A Smart Decision-Making Method for Maize Water and Fertilizer Management in Arid Zones Based on Radial Basis Neural Networks

Abstract

This invention discloses an intelligent water and fertilizer decision-making method for maize in arid regions based on radial basis function neural networks (RBNs), comprising the following steps: S1, distributed data acquisition; S2, construction of a RBN decision model; S3, real-time water and fertilizer decision generation; S4, execution of multi-mode water and fertilizer application; and S5, closed-loop feedback adjustment. The advantages of this invention compared to existing technologies are: by using a RBN to construct the water and fertilizer decision model, it can effectively fit the complex nonlinear relationship between multiple parameters such as soil moisture and fertility and the water and fertilizer requirements of maize. Compared with traditional linear empirical models, it significantly improves the accuracy of water and fertilizer decision-making for maize in arid regions and reduces water and fertilizer waste or insufficient supply caused by model bias.

Description

Technical Field

[0001] This invention relates to the field of intelligent water and fertilizer management technology in agriculture, specifically to an intelligent decision-making method for water and fertilizer management of maize in arid areas based on radial basis function neural networks. Background Technology

[0002] In arid regions, the precision of water and fertilizer application directly impacts crop yield and water resource utilization efficiency during maize cultivation. Traditional water and fertilizer decisions often rely on empirical formulas or simple linear models, which struggle to accurately depict the complex nonlinear coupling relationship between soil fertility, soil moisture, environmental temperature and humidity, and maize yield. This leads to significant biases in decision-making, easily resulting in excessive or insufficient water and fertilizer application.

[0003] On the other hand, traditional integrated water and fertilizer machines mostly adopt a crude application method with fixed time periods and fixed amounts, which cannot dynamically adjust the water and fertilizer ratio and application amount according to real-time soil parameters in the field. The precision of water and fertilizer application is insufficient, which not only wastes water and fertilizer resources, but also fails to meet the differentiated needs of corn at different growth stages. At the same time, the existing system has a single irrigation mode and lacks the ability to remotely and flexibly control it, which restricts the improvement of the intelligence level of modern agricultural machinery and equipment.

[0004] Radial basis function (RBF) neural networks possess strong nonlinear mapping capabilities, enabling them to approximate complex nonlinear functions with arbitrary precision. This makes them suitable for handling multi-input, strongly coupled, and nonlinear water and fertilizer decision-making problems in agricultural systems. Combining RBF neural networks with distributed sensing and multi-mode intelligent execution devices to construct an intelligent water and fertilizer decision-making system for maize in arid regions can effectively improve decision accuracy and application precision, achieving on-demand irrigation and fertilization. Summary of the Invention

[0005] The technical problem to be solved by this invention is to provide an intelligent decision-making method for water and fertilizer management of maize in arid areas based on radial basis function neural networks.

[0006] To solve the above-mentioned technical problems, the technical solution provided by this invention is: a smart decision-making method for water and fertilizer management of maize in arid areas based on radial basis function neural networks, comprising the following steps: S1. Distributed data acquisition: Multiple sensor nodes are deployed in cornfields in arid areas to collect soil moisture, soil fertility, conductivity of applied liquid, pH value, and environmental temperature and humidity parameters in real time, and transmit them to the main control unit. S2. Construction of Radial Basis Neural Network Decision Model: Using soil moisture, soil fertility, liquid EC value, pH value, and ambient temperature and humidity as input variables, and the optimal water and fertilizer application rate for maize as the output variable, a radial basis neural network model is constructed; historical field measurement data of maize in arid areas are used to train the network and determine the network parameters. S3. Real-time water and fertilizer decision generation: The main control unit inputs the field parameters collected in real time into the trained radial basis neural network model and outputs the recommended irrigation water and fertilizer amount for maize at the corresponding growth stage. S4. Multi-mode water and fertilizer application execution: The intelligent water and fertilizer application device receives decision instructions and uses three working modes: intelligent automatic irrigation, timed automatic irrigation, and mobile APP remote control irrigation. According to the decision results, it accurately adjusts the irrigation duration, flow rate and fertilizer solution ratio to complete the integrated water and fertilizer application. S5. Closed-loop feedback adjustment: After application, distributed sensors continuously monitor changes in soil parameters and re-input the feedback data into the decision-making model to dynamically correct the next water and fertilizer decision.

