A method for cooperative monitoring and adaptive step irrigation control

Abstract

The application discloses a kind of cooperative monitoring adaptive step-by-step sprinkling irrigation control method, step one, topography gradient, slope direction are regarded as rigid constraint condition, the three-dimensional mapping relationship between remote sensing image pixel and ground sensor measuring point is established;Step two, unmanned aerial vehicle keeps relative ground fixed height flight along topographic contour line, while completing the real-time splicing and exponential calculation of multispectral image at airborne end, uploads the grid level data after fusion;Step three, slope direction radiation difference, evapotranspiration difference caused by mountain wind are regarded as model input feature, using online learning mechanism to update model parameters in real time, carry out adaptive adjustment of irrigation decision;Step four, upper layer carries out global path planning based on digital chassis, lower layer uses terrain adaptive synchronous control law to carry out local trajectory tracking;Step five, four-legged robot dog is used as mobile relay node, and moves to optimal position to forward instruction automatically in communication blind area.

Description

Technical Field

[0001] This invention belongs to the field of sprinkler irrigation technology, specifically, it relates to a collaborative monitoring adaptive cascade sprinkler irrigation control method. Background Technology

[0002] Mountainous areas account for approximately 40% of my country's arable land. Terraced fields are characterized by fragmented plots, significant elevation differences, diverse slope aspects, and crisscrossing ridges. Current irrigation methods primarily include traditional flood irrigation (water resource utilization rate <40%) and manual mobile sprinkler irrigation, both of which are inefficient and labor-intensive. While large-scale sprinkler irrigation machines and soil moisture-based automatic control have been widely adopted in plains areas, they are ill-suited to the complex terrain of terraced fields. Related technological fields, such as air-to-ground collaborative monitoring, have been applied to crop monitoring, but face challenges in data spatiotemporal registration and crude sampling strategies. Multi-machine collaborative control is industrially mature but assumes flat ground and fails to consider elevation changes and slope dynamics. While mountain communication technologies like LoRa and Mesh exist, static deployment cannot handle mobile nodes. AI irrigation decision-making often relies on offline, fixed models, which cannot adapt to dynamic changes in mountain microclimates. Predictive maintenance has limited application in agricultural equipment due to cost sensitivity and harsh environmental conditions.

[0003] Existing technologies face five major challenges in mountain terraced field scenarios: At the perception layer, failure in spatiotemporal registration of multi-source data leads to AI decisions based on incorrect mappings; fixed-altitude flight results in uneven monitoring resolution; and sensors experience severe data drift in harsh environments. At the decision-making layer, offline model solidification leads to performance differences exceeding 30% between sunny and shady slopes; the correlation between static attribute-based management zone division and actual crop water requirements is less than 0.6; and reactive decision-making often results in ineffective irrigation due to rainfall immediately following irrigation. At the execution layer, sprinkler irrigation machines cannot cross field ridges, are prone to tipping over on slopes, multi-machine collaborative control suffers from formation instability in three-dimensional terrain, and water pressure fluctuations in self-pressurized water supply systems cause flow control errors exceeding 30%. At the communication layer, 58% of farmers reported unstable signals, and the lack of a unified protocol for heterogeneous devices creates information silos. At the operation layer, 87% of farmers found the operation complex, and 73% of malfunctions went unanswered for 48 hours. The core root cause of these problems lies in the fact that existing technologies simplify the complex three-dimensional scenario of mountain terraced fields into a two-dimensional problem, causing the entire chain from perception and decision-making to execution to fail in real-world scenarios. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the present invention aims to provide a collaborative monitoring and adaptive cascade sprinkler irrigation control method.

[0005] To achieve the aforementioned objectives, the technical solution adopted by this invention includes: Step 1: Using terrain slope and aspect as rigid constraints, establish a three-dimensional mapping relationship between remote sensing image pixels and ground sensor measurement points.

[0006] Specifically, it includes: A1: High-altitude UAV LiDAR terrain scanning A high-altitude fixed-wing UAV, equipped with a linear frequency modulated continuous wave LiDAR, performs a full-coverage scan of the terraced fields following a terrain-following flight path. The flight altitude is 150-200 meters above the ground (ensuring a point cloud density of ≥15 points / square meter), with a flight path spacing of 50 meters (ensuring 30% lateral overlap). The LiDAR equipment used is a RIEGL VUX-1LR or similar product, with a laser emission frequency ≥400kHz and a scanning angle of ±30°. The raw point cloud data P is acquired. , , , Point cloud density target: ≥15 points / square meter (actual density can reach 18-22 points / square meter).

[0007] A2: Gradiently Enhanced Triangular Mesh Filtering Based on Slope Adaptation A slope-adaptive progressive densification triangular mesh filtering algorithm is used to distinguish between ground points and vegetation points: an initial sparse triangular mesh is constructed, and the distance from each point P to be judged to the triangular mesh is calculated. and slope angle , in The elevation of the point to be determined. It is the elevation interpolated within the triangulation network. , like < and < If P is determined to be a ground point, it is added to the triangulation network, where the dynamic threshold is... Adaptive adjustment based on local topographic standard deviation: , The basic slope angle threshold is set at 15°; k: adjustment coefficient, with a value range of 0.5-1.5, which is adaptively adjusted according to the roughness of the terrain (determined by the ratio of the local elevation standard deviation to the global standard deviation). Local area elevation standard deviation, calculation window size 15m × 15m; Parameter values: = 15°, k = 0.8, The standard deviation of elevation within a 5-meter radius of P; Intelligent extraction of terraced field boundaries Based on the filtered ground point cloud, a digital elevation model was constructed using Kriging interpolation with a grid resolution of 0.5 m × 0.5 m; the elevation gradient magnitude of each grid cell was calculated. Where x and y are planar coordinates, Let x be the partial derivative of elevation in the x-direction. Let be the partial derivative of the elevation in the y-direction.

[0008] A significant peak in G(x,y) appears at the field ridge. Canny dual-threshold segmentation is used: high threshold... = 1.5 meters, low threshold = 0.8 meters, extract the field boundary line. Perform Douglas-Puck thinning (tolerance) on the boundary line. (meters), generate a vector image of terraced fields. Each field contains attributes: boundary coordinate sequence, average elevation, average slope, and aspect.

[0009] A4: BeiDou grid coding mapping and terrain factor calculation Align the terraced field vector map with the BeiDou grid coding system, with a grid size of 0.0001 degrees longitude difference × 0.0001 degrees latitude difference (approximately 10 meters × 10 meters), forming a two-level spatial index of field-grid. Calculate the following topographic factors for each grid: slope : Calculated by fitting a local plane; Slope The direction of the projection of the slope normal onto the horizontal plane; Terrain Humidity Index (TWI): TWI = ln(α / tanβ), where α is the catchment area per unit contour length and β is the slope; Solar radiation correction factor = 1 + 0.2·cos( - )·tan( ),in It is the azimuth of the sun.

[0010] Step 2: The drone flies at a fixed altitude relative to the ground along the terrain contour lines to ensure consistent ground resolution; at the same time, the multispectral image is stitched and the index is calculated in real time on the airborne terminal, and only the fused grid-level data is uploaded.

[0011] Specifically, including: B1: Low-altitude drone swarm terrain-following flight path planning Based on the DEM data generated in step A3, a terrain-following flight path is planned for each low-altitude rotary-wing UAV. The UAV's flight altitude remains fixed relative to the ground. = 50 meters, actual flight altitude is: Where δ(x,y) is the safety margin: 10 meters on steep slopes and 5 meters on gentle slopes. The flight path design ensures an 80% forward overlap and a 70% lateral overlap to guarantee the accuracy of the 3D reconstruction. This refers to the ground elevation.

[0012] B2: Multispectral Data Acquisition and Real-time Preprocessing The drone swarm synchronously collects multispectral data following the terrain and flight path. Each drone records its precise pose at the moment of exposure in real time (centimeter-level positioning via RTK-GPS and attitude angle via IMU), and packages and stores the raw images and pose data. Simultaneously, the onboard edge computing unit performs pixel-by-pixel radiometric calibration and atmospheric correction, generating reflectance image patches with geographic tags, but does not perform cross-drone stitching.

[0013] The airborne edge computing unit performs the following preprocessing in real time: Radiation calibration: converting DN values ​​to reflectivity Atmospheric correction: Dark target subtraction is used to eliminate atmospheric effects. Image stitching: Real-time generation of orthophoto mosaic based on POS data.

[0014] B3: After all drones complete their flight path missions, they transmit the pre-processed image blocks to the edge computing base station via a high-speed data transmission link (5G or high-bandwidth self-organizing network). The base station then performs joint adjustment and orthophoto mosaicking based on POS data on the ground to generate a reflectance mosaic map of the entire terraced field area. Vegetation index calculation and crop moisture inversion The following key indices are calculated in real time on the airborne end: Normalized Difference Vegetation Index (NDVI): ,in, Near-infrared reflectivity Red light band reflectivity; Normalized Moisture Index (NDWI): ,in, Green light band reflectivity Near-infrared reflectivity; Crop water retrieval model (simplified forward inversion based on the PROSAIL radiative transfer model): Where CWC represents the canopy water content (mm). Canopy temperature (obtained from a thermal infrared sensor). Parameter values ​​(calibrated in a cornfield): =-0.15、 =0.92、 =0.28、 =0.18、 =0.006.

[0015] B4: Quadruped Robot Dog Ground Fine Sampling and Path Planning The quadruped robot dog has a foot force sensing system, with a six-dimensional force / torque sensor (ATIMini45 or similar product) installed on each foot, with a sampling frequency of 1kHz; The sensors are connected to the robot dog's main controller (Intel NUC+ real-time kernel) via an EtherCAT bus; When a hard surface is detected, the robot dog marks the grid as an obstacle (cost value 255) in the local cost map, triggering the DWA local path planner to replan the obstacle avoidance path; When soft soil is detected, it is marked as a high-risk area (cost value 150), and the planner will prioritize detouring; if detouring is not possible, the walking speed will be reduced (from 1.0m / s to 0.3m / s) and the foot contact area will be increased; The quadruped robot dog, equipped with an insertable soil moisture sensor (based on the TDR time-domain reflectometry principle) and a portable weather station, performs ground sampling according to a grid-based stratified sampling strategy. At least three representative character grids should be drawn from each field (one for each of the high, medium, and low water zones).

[0016] Within each sampling grid, the robot dog autonomously selects 3 sampling points (avoiding rocks and crop roots).

[0017] Measurement point depth: 10cm, 20cm, and 30cm (three layers).

[0018] The robot dog uses foot-based force sensors to determine soil hardness and automatically avoids areas with rocks. Sampling data includes: soil volumetric water content. Soil temperature and soil electrical conductivity (EC).

