A south large irrigation district intelligent system and method based on space-ground-water three-dimensional perception

Abstract

The application discloses a kind of based on air-ground-water three-dimensional perception southern large irrigation district intelligent system and method, system includes air-ground-water three-dimensional data acquisition layer, internet of things platform layer, management area decision layer and production area execution layer, management area decision layer includes multi-source heterogeneous data space-time fusion module, to adaptive Kalman filtering and space-time Kriging interpolation realizes the two-stage space-time fusion of air-based, ground, water body three types of sensor data Generation full irrigation area gridded physical quantity distribution field, and model online calibration and decision confidence evaluation module, to variational bayes inference and extended Kalman filtering realizes model parameter posterior inference and state real-time prediction, when parameter relative uncertainty is over threshold, it is automatically switched to safe scheduling mode, still include based on Saint-Venant equation group's four prediction function hierarchical progressive execution and based on FAO Penman-Monteith formula's water resource supply and demand balance optimization;The method is suitable for the intelligent construction of southern large irrigation district.

Description

Technical Field

[0001] This invention belongs to the field of intelligent irrigation technology, and in particular relates to an intelligent system and method for large-scale irrigation districts in southern China based on three-dimensional perception of air, ground and water. Background Technology

[0002] Large irrigation districts in southern China are generally characterized by complex terrain, dense river networks, diverse water sources, and dispersed irrigation areas. They face challenges in water resource allocation and management, including difficulties in effectively integrating multi-source monitoring data, seasonal drift in model parameters leading to decision-making biases, and rigid sensor acquisition strategies resulting in insufficient or redundant data during critical periods. As the requirements for refined and intelligent irrigation district management continue to increase, a key technological bottleneck restricting the improvement of irrigation district intelligence has emerged: how to unify aerial survey data from UAVs, ground-based fixed-site monitoring data, and water body hydrological and water quality data into a single spatiotemporal framework for collaborative analysis, and how to make reliable decisions when model parameters are uncertain.

[0003] In existing technologies, irrigation district information systems typically employ single-dimensional monitoring methods, such as deploying sensor networks only on the ground or using satellite remote sensing for large-scale monitoring, lacking a systematic fusion solution for multi-dimensional sensing data. While some solutions mention digital twin technology, the data base construction process often involves simply stitching together or weighted averaging three-dimensional heterogeneous data from different sources and with different spatiotemporal resolutions, without designing specific noise reduction and interpolation processes tailored to sensor noise characteristics and spatial distribution features. Regarding model parameter calibration, existing solutions mostly employ periodic manual calibration, where technicians manually adjust model parameters based on measured data at fixed time intervals. This method is time-consuming and has a slow response, making it difficult to adapt to the rapid changes in crop water requirements in irrigation districts due to weather conditions and growth stages. Furthermore, existing scheduling and decision-making systems typically assume that model outputs are always reliable, lacking a quantitative assessment mechanism for model prediction confidence. When sensor failures or data gaps cause the model to deviate from reality, they cannot automatically identify low-confidence scenarios and adopt conservative strategies, posing a risk to decision-making security.

[0004] The existing technologies mentioned above have the following shortcomings: First, single-dimensional monitoring and simple data fusion methods cannot effectively eliminate the interference of various sensor observation noises. There is a lack of spatiotemporal benchmark consistency between airborne data and ground / water body data, which directly affects the construction quality of the digital twin data base and the accuracy of subsequent analysis and decision-making. Second, manual parameter calibration methods have long update cycles and high labor intensity. During irrigation, when crop water requirement model parameters change rapidly with crop growth stages, the deviation between the fixed parameter model and the actual physical process will accumulate over time, leading to a gradual decline in the applicability of the scheduling scheme. Third, there is a lack of quantitative evaluation methods for model prediction confidence. When the model is unreliable, the scheduling system still blindly relies on the model output, which may generate scheduling instructions that do not conform to the actual safety constraints of the irrigation area, affecting the safety of project operation. Therefore, it is necessary to design a smart system and method for large-scale irrigation areas in southern China based on three-dimensional perception of air, ground, and water to solve the above problems. Summary of the Invention

[0005] The technical problem to be solved by this invention is to provide a smart system and method for large-scale irrigation areas in southern China based on three-dimensional perception of air, ground, and water. It aims to overcome the shortcomings of existing smart irrigation systems, such as inconsistent spatiotemporal benchmarks for multi-source heterogeneous data, delayed updates of model parameters due to reliance on manual calibration, and lack of prediction confidence assessment mechanisms for scheduling decisions. It features the characteristics of achieving spatiotemporal normalization of three-dimensional data through two-level fusion of adaptive Kalman filtering and spatiotemporal Kriging, and achieving online calibration of model parameters and uncertainty-driven safety degradation decision-making through the joint use of variational Bayesian inference and extended Kalman filtering.

[0006] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: A smart system for large-scale irrigation districts in southern China based on three-dimensional perception of air, ground, and water includes a three-dimensional data acquisition layer of air, ground, and water, an Internet of Things platform layer, a management area decision-making layer, and a production area execution layer. The air-ground-water three-dimensional data acquisition layer includes an air-end sensing subsystem, a ground-end monitoring subsystem, and a water-end sensing subsystem. Data collected by each subsystem is uploaded to the management area decision-making layer through the Internet of Things platform layer. The aerial sensing subsystem includes several sets of compound-wing UAVs and several sets of multi-rotor UAVs, all equipped with automated hangars and equipped with dual-light pods, multispectral cameras and searchlight loudspeakers, used to perform large-scale inspections of irrigation areas, monitoring of crop planting structures and emergency response; The ground-based monitoring subsystem includes several soil moisture monitoring stations, paddy field water level monitoring stations, video monitoring stations, and local control units for pump stations / gates. It covers the entire irrigation area according to the layout of irrigation patches and is used to realize the monitoring of soil moisture, field water level, engineering status, and local control. The water body sensing subsystem connects to water and rainfall monitoring stations, water quality monitoring stations, integrated measurement and control gates, flow measurement devices, and flow and pressure regulating valves. It is deployed at reservoirs, main canals, and water distribution points to realize the sensing of water source quantity, water quality parameters, and water flow rate. The IoT platform layer adopts a hybrid networking communication architecture, including a production control network, a management network, and a communication subsystem. The production control network connects the dispatch center, management station and key pump stations / gates via self-built optical cables to ensure the real-time transmission of control commands; The management network connects the dispatch center to the management station and the management station to the pumping station via leased operator lines, and is used to carry video surveillance and management information transmission; The communication subsystem realizes unified conversion of NB-IoT, 4G / 5G and MQTTS protocols. Distributed monitoring points use NB-IoT to upload data, and the gate control unit uses 4G / 5G plus MQTTS encrypted protocol for transmission. The production control network and management network are physically isolated. The management area's decision-making level is built on a digital twin platform, including a data base, model library, knowledge base, and twin engine. The data base integrates geographic scene data with multi-level spatial resolution and multi-source sensor data, covering the entire irrigation area; The model library includes water resource allocation models, water demand prediction models, and engineering safety assessment models, supporting forecasting, early warning, simulation, and contingency planning functions; The knowledge base covers irrigation district scheduling rules, engineering parameters, and crop water requirement quotas; The twin engine enables data fusion, simulation, and intelligent analysis; The production execution layer includes edge computing nodes and execution control units: Edge computing nodes are deployed at key control points to perform local preprocessing of scheduling instructions; The execution control unit includes a pump station control module, a gate control module, and a flow and pressure regulating valve control module. It adopts a dual-mode interlock of local control and remote control and is used to execute unit start-up and shutdown, flow regulation, and opening control.

[0007] Furthermore, the aerial sensing subsystem is equipped with two sets of compound-wing UAVs and four sets of multi-rotor UAVs; the compound-wing UAV hangars are deployed at the Wenfang branch pumping station in the north of the irrigation area and the Zoumabei management station in the south, with an effective patrol radius of not less than 25km; the multi-rotor UAV hangars are deployed in the reservoir dam site area and key towns, with an effective patrol radius of not less than 7km; the multispectral camera supports five multispectral channels and one RGB channel, used to identify the planting structure and growth status of navel oranges and rice.

[0008] Furthermore, the ground-based monitoring subsystem includes 44 soil moisture monitoring stations, configured with one dryland station and three orchard stations per irrigation area, to monitor soil temperature, humidity, and volumetric water content; there are 10 paddy field water level monitoring stations, using non-contact water level gauges with a range of 0 to 10m and a distance measurement accuracy of ±2mm.