[0007] Furthermore, in step S1, the distributed sensor nodes include soil moisture sensors, soil nitrogen, phosphorus and potassium sensors, EC sensors, pH sensors, and air temperature and humidity sensors. Each sensor is connected to the main control unit through a wireless communication module, and the acquisition frequency can be dynamically adjusted according to the corn growth period.

[0008] Furthermore, in step S2, the radial basis neural network adopts a three-layer network structure. The input layer corresponds to the field environment and soil parameters, the hidden layer uses radial basis functions for nonlinear transformation, and the output layer outputs two types of decision values: irrigation water amount and fertilizer amount.

[0009] Furthermore, in step S2, the model training process uses unsupervised learning to determine the basis function center and supervised learning to solve the output weights. The training samples cover field measured data and corresponding yield data of maize at different growth stages.

[0010] Furthermore, in step S4, under the intelligent automatic irrigation mode, the system automatically starts and stops the irrigation equipment and adjusts the water-fertilizer ratio based on the decision results output in real time by the radial basis neural network, and automatically triggers irrigation when the soil parameters are lower than the appropriate threshold.

[0011] Furthermore, in step S4, under the timed automatic irrigation mode, the user presets the irrigation period and the basic application range. The system then fine-tunes the actual application amount within the preset period by combining real-time soil parameters, taking into account both management habits and precise application.

[0012] Furthermore, in step S4, under the mobile APP remote irrigation mode, the user remotely issues irrigation, pause, or dosage adjustment commands through a mobile terminal, and the system executes them instantly and transmits back the equipment operating status and field parameters.

[0013] Furthermore, in step S4, the intelligent water and fertilizer application device monitors the output liquid parameters in real time through EC sensors and pH sensors during irrigation, dynamically fine-tunes the fertilizer solution intake, and ensures that the water and fertilizer concentration meets the decision-setting value.

[0014] Furthermore, in step S5, the closed-loop feedback adjustment evaluates the application effect based on the soil moisture and fertility change data within 24 hours after application, and automatically corrects the model output correction coefficient to ensure that the decision-making continuously adapts to the actual field conditions.

[0015] Furthermore, the radial basis function neural network model trains corresponding sub-models for different growth stages of maize, and the system automatically switches and calls them according to the growth stage to ensure the pertinence of decisions for each growth stage.

[0016] With the above structure, the present invention has the following advantages: 1. The water and fertilizer decision model is constructed by using radial basis neural network, which can effectively fit the complex nonlinear relationship between soil moisture, fertility and other parameters and maize water and fertilizer requirements. Compared with the traditional linear empirical model, it significantly improves the accuracy of maize water and fertilizer decision-making in arid areas and reduces water and fertilizer waste or insufficient supply caused by model bias.

[0017] 2. By collecting key parameters such as soil moisture, liquid EC, and pH value in real time through distributed multi-sensors, and combining them with the precise flow control of intelligent water and fertilizer application devices, the transformation of water and fertilizer application from extensive to intensive can be realized, meeting the on-demand supply of corn at different growth stages and improving the efficiency of fertilizer and water resource utilization.

[0018] 3. It integrates three irrigation modes: intelligent automatic, timed automatic, and mobile APP remote control, taking into account the needs of automated operation and flexible manual intervention, adapting to different field management scenarios, and improving the intelligence and ease of use of modern agricultural machinery and equipment.

[0019] 4. Introduce a closed-loop feedback adjustment mechanism to continuously optimize decision output using soil monitoring data after application, enabling the model to adapt to changes in the field environment and maintain stable decision accuracy over the long term. Detailed Implementation

[0020] The present invention will now be described in further detail.