[0019] B5: Adaptive Fusion of Grid-Level Multi-Source Data The CWC data retrieved from the drone and the actual measurements from the robot dog The data is uniformly mapped onto the BeiDou grid. For each grid... The state of soil moisture after fusion Calculated using the Bayesian weighted average method: in: The surface soil moisture is retrieved from remote sensing and corrected for ground data. To measure soil moisture for the robot dog This is the variance of remote sensing inversion (calculated based on image quality). The ground sampling variance (calculated based on the coefficient of variation of the sampling points). Remote sensing inversion variance It is dynamically determined by the following formula: in Based on the basic variance (taken as 0.02). , This represents the pixel percentage of clouds and shadows within the grid. The penalty coefficients for clouds and shadows are λ1=2.0 and λ2=1.5.

[0020] Step 3: Using the differences in slope aspect radiation and evapotranspiration caused by mountain winds as input features of the model, the model parameters are updated in real time using an online learning mechanism to adaptively adjust irrigation decisions.

[0021] C1: Dynamic Calculation of Crop Water Requirements For each grid The reference crop evapotranspiration ET0 was calculated using the Penman-Monteith formula: in: Net radiation (W / m 2 The crop canopy temperature, calculated from the crop canopy temperature retrieved in step B3 and corrected for slope aspect, is as follows: Where G is the soil heat flux (W / m³) 2 (Approximately 0.1) , T represents the average air temperature (°C), measured by a weather sensor carried by the robot dog, taking into account the decrease in elevation. Wind speed at 2 meters (m / s), corrected for terrain: z0 is the surface roughness; The saturated vapor pressure is (kPa). This represents the actual water vapor pressure (kPa). Δ is the slope of the saturated water vapor pressure curve (kPa / ℃). γ is the wet / dry constant (kPa / ℃).

[0022] Crop coefficient Adjustments are made dynamically based on crop growth stages. The current growth stage is determined using phenological information retrieved from time-series UAV imagery and the cumulative growth days method. When GDD reaches the threshold, a table lookup and update are performed. value.

[0023] Step C1 calculates the value of each grid cell in hourly increments (on the hour). The value will be calculated and stored in the grid status database, with fields including: grid ID, timestamp, etc. , T wait.

[0024] C2: Soil Moisture Dynamics Prediction Based on Particle Filtering A particle filter algorithm is used to dynamically predict soil moisture in a grid, addressing the difficulty in obtaining parameters for traditional mechanistic models. The state equation and observation equation are as follows: Equation of state (soil water balance): ,in, Let t be the soil moisture content. This refers to the rainfall during the specified period. Irrigation volume during the period This represents crop evapotranspiration over a given period of time. This represents the amount of deep seepage within a given time period. Surface runoff during the time period This is process noise; Observation equation: in For process noise, To observe noise.

[0025] Particle filtering execution steps: Initialization: Generate N=500 particles {θ (i) 0}, weight w (i) 0 = 1 / N Prediction: Update θ for each particle based on the equation of state. (i) _{t+1|t}, Update: When a new observation y_{t+1} is obtained, calculate the particle weights: ,in, Prior particle weights, , For a one-step prediction, R is the observation noise variance, 2π is twice the value of pi, and exp() is an exponential function. Resampling: Resample according to weights to avoid particle degradation. State estimation: .

[0026] The particle filter prediction module in step C2 executes every 30 minutes, reading the latest data from the database. Based on soil moisture values ​​and weather forecast data, predict soil moisture changes 24 hours in advance.

[0027] C3: Spatiotemporal Dynamic Management Zone Division Based on the predicted 24-hour soil moisture distribution using particle filtering, fuzzy C-means clustering algorithm is employed to dynamically divide irrigation management zones. The clustering feature vectors include: Current soil moisture Predicting soil moisture Topographic humidity index (TWI) and crop coefficient , Clustering objective function: Where c is the number of clusters (3-5), m is the fuzzy index (2), n is the number of samples, and j is the category index. For membership degree, As cluster center, The eigenvectors are used. The membership matrix of each grid is obtained through iterative optimization, and the grids are assigned to the management areas with the highest membership degrees.

[0028] C4: Irrigation Urgency Calculation and Prescription Map Generation For each grid Calculate the Irrigation Urgency Index (IUI): in: Field holding capacity The wilting coefficient, For the predicted soil moisture 24 hours later, Evapotranspiration of grid crops The crop sensitivity coefficient is set at 1.2 for corn and 1.0 for wheat. The IUI distribution of each grid within a field is statistically analyzed.

[0029] For fields Calculate the comprehensive irrigation decision index : in Grid area weights. Decision rule: like If the score is greater than 1.5, the field will be added to the irrigation queue, with priority... = Otherwise, the field will not be irrigated; For fields included in the irrigation program, the target irrigation volume is... Determined by the following formula: in: The target soil moisture content is 0.8θ_fc. This represents the average soil moisture content of the field. The planned wetting layer depth is 0.3-0.5 meters. The irrigation water utilization coefficient is taken as 0.85-0.9.

[0030] Step 4: The upper layer performs global path planning based on the digital chassis, while the lower layer uses terrain-adaptive synchronous control law to achieve local trajectory tracking.

[0031] Specifically, it includes: D1: Kinematic and Dynamic Modeling of Sprinkler Irrigation Machines Each sprinkler system consists of a powered main tower and several driven spans. A kinematic model of the sprinkler system is established: Where (x, y) is the position of the main tower crane, θ is the heading angle, and v is the linear velocity. ω is the angular velocity.

[0032] The dynamic model considers the gravity component of the slope: in As the driving force, For rolling resistance, For slope, For the direction of travel, It is a slope.

[0033] D2: Cross-plot global path planning based on an improved A* algorithm When a sprinkler needs to move from the current field to a target field, a 3D accessibility cost map is constructed based on the DEM generated in step A3: in: Local elevation changes Maximum obstacle clearance height of sprinkler irrigation machine; surface roughness. Elevation gradient magnitude, Slope angle; The first penalty is for steep slopes (slope weighting coefficient λ1=2.0). The second penalty exceeds the obstacle-crossing height of the sprinkler machine. (0.3) Field ridges (elevation change weighting coefficient λ2=5.0) The third penalty is for surface roughness (roughness weighting coefficient λ3=1.0). An improved A* algorithm is used to search for the optimal transfer path on the cost map, with node expansion considering eight-neighbor directions. The heuristic function uses an improved Euclidean distance that takes elevation into account. ,in , Current node planar coordinates, , Target node planar coordinates Current node elevation, The target node elevation is set, and α is the elevation weighting coefficient, which is set to 0.5. The algorithm output is a sequence of path points. m is the total number of path points, and each path It is a three-dimensional coordinate vector: , The kth path point, It is the x-coordinate of the path point. It is the y-coordinate of the path point. It is the elevation of the path point.

[0034] D3: Field Operation Path Planning For a single terraced field Based on its boundary shape and sprinkler arrangement, plan the sprinkler operation path: For rectangular or near-rectangular fields: use a zigzag reciprocating path, with the path spacing equal to the spray width (15-20 meters). Irregular plots: Employ a spiral path with inward contraction of boundaries. The job path is represented by parametric equations: ,in This is the total path length. The operation path function, s is the path arc length parameter, x(s) is the x-coordinate function, y(s) is the y-coordinate function, and θ(s) is the heading angle function; D4: Distributed cooperative control based on leader-follower The system adopts a leader-follower distributed collaborative control architecture, designating one sprinkler as the leader and the other sprinklers as followers. The leader is responsible for global path tracking, while the followers are responsible for maintaining their relative positions with the leader.

[0035] Master Navigator Control Law (based on chain control theory): in:( , , () represents the lateral deviation, longitudinal deviation, and heading deviation relative to the reference path. , For reference speed The control gain was adjusted through field testing: =0.8, =1.2, =2.5, Follower control law (based on cascaded PI control): in: The azimuth angle of the follower relative to the navigator. Distance deviation: = - , This is the actual spacing. For the desired spacing, For heading deviation; PI controller parameters: 1 = 0.5 1 = 0.1 2 = 0.8 2 = 0.15.

[0036] D5: Terrain Adaptive Synchronization Control Compensation When the sprinkler is on a slope, the control law incorporates a terrain compensation term: Speed ​​compensation: Where μ is the rolling resistance coefficient (taken as 0.15). For slope, For the direction of travel, For slope direction. The compensation item ensures increased driving force when going uphill and limits speed to prevent loss of control when going downhill.

[0037] Synchronous control across spans: The relative torsion angle between each span of the sprinkler system is detected by angle sensors (sensor selection: magnetic absolute encoder (AS5048A, resolution 0.05°)). The drive wheel speed is adjusted via frequency conversion, with the control target being |δθ| < 3°. Control law: in The synchronization gain is set to 0.3. This is the velocity correction amount for the i-th span.

[0038] D6: Variable irrigation control based on water pressure feedforward-feedback To address the issue of large water pressure fluctuations in mountainous areas, a feedforward-feedback composite control strategy is adopted to adjust the sprinkler head flow rate. Feedforward control: Calculate the theoretical water pressure based on the current elevation z. : in For water source pressure, For water source elevation, This refers to losses along the pipeline.

[0039] Feedback control: based on the measured values ​​of the flow sensor With target value PID control is applied to the deviation: ,in, It is the proportional gain coefficient. It is the integral gain coefficient. It is the differential gain coefficient. It is the target traffic. This is the actual measured traffic volume. It's a flow deviation. It is the deviation integral. It is the differential of the deviation, and t is time; Parameter values: =0.8, =0.15, =0.05.

[0040] Final PWM duty cycle: in This is the maximum flow rate of the nozzle. This is the rated working pressure.

[0041] This also includes the navigator being responsible for global path tracking, and its walking speed. The follower speed is determined by the control law in step D4. The control law is determined by step D4, but fine-tuning at the grid level is allowed. When the follower is about to enter a grid At that time, based on the target irrigation volume of the grid. Calculate the ideal speed : in This is the overlap coefficient (considering spray overlap). It's the nozzle flow rate. It is the width of the spray pattern. That is the grid width. If | - If the speed is greater than 0.1 m / s, the follower sends a speed adjustment request to the navigator. The navigator, while maintaining overall formation stability, adjusts the global speed appropriately. ; Speed ​​fine-tuning is limited to ±15%. Exceeding this range will trigger formation reorganization (followers will temporarily leave the formation and rejoin the formation after completing the current grid).

[0042] Step 5: Use the quadruped robot dog as a mobile relay node to automatically move to the optimal position in the communication blind spot to forward commands; at the same time, design a three-level degradation mode to ensure that the system can still operate to a limited extent when the communication part is interrupted.