[0009] Furthermore, the water body sensing subsystem connects to data from 95 water and rainfall monitoring stations; there are 5 water quality monitoring stations, deployed at the intake and branch outlets of the main canal, which use chemical methods to monitor pH, turbidity, ammonia nitrogen, and total phosphorus, with a data transmission cycle of no more than 1 hour; there are 347 integrated measurement and control gates, 169 flow measurement devices, and 49 flow and pressure regulating valves.

[0010] Furthermore, in the IoT platform layer, the production control network is configured with a 100Mbps bandwidth switch; the dedicated line bandwidth from the dispatch center to the management station in the management network is no less than 300Mbps, and the dedicated line bandwidth from the management station to the pump station is no less than 50Mbps; the minimum communication baud rate of the communication subsystem is 19.2K.

[0011] Furthermore, the data base integrates 1m resolution DOM large scene data, 0.2m resolution refined scene data, and 3cm resolution oblique photography model, covering an area of ​​3671km² in the irrigation district; the model library includes water resource allocation model, water demand prediction model, and engineering safety assessment model.

[0012] Preferably, the management area decision-making layer further includes a multi-source heterogeneous data spatiotemporal fusion module, which performs two-level fusion processing on the multi-sensor observation data output by the air-ground-water three-dimensional data acquisition layer, including adaptive Kalman filtering and spatiotemporal Kriging interpolation.

[0013] Preferably, the first stage is an adaptive Kalman filter, which uses time-series observation data from each monitoring station in the irrigation area as input and defines a state vector. ,in This refers to the soil volumetric water content. For soil temperature, For the water level of the rice paddies, For channel traffic, This refers to the ammonia nitrogen concentration. Total phosphorus concentration, subscript Indicates the first Each sampling time; the state evolution equation is: The observation equation is ,in Here is the state transition matrix. For the observation matrix, and The process noise and observation noise are respectively represented by their covariance matrices. and ; with new information sequence To drive this process, the noise covariance is adaptively adjusted using the following formula: ; ; in, Here is the Kalman gain matrix. For the prediction error covariance matrix, and The adaptive learning rate is set to a value between 0.01 and 0.1.

[0014] Preferably, the second stage is spatiotemporal kriging interpolation, using the filtered state estimates of each station output from the first-stage adaptive Kalman filter as input, and constructing an empirical variogram function for each physical quantity: ; in For station spacing, The distance is The number of sites and Positions and The filtered state value at the location; an exponential variogram model is used. Fit, where Value of a nugget. For the partial sill value, For variable range; ordinary kriging interpolation is performed based on the fitted variogram function for any unsensored location within the irrigation area. Estimate its physical quantity value: ; in The number of neighboring sites participating in the interpolation. The interpolation weights are obtained by solving the Kriging equations, with the following constraints: The interpolation results generate a rasterized spatial distribution field of the entire irrigation area, which is then input into the data base of the digital twin platform.

[0015] Furthermore, in the spatiotemporal fusion module for multi-source heterogeneous data, the adaptive learning rate of the adaptive Kalman filter... and The following rules apply to different time periods: During non-irrigation periods... , During irrigation or flood season, , This is to accelerate the rate of tracking new information changes.

[0016] Preferably, the management area decision-making layer further includes an online model calibration and decision confidence assessment module, which is implemented based on variational Bayesian inference and extended Kalman filtering. The variational Bayesian inference uses a sliding time window. Using historical forecast-observation residual sequences as evidence, the model parameter vectors were analyzed. Perform posterior inference, where For crop coefficients, Soil infiltration rate, This is the channel roughness coefficient. These are the parameters of the soil moisture characteristic curve. The crop water sensitivity coefficient The sliding window length; in variational distribution Approaching the true posterior ,in For variational parameters, Given the observation dataset within the window; update the variational parameters by maximizing the lower bound of evidence. : ; in For Kullback-Leibler divergence, The prior distribution is established based on knowledge base scheduling rules and engineering experience parameters; parameter updates are triggered when background conditions and observation completeness conditions are met. The triggering conditions are: no sensor fault alarms for 72 consecutive hours and the effective observation sample size is not less than 90% of the theoretical sampling number within the window.

[0017] Preferably, the extended Kalman filter uses variational Bayesian inference to determine the posterior mean of the output parameters. As model parameters, real-time recursive predictions are performed on the hydraulic and soil hydrodynamic states of the irrigation area: The decision confidence assessment module calculates the relative uncertainty of the parameters posteriorly before each scheduling instruction is generated: ; in For parameters The posterior variance, Let be the posterior mean of the parameters; when the uncertainty of any parameter is... Exceeding the preset threshold When the model prediction is determined to be in a low-confidence state, the system automatically switches to a rule-based safe scheduling mode, suspends the use of model optimization output, and executes control according to preset conservative scheduling rules; when all parameters are... When all values ​​fall below the corresponding threshold, model-driven optimization scheduling is resumed.

[0018] Furthermore, the length of the sliding window Adaptive seasonal adjustment: Non-irrigation period Days, irrigation period Heaven, flood season Day; parameter uncertainty threshold Set as: , , , , .

[0019] Preferably, a smart irrigation method for large-scale irrigation districts in southern China based on three-dimensional perception of air, ground, and water includes the following steps: S1: A three-dimensional data acquisition layer (air-ground-water) is used to acquire all-element data. At the air end, drones perform inspections in autonomous and specialized modes to obtain data on irrigation area topography and crop distribution. At the ground end, soil moisture monitoring stations collect soil temperature, humidity, and volumetric water content at set intervals, paddy field water level monitoring stations continuously monitor field water levels, and video monitoring stations capture real-time images of the project's operational status. At the water body end, water and rainfall monitoring stations, water quality monitoring stations, and water measurement equipment collect data on water source quantity, quality, and flow rate at set intervals. All collected data is uploaded via the IoT platform layer. S2: The IoT platform layer performs protocol conversion and secure transmission of collected data. The communication subsystem unifies the conversion of heterogeneous protocols such as NB-IoT, 4G / 5G and MQTTS. The production control network transmits control commands, and the management network transmits video and management data. Data security is ensured through physical isolation and encrypted transmission. S3: The management area's decision-making level conducts analysis and decision-making based on the digital twin platform. The digital twin platform processes the uploaded data, updates the overall data base, and uses a knowledge base to establish mapping relationships between water conditions, engineering conditions, agricultural conditions, and water quality. Based on the model library, it conducts water resource supply and demand balance simulations, irrigation scheduling plan rehearsals, and disaster risk assessments, and generates scheduling instructions. S4: The production area execution layer responds quickly and executes scheduling commands automatically. Edge computing nodes preprocess scheduling commands locally and control the start and stop of pump station units, gate opening adjustment, and pressure adjustment of flow regulating and pressure regulating valves through a dual-mode interlocking mechanism of local and remote control. During the execution process, each monitoring device provides real-time feedback on the execution effect, forming a closed-loop control. S5: The digital twin platform continuously tracks the operation status of the irrigation area, compares the actual operation data with the simulation results, analyzes the causes of deviations, and dynamically optimizes model parameters and scheduling strategies; it also regularly compiles system operation indicators to evaluate water resource utilization efficiency and engineering operation reliability.

[0020] Furthermore, in step S1, the routine cycle for drone inspections is 7 days, which is increased to 1 to 3 days during irrigation and flood seasons; the data collection cycle for soil moisture monitoring is once every 5 to 15 minutes, which is adjusted to once every 1 to 3 minutes during irrigation season.

[0021] Furthermore, in step S2, data transmission adopts the MQTTS encryption protocol, combined with physical isolation design, to build a security protection system, and the system security protection level reaches Level 3 of the Information Security Protection Standard.

[0022] Furthermore, in step S3, water resource allocation follows the principles of high water utilization, low water utilization, and superior water utilization, prioritizing the use of water diversion projects, then activating water storage projects, and finally calling upon water lifting projects.

[0023] Furthermore, in step S4, the gate control adopts a local control and remote control interlocking mode. The remote control issues opening or flow commands through the digital twin platform, and the control response time does not exceed 3 seconds. The pump station control supports unit start-up and shutdown, flow regulation and automatic fault alarm, and realizes joint scheduling of multiple pump stations.

