ±0.3℃ / ±2%RH, I2C communication interface; SCD30 CO2 sensor: measurement range 0-40000ppm, accuracy ±(30ppm+3%), supports automatic calibration; TSL2591 light sensor: measurement range 0.1~188000lux, accuracy ±10%, spectral response 400-700nm; TDR-350 soil moisture content sensor: buried depth 20cm (main root distribution layer), measurement range 0-50%vol, accuracy ±1%vol; (3) Multi-source auxiliary data interface: Weather data interface: through HTTP protocol to dock the national weather data platform API, need to complete real-name authentication to obtain API key, data update frequency 2 hours / time, format JSON; UAV data interface: through 4G module (model SIM7600) to receive the crown layer temperature data transmitted by UAV, UAV flight parameters are "height 5m, route Z-shaped, speed 2m / s", each flight 1 time at 9:00, 15:00; (4) Data transmission method: Sensor data: LoRa gateway (transmission distance ≤1km, rate 125kbps, power supply 220VAC) is used to realize low-power transmission, 4G module is used in remote areas; Multi-source data: weather data is transmitted through Ethernet, UAV data is transmitted through 4G, both are real-time pushed to the edge computing nodes of the feature processing layer; (5) Collection frequency adaptation: dynamically adjust according to growth stages and time periods to ensure key data density;

[0036] 2. Feature processing layer (algorithm's "data preprocessing center") Function positioning: clean, complete, and standardize the data from the collection layer to build a "crop type-growth stage-zone" three-dimensional feature library required by the algorithm, providing high-quality input for the algorithm; Core modules and technical details: (1) Three-dimensional feature library construction: Construction logic: according to the three-dimensional dimensions of "crop type (tomato / cucumber / strawberry)-growth stage (seedling stage / flowering stage / fruiting stage)-zone type (center / transition / edge)", store the optimal range of physiological / environmental indicators under each dimension (based on crop variety manual and field test data); Example (tomato):

[0037] Feature library update: support users to upload new crop variety test data through the platform application layer to expand the feature library (such as adding the optimal range of "strawberry-fruiting stage-center zone"); (2) Data preprocessing process: Outlier rejection: use 3σ criterion (based on the calculation of mean μ and standard deviation σ of the data in the past 24 hours for the zone, reject data outside μ±3σ), for example, if the SPAD value of a certain zone suddenly drops to 30 (μ=47, σ=3, 3σ range 38-56), it is determined as an outlier and is rejected; Temporal and spatial interpolation completion: Spatial completion: use Kriging interpolation method (the variogram model is Gaussian model), complete single-point sensor failure data in the zone (such as one soil sensor offline in a certain zone, based on the interpolation of data in the surrounding 3 zones, error ≤±3%); Time completion: use linear interpolation method to complete short-term missing data (≤2 hours), and more than 2 hours will trigger sensor failure alarm; Standardization: convert all data to deviation coefficients (Pi, Ei, E weather , E canopy ) in the range of 0-1, calculation method same as algorithm step S2; (3) Data output: preprocessed deviation coefficients (Pi, Ei, E weather , E canopy ), optimal range of current zone of three-dimensional feature library, real-time push to model algorithm layer.

[0038] 3. Model algorithm layer (algorithm's "deployment and calculation center") Function orientation: Deploy the aforementioned "environmental regulation model algorithm integrated with crop physiology", execute steps S1-S4, and generate partition regulation instructions; Hardware and software configuration: (1) Hardware: Edge computing node (model NVIDIA Jetson AGX Xavier, CPU 8 cores, GPU 512 cores, supporting parallel computing), memory 32GB (ensuring algorithm iteration speed), storage 1TB SSD (storing historical data and algorithm parameters); (2) Software: The algorithm is written in Python 3.9, based on the TensorFlow framework to realize self-learning iteration (gradient descent algorithm), and is deployed through Docker containerization (ensuring cross-platform compatibility); (3) Linkage logic of algorithm and platform: Data input: Real-time receive of bias coefficient and optimal range data from feature processing layer; Calculation execution: Automatically execute calculation according to steps S1-S4, update D value every 5 minutes; Instruction output: Push the regulation instructions (such as "edge partition CO2 insufficient → start CO2 generator, gas production 4m³ / h, duration 20 minutes") to the platform application layer; Parameter feedback: Push the self-learning optimized Wi0 / Wi1 to the feature processing layer and platform application layer to realize full-platform parameter synchronization.

