A multi-target dynamic precision fertilization control parameter optimization method for water and fertilizer integrated machine

Abstract

The application discloses a kind of water and fertilizer integrated machine multi-objective dynamic precision fertilization control parameter optimization method, it is related to agricultural irrigation and fertilization control technical field.The method is by constructing "crop nutrient supply-water and fertilizer resource saving-soil physical and chemical balance-environment adaptation" four-dimensional multi-objective optimization index system, integrates multi-source sensor array real-time acquisition soil, crop, environmental data, using improved non-dominated sorting genetic algorithm III (NSGA-III) to fertilizer ratio concentration, irrigation flow, fertilization duration, stirring frequency and so on Control parameter is dynamically optimized, and continuously improves adaptability by self-learning iteration mechanism.The application solves the technical defects that existing water and fertilizer integrated machine fertilization control target is single, parameter adjustment lag, poor adaptability, realizes precision fertilization under multi-objective balance, substantially improves water and fertilizer utilization rate, reduces labor intensity, adapts to a variety of scenes such as field planting, facility agriculture, orchard planting, provides strong technical support for precision agriculture development.

Description

Technical Field

[0001] This invention relates to the field of agricultural irrigation and fertilization technology, specifically to a method for optimizing control parameters of an integrated water and fertilizer machine based on multi-sensor data fusion and an improved intelligent algorithm, which is particularly suitable for dynamic fertilization scenarios under complex soil conditions. Background Technology

[0002] Existing fertigation systems mostly use a "static parameter input + fixed ratio" mode for fertilizer control, which has three major drawbacks: Single-minded focus: Taking only "fertilizer application per acre" as the core objective without considering the real-time nutrient needs of crops, water conservation, soil physicochemical balance, and environmental adaptability (such as the impact of temperature and humidity on fertilizer efficiency) can easily lead to fertilizer waste or crop nutrient imbalance. Parameter adjustment lag: Relying on manually preset crop growth cycle parameters, it cannot respond to dynamic factors such as soil moisture, crop physiological state (such as leaf nitrogen content), and environmental changes (rainfall, high temperature), resulting in insufficient accuracy; Limitations of optimization algorithms: They often use simple proportional control or single-objective optimization algorithms, which are difficult to balance the conflict of multiple objectives such as "nutrient standard, water and fertilizer conservation, and soil protection", and have poor adaptability (different crops and plots require repeated manual adjustment of parameters).

[0003] Existing technologies cannot meet the demands of modern precision agriculture for "dynamic adaptation, multi-objective balancing, and intelligent self-optimization," and there is an urgent need for a method that can dynamically respond to multi-dimensional data and automatically optimize control parameters. Summary of the Invention

[0004] Therefore, the purpose of this invention is to provide a method for optimizing the multi-objective dynamic precision fertilization control parameters of an integrated water and fertilizer machine, which aims to solve the problems existing in the background technology, such as the single fertilization control objective, the lag in parameter adjustment, the limitation of optimization algorithms, and the poor overall adaptability.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: A method for optimizing multi-objective dynamic precision fertilization control parameters in an integrated water and fertilizer machine includes the following steps: Step 1: Construct a four-dimensional, multi-objective optimization index system covering "crop nutrient supply - water and fertilizer resource conservation - soil physicochemical balance - environmental adaptation". This system specifically includes core indicators such as crop nutrient precision supply rate, water and fertilizer resource utilization rate, soil physicochemical balance, and environmental adaptability.

[0006] Step 2: Deploy a multi-source sensor array for real-time data collection of soil, crop, and environmental data. Specifically, in-situ soil sensors collect soil moisture, pH, EC values, and nitrogen, phosphorus, and potassium mass fractions; spectral crop sensors, combined with image recognition technology, collect crop leaf nitrogen content, plant height, and leaf area index; and environmental sensors collect air temperature, humidity, light intensity, and rainfall.

[0007] Step 3: The key control parameters of the integrated water and fertilizer machine are dynamically optimized using the improved Non-dominated Sorting Genetic Algorithm III (NSGA-III). The improvements include: introducing an adaptive crossover and mutation operator to improve the convergence speed of the algorithm, and adding a constraint penalty mechanism to effectively eliminate invalid solutions that do not meet actual production requirements.

[0008] Step 4: Preprocess the real-time data collected in Step 2 through an edge computing gateway, and input the preprocessed data, along with preset crop variety parameters and soil basic parameters, into the improved NSGA-Ⅲ algorithm to calculate and output a set of optimal control parameter combinations. The control parameters include at least fertilizer concentration, irrigation flow rate, fertilization duration, and stirring frequency.

[0009] Step 5: Transmit the optimal control parameter combination obtained in Step 4 to the PLC (Programmable Logic Controller), which will drive the integrated water and fertilizer machine to perform precision fertilization.

