A distributed corn water and fertilizer integrated application system and method based on an intelligent decision model

Abstract

The application discloses a distributed corn water and fertilizer integrated application system based on an intelligent decision model, adopts a "cloud-edge-end" three-level distributed architecture, and comprises a field sensing layer, an edge decision layer, a cloud control layer and a partition execution layer; the field sensing layer is used for collecting field multi-source environment and crop data and transmitting the data to the edge decision layer; the edge decision layer is arranged in each field control unit and independently generates water and fertilizer application decision instructions based on local data; the cloud control layer is used for global data convergence, model training and optimization scheduling; and the partition execution layer executes partition variable irrigation and fertilizer distribution work according to the decision instructions. Compared with the prior art, the application has the advantages that: 1. the distributed architecture has high reliability; 2. double driving decision improves precision; 3. offline operation has strong adaptability; 4. partition variable adapts to heterogeneity; 5. closed-loop optimization continuously evolves; and 6. agronomy is deeply fused.

Description

Technical Field

[0001] This invention relates to the field of precision irrigation and intelligent fertilization technology in agriculture, specifically to a distributed integrated water and fertilizer application system and method for corn based on an intelligent decision-making model. Background Technology

[0002] Corn is one of my country's three major grain crops, with an annual planting area of ​​approximately 650 million mu (16.7 million hectares), primarily produced in the Huang-Huai-Hai Plain and the Northeast Plain. Traditional corn cultivation generally employs an extensive management model of flood irrigation and surface fertilization, which results in low water and fertilizer utilization rates, significant resource waste, and high environmental pollution risks.

[0003] Integrated water and fertilizer management technology organically combines irrigation and fertilization, delivering dissolved fertilizer evenly to the crop root zone via a pipeline system, significantly improving water and fertilizer use efficiency. However, existing integrated water and fertilizer management systems mostly employ a centralized control architecture, with a central control cabinet managing all equipment, which has the following technical drawbacks: First, centralized architectures are prone to single-point-of-failure risks; a failure of the main control equipment can paralyze the entire field, making it difficult to meet the reliability requirements of large-scale contiguous planting. Second, decision-making models often use fixed thresholds or simple empirical formulas, failing to fully consider the spatial heterogeneity of the field and dynamic weather changes, resulting in limited decision-making accuracy. Third, existing intelligent decision-making systems are mostly pure mechanistic models or pure data-driven models. Mechanistic models are difficult to calibrate parameters and have poor adaptability, while data-driven models lack agronomic mechanism constraints and are prone to decision results that violate common sense. Fourth, most systems rely on cloud-based online operation, and cannot function properly when network communication is interrupted, making them unsuitable for remote farmland with poor network conditions.

[0004] To address the aforementioned issues, some research in the industry has attempted to introduce distributed architecture and intelligent algorithms. However, in practical applications, problems remain, such as the decision-making model not being closely integrated with the agronomical needs of maize, the distributed collaboration mechanism being imperfect, and the closed-loop feedback capability being insufficient. These issues make it difficult to meet the actual needs of precise water and fertilizer management throughout the entire growth period of maize. Summary of the Invention

[0005] To address the shortcomings of existing technologies, the present invention aims to provide a distributed corn fertigation system and method based on an intelligent decision-making model.

[0006] To address the aforementioned technical problems, the present invention provides the following technical solution: a distributed maize fertigation system based on an intelligent decision-making model, employing a three-tiered distributed architecture of "cloud-edge-device," comprising a field perception layer, an edge decision-making layer, a cloud control layer, and a zone execution layer; the field perception layer is used to collect multi-source environmental and crop data in the field and transmit it to the edge decision-making layer; the edge decision-making layer is deployed in each field control unit and independently generates fertigation decision instructions based on local data; the cloud control layer is used for global data aggregation, model training, and optimization scheduling; and the zone execution layer executes zone-specific variable irrigation and fertilizer application operations according to the decision instructions.

[0007] Furthermore, the field sensing layer includes soil monitoring nodes, meteorological monitoring stations, and crop growth acquisition units; the soil monitoring nodes are deployed in each zone according to a grid to collect soil volume moisture content and soil electrical conductivity at three layers: 10cm, 20cm, and 40cm; the meteorological monitoring station collects air temperature and humidity, light intensity, wind speed, and rainfall; the crop growth acquisition unit acquires maize canopy NDVI and plant height data through near-ground spectral sensors.