[0043] Specifically, it includes: E1: Construction of Heterogeneous Mesh Self-Organizing Networks The system uses the LoRaMesh protocol to build a self-organizing network, and the network nodes include: Fixed nodes: Edge computing base stations (gateways) Backbone mobile nodes: Low-altitude unmanned aerial vehicle (UAV) swarm (airborne relay) Ground mobile nodes: quadruped robot dog (ground relay), sprinkler cluster (terminal) Each node periodically broadcasts a beacon frame containing its node ID, location, remaining battery power, and link quality. The routing table is dynamically maintained using the OLSR (Optimized Link-State Routing) protocol.

[0044] Hardware connection: The main controller (STM32H7 series MCU) of each sprinkler is connected through two independent SPI interfaces. The LoRa RF module (SX1262, 470MHz) is used for data communication, and the UWB ranging module (DW1000, channel 2) is used for relative positioning. The two modules use independent clock sources and power management to avoid mutual interference.

[0045] Protocol interaction mechanism: Timing slotting: The time axis is divided into 10ms time slots. Odd-numbered time slots are used for LoRa data transmission and reception, and even-numbered time slots are used for UWB ranging.

[0046] Synchronization Trigger: The main controller of the sprinkler machine generates a synchronization pulse through the PWM output pin, which simultaneously triggers the time slot counters of the LoRa module and the UWB module to reset.

[0047] Data fusion: The relative distance and azimuth angle measured by the UWB module are transmitted to the LoRa protocol stack through internal shared memory and uploaded along with the status data.

[0048] E2: Dynamic Relay Node Deployment Optimization When a communication link quality degradation is detected in the area where the sprinkler is located (RSSI < -85dBm or packet loss rate > 5%), the system automatically dispatches the nearest quadruped robot dog to the optimal relay position. The optimal position is determined by solving the following optimization problem: ,in, It is a candidate position for the robot dog. It refers to the link quality from the robot dog to the base station. It's about the quality of the link from the robot dog to the sprinkler machine. Minimum value function It's a maximization problem.

[0049] Constraints: The robot dog's travel distance must be less than the battery life limit, and the location must be within a passable area. The particle swarm optimization algorithm is used to solve the problem in real time, with a population size of 20 and 50 iterations, and the optimal relay point coordinates are output.

[0050] E3: Three-level communication degradation mechanism Each node monitors the communication link quality in real time and triggers a degradation based on the Comprehensive Communication Quality Index (CQI): Where RSSI_max = -50dBm, SNR_max = 30dB, w1 = 0.4, w2 = 0.3, and w3 = 0.3. Maximize the quality of the weaker link between the two links to ensure overall communication reliability.

[0051] Level 1 Degradation (CQI < 0.6): The sprinkler is disconnected from the edge base station, but the sprinklers maintain LoRaMesh interconnection. It enters cluster autonomy mode, where the current lead sprinkler assumes local decision-making functions and continues operation based on a pre-cached irrigation prescription map. The operation status is cached locally.

[0052] Level 2 Degradation (CQI < 0.4): Communication between sprinkler irrigation units is interrupted, and each sprinkler enters standby autonomous mode. It continues operation based on the currently stored irrigation prescription for the current field and the pre-programmed emergency path, remaining in place after completing the current field's task. Simultaneously, a robot dog is activated to actively search for and move to the disconnected area to attempt to restore communication.

[0053] Level 3 Degradation (CQI < 0.2 or no communication for 5 consecutive minutes): Trigger the safety stop procedure: stop walking, shut off the nozzle, record the current breakpoint coordinates and time, and send an emergency distress signal via BeiDou short message (only for models equipped with BeiDou terminals).

[0054] E4: Resuming interrupted downloads and task recovery Once communication is restored, each node uploads its work logs (including irrigated areas, traveled paths, and abnormal events) to the edge computing base station during the interruption. A timestamp vector method is used to resolve concurrent conflicts: each node maintains a local vector clock, and upon receiving the logs, the base station merges the global task progress using a distributed consensus protocol, re-plans the remaining tasks, and distributes them.

[0055] Step 6: Design a voice interaction interface and a predictive maintenance system. Control the system through natural language voice commands. The system uses machine learning to predict equipment failures and provide early warnings.

[0056] F1: Voice Interaction Interface Design Develop a lightweight speech recognition module that supports dialect adaptation. The preset command set includes: Start irrigating plot XX, check soil moisture, stop all equipment, and report the malfunction. Voice commands are processed by the edge computing base station's end-side speech recognition model (based on convolutional neural network + connection-time classification), requiring no network connection and with a response time of less than 1 second.

[0057] F2: Visualized Digital Twin Interface Build digital twin models of the terraced fields on mobile apps and the web, and display them in real time: Real-time video stream from drones Soil moisture distribution heat map Real-time location and trajectory of the sprinkler irrigation machine Device battery level and communication status. The interface adopts a minimalist design, retaining only the core control buttons, and supports one-click activation of preset irrigation modes.

[0058] F3: Equipment Health Monitoring and Predictive Maintenance Collect operational data from each device (motor current, vibration, temperature, battery cycle count) to construct a device health index: m is the number of features, and j is the feature index. It is the square of the characteristic deviation. It is the weighted squared deviation. It is a normalization factor; in For the current eigenvalue, This is the baseline value within the normal range. The feature weights are defined as follows. An isolation forest algorithm is used to detect anomalies: randomly selected samples are used to construct isolation trees, and the anomaly score s(x,n) is determined by the path length. When HI < 0.7 or an anomaly is detected, the system automatically sends a warning message to the maintenance personnel's mobile phone, along with a fault cause analysis and repair suggestions.

[0059] F4: Integrated Water and Fertilizer Expansion Interface The system has a reserved interface for a fertilizer applicator, supporting automatic liquid fertilizer injection. Fertilization decisions are based on: Soil nutrient sensor (optional) measures nitrogen, phosphorus, and potassium content. Crop nutrient requirements during different growth stages User-preset fertilization plan Coordinated control of fertilizer and irrigation amounts: delaying application after irrigation begins Start the fertilizer pump within minutes to ensure that fertilizer enters the root zone with the water, reducing deep seepage losses.

[0060] Compared with the prior art, the advantages of the present invention include: (1) The present invention provides a collaborative monitoring adaptive cascade sprinkler irrigation control method. Existing technologies simplify three-dimensional terrain into two-dimensional plane processing. The present invention is the first to take terrain dynamics constraints as the core element throughout the entire process, forming a complete technical chain from data acquisition (terrain following), fusion (elevation gradient registration), decision-making (slope aspect radiation correction) to control (slope dynamics compensation), realizing true three-dimensional smart irrigation. (2) The present invention provides a collaborative monitoring adaptive cascade sprinkler irrigation control method. Most existing AI models are trained offline and then fixed and deployed. The present invention introduces a particle filter online update mechanism, which enables the model to dynamically adjust parameters according to real-time monitoring data, solves the model failure problem caused by the spatiotemporal variability of mountain micro-meteorology, and realizes the self-evolution of decision-making. (3) The present invention provides a collaborative monitoring adaptive cascade sprinkler control method. Existing multi-machine collaborative methods are mostly flat structures. The present invention proposes a global-local two-level decoupled control architecture. The upper layer is based on a digital chassis for strategic planning (where to water), and the lower layer uses terrain adaptive control law to realize tactical execution (how to water). The hierarchy is clear and the robustness is strong. It solves the problem of computational explosion in multi-machine collaborative control under complex terrain. (4) The present invention provides a collaborative monitoring adaptive cascade sprinkler irrigation control method. Existing systems are reliable by default. The present invention takes into account the reality of poor communication in mountainous areas and designs a mobile relay dynamic optimization and a three-level degradation mechanism to enable the system to have fault tolerance capability of not stopping work when the network is interrupted, thus solving the reliability bottleneck of intelligent irrigation in mountainous areas from the root. (5) The collaborative monitoring adaptive cascade sprinkler irrigation control method provided by this invention addresses the pain points of farmers being afraid to use it or not knowing how to use it. It innovatively introduces voice interaction and predictive maintenance, simplifies complex system operations into natural language commands, and transforms passive fault repair into proactive early warning services, truly bridging the last mile of technology implementation. Attached Figure Description

[0061] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0062] Figure 1 This is a flowchart of a collaborative monitoring adaptive cascade sprinkler irrigation control method according to the present invention. Detailed Implementation

[0063] In view of the shortcomings of the prior art, the inventors of this invention, through long-term research and extensive practice, have proposed the technical solution of this invention. The technical solution, its implementation process, and principles will be further explained below with reference to the accompanying drawings and specific implementation examples in the embodiments of this application.

[0064] It should be noted that the embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention. The described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, the present invention covers any substitutions, modifications, equivalent methods and solutions made on the spirit, principles and scope of the present invention as defined by the claims. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0065] In the description of this application, the terms "first," "second," "third," and similar terms do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Similarly, the terms "one" or "a" do not indicate a quantity limitation, but rather indicate the presence of at least one. Terms such as "include" or "contain" mean that the elements or objects preceding "include" cover the elements or objects listed following "include" or their equivalents, and do not exclude other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect.

[0066] In the description of this application, the terms center, up, down, front, back, left, right, vertical, horizontal, top, bottom, inside, outside, etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this application. In addition, when using positional terms such as sides, outer side, top and bottom, etc., it should be understood that they are used only for the convenience of understanding and description, taking into account that the structure may be oriented to other positions.

[0067] In the description of this application, unless otherwise expressly specified and limited, the technical or scientific terms used shall have the ordinary meaning understood by a person with ordinary skills in the art to which this application pertains. Terms such as installation, connection, and linking shall be interpreted broadly, for example, they may refer to fixed connection, detachable connection, mating connection, or integral connection. For a person skilled in the art, the specific meaning of the above terms in this application may be understood according to the specific circumstances.

[0068] The present invention aims to introduce and explain the structural composition and the coordination relationship between the components of the collaborative monitoring adaptive cascade sprinkler irrigation control method. Unless otherwise specified, the size, material and manufacturing process of each component in the collaborative monitoring adaptive cascade sprinkler irrigation control method of the present invention can be selected according to specific circumstances, and no special limitations or explanations are made here.

[0069] Furthermore, to provide the public with a better understanding of the present invention, certain specific details are described in detail in the following description of the invention. However, those skilled in the art will fully understand the invention even without these detailed descriptions.

[0070] Example Please see Figure 1 Step 1: Using terrain slope and aspect as rigid constraints, establish a three-dimensional mapping relationship between remote sensing image pixels and ground sensor measurement points.