[0010] Step 6: After each fertilization operation, the system records the "control parameter combination" used and its corresponding "actual effect data". If the error between the actual effect and the preset target index exceeds the allowable threshold, the system automatically updates the algorithm's training dataset, thereby initiating a self-learning iterative optimization mechanism to continuously bring the system parameters closer to the optimal state.

[0011] Furthermore, an integrated multi-source sensor array is deployed to collect three types of key data in real time: soil in-situ sensors collect soil moisture, pH value, EC value, and nitrogen, phosphorus, and potassium mass fractions (N / P / K detection accuracy ≤ ±5%, soil moisture detection range 0-100% Vol, accuracy ≤ ±2% Vol); spectral crop sensors (detection wavelength 400-1000 nm) combined with image recognition collect crop leaf nitrogen content (error ≤ ±0.1%), plant height, and leaf area index; environmental sensors collect air temperature, humidity, light intensity, and rainfall. Simultaneously, outliers are removed using the 3σ criterion, and the data is mapped to the [0,1] interval using Min-Max standardization, and transmitted to the edge computing gateway via the MQTT protocol (transmission latency ≤ 1 second). This multi-dimensional data collection captures the dynamic changes in soil, crops, and the environment, and preprocessing ensures data accuracy and real-time performance; providing comprehensive and reliable data support for subsequent parameter optimization, and overcoming the shortcomings of traditional methods that rely on manual presets and cannot respond to dynamic factors.

[0012] Furthermore, an improved non-dominated sorting genetic algorithm III (NSGA-III) is used to optimize the control parameters. This algorithm introduces adaptive crossover and mutation operators (crossover probability P_c = P_c0 × exp (-|f_avg - f_max| / (k×f_max)), mutation probability P_m = P_m0 × exp (-|f_avg - f_min| / (k×f_min)), where P_c0 takes values ​​of 0.7-0.9, P_m0 takes values ​​of 0.01-0.05, and k takes values ​​of 0.1-0.3) to improve the convergence speed. A constraint penalty mechanism is constructed by improving the fitness function (F'(x) = F (x) - λ×∑max (0, g_i (x)), where λ takes values ​​of 10-50), and an environmental factor compensation module is added. By improving the algorithm to balance multi-objective conflicts, constraining parameter compliance, and compensating for the impact of environmental factors on fertilizer efficiency, the algorithm converges faster and has fewer invalid solutions. It can dynamically adjust parameters for scenarios such as high temperature (>35℃) and rainfall (>10 mm / 24h), reduce fertilizer volatilization and nutrient leaching loss, and output the optimal parameter combination that is more in line with the actual scenario.

[0013] Furthermore, the optimal combination of control parameters must adhere to crop type adaptation rules. The system determines the crop type through user presets or leaf morphology image recognition: for shallow-rooted crops, the fertilization time is controlled at 20-30 minutes and the stirring frequency at 40-50 r / min; for deep-rooted crops, the fertilization time is controlled at 40-60 minutes and the stirring frequency at 60-70 r / min. Control parameters include fertilizer concentration, irrigation flow rate, fertilization time, and stirring frequency. Parameters are adjusted according to the root distribution characteristics of different crops to match nutrient absorption efficiency; this improves the targeting of fertilization and avoids the problem of nutrient excess in shallow-rooted crops and nutrient deficiency in deep-rooted crops caused by uniform parameters, thus adapting to various crop planting scenarios.

[0014] Furthermore, an active soil pH adjustment step is added: when the soil in-situ sensor detects a pH value < 5.5, the PLC controller drives the device to inject calcium carbonate solution, with the addition amount V = (5.5 - pH_actual) × S × k1 (k1 takes a value of 0.02-0.03) and not exceeding 5% of the total fertilizer solution volume; when the pH value > 7.5, phosphoric acid solution is injected, with the addition amount V = (pH_actual - 7.5) × S × k2 (k2 takes a value of 0.015-0.025). This acid-base regulator precisely corrects the soil pH value, maintaining soil physicochemical balance; it avoids the inhibition of crop growth by soil acidification or alkalization, ensures nutrient absorption efficiency, and extends the soil's arable lifespan.

[0015] Furthermore, a self-learning iterative optimization mechanism is established: after each task is completed, "control parameter combination - actual effect data" is recorded. When the average error of three consecutive tasks is ≤3%, the algorithm training set update frequency is reduced to once every 7 days; when the single error is >8%, an emergency update is triggered to retrain and verify the parameters; the training set update uses a weighted average method to retain historical valid data, with the weight of historical data decaying at 0.9 per month. The algorithm model is continuously optimized through feedback from actual effects, strengthening the adaptability of parameters to real-world scenarios; achieving dynamic improvement in parameter accuracy, reducing the frequency of manual debugging, and gradually enhancing adaptability over long-term use, making it suitable for various planting scenarios such as field crops, facility agriculture, and orchards.