[0024] Preferably, in step S3, the processing of the uploaded data includes spatiotemporal fusion processing of multi-source heterogeneous data, which includes: First-stage adaptive Kalman filter: Using time-series observation data from each monitoring station in the irrigation district as input, a state vector is constructed. ,in This refers to the soil volumetric water content. For soil temperature, For the water level of the rice paddies, For channel traffic, This refers to the ammonia nitrogen concentration. The total phosphorus concentration is; the state evolution equation is: The observation equation is ,in and covariance matrix and Based on the new information sequence Adaptive updates; Second-level spatiotemporal Kriging interpolation: Using the Kalman filter outputs of each station as input, an empirical variogram is constructed. An exponential variogram model is fitted, and ordinary kriging interpolation is performed. The interpolation results generate a rasterized spatial distribution field of the entire irrigation area.

[0025] Preferably, step S3, before generating the scheduling instructions, further includes online model calibration and decision confidence assessment steps: With sliding time window Using historical forecast-observation residual sequences as evidence, the model parameter vectors were analyzed. Perform variational Bayesian inference, where For crop coefficients, Soil infiltration rate, This is the channel roughness coefficient. These are the parameters of the soil moisture characteristic curve. The crop water sensitivity coefficient; expressed in variational distribution. Approximating the true posterior, the variational parameters are updated by maximizing the lower bound of evidence. The parameter update trigger conditions are: no sensor fault alarms for 72 consecutive hours, and the effective observation sample size is not less than 90% of the theoretical sampling number within the window; With the posterior mean of the parameters Substitute the extended Kalman filter to perform real-time recursive prediction of the hydraulic and soil hydrodynamic states of the irrigation area; Before each scheduling instruction is generated, the relative uncertainty of each parameter is calculated. When any Exceeding the preset threshold When the model is in a low-confidence state, switch to rule-based safe scheduling mode; when all When the value falls below the threshold, model-driven optimization scheduling is resumed.

[0026] Preferably, in step S3, the functions of forecasting, early warning, rehearsal, and contingency plan are executed in a progressive manner as follows: Forecast: in the future Using daily weather forecast data, upstream basin rainfall forecast data, and reservoir water storage status as inputs, the model outputs the future water inflow forecast data for the headworks section. The process of water flow during the day , Simultaneously, using historical irrigation data and crop growth stages as inputs, the water demand prediction model outputs the future water demand forecasts for each irrigation area. Daily water demand ,in For the film filling number, ; Warning: Calculate the daily water supply shortfall separately for each irrigation area. ,in For film filling The water utilization coefficient of the canal system The daily time step is set; a three-level early warning threshold is set when... A value less than 0.15 indicates a yellow alert, a value between 0.15 and 0.30 indicates an orange alert, and a value greater than 0.30 indicates a red alert. Preliminary simulation: Using at least two candidate scheduling schemes as input, each scheme includes the planned opening sequence of each gate and the planned flow sequence of the pumping station. A hydrodynamic simulation model is constructed based on the one-dimensional Saint-Venant equations for the channel. ; ; in, The cross-sectional area of ​​the water passage. For traffic, For water level, For lateral inflow, It is the acceleration due to gravity. For friction ratio reduction, The canal bottom gradient is determined; the Preissmann four-point implicit difference scheme is used for discretization and solution, and the water arrival time, water loss rate and irrigation area water satisfaction rate of each scheme are output; Contingency Plan: With the objectives of minimizing water loss rate and maximizing irrigation water supply satisfaction rate, the simulation results of candidate schemes are non-dominated and ranked. The scheme on the Pareto front is selected as the recommended scheme, while the other schemes on the front are retained as alternative schemes.

[0027] Preferably, in step S3, the water resource supply and demand balance simulation includes: The entire irrigation district is divided into Each computing unit, unit Daily water requirement of crops ,in For unit The crop coefficient, Reference evapotranspiration calculated using the FAO Penman-Monteith formula; unit Net irrigation water demand Gross water requirement for irrigation ,in For effective precipitation, For unit The corresponding canal system water utilization coefficient; With the total water supply of the irrigation district and the allowable flow of each channel as constraints, and minimizing the sum of squares of the difference between the actual water supply and the gross water demand of each unit as the objective, the sequential quadratic programming method is used to solve the water allocation optimization scheme.

[0028] Preferably, in step S1, the acquisition period of each sensor is dynamically adjusted according to the following adaptive rule: For soil moisture monitoring stations, the data collection cycle is... Adaptive adjustment as follows: ; in, For the first The acquisition cycle of each adjustment cycle, This represents the absolute value of the change in soil volumetric water content between two consecutive data collections. The threshold for water content change is taken as... , Let [the coefficient] be the adjustment factor. , and These are the lower and upper scaling factors for the acquisition period, respectively.

[0029] Furthermore, for water quality monitoring stations, the sampling period is fixed at 60 minutes during non-irrigation periods and 30 minutes during irrigation periods. When any water quality parameter exceeds 80% of the preset standard value, the sampling period is automatically shortened to 15 minutes.

[0030] Preferably, a computer device includes a memory and a processor, which are interconnected. The memory stores computer instructions, and the processor executes the computer instructions to perform the aforementioned intelligent irrigation method for large-scale irrigation areas in southern China based on three-dimensional perception of air, land, and water.

[0031] Preferably, a computer-readable storage medium stores computer instructions that enable a computer to execute the aforementioned intelligent method for large-scale irrigation areas in southern China based on three-dimensional perception of air, ground, and water.

[0032] The beneficial effects of this invention are as follows: 1. This invention constructs a three-dimensional data acquisition layer (air-ground-water) and a spatiotemporal fusion module for multi-source heterogeneous data. It uses adaptive Kalman filtering to perform site-by-site noise reduction on the time-series observation data of each sensor station, uses the innovation sequence to drive the real-time adaptive adjustment of the noise covariance matrix, and uses the filtered state estimate as input to fit the empirical variogram through spatiotemporal Kriging interpolation and perform ordinary Kriging interpolation to generate a gridded physical quantity distribution field of the entire irrigation area. This achieves spatiotemporal unification and noise elimination of three-dimensional heterogeneous sensor data, and solves the defects of existing technologies, such as inconsistent spatiotemporal benchmarks of multi-source data and ineffective filtering of sensor observation noise, which limit the quality of digital twin data base construction. It improves the fidelity and spatial coverage integrity of the digital twin irrigation area data base.

[0033] 2. This invention establishes an online model calibration and decision confidence assessment module. Using the historical forecast observation residual sequence within a sliding time window as evidence, it employs variational Bayesian inference to perform posterior inference on the parameters of the irrigation district hydraulic and soil hydrodynamic models. By maximizing the lower bound of evidence to update the variational parameters, and substituting the posterior mean of the parameters into the extended Kalman filter, it achieves real-time recursive prediction of the state after parameter calibration. Simultaneously, by calculating the relative uncertainty of the posterior of each parameter, and determining that the model is in a low-confidence state when the uncertainty of any parameter exceeds a threshold, it automatically switches to a rule-based safe scheduling mode. This solves the defects of existing technologies, such as long manual calibration cycles for model parameters, slow response, and lack of quantitative assessment mechanisms for prediction confidence. It realizes continuous adaptive calibration of model parameters according to the operating state of the irrigation district and safe degraded scheduling under abnormal operating conditions.

[0034] 3. This invention utilizes the hierarchical execution of forecasting, early warning, pre-drilling, and contingency planning functions within a digital twin platform. Based on water demand forecasting and water availability prediction, it calculates water supply gaps for each irrigation area and sets three levels of early warning thresholds. A hydrodynamic simulation model using the one-dimensional Saint-Venant equations of the canal is used to numerically simulate multiple candidate scheduling schemes. Candidate schemes are non-dominated and ranked based on minimizing water loss rate and maximizing irrigation area water supply satisfaction rate, with the scheme on the Pareto front selected as the recommended contingency plan. Simultaneously, it combines the calculation of crop reference evapotranspiration using the FAO Penman-Monteith formula with a sequential quadratic programming solution to optimize water allocation. This addresses the shortcomings of existing irrigation district scheduling methods that rely on experience-based judgment and lack quantitative pre-drilling and comparison methods, achieving a complete closed loop from data perception to scheduling decision-making, and improving the scientific and precise level of irrigation district water resource allocation. Attached Figure Description

[0035] Figure 1 This is a schematic diagram of the system structure according to an embodiment of the present disclosure; Figure 2 This is a schematic diagram of the method flow of an embodiment of this disclosure; Figure 3 This is a schematic diagram comparing the noise reduction effect of adaptive Kalman filtering on time-series observations of soil volumetric water content according to embodiments of this disclosure; Figure 4 This is a schematic diagram illustrating the tracking of channel flow prediction accuracy and parameter uncertainty using variational Bayesian inference combined with extended Kalman filtering according to an embodiment of this disclosure. Figure 5 A schematic diagram of the structure of a computer device provided in an embodiment of this disclosure. Detailed Implementation

