Intelligent irrigation control system and method based on meteorological environment acquisition

Abstract

The invention relates to the technical field of agricultural intelligent irrigation, in particular to an intelligent irrigation control system and method based on meteorological environment acquisition, and the system comprises a sensing layer, a transmission layer, a control layer, a cloud layer and an execution layer. Then meteorological environment data collected by the sensing layer is transmitted to the control layer through a wireless transmitting module and a wireless receiving module, and a controller of the control layer is integrated with an edge computing unit, a communication unit and a decision-making unit; the problems that an existing agricultural meteorological monitoring and irrigation control system is high in network transmission load coefficient, and the edge computing capability of sensor nodes cannot be effectively utilized; the problem that an irrigation control algorithm based on a static threshold mechanism or a single statistical model lacks adaptive processing of rainfall time sequence prediction and dynamic response ability of specific requirements of crop growth stages is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of agricultural intelligent irrigation, and specifically to an intelligent irrigation control system and method based on meteorological environment collection. Background Art

[0002] In modern agricultural production, the monitoring of meteorological environment and irrigation control are crucial for improving crop yields and optimizing resource utilization efficiency. Existing agricultural meteorological monitoring and irrigation control systems form a complete closed-loop of data collection, processing, and response through the collaborative work of a series of structured components. First, meteorological data collection terminals distributed in farmlands, including various sensor nodes, collect key meteorological parameters in real time. Next, through the "edge-side collection-cloud processing" architecture, the data collected by sensor nodes is transmitted to the cloud for centralized processing. On the cloud or local server, irrigation control algorithms based on static threshold mechanisms or single statistical models analyze the received data to make irrigation decisions. In addition, the system also relies on regional forecast data provided by meteorological departments to assist in decision-making. Finally, according to the above analysis results, the irrigation control system will execute corresponding irrigation operations.

[0003] However, the existing agricultural meteorological monitoring and irrigation control systems have the following limitations: First, the "edge-side collection-cloud processing" mode adopted by traditional meteorological collection terminals leads to a high network transmission load coefficient and fails to effectively utilize the edge computing capabilities of sensor nodes, thus increasing system energy consumption and latency. Second, current mainstream irrigation control algorithms are generally based on static threshold mechanisms or single statistical models, lacking adaptive processing for precipitation time series prediction and dynamic response capabilities for the specific needs of crop growth stages. In addition, relying on regional forecast data provided by meteorological departments, its spatial resolution is low and it is difficult to accurately reflect the specific microclimate characteristics of farmlands, limiting the accuracy of irrigation decisions.

[0004] Therefore, it is necessary to invent an intelligent irrigation control system and method based on meteorological environment collection to solve the above problems. Summary of the Invention

[0005] In order to solve the problems existing in the existing agricultural meteorological monitoring and irrigation control systems, such as high network transmission load coefficient, failure to effectively utilize the edge computing capabilities of sensor nodes, lack of adaptive processing for precipitation time series prediction and dynamic response capabilities for the specific needs of crop growth stages in irrigation control algorithms based on static threshold mechanisms or single statistical models, and low spatial resolution of relying on regional forecast data and being difficult to accurately reflect the specific microclimate characteristics of farmlands, the present invention provides an intelligent irrigation control system and method based on meteorological environment collection.

[0006] The present invention is implemented by adopting the following technical solutions:

[0007] An intelligent irrigation control method based on meteorological environment collection, and the method is implemented by the following steps:

[0008] S1. Collect historical meteorological environment data of the target area, and construct and train a lightweight LSTM model;

[0009] S2. Predict the rainfall on the t-th day in the future through the trained lightweight LSTM model, where t is a positive integer;

[0010] S3. Call the predicted rainfall on the t-th day in the future obtained by the meteorological system API, and perform data fusion with the predicted rainfall on the t-th day in the future output by the lightweight LSTM model to generate the improved predicted rainfall on the t-th day in the future;

[0011] S4. Set corresponding soil moisture threshold ranges, optimal soil moisture, and irrigation times according to different growth stages of the target crop, and combine real-time soil moisture data and the improved predicted rainfall on the t-th day in the future to generate dynamic irrigation decision instructions to achieve intelligent irrigation.

[0012] Further, the construction and training of the lightweight LSTM model includes:

[0013] First, normalize the collected historical meteorological environment data and divide it into a training set and a test set;

[0014] Then, use the Keras high-level API in TensorFlow to construct a lightweight LSTM model, define the architecture of the lightweight LSTM model, select the mean squared error MSE as the loss function, compile the model using the Adam optimizer, perform network training on the training set, and evaluate the performance of the model on the test set after training;

[0015] Finally, use TensorFlow Lite to quantize the trained model to FP16 precision and import it into STM32Cube.AI to generate a C language static library.