[0016] Compared with the prior art, the present invention has the following significant advantages: Multi-objective balance optimization significantly improves resource utilization: Through a four-dimensional multi-objective indicator system and improved algorithms, it achieves synergistic optimization of precise crop nutrient supply, water and fertilizer conservation, soil protection, and environmental adaptation, increasing fertilizer utilization by 15%-20% and saving water resources by 10%-15%, thus solving the core defect of existing technologies that "focus on one thing but lose another".

[0017] Strong dynamic adaptability and high accuracy: It integrates multi-source sensors to collect data in real time, and the parameter adjustment response time is ≤5 seconds. It can quickly respond to the dynamic changes of soil, crops and environment without human intervention. The crop nutrient precision supply rate is ≥95%, which is far higher than the existing technology (≤85%).

[0018] The algorithm boasts excellent performance and wide adaptability: the improved NSGA-Ⅲ algorithm has a convergence speed that is more than 30% faster, effectively avoiding local optima. Combined with crop type adaptation rules, active soil pH adjustment, and environmental factor compensation modules, it can adapt to different crops such as leafy vegetables, melons and fruits, fruit trees, different soil types such as loam and sandy soil, and different planting scenarios such as open fields, greenhouses, and orchards.

[0019] Self-learning and iterative optimization with low operating threshold: The training set is updated through a closed-loop data mechanism of "parameters-effects", which continuously improves the adaptability of the algorithm. There is no need to repeatedly adjust parameters manually for different crops and plots. Ordinary farmers can operate it, reducing the cost of promotion and application.

[0020] High compatibility and low modification difficulty: It can be directly embedded into existing PLC-controlled water and fertilizer integrated machines without modifying the hardware structure. Only the software algorithm needs to be upgraded and the necessary sensors need to be added. The modification cost is low and it is easy to promote and apply to existing agricultural equipment.

[0021] Significant soil protection effects: By actively regulating soil pH and controlling physicochemical balance, soil acidification, compaction, and salinization are avoided, extending soil tillage life and meeting the needs of green agricultural development. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 A flowchart provided for an embodiment of the present invention; Figure 2 A diagram illustrating the four-dimensional multi-objective optimization index system provided in this embodiment of the invention; Figure 3 The system architecture and algorithm optimization flowchart provided for embodiments of the present invention.

[0024] The accompanying drawings have illustrated specific embodiments of the invention, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the invention in any way, but rather to illustrate the concept of the invention to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] To make the technical solutions and advantages of the present invention clearer, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0027] Please see Figures 1 to 3 As shown, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. The described embodiments are intended to make the technical solutions of the present invention easier for those skilled in the art to understand and implement, and are not intended to limit the scope of protection of the present invention. This embodiment takes the field corn planting scenario in northern China as an example to illustrate the specific application process of the method of the present invention in detail, covering the entire process from preliminary preparation to operation execution and iterative optimization, to ensure that all technical solutions in the claims are fully disclosed and reflected.

[0028] In the early stages of implementation, it is necessary to confirm the basic parameters of the planting scenario and deploy the equipment. The selected field corn planting area is 2000m², the soil type is sandy loam, and the field water holding capacity is 28% Vol according to the preliminary test. Based on the root distribution characteristics of corn, it is preset to be a deep-rooted crop. The growth period will go through three key stages: seedling stage, jointing stage and grain filling stage. The nutrient and water requirements of each stage are significantly different. In the equipment deployment phase, an integrated multi-source sensor array needs to be set up according to the density requirements. Among them, the soil in-situ sensor is selected with a model that has multi-parameter synchronous detection function. Its N / P / K detection accuracy is actually ±4.5% (meeting the requirement of ≤±5%), and the soil moisture detection range covers 0-100% Vol. The detection accuracy after calibration is ±1.7% Vol (meeting the standard of ≤±2% Vol). This type of sensor is buried in the soil tillage layer (about 25cm deep) at a density of 1 per 100m². The spectral crop sensor is selected with a detection wavelength range of 400-1000 nm. It is equipped with a high-definition industrial camera to realize image recognition function. In actual detection, the leaf nitrogen content error can be controlled within ±0.08% (below the error requirement of ≤±0.1%). The sensor is installed on a movable bracket, and the height of the bracket is adjusted to 1.8m away from the corn canopy to ensure that the detection range covers the entire planting area. The environmental sensor is installed on the weather station bracket in the center of the planting area, which can collect air temperature, humidity, light intensity and rainfall data in real time. It should be noted that the aforementioned sensors can also be installed in front of the mobile fertigation unit, collecting data while it is being moved, while the fertilization pipeline is installed at the rear of the mobile fertigation unit.