[0036] Example 1: like Figure 1 As shown, a smart system for large-scale irrigation districts in southern China based on three-dimensional perception of air, ground, and water includes a three-dimensional data acquisition layer of air, ground, and water, an Internet of Things platform layer, a management area decision-making layer, and a production area execution layer. The air-ground-water three-dimensional data acquisition layer includes an air-end sensing subsystem, a ground-end monitoring subsystem, and a water-end sensing subsystem. Data collected by each subsystem is uploaded to the management area decision-making layer through the Internet of Things platform layer. The aerial sensing subsystem includes several sets of compound-wing UAVs and several sets of multi-rotor UAVs, all equipped with automated hangars and equipped with dual-light pods, multispectral cameras and searchlight loudspeakers, used to perform large-scale inspections of irrigation areas, monitoring of crop planting structures and emergency response; The ground-based monitoring subsystem includes several soil moisture monitoring stations, paddy field water level monitoring stations, video monitoring stations, and local control units for pump stations / gates. It covers the entire irrigation area according to the layout of irrigation patches and is used to realize the monitoring of soil moisture, field water level, engineering status, and local control. The water body sensing subsystem connects to water and rainfall monitoring stations, water quality monitoring stations, integrated measurement and control gates, flow measurement devices, and flow and pressure regulating valves. It is deployed at reservoirs, main canals, and water distribution points to realize the sensing of water source quantity, water quality parameters, and water flow rate. The IoT platform layer adopts a hybrid networking communication architecture, including a production control network, a management network, and a communication subsystem. The production control network connects the dispatch center, management station and key pump stations / gates via self-built optical cables to ensure the real-time transmission of control commands; The management network connects the dispatch center to the management station and the management station to the pumping station via leased operator lines, and is used to carry video surveillance and management information transmission; The communication subsystem realizes unified conversion of NB-IoT, 4G / 5G and MQTTS protocols. Distributed monitoring points use NB-IoT to upload data, and the gate control unit uses 4G / 5G plus MQTTS encrypted protocol for transmission. The production control network and management network are physically isolated. The management area's decision-making level is built on a digital twin platform, including a data base, model library, knowledge base, and twin engine. The data base integrates geographic scene data with multi-level spatial resolution and multi-source sensor data, covering the entire irrigation area; The model library includes water resource allocation models, water demand prediction models, and engineering safety assessment models, supporting forecasting, early warning, simulation, and contingency planning functions; The knowledge base covers irrigation district scheduling rules, engineering parameters, and crop water requirement quotas; The twin engine enables data fusion, simulation, and intelligent analysis; The production execution layer includes edge computing nodes and execution control units: Edge computing nodes are deployed at key control points to perform local preprocessing of scheduling instructions; The execution control unit includes a pump station control module, a gate control module, and a flow and pressure regulating valve control module. It adopts a dual-mode interlock of local control and remote control and is used to execute unit start-up and shutdown, flow regulation, and opening control.

[0037] Furthermore, the aerial sensing subsystem is equipped with two sets of compound-wing UAVs and four sets of multi-rotor UAVs; the compound-wing UAV hangars are deployed at the Wenfang branch pumping station in the north of the irrigation area and the Zoumabei management station in the south, with an effective patrol radius of not less than 25km; the multi-rotor UAV hangars are deployed in the reservoir dam site area and key towns, with an effective patrol radius of not less than 7km; the multispectral camera supports five multispectral channels and one RGB channel, used to identify the planting structure and growth status of navel oranges and rice.

[0038] Furthermore, the ground-based monitoring subsystem includes 44 soil moisture monitoring stations, configured with one dryland station and three orchard stations per irrigation area, to monitor soil temperature, humidity, and volumetric water content; there are 10 paddy field water level monitoring stations, using non-contact water level gauges with a range of 0 to 10m and a distance measurement accuracy of ±2mm.

[0039] Furthermore, the water body sensing subsystem connects to data from 95 water and rainfall monitoring stations; there are 5 water quality monitoring stations, deployed at the intake and branch outlets of the main canal, which use chemical methods to monitor pH, turbidity, ammonia nitrogen, and total phosphorus, with a data transmission cycle of no more than 1 hour; there are 347 integrated measurement and control gates, 169 flow measurement devices, and 49 flow and pressure regulating valves.

[0040] Furthermore, in the IoT platform layer, the production control network is configured with a 100Mbps bandwidth switch; the dedicated line bandwidth from the dispatch center to the management station in the management network is no less than 300Mbps, and the dedicated line bandwidth from the management station to the pump station is no less than 50Mbps; the minimum communication baud rate of the communication subsystem is 19.2K.

[0041] Furthermore, the data base integrates 1m resolution DOM large scene data, 0.2m resolution refined scene data, and 3cm resolution oblique photography model, covering an area of ​​3671km² in the irrigation district; the model library includes water resource allocation model, water demand prediction model, and engineering safety assessment model.

[0042] Preferably, the management area decision-making layer further includes a multi-source heterogeneous data spatiotemporal fusion module, which performs two-level fusion processing on the multi-sensor observation data output by the air-ground-water three-dimensional data acquisition layer, including adaptive Kalman filtering and spatiotemporal Kriging interpolation.

[0043] Preferably, the first stage is an adaptive Kalman filter, which uses time-series observation data from each monitoring station in the irrigation area as input and defines a state vector. ,in This refers to the soil volumetric water content. For soil temperature, For the water level of the rice paddies, For channel traffic, This refers to the ammonia nitrogen concentration. Total phosphorus concentration, subscript Indicates the first Each sampling time; the state evolution equation is: The observation equation is ,in Here is the state transition matrix. For the observation matrix, and The process noise and observation noise are respectively represented by their covariance matrices. and ; with new information sequence To drive this process, the noise covariance is adaptively adjusted using the following formula: ; ; in, Here is the Kalman gain matrix. For the prediction error covariance matrix, and The adaptive learning rate is set to a value between 0.01 and 0.1.

[0044] Preferably, the second stage is spatiotemporal kriging interpolation, using the filtered state estimates of each station output from the first-stage adaptive Kalman filter as input, and constructing an empirical variogram function for each physical quantity: ; in For station spacing, The distance is The number of sites and Positions and The filtered state value at the location; an exponential variogram model is used. Fit, where Value of a nugget. For the partial sill value, For variable range; ordinary kriging interpolation is performed based on the fitted variogram function for any unsensored location within the irrigation area. Estimate its physical quantity value: ; in The number of neighboring sites participating in the interpolation. The interpolation weights are obtained by solving the Kriging equations, with the following constraints: The interpolation results generate a rasterized spatial distribution field of the entire irrigation area, which is then input into the data base of the digital twin platform.

[0045] Furthermore, in the spatiotemporal fusion module for multi-source heterogeneous data, the adaptive learning rate of the adaptive Kalman filter... and The following rules apply to different time periods: During non-irrigation periods... , During irrigation or flood season, , This is to accelerate the rate of tracking new information changes.

[0046] Preferably, the management area decision-making layer further includes an online model calibration and decision confidence assessment module, which is implemented based on variational Bayesian inference and extended Kalman filtering. The variational Bayesian inference uses a sliding time window. Using historical forecast-observation residual sequences as evidence, the model parameter vectors were analyzed. Perform posterior inference, where For crop coefficients, Soil infiltration rate, This is the channel roughness coefficient. These are the parameters of the soil moisture characteristic curve. The crop water sensitivity coefficient The sliding window length; in variational distribution Approaching the true posterior ,in For variational parameters, Given the observation dataset within the window; update the variational parameters by maximizing the lower bound of evidence. : ; in For Kullback-Leibler divergence, The prior distribution is established based on knowledge base scheduling rules and engineering experience parameters; parameter updates are triggered when background conditions and observation completeness conditions are met. The triggering conditions are: no sensor fault alarms for 72 consecutive hours and the effective observation sample size is not less than 90% of the theoretical sampling number within the window.

[0047] Preferably, the extended Kalman filter uses variational Bayesian inference to determine the posterior mean of the output parameters. As model parameters, real-time recursive predictions are performed on the hydraulic and soil hydrodynamic states of the irrigation area: The decision confidence assessment module calculates the relative uncertainty of the parameters posteriorly before each scheduling instruction is generated: ; in For parameters The posterior variance, Let be the posterior mean of the parameters; when the uncertainty of any parameter Exceeding the preset threshold When the model prediction is determined to be in a low-confidence state, the system automatically switches to a rule-based safe scheduling mode, suspends the use of model optimization output, and executes control according to preset conservative scheduling rules; when all parameters are... When all values ​​fall below the corresponding threshold, model-driven optimization scheduling is resumed.