[0071] Specifically, it includes: A1: High-altitude UAV LiDAR terrain scanning A high-altitude fixed-wing UAV, equipped with a linear frequency modulated continuous wave LiDAR, performs a full-coverage scan of the terraced fields following a terrain-following flight path. The flight altitude is 150-200 meters above the ground (ensuring a point cloud density of ≥15 points / square meter), with a flight path spacing of 50 meters (ensuring 30% lateral overlap). The LiDAR equipment used is a RIEGL VUX-1LR or similar product, with a laser emission frequency ≥400kHz and a scanning angle of ±30°. The raw point cloud data P is acquired. , , , Point cloud density target: ≥15 points / square meter (actual density can reach 18-22 points / square meter).

[0072] A2: Slope-adaptive progressive densification triangular mesh filtering (the ground surface composed of ground points is continuous and smooth, while non-ground points such as vegetation points and building points are suspended above the ground surface. By constructing an initial sparse ground triangular mesh, and then iteratively adding points that meet the constraints to the triangular mesh and densifying it, a complete ground model is finally obtained.)

[0073] In each iteration, it is necessary to determine whether the point P to be judged should be added to the triangulation network (ground point) or filtered out (non-ground point). The judgment is based on two indicators: the perpendicular distance from the point to the triangulation network and the slope angle from the point to the triangulation network. A slope-adaptive progressive densification triangular mesh filtering algorithm is used to distinguish between ground points and vegetation points: an initial sparse triangular mesh is constructed, and the distance from each point P to be judged to the triangular mesh is calculated. and slope angle , in The elevation of the point to be determined. It is the elevation interpolated within the triangulation network. , like < and < If P is determined to be a ground point, it is added to the triangulation network, where the dynamic threshold is... Adaptive adjustment based on local topographic standard deviation: , The basic slope angle threshold is set at 15°; k: adjustment coefficient, with a value range of 0.5-1.5, which is adaptively adjusted according to the roughness of the terrain (determined by the ratio of the local elevation standard deviation to the global standard deviation). Local area elevation standard deviation, calculation window size 15m × 15m; Parameter values: = 15°, k = 0.8, The standard deviation of elevation within a 5-meter radius of P; Intelligent extraction of terraced field boundaries (The essential characteristic of terraced fields is that they are flat fields at different elevations connected by ridges. The terrain changes drastically at these ridges, resulting in abrupt changes in elevation. By calculating the rate of change of elevation (i.e., gradient) of the land surface, these abrupt changes can be located, thus extracting the field boundaries.) Based on the filtered ground point cloud, a digital elevation model was constructed using Kriging interpolation with a grid resolution of 0.5 m × 0.5 m; the elevation gradient magnitude of each grid cell was calculated. Where x and y are planar coordinates, Let x be the partial derivative of elevation in the x-direction. Let be the partial derivative of the elevation in the y-direction.

[0074] A significant peak in G(x,y) appears at the field ridge. Canny dual-threshold segmentation is used: high threshold... = 1.5 meters, low threshold = 0.8 meters, extract the field boundary line. Perform Douglas-Puck thinning (tolerance) on the boundary line. (meters), generate a vector image of terraced fields. Each field contains attributes: boundary coordinate sequence, average elevation, average slope, and aspect.

[0075] In step A3, Canny dual-threshold segmentation is used to extract the field ridge boundaries: High threshold = 1.5: Corresponds to a slope of approximately 56.3°. Grids exceeding this value are identified as field ridges (strong edges).

[0076] low threshold = 0.8: corresponds to a slope of approximately 38.7°. Between and The grid between them is only identified as a field ridge when it is connected to a high-threshold grid.

[0077] The threshold was determined by measuring the typical slope range of the terraced field ridges (45°~70°) in the field, combining the noise level of the DEM data (gradient noise of about 0.14 due to an elevation error of about 0.1 meters), and optimizing it through experiments.

[0078] A4: BeiDou grid coding mapping and terrain factor calculation Align the terraced field vector map with the BeiDou grid coding system, with a grid size of 0.0001 degrees longitude difference × 0.0001 degrees latitude difference (approximately 10 meters × 10 meters), forming a two-level spatial index of field-grid. Calculate the following topographic factors for each grid: slope : Calculated by fitting a local plane; Slope The direction of the projection of the slope normal onto the horizontal plane; Terrain Humidity Index (TWI): TWI = ln(α / tanβ), where α is the catchment area per unit contour length and β is the slope; Solar radiation correction factor = 1 + 0.2·cos( - )·tan( ),in It is the azimuth of the sun.

[0079] The solar azimuth angle is the angle measured from true north (N) of the observation point, rotating clockwise along the horizon, to the direction of the sun's rays projected onto the horizon. Its range is 0° to 360°, where 0° indicates the sun is due north (possibly only in high-latitude regions near the North Pole), 90° indicates the sun is due east, 180° indicates the sun is due south (at noon in the Northern Hemisphere), and 270° indicates the sun is due west.

[0080] Step 2: The drone flies at a fixed altitude relative to the ground along the terrain contour lines to ensure consistent ground resolution; at the same time, the multispectral image is stitched and the index is calculated in real time on the airborne terminal, and only the fused grid-level data is uploaded.

[0081] Specifically, including: B1: Low-altitude drone swarm terrain-following flight path planning In terraced field scenarios, the terrain is dramatically undulating. If a fixed altitude is used for flight, the drone will be too close to the ground when flying over ridges (risk of collision), and too far from the ground when flying over valleys (reduced image resolution). The core idea of ​​terrain-following flight is that the drone maintains a constant altitude relative to the ground, flying along the contour lines of the terrain, thereby ensuring: Safety: Always maintain a safe distance from the ground; Data consistency: The ground resolution (GSD) of the multispectral imagery remains constant; Stable overlap: The overlap rates in both heading and lateral directions are unaffected by terrain undulations.

[0082] Based on the DEM data generated in step A3, a terrain-following flight path is planned for each low-altitude rotary-wing UAV. The UAV's flight altitude remains fixed relative to the ground. = 50 meters, actual flight altitude is: Where δ(x,y) is the safety margin: 10 meters on steep slopes and 5 meters on gentle slopes. The flight path design ensures an 80% forward overlap and a 70% lateral overlap to guarantee the accuracy of the 3D reconstruction. This refers to the ground elevation.

[0083] B2: Multispectral Data Acquisition and Real-time Preprocessing The drone swarm synchronously collects multispectral data following the terrain and flight path. Each drone records its precise pose at the moment of exposure in real time (centimeter-level positioning via RTK-GPS and attitude angle via IMU), and packages and stores the raw images and pose data. Simultaneously, the onboard edge computing unit performs pixel-by-pixel radiometric calibration and atmospheric correction, generating reflectance image patches with geographic tags, but does not perform cross-drone stitching.

[0084] The airborne edge computing unit performs the following preprocessing in real time: Radiation calibration: converting DN values ​​to reflectivity Atmospheric correction: Dark target subtraction is used to eliminate atmospheric effects. Image stitching: Real-time generation of orthophoto mosaic based on POS data.

[0085] B3: After all drones complete their flight path missions, they transmit the pre-processed image blocks to the edge computing base station via a high-speed data transmission link (5G or high-bandwidth self-organizing network). The base station then performs joint adjustment and orthophoto mosaicking based on POS data on the ground to generate a reflectance mosaic map of the entire terraced field area. Vegetation index calculation and crop moisture inversion The following key indices are calculated in real time on the airborne end: Normalized Difference Vegetation Index (NDVI): ,in, Near-infrared reflectivity Red light band reflectivity; Normalized Moisture Index (NDWI): ,in, Green light band reflectivity Near-infrared reflectivity; Crop water retrieval model (simplified forward inversion based on the PROSAIL radiative transfer model): Where CWC represents the canopy water content (mm). Canopy temperature (obtained from a thermal infrared sensor). Parameter values ​​(calibrated in a cornfield): =-0.15、 =0.92、 =0.28、 =0.18、 =0.006. The physical basis of NDWI is the absorption characteristics of the spectrum by vegetation moisture: Green light band: Not sensitive to changes in vegetation moisture, used as a reference benchmark. Near-infrared band: Sensitive to vegetation moisture content—the higher the moisture content, the stronger the near-infrared absorption and the lower the reflectivity. NDWI uses the difference between these two bands to quantify vegetation moisture status: When vegetation has sufficient moisture: Lower (moisture absorption). Relatively stable → NDWI is positive. When vegetation is under water stress (water shortage): As water absorption decreases, the NDWI value decreases or even turns negative.

[0086] Value range: [-1, 1] Ratio characteristics: Through normalization (dividing by the sum), the influence of factors such as solar altitude angle, atmospheric conditions, and terrain shadows is partially eliminated. Nonlinearity: NDVI has a nonlinear relationship with vegetation cover, being sensitive at low cover and easily saturated at high cover.

[0087] B4: Quadruped Robot Dog Ground Fine Sampling and Path Planning The quadruped robot dog has a foot force sensing system, with a six-dimensional force / torque sensor (ATIMini45 or similar product) installed on each foot, with a sampling frequency of 1kHz; The sensors are connected to the robot dog's main controller (Intel NUC+ real-time kernel) via an EtherCAT bus; When a hard surface is detected, the robot dog marks the grid as an obstacle (cost value 255) in the local cost map, triggering the DWA local path planner to replan the obstacle avoidance path; When soft soil is detected, it is marked as a high-risk area (cost value 150), and the planner will prioritize detouring; if detouring is not possible, the walking speed will be reduced (from 1.0m / s to 0.3m / s) and the foot contact area will be increased; The quadruped robot dog, equipped with an insertable soil moisture sensor (based on the TDR time-domain reflectometry principle) and a portable weather station, performs ground sampling according to a grid-based stratified sampling strategy. At least three representative character grids should be drawn from each field (one for each of the high, medium, and low water zones).

[0088] Within each sampling grid, the robot dog autonomously selects 3 sampling points (avoiding rocks and crop roots).

[0089] Measurement point depth: 10cm, 20cm, and 30cm (three layers).

[0090] The robot dog uses foot-based force sensors to determine soil hardness and automatically avoids areas with rocks. Sampling data includes: soil volumetric water content. Soil temperature and soil electrical conductivity (EC).

[0091] B5: Adaptive Fusion of Grid-Level Multi-Source Data In terraced field environments, a single data source has limitations: Remote sensing inversion data ( The spatial coverage is complete (each grid cell has a value), but its accuracy is unstable due to the influence of the atmosphere, clouds, and terrain shadows. Ground measurement data ( High accuracy (sensor in direct contact with soil), but sparse sampling (only some grid cells have values). The core idea of ​​the Bayesian weighted average method is to dynamically allocate weights based on the uncertainty (variance) of the two data sources, with the data source with lower uncertainty (higher accuracy) receiving greater weight.

