Real-time monitoring and control method and system for water and fertilizer irrigation based on edge computing

Abstract

This invention relates to the field of smart agriculture technology, and in particular to a method and system for real-time monitoring and control of water and fertilizer irrigation based on edge computing. The method includes: deploying edge computing nodes and dividing the area into sub-regions to collect and preprocess data; acquiring information tailored to the characteristics of different regions to provide accurate data support for personalized water and fertilizer control; generating instruction sets by inputting local data into a local water and fertilizer demand prediction model to anticipate crop water and fertilizer requirements and plan control strategies in advance; generating a global collaborative irrigation map by sharing instructions between adjacent nodes to coordinate global irrigation needs, balance water and fertilizer allocation across regions, and prevent localized irrigation imbalances from impacting the overall farmland ecosystem; forming a sequence of water and fertilizer execution parameters by deconstructing and recombining instructions based on the global map to ensure error-free irrigation execution and improve water and fertilizer utilization efficiency; and controlling irrigation by issuing parameters and uploading the results data to enhance the intelligence level of subsequent water and fertilizer control.

Description

Technical Field

[0001] This invention relates to the field of smart agriculture technology, and in particular to a method and system for real-time monitoring and control of water and fertilizer irrigation based on edge computing. Background Technology

[0002] Existing water and fertilizer irrigation regulation methods typically employ irrigation strategies based on single sensor thresholds or centralized processing via cloud platforms; however, when farmland areas are large and crop varieties are diverse, existing methods face fundamental limitations.

[0003] Methods based on single-sensor thresholds rely solely on local point data, failing to perceive differences in soil moisture transport and crop water requirements between adjacent areas. In farmland with significant soil spatial variability, single-point data is insufficient to represent the true water demand of the entire irrigation district, easily leading to over-irrigation in some areas and under-irrigation in others. While cloud-based centralized processing methods can integrate multi-source data, the data transmission chain is long, with the complete closed loop from data acquisition and cloud processing to command issuance typically taking several minutes to tens of minutes. Furthermore, crop water and fertilizer requirements may change during periods of rapid light and temperature variations, and cloud computing delays can cause control commands to be out of sync with actual crop needs. Additionally, in centralized architectures, all edge sensors and execution terminals communicate directly with the central cloud platform; when network connectivity is unstable, the entire irrigation system faces the risk of paralysis. Therefore, this invention proposes a real-time monitoring and control method and system for water and fertilizer irrigation based on edge computing. Summary of the Invention

[0004] The purpose of this invention is to solve the problems in the background art, and to propose a real-time monitoring and control method and system for water and fertilizer irrigation based on edge computing.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: The first aspect of this invention provides a method for real-time monitoring and control of water and fertilizer irrigation based on edge computing, comprising: S1. Deploy multiple edge computing nodes within the farmland area. Each edge computing node is responsible for collecting soil moisture data and crop physiological and ecological data of its corresponding sub-region and performing preprocessing, thereby constructing and storing local monitoring time-series data streams. S2. Each edge computing node inputs the local monitoring time series data stream into the pre-deployed local water and fertilizer demand prediction model. By analyzing the time series variation characteristics of crop water and fertilizer demand, a set of water and fertilizer balance control instructions for sub-regions is obtained. S3. Adjacent edge computing nodes share their respective water and fertilizer balance control instruction sets through a self-organizing grid network, and generate a global collaborative irrigation map through instruction integration and consensus decision-making collaborative processing. S4. Each edge computing node deconstructs and reorganizes the instructions in the water and fertilizer balance control instruction set according to the global collaborative irrigation map to form a water and fertilizer execution parameter sequence. S5. The edge computing node sends the water and fertilizer execution parameter sequence to the corresponding irrigation execution terminal, controls the water and fertilizer irrigation process in real time, and simultaneously uploads the execution result of this irrigation and the stored local monitoring time series data stream to the central cloud platform.

[0006] Furthermore, the process of dividing the sub-region corresponding to each edge computing node in S1 includes: Obtain digital elevation models and spatial distribution maps of soil texture in farmland areas, and divide farmland into several hydrological response units based on surface runoff simulation algorithms; Obtain a crop planting distribution map of farmland areas and divide consecutive plots of land with the same variety and consistent growth cycle into a crop management unit; The hydrological response unit and the crop management unit are spatially overlaid and analyzed, and their intersection is taken as the basic irrigation unit. Constrained by the wireless communication coverage radius and data processing capability of edge computing nodes, spatially adjacent and similar basic irrigation units are aggregated into a sub-region.

[0007] Further, S1 generates a local monitoring time-series data stream, including: Each edge computing node collects soil moisture data and crop physiological and ecological data of each basic irrigation unit in its corresponding sub-region in a cyclical manner at a preset sampling period. Edge computing nodes add timestamps to each raw data packet corresponding to the collected soil moisture data and crop physiological and ecological data, forming a raw monitoring sequence with time stamps; The original monitoring sequence with time stamps is input into a locally deployed Kalman filter to obtain a smoothed monitoring sequence; A cubic spline interpolation operation is performed on the smoothed monitoring sequence to construct a temporally continuous and synchronous local monitoring time series data stream.

[0008] Furthermore, S2 obtains a set of water and fertilizer balance control instructions for the sub-region, including: The local monitoring time-series data stream is analyzed, and the first feature value representing the crop water deficit state and the second feature value representing the crop nutrient demand are extracted from the local monitoring time-series data stream. The first feature value and the second feature value are combined into a fused feature vector, and the fused feature vector is input into the local water and fertilizer demand prediction model deployed thereon. The local water and fertilizer demand prediction model performs forward propagation calculation on the input fused feature vector and outputs the water demand probability density distribution curve and fertilizer demand probability density distribution curve of the sub-region. Input the water demand probability density distribution curve and the fertilizer demand probability density distribution curve into the joint decoder; The joint decoder is used to detect and determine whether the event of generating the water and fertilizer balance regulation command has been triggered. Based on this triggered event, assign a priority value and a timeliness label to the generated water and fertilizer balance control command; All water and fertilizer balance control instructions generated during the current monitoring period, along with their respective instruction priority values ​​and timeliness tags, will be combined and encapsulated into a water and fertilizer balance control instruction set for this sub-region.

[0009] Furthermore, the generation of a global collaborative irrigation map in S3 includes: Each edge computing node establishes a self-organizing mesh network connection with other edge computing nodes within the wireless communication coverage radius according to a preset neighbor discovery protocol; Each edge computing node will generate a set of water and fertilizer balance control instructions and broadcast them to all adjacent edge computing nodes via a self-organizing grid network. Each edge computing node receives the water and fertilizer balance control instruction set broadcast by all neighboring edge computing nodes, and merges it with its own instruction set to form a candidate instruction pool awaiting consensus. Using the candidate instruction pool as input, an I-PBFT-based consensus mechanism is initiated. In this consensus mechanism, the instruction priority value carried by each instruction requiring water and fertilizer balance regulation is converted into the voting weight of the edge computing node to which the instruction belongs, and participates in the latest block generation and verification voting in the consensus process. After the consensus mechanism is executed, all edge computing nodes reach a consensus on the global execution order of instructions in the candidate instruction pool, and arrange the agreed instructions according to the consensus order to form a globally ordered instruction list; Generate a global collaborative irrigation map based on a globally ordered list of instructions.

[0010] Furthermore, the step of generating a global collaborative irrigation map based on a globally ordered instruction list includes: The edge computing node resolves the globally ordered instruction list, extracts the target irrigation area identifier corresponding to each instruction and the instruction sequence number in the globally ordered instruction list from the globally ordered instruction list; Based on the target irrigation area identifier, the spatial boundary vector data and irrigation facility topology data of each target irrigation area are retrieved from the farmland geographic information database stored locally or in the cloud, in order to determine whether any two target irrigation areas have a spatial adjacency relationship or a hydraulic connectivity relationship. Each target irrigation area is defined as a node in the map, and the instruction sequence number and spatial coordinates of the target irrigation area are associated and stored for each map node. Based on spatial adjacency and hydraulic connectivity, create connecting edges between the corresponding graph nodes; All graph nodes and all connecting edges are combined and encapsulated to generate a global collaborative irrigation graph. The global collaborative irrigation graph is used to reflect the water and fertilizer demand status of the entire farmland and the global command priority ranking of irrigation tasks in real time.