[0008] Furthermore, the edge decision layer incorporates a mechanism-data dual-driven intelligent decision-making model, which couples the FAO-56 Penman-Monteith crop water requirement mechanism model with the LSTM time-series prediction algorithm. The mechanism model is used to calculate the reference crop evapotranspiration and the maize crop coefficient, while the LSTM algorithm is used to correct prediction biases based on historical data. The outputs of the two are weighted and fused to generate the final irrigation amount decision.

[0009] Furthermore, the intelligent decision-making model also includes a maize growth period knowledge base and a water and fertilizer coupling optimization module; the growth period knowledge base stores the water and fertilizer demand thresholds for six stages of maize growth: seedling stage, jointing stage, large trumpet stage, tasseling and silking stage, grain filling stage, and maturity stage; the water and fertilizer coupling optimization module adopts the NSGA-II multi-objective optimization algorithm to solve for the optimal nitrogen, phosphorus and potassium ratio and irrigation quota with the objectives of maximizing yield and minimizing water and fertilizer input.

[0010] Furthermore, the edge decision layer has the ability to operate independently offline, with a built-in edge computing gateway and local data storage unit; when cloud communication is interrupted, the edge node continues to perform water and fertilizer management based on local historical data and preset rules, and automatically synchronizes data and decision logs after communication is restored.

[0011] Furthermore, the zonal execution layer includes a main water supply pipeline, a zonal solenoid valve group, a Venturi fertilizer applicator group, a fertilizer mother liquor tank, and a drip irrigation network; each field is zonal and equipped with an independent water inlet solenoid valve and fertilizer channel, supporting independent irrigation start / stop, flow regulation, and fertilizer concentration control; the fertilizer mother liquor tank is equipped with at least two independent storage tanks for nitrogen, phosphorus, and potassium, and the fertilizer absorption ratio is precisely adjusted by an electromagnetic metering pump.

[0012] Furthermore, the cloud-based management and control layer includes a data platform, a model training module, and a global optimization and scheduling module. The data platform aggregates the perception and execution data of all edge nodes to construct a soil-crop-environment spatiotemporal database. The model training module uses a sliding window mechanism to periodically update the parameters of the LSTM prediction model. The global optimization and scheduling module optimizes the irrigation sequence of each zone based on water resource constraints and peak and off-peak electricity periods.

[0013] Furthermore, the system also includes a closed-loop feedback verification mechanism; after each application of water and fertilizer, the field sensing layer continuously monitors changes in soil moisture and electrical conductivity, and the edge decision layer compares the deviation between the measured value and the predicted value. When the deviation exceeds the 15% threshold, the model parameters are automatically corrected and the next decision is adjusted.

[0014] A method for integrated water and fertilizer application of corn based on the system includes the following steps: S1: Divide the target cornfield into several management zones, and deploy a set of edge decision units and sensing nodes in each zone; S2: The sensing nodes collect soil, meteorological and crop data at a set frequency and transmit them to the local edge decision unit; S3: The edge decision-making unit calls the dual-drive intelligent decision-making model to make decisions based on the current fertilization cost zone irrigation volume, fertilizer application volume and water-fertilizer ratio during the current growing season. S4: The edge decision unit sends control commands to the partition execution layer to start the corresponding solenoid valves and fertilizer pumps to perform integrated water and fertilizer application; S5: After application, collect feedback data, calculate decision bias and correct model parameters; data is synchronously uploaded to the cloud. S6: The cloud periodically performs global model iteration training and scheduling optimization, and distributes the updated model parameters to each edge node.

[0015] Furthermore, the formula for calculating irrigation volume in step S3 is as follows: I=Kc×ET0×(θfc-θactual) / η; Where I is the irrigation amount, Kc is the maize crop coefficient, ET0 is the reference crop evapotranspiration, θfc is the field water holding capacity, θactual is the measured soil volumetric water content, and η is the irrigation water use coefficient; ET0 is the base value calculated by the FAO-56 model and then corrected by the LSTM model based on the meteorological data of the past 7 days.