[0092] The CWC data retrieved from the drone and the actual measurements from the robot dog The data is uniformly mapped onto the BeiDou grid. For each grid... The state of soil moisture after fusion Calculated using the Bayesian weighted average method: in: The surface soil moisture is retrieved from remote sensing and corrected for ground data. To measure soil moisture for the robot dog This is the variance of remote sensing inversion (calculated based on image quality). The ground sampling variance (calculated based on the coefficient of variation of the sampling points). Remote sensing inversion variance It is dynamically determined by the following formula: in Based on the basic variance (taken as 0.02). , This represents the pixel percentage of clouds and shadows within the grid. The penalty coefficients for clouds and shadows are λ1=2.0 and λ2=1.5.

[0093] Step 3: Using the differences in slope aspect radiation and evapotranspiration caused by mountain winds as input features of the model, the model parameters are updated in real time using an online learning mechanism to adaptively adjust irrigation decisions.

[0094] C1: Dynamic Calculation of Crop Water Requirements For each grid The reference crop evapotranspiration ET0 was calculated using the Penman-Monteith formula: in: Net radiation (W / m 2 The crop canopy temperature, calculated from the crop canopy temperature retrieved in step B3 and corrected for slope aspect, is as follows: Where G is the soil heat flux (W / m³) 2(Approximately 0.1) , T represents the average air temperature (°C), measured by a weather sensor carried by the robot dog, taking into account the decrease in elevation. Wind speed at 2 meters (m / s), corrected for terrain: z0 is the surface roughness; The saturated vapor pressure is (kPa). This represents the actual water vapor pressure (kPa). Δ is the slope of the saturated water vapor pressure curve (kPa / ℃). γ is the wet / dry constant (kPa / ℃).

[0095] Crop coefficient Adjustments are made dynamically based on crop growth stages. The current growth stage is determined using phenological information retrieved from time-series UAV imagery and the cumulative growth days method. When GDD reaches the threshold, a table lookup and update are performed. value.

[0096] Step C1 calculates the value of each grid cell in hourly increments (on the hour). The value will be calculated and stored in the grid status database, with fields including: grid ID, timestamp, etc. , T wait.

[0097] C2: Soil Moisture Dynamics Prediction Based on Particle Filtering A particle filter algorithm is used to dynamically predict soil moisture in a grid, solving the problem of difficulty in obtaining parameters for traditional mechanistic models. State equations and observation equations are as follows: Particle filtering is a sequential Monte Carlo method based on Bayesian theory, suitable for state estimation of nonlinear, non-Gaussian systems. In soil moisture prediction: Equation of state: describes the change of soil moisture over time (water balance). Observation equation: describes the relationship between sensor observations and actual soil moisture. Dynamic prediction is achieved by approximating the posterior probability distribution of soil moisture using a large number of weighted random particles (each particle representing a possible soil moisture value).

[0098] Equation of state (soil water balance): ,in, Let t be the soil moisture content. This refers to the rainfall during the specified period. Irrigation volume during the period This represents crop evapotranspiration over a given period of time. This represents the amount of deep seepage within a given time period. Surface runoff during the time period This is process noise; Observation equation: in For process noise, To observe noise.

[0099] Particle filtering execution steps: Initialization: Generate N=500 particles {θ (i) 0}, weight w (i) 0 = 1 / N Prediction: Update θ for each particle based on the equation of state. (i) _{t+1|t}, Update: When a new observation y_{t+1} is obtained, calculate the particle weights: ,in, Prior particle weights, , For a one-step prediction, R is the observation noise variance, 2π is twice the value of pi, and exp() is an exponential function. Resampling: Resample according to weights to avoid particle degradation. State estimation: , The particle weight update formula is the core of the particle filtering algorithm—it re-evaluates the degree of matching between each particle (i.e., each possible soil moisture state) and the true state based on newly acquired observation data.

[0100] The particle filter prediction module in step C2 executes every 30 minutes, reading the latest data from the database. Based on soil moisture values ​​and weather forecast data, predict soil moisture changes 24 hours in advance.

[0101] C3: Spatiotemporal Dynamic Management Zone Division Fuzzy C-means clustering is a soft clustering method that allows each sample (grid) to belong to multiple categories (administrative areas) with varying degrees of membership. This differs from hard clustering (such as K-means) where each sample can only belong to one category.

[0102] In the delineation of irrigation management zones, this fuzziness has a clear physical meaning: soil moisture and crop growth change continuously in space, and the boundaries between fields are often gradual rather than abrupt. A grid may simultaneously possess characteristics of two adjacent management zones, and fuzzy clustering can more realistically reflect this transitional nature.

[0103] Based on the predicted 24-hour soil moisture distribution using particle filtering, fuzzy C-means clustering algorithm is employed to dynamically divide irrigation management zones. The clustering feature vectors include: Current soil moisture Predicting soil moisture Topographic humidity index (TWI) and crop coefficient , Clustering objective function: Where c is the number of clusters (3-5), m is the fuzzy index (2), n is the number of samples, and j is the category index. For membership degree, As cluster center, The eigenvectors are used. The membership matrix of each grid is obtained through iterative optimization, and the grids are assigned to the management areas with the highest membership degrees.

[0104] C4: Irrigation Urgency Calculation and Prescription Map Generation For each grid Calculate the Irrigation Urgency Index (IUI): in: Field holding capacity The wilting coefficient, For the predicted soil moisture 24 hours later, Evapotranspiration of grid crops The crop sensitivity coefficient is set at 1.2 for corn and 1.0 for wheat. The IUI distribution of each grid within a field is statistically analyzed.

[0105] For fields Calculate the comprehensive irrigation decision index : in Grid area weights. Decision rule: like If the score is greater than 1.5, the field will be added to the irrigation queue, with priority... = Otherwise, the field will not be irrigated; For fields included in the irrigation program, the target irrigation volume is... Determined by the following formula: in: The target soil moisture content is 0.8θ_fc. This represents the average soil moisture content of the field. The planned wetting layer depth is 0.3-0.5 meters. The irrigation water utilization coefficient is taken as 0.85-0.9.

[0106] Step 4: The upper layer performs global path planning based on the digital chassis, while the lower layer uses terrain-adaptive synchronous control law to achieve local trajectory tracking.

[0107] Specifically, it includes: D1: Kinematic and Dynamic Modeling of Sprinkler Irrigation Machines Each sprinkler system consists of a powered main tower and several driven spans. A kinematic model of the sprinkler system is established: Where (x, y) is the position of the main tower crane, θ is the heading angle, and v is the linear velocity. ω is the angular velocity.

[0108] The dynamic model considers the gravity component of the slope: in As the driving force, For rolling resistance, For slope, For the direction of travel, It is a slope.

[0109] D2: Cross-plot global path planning based on an improved A* algorithm A 3D accessibility cost map is the foundation of path planning. It quantifies terrain features into access costs, guiding algorithms to find the optimal transfer path. In a terraced field environment, sprinkler irrigation machines need to traverse complex terrain when moving between different plots: flat fields, steep ridges, abrupt elevation changes, etc. The 3D accessibility cost map quantifies the accessibility difficulty of each grid into a cost value C(x,y); the higher the cost, the more difficult the location is to access. Path planning algorithms (such as A*) will select the path with the lowest cumulative cost when searching for a path.

[0110] The cost function of the A* algorithm is: , g(n): The actual travel cost from the starting point to the current node n (accumulated from the three-dimensional traversability cost map in step D2). h(n): A heuristic function that estimates the minimum cost from the current node n to the target node. The heuristic function guides the search direction, prioritizing the exploration of promising paths that may lead to the target and avoiding blind searches. When the heuristic function satisfies admissibility (i.e., the estimated cost is no greater than the actual cost), the A* algorithm guarantees finding the optimal path.

[0111] When a sprinkler needs to move from the current field to a target field, a 3D accessibility cost map is constructed based on the DEM generated in step A3: in: Local elevation changes Maximum obstacle clearance height of sprinkler irrigation machine; surface roughness. Elevation gradient magnitude, Slope angle; The first penalty is for steep slopes (slope weighting coefficient λ1=2.0). The second penalty exceeds the obstacle-crossing height of the sprinkler machine. (0.3) Field ridges (elevation change weighting coefficient λ2=5.0) The third penalty is for surface roughness (roughness weighting coefficient λ3=1.0). An improved A* algorithm is used to search for the optimal transition path on the cost map, with node expansion considering eight-neighbor directions. The heuristic function, a core component of the A* algorithm, estimates the cost from the current node to the target node, guiding the search direction. The heuristic function employs an improved Euclidean distance that considers elevation. ,in , Current node planar coordinates, , Target node planar coordinates Current node elevation, The target node elevation is set, and α is the elevation weighting coefficient, which is set to 0.5. The algorithm output is a sequence of path points. m is the total number of path points, and each path It is a three-dimensional coordinate vector: , The kth path point, It is the x-coordinate of the path point. It is the y-coordinate of the path point. This refers to the elevation of the path points. After the A* algorithm completes its search on the 3D drivability cost map, it generates an optimal path from the starting point to the target point. This path consists of a series of discrete path points, each containing planar coordinates and elevation information, forming the actual trajectory of the sprinkler machine.

[0112] D3: Field Operation Path Planning When a sprinkler irrigation machine operates within a single terraced field, it needs to cover the entire field according to a certain pattern, while ensuring uniform irrigation without overlap or omissions. Parametric equations represent the continuous three-dimensional operating trajectory as a function of the path arc length s, providing reference input for subsequent path tracking control.

[0113] For a single terraced field Based on its boundary shape and sprinkler arrangement, plan the sprinkler operation path: For rectangular or near-rectangular fields: use a zigzag reciprocating path, with the path spacing equal to the spray width (15-20 meters). Irregular plots: Employ a spiral path with inward contraction of boundaries. The job path is represented by parametric equations: ,in This is the total path length. The operation path function, s is the path arc length parameter, x(s) is the x-coordinate function, y(s) is the y-coordinate function, and θ(s) is the heading angle function; D4: Distributed cooperative control based on leader-follower The system adopts a leader-follower distributed collaborative control architecture, designating one sprinkler as the leader and the other sprinklers as followers. The leader is responsible for global path tracking, while the followers are responsible for maintaining their relative positions with the leader.