[0011] Furthermore, the determination of whether two target irrigation areas have a spatial adjacency or hydraulic connectivity based on spatial boundary vector data and irrigation facility topology data includes: For spatial boundary vector data, analyze the spatial adjacency relationship between any two target irrigation areas. When the boundaries of the two target irrigation areas overlap or the distance is less than the preset spatial coupling threshold, it is determined that the two target irrigation areas have a spatial adjacency relationship. For irrigation facility topology data, analyze whether there is a direct hydraulic connection between any two target irrigation areas. If so, determine that the two target irrigation areas have a hydraulic connection and determine the direction of the hydraulic connection based on the water flow direction constraint.

[0012] Furthermore, the water and fertilizer execution parameter sequence formed in S4 includes: From the generated global collaborative irrigation map, the first instruction to be executed, which is assigned to this sub-region and has the highest global instruction priority, is parsed out; Based on the target irrigation area identifier carried by the first instruction to be executed, retrieve the pre-determined soil infiltration characteristic curve and fertilizer transport parameters of the target irrigation area from the local database; The target water requirement value, target fertilizer requirement value, soil infiltration characteristic curve, and fertilizer transport parameters in the first instruction to be executed are input into the locally deployed fluid dynamics simulation model. The fluid dynamics simulation model takes the uniform distribution of water and fertilizer in the root zone soil of the target irrigation area and the minimization of deep leakage as the objective function, and takes the valve opening sequence, fertilizer pump duty cycle sequence and total execution time as the variables to be solved, and performs inverse solution iterative calculation. The fluid dynamics simulation model outputs the valve opening adjustment sequence, fertilizer pump duty cycle adjustment sequence, and total execution time of this irrigation to optimize the objective function. These three are combined into a water and fertilizer execution parameter sequence for the target irrigation area.

[0013] Further, S5 includes: Based on the generated water and fertilizer execution parameter sequence, it is distributed to the irrigation execution terminal corresponding to the target irrigation area through a self-organizing grid network; The irrigation execution terminal receives and parses the water and fertilizer execution parameter sequence, and performs irrigation in conjunction with the local real-time clock unit. Real-time recording of valve opening degree, actual flow rate, and execution time; generating actual execution log. During and after the irrigation process, the irrigation execution terminal will send the actual execution log back to the corresponding edge computing node. The actual execution logs are associated and packaged with the local monitoring time-series data stream corresponding to this irrigation to form training sample data with execution effect labels; The training sample data is uploaded to the central cloud platform, which then aggregates the training sample data uploaded by all edge computing nodes to build a global sample library. The central cloud platform adopts a federated learning framework and uses a global sample database to incrementally train and optimize the local water and fertilizer demand prediction model.

[0014] A second aspect of the present invention provides a real-time monitoring and control system for water and fertilizer irrigation based on edge computing, comprising: Edge data acquisition and preprocessing module: Multiple edge computing nodes are deployed in the farmland area. Each edge computing node is responsible for collecting soil moisture data and crop physiological and ecological data of its corresponding sub-region and preprocessing them, thereby constructing and storing local monitoring time series data streams; Water and fertilizer balance regulation instruction set generation module: Each edge computing node inputs the local monitoring time series data stream into the pre-deployed local water and fertilizer demand prediction model. By analyzing the time series change characteristics of crop water and fertilizer demand, a water and fertilizer balance regulation instruction set for the sub-region is obtained. Command Coordination and Decision Integration Module: Adjacent edge computing nodes share their respective water and fertilizer balance control command sets through a self-organizing grid network, and generate a global collaborative irrigation map through command integration and consensus decision-making collaborative processing; Instruction deconstruction and reorganization execution module: Each edge computing node deconstructs and reorganizes the instructions in the water and fertilizer balance regulation instruction set according to the global collaborative irrigation map to form a water and fertilizer execution parameter sequence; Execution feedback and cloud upload module: The edge computing node sends the water and fertilizer execution parameter sequence to the corresponding irrigation execution terminal, controls the water and fertilizer irrigation process in real time, and simultaneously uploads the execution result of this irrigation and the stored local monitoring time series data stream to the central cloud platform.

[0015] Compared with existing technologies, the beneficial effects of the present invention in providing a real-time monitoring and control method and system for water and fertilizer irrigation based on edge computing are as follows: 1) By deploying multiple edge computing nodes in farmland and dividing it into sub-regions, data from different regions can be collected and preprocessed, providing a reliable foundation for subsequent analysis. This adapts to the complex environment of farmland, enables precise local monitoring, and improves data availability. By inputting local data into a local water and fertilizer demand prediction model to generate a water and fertilizer balance control instruction set, the timing changes in crop water and fertilizer demand can be analyzed, and demand can be predicted in advance, making the control more forward-looking, improving resource utilization efficiency, and ensuring healthy crop growth. 2) By sharing instruction sets among adjacent nodes and generating a global collaborative irrigation map, global collaborative decision-making is achieved, optimizing overall resource allocation and improving the scientific and rational nature of farmland management. Based on the global map, instructions are deconstructed and recombined to form a sequence of water and fertilizer execution parameters, ensuring precise and effective irrigation execution, enabling water and fertilizer to be distributed rationally as needed, and improving crop absorption efficiency. By issuing the water and fertilizer execution parameter sequence to regulate irrigation and uploading the results and data, real-time precise regulation is achieved, and the data is transmitted to the central cloud platform to provide a basis for model optimization. Attached Figure Description

[0016] Figure 1 This is a flowchart of the real-time monitoring and control method for water and fertilizer irrigation based on edge computing proposed in this invention.

[0017] Figure 2 This is a block diagram of the real-time monitoring and control system for water and fertilizer irrigation based on edge computing proposed in this invention. Detailed Implementation

[0018] 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 embodiments of the present invention, and not all embodiments. 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.

[0019] Please see Figure 1 This invention provides a real-time monitoring and control method for water and fertilizer irrigation based on edge computing, comprising: S1. Deploy multiple edge computing nodes within the farmland area. Each edge computing node is responsible for collecting soil moisture data and crop physiological and ecological data of its corresponding sub-region and performing preprocessing, thereby constructing and storing local monitoring time-series data streams. S2. Each edge computing node inputs the local monitoring time series data stream into the pre-deployed local water and fertilizer demand prediction model. By analyzing the time series variation characteristics of crop water and fertilizer demand, a set of water and fertilizer balance control instructions for sub-regions is obtained. S3. Adjacent edge computing nodes share their respective water and fertilizer balance control instruction sets through a self-organizing grid network, and generate a global collaborative irrigation map through instruction integration and consensus decision-making collaborative processing. S4. Each edge computing node deconstructs and reorganizes the instructions in the water and fertilizer balance control instruction set according to the global collaborative irrigation map to form a water and fertilizer execution parameter sequence. S5. Edge computing nodes distribute water and fertilizer execution parameter sequences to the corresponding irrigation execution terminals, control the water and fertilizer irrigation process in real time, and simultaneously upload the execution results of this irrigation and the stored local monitoring time series data stream to the central cloud platform for incremental training and optimization of the local water and fertilizer demand prediction model.

[0020] In this embodiment of the invention, the detailed implementation steps of step S1, which involves deploying multiple edge computing nodes within a farmland area, with each edge computing node responsible for collecting soil moisture data and crop physiological and ecological data for its corresponding sub-region and performing preprocessing, thereby constructing and storing a local monitoring time-series data stream, include:

[0021] S11. Each edge computing node collects soil moisture data and crop physiological and ecological data of each basic irrigation unit in its corresponding sub-region in a cyclical manner at a preset sampling period. Among them, soil moisture data includes soil volumetric water content, matrix potential and electrical conductivity, and crop physiological and ecological data includes stem flow rate, canopy temperature and leaf surface humidity.