[0016] With the above structure, this invention has the following advantages: 1. Distributed architecture with high reliability: Adopting a three-tier "cloud-edge-device" architecture, each partition's edge node makes independent decisions and executes independently, avoiding the risk of a single point of failure in a centralized system causing a complete system paralysis. A single node failure only affects the corresponding partition, while the remaining partitions can operate normally, significantly improving the overall system reliability.

[0017] 2. Dual-drive decision-making for improved accuracy: The mechanistic model and the data-driven model complement each other. The FAO-56 mechanistic model provides the theoretical basis for agronomy, while the LSTM data model corrects deviations based on actual data. The fusion of the two improves the decision-making accuracy by more than 18% compared to the pure mechanistic model, while the pure data model avoids abnormal decisions that violate agronomic common sense.

[0018] 3. Offline operation and strong adaptability: Edge nodes have built-in complete decision-making capabilities and data storage, do not rely on cloud online operation, can work normally when the network is interrupted, and are suitable for remote farmland and large-scale planting scenarios with poor network conditions.

[0019] 4. Zonal variables to adapt to heterogeneity: Supports the application of field-level zonal variables, which can make decisions based on the differences in soil texture, soil fertility level and crop growth in different zones, adapt to the spatial heterogeneity in the field, and improve water and fertilizer use efficiency compared with a uniform application mode.

[0020] 5. Closed-loop optimization and continuous evolution: Construct a complete closed loop of "perception-decision-execution-feedback-correction". The system continuously optimizes itself during operation. The longer it is used, the richer the data accumulation and the higher the decision accuracy.

[0021] 6. Deep integration of agronomy: It has a built-in knowledge base of the entire growth cycle of maize, and the decision-making process fully combines the growth and development law of maize. During the critical growth period, it focuses on ensuring the supply of water and fertilizer, and moderately controls the supply during non-critical periods. It conforms to the principles of maize cultivation and can achieve water and fertilizer conservation under the premise of stable yield. Detailed Implementation

[0022] The present invention will now be described in further detail.

[0023] A distributed fertigation system for maize based on an intelligent decision-making model adopts a three-level distributed architecture of "cloud-edge-device", including a field perception layer, an edge decision-making layer, a cloud control layer, and a zone execution layer.

[0024] The field sensing layer consists of soil monitoring nodes, meteorological monitoring stations, and crop growth data collection units. Soil monitoring nodes are deployed at a grid density of 50-100 mu (approximately 3.3-6.7 hectares), with each node collecting soil volumetric moisture content and soil electrical conductivity (EC) values ​​at depths of 10cm, 20cm, and 40cm. The collection frequency can be configured to 15-60 minutes per instance. One meteorological monitoring station is set up every 500 mu (approximately 33 hectares) to collect data on air temperature, relative humidity, photosynthetically active radiation, wind speed and direction, and rainfall. The crop growth data collection units utilize near-ground multispectral sensors to periodically acquire normalized difference vegetation index (NDVI) and plant height data for the maize canopy, used to assess crop growth status and nutrient stress levels.

[0025] The edge decision layer is deployed in each field control unit, with each unit managing an independent zone. It includes an edge computing gateway, local data storage, and decision algorithm modules. The core is a mechanism-data dual-driven intelligent decision-making model: the mechanism part uses the FAO-56 Penman-Monteith formula to calculate the reference crop evapotranspiration ET0, and combines it with the maize growth stage crop coefficient Kc to calculate the actual crop water requirement; the data-driven part uses an LSTM long short-term memory neural network to correct biases in the mechanism model output based on nearly 30 days of historical meteorological, soil, and irrigation data. The two outputs are weighted and fused at a 7:3 ratio to generate the final irrigation decision.

[0026] The decision-making model also integrates a maize growth stage knowledge base, covering water and fertilizer demand parameters for six key growth stages: seedling stage, jointing stage, large trumpet stage, tasseling and silking stage, grain filling stage, and maturity stage. These parameters include upper and lower limits of suitable soil moisture content, nitrogen, phosphorus, and potassium requirements, and EC value thresholds. The water and fertilizer coupling optimization module employs the NSGA-II multi-objective optimization algorithm, aiming to maximize yield, minimize water and fertilizer input, and minimize nitrogen leaching, to solve for the optimal irrigation quota and fertilizer ratio scheme for each growth stage.