[0016] Further, the data fusion is achieved through the following formula:

[0017] ;

[0018] Where:

[0019] : The improved predicted rainfall on the t-th day in the future;

[0020] : The predicted rainfall on the t-th day in the future output by the lightweight LSTM model;

[0021] : The predicted rainfall for the t-th day in the future obtained from the meteorological system API;

[0022] : The weight coefficient.

[0023] Furthermore, the weight coefficient is obtained through the following method:

[0024] First, the meteorological environment data for the current day and the past N days are collected in real time, where N is a positive integer greater than 1;

[0025] Then, the rainfall for the past N days is predicted through the trained lightweight LSTM model, and the error variance of the predicted rainfall output by the lightweight LSTM model for the past N days is calculated by comparing it with the rainfall for the past N days collected in real time; the predicted rainfall for the past N days obtained by calling the meteorological system API is used, and the error variance of the predicted rainfall obtained by the meteorological system API for the past N days is calculated by comparing it with the rainfall for the past N days collected in real time; the meteorological environment data for the past N days, the predicted rainfall data output by the lightweight LSTM model for the past N days, and the predicted rainfall data obtained by the meteorological system API for the past N days are all dynamically updated through the sliding window mechanism, and the coverage period of the sliding window is shifted backward day by day to ensure that there is no overlap between the historical data and the current prediction period;

[0026] Finally, the weight coefficient is calculated through the following weight coefficient formula :

[0027] ;

[0028] Where:

[0029] : The error variance of the predicted rainfall output by the lightweight LSTM model for the past N days;

[0030] : The error variance of the predicted rainfall obtained by the meteorological system API for the past N days;

[0031] In the initial stage, that is, when the historical data in the sliding window is less than N days, equal weight fusion is adopted, that is = 0.5. After the data in the sliding window accumulates to N days, the weight coefficient is dynamically updated according to the weight coefficient formula.

[0032] Furthermore, the meteorological environment data includes wind speed, wind direction, light intensity, air temperature, air humidity, soil moisture, and rainfall.

[0033] Furthermore, the dynamic irrigation decision includes:

[0034] If the soil humidity on the current day is equal to or higher than the lower limit of the set soil humidity threshold range, including being equal to or higher than the upper limit of the set soil humidity threshold range, irrigation is not initiated.

[0035] If the soil humidity on the current day is lower than the lower limit of the set soil humidity threshold range, a judgment is made based on the improved predicted rainfall for the next day and the improved predicted rainfall for the day after the next day.

[0036] Furthermore, the judgment based on the improved predicted rainfall for the next day and the improved predicted rainfall for the day after the next day includes:

[0037] If the improved predicted rainfall for the next day or the improved predicted rainfall for the day after the next day is moderate rain or above, i.e., ≥10 mm / day, irrigation is not initiated.

[0038] If the improved predicted rainfall for the next day or the improved predicted rainfall for the day after the next day is light rain, i.e., 0.1 - 9.9 mm / day, irrigation is initiated until the lower limit of the soil humidity threshold range.

[0039] If the improved predicted rainfall for the next day or the improved predicted rainfall for the day after the next day is no rain or trace rainfall, i.e., <0.1 mm / day, irrigation is initiated until the optimal soil humidity.

[0040] An intelligent irrigation control system based on meteorological environment collection, which is implemented based on the intelligent irrigation control method described in the present invention, includes:

[0041] A sensing layer for real-time collecting meteorological environment data of a target area through multiple sensors, where the multiple sensors include a wind speed sensor, a wind direction sensor, a soil moisture sensor, a multi-element sensor integrating light, temperature and humidity, and a rainfall sensor.

[0042] A transmission layer for transparently transmitting the meteorological environment data collected by the sensing layer to the control layer through a wireless sending module and a wireless receiving module.

[0043] The control layer is used to convert the data transmitted by the transmission layer through a signal conversion module and then process the data through a controller. The controller integrates an edge computing unit, a communication unit, and a decision-making unit. The edge computing unit is used to run a lightweight LSTM model to predict the rainfall on the t-th day in the future and fuse and call the predicted rainfall on the t-th day obtained from the meteorological system API to generate an improved predicted rainfall on the t-th day, where t is a positive integer. The communication unit is used on the one hand to call the predicted rainfall on the t-th day obtained from the meteorological system API through a two-way wireless communication module, and on the other hand to send the data received and generated by the control layer to the cloud layer through the two-way wireless communication module. The decision-making unit is used to receive the dynamic irrigation decision instruction from the cloud layer through the two-way wireless communication module and send it to the execution layer.