[0021] A smart decision-making method for water and fertilizer management of maize in arid areas based on radial basis function neural networks includes the following steps: S1. Distributed data acquisition: Multiple sensor nodes are deployed in cornfields in arid areas to collect soil moisture, soil fertility, conductivity of applied liquid, pH value, and environmental temperature and humidity parameters in real time, and transmit the collected data to the main control unit. S2. Construction of Radial Basis Neural Network Decision Model: Using collected soil moisture, soil fertility, liquid EC value, pH value, and ambient temperature and humidity as input variables, and the optimal water and fertilizer application rate per maize plant as the output variable, a radial basis neural network model is constructed; historical field data of maize in arid areas are used to train the network and determine the network center, width, and output weight parameters. S3. Real-time water and fertilizer decision generation: The main control unit inputs the field parameters collected in real time into the trained radial basis neural network model and outputs the recommended irrigation water and fertilizer amount for maize at the corresponding growth stage. S4. Multi-mode water and fertilizer application execution: The intelligent water and fertilizer application device receives decision instructions and supports three working modes: intelligent automatic irrigation, timed automatic irrigation, and mobile APP remote control irrigation. It accurately adjusts the irrigation duration, flow rate and fertilizer solution ratio according to the decision results to complete the integrated water and fertilizer application. S5. Closed-loop feedback adjustment: After application, distributed sensors continuously monitor changes in soil parameters and re-input the feedback data into the decision-making model to dynamically correct the next water and fertilizer decision.

[0022] The present invention will be further described below with reference to the embodiments, but the scope of protection of the present invention is not limited to the following embodiments.

[0023] Example 1: Water and fertilizer decision-making for maize during the jointing stage in arid areas under intelligent automatic irrigation mode This embodiment is applied to field management of maize during the jointing stage in the arid Northwest region. The specific implementation process is as follows: In a corn experimental field of approximately 30 mu (about 2 hectares), six sets of distributed sensor nodes were deployed at a density of one set per 5 mu (about 0.3 hectares). Each set of nodes integrated a soil moisture sensor, a soil nitrogen, phosphorus, and potassium sensor, an EC (extracorporeal membrane oxygenation) sensor, a pH sensor, and an air temperature and humidity sensor, with a data acquisition frequency set to once every 30 minutes. All sensors transmitted data to the main control unit in the field control cabinet via a LoRa wireless module.

[0024] The main control unit incorporates a pre-trained radial basis function neural network decision-making model. The model's input variables include eight parameters: soil volumetric moisture content, soil available nitrogen content, soil available phosphorus content, soil available potassium content, irrigation solution EC value, irrigation solution pH value, air temperature, and air humidity. The output variables are the amount of water used in a single irrigation and the amount of nitrogen, phosphorus, and potassium fertilizer applied. The model training samples consist of field-measured data from the maize jointing stage and corresponding yield data from the past three years in this region. After training, the model's decision error was controlled within a reasonable range.

[0025] The system operates in intelligent automatic irrigation mode. When the main control unit detects that the soil volumetric moisture content is lower than the suitable threshold for corn jointing, it automatically triggers the decision-making process. The eight input parameters at the current moment are input into the radial basis function neural network model, and after nonlinear mapping calculation, the recommended irrigation water volume and nitrogen, phosphorus, and potassium fertilizer ratio are output.

[0026] The main control unit sends decision commands to the intelligent water and fertilizer application device. Based on these commands, the device adjusts the opening of the inlet solenoid valve, the ratio of the fertilizer absorption channels, and the pressure of the main irrigation network to precisely control the concentration and total flow rate of the water-fertilizer mixture. During irrigation, the EC sensor and pH sensor monitor the output liquid parameters in real time. If the parameters deviate from the set decision range, the device dynamically fine-tunes the fertilizer absorption rate to ensure application accuracy.

[0027] After irrigation, sensors continuously monitor changes in soil moisture and fertility over 24 hours, feeding the data back to the decision-making model. The model automatically adjusts its output correction coefficients based on the actual changes in soil parameters, which are then used for the next decision calculation, forming a closed-loop optimization.