[0027] The edge decision-making layer has complete offline operation capabilities, storing historical data and decision rules locally. When communication with the cloud is interrupted, it automatically switches to local independent operation mode, continuously executing water and fertilizer management based on cached data and preset strategies, ensuring normal basic functions without relying on the network.

[0028] The cloud-based management and control layer is deployed on a cloud server and includes a data platform, a model training module, and a global optimization and scheduling module. The data platform aggregates sensing data, execution logs, and decision records uploaded by all edge nodes to construct a spatiotemporal sequence database. The model training module uses a sliding window mechanism, retraining the LSTM prediction model every 7 days based on new data and distributing the updated model parameters to each edge node. The global optimization and scheduling module considers constraints such as water supply capacity, peak and off-peak electricity prices, and crop stress levels, using a genetic algorithm to optimize the irrigation timing of each zone, achieving efficient utilization of global resources.

[0029] The zoned execution layer includes the main water supply pipeline, zoned solenoid valve assemblies, Venturi fertilizer applicator assemblies, fertilizer stock solution tanks, and the drip irrigation network system. Each management zone is equipped with an independent inlet solenoid valve, supporting independent start / stop and flow rate adjustment. At least two fertilizer stock solution tanks are installed, one for storing nitrogen fertilizer stock solution and the other for storing a mixed phosphorus and potassium stock solution. These tanks are equipped with electromagnetic metering pumps and Venturi fertilizer applicators, allowing for precise adjustment of fertilizer absorption ratios and concentrations. The drip irrigation network uses labyrinth-type pressure-compensated drippers with a flow rate of 2.0-2.5 L / h, spaced 30 cm apart, laid between corn rows.

[0030] The system employs a closed-loop feedback verification mechanism: within 24 hours after each water and fertilizer application, soil monitoring nodes remeasure soil moisture and EC values. The edge decision layer compares the measured values ​​with the predicted values ​​and calculates the relative deviation. When the deviation exceeds a 15% threshold, the model parameters are automatically adaptively corrected, and the next decision-making scheme is adjusted, forming a continuously optimized closed-loop control.

[0031] The corresponding application method includes the following steps: S1: Field Zoning Deployment. The target cornfield is divided into several management zones based on topography, soil type, and crop variety, with each zone covering 50-100 mu (approximately 3.3-6.7 hectares). Each zone is equipped with a perimeter decision unit, soil monitoring nodes, and zone execution equipment, while weather stations are deployed on a regionally shared basis.

[0032] S2: Multi-source data acquisition. Sensing nodes collect data on soil moisture content (three layers), EC value, meteorological parameters, and crop growth at a set frequency. The data is transmitted to the local edge decision unit via LoRa or 4G for preliminary processing and storage.

[0033] S3: Local Intelligent Decision Making. The edge decision unit determines the current corn growth stage, calls the corresponding knowledge base parameters, runs the dual-drive decision model to calculate the irrigation amount, fertilizer amount and nitrogen, phosphorus and potassium ratio for this area, and generates water and fertilizer integration operation instructions.

[0034] S4: Zoned Application. The edge decision unit sends control commands to the zoned execution layer, opening the inlet solenoid valves according to the set sequence and starting the corresponding fertilizer metering pumps. The fertilizer concentrate is mixed into the irrigation water through a Venturi device and evenly delivered to the corn root zone via the drip irrigation network. During application, pipeline pressure, flow rate, and EC value are monitored in real time to ensure parameter stability.

[0035] S5: Feedback Correction and Data Synchronization. After application, soil feedback data is collected, decision bias is calculated, and model parameters are automatically corrected. Sensing data, decision records, and execution logs are synchronously uploaded to the cloud-based management layer.

[0036] S6: Cloud-based global optimization. The cloud periodically performs data aggregation and analysis, iterative model training, and global scheduling optimization, and distributes updated model parameters and scheduling schemes to each edge node to achieve continuous system optimization.

[0037] The present invention will be further described in detail below with reference to specific embodiments.

[0038] Example 1: Large-scale planting and application of summer maize in the Huang-Huai-Hai Plain This embodiment uses a 300-mu summer maize planting base in Huojia County, Xinxiang City, Henan Province as the application scenario. The variety is Zhengdan 958, the soil type is alluvial soil, the texture is medium loam, the field water holding capacity is 24% (volume water content), and the organic matter content of the topsoil is 12.5g / kg.