[0044] The cloud layer is used to generate a dynamic irrigation decision instruction through the Internet of Things cloud platform and realize two-way data interaction with the controller through a two-way wireless communication module.

[0045] The execution layer is used to receive the irrigation decision instruction from the control layer and control the irrigation solenoid valve to perform the irrigation action through a drive module.

[0046] Further, the wireless transmission module is an RS485-to-LoRa wireless transmission module; the wireless reception module is an RS485-to-LoRa wireless reception module; the signal conversion module is an RS485-to-TTL module; the controller is an STM32 single-chip microcomputer; the two-way wireless communication module is a 4G DTU wireless module; the drive module is an H-bridge PWM drive module.

[0047] Further, the input end of the RS485-to-LoRa wireless transmission module is unidirectionally electrically connected to the output ends of a wind speed sensor, a wind direction sensor, a soil moisture sensor, a multi-factor sensor integrating light, temperature and humidity, and a rainfall sensor respectively, and the output end of the RS485-to-LoRa wireless transmission module is unidirectionally wirelessly connected to the input end of the RS485-to-LoRa wireless reception module; the output end of the RS485-to-LoRa wireless reception module is unidirectionally electrically connected to the input end of the RS485-to-TTL module; the output end of the S485-to-TTL module is unidirectionally electrically connected to the input end of the STM32 single-chip microcomputer; the output end of the STM32 single-chip microcomputer is unidirectionally electrically connected to the input end of the H-bridge PWM drive module, and the STM32 single-chip microcomputer is bidirectionally electrically connected to the 4G DTU wireless module; the 4G DTU wireless module is bidirectionally wirelessly connected to the Internet of Things cloud platform; the output end of the H-bridge PWM drive module is unidirectionally electrically connected to the input end of the irrigation solenoid valve.

[0048] The present invention provides an intelligent irrigation control system and method based on meteorological environment collection, constructs a meteorological environment perception system and a dynamic irrigation decision-making system based on edge intelligence, integrates multi-factor sensor data of wind speed, wind direction, light intensity, air temperature, air humidity, soil moisture and rainfall through an embedded edge computing architecture, deploys a lightweight LSTM model on the controller to realize end-side rainfall prediction, and fuses meteorological system API data to optimize the prediction accuracy. Combining the differential thresholds of crop growth stages and rainfall response strategies, namely stopping irrigation during moderate rain, compensating during light rain, and irrigating without rain, an adaptive irrigation decision-making logic is constructed. At the same time, this system realizes low-power intelligent control based on the STM32 single-chip microcomputer, significantly improves the water resource utilization efficiency, is applicable to the precise irrigation scenario in the farmland environment, and is of great significance to promoting the improvement of the intelligent water-saving irrigation technology level, and has good practicability and popularization prospects. Brief Description of the Drawings

[0049] Figure 1 is a schematic diagram of the structural composition of the intelligent irrigation control system in the present invention.

[0050] Figure 2 is a flowchart for obtaining the predicted rainfall on the future t-th day after improvement in the present invention.

[0051] Figure 3 is a flowchart for generating dynamic irrigation decisions in the present invention. Detailed Embodiments

[0052] The present invention will be further described in detail below through specific embodiments. It should be noted that the present invention is not limited to the following embodiments.

[0053] Embodiment 1

[0054] An intelligent irrigation control method based on meteorological environment collection, which is implemented by the following steps:

[0055] S1. Collect historical meteorological environment data of the target area in the past three years through a wind speed sensor, a wind direction sensor, a soil moisture sensor, a multi-factor sensor integrating light and temperature and humidity, and a rain sensor. The meteorological environment data includes wind speed, wind direction, light intensity, air temperature, air humidity, soil moisture and rainfall, and construct and train a lightweight LSTM model.

[0056] Collecting meteorological environment data through a multi-sensor network solves the problem that the existing agricultural meteorological monitoring and irrigation control system relies on a single data source, ensures the comprehensiveness and accuracy of the data, improves the scientificity of irrigation decisions, and provides a solid foundation for subsequent data processing and analysis.