[0039] 4. Platform application layer (algorithm "execution and feedback center") Function orientation: Issue instructions generated by the algorithm to the execution mechanism, collect regulation effects and feedback to the algorithm, and provide visualization and interaction functions for users; Core modules and technical details: (1) Partition regulation instruction issuing module: Execution mechanism deployment: Configure independent execution mechanism for each partition to ensure accurate partition regulation; Supplementary light: 2 per partition, 100W per, spectrum 660nm (red light, promoting photosynthesis) + 450nm (blue light, regulating morphology), supporting PWM dimming (power 0-100% adjustable); Drip irrigation system: 1 electromagnetic valve per partition, flow rate 2L / h, supporting timing / quantitative control; CO2 generator: 1 branch pipeline per partition (diameter 20mm), total generator is gas type (gas production 5m³ / h), CO2 supply of each partition is controlled by valve; Ventilator / heating pipe: 1 ventilator (power 50W) and 1 heating pipe (power 2kW) per partition to avoid direct irradiation of crops; Instruction transmission protocol: MQTT protocol (QoS level 2, ensure that the instruction is delivered once, no loss, no repetition), communicate with the actuator controller (model STM32F103) through the LoRa gateway; Cross-zone coordination: When ≥3 adjacent zones trigger the same regulation (such as cooling in summer high temperature multi-zone), start the "master-slave" coordination mode; Master zone: Select the central zone (low load of actuator), execute the instruction first; Slave zone: Delay 10 minutes to execute, avoid simultaneous start causing power grid load surge (such as from 20kW to 60kW); (2) Effect feedback module: Feedback timing: 1 hour after the execution of the regulation instruction, collect the physiological data (SPAD value, leaf water potential, stomatal conductance) of the zone; Effect determination: Calculate ΔPi=(Pi after regulation-Pi before regulation) / Pi before regulation, ΔPi>0.1 is determined as "regulation effective", ≤0.1 is determined as "invalid"; Secondary regulation: When invalid, automatically increase the associated index Wi0.08-0.1 of the zone (such as CO2 Wi from 0.24 to 0.32), recalculate the D value and issue the instruction; If it is invalid for 3 times in a row, push the "zone sensor failure" warning (PC pop-up window + mobile APP push); (3) Multi-source data visualization module: Visualization function: Develop the front-end interface based on Vue.js, support "zone map-index curve-time axis" linkage; Zone map: Real-time display of physiological / environmental indicators of each zone (color annotation: green=normal, yellow=mild deviation, red=stress); Index curve: Show the correlation curve of a single zone's certain index (such as SPAD value) and canopy temperature (such as SPAD value decreases with the increase of canopy temperature); Comparison chart: Columnar comparison chart of CO2 concentration, humidity and other indicators of multiple zones at the same time; Report export: Support exporting "regulation effect report" by growth period, including ΔPi, resource consumption (water, electricity, CO2), yield contribution degree (based on physiological indicators), format: Excel / PDF; (4) Self-learning parameter management module: Parameter record: Store the Wi0 / Wi1 optimization records of the past 30 days (including pre-optimization value, post-optimization value, ΔPi change rate), support date query; Parameter operation: Support users to "manually lock" key parameters (such as the result period CO2 Wi1, avoid self-learning to cause parameter abnormalities), "one key rollback" to the historical optimal version (such as rollback to the weight combination with the highest ΔPi). Optimization suggestion: Based on historical data, push parameter optimization suggestions to users (such as "Tomato flowering period edge partition humidity Wi1 is recommended to be increased to 0.22").

[0040] Embodiment 1: Application of North China winter-spring tomato in sunlight greenhouse (algorithm + platform collaborative verification) This embodiment takes a sunlight greenhouse (span 10 m, length 50 m, total area 500 m 2 ) as the scene, the crop is winter-spring tomato (variety "Zhuzi 105"), the implementation period is from March 1, 2024 to May 29, 2024 (whole growth period 90 days), and the effect difference between the "algorithm + platform" of the application (test group) and the traditional "static threshold + global platform" (control group) is compared.

[0041] I. Algorithm and platform deployment Algorithm parameter initialization: Three-dimensional feature library: import the optimal range of "tomato-winter-spring-10 partitions" (10 partitions: 1-3 for the center, 4-7 for the transition, and 8-10 for the edge); Initial weight: Wi0 (center 1.0, transition 0.9, edge 0.8), Wi1 (set according to Table 1), Wi2 (initially 1.0, to be corrected by multi-source data); Stress threshold: the basic threshold is set according to Table 2, and the correction rule is enabled.