[0039] 1. System deployment; The 300-mu (approximately 20 hectares) plot of land was divided into three management zones, each approximately 100 mu (approximately 6.7 hectares). Each zone was equipped with one edge decision unit, comprising an edge computing gateway and a local data storage and control module. Within each zone, three soil monitoring nodes were deployed in a grid pattern to collect soil volumetric moisture content and EC (excessive organic matter) values ​​at depths of 10cm, 20cm, and 40cm, respectively, with a sampling frequency of 30 minutes per session. An automatic weather station was installed throughout the site to monitor air temperature and humidity, sunlight, wind speed, and rainfall.

[0040] Zoned execution layer configuration: Each zone is equipped with an independent inlet solenoid valve (DN100), and two fertilizer mother liquor tanks (1000L capacity) to store urea solution and potassium dihydrogen phosphate mixture respectively, along with an electromagnetic metering pump and a Venturi fertilizer applicator. Drip irrigation pipes are laid along the corn rows with a row spacing of 60cm, a dripper spacing of 30cm, and a dripper flow rate of 2.3L / h.

[0041] The cloud-based management layer is deployed on Alibaba Cloud servers, receiving data uploads from three partitions and running global optimization and model training.

[0042] 2. Decision model parameter configuration; The growth period of corn is divided into: seedling stage (sowing to jointing, about 25 days), jointing stage (jointing to large trumpet stage, about 20 days), large trumpet stage (large trumpet stage to tasseling, about 15 days), tasseling and silking stage (tasseling to silking end, about 10 days), grain filling stage (silking to milk stage, about 30 days), and maturity stage (milk stage to harvest, about 20 days).

[0043] The crop coefficients Kc for each growth stage are as follows: seedling stage 0.35, jointing stage 0.75, large trumpet stage 1.10, tasseling and silking stage 1.20, grain filling stage 1.05, and maturity stage 0.70.

[0044] Soil moisture control limits: 60% field capacity during seedling stage, 65% during jointing stage, 70% during large trumpet stage, 75% during tasseling and silking stage, 70% during grain filling stage, and 60% during maturity stage.

[0045] Total nitrogen, phosphorus, and potassium application rate: 210 kg / hm² (pure N) 2 P2O5 90kg / hm 2 K2O 120kg / hm 2 The distribution ratios based on the growth period are as follows: seedling stage 10%, jointing stage 20%, large trumpet stage 35%, tasseling and silking stage 20%, and grain filling stage 15%.

[0046] Dual-drive model weight configuration: mechanistic model accounts for 70%, LSTM correction term accounts for 30%. LSTM model input features include the daily maximum and minimum temperatures, average humidity, sunshine duration, wind speed, rainfall, and soil moisture content of the previous day for the past 7 days.

[0047] 3. Operation process; Taking the typical operation during the large trumpet mouth stage as an example: At 8:00 a.m. on a certain day, the soil monitoring node measured the soil volumetric water content at a depth of 20cm to be 15.8%, which is 65.8% of the field capacity. This is lower than the control limit of 70%, triggering an irrigation decision.

[0048] The edge decision unit called the FAO-56 model to calculate the reference crop evapotranspiration ET0 = 5.2 mm / d the previous day, Kc = 1.10 at the large bell mouth stage, and the actual crop evapotranspiration ETc = 5.72 mm.

[0049] The LSTM model predicts a correction of +0.4 mm based on the meteorological data of the past 7 days. After weighted fusion, the final estimated daily water consumption is 5.72×0.7+(5.72+0.4)×0.3=5.84 mm.

[0050] Calculate irrigation amount: The soil moisture content needs to be increased from 15.8% to 16.8% (70% field capacity), the planned wetting layer is 40cm, and the irrigation water utilization coefficient is 0.90.

[0051] Irrigation volume I = (0.168 - 0.158) × 0.4 × 1000 / 0.90 = 44.4 m³ 3 / mu.