[0057] The constructing and training of the lightweight LSTM model includes:

[0058] First, normalize the collected historical meteorological environment data and divide it into a training set and a test set;

[0059] Then, use the Keras high-level API in TensorFlow to build a lightweight LSTM model, define the architecture of the lightweight LSTM model, select the mean squared error (MSE) as the loss function, compile the model using the Adam optimizer, train the network on the training set, and after training is completed, evaluate the performance of the model on the test set;

[0060] By building a lightweight LSTM model with TensorFlow and Keras and quantifying the model using TensorFlow Lite, the computational load is reduced, making it suitable for running on edge computing terminals, and solving the problems of latency and energy consumption caused by the existing agricultural meteorological monitoring and irrigation control system relying on cloud processing;

[0061] Finally, use TensorFlow Lite to quantify the trained model to FP16 precision and import it into STM32Cube.AI to generate a C language static library;

[0062] The model is quantified to FP16 precision and a C language static library is generated, which is convenient for deployment on embedded devices, improving the real-time performance and low-power characteristics of the system.

[0063] S2. Predict the rainfall on the first day and the rainfall on the second day in the future through the trained lightweight LSTM model.

[0064] S3. As shown in, call the meteorological system API to obtain the predicted rainfall on the first day and the predicted rainfall on the second day in the future, and perform data fusion with the predicted rainfall on the first day and the predicted rainfall on the second day output by the lightweight LSTM model to generate the improved predicted rainfall on the first day and the predicted rainfall on the second day in the future; Figure 2 The data fusion is achieved through the following formula:

[0065] ;

[0066] ;

[0067] Where:

[0068] t = 1, 2;

[0069] : The predicted rainfall on the improved t-th day in the future;

[0070] : The predicted rainfall on the t-th day in the future output by the lightweight LSTM model;

[0071] : The predicted rainfall on the t-th day in the future obtained from the meteorological system API;

[0072] : The weight coefficient.

[0073] First, collect the meteorological environment data of the current day and the past 6 days in real time;

[0074] Then, predict the rainfall of the past 6 days through the trained lightweight LSTM model, and calculate the error variance of the predicted rainfall output by the lightweight LSTM model in the past 6 days by comparing it with the rainfall of the past 6 days collected in real time; call the predicted rainfall of the past 6 days obtained from the meteorological system API, and calculate the error variance of the predicted rainfall obtained from the meteorological system API in the past 6 days by comparing it with the rainfall of the past 6 days collected in real time; the meteorological environment data of the past 6 days, the predicted rainfall data output by the lightweight LSTM model in the past 6 days, and the predicted rainfall data obtained from the meteorological system API in the past 6 days are all dynamically updated through a sliding window with a coverage period of the past 6 days, and the window moves one day later to ensure that the historical data does not overlap with the current prediction period;

[0075] Finally, calculate the weight coefficient through the following weight coefficient formula :

[0076] ;

[0077] ;

[0078] ;

[0079] ;

[0080] ;

[0081] ;

[0082] ;

[0083] Where:

[0084] N = 6;

[0085] i = 1, 2, 3, 4, 5, 6;

[0086] : The error variance of the predicted rainfall output by the lightweight LSTM model in the past N days;

[0087] : The error variance of the predicted rainfall obtained from the meteorological system API in the past N days;

[0088] : The predicted rainfall error output by the lightweight LSTM model on the i-th day in the past;

[0089] : The predicted rainfall error obtained from the meteorological system API on the i-th day in the past;

[0090] : The average error of the output of the lightweight LSTM model in the past N days;

[0091] : The average error obtained from the meteorological system API in the past N days;

[0092] : The predicted rainfall output by the lightweight LSTM model on the i-th day in the past;

[0093] : The predicted rainfall obtained from the meteorological system API on the i-th day in the past;

[0094] : The actual rainfall on the i-th day in the past collected by the rain gauge sensor.

[0095] In the initial stage, that is, when the historical data in the sliding window is less than 6 days, equal-weight fusion is adopted, that is = 0.5. After the data in the sliding window accumulates to 6 days, the weight coefficient is dynamically updated according to the weight coefficient formula. In this embodiment, the meteorological environment data in the past 6 days, the predicted rainfall data output by the lightweight LSTM model in the past 6 days, and the predicted rainfall data obtained from the meteorological system API in the past 6 days have been obtained.

[0096] By fusing the prediction results of the lightweight LSTM model and the meteorological system API, and dynamically adjusting the weight coefficient in combination with the historical error variance, the prediction accuracy is improved, and the problems of poor adaptability of the traditional static threshold mechanism and single statistical model, and the problem that it is difficult to accurately reflect the specific farmland microclimate characteristics due to the low spatial resolution of the regional-level forecast data are solved.