[0022] S12. The edge computing node, based on its local real-time clock unit, timestamps each raw data packet corresponding to the collected soil moisture data and crop physiological and ecological data, forming a raw monitoring sequence with time stamps.

[0023] S13. Input the original monitoring sequence with time stamps into the locally deployed Kalman filter to obtain the smoothed monitoring sequence;

[0024] Specifically, the generated original monitoring sequence with time stamps is input into the Kalman filter as the observation sequence; state equation and observation equation are established, where the state equation characterizes the dynamic change law of soil moisture or crop physiological parameters, adopts a first-order linear model, and the state vector at the current moment is equal to the state vector at the previous moment multiplied by the state transition matrix plus the process noise, while the observation equation characterizes the relationship between the sensor measurement value and the true state, and the observation vector at the current moment is equal to the state vector at the current moment multiplied by the observation matrix plus the observation noise;

[0025] The Kalman filter performs iterative calculations, including two stages: time update and measurement update. In the time update stage, based on the state estimate and state transition matrix of the previous time step, the prior state estimate of the current time step is predicted, and the prior estimate covariance is calculated. In the measurement update stage, based on the observation value and observation matrix of the current time step, the Kalman gain is calculated, and the prior state estimate is corrected using the Kalman gain to obtain the posterior state estimate of the current time step, and the posterior estimate covariance is updated. For outlier impulse noise caused by sensor interruptions, since the deviation between the observed value and the state prediction value is extremely large, the Kalman gain will approach zero, making the corrected posterior state estimate almost entirely dependent on the state prediction value, thus effectively eliminating outliers. After multiple iterations, a smoothed monitoring sequence is output, in which each data point is the optimal estimate based on historical data and the current observation.

[0026] S14. Perform cubic spline interpolation on the smoothed monitoring sequence to uniformly interpolate and align data from different sensors and different original sampling frequencies onto the same standard time axis, thereby constructing a local monitoring time-series data stream that is continuous and synchronous in time. The local monitoring time-series data stream serves as the standardized input for the subsequent local water and fertilizer demand prediction model. Cubic spline interpolation is an interpolation method that fits discrete data points using a series of piecewise cubic polynomials, ensuring that the function values, first derivatives, and second derivatives are continuous in each segment interval, thus making the overall fitted curve smooth.

[0027] The process of dividing the sub-region corresponding to each edge computing node in S1 includes: P1. Obtain the digital elevation model and soil texture spatial distribution map of the farmland area, and divide the farmland into several hydrological response units based on the surface runoff simulation algorithm; Understandably, a digital elevation model is a ground elevation dataset stored in the form of a regular grid or an irregular triangular network, with each grid point recording the elevation value at that location; a soil texture spatial distribution map is soil classification data stored in the form of vector polygons or raster, with each region labeled with its soil texture type. The digital elevation model is imported into a surface runoff simulation algorithm, which is based on the D8 single-direction algorithm or the MFD multi-direction algorithm, to calculate the runoff accumulation and direction of each grid cell. For each grid cell, the outflow direction of the water flow is determined based on its slope, aspect, and elevation difference with adjacent cells. Based on the spatial distribution of the runoff accumulation, consecutive grid cells with similar runoff paths and runoff accumulation within the same range are merged into a hydrological response unit. Meanwhile, by combining the spatial distribution map of soil texture, continuous areas with the same range of soil permeability coefficients are designated as soil texture zones; the hydrological response units are spatially superimposed with the soil texture zones, and the intersection area of ​​the two is taken as the final hydrological response unit; each hydrological response unit has similar surface runoff paths and soil moisture infiltration characteristics, providing a hydrological basis for the subsequent calculation of irrigation water demand for the unit.

[0028] P2. Obtain a crop planting distribution map of farmland areas and divide continuous plots of land with the same variety and consistent growth cycle into a crop management unit. Understandably, crop planting distribution maps are obtained through satellite remote sensing image classification or field surveys, and the planting boundaries and variety information of each crop are recorded in vector polygon form. The crop variety label and sowing date of each polygon plot are extracted from the crop planting distribution map. Consecutive plots planted with the same variety and whose sowing dates differ by no more than a preset number of days are merged into one crop management unit. Adjacent plots planted with different crops are divided into different crop management units. Plots of the same variety but with significantly different sowing dates are also divided into different crop management units due to their different growth cycles and differences in water and fertilizer requirements. The division results of crop management units are stored in vector layer form, and each unit contains agronomic attributes such as crop variety, sowing date, and expected harvest date. The crop management units reflect the physiological needs of the crops themselves.

[0029] P3. Spatial overlay analysis of hydrological response units and crop management units is performed, and the intersection of the two is taken as the basic irrigation unit, including: The obtained hydrological response unit vector layers and crop management unit vector layers are imported into the geographic information system software, and spatial overlay analysis is performed. The spatial overlay analysis uses intersection operation to overlay the two unit vector layers to generate a new polygon layer. Each newly generated polygon layer represents the intersection area of ​​the hydrological response unit and the crop management unit. For each intersection area, its corresponding hydrological response attributes are extracted, including runoff accumulation level and soil permeability coefficient, as well as crop management attributes, including crop variety and sowing date. The intersection area is defined as a basic irrigation unit. The basic irrigation unit also has hydrological consistency, that is, similar runoff and infiltration characteristics within the basic irrigation unit, and agronomic consistency, that is, the same variety of crop with the same growth cycle is planted within the basic irrigation unit. Understandably, the basic irrigation unit is the smallest spatial unit for water and fertilizer management. Subsequent sensor deployment and irrigation execution are all based on the basic irrigation unit. The collection of all basic irrigation units covers the entire farmland area, and the boundaries between the units do not overlap.

[0030] P4. The process of aggregating spatially adjacent and similarly attributed basic irrigation units into a sub-region, constrained by the wireless communication coverage radius and data processing capability of edge computing nodes, is as follows: The device parameters of each edge computing node are obtained, including the maximum wireless communication coverage radius of the wireless communication submodule and the rated data processing capacity of the central processing unit. The rated data processing capacity is expressed as the number of sampled data points that can be processed per second or the number of instructions that can be issued per second. Taking the generated basic irrigation units as input, a clustering algorithm is used to aggregate spatially adjacent and similar basic irrigation units into clusters. Specifically, spatial adjacency is obtained by calculating the Euclidean distance between the centroids of two basic irrigation units and determining whether it is less than the wireless communication coverage radius. Attribute similarity is determined by comparing the soil texture type and crop variety of the basic irrigation units. The constraints of the clustering algorithm include: all basic irrigation units within each cluster are fully covered by the wireless communication radius of an edge computing node, meaning the distance from any basic irrigation unit within the cluster to the candidate location of the edge computing node at the cluster center is no greater than the wireless communication coverage radius; simultaneously, the sum of the total data sampling frequency and the total command issuance frequency of all basic irrigation units within the cluster does not exceed the rated data processing capacity of the edge computing node; the clustering algorithm runs iteratively until all basic irrigation units are assigned to clusters that satisfy the constraints; each cluster is a sub-region, and an edge computing node is deployed at the cluster center.