[0048] Furthermore, the length of the sliding window Adaptive seasonal adjustment: Non-irrigation period Days, irrigation period Heaven, flood season Day; parameter uncertainty threshold Set as: , , , , .

[0049] like Figure 2 As shown, a smart irrigation method for large-scale irrigation districts in southern China based on three-dimensional perception of air, ground, and water includes the following steps: S1: A three-dimensional data acquisition layer (air-ground-water) is used to acquire all-element data. At the air end, drones perform inspections in autonomous and specialized modes to obtain data on irrigation area topography and crop distribution. At the ground end, soil moisture monitoring stations collect soil temperature, humidity, and volumetric water content at set intervals, paddy field water level monitoring stations continuously monitor field water levels, and video monitoring stations capture real-time images of the project's operational status. At the water body end, water and rainfall monitoring stations, water quality monitoring stations, and water measurement equipment collect data on water source quantity, quality, and flow rate at set intervals. All collected data is uploaded via the IoT platform layer. S2: The IoT platform layer performs protocol conversion and secure transmission of collected data. The communication subsystem unifies the conversion of heterogeneous protocols such as NB-IoT, 4G / 5G and MQTTS. The production control network transmits control commands, and the management network transmits video and management data. Data security is ensured through physical isolation and encrypted transmission. S3: The management area's decision-making level conducts analysis and decision-making based on the digital twin platform. The digital twin platform processes the uploaded data, updates the overall data base, and uses a knowledge base to establish mapping relationships between water conditions, engineering conditions, agricultural conditions, and water quality. Based on the model library, it conducts water resource supply and demand balance simulations, irrigation scheduling plan rehearsals, and disaster risk assessments, and generates scheduling instructions. S4: The production area execution layer responds quickly and executes scheduling commands automatically. Edge computing nodes preprocess scheduling commands locally and control the start and stop of pump station units, gate opening adjustment, and pressure adjustment of flow regulating and pressure regulating valves through a dual-mode interlocking mechanism of local and remote control. During the execution process, each monitoring device provides real-time feedback on the execution effect, forming a closed-loop control. S5: The digital twin platform continuously tracks the operation status of the irrigation area, compares the actual operation data with the simulation results, analyzes the causes of deviations, and dynamically optimizes model parameters and scheduling strategies; it also regularly compiles system operation indicators to evaluate water resource utilization efficiency and engineering operation reliability.

[0050] Furthermore, in step S1, the routine cycle for drone inspections is 7 days, which is increased to 1 to 3 days during irrigation and flood seasons; the data collection cycle for soil moisture monitoring is once every 5 to 15 minutes, which is adjusted to once every 1 to 3 minutes during irrigation season.

[0051] Furthermore, in step S2, data transmission adopts the MQTTS encryption protocol, combined with physical isolation design, to build a security protection system, and the system security protection level reaches Level 3 of the Information Security Protection Standard.

[0052] Furthermore, in step S3, water resource allocation follows the principles of high water utilization, low water utilization, and superior water utilization, prioritizing the use of water diversion projects, then activating water storage projects, and finally calling upon water lifting projects.

[0053] Furthermore, in step S4, the gate control adopts a local control and remote control interlocking mode. The remote control issues opening or flow commands through the digital twin platform, and the control response time does not exceed 3 seconds. The pump station control supports unit start-up and shutdown, flow regulation and automatic fault alarm, and realizes joint scheduling of multiple pump stations.

[0054] Preferably, in step S3, the processing of the uploaded data includes spatiotemporal fusion processing of multi-source heterogeneous data, which includes: First-stage adaptive Kalman filter: Using time-series observation data from each monitoring station in the irrigation district as input, a state vector is constructed. ,in This refers to the soil volumetric water content. For soil temperature, For the water level of the rice paddies, For channel traffic, This refers to the ammonia nitrogen concentration. The total phosphorus concentration is given by the state evolution equation. The observation equation is ,in and covariance matrix and Based on the new information sequence Adaptive updates; Second-level spatiotemporal Kriging interpolation: Using the Kalman filter outputs of each station as input, an empirical variogram is constructed. An exponential variogram model is fitted, and ordinary kriging interpolation is performed. The interpolation results generate a rasterized spatial distribution field of the entire irrigation area.

[0055] Preferably, step S3, before generating the scheduling instructions, further includes online model calibration and decision confidence assessment steps: With sliding time window Using historical forecast-observation residual sequences as evidence, the model parameter vectors were analyzed. Perform variational Bayesian inference, where For crop coefficients, Soil infiltration rate, This is the channel roughness coefficient. These are the parameters of the soil moisture characteristic curve. The crop water sensitivity coefficient; expressed in variational distribution. Approximating the true posterior, the variational parameters are updated by maximizing the lower bound of evidence. The parameter update trigger conditions are: no sensor fault alarms for 72 consecutive hours, and the effective observation sample size is not less than 90% of the theoretical sampling number within the window; With the posterior mean of the parameters Substitute the extended Kalman filter to perform real-time recursive prediction of the hydraulic and soil hydrodynamic states of the irrigation area; Before each scheduling instruction is generated, the relative uncertainty of each parameter is calculated. When any Exceeding the preset threshold When the model is in a low-confidence state, switch to rule-based safe scheduling mode; when all When the value falls below the threshold, model-driven optimization scheduling is resumed.

[0056] Preferably, in step S3, the functions of forecasting, early warning, rehearsal, and contingency plan are executed in a progressive manner as follows: Forecast: in the future Using daily weather forecast data, upstream basin rainfall forecast data, and reservoir water storage status as inputs, the model outputs the future water inflow forecast data for the headworks section. The process of water flow during the day , Simultaneously, using historical irrigation data and crop growth stages as inputs, the water demand prediction model outputs the future water demand forecasts for each irrigation area. Daily water demand ,in Number the film. ; Warning: Calculate the daily water supply shortfall separately for each irrigation area. ,in For film filling The water utilization coefficient of the canal system The daily time step is set; a three-level early warning threshold is set when... A value less than 0.15 indicates a yellow alert, a value between 0.15 and 0.30 indicates an orange alert, and a value greater than 0.30 indicates a red alert. Preliminary simulation: Using at least two candidate scheduling schemes as input, each scheme includes the planned opening sequence of each gate and the planned flow sequence of the pumping station. A hydrodynamic simulation model is constructed based on the one-dimensional Saint-Venant equations for the channel. ; ; in, The cross-sectional area of ​​the water passage. For traffic, For water level, For lateral inflow, It is the acceleration due to gravity. For friction ratio reduction, The canal bottom gradient is determined; the Preissmann four-point implicit difference scheme is used for discretization and solution, and the water arrival time, water loss rate and irrigation area water satisfaction rate of each scheme are output; Contingency Plan: With the objectives of minimizing water loss rate and maximizing irrigation water supply satisfaction rate, the simulation results of candidate schemes are non-dominated and ranked. The scheme on the Pareto front is selected as the recommended scheme, while the other schemes on the front are retained as alternative schemes.

[0057] Preferably, in step S3, the water resource supply and demand balance simulation includes: The entire irrigation district is divided into Each computing unit, unit Daily water requirement of crops ,in For unit The crop coefficient, Reference evapotranspiration calculated using the FAO Penman-Monteith formula; unit Net irrigation water demand Gross water requirement for irrigation ,in For effective precipitation, For unit The corresponding canal system water utilization coefficient; With the total water supply of the irrigation district and the allowable flow of each channel as constraints, and minimizing the sum of squares of the difference between the actual water supply and the gross water demand of each unit as the objective, the sequential quadratic programming method is used to solve the water allocation optimization scheme.

[0058] Preferably, in step S1, the acquisition period of each sensor is dynamically adjusted according to the following adaptive rule: For soil moisture monitoring stations, the data collection cycle is... Adaptive adjustment as follows: ; in, For the first The acquisition cycle of each adjustment cycle, This represents the absolute value of the change in soil volumetric water content between two consecutive data collections. The threshold for water content change is taken as follows: , The adjustment coefficient is taken as... , and These are the lower and upper scaling factors for the acquisition period, respectively.