[0097] S4. As Figure 3 shown, set the corresponding soil moisture threshold range, optimal soil moisture and irrigation time according to the different growth stages of the target crop, and combine the real-time soil moisture data and the predicted rainfall on the t-th day in the future after improvement to generate a dynamic irrigation decision instruction to achieve intelligent irrigation.

[0098] The target crop in this embodiment is spring wheat. As shown in Table 1, the soil moisture control standards for the irrigation growth stages of spring wheat, the growth periods of spring wheat are divided into emergence stage, tillering stage, jointing stage, flowering stage, filling stage, and maturity stage; at the emergence stage, the lower limit of the soil moisture threshold range of spring wheat is 55%, the optimal soil moisture is 65%, and the upper limit of the soil moisture threshold range is 75%; at the tillering stage, the lower limit of the soil moisture threshold range of spring wheat is 60%, the optimal soil moisture is 66.5%, and the upper limit of the soil moisture threshold range is 73%; at the jointing stage, the lower limit of the soil moisture threshold range of spring wheat is 75%, the optimal soil moisture is 77.5%, and the upper limit of the soil moisture threshold range is 80%; at the flowering stage, the lower limit of the soil moisture threshold range of spring wheat is 75%, the optimal soil moisture is 80%, and the upper limit of the soil moisture threshold range is 85%; at the filling stage, the lower limit of the soil moisture threshold range of spring wheat is 65%, the optimal soil moisture is 70%, and the upper limit of the soil moisture threshold range is 75%; at the maturity stage, the lower limit of the soil moisture threshold range of spring wheat is 65%, the optimal soil moisture is 67.5%, and the upper limit of the soil moisture threshold range is 70%.

[0099] Table 1 Soil Moisture Control Standards for the Irrigation Growth Stages of Spring Wheat

[0100] Growth date (month.day) Growth stage Lower limit of soil moisture threshold range ( / %) Optimal soil moisture ( / %) Upper limit of soil moisture threshold range ( / %) 4.22-5.14 Emergence stage 55 65 75 5.15-5.27 Tillering stage 60 66.5 73 5.28-6.09 Jointing stage 75 77.5 80 6.10-6.25 Flowering stage 75 80 85 6.26-7.05 Filling stage 65 70 75 7.06-7.22 Maturity stage 65 67.5 70

[0101] The dynamic irrigation decision-making includes:

[0102] If the soil moisture on the current day is equal to or higher than the lower limit of the soil moisture threshold range corresponding to spring wheat, including equal to or higher than the upper limit of the set soil moisture threshold range, irrigation is not started;

[0103] If the soil moisture on the current day is lower than the lower limit of the soil moisture threshold range corresponding to spring wheat, it is judged according to the predicted rainfall on the improved first day in the future and the predicted rainfall on the improved second day in the future.

[0104] The judgment according to the predicted rainfall on the improved first day in the future and the predicted rainfall on the improved second day in the future includes:

[0105] If the predicted rainfall on the improved first day in the future or the predicted rainfall on the improved second day in the future is moderate rain or above, that is, ≥10 mm / day, irrigation is not started;

[0106] If the predicted rainfall on the improved first day in the future or the predicted rainfall on the improved second day in the future is light rain, that is, 0.1 - 9.9 mm / day, irrigation is started until the lower limit of the soil moisture threshold range;

[0107] If the predicted rainfall on the first day in the future after improvement or the predicted rainfall on the second day in the future after improvement is rainless or trace rainfall, i.e., <0.1 mm / day, then start irrigation until the optimal soil moisture is reached.

[0108] The rainfall levels of moderate rain, light rain, rainless, and trace rainfall are divided according to the national standard GB / T 28592-2012 "Precipitation Grades".

[0109] Dynamically adjusting the irrigation strategy according to soil moisture, crop growth stage, and predicted rainfall solves the problem of lack of flexibility in the traditional static threshold mechanism. Through the strategies of stopping irrigation during moderate rain, compensating during light rain, and irrigating during rainless days, the water resource utilization efficiency is optimized, precise irrigation is achieved, and the problems of over-irrigation or under-irrigation are avoided.

[0110] Embodiment 2

[0111] An intelligent irrigation control system based on meteorological environment collection, as Figure 1 shown, includes:

[0112] The sensing layer is used to collect meteorological environment data of the target area in real time through multiple sensors. The multiple sensors include a wind speed sensor, a wind direction sensor, a soil moisture sensor, a multi-factor sensor integrating light, temperature, and humidity, and a rainfall sensor.