[0031] In this embodiment of the invention, each edge computing node of S2 inputs the local monitoring time-series data stream into the pre-deployed local water and fertilizer demand prediction model. By analyzing the time-series variation characteristics of crop water and fertilizer demand, the detailed implementation steps of obtaining the water and fertilizer balance regulation instruction set for the sub-region include: S21. Analyze the local monitoring time-series data stream and extract the first feature value representing the crop water deficit state and the second feature value representing the crop nutrient demand from the local monitoring time-series data stream. The first feature value includes the maximum daily fluctuation of the crop stem flow rate at noon and the cumulative difference between the canopy temperature and the air temperature. The second feature value includes the slope of the change of soil electrical conductivity in the root zone per unit time and the duration of leaf surface humidity. Specifically, the local monitoring time-series data stream is analyzed, which contains time series data from multiple channels, including stem flow rate, canopy temperature, air temperature, soil conductivity, and leaf surface humidity. For the first characteristic value of crop water deficit state, the maximum daily fluctuation of crop stem flow rate at noon is extracted. This includes: using a 24-hour window, locating the time period from 12 noon to 2 pm every day, finding the highest and lowest values ​​of stem flow rate during this time period, and calculating the difference between the two as the maximum daily fluctuation value; at the same time, the process of extracting the cumulative difference between canopy temperature and air temperature is as follows: calculating the difference between canopy temperature and air temperature at each sampling time, taking the positive value of the difference, that is, only keeping the part greater than zero, and then accumulating it on an hourly basis to obtain the hourly cumulative difference heat value; For the second characteristic value of crop nutrient requirements, the slope of the change in soil electrical conductivity in the root zone over a unit of time is extracted. This includes: taking the soil electrical conductivity time series of the past hour, performing linear fitting on the series using the least squares method, and the slope of the resulting straight line is the slope of change. At the same time, the process of extracting the duration of leaf surface humidity is as follows: monitoring the leaf surface humidity sensor values, starting timing when the value exceeds the preset wetting threshold, and stopping timing when the value falls below the preset drying threshold. This time length is the duration. Among them, the wetting threshold is the lowest sensor reading threshold value for determining that there is liquid water on the leaf surface or that it is in a high humidity state; the drying threshold is the highest sensor reading threshold value for determining that the moisture on the leaf surface has evaporated and returned to a dry state.

[0032] S22. Combine the first feature value and the second feature value into a fused feature vector, and input the fused feature vector into the local water and fertilizer demand prediction model it is deployed. The local water and fertilizer demand prediction model is a deep learning model built on a long short-term memory network. Its network structure includes an input layer, two long short-term memory hidden layers, and a fully connected output layer. The number of nodes in the input layer is the number of features extracted from the local monitoring time-series data stream. For example, four nodes correspond to the maximum daily fluctuation amplitude, the cumulative difference, the slope of conductivity change, and the duration of leaf surface humidity, respectively. The first long short-term memory hidden layer contains 128 memory units, each with an input gate, a forget gate, and an output gate structure, used to learn the short-term fluctuation pattern of the time series. The second long short-term memory hidden layer contains 64 memory units, used to further abstract and extract the long-term dependencies of the time series. The fully connected output layer contains two neurons, which output the instantaneous values ​​of water demand probability and fertilizer demand probability, respectively. The training process of the model is as follows: collect historical monitoring data of the sub-region over the past three years, including soil moisture, crop physiological and ecological data, and actual irrigation records; use the feature values ​​extracted from the historical monitoring data as input samples, and use whether irrigation was carried out and the amount of irrigation in the actual irrigation records as labels; use mean squared error as the loss function, use the Adam optimizer for backpropagation training, and iteratively update the network weight parameters until the loss function converges; after training, store the model parameters in the non-volatile memory of the edge computing node.

[0033] S23. The specific steps of the local water and fertilizer demand prediction model to perform forward propagation calculation on the input fused feature vector and output the water demand probability density distribution curve of the sub-region within the first preset time window and the fertilizer demand probability density distribution curve within the second preset time window are as follows: The trained local water and fertilizer demand prediction model is deployed on edge computing nodes. During real-time operation, each new set of fused feature vectors is input into the local water and fertilizer demand prediction model. The local water and fertilizer demand prediction model first passes the fused feature vectors through the input layer to the first long short-term memory hidden layer. In the first hidden layer, each memory unit updates its current cell state and hidden state based on the current input and the hidden state of the previous time step through calculations via the input gate, forget gate, and output gate. The updated hidden state is then passed as the output to the second long short-term memory hidden layer. The second hidden layer performs similar calculations to further extract higher-order features. Finally, the fully connected output layer receives the output of the second hidden layer and generates two probability density distribution curves through linear transformation and the Softmax activation function. The water demand probability density distribution curve represents the probability value of irrigation required at each hour within the first preset time window, such as the next 6 hours. The fertilizer demand probability density distribution curve represents the probability value of fertilization required at each hour within the second preset time window, such as the next 12 hours. Both curves are plotted with time on the horizontal axis and probability values ​​on the vertical axis.

[0034] S24. Input the water demand probability density distribution curve and the fertilizer demand probability density distribution curve into the joint decoder, where the joint decoder simultaneously monitors the instantaneous values ​​of the two curves.

[0035] S25. Use the joint decoder to detect and determine whether the water and fertilizer balance control command generation event is triggered: When the joint decoder detects that the instantaneous values ​​of water demand probability and fertilizer demand probability both exceed their respective preset dynamic start thresholds at the same time, the water and fertilizer balance control command generation event is triggered. Specifically, the joint decoder maintains two preset dynamic activation thresholds: a water requirement activation threshold and a fertilizer requirement activation threshold. These two preset dynamic activation thresholds are dynamically adjusted according to the season and crop growth stage. The water requirement activation threshold is the minimum instantaneous value of the water requirement probability required to trigger the generation of an irrigation command, used to determine whether the crop has entered a state that requires irrigation. The fertilizer requirement activation threshold is the minimum instantaneous value of the fertilizer requirement probability required to trigger the generation of a fertilization command, used to determine whether the crop has entered a state that requires fertilization. The joint decoder receives the output water demand probability density distribution curve and fertilizer demand probability density distribution curve in real time. For the water demand probability density curve, the decoder monitors its instantaneous value at the current moment, i.e., the vertical axis value corresponding to the water demand probability density curve at the current time point. Similarly, it monitors the instantaneous value of the fertilizer demand probability density curve at the current moment. The decoder internally sets up a state machine, which is initially in a waiting state. At each sampling moment, the decoder compares the instantaneous water demand value at the current moment with the water demand initiation threshold, and at the same time compares the instantaneous fertilizer demand value with the fertilizer demand initiation threshold. When the instantaneous water demand value is greater than or equal to the water demand initiation threshold, and the instantaneous fertilizer demand value is greater than or equal to the fertilizer demand initiation threshold, the state machine jumps from the waiting state to the trigger state, and records the current moment as the trigger moment. The decoder generates a trigger signal, marking that the generation event of the water demand and fertilizer balance control command has been triggered. If only one condition is met or both conditions are not met, the state machine remains in the waiting state and continues to monitor the next moment.

[0036] S26. Based on this triggering event, assign a priority value and a timeliness label to the generated water and fertilizer balance control command: obtain the weighted sum of the magnitude of the instantaneous water demand probability exceeding the water demand start-up threshold and the magnitude of the instantaneous fertilizer demand probability exceeding the fertilizer start-up threshold, assign a priority value to the generated water and fertilizer balance control command, and assign a timeliness label to the generated water and fertilizer balance control command based on the intersection of the peak time of the water demand probability density distribution curve and the peak time of the fertilizer demand probability density distribution curve, wherein the timeliness label indicates that the command must be executed before the specified deadline.

[0037] S27. All water and fertilizer balance control instructions generated during the current monitoring period, along with their respective instruction priority values ​​and timeliness tags, will be combined and encapsulated into a water and fertilizer balance control instruction set for this sub-region.