[0028] Example 2: Water and fertilizer management during the corn tasseling stage using timed automatic irrigation combined with remote control via mobile app This embodiment is applied to water and fertilizer management during the tasseling stage of maize in arid areas, and adopts a working mode that mainly uses timed automatic irrigation and is supplemented by remote control via mobile APP.

[0029] Similar to Example 1, the field-distributed data collection device increases the sensor collection frequency to once every 15 minutes, taking into account the high water and fertilizer demand during the corn tasseling stage. Users can preset two irrigation periods each day, morning and evening, and set the basic irrigation water and fertilizer application ranges via a mobile app.

[0030] Thirty minutes before the preset irrigation period, the main control unit calls the radial basis function neural network model to calculate the optimal water and fertilizer application rate based on the current soil moisture, fertility and environmental parameters. It automatically fine-tunes the actual application value within the user-defined basic range, ensuring both the management habit of scheduled operation and the dynamic optimization of application rate.

[0031] During irrigation, users can view field soil parameters, equipment operating status, and the progress of water and fertilizer application in real time via a mobile app. In case of sudden weather changes or special field conditions, users can remotely pause irrigation, adjust the dosage, or issue an additional irrigation command via the app. Upon receiving a remote control command from the app, the system prioritizes executing the manual intervention command and transmits the execution result back to the app interface in real time.

[0032] After each irrigation, the system automatically generates an application record and stores it in the historical database. The radial basis function neural network model periodically uses newly added historical data for incremental learning, continuously improving the adaptability of the decision-making model to local soil and climate conditions.

[0033] Example 3: Validation and application of water and fertilizer decision-making throughout the entire growth period of maize in arid areas This embodiment conducts full-growth-cycle verification in a contiguous maize planting base in an arid region, covering six main growth stages: sowing, seedling, jointing, tasseling, grain filling, and maturity.

[0034] Within the base, experimental and control groups were divided. The experimental group adopted the radial basis function neural network intelligent decision-making method of this invention, while the control group adopted the local traditional experience-based water and fertilizer management method. Identical distributed data acquisition devices were deployed in both groups' fields. Data from the experimental group was used for real-time decision-making, while data from the control group was only used for recording and comparison.

[0035] For different growth stages, the radial basis function neural network model uses training samples corresponding to each growth stage to model separately, ensuring that the decisions at each stage are targeted. Throughout the entire growth period, the system automatically switches the corresponding decision model according to the water and fertilizer requirements of maize at each stage. During the seedling stage, the system focuses on conserving moisture and promoting root growth to control the amount of water and fertilizer. During the jointing and tasseling stages, the supply intensity is increased. During the grain filling stage, the appropriate water and fertilizer levels are maintained. During the maturity stage, the supply is gradually reduced.

[0036] Throughout the entire reproductive cycle, the system supports flexible switching between three irrigation modes: intelligent automatic mode for daily operation; timed automatic mode for critical reproductive periods to ensure supply rhythm; and remote control mode via mobile APP for on-site dispatching by management personnel when they are out on patrol.

[0037] During the harvest period, yield comparison showed that the experimental group had a higher corn yield than the control group, while the amount of irrigation water and fertilizer used was lower than that of the control group. This verified the practical effect of the method of the present invention in improving decision-making accuracy and increasing water and fertilizer utilization efficiency.

