Irrigation area hydrological management and control system based on sky-ground integrated perception and AI

Abstract

The invention belongs to the field of hydrological management and control, and particularly relates to an irrigation area hydrological management and control system based on sky-ground integrated perception and AI, which comprises a sky-ground integrated perception layer, an edge calculation layer, an AI analysis decision-making layer and an intelligent management and control execution layer which are in communication connection in sequence, a water conservancy large model, a deep learning algorithm combination, a digital twin platform and a decision optimization algorithm are integrated, and the intelligent management and control execution layer comprises a control device and a visual monitoring device to form a full-link management and control system of'global perception-edge processing-intelligent decision-linkage management and control '. When the deviation absolute value of the hydrological parameter relative to the preset threshold value exceeds 10%, an abnormal response is triggered. Therefore, comprehensive coverage, stable transmission and safe storage of operation data of whole-domain hydrology, water quality and water conservancy facilities in an irrigation area are achieved, and the problems that due to the fact that traditional monitoring depends on single equipment, the number of coverage blind areas is large, data transmission is prone to being interrupted, and information is prone to being lost or leaked are solved.

Description

Technical Field

[0001] This invention belongs to the field of hydrological management and control, and in particular relates to a hydrological management and control system for irrigation districts based on integrated sky-ground perception and AI. Background Technology

[0002] Hydrological management in irrigation districts is crucial for ensuring agricultural production and the rational use of water resources. Existing technologies mostly collect hydrological data through single or partial combinations of ground sensors and satellite remote sensing, supplemented by basic data processing algorithms or simple AI models to achieve monitoring and scheduling. Some solutions attempt to integrate domestically produced hardware equipment with open-source algorithm frameworks, which has initially improved the coverage and processing efficiency of data collection.

[0003] However, existing technologies generally suffer from the core problem of insufficient decision-making efficiency. Even with the introduction of basic intelligent algorithms, the lack of spatiotemporal fusion of multi-source data and the limited ability of models to capture the spatiotemporal characteristics and global correlations of hydrological data lead to delayed identification of abnormal events and long time consumption in generating scheduling schemes, making it difficult to adapt to the needs of modern irrigation district dynamic management for real-time response and accurate decision-making. Summary of the Invention

[0004] The purpose of this invention is to address the core problem of insufficient decision-making efficiency in the existing technologies mentioned in the background section. Even with the introduction of basic intelligent algorithms, the lack of spatiotemporal fusion of multi-source data and the limited ability of models to capture the spatiotemporal characteristics and global correlations of hydrological data lead to delayed identification of abnormal events and long time consumption in generating scheduling schemes, making it difficult to adapt to the needs of modern irrigation district dynamic management for real-time response and accurate decision-making. This invention provides an irrigation district hydrological management and control system based on integrated sky-ground perception and AI.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a hydrological management and control system for irrigation districts based on integrated sky-ground perception and AI, employing a cloud-edge-device layered distributed architecture, including a sky-ground integrated perception layer, an edge computing layer, an AI analysis and decision-making layer, and an intelligent management and control execution layer connected in sequence. Each layer interacts with data through a standard interface, with a data transmission rate ≥100Mbps. The integrated sky-ground perception layer is a three-dimensional monitoring network encompassing sky, air, ground, water, and engineering, composed of satellites, drones, ground sensors, underwater sensors, and engineering sensors. Each device coordinates monitoring through a spatiotemporal synchronization calibration mechanism, and the trigger condition for satellite and drone data completion is ground sensor data. The missing time is ≥5min; the edge computing layer is built with dual-chip collaboration and AI acceleration as the core to construct edge processing nodes and realize lightweight data processing; the AI ​​analysis and decision-making layer is built on the technology stack, integrating water conservancy big data model, deep learning algorithm combination, digital twin platform and decision optimization algorithm, and the physical scene mapping accuracy of digital twin platform is ≥95%; the intelligent control and execution layer includes control equipment and visual monitoring equipment to realize command implementation and status feedback; data transmission adopts MQTT and national cryptographic SM4 encryption protocol to form a full-link control system of "full-domain perception-edge processing-intelligent decision-making-linked control", and an abnormal response is triggered when the absolute value of the deviation of hydrological parameters from the preset threshold exceeds 10%.

[0006] Furthermore, the standard interface is compatible with MQTT, Modbus and HTTP protocols, with the priority being to use MQTT protocol first, automatically switch to Modbus protocol when communication is abnormal, and switch to HTTP protocol when abnormality occurs again; the spatiotemporal synchronization calibration mechanism consists of "satellite time synchronization + timestamp alignment algorithm", with satellite time synchronization error ≤1ms and calibration cycle ≤30s.