[0038] In this embodiment of the invention, the specific implementation method of the adjacent edge computing nodes of S3 sharing their respective water and fertilizer balance control instruction sets through a self-organizing grid network, and generating a global collaborative irrigation map through instruction integration and consensus decision-making collaborative processing is as follows: S31. Each edge computing node uses its own wireless communication submodule to establish a self-organizing mesh network connection with other edge computing nodes within the wireless communication coverage radius according to the preset neighbor discovery protocol. The process of establishing a self-organizing grid network connection is as follows: After each edge computing node powers on, its wireless communication submodule begins scanning beacon signals on a preset industrial, scientific, and medical frequency band. The edge computing node periodically broadcasts Hello messages based on a neighbor discovery protocol, such as an improved version of the on-demand distance vector routing protocol. The Hello message contains the edge computing node's device identifier, current geographic coordinates, and remaining battery power. When edge computing node B receives a Hello message broadcast by edge computing node A, edge computing node B calculates the signal strength indicator value between itself and edge computing node A. If this value is greater than a preset communication quality threshold, edge computing node A is considered a reachable neighbor. It should be noted that the calculation process for the signal strength indicator value is as follows: the wireless receiver chip of edge computing node B measures the power level of the wireless signal received from edge computing node A, takes the common logarithm of the ratio of this power level value to a 1 milliwatt reference power, and multiplies it by 10 to obtain the signal strength indicator value in dBm. The communication quality threshold is the minimum signal strength requirement to ensure successful data packet reception and decoding, and is a preset value based on the receiving sensitivity and bit error rate requirements of the wireless communication chip. Edge computing node B records the information of edge computing node A in its local neighbor table and unicasts an acknowledgment message to edge computing node A. After receiving the acknowledgment message, edge computing node A adds edge computing node B to its neighbor table. After multiple rounds of broadcasting and acknowledgment, each edge computing node establishes a neighbor table containing all other edge computing nodes within the wireless communication coverage radius. Based on the neighbor relationships obtained from the neighbor table, edge computing nodes automatically form a network through the wireless mesh network protocol, forming a multi-hop interconnected self-organizing mesh network topology.

[0039] S32. Each edge computing node will broadcast the generated water and fertilizer balance control instruction set to all adjacent edge computing nodes via the self-organizing grid network.

[0040] S33. Each edge computing node receives the water and fertilizer balance control instruction set broadcast by all neighboring edge computing nodes, and merges it with its own instruction set (merging the received instruction set with its own generated instruction set) to form a candidate instruction pool awaiting consensus.

[0041] S34. Each edge computing node takes the candidate instruction pool as input and starts a consensus mechanism based on I-PBFT, where I-PBFT stands for Improved Practical Byzantine Fault Tolerance Algorithm. In this consensus mechanism, the instruction priority value carried by each instruction requiring water and fertilizer balance regulation is converted into the voting weight of the edge computing node to which the instruction belongs, and participates in the latest block generation and verification voting in the consensus process. Specifically, after forming a candidate instruction pool, each edge computing node initiates a consensus process based on an improved practical Byzantine fault-tolerant algorithm. All edge computing nodes select one as the leader edge computing node for this round of consensus through a round of hash calculation and comparison. The leader edge computing node extracts unprocessed instructions from the candidate instruction pool, preliminarily sorts them according to their reception time, packages them into a candidate block, and broadcasts this candidate block to all other edge computing nodes, i.e., replica edge computing nodes. Each replica edge computing node, upon receiving a candidate block, verifies each instruction within the block. Verification includes checking the authenticity and validity of the digital signature of the edge computing node from which the instruction originated, whether the instruction priority value is within a preset range, and whether the deadline indicated by the timeliness tag is later than the current time. After verification, the replica edge computing node calculates its own voting weight for the candidate block based on the priority value of each instruction. The specific calculation method is as follows: the replica edge computing node sums the priority values ​​of all instructions in the candidate block to obtain the total instruction priority value. Then, the replica edge computing node decides to support or oppose the candidate block based on its own local strategy, such as if the candidate block contains instructions it generated, it tends to support it, otherwise it remains neutral. The weight coefficient of its vote is equal to the sum of the instruction priority values ​​held by the replica edge computing node itself. The replica edge computing node broadcasts a preparation message with weighted coefficients to all other edge computing nodes. Each edge computing node collects preparation messages from other edge computing nodes. When the sum of the weights represented by the collected preparation messages exceeds two-thirds of the total weight of all edge computing nodes, the edge computing node broadcasts a commit message. After the edge computing nodes have collected enough commit messages, the candidate block is officially confirmed as the latest block on the blockchain. The order of instructions in the latest block is the global execution order. All edge computing nodes arrange the confirmed instructions according to the latest block order to form a globally ordered instruction list.

[0042] S35. After the consensus mechanism is executed, all edge computing nodes reach a consensus on the global execution order of instructions in the candidate instruction pool, and arrange the agreed instructions according to the consensus order to form a globally ordered instruction list.

[0043] S36. Generate a global collaborative irrigation map based on the globally ordered instruction list, including: S361. The edge computing node resolves the globally ordered instruction list, extracts the target irrigation area identifier corresponding to each instruction and the instruction sequence number in the globally ordered instruction list from the globally ordered instruction list, where the instruction sequence number is used to characterize the global execution sequence of the instruction.

[0044] S362. Based on the target irrigation area identifier, retrieve the spatial boundary vector data and irrigation facility topology data of each target irrigation area from the farmland geographic information database stored locally or in the cloud, to determine whether any two target irrigation areas have a spatial adjacency relationship or a hydraulic connectivity relationship; wherein, the irrigation facility topology data includes the irrigation pipeline connection relationship and water flow direction constraints.

[0045] S363. Define each target irrigation area as a node in the map, and associate and store the instruction sequence number and spatial coordinates of the target irrigation area with each map node, including: For spatial boundary vector data, analyze the spatial adjacency relationship between any two target irrigation areas. When the boundaries of the two target irrigation areas overlap or the distance is less than the preset spatial coupling threshold, it is determined that the two target irrigation areas have a spatial adjacency relationship. For irrigation facility topology data, analyze whether there is a direct hydraulic connection between any two target irrigation areas. If there is, determine that the two target irrigation areas have a hydraulic connection and determine the direction of hydraulic connection based on the water flow direction constraint. Specifically, when defining a graph node, the edge computing node creates an empty directed graph data structure in its local memory. This data structure is stored in the form of an adjacency list to optimize query efficiency. The edge computing node traverses each instruction in the globally ordered instruction list and extracts the corresponding target irrigation area identifier from each instruction. The edge computing node searches in the graph data structure to see if a graph node corresponding to the target irrigation area identifier already exists. If it does not exist, the edge computing node creates a new graph node and sets the target irrigation area identifier as the unique identifier of the graph node. The edge computing node associates and stores two core attributes for the graph node: the first attribute is the instruction sequence number of the target irrigation area in the globally ordered instruction list, which is stored as an integer value in the attribute field of the graph node; the second attribute is the geographic coordinates of the center point of the target irrigation area obtained from the farmland geographic information database, including longitude and latitude values, stored as floating-point numbers.

[0046] S364. Based on spatial adjacency and hydraulic connectivity, create connecting edges between the corresponding graph nodes, and attach edge type labels and direction attributes to each connecting edge. The edge type labels are used to distinguish between spatial adjacency and hydraulic connectivity. The direction attributes are determined according to the order of the instruction sequence in spatial adjacency and according to the direction of water flow in hydraulic connectivity. Specifically, when determining spatial adjacency, the edge computing node traverses all graph node pairs in the directed graph data structure. For each graph node pair, designated as the first graph node and the second graph node, the edge computing node retrieves the spatial boundary vector data of the first target irrigation area corresponding to the first graph node and the spatial boundary vector data of the second target irrigation area corresponding to the second graph node from the farmland geographic information database. A planar scanning algorithm is used to calculate and analyze the positional relationship between the two spatial boundary vector polygons. If the boundaries of the two polygons share an edge or have an overlapping area, the edge computing node directly determines that the first graph node and the second graph node are spatially adjacent. If the boundaries of the two polygons do not intersect, but the Euclidean distance between the center point of the first graph node and the center point of the second graph node is less than a preset spatial coupling threshold, such as 50 meters, the edge computing node also determines that the two are spatially adjacent. Here, the spatial coupling threshold refers to the maximum critical distance value at which two independent irrigation areas are considered to be likely to have mutual influence in spatial location. When determining hydraulic connectivity, the edge computing node extracts a directed graph model of the irrigation pipeline network from the irrigation facility topology data. The nodes of this directed graph model are physical pipeline connection points, and the edges are pipeline segments with water flow direction constraints. The edge computing node associates the first graph node with the nearest pipeline network node and the second graph node with the nearest pipeline network node. Using a depth-first search algorithm, starting from the pipeline network node associated with the first graph node, the node traverses along the edge direction. If the search reaches the pipeline network node associated with the second graph node, it is determined that there is a hydraulic connectivity relationship between the first graph node and the second graph node, and the water flow direction is determined to be from the first graph node to the second graph node based on the search path. When creating connecting edges, for spatial adjacency, the edge computing node creates an undirected edge between the first and second graph nodes, but adds a direction attribute to this edge, with the direction pointing from the graph node with the smaller instruction sequence number to the graph node with the larger instruction sequence number; for hydraulic connectivity, the edge computing node creates a directed edge between the first and second graph nodes, with the edge direction consistent with the water flow direction; the edge computing node adds an edge type label to each connecting edge, with the label value being either spatial adjacency or hydraulic connectivity, for subsequent querying and calculation.