[0042] Platform hardware deployment: Data acquisition layer: 10 partitions each deploy 1 set of physiological / environmental sensors, unmanned aerial vehicle DJI Mavic 3T flies every day at 9:00 and 15:00, and the meteorological interface is connected to the meteorological data platform; Feature processing layer: edge computing node NVIDIA Jetson AGX Xavier, deploy three-dimensional feature library and data preprocessing software; Model algorithm layer: Docker containerized deployment of the algorithm of the application, iteration period every day at 2:00 am; Application layer: 10 partitions each deploy 2 light supplement lamps, 1 drip irrigation electromagnetic valve, and 1 CO2 branch pipe, and the user end (PC + mobile phone APP) installs visual software.

[0043] II. Algorithm and platform collaborative regulation process 1. Seedling stage (3.1-3.30); verification of spatio-temporal heterogeneity adaptation Key event (3.1010:00): (1) Data acquisition layer: the temperature of the edge zone 8 is 22°C (Ei=0, lower than the optimal range 23-26°C of the zone), and the temperature of the center zone 1 is 24°C (Ei=0.33); (2) Algorithm step S1: the temperature of the edge zone 8 is Wi0=0.8 (edge) +0.05 (daytime)=0.85; (3) Algorithm step S2: Wi1 (seedling stage temperature Wi1=0.25), Wi2=1.0 (no abnormal weather), and the comprehensive Wi=0.85*0.25*1.0=0.2125; D=0.2125*0*0.33≈0.04<0.6→trigger control; (4) Platform application layer: only the heating pipe (2kW) of the zone 8 is started, and the temperature rises to 24°C (Ei=0.33) after 20 minutes, and D=0.2125*0.33*0.33≈0.61→stop heating; Seedling stage effect: the SPAD value of the edge zone 8 is 46.5 on average (43.2 of the control group), and the difference with the center zone 1 is reduced from 4.8 to 1.2, and the power consumption is saved by 42% compared with the control group (global heating).

[0044] 2. Flowering stage (3.31-4.30); multi-source fusion and self-learning verification Key event (4.188:00): (1) Data acquisition layer: meteorological forecast 24-hour rainfall probability 70%, unmanned aerial vehicle detects canopy temperature of zone 5 29°C; (2) Algorithm step S3: high humidity stress threshold is adjusted from “stomatal conductance≤100” to “≤80”, and drought stress threshold is adjusted from “leaf water potential≤-0.8” to “≤-0.7”; (3) Algorithm step S2: humidity of zone 5 Wi=0.9 (transition zone)*0.2 (flowering stage Wi1)*1.2 (rainfall correction)=0.216, temperature Wi=0.9*0.05*1.2 (high temperature correction)=0.054; D=0.216*0*1+0.054*0*1≈0.06<0.6→trigger ventilation+cooling; (4) Platform application layer: start the ventilation fan and heating pipe (auxiliary dehumidification) of zone 5, and the humidity is reduced from 85% to 70% after 1 hour, and the canopy temperature is reduced to 27°C; (5) Algorithm step S4 (4.21 early morning): extract data from 4.15-4.20, find that humidity control ΔPi is 0.09<0.15, increase humidity Wi1 from 0.2 to 0.23; test zone 3 for 3 days, ΔPi rises to 0.17, and 4.24 is promoted to the whole greenhouse; Flowering period effect: flower drop rate decreased from 18% in the control group to 2.3%, and CO2 utilization rate increased by 40% (avoiding CO2 waste caused by high humidity).

[0045] 3. Results period (5.1-5.29); closed-loop feedback verification Key events (5.15 14:00): (1) Data acquisition layer: partition 2, 3 (adjacent central partition) leaf water potential -0.82 MPa (triggering drought stress), canopy temperature 30℃; (2) Algorithm step S3: drought stress threshold is corrected to -0.7 MPa, and soil moisture content Wi is increased to 0.4; (3) Platform application layer: start drip irrigation in partition 2 and 3 (30 minutes / time, interval 2 hours), physiological data is collected after 1 hour: leaf water potential in partition 2 is -0.7 MPa (ΔPi=0.14), and that in partition 3 is -0.72 MPa (ΔPi=0.12) → both are effective; (4) Algorithm step S4: soil moisture content Wi1 is increased from 0.17 to 0.18, and subsequent ΔPi is stabilized at 0.15-0.18; Results period effect: tomato single fruit weight 185g (control group 150g), total yield 8200kg / acre (control group 6500kg / acre), increased by 26.2%.