[0029] In terms of hardware configuration, the edge computing gateway adopts a model that supports industrial-grade IoT protocols, has data storage and preprocessing capabilities, and is compatible with the MQTT protocol; the PLC controller uses a mainstream medium-sized programmable controller, which can simultaneously connect to the fertilizer mixing pump, irrigation solenoid valve, stirring motor, and acid-base regulator addition pump of the integrated water and fertilizer machine to realize multi-device collaborative control; the improved non-dominated sorting genetic algorithm III (NSGA-III) is pre-deployed in the built-in processor of the edge computing gateway, with the initial parameters of the algorithm set to a population size of 120 and 60 iterations, laying the foundation for subsequent parameter optimization.

[0030] The first step in the implementation process is to construct a four-dimensional, multi-objective optimization indicator system. This system includes four core indicators: crop nutrient precision supply rate, water and fertilizer resource utilization rate, soil physicochemical balance, and environmental adaptability. The definition of each indicator needs to be clarified in conjunction with the actual needs of maize planting: The crop nutrient precision supply rate is based on the standard physiological parameters of maize at different growth stages, quantifying the degree of conformity between the actual measured leaf nitrogen content, plant height, and leaf area index and the target values. For example, the target value for maize leaf nitrogen content during the seedling stage is 3.0%, the target value for plant height is 20cm, and the target value for leaf area index is 0.6; during the fruit expansion stage (maize grain filling stage), the target value for leaf nitrogen content is 3.8%, the target value for plant height is 180cm, and the target value for leaf area index is 3.5. The water and fertilizer resource utilization rate is calculated by the ratio of the total amount of nutrients actually absorbed by the crop to the total amount of fertilizer applied, and the ratio of the amount of water absorbed by the crop to the total amount of irrigation, which directly reflects the efficiency of resource utilization. The soil physicochemical balance is defined by the percentage of time that the soil pH value is maintained in the suitable range of 5.5-7.5 and the EC value is stable at 1.2-2.0 mS / cm. The proportion within a reasonable range serves as the evaluation criterion to ensure soil environmental stability. Environmental adaptability is quantified by the ratio of actual fertilizer efficiency to fertilizer efficiency under ideal conditions (temperature 25-30℃, humidity 60-70%, no rainfall), reflecting the ability to adapt to environmental changes. Simultaneously, the weights of this indicator system can be dynamically adjusted according to the corn's growth stage. When corn is in the seedling stage, as this is a critical period for the formation of the crop's growth foundation, the weight of the crop nutrient precision supply rate is set to 0.4, and the weights of the other three indicators are each set to 0.2. After entering the grain-filling stage, corn's nutrient demand surges while also requiring resource conservation. The system can automatically trigger weight adjustments through leaf morphology data collected by spectral sensors and ear development status identified by image recognition. At this time, the weight of the crop nutrient precision supply rate increases to 0.5, the weight of water and fertilizer resource utilization rate is set to 0.3, and the weights of soil physicochemical balance and environmental adaptability are each set to 0.1. If manual intervention is required, the weight parameters can also be manually adjusted through the user terminal to ensure that the indicator system always aligns with the crop's growth needs.

[0031] The second step is multi-source data acquisition and preprocessing. During a certain operating cycle in the corn seedling stage, the soil data collected in real time by the in-situ soil sensor includes: soil moisture 16% Vol, pH value 6.3, EC value 1.5 mS / cm, nitrogen content 22 mg / kg, phosphorus content 10 mg / kg, and potassium content 28 mg / kg; the spectral crop sensor combined with image recognition technology collected corn leaf nitrogen content 2.8%, plant height 18 cm, and leaf area index 0.55; the environmental sensor collected real-time environmental data including air temperature 26℃, humidity 62%, light intensity 7500 lx, and 24-hour rainfall 0 mm. After data collection, preprocessing is required. First, the 3σ criterion is used to remove outliers. Taking soil moisture data as an example, the mean of this batch of soil moisture data is calculated to be 16% Vol, and the standard deviation is 1.8% Vol. Based on this, the outlier judgment range is determined to be the mean ± 3 times the standard deviation (i.e., 16% Vol ± 5.4% Vol). Data exceeding the range of 10.6% Vol - 21.4% Vol are considered outliers. All data collected in this batch fall within this range, and no outliers need to be removed. Next, the Min-Max standardization method is used to map the data to the [0,1] interval. Taking soil moisture data standardization as an example, the original data x = 16% Vol, the minimum value of this type of data x_min = 0% Vol, and the maximum value x_max = 100% Vol. Substituting these values ​​into the standardization formula x' = (x - x_min) / (x_max - ... The standardized data x' = (16-0) / (100-0) = 0.16 was calculated using x_min. Other types of data (such as leaf nitrogen content and air temperature) were also standardized using the same method. Finally, the preprocessed data was transmitted to the edge computing gateway via the MQTT protocol. The actual transmission latency was 0.9 seconds, which meets the requirement of ≤1 second transmission and provides high-quality data input for subsequent algorithm optimization.