[0047] S365. Combine and encapsulate all graph nodes and all connecting edges to generate a global collaborative irrigation graph data structure containing node attribute sets and edge attribute sets, and store this graph data structure in local memory; wherein, the global collaborative irrigation graph is used to reflect the water and fertilizer demand status of the entire farmland and the global command priority sorting of irrigation tasks in real time.

[0048] In this embodiment of the invention, the specific implementation method for each edge computing node in S4 to deconstruct and reorganize the instructions in the water and fertilizer balance control instruction set based on the global collaborative irrigation map to form a water and fertilizer execution parameter sequence includes: S41. From the generated global collaborative irrigation map, parse out the first instruction to be executed that is assigned to this sub-region and has the highest global instruction priority. Understandably, in the generated global collaborative irrigation map, each map node corresponds to a target irrigation area; each map node is associated with and stores the instruction sequence number for that target irrigation area. This instruction sequence number is a globally unique sequential number determined by a consensus mechanism, with smaller numbers indicating higher global instruction priority; each edge computing node maintains a queue of tasks to be executed in its local memory, initially empty; after the global collaborative irrigation map is updated, the edge computing node initiates the task parsing process, traversing all map nodes in the global collaborative irrigation map; for each map node, the edge computing node extracts its corresponding target irrigation area identifier and determines... The edge computing node determines whether the target irrigation area belongs to a sub-region under its responsibility. If the result is yes, the edge computing node further extracts the instruction sequence number associated with the graph node. The edge computing node sorts and compares the instruction sequence numbers of all graph nodes belonging to its sub-region, and selects the graph node with the smallest instruction sequence number. The water and fertilizer balance control instruction corresponding to this graph node is the first instruction to be executed, which is assigned to this sub-region and has the highest global instruction priority. The edge computing node copies this first instruction to be executed from the global collaborative irrigation graph and pushes it to the head of the local task queue, awaiting processing by subsequent steps.

[0049] S42. Based on the target irrigation area identifier carried by the first instruction to be executed, retrieve the pre-measured soil infiltration characteristic curve and fertilizer transport parameters of the target irrigation area from the local database. The soil infiltration characteristic curve represents the relationship between soil moisture infiltration rate and time, and the fertilizer transport parameters include solute dispersion coefficient and adsorption coefficient.

[0050] S43. Input the target water demand value, target fertilizer demand value, soil infiltration characteristic curve, and fertilizer transport parameters from the first instruction to be executed into the locally deployed fluid dynamics simulation model. Understandably, the first instruction to be executed contains the generated instruction priority value, timeliness tag, and specific operation parameters. When encapsulating the water and fertilizer balance control instruction set, each instruction, in addition to binding the instruction priority value and timeliness tag, is also bound to the quantified value obtained by decoding the water demand probability density distribution curve and the fertilizer demand probability density distribution curve. Specifically, the water demand value corresponding to the peak of the water demand probability density distribution curve is used as the target water demand value, and the fertilizer demand value corresponding to the peak of the fertilizer demand probability density distribution curve is used as the target fertilizer demand value. The two quantified values ​​and the instruction identifier are encapsulated together in the instruction data structure. The instruction identifier is a unique identification code assigned to the generated water and fertilizer balance control instruction, which is composed of the generation timestamp and the edge computing node number and is used to uniquely identify and index the instruction in the entire system. When the first instruction to be executed is parsed, the target water demand value and the target fertilizer demand value can be directly read from the data structure of the water and fertilizer balance control instruction. The fluid dynamics simulation model is a simplified solution model based on the Richards equation and the one-dimensional convection-dispersion equation. The Richards equation characterizes the movement of water in unsaturated soil, and its basic form is that the rate of change of water content over time is equal to the spatial gradient of water flux. The one-dimensional convection-dispersion equation characterizes the transport of solutes in the soil, including the convection term that flows with water and the dispersion term caused by the concentration gradient. The fluid dynamics simulation model discretizes the soil profile into several layers, each assuming homogeneous soil properties. The implicit finite difference method is used to discretize the Richards equation, transforming the partial differential equation into a system of nonlinear algebraic equations, which are then solved using the Newton-Raphson iteration method. For the one-dimensional convection-dispersion equation, the method of characteristics is used for decoupling, treating the convection and dispersion terms separately. The fluid dynamics simulation model is pre-compiled and optimized to achieve the solution.

[0051] S44. The fluid dynamics simulation model takes the uniform distribution of water and fertilizer in the root zone soil of the target irrigation area and the minimization of deep leakage as the objective function, and takes the valve opening sequence, fertilizer pump duty cycle sequence and total execution time as the variables to be solved, and performs inverse solution iterative calculation. Specifically, the fluid dynamics simulation model transforms the inverse problem into an optimization problem; it initializes the valve opening sequence to a constant, such as 50%, the fertilizer pump duty cycle sequence to a constant, such as 30%, and the total execution time to an empirical value, such as 60 minutes; it then substitutes these initial values ​​into the fluid dynamics simulation model for forward simulation, simulating the distribution of soil moisture and nutrients in the root zone within the future total execution time; finally, it calculates the error between the simulation results and the target distribution, which requires moisture and nutrients to be uniformly distributed within the root zone depth range, and... The deep leakage is minimized; based on the error, the gradient descent method is used to adjust the valve opening sequence, fertilizer pump duty cycle sequence, and total execution time; after adjustment, a forward simulation is performed again to calculate the new error; this process is repeated iteratively until the error reaches the maximum number of iterations; the final output valve opening adjustment sequence is a curve that changes over time, for example, the opening increases linearly from 0 to 80% in the first 10 minutes, remains at 80% for the middle 40 minutes, and decreases linearly to 0 in the last 10 minutes; the fertilizer pump duty cycle adjustment sequence is also a curve that changes over time; the total execution time is the duration of the entire process.

[0052] S45. The fluid dynamics simulation model outputs the valve opening adjustment sequence over time, the fertilizer pump duty cycle adjustment sequence over time, and the total execution time of this irrigation, which optimize the objective function. These three are combined into a water and fertilizer execution parameter sequence for the target irrigation area.

[0053] In this embodiment of the invention, the detailed implementation steps of the S5 edge computing node sending the water and fertilizer execution parameter sequence to the corresponding irrigation execution terminal, controlling the water and fertilizer irrigation process in real time, and simultaneously uploading the execution result of this irrigation and the stored local monitoring time series data stream to the central cloud platform include: S51. Based on the generated water and fertilizer execution parameter sequence, it is distributed to the irrigation execution terminal corresponding to the target irrigation area through a self-organizing grid network; wherein, the irrigation execution terminal is a low-power intelligent valve controller integrating a wireless communication module, a high-precision flow meter and an electrically controlled regulating valve; the low-power intelligent valve controller can receive and parse the water and fertilizer execution parameter sequence, and realize precise synchronization with the instruction cycle of the edge computing node based on the local real-time clock unit.