[0046] III. Comparison of implementation effects

[0047] Example 2: Application of cucumber greenhouse in southern region in summer (algorithm adaptability verification) This example takes a cucumber greenhouse (span 8m, length 40m, total area 320m 2 ) as the scene, the crop is summer cucumber (variety "Jinchun 4"), the implementation period is from June 1, 2024 to July 30, 2024 (whole growth period 60 days), and the adaptability of the algorithm and platform to high temperature and high humidity environment is verified.

[0048] I. Key adaptation adjustment Algorithm adjustment: Three-dimensional feature library: the optimal temperature range of cucumber fruiting period central partition is 23-26℃, and that of edge partition is 24-27℃; Weight correction: temperature Wi0 is additionally +0.1 in high temperature period (11:00-15:00), and humidity Wi2=1.3 (high humidity risk); Stress threshold: high humidity stress threshold is adjusted from "stomatal conductance ≤100" to "≤70".

[0049] Platform adjustment: Data acquisition layer: Added zoned shade net controller (1 set per zone), and added noon 12:00 flight time for drones; Application layer: The actuator adds a "wet curtain-fan" cooling system (1 set per zone).

[0050] II. Core Effects During the high-temperature period of summer (June 20 - July 10), the algorithm coordinated the control of "shading + wet curtain + ventilation" and the platform achieved independent execution of zones. The temperature in the central zone was stabilized at 25-26℃ (compared to 28-32℃ in the control group), and the total cucumber yield was 6800 kg / mu (compared to 5000 kg / mu in the control group), an increase of 36%. The incidence of diseases and pests (downy mildew) decreased from 25% to 5%, verifying the adaptability of the algorithm and the platform.

[0051] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of protection claimed by the present invention. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. An algorithm for modeling environmental regulation of crop physiology, characterized in that, Comprising the following steps: S1. Crop physiological spatiotemporal heterogeneity adaptation: Obtain greenhouse subzone physiological data and subzone environmental data, subzone physiological data including chlorophyll relative content SPAD value, leaf water potential and stomatal conductance parameters, subzone environmental data including air temperature, humidity, CO2 concentration, light intensity parameters, assign subzone basic weight Wi0 according to the distance between subzone and crop growth center, Wi0=1.0 when the distance≤2m, Wi0=0.9 when 2m<distance≤4m, Wi0=0.8 when distance>4m, and correct Wi0 according to the time period, Wi0 of light / CO2 is additionally +0.05 during the day photosynthesis vigorous period 9:00-15:00, and Wi0 of temperature is additionally +0.05 during the night respiration period 22:00-6:00; S2. Multi-source data dynamic coupling: access to future 24-hour weather forecast data and unmanned aerial vehicle canopy temperature data, calculate the correction weight Wi2 of multi-source data, humidity Wi2=1.2 when rainfall probability ≥60%, canopy temperature ≥30℃ when temperature Wi2=1.2; combine with crop growth period weight Wi1, CO2 Wi1=0.3, leaf water potential Wi1=0.25 in flowering stage, CO2 Wi1=0.3, temperature Wi1=0.2 in result stage, calculate the comprehensive weight Wi=Wi0×Wi1×Wi2; standardize physiological data and environmental data into deviation coefficient Pi as physiological coefficient, Ei as environmental coefficient, generate partition control decision value through formula D=Σ(Wi×Pi×Ei×E weather ×E canopy ), E weather is weather deviation coefficient, E canopy is canopy temperature deviation coefficient, D<0.6 triggers control; S3. Stress adaptive regulation: preset partition physiological stress threshold, leaf water potential ≤-0.8 MPa as drought stress, SPAD ≤40 as nutrient stress, combined with dynamic correction threshold of multi-source data, high humidity stress threshold of stomatal conductance from ≤100 mmol / m 2 s to ≤80 mmol / m 2 s; trigger threshold to improve the associated index Wi to 1.2-1.5 times of the original weight, and generate targeted regulation strategy; S4. Self-learning iterative optimization: Extract the "control instruction-physiological response" data pairs of the previous 7 days at 2:00 am every day, and optimize Wi0 and Wi1 using gradient descent algorithm, the optimization target is ΔPi after control, (Pi after control-Pi before control) / Pi before control, the average value≥0.15; if a certain index is continuously for 3 days ΔPi<0.1, the index Wi1 is increased by 0.03-0.05, until ΔPi≥0.

1.

2. The integrated plant physiology environmental regulation model algorithm of claim 1, wherein: The method for standardization processing in step S2 is: If the measured value is < the lower limit of the optimal range of the index, the deviation coefficient = 0; If the measured value is > the upper limit, the deviation coefficient = 1; If it is within the range, the deviation coefficient = (measured value-lower limit) / (upper limit-lower limit); The growth period weight Wi1 satisfies that the sum of all indexes Wi1 in the same subzone is 1.