[0032] The third step is to optimize the control parameters of the integrated water and fertilizer machine using an improved non-dominated sorting genetic algorithm III (NSGA-III). This improved algorithm includes three core technical solutions. The first step is to introduce an adaptive crossover and mutation operator to improve convergence speed. The specific calculation method needs to be determined based on the fitness data of the current population. In this embodiment, the initial crossover probability P_c0 is set to 0.8 (within the range of 0.7-0.9), the initial mutation probability P_m0 is set to 0.03 (within the range of 0.01-0.05), and the adjustment coefficient k is set to 0.2 (within the range of 0.1-0.3). The calculated average fitness of the current population is f_avg=0.72, the maximum fitness is f_max=0.90, and the minimum fitness is f_min=0.55. Substituting these parameters into the crossover probability formula P_c = P_c0 × exp (-|f_avg - f_max| / (k×f_max)), we get P_c=0.8×exp (-|0.72-0.90| / (0.2×0.90))=0.8×exp (-0.18 / 0.18)=0.8×exp (-1)≈0.8×0.37≈0.29; Substituting into the mutation probability formula P_m = P_m0 × exp (-|f_avg - f_min| / (k×f_min)), we get P_m=0.03×exp (-|0.72-0.55| / (0.2×0.55))=0.03×exp (-0.17 / 0.11)≈0.03×exp (-1.55)≈0.03×0.21≈0.006. By adaptively adjusting the crossover and mutation probabilities, the convergence efficiency of the algorithm is effectively improved, avoiding getting trapped in local optima. The second aspect is the addition of a constraint penalty mechanism to eliminate invalid solutions. This mechanism is implemented by improving the fitness function, which is F'(x) = F(x) -λ×∑max(0, g_i(x)). The original fitness function F(x) is calculated based on a four-dimensional index system, and the penalty coefficient λ is set to 30 (within the range of 10-50). The constraint function g_i(x) follows the rule that g_i(x) = 0 when the control parameter x satisfies the constraint, and g_i(x) is the absolute value of the parameter deviating from the constraint boundary when it does not. In this embodiment, the constraints include: solid water-soluble fertilizer concentration ≤ 30%, liquid fertilizer dilution concentration ≥ 5% (liquid fertilizer is used in this scenario, so the dilution concentration must meet the requirements), and corn planting uses sprinkler irrigation, therefore the irrigation flow rate must be ≤ 50 L / h. m², and the soil moisture must be ≤ 80% of the field capacity (i.e., 28% Vol × 80% = 22.4% Vol). After verification, all parameters in this batch of optimization process meet the above constraints. Therefore, g_i (x) = 0. The improved fitness function F'(x) has the same value as the original fitness function F (x). If the parameters exceed the constraint range, the penalty mechanism will remove invalid solutions by reducing the fitness function value to ensure parameter compliance. The third item is the addition of an "environmental factor compensation module". When the environmental sensor detects an air temperature > 35℃, the algorithm will automatically increase the irrigation flow rate by 10%-15% and shorten the fertilization time by 20%-25% based on the optimized value to compensate for the loss of fertilizer efficiency due to high temperature. When the 24-hour rainfall is detected to be > 10 mm, the algorithm will increase the fertilizer concentration by 5%-8% based on the optimized value to avoid nutrient loss caused by rainwater leaching. The compensation range is calculated using the formula ΔP = α×ΔT + β×R, where ΔP is the parameter compensation ratio, α is the temperature influence coefficient (value 0.005-0.01, 0.008 in this embodiment), ΔT is the difference between the actual temperature and 30℃, β is the rainfall influence coefficient (value 0.008-0.012, 0.01 in this embodiment), and R is the 24-hour rainfall (mm). In this batch of environmental data, the temperature is 26℃ and the rainfall is 0 mm, so there is no need to trigger the compensation mechanism. The algorithm outputs parameters according to the basic optimization logic.