[0054] S52. The irrigation execution terminal receives and parses the water and fertilizer execution parameter sequence. Based on the valve opening adjustment sequence and fertilizer pump duty cycle adjustment sequence, and combined with the local real-time clock unit, it controls the opening of the electronic control valve and the start and stop of the fertilizer pump on the time axis, and monitors the actual flow in real time through a high-precision flow meter. At the same time, it records the valve opening, actual flow and execution time in real time and generates an actual execution log. Understandably, the microcontroller inside the irrigation execution terminal receives the water and fertilizer execution parameter sequence, parses it, and stores it in memory. The microcontroller integrates a local real-time clock unit, which is synchronized with the time synchronization mechanism of the edge computing node. At the start of execution, the microcontroller reads the current time and finds the target valve opening value corresponding to that moment in the water and fertilizer execution parameter sequence. The microcontroller drives the motor of the electrically controlled regulating valve through a pulse-width modulation signal, while simultaneously using the valve's built-in angle sensor to provide real-time feedback on the actual opening, forming a proportional-integral-derivative closed-loop control that ensures the actual opening quickly follows the target valve opening value. Similarly, based on the target duty cycle value of the fertilizer pump corresponding to that moment in the water and fertilizer execution parameter sequence, the microcontroller drives the DC motor of the fertilizer pump through another pulse-width modulation signal, controlling its speed and thus the fertilizer injection rate. A high-precision flow meter is connected in series in the irrigation pipeline to measure the instantaneous flow rate in real time and accumulate the total flow rate. The microcontroller compares the actual flow rate with the theoretical flow rate; if the deviation exceeds the preset flow range, it adjusts the valve opening or pump duty cycle to compensate.

[0055] S53. During and after the irrigation process, the irrigation execution terminal will send the actual execution log back to the edge computing node to which it belongs.

[0056] S54. Associate and package the actual execution log with the local monitoring time series data stream corresponding to this irrigation to form training sample data with execution effect labels; Specifically, during irrigation, the edge computing node continuously receives the actual execution logs, including the actual valve opening, actual flow rate, and fertilizer pump status recorded every 100 milliseconds. Simultaneously, the edge computing node retrieves the historical local monitoring time-series data stream corresponding to this irrigation from its local database. This data stream covers the time range from 2 hours before irrigation begins to 2 hours after irrigation ends. The actual execution logs and historical monitoring time-series data streams are merged along the same time axis to form a multi-dimensional time-series data matrix. Key features are extracted from this multi-dimensional time-series data matrix as execution effect labels. For example, the difference between the average root zone soil volumetric water content and the target water content within 1 hour after irrigation ends is calculated. If the difference is within ±5%, the execution effect label is good; if the difference exceeds 5%, the execution effect is poor. The execution effect label is appended to the merged multi-dimensional time-series data matrix to form a complete training sample data.

[0057] S55. Upload the training sample data to the central cloud platform. The central cloud platform collects the training sample data uploaded by all edge computing nodes and builds a global sample library.

[0058] S56. The central cloud platform adopts a federated learning framework and uses a global sample library to incrementally train and optimize the local water and fertilizer demand prediction model. Understandably, the central cloud platform runs an aggregation server. After all edge computing nodes complete a round of irrigation and upload training sample data, the aggregation server initiates the federated learning training process, including: the aggregation server randomly selects a portion of edge computing nodes as participating nodes for this training; the aggregation server sends the initial parameters of the latest local water and fertilizer demand prediction model to the participating nodes; each participating node, upon receiving the initial parameters, loads them into its local water and fertilizer demand prediction model; several rounds of gradient descent calculations are performed using the local training sample data to obtain the model parameter update gradient; the participating nodes encrypt the calculated update gradient and send it back to the aggregation server; after collecting a sufficient number of update gradients, the aggregation server uses a federated averaging algorithm to perform a weighted average of all received gradients to calculate the global gradient; the aggregation server uses the global gradient to update the global model parameters; after the update is completed, the aggregation server encrypts and distributes the new global model parameters back to all edge computing nodes; after receiving the new parameters, the edge computing nodes replace the old parameters of their local water and fertilizer demand prediction models; the original monitoring data is retained in the local storage of the edge computing nodes throughout the entire process.

[0059] Please see Figure 2 This invention provides a real-time monitoring and control system for water and fertilizer irrigation based on edge computing, comprising: Edge data acquisition and preprocessing module: Multiple edge computing nodes are deployed in the farmland area. Each edge computing node is responsible for collecting soil moisture data and crop physiological and ecological data of its corresponding sub-region and preprocessing them, thereby constructing and storing local monitoring time series data streams; Water and fertilizer balance regulation instruction set generation module: Each edge computing node inputs the local monitoring time series data stream into the pre-deployed local water and fertilizer demand prediction model. By analyzing the time series change characteristics of crop water and fertilizer demand, a water and fertilizer balance regulation instruction set for the sub-region is obtained. Command Coordination and Decision Integration Module: Adjacent edge computing nodes share their respective water and fertilizer balance control command sets through a self-organizing grid network, and generate a global collaborative irrigation map through command integration and consensus decision-making collaborative processing; Instruction deconstruction and reorganization execution module: Each edge computing node deconstructs and reorganizes the instructions in the water and fertilizer balance regulation instruction set according to the global collaborative irrigation map to form a water and fertilizer execution parameter sequence; Execution feedback and cloud upload module: The edge computing node sends the water and fertilizer execution parameter sequence to the corresponding irrigation execution terminal, controls the water and fertilizer irrigation process in real time, and simultaneously uploads the execution result of this irrigation and the stored local monitoring time series data stream to the central cloud platform.

[0060] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0061] It should be noted that all formulas in this manual are calculated by removing dimensions and taking their numerical values. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters and thresholds in the formulas are set by those skilled in the art according to the actual situation.

[0062] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. A real-time monitoring and control method for water and fertilizer irrigation based on edge computing, characterized in that, include: S1. Deploy multiple edge computing nodes within the farmland area. Each edge computing node is responsible for collecting soil moisture data and crop physiological and ecological data of its corresponding sub-region and performing preprocessing, thereby constructing and storing local monitoring time-series data streams. S2. Each edge computing node inputs the local monitoring time series data stream into the pre-deployed local water and fertilizer demand prediction model. By analyzing the time series variation characteristics of crop water and fertilizer demand, a set of water and fertilizer balance control instructions for sub-regions is obtained. S3. Adjacent edge computing nodes share their respective water and fertilizer balance control instruction sets through a self-organizing grid network, and generate a global collaborative irrigation map through instruction integration and consensus decision-making collaborative processing. S4. Each edge computing node deconstructs and reorganizes the instructions in the water and fertilizer balance control instruction set according to the global collaborative irrigation map to form a water and fertilizer execution parameter sequence. S5. The edge computing node sends the water and fertilizer execution parameter sequence to the corresponding irrigation execution terminal, controls the water and fertilizer irrigation process in real time, and simultaneously uploads the execution result of this irrigation and the stored local monitoring time series data stream to the central cloud platform.

2. The real-time monitoring and control method for water and fertilizer irrigation based on edge computing according to claim 1, characterized in that, The process of dividing the sub-regions corresponding to each edge computing node in S1 includes: Obtain digital elevation models and spatial distribution maps of soil texture in farmland areas, and divide farmland into several hydrological response units based on surface runoff simulation algorithms; Obtain a crop planting distribution map of farmland areas and divide consecutive plots of land with the same variety and consistent growth cycle into a crop management unit; The hydrological response unit and the crop management unit are spatially overlaid and analyzed, and their intersection is taken as the basic irrigation unit. Constrained by the wireless communication coverage radius and data processing capability of edge computing nodes, spatially adjacent and similar basic irrigation units are aggregated into a sub-region.

3. The real-time monitoring and control method for water and fertilizer irrigation based on edge computing according to claim 2, characterized in that, The S1 generates a local monitoring time-series data stream, including: Each edge computing node collects soil moisture data and crop physiological and ecological data of each basic irrigation unit in its corresponding sub-region in a cyclical manner at a preset sampling period. Edge computing nodes add timestamps to each raw data packet corresponding to the collected soil moisture data and crop physiological and ecological data, forming a raw monitoring sequence with time stamps; The original monitoring sequence with time stamps is input into a locally deployed Kalman filter to obtain a smoothed monitoring sequence; A cubic spline interpolation operation is performed on the smoothed monitoring sequence to construct a temporally continuous and synchronous local monitoring time series data stream.