[0113] The original data of each sensor are communicated in the RTU mode on the MODBUS serial link, and the data are distinguished according to different target device addresses. Physically, they are connected to the same RS485 bus.

[0114] The transmission layer is used to transmit the meteorological environment data collected by the sensing layer to the control layer through the wireless sending module and the wireless receiving module by means of spread spectrum technology; the wireless sending module is an RS485 to LoRa wireless sending module; the wireless receiving module is an RS485 to LoRa wireless receiving module; the working frequency band of the spread spectrum technology is 470 MHz, with the characteristics of a low-power wide area network; the input end of the RS485 to LoRa wireless sending module is unidirectionally electrically connected to the output ends of the wind speed sensor, the wind direction sensor, the soil moisture sensor, the multi-factor sensor integrating light, temperature, and humidity, and the rainfall sensor respectively, and the output end of the RS485 to LoRa wireless sending module is unidirectionally wirelessly connected to the input end of the RS485 to LoRa wireless receiving module.

[0115] The control layer is used to convert the data transmitted by the transmission layer through the signal conversion module and then process the data through the controller. The signal conversion module is an RS485 to TTL module; the output end of the RS485 to LoRa wireless receiving module is unidirectionally electrically connected to the input end of the RS485 to TTL module; the controller is an STM32 single-chip microcomputer, which serves as an intelligent decision-making center and integrates an edge computing unit, a communication unit, and a decision-making unit; the output end of the S485 to TTL module is unidirectionally electrically connected to the input end of the STM32 single-chip microcomputer; after the RS485 to LoRa wireless receiving module synchronously decodes the data of the RS485 to LoRa wireless transmitting module, it adapts the level signal to the 3.3V TTL standard through the RS485 to TTL module to establish communication with the STM32 single-chip microcomputer.

[0116] The cloud layer is used to generate dynamic irrigation decision instructions through the Internet of Things cloud platform and realize two-way data interaction with the controller through the two-way wireless communication module; the two-way wireless communication module is a 4G DTU wireless module with a built-in TCP / IP protocol stack; the STM32 single-chip microcomputer is two-way electrically connected to the 4G DTU wireless module; the 4G DTU wireless module is two-way wirelessly connected to the Internet of Things cloud platform; the data of multiple sensors and the predicted rainfall push the structured JSON data packet containing the timestamp and device ID to the Internet of Things cloud platform through the 4G DTU wireless module via the MQTT protocol. The Internet of Things cloud platform has the functions of data visualization storage and irrigation strategy parameter configuration.

[0117] The edge computing unit is used to run a lightweight LSTM model to predict the rainfall in the next two days and fuse and call the predicted rainfall in the next two days obtained from the meteorological system API to generate an improved predicted rainfall in the next two days.

[0118] The communication unit is used, on the one hand, to send the data received by and generated by the control layer to the cloud layer through the two-way wireless communication module; on the other hand, it is used to obtain the predicted rainfall in the next two days by calling the meteorological system API through the two-way wireless communication module. That is, the STM32 single-chip microcomputer activates 4G communication by connecting to the 4G DTU wireless module, configuring the APN and network parameters, controls the 4G DTU wireless module to send an HTTP GET request to access the weather API interface through AT commands, receives the JSON format response data, parses the JSON weather data for the next two days returned, and the parsed data starts with the symbol "$" and ends with the symbol "*", and the strings in the middle are separated by ",". For example, "$DAILY,11,9,30,24,11.3,NW,5,87,9,4,30,23,0.0,W,4, 85*" represents the forecast data for the first and second days in the future. Among them, "11,9,30,24,11.3,NW,5,87," represents the forecast data for the first day in the future. The 8 groups of data respectively represent the daytime weather code, evening weather code, highest temperature, lowest temperature, precipitation, wind direction, wind force level, and humidity. And the precipitation data is 11.3, with the unit of mm. The following 8 groups of data are the forecast data for the second day in the future, with the same meaning as above. "$DAILY" is the guide, and the 16 groups of data are separated by ",".

[0119] The decision-making unit is used to receive the dynamic irrigation decision-making instructions from the cloud layer through the two-way wireless communication module and send them to the execution layer.