[0052] Water and fertilizer ratio: Nitrogen fertilizer accounts for 35% of the total nitrogen during the large trumpet stage. The amount of nitrogen applied in this irrigation is 210 × 0.35 / 3 = 24.5 kg / hm². 2 =1.63 kg / mu, which translates to 3.54 kg / mu of urea (containing 46% N). Phosphorus and potassium fertilizers should be applied simultaneously in proportion.

[0053] The edge decision unit issues instructions to open the zone water inlet valve, start the No. 1 nitrogen fertilizer metering pump and the No. 2 phosphorus and potassium metering pump, and adjust the fertilizer absorption ratio to maintain the EC value of irrigation water at 1.8-2.0 mS / cm.

[0054] Irrigation lasted approximately 4 hours. After application, the valves and fertilizer pumps were shut off. The soil moisture content was retested at 8:00 AM the following day and found to be 16.5%, with a deviation of 1.8%, which is within the allowable range. Therefore, this decision was recorded as valid.

[0055] 4. Application effects; Results from the entire growth period show that, compared with the local traditional flood irrigation + topdressing fertilization model: Irrigation water volume from 320m 3 / mu decreased to 268m 3 / mu, saving 16.25% of water; The amount of chemical fertilizer applied (pure) was reduced from 38 kg / mu to 33.1 kg / mu, resulting in fertilizer savings of 12.9%. The yield of corn per mu increased from 652 kg to 698 kg, an increase of 7.06%; The cost savings per mu (unit of land area) increased income by approximately 135 yuan.

[0056] Example 2: Zonal Application of Spring Maize Sloping Farmland in Northeast China This example uses a 200-mu (approximately 33.3 hectares) sloping farmland in Siping City, Jilin Province, as a scenario for spring corn planting. The variety is Xianyu 335. The terrain has a slope of 5°-8°, and the soil exhibits significant spatial heterogeneity, with sandy loam on the upper slope and medium loam on the lower slope.

[0057] 1. Differentiated partitioned deployment; The 200-mu area was divided into four zones based on topography and soil type: two zones of 45 mu each on the uphill slope (sandy loam soil), and two zones of 55 mu each on the downhill slope (medium loam soil). Each zone was independently equipped with edge decision units and soil monitoring nodes, and different soil hydraulic parameters were configured for each zone: uphill sandy loam soil had a field water holding capacity of 20% and a saturated hydraulic conductivity of 15 mm / h; downhill medium loam soil had a field water holding capacity of 25% and a saturated hydraulic conductivity of 8 mm / h.

[0058] In response to the characteristics of runoff loss on slopes, the upslope zoning adopts a strategy of small, frequent irrigations, reducing the amount of irrigation per irrigation by 20% and increasing the frequency of irrigation; the downslope zoning appropriately increases the amount of irrigation per irrigation and reduces the number of irrigations.

[0059] 2. Distributed collaborative scheduling; The cloud-based global optimization and scheduling module prioritizes irrigation for uphill zones based on the hydrological patterns of the slope. Excess infiltration water can replenish soil moisture on the downhill slope, improving overall utilization efficiency. Simultaneously, considering the limitations of well water supply capacity, a staggered irrigation mechanism is adopted for the four zones to avoid insufficient pressure caused by simultaneous water supply.

[0060] 3. Extreme weather response; In response to the spring drought and intermittent low temperatures in Northeast China, the decision-making model has added extreme weather correction rules: when there is no effective rainfall predicted for the next 3 days and the soil moisture content continues to decline, preventive irrigation will be started in advance; when the average daily temperature is below 15℃ for 3 consecutive days, the irrigation amount will be reduced and nitrogen application will be suspended to avoid corn seedling diseases caused by low temperature and high humidity.

[0061] 4. Application effects; One growing season operation showed that: The overall irrigation uniformity of sloping farmland has increased from the traditional 68% to 89%; In sandy loam soil plots on slopes, the emergence rate of corn seedlings increased by 12%, and the lodging rate in the later stages decreased by 8%. The average water and fertilizer utilization rate of the whole field increased by 18%, and the yield variation coefficient decreased from 15.2% to 8.7%.

[0062] Example 3: Improved water and fertilizer integration application for maize cultivation in saline-alkali land This example uses 150 mu of coastal saline-alkali land in Dongying City, Shandong Province as a scenario for corn planting. The soil salinity is 0.3%-0.5%, which is considered moderately saline-alkali land. The salt-tolerant corn variety Denghai 605 was selected.