4. The real-time monitoring and control method for water and fertilizer irrigation based on edge computing according to claim 1, characterized in that, S2 obtains a set of water and fertilizer balance control instructions for the sub-region, including: The local monitoring time-series data stream is analyzed, and the first feature value representing the crop water deficit state and the second feature value representing the crop nutrient demand are extracted from the local monitoring time-series data stream. The first feature value and the second feature value are combined into a fused feature vector, and the fused feature vector is input into the local water and fertilizer demand prediction model deployed thereon. The local water and fertilizer demand prediction model performs forward propagation calculation on the input fused feature vector and outputs the water demand probability density distribution curve and fertilizer demand probability density distribution curve of the sub-region. Input the water demand probability density distribution curve and the fertilizer demand probability density distribution curve into the joint decoder; The joint decoder is used to detect and determine whether the event of generating the water and fertilizer balance regulation command has been triggered. Based on this triggered event, assign a priority value and a timeliness label to the generated water and fertilizer balance control command; All water and fertilizer balance control instructions generated during the current monitoring period, along with their respective instruction priority values ​​and timeliness tags, will be combined and encapsulated into a water and fertilizer balance control instruction set for this sub-region.

5. The real-time monitoring and control method for water and fertilizer irrigation based on edge computing according to claim 1, characterized in that, The generation of a global collaborative irrigation map in S3 includes: Each edge computing node establishes a self-organizing mesh network connection with other edge computing nodes within the wireless communication coverage radius according to a preset neighbor discovery protocol; Each edge computing node will generate a set of water and fertilizer balance control instructions and broadcast them to all adjacent edge computing nodes via a self-organizing grid network. Each edge computing node receives the water and fertilizer balance control instruction set broadcast by all neighboring edge computing nodes, and merges it with its own instruction set to form a candidate instruction pool awaiting consensus. Using the candidate instruction pool as input, an I-PBFT-based consensus mechanism is initiated. In this consensus mechanism, the instruction priority value carried by each instruction requiring water and fertilizer balance regulation is converted into the voting weight of the edge computing node to which the instruction belongs, and participates in the latest block generation and verification voting in the consensus process. After the consensus mechanism is executed, all edge computing nodes reach a consensus on the global execution order of instructions in the candidate instruction pool, and arrange the agreed instructions according to the consensus order to form a globally ordered instruction list; Generate a global collaborative irrigation map based on a globally ordered list of instructions.

6. The real-time monitoring and control method for water and fertilizer irrigation based on edge computing according to claim 5, characterized in that, The step of generating a global collaborative irrigation map based on a globally ordered instruction list includes: The edge computing node resolves the globally ordered instruction list, extracts the target irrigation area identifier corresponding to each instruction and the instruction sequence number in the globally ordered instruction list from the globally ordered instruction list; Based on the target irrigation area identifier, the spatial boundary vector data and irrigation facility topology data of each target irrigation area are retrieved from the farmland geographic information database stored locally or in the cloud, in order to determine whether any two target irrigation areas have a spatial adjacency relationship or a hydraulic connectivity relationship. Each target irrigation area is defined as a node in the map, and the instruction sequence number and spatial coordinates of the target irrigation area are associated and stored for each map node. Based on spatial adjacency and hydraulic connectivity, create connecting edges between the corresponding graph nodes; All graph nodes and all connecting edges are combined and encapsulated to generate a global collaborative irrigation graph. The global collaborative irrigation graph is used to reflect the water and fertilizer demand status of the entire farmland and the global command priority ranking of irrigation tasks in real time.

7. The real-time monitoring and control method for water and fertilizer irrigation based on edge computing according to claim 6, characterized in that, The determination of whether two target irrigation areas have a spatial adjacency or hydraulic connectivity based on spatial boundary vector data and irrigation facility topology data includes: For spatial boundary vector data, analyze the spatial adjacency relationship between any two target irrigation areas. When the boundaries of the two target irrigation areas overlap or the distance is less than the preset spatial coupling threshold, it is determined that the two target irrigation areas have a spatial adjacency relationship. For irrigation facility topology data, analyze whether there is a direct hydraulic connection between any two target irrigation areas. If so, determine that the two target irrigation areas have a hydraulic connection and determine the direction of the hydraulic connection based on the water flow direction constraint.

8. The real-time monitoring and control method for water and fertilizer irrigation based on edge computing according to claim 1, characterized in that, The water and fertilizer execution parameter sequence formed in S4 includes: From the generated global collaborative irrigation map, the first instruction to be executed, which is assigned to this sub-region and has the highest global instruction priority, is parsed out; Based on the target irrigation area identifier carried by the first instruction to be executed, retrieve the pre-determined soil infiltration characteristic curve and fertilizer transport parameters of the target irrigation area from the local database; The target water requirement value, target fertilizer requirement value, soil infiltration characteristic curve, and fertilizer transport parameters in the first instruction to be executed are input into the locally deployed fluid dynamics simulation model. The fluid dynamics simulation model takes the uniform distribution of water and fertilizer in the root zone soil of the target irrigation area and the minimization of deep leakage as the objective function, and takes the valve opening sequence, fertilizer pump duty cycle sequence and total execution time as the variables to be solved, and performs inverse solution iterative calculation. The fluid dynamics simulation model outputs the valve opening adjustment sequence, fertilizer pump duty cycle adjustment sequence, and total execution time of this irrigation to optimize the objective function. These three are combined into a water and fertilizer execution parameter sequence for the target irrigation area.

9. The real-time monitoring and control method for water and fertilizer irrigation based on edge computing according to claim 1, characterized in that, S5 includes: Based on the generated water and fertilizer execution parameter sequence, it is distributed to the irrigation execution terminal corresponding to the target irrigation area through a self-organizing grid network; The irrigation execution terminal receives and parses the water and fertilizer execution parameter sequence, and performs irrigation in conjunction with the local real-time clock unit. Real-time recording of valve opening degree, actual flow rate, and execution time; generating actual execution log. During and after the irrigation process, the irrigation execution terminal will send the actual execution log back to the corresponding edge computing node. The actual execution logs are associated and packaged with the local monitoring time-series data stream corresponding to this irrigation to form training sample data with execution effect labels; The training sample data is uploaded to the central cloud platform, which then aggregates the training sample data uploaded by all edge computing nodes to build a global sample library. The central cloud platform adopts a federated learning framework and uses a global sample database to incrementally train and optimize the local water and fertilizer demand prediction model.

10. A real-time monitoring and control system for water and fertilizer irrigation based on edge computing, characterized in that, The system, applied to the edge computing-based real-time monitoring and control method for water and fertilizer irrigation as described in any one of claims 1-9, comprises: Edge data acquisition and preprocessing module: Multiple edge computing nodes are deployed in the farmland area. Each edge computing node is responsible for collecting soil moisture data and crop physiological and ecological data of its corresponding sub-region and preprocessing them, thereby constructing and storing local monitoring time series data streams; Water and fertilizer balance regulation instruction set generation module: Each edge computing node inputs the local monitoring time series data stream into the pre-deployed local water and fertilizer demand prediction model. By analyzing the time series change characteristics of crop water and fertilizer demand, a water and fertilizer balance regulation instruction set for the sub-region is obtained. Command Coordination and Decision Integration Module: Adjacent edge computing nodes share their respective water and fertilizer balance control command sets through a self-organizing grid network, and generate a global collaborative irrigation map through command integration and consensus decision-making collaborative processing; Instruction deconstruction and reorganization execution module: Each edge computing node deconstructs and reorganizes the instructions in the water and fertilizer balance regulation instruction set according to the global collaborative irrigation map to form a water and fertilizer execution parameter sequence; Execution feedback and cloud upload module: The edge computing node sends the water and fertilizer execution parameter sequence to the corresponding irrigation execution terminal, controls the water and fertilizer irrigation process in real time, and simultaneously uploads the execution result of this irrigation and the stored local monitoring time series data stream to the central cloud platform.

Images