[0120] The execution layer is used to receive the irrigation decision-making instructions from the control layer and control the irrigation solenoid valve to perform the irrigation action through the drive module. The drive module is an H-bridge PWM drive module; the output end of the STM32 single-chip microcomputer is unidirectionally electrically connected to the input end of the H-bridge PWM drive module; the output end of the H-bridge PWM drive module is unidirectionally electrically connected to the input end of the irrigation solenoid valve; when the STM32 single-chip microcomputer outputs its corresponding high-level control signal, the H-bridge PWM drive module starts the irrigation solenoid valve to work through the PWM signal. When the STM32 single-chip microcomputer outputs its corresponding low-level control signal, the H-bridge PWM drive module stops the irrigation solenoid valve from working through the PWM signal; the start and stop actions of the irrigation solenoid valve achieve intelligent water-saving irrigation based on the meteorological environment collection.

[0121] In the specific implementation process, the model of the STM32 single-chip microcomputer is STM32F746.

[0122] In summary, the present invention provides a method for collecting meteorological data, predicting rainfall, fusing external meteorological data, and generating dynamic irrigation decisions, as well as an intelligent irrigation control system that works in coordination with a sensing layer, a transmission layer, a control layer, a cloud layer, and an execution layer, realizing the full-process automation from data collection to irrigation execution, and solving the problems of high network transmission load and high energy consumption in existing agricultural meteorological monitoring and irrigation control systems.

[0123] In the description of the present invention, it should be understood that the indicated orientation or positional relationship is based on the orientation or positional relationship shown in the drawings, and is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the present invention.

[0124] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention, and the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent irrigation control method based on meteorological environment collection, characterized in that: This method is implemented by the following steps: S1. Collect historical meteorological and environmental data of the target area, build and train a lightweight LSTM model; S2, predict the rainfall on the tth day in the future through the trained lightweight LSTM model, where t is a positive integer; S3, calling the meteorological system API to obtain the predicted rainfall for the tth day in the future, and fusing the data with the predicted rainfall for the tth day in the future output by the lightweight LSTM model to generate an improved predicted rainfall for the tth day in the future; S4. Set the corresponding soil moisture threshold range, optimal soil moisture and irrigation time according to the different growth stages of the target crops, combine the real-time soil moisture data and the improved predicted rainfall on the tth day in the future, and generate dynamic irrigation decision instructions to achieve intelligent irrigation.

2. The intelligent irrigation control method based on meteorological environment collection according to claim 1 is characterized in that: The construction and training of the lightweight LSTM model includes: Firstly, the collected historical meteorological environment data are normalized and divided into training set and test set; Then, use Keras high-level API in TensorFlow to build a lightweight LSTM model, define the architecture of the lightweight LSTM model, select mean square error MSE as the loss function, compile the model using Adam optimizer, perform network training on the training set, and evaluate the model performance on the test set after training. Finally, use TensorFlow Lite to quantize the trained model to FP16 precision and import it into STM32Cube.AI to generate a C language static library.

3. The intelligent irrigation control method based on meteorological environment collection according to claim 1 is characterized in that: The data fusion is achieved through the following formula: ； in: : The improved predicted rainfall for the next t day; : The predicted rainfall for the next t day output by the lightweight LSTM model; : The predicted rainfall for the next t day obtained by the weather system API; : Weight coefficient.

4. The intelligent irrigation control method based on meteorological environment collection according to claim 3 is characterized in that: The weight coefficient It is obtained by: First, the meteorological environment data of the current day and the past N days are collected in real time, where N is a positive integer greater than 1; Then, the rainfall in the past N days is predicted by the trained lightweight LSTM model, and the error variance of the predicted rainfall output by the lightweight LSTM model in the past N days is obtained by comparing it with the real-time rainfall in the past N days. The forecast rainfall for the past N days is obtained by calling the meteorological system API, and the forecast rainfall obtained by the meteorological system API for the past N days is calculated by comparing it with the real-time rainfall for the past N days; the meteorological environment data for the past N days, the forecast rainfall data output by the lightweight LSTM model for the past N days, and the forecast rainfall data obtained by the meteorological system API for the past N days are all dynamically updated through a sliding window mechanism, and the coverage period of the sliding window is shifted backward day by day to ensure that the historical data does not overlap with the current forecast period; Finally, the weight coefficient is calculated by the following weight coefficient formula: : ； in: : Error variance of the predicted rainfall output by the lightweight LSTM model over the past N days; : Error variance of the predicted rainfall obtained by the weather system API in the past N days; In the initial stage, when the historical data in the sliding window is less than N days, equal weight fusion is adopted, that is, =0.5, after the data in the sliding window accumulates to N days, the weight coefficient is dynamically updated according to the weight coefficient formula.

5. The intelligent irrigation control method based on meteorological environment collection according to claim 4 is characterized in that: The meteorological environment data include wind speed, wind direction, light intensity, air temperature, air humidity, soil moisture and rainfall.