[0033] The fourth step is the determination and output of the optimal control parameter combination. The edge computing gateway inputs the preprocessed multi-source data, maize variety parameters (such as a growth cycle of 115 days and critical fertilizer requirements at different growth stages), and basic soil parameters (field water holding capacity of sandy loam and initial nitrogen, phosphorus, and potassium content) into the improved NSGA-Ⅲ algorithm. After iterative calculation, the algorithm outputs the optimal control parameter combination. The output process must also meet the "crop type adaptation rule"—since maize is a deep-rooted crop, the fertilization duration and stirring frequency must be set to 60-70 r / min according to the rule. The final output parameters are: fertilizer concentration (nitrogen, phosphorus, and potassium ratio 3:1:2) 12% (liquid fertilizer, meeting the constraint of ≥5%), irrigation flow rate 42 L / h. m² (sprinkler irrigation mode, meeting ≤50 L / h) With constraints of m² and a stirring frequency of 65 r / min (which meets the requirements of 60-70 r / min for deep-rooted crops), this parameter combination can simultaneously meet the optimization objectives of the four-dimensional index system, taking into account nutrient supply, resource conservation, and environmental adaptation.

[0034] The fifth step is the execution of fertilization. The edge computing gateway transmits the above optimal control parameter combination to the PLC controller via the industrial bus. After receiving the parameters, the PLC controller drives the various execution components of the integrated water and fertilizer machine to work together according to the preset logic: First, it controls the fertilizer mixing pump to prepare nitrogen, phosphorus, and potassium liquid fertilizer at a concentration of 12%, and at the same time starts the stirring motor to stir the fertilizer solution at a frequency of 65 r / min to ensure uniform fertilizer composition; after stirring is completed, the PLC controller opens the solenoid valve of the sprinkler system at 42 L / h. The irrigation flow rate of m² is used to start the sprinkler irrigation operation, and the fertilizer channel is opened at the same time to deliver the fertilizer solution to the corn root zone along with the irrigation water. During the operation, the sensors continue to collect data in real time. If the soil moisture is found to be close to 22.4% Vol (80% of field capacity) or the fertilizer solution concentration fluctuates, the PLC controller will dynamically fine-tune the irrigation flow rate or fertilizer pump speed according to the feedback data to ensure that the operation accuracy meets the requirements.

[0035] The sixth step is self-learning and iterative optimization. After the fertilization operation is completed, the system automatically records the "control parameter combination - actual effect data". The actual effect data includes: the nitrogen content of corn leaves increased to 2.95%, the plant height increased to 19.8cm, the leaf area index increased to 0.59, the water and fertilizer resource utilization rate reached 83%, the soil pH value was maintained in the range of 6.2-6.4, and the environmental adaptability evaluation was 0.94. The error between the actual effect and the target index was then calculated. The average error for this batch was 2.5%, which meets the condition that the average error of three consecutive fertilization operations is ≤3%. Therefore, according to the rules, the update frequency of the algorithm training set was reduced to once every 7 days. If a single operation error >8% occurs in the future, the system will automatically trigger an emergency update, re-call the improved NSGA-Ⅲ algorithm to train and optimize the parameters, and verify the optimized parameters through small-scale experiments. The next fertilization operation will be executed only after the verification is successful. During the update of the training set, a weighted average method is used to retain historical effective data. The weight of historical data gradually decreases over time with a decay coefficient of 0.9 / month. For example, the weight of historical data from 3 months ago is 0.9×0.9×0.9=0.729. This ensures that recent effective data plays a dominant role in algorithm optimization, while avoiding the loss of experience caused by completely discarding historical data. Through continuous self-learning iteration, the adaptability of the algorithm to this corn planting area will gradually improve, and the parameter optimization accuracy will also continuously improve.

[0036] Furthermore, the method of this invention also includes an "active soil pH adjustment step." Before a fertilization operation, if the soil in-situ sensor detects a soil pH value of 5.3 < 5.5, the PLC controller will automatically trigger the operation of the acid-base regulator addition device: First, the amount of calcium carbonate solution to be added is calculated according to the formula V = (5.5 - pH_actual) × S × k1, where pH_actual = 5.3 (actual soil pH value), S = 2000 m² (fertilization area), and k1 is set to 0.025 (within the range of 0.02-0.03). Substituting these values ​​into the formula, we get V = (5.5 - 5.3) × 2000 × 0.025 = 10 L. Then, the total volume of the fertilizer solution is detected to be 200 L, and the addition amount is calculated to be 10 L / 200 L = 5%, which does not exceed the 5% limit of the total volume of the fertilizer solution. Therefore, the PLC controller drives the addition device to inject 10 L of fertilizer solution into the fertilizer solution. After thoroughly mixing the calcium carbonate solution, apply the fertilizer. Adjust the soil pH to stabilize at 5.6, restoring it to the suitable range. If the soil pH is detected to be >7.5, calculate the amount of phosphoric acid solution to be added using the formula V = (pH_actual - 7.5)×S×k2 (k2 is 0.015-0.025). Similarly, control the amount added to not exceed 5% of the total fertilizer solution volume. Actively adjust to ensure that the soil physicochemical balance meets the optimized index requirements.