[0063] 1. System modification and configuration; The system adds soil salinity monitoring functionality to the conventional system, with salinity sensors added to soil nodes to monitor the conductivity of soil leachate in real time. A dedicated salt-leaching program is added to the execution layer, enabling large-scale irrigation salt-leaching operations.

[0064] The decision-making model adds a salt stress constraint module. When the soil EC value exceeds the salt tolerance threshold of maize (1.2 mS / cm in the seedling stage and 2.0 mS / cm in the mid-to-late growth stage), it automatically triggers salt leaching irrigation to wash the salt down to below the root zone.

[0065] The fertilizer ratio was adjusted to increase the proportion of potassium fertilizer and appropriately reduce the concentration of nitrogen fertilizer. A strategy of applying small amounts of fertilizer multiple times was adopted to avoid excessive local salinity caused by a single application of fertilizer.

[0066] 2. Coordinated regulation of water, fertilizer, and salt; The decision-making model enables coordinated regulation of water, fertilizer, and salt. During the seedling stage, the main focus is on controlling salt content and protecting seedlings. This involves frequent, light irrigation to maintain stable soil salinity and applying light fertilizer to the seedlings. After the jointing stage, gradually increase the supply of water and fertilizer as the salt tolerance of corn improves; Before each fertilization, irrigate the soil with clean water for 30 minutes to moisten it. After fertilization, flush the pipes with clean water to prevent fertilizer salts from clogging the drippers.

[0067] 3. Application effects; The soil salinity in the root zone of maize was consistently controlled below 0.35% throughout the entire growth period, and no symptoms of salt stress were observed. Compared to traditional local farming methods, it saves 21% on water and 16% on fertilizer. The yield of corn in saline-alkali land reached 580 kg per mu, an increase of 23% compared with the control field.

[0068] Working Principle Summary: This invention utilizes a distributed architecture to decentralize decision-making capabilities to the edge of the field. Each zone independently perceives, makes decisions, and executes, fundamentally solving the reliability problem of single-point-of-failure in centralized systems. The mechanism-data dual-driven decision-making model retains the agronomical theoretical foundation of crop water and fertilizer requirements while continuously learning and correcting from actual data through machine learning, achieving a dynamic balance between theory and practice. The closed-loop feedback mechanism enables the system to self-evolve, with accuracy increasing over time. Zoned variable application adapts to the spatial heterogeneity of the field, allowing for differentiated management based on different soil, topography, and growth conditions, truly achieving on-demand supply.

[0069] The present invention and its embodiments have been described above. This description is not restrictive, and the actual structure is not limited thereto. In conclusion, if those skilled in the art, inspired by this description, design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the scope of protection of this invention.

Claims

1. A distributed fertigation system for corn based on an intelligent decision-making model, characterized in that, The system adopts a three-tiered distributed architecture of "cloud-edge-device", including a field perception layer, an edge decision layer, a cloud control layer, and a zone execution layer. The field perception layer is used to collect multi-source environmental and crop data in the field and transmit it to the edge decision layer. The edge decision layer is deployed in each field control unit and independently generates water and fertilizer application decision instructions based on local data. The cloud control layer is used for global data aggregation, model training, and optimization scheduling. The zone execution layer executes zone-specific variable irrigation and fertilizer application operations according to the decision instructions.

2. The distributed fertigation system for maize based on an intelligent decision-making model according to claim 1, characterized in that, The field sensing layer includes soil monitoring nodes, meteorological monitoring stations, and crop growth acquisition units. Soil monitoring nodes are deployed in each zone in a grid pattern to collect soil volumetric moisture content and soil electrical conductivity data at 10cm, 20cm, and 40cm depths. Meteorological monitoring stations collect air temperature and humidity, light intensity, wind speed, and rainfall data. Crop growth acquisition units acquire maize canopy NDVI and plant height data through near-ground spectral sensors.

3. A distributed fertigation system for maize based on an intelligent decision-making model according to claim 1, characterized in that, The edge decision layer incorporates a mechanism-data dual-driven intelligent decision-making model, which couples the FAO-56 Penman-Monteith crop water requirement mechanism model with the LSTM time-series prediction algorithm. The mechanism model is used to calculate the reference crop evapotranspiration and the maize crop coefficient, while the LSTM algorithm is used to correct prediction biases based on historical data. The outputs of the two are weighted and fused to generate the final irrigation amount decision.