[0114] Master Navigator Control Law (based on chain control theory): in:( , , () represents the lateral deviation, longitudinal deviation, and heading deviation relative to the reference path. , For reference speed The control gain was adjusted through field testing: =0.8, =1.2, =2.5, Follower control law (based on cascaded PI control): in: The azimuth angle of the follower relative to the navigator. Distance deviation: = - , This is the actual spacing. For the desired spacing, For heading deviation; PI controller parameters: 1 = 0.5 1 = 0.1 2 = 0.8 2 = 0.15.

[0115] D5: Terrain Adaptive Synchronization Control Compensation When the sprinkler is on a slope, the control law incorporates a terrain compensation term: Speed ​​compensation: Where μ is the rolling resistance coefficient (taken as 0.15). For slope, For the direction of travel, For slope direction. The compensation item ensures increased driving force when going uphill and limits speed to prevent loss of control when going downhill.

[0116] Synchronous control across spans: The relative torsion angle between each span of the sprinkler system is detected by angle sensors (sensor selection: magnetic absolute encoder (AS5048A, resolution 0.05°)). The drive wheel speed is adjusted via frequency conversion, with the control target being |δθ| < 3°. Control law: in The synchronization gain is set to 0.3. This is the velocity correction amount for the i-th span.

[0117] D6: Variable irrigation control based on water pressure feedforward-feedback In terraced irrigation in mountainous areas, water pressure varies drastically with elevation (approximately 0.1 MPa decreases for every 10 meters increase in elevation). Additionally, factors such as pipe friction and the simultaneous operation of multiple sprinklers also cause pressure fluctuations. PID feedback control, by real-time detection of the deviation between the actual flow rate and the target flow rate, calculates the required sprinkler opening adjustment, enabling the actual flow rate to quickly and stably track the target value.

[0118] To address the issue of large water pressure fluctuations in mountainous areas, a feedforward-feedback composite control strategy is adopted to adjust the sprinkler head flow rate. Feedforward control: Calculate the theoretical water pressure based on the current elevation z. : in For water source pressure, For water source elevation, This refers to losses along the pipeline.

[0119] Feedback control: based on the measured values ​​of the flow sensor With target value PID control is applied to the deviation: ,in, It is the proportional gain coefficient. It is the integral gain coefficient. It is the differential gain coefficient. It is the target traffic. This is the actual measured traffic volume. It's a flow deviation. It is the deviation integral. It is the differential of the deviation, and t is time; Parameter values: =0.8, =0.15, =0.05.

[0120] Final PWM duty cycle: in This is the maximum flow rate of the nozzle. This is the rated working pressure.

[0121] This also includes the navigator being responsible for global path tracking, and its walking speed. The follower speed is determined by the control law in step D4. The control law is determined by step D4, but fine-tuning at the grid level is allowed. When the follower is about to enter a grid At that time, based on the target irrigation volume of the grid. Calculate the ideal speed : in This is the overlap coefficient (considering spray overlap). It's the nozzle flow rate. It is the width of the spray pattern. That is the grid width. If | - If the speed is greater than 0.1 m / s, the follower sends a speed adjustment request to the navigator. The navigator, while maintaining overall formation stability, adjusts the global speed appropriately. The core requirement of variable irrigation is that the amount of water applied to each grid by the sprinkler as it passes through it must be equal to the target irrigation amount specified in the irrigation prescription map. The water volume sprayed per unit time by a sprinkler irrigation machine is determined by the nozzle flow rate. The area covered per unit time is determined by the spray width. The walking speed (vv) is the determining factor. Therefore, given the nozzle flow rate and spray width, the walking speed determines the amount of water applied per unit area—the slower the speed, the more water is applied; the faster the speed, the less water is applied.

[0122] The ideal speed formula is based on this physical relationship, and it calculates the amount of irrigation needed to achieve the target amount. The required walking speed.

[0123] Speed ​​fine-tuning is limited to ±15%. Exceeding this range will trigger formation reorganization (followers will temporarily leave the formation and rejoin the formation after completing the current grid).

[0124] The ideal speed formula is based on this physical relationship, and it calculates the amount of irrigation needed to achieve the target amount. The required walking speed.

[0125] Step 5: Use the quadruped robot dog as a mobile relay node to automatically move to the optimal position in the communication blind spot to forward commands; at the same time, design a three-level degradation mode to ensure that the system can still operate to a limited extent when the communication part is interrupted.

[0126] Specifically, it includes: E1: Construction of Heterogeneous Mesh Self-Organizing Networks The system uses the LoRaMesh protocol to build a self-organizing network, and the network nodes include: Fixed nodes: Edge computing base stations (gateways) Backbone mobile nodes: Low-altitude unmanned aerial vehicle (UAV) swarm (airborne relay) Ground mobile nodes: quadruped robot dog (ground relay), sprinkler cluster (terminal) Each node periodically broadcasts a beacon frame containing its node ID, location, remaining battery power, and link quality. The routing table is dynamically maintained using the OLSR (Optimized Link-State Routing) protocol.

[0127] Hardware connection: The main controller (STM32H7 series MCU) of each sprinkler is connected through two independent SPI interfaces. The LoRa RF module (SX1262, 470MHz) is used for data communication, and the UWB ranging module (DW1000, channel 2) is used for relative positioning. The two modules use independent clock sources and power management to avoid mutual interference.

[0128] Protocol interaction mechanism: Timing slotting: The time axis is divided into 10ms time slots. Odd-numbered time slots are used for LoRa data transmission and reception, and even-numbered time slots are used for UWB ranging.

[0129] Synchronization Trigger: The main controller of the sprinkler machine generates a synchronization pulse through the PWM output pin, which simultaneously triggers the time slot counters of the LoRa module and the UWB module to reset.

[0130] Data fusion: The relative distance and azimuth angle measured by the UWB module are transmitted to the LoRa protocol stack through internal shared memory and uploaded along with the status data.

[0131] E2: Dynamic Relay Node Deployment Optimization In mountainous terraced fields, due to terrain obstruction, sprinkler irrigation machines may be in communication dead zones (low RSSI or high packet loss rate). In this case, the system dispatches a quadruped robot dog to move to an optimal relay position, acting as a communication relay node to forward data between the edge computing base station and the sprinkler irrigation machine.

[0132] The optimal relay location needs to satisfy two objectives simultaneously: The link quality from the robot dog to the base station should be as good as possible (RSSI should be as high as possible). The link quality from the robot dog to the sprinkler machine should be as good as possible (RSSI should be as high as possible). This is a classic min-max optimization problem—maximizing the quality of the worse link between two links to ensure overall communication reliability.

[0133] When a communication link quality degradation is detected in the area where the sprinkler is located (RSSI < -85dBm or packet loss rate > 5%), the system automatically dispatches the nearest quadruped robot dog to the optimal relay position. The optimal position is determined by solving the following optimization problem: ,in, It is a candidate position for the robot dog. It refers to the link quality from the robot dog to the base station. It's about the quality of the link from the robot dog to the sprinkler machine. Minimum value function It's a maximization problem.

[0134] Constraints: The robot dog's travel distance must be less than the battery life limit, and the location must be within a passable area. The particle swarm optimization algorithm is used to solve the problem in real time, with a population size of 20 and 50 iterations, and the optimal relay point coordinates are output.

[0135] E3: Three-level communication degradation mechanism Each node monitors the communication link quality in real time and triggers a degradation based on the Comprehensive Communication Quality Index (CQI): Where RSSI_max = -50dBm, SNR_max = 30dB, w1 = 0.4, w2 = 0.3, and w3 = 0.3. Maximize the quality of the weaker link between the two links to ensure overall communication reliability.

[0136] Level 1 Degradation (CQI < 0.6): The sprinkler is disconnected from the edge base station, but the sprinklers maintain LoRaMesh interconnection. It enters cluster autonomy mode, where the current lead sprinkler assumes local decision-making functions and continues operation based on a pre-cached irrigation prescription map. The operation status is cached locally.

[0137] Level 2 Degradation (CQI < 0.4): Communication between sprinkler irrigation units is interrupted, and each sprinkler enters standby autonomous mode. It continues operation based on the currently stored irrigation prescription for the current field and the pre-programmed emergency path, remaining in place after completing the current field's task. Simultaneously, a robot dog is activated to actively search for and move to the disconnected area to attempt to restore communication.

[0138] Level 3 Degradation (CQI < 0.2 or no communication for 5 consecutive minutes): Trigger the safety stop procedure: stop walking, shut off the nozzle, record the current breakpoint coordinates and time, and send an emergency distress signal via BeiDou short message (only for models equipped with BeiDou terminals).

[0139] E4: Resuming interrupted downloads and task recovery Once communication is restored, each node uploads its work logs (including irrigated areas, traveled paths, and abnormal events) to the edge computing base station during the interruption. A timestamp vector method is used to resolve concurrent conflicts: each node maintains a local vector clock, and upon receiving the logs, the base station merges the global task progress using a distributed consensus protocol, re-plans the remaining tasks, and distributes them.

[0140] Step 6: Design a voice interaction interface and a predictive maintenance system. Control the system through natural language voice commands. The system uses machine learning to predict equipment failures and provide early warnings.

[0141] F1: Voice Interaction Interface Design Develop a lightweight speech recognition module that supports dialect adaptation. The preset command set includes: Start irrigating plot XX, check soil moisture, stop all equipment, and report the malfunction. Voice commands are processed by the edge computing base station's end-side speech recognition model (based on convolutional neural network + connection-time classification), requiring no network connection and with a response time of less than 1 second.

[0142] F2: Visualized Digital Twin Interface Build digital twin models of the terraced fields on mobile apps and the web, and display them in real time: Real-time video stream from drones Soil moisture distribution heat map Real-time location and trajectory of the sprinkler irrigation machine Device battery level and communication status. The interface adopts a minimalist design, retaining only the core control buttons, and supports one-click activation of preset irrigation modes.

[0143] F3: Equipment Health Monitoring and Predictive Maintenance The Health Index (HI) is a core indicator for predictive maintenance. It comprehensively assesses the degree of equipment degradation by monitoring multiple operating parameters and comparing them to baseline values ​​under normal conditions. The HI value ranges from 0 to 1. HI ≈ 1.0: The equipment is in good health, and all indicators are normal. HI ≈ 0.8~0.9: Slight degradation, monitoring is recommended. HI ≈ 0.6~0.7: Moderate degradation, scheduled maintenance. HI < 0.6: Severe degradation, requiring immediate maintenance. HI < 0.4: Extremely high risk of failure; shutdown and maintenance are recommended. Collect operational data from each device (motor current, vibration, temperature, battery cycle count) to construct a device health index: m is the number of features, and j is the feature index. It is the square of the characteristic deviation. It is the weighted squared deviation. It is a normalization factor; in For the current eigenvalue, This is the baseline value within the normal range. The feature weights are defined as follows. An isolation forest algorithm is used to detect anomalies: randomly selected samples are used to construct isolation trees, and the anomaly score s(x,n) is determined by the path length. When HI < 0.7 or an anomaly is detected, the system automatically sends a warning message to the maintenance personnel's mobile phone, along with a fault cause analysis and repair suggestions.