[0038] 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 smart decision-making method for water and fertilizer management of maize in arid regions based on radial basis function neural networks, characterized in that, Includes the following steps: S1. Distributed data acquisition: Multiple sensor nodes are deployed in cornfields in arid areas to collect soil moisture, soil fertility, conductivity of applied liquid, pH value, and environmental temperature and humidity parameters in real time, and transmit them to the main control unit. S2. Construction of Radial Basis Neural Network Decision Model: Using soil moisture, soil fertility, liquid EC value, pH value, and ambient temperature and humidity as input variables, and the optimal water and fertilizer application rate for maize as the output variable, a radial basis neural network model is constructed; historical field measurement data of maize in arid areas are used to train the network and determine the network parameters. S3. Real-time water and fertilizer decision generation: The main control unit inputs the field parameters collected in real time into the trained radial basis neural network model and outputs the recommended irrigation water and fertilizer amount for maize at the corresponding growth stage. S4. Multi-mode water and fertilizer application execution: The intelligent water and fertilizer application device receives decision instructions and uses three working modes: intelligent automatic irrigation, timed automatic irrigation, and mobile APP remote control irrigation. According to the decision results, it accurately adjusts the irrigation duration, flow rate and fertilizer solution ratio to complete the integrated water and fertilizer application. S5. Closed-loop feedback adjustment: After application, distributed sensors continuously monitor changes in soil parameters and re-input the feedback data into the decision-making model to dynamically correct the next water and fertilizer decision.

2. The intelligent decision-making method for water and fertilizer management of maize in arid areas based on radial basis function neural networks according to claim 1, characterized in that, In step S1, the distributed sensor nodes include soil moisture sensors, soil nitrogen, phosphorus and potassium sensors, EC sensors, pH sensors, and air temperature and humidity sensors. Each sensor is connected to the main control unit through a wireless communication module, and the acquisition frequency can be dynamically adjusted according to the corn growth period.

3. The intelligent decision-making method for water and fertilizer management of maize in arid areas based on radial basis function neural networks according to claim 1, characterized in that, In step S2, the radial basis function neural network adopts a three-layer network structure. The input layer corresponds to the field environment and soil parameters, the hidden layer uses radial basis functions for nonlinear transformation, and the output layer outputs two types of decision values: irrigation water amount and fertilizer amount.

4. The intelligent decision-making method for water and fertilizer management of maize in arid areas based on radial basis function neural networks according to claim 1, characterized in that, In step S2, the model training process uses unsupervised learning to determine the basis function center and supervised learning to solve the output weights. The training samples cover field measured data and corresponding yield data of maize at different growth stages.

5. The intelligent decision-making method for water and fertilizer management of maize in arid areas based on radial basis function neural networks according to claim 1, characterized in that, In step S4, under the intelligent automatic irrigation mode, the system automatically starts and stops the irrigation equipment and adjusts the water-fertilizer ratio based on the decision results output in real time by the radial basis neural network. When the soil parameters are lower than the appropriate threshold, irrigation is automatically triggered.

6. The intelligent decision-making method for water and fertilizer management of maize in arid areas based on radial basis function neural networks according to claim 1, characterized in that, In step S4, under the timed automatic irrigation mode, the user presets the irrigation period and basic application range. The system then fine-tunes the actual application amount within the preset period by combining real-time soil parameters, taking into account both management habits and precise application.

7. The intelligent decision-making method for water and fertilizer management of maize in arid areas based on radial basis function neural networks according to claim 1, characterized in that, In step S4, under the mobile APP remote irrigation mode, the user remotely issues irrigation, pause, or dosage adjustment commands through a mobile terminal, and the system executes them in real time and transmits back the equipment operating status and field parameters.

8. The intelligent decision-making method for water and fertilizer management of maize in arid areas based on radial basis function neural networks according to claim 1, characterized in that, In step S4, the intelligent water and fertilizer application device monitors the output liquid parameters in real time through EC sensor and pH sensor during irrigation, dynamically fine-tunes the fertilizer solution intake, and ensures that the water and fertilizer concentration meets the decision setting value.

9. The intelligent decision-making method for water and fertilizer management of maize in arid areas based on radial basis function neural networks according to claim 1, characterized in that, In step S5, the closed-loop feedback adjustment evaluates the application effect based on the soil moisture and fertility change data within 24 hours after application, and automatically corrects the model output correction coefficient to ensure that the decision-making continuously adapts to the actual field conditions.

10. A method for intelligent water and fertilizer decision-making for maize in arid regions based on radial basis function neural networks according to claim 1, characterized in that, The radial basis function neural network model trains corresponding sub-models for different growth stages of maize, and the system automatically switches and calls them according to the growth stage to ensure the targeted nature of decisions for each growth stage.