[0059] Furthermore, for water quality monitoring stations, the sampling period is fixed at 60 minutes during non-irrigation periods and 30 minutes during irrigation periods. When any water quality parameter exceeds 80% of the preset standard value, the sampling period is automatically shortened to 15 minutes.

[0060] Preferably, a computer device includes a memory and a processor, which are interconnected. The memory stores computer instructions, and the processor executes the computer instructions to perform the aforementioned intelligent irrigation method for large-scale irrigation areas in southern China based on three-dimensional perception of air, land, and water.

[0061] Preferably, a computer-readable storage medium stores computer instructions that enable a computer to execute the aforementioned intelligent method for large-scale irrigation areas in southern China based on three-dimensional perception of air, ground, and water.

[0062] Example 2: This embodiment uses the Meijiang Irrigation District in Jiangxi Province as the application scenario. The irrigation district covers an area of ​​3671 km², involving 22 townships and divided into 5 irrigation areas: 2 in the north, 2 in the south, and 1 in the west. Water sources include reservoirs, weirs, and pumping stations. The experimental platform is based on an air-ground-water three-dimensional data acquisition system actually deployed in the irrigation district: At the air end, 2 composite-wing UAVs equipped with multispectral cameras and 4 multi-rotor UAVs are deployed; at the ground end, 44 soil moisture monitoring stations and 10 paddy field water level monitoring stations are deployed; and at the water body end, 95 hydrological and rainfall monitoring stations and 5 water quality monitoring stations are connected. The IoT platform layer adopts a production control network, including self-built optical cables and 100Mbps switches, and a management network, including dedicated lines from operators, with 300Mbps from the dispatch center to the management station and 50Mbps from the management station to the pumping station. The communication subsystem uniformly converts NB-IoT, 4G / 5G, and MQTTS protocols. The management area's decision-making level deployed a digital twin platform, which consists of a data base consisting of a 1m resolution DOM, a 0.2m resolution detailed scene, and a 3cm resolution oblique photogrammetry model.

[0063] Parameter settings: Adaptive Kalman filter learning rate during non-irrigation period , Irrigation period , Spacetime Kriging Variable right Take 8.2km, pair Take 12.5 km. VBI sliding window length during non-irrigation period. Days, Irrigation period Weather, flood season Heaven. Parameter uncertainty threshold. , , , , Soil moisture monitoring station data collection cycle adaptive adjustment factor , , , Comparison Method 1 uses a conventional moving average filter with a fixed parameter model. The parameters are manually calibrated once per quarter, and the crop coefficient is fixed. In contrast, method 2 uses a simple weighted average for data fusion, with equal weights for each sensor, and employs empirical rules for scheduling decisions, without relying on digital twin model rehearsals.

[0064] Example Implementation Process: Real-world operational data was collected from April 1st to June 29th, 2025 (90 days total), covering a 30-day non-irrigation period, a 40-day irrigation period, and a 20-day flood season. Data was analyzed at a 10-minute frequency from the soil moisture monitoring station. Timing and gate flow The AKF filtering method of this approach and the moving average filtering method of comparison method 1 were performed respectively. For the digital twin model parameters, the VBI+EKF online calibration method of this approach and the fixed parameter method of comparison method 1 were performed respectively. For the scheduling scheme, the Saint-Venant equation pre-simulation plus Pareto comparison method of this approach and the empirical rule scheduling method of comparison method 2 were performed respectively. The filtering error, prediction error, water quantity satisfaction rate of the scheduling scheme, water loss rate, and model uncertainty changes for each irrigation area were recorded. Statistics were compiled every 10 days, for a total of 9 statistical time points. Specific data are shown in Tables 1 and 2 below.

[0065] Table 1: Statistical analysis of various evaluation indicators for the five irrigation patches using this method;

[0066] Table 2: Comparison of indicators between this method and the comparative method for each sac;

[0067] As can be seen from the data in Table 1, the AKF adaptive Kalman filter of this method is effective for soil volumetric water content in each irrigation patch. The root mean square error (RMSE) of the filters was controlled between 0.008 and 0.012, with an average of 0.0096 for the entire irrigation area, affecting canal flow. The filtering RMSE was controlled within the range of 1.2 to 1.7 m³ / h, with an average of 1.40 m³ / h across the entire irrigation area, indicating that AKF maintained stable noise reduction performance under various operating conditions, including non-irrigated, irrigated, and flood season conditions. The average relative error of STK spatiotemporal kriging interpolation across the entire irrigation area was 4.7%, with higher interpolation accuracy in the southern area than in the northern area. This difference is mainly due to the higher sensor deployment density in the southern area, consistent with the physical law that kriging interpolation accuracy is positively correlated with station density. The convergence time for VBI parameter online calibration was between 65 and 75 hours, and the relative uncertainty of the parameters was... It remained at a level between 0.16 and 0.24, and never exceeded the threshold throughout the entire process. This indicates that the model parameters are stable in the posterior inference, eliminating the need for frequent security degradation triggers. The simulation calculation time is between 3.3 and 4.1 seconds, meeting the real-time simulation requirements of the digital twin platform for scheduling schemes.

[0068] As can be seen from the comparison data in Table 2, the AKF filtering method presented in this paper is superior to the moving average method in comparison method 1. The average improvement in filter RMSE was 52.9%, with a significant improvement of 60.3% during periods of sensor anomalies. This is because the moving average method does not specifically suppress abnormal noise, while AKF automatically enhances the filter strength during abnormal periods through innovation-driven adaptive covariance updates. This method combines VBI and EKF. The predicted RMSE shows an average improvement of 56.5% compared to the fixed-parameter model, with the improvement gradually increasing over time. In the short term, the improvement is 55.7%, but in long-term operation across seasons, the prediction bias of the fixed-parameter model continues to accumulate. This method uses a sliding window of VBI for online calibration to continuously correct parameters, maintaining stable prediction accuracy. Regarding scheduling performance, this method, through Saint Vincent's equations and Pareto front comparison, achieves an average water supply satisfaction rate of 92.9%, higher than the 77% of empirical rule-based scheduling, and an average water loss rate of 8.9%, lower than the 19.9% ​​of empirical scheduling. During periods of sensor anomalies, this method triggers a safety degradation mechanism through low-confidence detection, maintaining a water supply satisfaction rate of 89.8% and a water loss rate of 14.5%, both significantly better than the 68.2% and 31.2% of empirical scheduling, respectively. These data demonstrate that this method has significant advantages in noise reduction, parameter tracking, prediction accuracy, and fault tolerance under abnormal operating conditions.

[0069] like Figure 3 As shown, multi-source data fusion schemes in irrigation districts typically use simple weighted averages or moving averages to preprocess sensor observation data, lacking dynamic adaptive capabilities to address changes in sensor noise statistical characteristics under varying operating conditions. When an irrigation district transitions from a non-irrigation period to an irrigation period, the rate of change in soil moisture content increases significantly. The fixed noise covariance of a static filter cannot track this change; that is, setting the covariance too high will lead to filtering lag, while setting it too low will result in excessive residual noise.

[0070] This scheme uses the new information sequence as a driver to adjust the process noise covariance matrix in real time. and observation noise covariance matrix During the irrigation period, the AKF filter automatically amplifies the process covariance to accelerate tracking, and during the non-irrigation period, it automatically reduces the covariance to enhance smoothness. As shown in Figure A, the AKF filter curve closely follows the changes in the true value during the irrigation period, effectively suppresses the sudden amplification of abnormal noise during periods of sensor anomalies, and maintains a stable noise reduction effect during non-irrigation periods. Its root mean square error is significantly lower than that of the conventional moving average method, which intuitively verifies the effectiveness of the adaptive noise covariance adjustment mechanism.

[0071] like Figure 4 As shown, traditional irrigation district scheduling models operate with fixed parameters. As crop growth stages and environmental conditions change, the model parameters gradually deviate from the actual physical state, leading to an accumulation of flow prediction errors over time. This scheme uses variational Bayesian inference to dynamically update the posterior distribution of model parameters within a sliding window, and uses the posterior mean of the parameters to drive extended Kalman filtering for state prediction. The upper subfigure shows that the predicted flow curve of the VBI+EKF method closely matches the measured values, with a narrow prediction band that covers the measured values. In contrast, the fixed-parameter model's predictions begin to show significant deviations after day 15, reaching a deviation of over 0.2 m³ / s by day 25. The lower subfigure shows the crop coefficient. uncertainty The trajectory of change over time: In the initial stage when the model parameters are still accurate At the threshold The following operation was performed; after entering a new irrigation cycle, the parameter uncertainty naturally increased, but remained within a controllable range; around day 24, continuous rainfall caused changes in the model input. The brief breach of the threshold triggered a safety downgrade switch, which then subsided after new observation data accumulated. This demonstrates the safety feature of this scheme, which proactively enters a conservative scheduling mode when the model becomes inaccurate.