[0144] F4: Integrated Water and Fertilizer Expansion Interface The system has a reserved interface for a fertilizer applicator, supporting automatic liquid fertilizer injection. Fertilization decisions are based on: Soil nutrient sensor (optional configuration) measures nitrogen, phosphorus, and potassium content, crop nutrient requirements at different growth stages, and user-preset fertilization plans. Coordinated control of fertilizer and irrigation amounts: delaying application after irrigation begins Start the fertilizer pump within minutes to ensure that fertilizer enters the root zone with the water, reducing deep seepage losses.

[0145] It should be understood that the above embodiments are only for illustrating the technical concept and features of the present invention, and are intended to enable those skilled in the art to understand the content of the present invention and implement it accordingly. It should not be considered that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, several simple deductions or substitutions can be made without departing from the concept of the present invention. All equivalent changes or modifications made in accordance with the spirit and essence of the present invention should be covered within the protection scope of the present invention.

Claims

1. A collaborative monitoring and adaptive cascade sprinkler irrigation control method, characterized in that: include: Step 1: Using terrain slope and aspect as rigid constraints, establish a three-dimensional mapping relationship between remote sensing image pixels and ground sensor measurement points; Step 2: The UAV flies at a fixed altitude relative to the ground along the terrain contour lines, while the onboard device performs real-time stitching and index calculation of multispectral images and uploads the fused grid-level data. Step 3: Using the differences in slope aspect radiation and evapotranspiration caused by mountain winds as input features of the model, the model parameters are updated in real time using an online learning mechanism to adaptively adjust irrigation decisions; Step 4: The upper layer performs global path planning based on the digital chassis, while the lower layer uses terrain-adaptive synchronous control law for local trajectory tracking; Step 5: Use a quadruped robot dog as a mobile relay node to automatically move to the optimal position in the communication blind spot to forward instructions; at the same time, design a three-level degradation mode to ensure that the system can still operate to a limited extent when the communication part is interrupted.

2. The collaborative monitoring adaptive cascade sprinkler irrigation control method according to claim 1, characterized in that: It also includes step six, designing a voice interaction interface and a predictive maintenance system, which controls the system through natural language voice commands. The system uses machine learning to predict equipment failures and provide early warnings.

3. The collaborative monitoring and adaptive cascade sprinkler irrigation control method according to claim 1, characterized in that: Step one specifically includes: A1: High-altitude UAV LiDAR terrain scanning: A high-altitude fixed-wing UAV equipped with a linear frequency modulated continuous wave LiDAR performs a full-coverage scan of the terraced fields following a terrain-following flight path to obtain raw point cloud data P={( , , , Point cloud density target: ≥15 points / square meter; A2: Gradiently Enhanced Triangular Mesh Filtering Based on Slope Adaptation A slope-adaptive progressive densification triangular mesh filtering algorithm is used to distinguish between ground points and vegetation points: an initial sparse triangular mesh is constructed, and the distance from each point P to be judged to the triangular mesh is calculated. and slope angle , in The elevation of the point to be determined. It is the elevation interpolated within the triangulation network. ， like < and < If P is determined to be a ground point, it is added to the triangulation network, where the dynamic threshold is... Adaptive adjustment based on local topographic standard deviation: , The basic slope angle threshold is set to 15°; k: adjustment coefficient, with a value range of 0.5-1.5, which is adaptively adjusted according to the roughness of the terrain; Local area elevation standard deviation, calculation window size 15m × 15m; Parameter values: = 15°, k = 0.8, The standard deviation of elevation within a 5-meter radius of P; Intelligent extraction of terraced field boundaries Based on the filtered ground point cloud, a digital elevation model was constructed using Kriging interpolation with a grid resolution of 0.5 m × 0.5 m; the elevation gradient magnitude of each grid cell was calculated. Where x and y are planar coordinates, Let x be the partial derivative of elevation in the x-direction. The partial derivative of elevation in the y-direction; A significant peak appears in G(x,y) at the field ridge, so Canny dual-threshold segmentation is used: high threshold... = 1.5 meters, low threshold = 0.8 meters, extract the field boundary lines, perform Douglas-Puk thinning on the boundary lines, and generate a vector map of the terraced fields. Each field contains the following attributes: boundary coordinate sequence, average elevation, average slope, and aspect. A4: BeiDou grid coding mapping and terrain factor calculation Align the terraced field vector map with the BeiDou grid encoding system, and calculate the following terrain factors for each BeiDou grid: slope : Calculated by fitting a local plane; Slope The direction of the projection of the slope normal onto the horizontal plane; Terrain Humidity Index (TWI): TWI = ln(α / tanβ), where α is the catchment area per unit contour length and β is the slope; Solar radiation correction factor = 1 + 0.2·cos( - )·tan( ),in It is the azimuth of the sun.

4. The collaborative monitoring adaptive cascade sprinkler irrigation control method according to claim 3, characterized in that: Step two specifically includes: B1: Low-altitude drone swarm terrain-following flight path planning Based on the DEM data generated in step A3, a terrain-following flight path is planned for each low-altitude rotary-wing UAV; the UAV's flight altitude remains a fixed relative altitude to the ground. = 50 meters, actual flight altitude is: Where δ(x,y) is the safety margin: 10 meters on steep slopes of the field ridge and 5 meters on gentle slopes; B2: Multispectral Data Acquisition and Real-time Preprocessing The drone swarm synchronously collects multispectral data by following the terrain and flight path. Each drone records the precise pose at the exposure moment in real time and packages and stores the original image and pose data. At the same time, the airborne edge computing unit completes pixel-by-pixel radiometric calibration and atmospheric correction to generate reflectance image blocks with geographic tags. B3: After all the drones have completed their flight path missions, they will transmit the pre-processed image blocks to the edge computing base station via a high-speed data transmission link. The edge computing base station will then perform joint adjustment and orthophoto mosaic based on POS data on the ground to generate a reflectance mosaic map of the entire terraced area. Vegetation index calculation and crop moisture inversion The following key indices are calculated in real time on the airborne end: Normalized Difference Vegetation Index (NDVI): ,in, Near-infrared reflectivity Red light band reflectivity; Normalized Moisture Index (NDWI): ,in, Green light band reflectivity Near-infrared reflectivity; Crop water inversion model: Where CWC is the canopy water content, Canopy temperature; B4: Quadruped Robot Dog Ground Fine Sampling and Path Planning The quadruped robot dog has a foot force sensing system, including a six-dimensional force / torque sensor installed on each foot. When a hard surface is detected, the robot dog marks the Beidou grid as an obstacle in the local cost map, triggering the DWA local path planner to replan the obstacle avoidance path. When soft soil is detected, it is marked as a high-risk area, and the planner will prioritize detouring; if detouring is not possible, the walking speed will be reduced and the contact area of ​​the feet with the ground will be increased. A quadruped robot dog carrying an insertable soil moisture sensor and a portable weather station conducted ground sampling using a stratified sampling strategy based on the BeiDou grid network. The robot dog uses a foot-based force sensing system to detect and determine soil hardness, automatically avoiding areas with rocks. Data collected includes soil volumetric water content. Soil temperature, soil electrical conductivity (EC); B5: Adaptive Fusion of Grid-Level Multi-Source Data The CWC data retrieved from the drone and the actual measurements from the robot dog The data is uniformly mapped to the BeiDou grid, and for each BeiDou grid... The state of soil moisture after fusion Calculated using the Bayesian weighted average method: in: The surface soil moisture is retrieved from remote sensing and corrected for ground data. To measure soil moisture for the robot dog For remote sensing inversion variance, For ground sampling variance, Remote sensing inversion variance It is dynamically determined by the following formula: ， in Based on the variance, , This represents the pixel percentage of clouds and shadows within the BeiDou grid. The penalty coefficients for clouds and shadows are λ1=2.0 and λ2=1.

5.

5. The collaborative monitoring adaptive cascade sprinkler irrigation control method according to claim 4, characterized in that: Step three includes: C1: Dynamic Calculation of Crop Water Requirements For each grid The reference crop evapotranspiration ET0 was calculated using the Penman-Monteith formula: in: Net radiation is calculated from the crop canopy temperature retrieved in step B3, and corrected for slope aspect: Where G is the soil heat flux. T represents the average temperature, measured by a weather sensor carried by the robot dog, taking into account the decrease in elevation. Wind speed at a height of 2 meters, corrected for terrain: z0 is the surface roughness; The saturated vapor pressure, This is the actual water vapor pressure. Δ is the slope of the saturated water vapor pressure curve. γ is the wet / dry constant; Crop coefficient Based on dynamic adjustments according to crop growth stages, the current growth stage is determined using phenological information retrieved from time-series UAV imagery and the cumulative growth days method. When GDD reaches the threshold, a table lookup and update are performed. value; Step C1 calculates the value of each grid cell on an hourly basis. The value will be calculated and stored in the grid status database, with fields including: grid ID, timestamp, etc. , T ; C2: Soil Moisture Dynamics Prediction Based on Particle Filtering The particle filter algorithm is used to dynamically predict soil moisture in gridded soil, solving the problem of difficulty in obtaining parameters in traditional mechanistic models. The state equation is: ,in, Let t be the soil moisture content. This refers to the rainfall during the specified period. Irrigation volume during the period This represents crop evapotranspiration over a given period of time. This represents the amount of deep seepage within a given time period. Surface runoff during the time period This is process noise; Observation equation: in For process noise, To observe the noise, Particle filtering execution steps: Initialization: Generate N=500 particles {θ (i) 0}, weight w (i) 0 = 1 / N Prediction: Update θ for each particle based on the equation of state. (i) _{t+1|t}, Update: When a new observation y_{t+1} is obtained, calculate the particle weights: ,in, Prior particle weights, , For a one-step prediction, R is the observation noise variance, 2π is twice the value of pi, and exp() is an exponential function. Resampling: Resample according to weights to avoid particle degradation. State estimation: ; C3: Spatiotemporal Dynamic Management Zone Division Based on the predicted 24-hour soil moisture distribution using particle filtering, fuzzy C-means clustering algorithm is employed to dynamically divide irrigation management zones. The clustering feature vectors include: Current soil moisture Predicting soil moisture Topographic humidity index (TWI) and crop coefficient , Clustering objective function: Where c is the number of clusters, m is the fuzzy index, n is the number of samples, and j is the category index. For membership degree, As cluster center, For feature vectors; C4: Irrigation Urgency Calculation and Prescription Map Generation For each grid cell G_ij, calculate the irrigation urgency index IUI: in: Field holding capacity The wilting coefficient, For the predicted soil moisture 24 hours later, Evapotranspiration of grid crops The crop sensitivity coefficient is calculated by statistically analyzing the IUI distribution of each grid within a field, using the field as the unit. For fields Calculate the comprehensive irrigation decision index : in The grid area is the weight; Decision-making rules: like If the score is greater than 1.5, the field will be added to the irrigation queue, with priority... = Otherwise, the field will not be irrigated; For fields included in the irrigation program, the target irrigation volume is... Determined by the following formula: in: Target soil moisture content This represents the average soil moisture content of the field. To plan the depth of the wetting layer, This is the irrigation water utilization coefficient.