3. The integrated plant physiology environmental regulation model algorithm of claim 1, wherein: The targeted control strategy in step S3 includes: subzone drip irrigation + leaf surface water spraying with a frequency of 2 hours / time and a duration of 30 minutes when drought stress occurs; fertilizer application reminder is pushed to the user end when nutrient stress occurs, and the Wi of SPAD value related environmental index is simultaneously increased.

4. The integrated plant physiology environmental regulation model algorithm of claim 1, wherein: The data pair screening in step S4 needs to exclude abnormal data containing sensor failure and power failure period data, and the sample size is≥100 groups; the optimized Wi0 and Wi1 need to be verified in 1 test subzone for 3 days, and when ΔPi is increased by≥10%, it is promoted to the whole greenhouse.

5. An integrated crop physiology environmental regulation algorithm model platform implementing the model algorithm of any one of claims 1-4, characterized in that, It comprises a data acquisition layer, a feature processing layer, a model algorithm layer and a platform application layer, and each layer cooperates to realize algorithm deployment and closed-loop control: The data collection layer comprises a partition physiological sensor group, a partition environment sensor group and a multi-source auxiliary data interface; the partition physiological sensor group is arranged every 50 m 2 1 partition, including an online SPAD sensor, a WP4-T leaf water potential meter and an AP4 stomatal conductance meter, with a collection frequency of once every 30 minutes in the seedling stage and once every 15 minutes in the flowering and fruiting stage; The multi-source auxiliary data interface accesses meteorological data through HTTP protocol and receives unmanned aerial vehicle canopy temperature data through 4G module; The feature processing layer constructs a "crop type-growth period-subregion" three-dimensional feature library, stores the optimal range of physiological / environmental indicators of each subregion, and the indicator deviation of the same crop in the same growth period in different subregions is ≤±5%. The collected data is subjected to outlier rejection, spatiotemporal interpolation completion and standardization processing, and Pi, Ei and E weather , E canopy are output. weather 0> The model algorithm layer deploys environmental control model algorithm, executes steps S1-S4, and generates subzone control decision value D and control instruction; The platform application layer includes multi-source data visualization module, subzone control instruction issuing module, effect feedback module and self-learning parameter management module; The instruction issuing module sends instructions to subzone execution mechanism through MQTT protocol, 2 light supplement lamps, 1 drip irrigation electromagnetic valve and 1 CO2 sub-pipeline valve for each subzone; The effect feedback module collects subzone physiological data 1 hour after control, and calculates ΔPi; The self-learning parameter management module stores Wi0 / Wi1 optimization records, and supports parameter rollback and locking.

6. The integrated plant physiology environmental regulation algorithmic model platform of claim 5, wherein: The partition environment sensor group of the data collection layer includes SHT31 temperature and humidity sensors, SCD30 CO2 sensors, TSL2591 light sensors, and TDR-350 soil moisture sensors.

7. The integrated plant physiology environmental regulation algorithmic model platform of claim 5, wherein: In the three-dimensional feature library of the feature processing layer, the optimal CO2 range of the center partition during the flowering period of tomatoes is 800-1000 ppm, and the optimal CO2 range of the edge partition is 780-980 ppm; the optimal temperature range of the center partition during the fruiting period of cucumbers is 23-26℃, and the optimal temperature range of the edge partition is 24-27℃.

8. The integrated plant physiology environmental regulation algorithmic model platform of claim 5, wherein: The partition control instruction issuing module of the platform application layer supports cross-partition coordination: when ≥3 adjacent partitions trigger the same stress, the "master-slave" coordination mode is started, the center partition is the master partition, the instruction is executed preferentially, and the slave partition is executed with a delay of 10 minutes to avoid excessive power grid load.

9. The integrated plant physiology environmental regulation algorithmic model platform of claim 5, wherein: The secondary control logic of the effect feedback module is: if ΔPi≤0.1 for a certain partition, the associated index Wi is increased to 0.08-0.1, and D is recalculated; if secondary control is ineffective for three consecutive times, a partition sensor fault warning is pushed to the user end.

10. The integrated plant physiology environmental regulation algorithmic model platform of claim 5, wherein: The multi-source data visualization module supports "partition map-index curve-time axis" linkage, can display the correlation curve of single-partition SPAD value and canopy temperature, a comparison chart of CO2 concentration at the same time for multiple partitions, and supports exporting a control effect report according to the growth period.

Images