4. A distributed fertigation system for maize based on an intelligent decision-making model according to claim 3, characterized in that, The intelligent decision-making model also includes a maize growth period knowledge base and a water and fertilizer coupling optimization module. The growth period knowledge base stores the water and fertilizer demand thresholds for six stages of maize growth: seedling stage, jointing stage, large trumpet stage, tasseling and silking stage, grain filling stage, and maturity stage. The water and fertilizer coupling optimization module uses the NSGA-II multi-objective optimization algorithm to solve for the optimal nitrogen, phosphorus and potassium ratio and irrigation quota with the objectives of maximizing yield and minimizing water and fertilizer input.

5. A distributed fertigation system for maize based on an intelligent decision-making model according to claim 1, characterized in that, The edge decision layer has the ability to operate independently offline, and has a built-in edge computing gateway and local data storage unit. When cloud communication is interrupted, the edge node continues to perform water and fertilizer management based on local historical data and preset rules. After communication is restored, the data and decision logs are automatically synchronized.

6. A distributed fertigation system for maize based on an intelligent decision-making model according to claim 1, characterized in that, The zonal execution layer includes a main water supply pipeline, a zonal solenoid valve group, a Venturi fertilizer applicator group, a fertilizer mother liquor tank, and a drip irrigation network; each field is equipped with an independent inlet solenoid valve and fertilizer channel, supporting independent irrigation start / stop, flow regulation, and fertilizer concentration control for each zone; the fertilizer mother liquor tank is equipped with at least two independent storage tanks for nitrogen, phosphorus, and potassium, and the fertilizer absorption ratio is precisely adjusted by an electromagnetic metering pump.

7. A distributed fertigation system for maize based on an intelligent decision-making model according to claim 1, characterized in that, The cloud-based management and control layer includes a data platform, a model training module, and a global optimization and scheduling module. The data platform aggregates the perception and execution data of all edge nodes to construct a soil-crop-environment spatiotemporal database. The model training module uses a sliding window mechanism to periodically update the parameters of the LSTM prediction model. The global optimization and scheduling module optimizes the irrigation sequence of each zone based on water resource constraints and peak and off-peak electricity periods.

8. A distributed fertigation system for maize based on an intelligent decision-making model according to claim 1, characterized in that, The system also includes a closed-loop feedback verification mechanism; after each application of water and fertilizer, the field sensing layer continuously monitors changes in soil moisture and electrical conductivity, and the edge decision layer compares the deviation between the measured value and the predicted value. When the deviation exceeds the 15% threshold, the model parameters are automatically corrected and the next decision is adjusted.

9. A method for integrated water and fertilizer application of corn based on the system described in any one of claims 1-8, characterized in that, Includes the following steps: S1: Divide the target cornfield into several management zones, and deploy a set of edge decision units and sensing nodes in each zone; S2: The sensing nodes collect soil, meteorological and crop data at a set frequency and transmit them to the local edge decision unit; S3: The edge decision-making unit calls the dual-drive intelligent decision-making model to make decisions based on the current fertilization cost zone irrigation volume, fertilizer application volume and water-fertilizer ratio during the current growing season. S4: The edge decision unit sends control commands to the partition execution layer to start the corresponding solenoid valves and fertilizer pumps to perform integrated water and fertilizer application; S5: After application, collect feedback data, calculate decision bias and correct model parameters; data is synchronously uploaded to the cloud. S6: The cloud periodically performs global model iteration training and scheduling optimization, and distributes the updated model parameters to each edge node.

10. The method according to claim 9, characterized in that, The formula for calculating irrigation volume in step S3 is: I=Kc×ET0×(θfc-θactual) / η; Where I is the irrigation amount, Kc is the maize crop coefficient, ET0 is the reference crop evapotranspiration, θfc is the field water holding capacity, θactual is the measured soil volumetric water content, and η is the irrigation water use coefficient; ET0 is the base value calculated by the FAO-56 model and then corrected by the LSTM model based on the meteorological data of the past 7 days.