6. The collaborative monitoring adaptive cascade sprinkler irrigation control method according to claim 1, characterized in that: Step four includes: D1: Kinematic and Dynamic Modeling of Sprinkler Irrigation Machines Each sprinkler system consists of a powered main tower and several driven spans. A kinematic model of the sprinkler system is established: Where (x, y) is the position of the main tower crane, θ is the heading angle, and v is the linear velocity. Angular velocity; The dynamic model considers the gravity component of the slope: in As the driving force, For rolling resistance, For slope, For the direction of travel, Slope direction; D2: Cross-plot global path planning based on an improved A* algorithm When a sprinkler needs to move from the current field to a target field, a 3D accessibility cost map is constructed based on the DEM generated in step A3: in: Local elevation changes Maximum obstacle clearance height of sprinkler irrigation machine; surface roughness. Elevation gradient magnitude, Slope angle; The first penalty is for steep slopes. The second penalty exceeds the obstacle-crossing height of the sprinkler machine. The ridges of the fields, The third penalty is for surface roughness. An improved A* algorithm is used to search for the optimal transfer path on the cost map. Node expansion considers eight-neighbor directions, and the heuristic function uses an improved Euclidean distance that takes elevation into account. ,in , Current node planar coordinates, , Target node planar coordinates Current node elevation, The target node elevation, where α is the elevation weighting coefficient, and the algorithm output is a sequence of path points. m is the total number of path points, and each path It is a three-dimensional coordinate vector: , The kth path point, It is the x-coordinate of the path point. It is the y-coordinate of the path point. It is the elevation of the path point; D3: Field Operation Path Planning For a single terraced field Based on its boundary shape and sprinkler arrangement, plan the sprinkler operation path: For rectangular or near-rectangular fields: use a zigzag reciprocating path, with the path spacing equal to the spray width. Irregular plots: Employ a spiral path with inward contraction of boundaries. The job path is represented by parametric equations: ,in This is the total path length. The operation path function, s is the path arc length parameter, x(s) is the x-coordinate function, y(s) is the y-coordinate function, and θ(s) is the heading angle function; D4: Distributed cooperative control based on leader-follower A leader-follower distributed collaborative control architecture is adopted, in which one sprinkler machine is set as the master leader and the other sprinkler machines are set as followers. The master leader is responsible for global path tracking, and the followers are responsible for maintaining their relative positions with the leader. Main Navigator Control Law: in:( , , () represents the lateral deviation, longitudinal deviation, and heading deviation relative to the reference path. For reference speed, The control gain was adjusted through field testing: =0.8, =1.2, =2.5, Follower control law: in: The azimuth angle of the follower relative to the navigator. Distance deviation: = - , This is the actual spacing. For the desired spacing, For heading deviation; PI controller parameters: 1 = 0.5 1 = 0.1 2 = 0.8 2 = 0.15; D5: Terrain Adaptive Synchronization Control Compensation When the sprinkler is on a slope, the control law incorporates a terrain compensation term: Speed ​​compensation: , Where μ is the rolling resistance coefficient. For slope, For the direction of travel, For slope aspect, the compensation item ensures increased driving force when going uphill and limits speed to prevent loss of control when going downhill; Synchronous control across spans: The relative torsional angle between each span of the sprinkler system is detected by angle sensors. The speed of the drive wheels is adjusted by frequency conversion speed regulation, with the control target being |δθ| < 3°. The control law is as follows: in For synchronization gain, This is the velocity correction amount for the i-th span. D6: Variable irrigation control based on water pressure feedforward-feedback To address the issue of large water pressure fluctuations in mountainous areas, a feedforward-feedback composite control strategy is adopted to adjust the sprinkler head flow rate. Feedforward control: Calculate the theoretical water pressure based on the current elevation z. : in For water source pressure, For water source elevation, This refers to losses along the pipeline. Feedback control: based on the measured values ​​of the flow sensor With target value PID control is applied to the deviation: ,in, It is the proportional gain coefficient. It is the integral gain coefficient. It is the differential gain coefficient. It is the target traffic. This is the actual measured traffic volume. It's a flow deviation. It is the deviation integral. It is the differential of the deviation, and t is time; Parameter values: =0.8, =0.15, =0.05; Final PWM duty cycle: in This is the maximum flow rate of the nozzle. This is the rated working pressure.

7. The collaborative monitoring adaptive cascade sprinkler irrigation control method according to claim 6, characterized in that: Step four includes: This also includes the navigator being responsible for global path tracking, and its walking speed. The follower speed is determined by the control law in step D4. The control law is determined by step D4, but fine-tuning at the grid level is allowed. When the follower is about to enter a grid At that time, based on the target irrigation volume of the grid. Calculate the ideal speed : in This is the overlap coefficient (considering spray overlap). It's the nozzle flow rate. It is the width of the spray pattern. It is the grid width, if | - If the speed is greater than 0.1 m / s, the follower sends a speed adjustment request to the navigator. The navigator, while maintaining overall formation stability, adjusts the global speed appropriately. .

8. The collaborative monitoring adaptive cascade sprinkler irrigation control method according to claim 6, characterized in that: Step five includes: E1: Construction of Heterogeneous Mesh Self-Organizing Networks The system uses the LoRaMesh protocol to build a self-organizing network, and the network nodes include: Fixed nodes: edge computing base stations, Backbone mobile nodes: Low-altitude drone swarm, Ground mobile nodes: quadruped robot dogs, sprinkler clusters, Each node periodically broadcasts a beacon frame containing the node ID, location, remaining battery power, and link quality, and dynamically maintains the routing table through the OLSR protocol; E2: Dynamic Relay Node Deployment Optimization When a decrease in the communication link quality in the area where the sprinkler is located is detected, the system automatically dispatches the nearest quadruped robot dog to the optimal relay position; E3: Three-level communication degradation mechanism Each node monitors the communication link quality in real time and triggers a degradation based on the Comprehensive Communication Quality Index (CQI): Where RSSI_max = -50dBm, SNR_max = 30dB, w1 = 0.4, w2 = 0.3, w3 = 0.3; Level 1 downgrade CQI < 0.6: The sprinkler is disconnected from the edge base station, but the sprinklers maintain LoRaMesh interconnection. The current lead sprinkler takes on the local decision-making function and continues to operate according to the pre-cached irrigation prescription map. The operation status is cached locally. Level 2 downgrade CQI < 0.4: Communication between sprinkler irrigation machines is interrupted. Each sprinkler irrigation machine enters single-machine autonomous mode and continues to work according to the current field irrigation prescription stored on the machine and the pre-programmed emergency path. After completing the current field, it stands by in place and at the same time activates the robot dog to actively search and move to the disconnected area to try to restore communication. Level 3 downgrade CQI < 0.2 or no communication for 5 consecutive minutes: trigger safety stop procedure: stop walking, shut off the nozzle, record the current breakpoint coordinates and time, and send an emergency distress signal via Beidou short message; E4: Resuming interrupted downloads and task recovery Once communication is restored, each node uploads the job logs from the interruption period to the edge computing base station. The timestamp vector method is used to resolve concurrent conflicts: each node maintains a local vector clock, and after receiving the logs, the base station merges the global task progress through a distributed consensus protocol, re-plans the remaining tasks, and distributes them.

9. The collaborative monitoring adaptive cascade sprinkler irrigation control method according to claim 2, characterized in that: Step six includes: F1: Voice Interaction Interface Design Develop a lightweight speech recognition module that supports dialect adaptation, with preset command sets including: Start irrigating plot XX, check soil moisture, stop all equipment, and report the malfunction. Voice commands are processed by the edge computing base station's end-side speech recognition model (based on convolutional neural network + connection-time classification), requiring no network connection and with a response time of less than 1 second; F2: Visualized Digital Twin Interface Build digital twin models of the terraced fields on mobile apps and the web, and display them in real time: Real-time video stream from drones Soil moisture distribution heat map Real-time location and trajectory of the sprinkler irrigation machine Device battery level and communication status. The interface adopts a minimalist design, retaining only the core control buttons, and supports one-click activation of preset irrigation modes; F3: Equipment Health Monitoring and Predictive Maintenance Collect operational data from each device (motor current, vibration, temperature, battery cycle count) to construct a device health index: m is the number of features, and j is the feature index. It is the square of the characteristic deviation. It is the weighted squared deviation. It is a normalization factor; in For the current eigenvalue, This is the baseline value within the normal range. The feature weights are used. Anomalies are detected using the isolated forest algorithm: randomly selected samples are used to construct an isolation tree, and the anomaly score s(x,n) is determined by the path length. When HI < 0.7 or an anomaly is detected, the system automatically sends a warning message to the maintenance personnel's mobile phone, along with a fault cause analysis and repair suggestions. F4: Integrated Water and Fertilizer Expansion Interface The system has a reserved interface for a fertilizer applicator, supporting automatic liquid fertilizer injection. Fertilization decisions are based on: Soil nutrient sensor (optional) measures nitrogen, phosphorus, and potassium content. Crop nutrient requirements during different growth stages User-preset fertilization plan Coordinated control of fertilizer and irrigation amounts: delaying application after irrigation begins Start the fertilizer pump within minutes to ensure that fertilizer enters the root zone with the water, reducing deep seepage losses.

10. The collaborative monitoring adaptive cascade sprinkler irrigation control method according to claim 8, characterized in that: The optimal position is determined by solving the following optimization problem: ,in, It is a candidate position for the robot dog. It refers to the link quality from the robot dog to the base station. It's about the quality of the link from the robot dog to the sprinkler machine. Minimum value function It's a maximization problem; Constraints: The robot dog's travel distance must be less than the battery life limit, and the location must be within a passable area. The particle swarm optimization algorithm is used to solve the problem in real time, with a population size of 20 and 50 iterations, and the optimal relay point coordinates are output.

Images