6. The intelligent irrigation control method based on meteorological environment collection according to claim 3 is characterized in that: The dynamic irrigation decision-making includes: If the soil moisture on that day is equal to or higher than the lower limit of the set soil moisture threshold range, including being equal to or higher than the upper limit of the set soil moisture threshold range, irrigation will not be started; If the soil moisture on that day is lower than the lower limit of the set soil moisture threshold range, a judgment is made based on the improved predicted rainfall for the first future day and the improved predicted rainfall for the second future day.

7. The intelligent irrigation control method based on meteorological environment collection according to claim 6 is characterized in that: The judging according to the improved predicted rainfall on the first day in the future and the improved predicted rainfall on the second day in the future includes: If the improved forecast rainfall for the first day in the future or the improved predicted rainfall for the next 2 days If it is moderate rain or above, i.e. ≥10 mm / day, irrigation will not be started; If the improved forecast rainfall for the first day in the future or the improved predicted rainfall for the next 2 days If it is light rain, i.e. 0.1-9.9 mm / day, irrigation is started to the lower limit of the soil moisture threshold range; If the improved forecast rainfall for the first day in the future or the improved predicted rainfall for the next 2 days If there is no rain or little rainfall, i.e. <0.1 mm / day, irrigation is started to reach the optimum soil moisture.

8. An intelligent irrigation control system based on meteorological environment collection, the system is implemented based on the intelligent irrigation control method based on meteorological environment collection as claimed in any one of claims 1 to 7, characterized in that: include: The perception layer is used to collect meteorological environment data of the target area in real time through multiple sensors, including wind speed sensors, wind direction sensors, soil moisture sensors, multi-factor sensors integrating light, temperature and humidity, and rainfall sensors; The transmission layer is used to transparently transmit the meteorological environment data collected by the perception layer to the control layer through the wireless transmission module and the wireless receiving module; The control layer is used to convert the data transmitted from the transmission layer into signals through the signal conversion module and then process the data through the controller, wherein the controller integrates an edge computing unit, a communication unit and a decision unit; the edge computing unit is used to run a lightweight LSTM model to predict the rainfall on the tth day in the future and fuse the predicted rainfall on the tth day in the future obtained by calling the meteorological system API to generate an improved predicted rainfall on the tth day in the future, wherein t is a positive integer; the communication unit is used to obtain the predicted rainfall on the tth day in the future by calling the meteorological system API through a two-way wireless communication module, and on the other hand, to send the data received and generated by the control layer to the cloud layer through the two-way wireless communication module; the decision unit is used to receive dynamic irrigation decision instructions from the cloud layer through the two-way wireless communication module and send them to the execution layer; The cloud layer is used to generate dynamic irrigation decision instructions through the IoT cloud platform and realize two-way data interaction with the controller through a two-way wireless communication module; The execution layer is used to receive irrigation decision instructions from the control layer and control the irrigation solenoid valve to perform irrigation actions through the drive module.

9. The intelligent irrigation control system based on meteorological environment collection according to claim 8 is characterized in that: The wireless transmission module is an RS485 to LoRa wireless transmission module; the wireless receiving module is an RS485 to LoRa wireless receiving module; the signal conversion module is an RS485 to TTL module; the controller is an STM32 single-chip microcomputer; the two-way wireless communication module is a 4G DTU wireless module; and the driving module is an H-bridge PWM driving module.

10. The intelligent irrigation control system based on meteorological environment collection according to claim 9, characterized in that: The input end of the RS485 to LoRa wireless transmitting module is unidirectionally electrically connected to the output ends of the wind speed sensor, the wind direction sensor, the soil moisture sensor, the multi-factor sensor integrating light and temperature and humidity, and the rainfall sensor, respectively, and the output end of the RS485 to LoRa wireless transmitting module is unidirectionally wirelessly connected to the input end of the RS485 to LoRa wireless receiving module; the output end of the RS485 to LoRa wireless receiving module is unidirectionally electrically connected to the input end of the RS485 to TTL module; the output end of the S485 to TTL module is unidirectionally electrically connected to the electrical input end of the STM32 single-chip microcomputer; the output end of the STM32 single-chip microcomputer is unidirectionally electrically connected to the input end of the H-bridge PWM driving module, and the STM32 single-chip microcomputer is bidirectionally electrically connected to the 4G DTU wireless module; the 4G DTU wireless module is bidirectionally wirelessly connected to the Internet of Things cloud platform; the output end of the H-bridge PWM driving module is unidirectionally electrically connected to the input end of the irrigation solenoid valve.

Images