[0037] This embodiment, through the implementation of the complete process described above, effectively solves the problems of traditional integrated water and fertilizer machines, such as single fertilization control targets, lagging parameter adjustments, and poor adaptability. In the application during the corn seedling stage, the crop nutrient precision supply rate is increased by 21% compared to traditional methods, the water and fertilizer resource utilization rate is increased by 28%, and the soil pH stability rate reaches 97%. Moreover, no manual parameter adjustments are required, significantly reducing labor intensity. At the same time, this method has strong adaptability. If the scenario is switched to facility agriculture cucumber planting (shallow-rooted crop), it is only necessary to preset the crop type through the user terminal or for the system to automatically determine it as a shallow-rooted crop through leaf morphology image recognition. The parameters can then be output according to the adaptation rules for shallow-rooted crops without large-scale adjustments to the core algorithm and hardware architecture. If applied to orchard planting scenarios, only the deployment density and detection parameters of the sensors need to be adjusted to achieve precise fertilization. This fully demonstrates the adaptability of this invention to various planting scenarios and provides strong technical support for the development of precision agriculture.

[0038] It should be understood that the present invention is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. A method for optimizing multi-objective dynamic precision fertilization control parameters in an integrated water and fertilizer machine, characterized in that: Includes the following steps: Step 1: Construct a four-dimensional, multi-objective optimization index system, which includes crop nutrient precision supply rate, water and fertilizer resource utilization rate, soil physicochemical balance, and environmental adaptability. Step 2: Deploy a multi-source sensor array to collect three types of data in real time: soil moisture, pH value, EC value and nitrogen, phosphorus and potassium mass fraction collected by in-situ soil sensors; crop leaf nitrogen content, plant height and leaf area index collected by spectral crop sensors combined with image recognition; and air temperature, humidity, light intensity and rainfall collected by environmental sensors. Step 3: The control parameters of the integrated water and fertilizer machine are optimized using the improved non-dominated sorting genetic algorithm III (NSGA-III). The improved NSGA-III algorithm includes: introducing an adaptive crossover and mutation operator to improve the convergence speed, and adding a constraint penalty mechanism to eliminate invalid solutions. Step 4: Preprocess the data collected in Step 2 through the edge computing gateway, input the preprocessed data, crop variety parameters, and basic soil parameters into the improved NSGA-Ⅲ algorithm, and output the optimal combination of control parameters, including fertilizer ratio concentration, irrigation flow rate, fertilization duration, and stirring frequency. Step 5: Transmit the optimal control parameter combination to the PLC controller, which then drives the integrated water and fertilizer machine to perform fertilization operations. Step 6: After the task is completed, record the "control parameter combination - actual effect data". If the error between the actual effect and the target index exceeds the preset threshold, update the algorithm training set to achieve self-learning iterative optimization.

2. The method for optimizing multi-objective dynamic precision fertilization control parameters of an integrated water and fertilizer machine according to claim 1, characterized in that, The soil in-situ sensor mentioned in step 2 is an integrated multi-parameter sensor with an N / P / K detection accuracy of ≤±5% and a soil moisture detection range of 0-100% Vol with a detection accuracy of ≤±2% Vol. The spectral crop sensor has a detection wavelength range of 400-1000 nm and a leaf nitrogen content detection error of ≤±0.1%.

3. The method for optimizing multi-objective dynamic precision fertilization control parameters of an integrated water and fertilizer machine according to claim 1, characterized in that, The specific calculation method of the adaptive crossover and mutation operator in step 3 is as follows: crossover probability P_c = P_c0 ×exp(-|f_avg - f_max| / (k×f_max)), mutation probability P_m = P_m0 ×exp(-|f_avg - f_min| / (k×f_min)); where P_c0 is the initial crossover probability (value 0.7-0.9), P_m0 is the initial mutation probability (value 0.01-0.05), f_avg is the average fitness of the current population, f_max is the maximum fitness of the current population, f_min is the minimum fitness of the current population, and k is the adjustment coefficient (value 0.1-0.3).

4. The method for optimizing multi-objective dynamic precision fertilization control parameters of an integrated water and fertilizer machine according to claim 1, characterized in that, The constraint penalty mechanism described in step 3 is implemented by improving the fitness function. The improved fitness function is: F'(x) = F(x) - λ×∑max(0, g_i(x)); where F(x) is the original fitness function, λ is the penalty coefficient (value 10-50), and g_i(x) is the constraint function. When the control parameter x satisfies the constraint, g_i(x) = 0; when it does not satisfy the constraint, g_i(x) is the absolute value of the parameter deviating from the constraint boundary. The constraints include: solid water-soluble fertilizer concentration ≤ 30%, liquid fertilizer dilution concentration ≥ 5%, irrigation flow rate ≤ 20 L / h·m² in drip irrigation mode, irrigation flow rate ≤ 50 L / h·m² in sprinkler irrigation mode, and soil moisture ≤ 80% of field capacity.