[0072] Example 3: like Figure 5As shown, this embodiment of the invention also provides a computer device, which includes one or more processors 10, a memory 20, and interfaces for connecting the various components, including high-speed interfaces and low-speed interfaces. The various components are interconnected via different buses and can be mounted on a common motherboard or otherwise installed as needed. The processors can process instructions executed within the computer device, including instructions stored in or on memory for displaying graphical information of a GUI on external input / output devices, such as display devices coupled to the interfaces. In some alternative embodiments, multiple processors and / or multiple buses can be used with multiple memories and multiple memory modules, if desired. Similarly, multiple computer devices can be connected, each providing some of the necessary operations, for example, as a server array, a group of blade servers, or a multiprocessor system.

[0073] Processor 10 may be a central processing unit, a network processor, or a combination thereof. Processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The programmable logic device may be a complex programmable logic device (CAMP), a field-programmable gate array (FPGA), a general-purpose array logic (GDA), or any combination thereof.

[0074] The memory 20 stores instructions executable by at least one processor 10 to cause the at least one processor 10 to perform the method shown in the above embodiments.

[0075] The memory 20 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the computer device. Furthermore, the memory 20 may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some alternative embodiments, the memory 20 may optionally include memory remotely located relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0076] The memory 20 may include volatile memory, such as random access memory; the memory may also include non-volatile memory, such as flash memory, hard disk or solid-state drive; the memory 20 may also include a combination of the above types of memory.

[0077] The computer device also includes a communication interface 30 for communicating with other devices or communication networks.

[0078] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code, which, when accessed and executed by the computer, processor, or hardware, implements the methods shown in the above embodiments.

[0079] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A smart system for large-scale irrigation districts in southern China based on three-dimensional perception of air, ground, and water, characterized in that: It includes a three-dimensional data acquisition layer (air-ground-water), an IoT platform layer, a management area decision-making layer, and a production area execution layer; The air-ground-water three-dimensional data acquisition layer includes an air-end sensing subsystem, a ground-end monitoring subsystem, and a water-end sensing subsystem. Data collected by each subsystem is uploaded to the management area decision-making layer through the Internet of Things platform layer. The IoT platform layer adopts a hybrid networking communication architecture, including a production control network, a management network, and a communication subsystem. The management area's decision-making level is built on a digital twin platform, including a data base, model library, knowledge base, and twin engine. The data base integrates geographic scene data with multi-level spatial resolution and multi-source sensor data, covering the entire irrigation area; The model library includes water resource allocation models, water demand prediction models, and engineering safety assessment models, supporting forecasting, early warning, simulation, and contingency planning functions; The knowledge base covers irrigation district scheduling rules, engineering parameters, and crop water requirement quotas; The twin engine enables data fusion, simulation, and intelligent analysis; The production execution layer includes edge computing nodes and execution control units: Edge computing nodes are deployed at key control points to perform local preprocessing of scheduling instructions; The execution control unit includes a pump station control module, a gate control module, and a flow and pressure regulating valve control module. It adopts a dual-mode interlock of local control and remote control and is used to execute unit start-up and shutdown, flow regulation, and opening control.

2. The intelligent system for large-scale irrigation areas in southern China based on three-dimensional perception of air, ground, and water, as described in claim 1, is characterized in that... The aerial sensing subsystem includes several sets of compound-wing UAVs and several sets of multi-rotor UAVs, all equipped with automated hangars and equipped with dual-light pods, multispectral cameras and searchlight loudspeakers, used to perform large-scale inspections of irrigation areas, monitoring of crop planting structures and emergency response; The ground-based monitoring subsystem includes several soil moisture monitoring stations, paddy field water level monitoring stations, video monitoring stations, and local control units for pump stations / gates. It covers the entire irrigation area according to the layout of irrigation patches and is used to realize the monitoring of soil moisture, field water level, engineering status, and local control. The water body sensing subsystem connects to water and rainfall monitoring stations, water quality monitoring stations, integrated measurement and control gates, flow measurement devices, and flow and pressure regulating valves. It is deployed at reservoirs, main canals, and water distribution points to realize the sensing of water source quantity, water quality parameters, and water flow.

3. The intelligent system for large-scale irrigation areas in southern China based on three-dimensional perception of air, ground, and water, as described in claim 1, is characterized in that... The production control network connects the dispatch center, management station and key pump stations / gates via self-built optical cables to ensure the real-time transmission of control commands; The management network connects the dispatch center to the management station and the management station to the pumping station via leased operator lines, and is used to carry video surveillance and management information transmission; The communication subsystem realizes unified conversion of NB-IoT, 4G / 5G and MQTTS protocols. Distributed monitoring points use NB-IoT to upload data, and the gate control unit uses 4G / 5G plus MQTTS encrypted protocol for transmission. The production control network and management network are physically isolated.

4. The intelligent system for large-scale irrigation areas in southern China based on three-dimensional perception of air, ground, and water, as described in claim 1, is characterized in that... The management area decision-making layer also includes a multi-source heterogeneous data spatiotemporal fusion module. This module performs two-level fusion processing on the multi-sensor observation data output by the air-ground-water three-dimensional data acquisition layer, including adaptive Kalman filtering and spatiotemporal Kriging interpolation.

5. The intelligent system for large-scale irrigation areas in southern China based on three-dimensional perception of air, ground, and water, as described in claim 4, is characterized in that... The first stage is an adaptive Kalman filter, which uses time-series observation data from each monitoring station in the irrigation district as input and defines a state vector. ,in This refers to the soil volumetric water content. For soil temperature, For the water level of the paddy fields, For channel traffic, This refers to the ammonia nitrogen concentration. Total phosphorus concentration, subscript Indicates the first Each sampling time; the state evolution equation is: The observation equation is ,in Here is the state transition matrix. For the observation matrix, and The process noise and observation noise are respectively represented by their covariance matrices. and ; with new information sequence To drive this process, the noise covariance is adaptively adjusted using the following formula: ； ； in, Here is the Kalman gain matrix. For the prediction error covariance matrix, and The adaptive learning rate is set to a value between 0.01 and 0.

1.

6. The intelligent system for large-scale irrigation areas in southern China based on three-dimensional perception of air, ground, and water, as described in claim 5, is characterized in that... The second stage is spatiotemporal Kriging interpolation, which uses the filtered state estimates of each station from the first-stage adaptive Kalman filter as input, and constructs an empirical variogram function for each physical quantity: ； in For station spacing, The distance is The number of sites and Positions and The filter state value at the location; Using an exponential variogram model Fit, where Value of a nugget. For the partial sill value, For variable range; ordinary kriging interpolation is performed based on the fitted variogram function for any unsensored location within the irrigation area. Estimate its physical quantity value: ； in The number of neighboring sites participating in the interpolation. The interpolation weights are obtained by solving the Kriging equations, with the following constraints: The interpolation results generate a rasterized spatial distribution field of the entire irrigation area, which is then input into the data base of the digital twin platform.

7. The intelligent system for large-scale irrigation areas in southern China based on three-dimensional perception of air, ground, and water, as described in claim 1, is characterized in that... The management area decision-making layer also includes an online model calibration and decision confidence assessment module, which is based on variational Bayesian inference and extended Kalman filtering. The variational Bayesian inference uses a sliding time window. Using historical forecast-observation residual sequences as evidence, the model parameter vectors were analyzed. Perform posterior inference, where For crop coefficients, Soil infiltration rate, This is the channel roughness coefficient. These are the parameters of the soil moisture characteristic curve. The crop water sensitivity coefficient The sliding window length; in variational distribution Approaching the true posterior ,in For variational parameters, Given the observation dataset within the window; update the variational parameters by maximizing the lower bound of evidence. : ； in For Kullback-Leibler divergence, The prior distribution is established based on knowledge base scheduling rules and engineering experience parameters; parameter updates are triggered when background conditions and observation completeness conditions are met. The triggering conditions are: no sensor fault alarms for 72 consecutive hours and the effective observation sample size is not less than 90% of the theoretical sampling number within the window.

8. The intelligent system for large-scale irrigation areas in southern China based on three-dimensional perception of air, ground, and water, as described in claim 7, is characterized in that... The extended Kalman filter uses variational Bayes to infer the posterior mean of the output parameters. As model parameters, real-time recursive predictions are performed on the hydraulic and soil hydrodynamic states of the irrigation area: The decision confidence assessment module calculates the relative uncertainty of the parameters posteriorly before each scheduling instruction is generated: ； in For parameters The posterior variance, Let be the posterior mean of the parameters; when the uncertainty of any parameter Exceeding the preset threshold When the model prediction is determined to be in a low-confidence state, the system automatically switches to a rule-based safe scheduling mode, suspends the use of model optimization output, and executes control according to preset conservative scheduling rules; when all parameters are... When all values ​​fall below the corresponding threshold, model-driven optimization scheduling is resumed.

9. A smart irrigation method for large-scale irrigation areas in southern China based on three-dimensional perception of air, ground, and water, as described in any one of claims 1-8, characterized in that... Includes the following steps: S1: A three-dimensional data acquisition layer (air-ground-water) is used to acquire all-element data. At the air end, drones perform inspections in autonomous and specialized modes to obtain data on irrigation area topography and crop distribution. At the ground end, soil moisture monitoring stations collect soil temperature, humidity, and volumetric water content at set intervals, paddy field water level monitoring stations continuously monitor field water levels, and video monitoring stations capture real-time images of the project's operational status. At the water body end, water and rainfall monitoring stations, water quality monitoring stations, and water measurement equipment collect data on water source quantity, quality, and flow rate at set intervals. All collected data is uploaded via the IoT platform layer. S2: The IoT platform layer performs protocol conversion and secure transmission of collected data. The communication subsystem unifies the conversion of heterogeneous protocols such as NB-IoT, 4G / 5G and MQTTS. The production control network transmits control commands, and the management network transmits video and management data. Data security is ensured through physical isolation and encrypted transmission. S3: The management area's decision-making level conducts analysis and decision-making based on the digital twin platform. The digital twin platform processes the uploaded data, updates the overall data base, and uses a knowledge base to establish mapping relationships between water conditions, engineering conditions, agricultural conditions, and water quality. Based on the model library, it conducts water resource supply and demand balance simulations, irrigation scheduling plan rehearsals, and disaster risk assessments, and generates scheduling instructions. S4: The production area execution layer responds quickly and executes scheduling commands automatically. Edge computing nodes preprocess scheduling commands locally and control the start and stop of pump station units, gate opening adjustment, and pressure adjustment of flow regulating and pressure regulating valves through a dual-mode interlocking mechanism of local and remote control. During the execution process, each monitoring device provides real-time feedback on the execution effect, forming a closed-loop control. S5: The digital twin platform continuously tracks the operation status of the irrigation area, compares the actual operation data with the simulation results, analyzes the causes of deviations, and dynamically optimizes model parameters and scheduling strategies; it also regularly compiles system operation indicators to evaluate water resource utilization efficiency and engineering operation reliability.

10. A smart irrigation method for large-scale irrigation areas in southern China based on three-dimensional perception of air, ground, and water, as described in claim 9, is characterized in that... In step S3, the processing of the uploaded data includes spatiotemporal fusion processing of multi-source heterogeneous data, which includes: First-stage adaptive Kalman filter: Using time-series observation data from each monitoring station in the irrigation district as input, a state vector is constructed. ,in This refers to the soil volumetric water content. For soil temperature, For the water level of the paddy fields, For channel traffic, This refers to the ammonia nitrogen concentration. The total phosphorus concentration is; the state evolution equation is: The observation equation is ,in and covariance matrix and Based on the new information sequence Adaptive updates; Second-level spatiotemporal Kriging interpolation: Using the Kalman filter outputs of each station as input, an empirical variogram is constructed. An exponential variogram model is fitted, and ordinary kriging interpolation is performed. The interpolation results generate a rasterized spatial distribution field of the entire irrigation area.

11. A smart irrigation method for large-scale irrigation areas in southern China based on three-dimensional perception of air, ground, and water, as described in claim 9, is characterized in that... Step S3, before generating scheduling instructions, also includes online model calibration and decision confidence assessment steps: With sliding time window Using historical forecast-observation residual sequences as evidence, the model parameter vectors were analyzed. Perform variational Bayesian inference, where For crop coefficients, Soil infiltration rate, This is the channel roughness coefficient. These are the parameters of the soil moisture characteristic curve. The crop water sensitivity coefficient; expressed in variational distribution. Approximating the true posterior, the variational parameters are updated by maximizing the lower bound of evidence. The parameter update trigger conditions are: no sensor fault alarms for 72 consecutive hours, and the effective observation sample size is not less than 90% of the theoretical sampling number within the window; With the posterior mean of the parameters Substitute the input into the extended Kalman filter to perform real-time recursive prediction of the hydraulic and soil hydrodynamic states of the irrigation area; Before each scheduling instruction is generated, the relative uncertainty of each parameter is calculated. When any Exceeding the preset threshold When the model is in a low-confidence state, switch to rule-based safe scheduling mode; when all When the value falls below the threshold, model-driven optimization scheduling is resumed.

12. A smart irrigation method for large-scale irrigation areas in southern China based on three-dimensional perception of air, ground, and water, as described in claim 9, is characterized in that... In step S3, the functions of forecasting, early warning, rehearsal, and contingency plan are executed in the following hierarchical order: Forecast: in the future Using daily weather forecast data, upstream basin rainfall forecast data, and reservoir water storage status as inputs, the model outputs the future water inflow forecast data for the headworks section. The process of water flow during the day , Simultaneously, using historical irrigation data and crop growth stages as inputs, the water demand prediction model outputs the future water demand forecasts for each irrigation area. Daily water demand ,in Number the film. ; Warning: Calculate the daily water supply shortfall separately for each irrigation area. ,in For film filling The water utilization coefficient of the canal system The daily time step is set; a three-level early warning threshold is set when... A value less than 0.15 indicates a yellow alert, a value between 0.15 and 0.30 indicates an orange alert, and a value greater than 0.30 indicates a red alert. Preliminary simulation: Using at least two candidate scheduling schemes as input, each scheme includes the planned opening sequence of each gate and the planned flow sequence of the pumping station. A hydrodynamic simulation model is constructed based on the one-dimensional Saint-Venant equations for the channel. ； ； in, The cross-sectional area of ​​the water passage. For traffic, For water level, For lateral inflow, It is the acceleration due to gravity. For friction ratio reduction, The canal bottom gradient is determined; the Preissmann four-point implicit difference scheme is used for discretization and solution, and the water arrival time, water loss rate and irrigation area water satisfaction rate of each scheme are output; Contingency Plan: With the objectives of minimizing water loss rate and maximizing irrigation water supply satisfaction rate, the simulation results of candidate schemes are non-dominated and ranked. The scheme on the Pareto front is selected as the recommended scheme, while the other schemes on the front are retained as alternative schemes.

13. A smart irrigation method for large-scale irrigation areas in southern China based on three-dimensional perception of air, ground, and water, as described in claim 9, is characterized in that... In step S3, the water resource supply and demand balance simulation includes: The entire irrigation district is divided into Each computing unit, unit Daily water requirement of crops ,in For unit The crop coefficient, Reference evapotranspiration calculated using the FAO Penman-Monteith formula; unit Net irrigation water demand Gross water requirement for irrigation ,in For effective precipitation, For unit The corresponding canal system water utilization coefficient; With the total water supply of the irrigation district and the allowable flow of each channel as constraints, and minimizing the sum of squares of the difference between the actual water supply and the gross water demand of each unit as the objective, the sequential quadratic programming method is used to solve the water allocation optimization scheme.

14. A smart irrigation method for large-scale irrigation areas in southern China based on three-dimensional perception of air, ground, and water, as described in claim 9, is characterized in that... In step S1, the acquisition period of each sensor is dynamically adjusted according to the following adaptive rules: For soil moisture monitoring stations, the data collection cycle is... Adaptive adjustment as follows: ； in, For the first The acquisition cycle of each adjustment cycle, This represents the absolute value of the change in soil volumetric water content between two consecutive data collections. The threshold for water content change. For adjustment coefficients, and These are the lower and upper scaling factors for the acquisition period, respectively.

15. A computer device, characterized in that, It includes a memory and a processor, which are interconnected and communicate with each other. The memory stores computer instructions, and the processor executes the computer instructions to perform the intelligent method for large irrigation areas in southern China based on three-dimensional perception of air, land and water, as described in any one of claims 9 to 14.

16. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to execute the intelligent method for large-scale irrigation areas in southern China based on three-dimensional perception of air, land, and water, as described in any one of claims 9 to 14.

Images