5. The method for optimizing multi-objective dynamic precision fertilization control parameters of an integrated water and fertilizer machine according to claim 1, characterized in that, The weights of the four-dimensional multi-objective optimization index system described in step 1 can be dynamically adjusted: when the crop growth period is the seedling stage, the weight of the crop nutrient precision supply rate is set to 0.4, and the weights of the other indicators are each set to 0.2; when the growth period is the fruit expansion / grain filling stage, the weight of the crop nutrient precision supply rate is set to 0.5, the weight of the water and fertilizer resource utilization rate is set to 0.3, and the weights of the other indicators are each set to 0.1; the weight adjustment is triggered by the user terminal or the system automatically identifying the crop growth period.

6. The method for optimizing multi-objective dynamic precision fertilization control parameters of an integrated water and fertilizer machine according to claim 1, characterized in that, The specific rules for self-learning iterative optimization in step 6 are as follows: when the average error between the actual effect of three consecutive fertilization operations and the target index is ≤3%, the algorithm training set update frequency is reduced to once / 7 days; when the error of a single operation is >8%, an emergency update is triggered, the algorithm is retrained and the optimization parameters are verified, and the next fertilization operation is executed after the verification is passed; when updating the training set, the weighted average method is used to retain historical valid data, and the weight of historical data decays over time with a decay coefficient of 0.9 / month.

7. The method for optimizing multi-objective dynamic precision fertilization control parameters of an integrated water and fertilizer machine according to claim 1, characterized in that, The improved NSGA-Ⅲ algorithm described in step 3 also includes an "environmental factor compensation module": when the air temperature is >35℃, the algorithm automatically increases the irrigation flow rate by 10%-15% and shortens the fertilization time by 20%-25% to compensate for fertilizer volatilization loss; when the rainfall is >10 mm / 24h, the algorithm increases the fertilizer ratio concentration by 5%-8% to avoid nutrient loss caused by rainwater leaching. The compensation range is calculated using the formula ΔP = α×ΔT + β×R, where ΔP is the parameter compensation ratio, α is the temperature influence coefficient (value 0.005-0.01), ΔT is the difference between the actual temperature and 30℃, β is the rainfall influence coefficient (value 0.008-0.012), and R is the 24-hour rainfall (mm).

8. The method for optimizing multi-objective dynamic precision fertilization control parameters of an integrated water and fertilizer machine according to claim 1, characterized in that, The output of the optimal control parameter combination described in step 4 must also meet the "crop type adaptation rule": for shallow-rooted crops, the fertilization time should be controlled at 20-30 minutes and the stirring frequency should be set at 40-50 r / min; for deep-rooted crops, the fertilization time should be controlled at 40-60 minutes and the stirring frequency should be set at 60-70 r / min; the crop type is preset by the user or automatically determined by the system through leaf morphology image recognition.

9. The method for optimizing multi-objective dynamic precision fertilization control parameters of an integrated water and fertilizer machine according to claim 1, characterized in that, It also includes a "soil pH active adjustment step": when the soil in-situ sensor detects a soil pH value < 5.5 in step 2, the PLC controller drives the acid-base regulator addition device to inject calcium carbonate solution into the fertilizer solution. The addition amount is V = (5.5 - pH_actual) × S × k1, and V does not exceed 5% of the current total volume of fertilizer solution; Where V is the volume of calcium carbonate solution (L), pH_actual is the actual soil pH value, S is the fertilization area (m²), and k1 is the adjustment coefficient (value 0.02-0.03); when the soil pH value is >7.5, phosphate solution is injected, and the amount added is V = (pH_actual - 7.5)×S×k2, where k2 is the adjustment coefficient (value 0.015-0.025).

10. The method for optimizing multi-objective dynamic precision fertilization control parameters of an integrated water and fertilizer machine according to claim 1, characterized in that, The data preprocessing described in step 4 includes: removing outliers using the 3σ criterion (data is considered outliers when it deviates from the mean by more than 3 times the standard deviation), mapping the data to the [0,1] interval using Min-Max standardization, with the standardization formula being x' = (x - x_min) / (x_max - x_min), where x is the original data, x_min is the minimum value of the data in this class, and x_max is the maximum value of the data in this class; the preprocessed data is transmitted to the edge computing gateway via the MQTT protocol.

Images