[0007] Furthermore, the technology stack of the AI ​​analysis and decision-making layer consists of a "deep learning framework + big data processing platform + simulation engine". The collaborative logic of the three is that the big data processing platform outputs preprocessed data to the deep learning framework, and inputs the inference results of the deep learning framework into the simulation engine. The physical scene mapping accuracy of the digital twin platform is based on the weighted calculation of coordinate deviation and hydrological parameter error. The weighting coefficient is a fixed proportion, and both coordinate deviation and hydrological parameter error are determined by the degree of deviation between the actual value and the simulation value.

[0008] Furthermore, the preset thresholds for the hydrological parameters are set based on the industry standards for irrigation district design and the 95th percentile of historical data statistically analyzed by irrigation district over the past 10 years; the AI ​​analysis and decision-making layer and the intelligent control and execution layer achieve instruction interaction through the MQTT extended protocol, and the simulation error of the digital twin platform is <5%.

[0009] Furthermore, the integrated space-ground sensing layer includes navigation satellites and high-resolution satellites at the celestial level, drones and their onboard lidar and multispectral cameras at the air level, water level gauges and flow meters at the ground level, water quality sensors and Doppler current meters at the water level, and gate opening sensors and pump station vibration sensors at the engineering level. Data fusion adopts a Kalman filter algorithm based on spatiotemporal weight adaptive adjustment of filter gain, and the monitoring accuracy after data fusion is ≤±2%.

[0010] Furthermore, the core hardware of the edge computing layer includes a main processor, an auxiliary computing chip, and integrates a 5G module, an NB-IoT module, and an AI acceleration module; the ResNet-50 model is distilled into a lightweight version of MobileNet-FP8, with an inference latency of ≤50ms for the lightweight model, and the recognition response time is the sum of the edge computing layer data preprocessing latency and inference latency, with an overall recognition response time of ≤100ms, achieving millisecond-level recognition of hazards such as piping and landslides.

[0011] Furthermore, the AI ​​analysis and decision-making layer adopts a dual-engine architecture consisting of a "model training engine + inference acceleration engine", which improves inference efficiency by more than 30% compared to a single-engine architecture. The deep learning algorithm combination is CNN-LSTM-Transformer, and the decision optimization algorithm is an improved NSGA-II multi-objective optimization algorithm, which is applied to the irrigation district water resource allocation and emergency response scheduling scenario. The algorithm output is directly mapped to scheduling instructions.

[0012] Furthermore, the core control equipment of the intelligent management and control execution layer includes a PLC, a frequency converter, and an intelligent gate controller. The auxiliary monitoring equipment includes an alarm system and video surveillance, supporting dual-mode control in both on-site and remote modes. When an anomaly is triggered, an audible and visual alarm, SMS, and APP push are activated simultaneously, with an alarm response delay of ≤3s. A hierarchical decision-making architecture is adopted, with an upstream and downstream linkage instruction transmission delay of <100ms, water distribution accuracy of ±5%, and a scheduling scheme update cycle of ≤5 minutes.

[0013] Furthermore, the integrated space-ground sensing layer adopts a hierarchical acquisition strategy: high-frequency acquisition once per minute for key monitoring points, standard acquisition once per 5 minutes to once per 15 minutes for general monitoring points, and low-frequency acquisition once per 30 minutes for auxiliary monitoring points; data transmission adopts a multi-mode solution of "BeiDou short message + NB-IoT + ZMESH self-organizing network", each monitoring point is equipped with a storage device with a storage capacity of ≥1TB, the data retention period is ≥90 days, and the data transmission success rate is ≥99.5%, which is the daily average statistical value after 24 hours of continuous operation.

[0014] A method for hydrological management and control in irrigation districts based on integrated sky-ground sensing and AI, using the system described in any one of claims 1-9, includes the following steps: S1. Collect hydrological, water quality and water conservancy facility operation data of irrigation area according to hierarchical collection strategy, and transmit them to the edge computing layer through multi-mode transmission and national cryptographic SM4 encryption. S2. The edge computing layer preprocesses the data and lightweights AI inference, initially screening out abnormal data with an accuracy of ≥85% before uploading it to the AI ​​analysis and decision-making layer; S3. Anomaly identification, hydrological prediction, and scheme simulation are achieved through deep learning algorithms and a digital twin platform, and the optimal scheme is generated using an improved NSGA-II algorithm; S4. The intelligent control execution layer receives instructions to link and control facilities, pushes alarm information and handling status, and the facility control response delay is ≤500ms; the improved NSGA-II algorithm outputs Pareto optimal solution set through 100-200 iterations, balancing the three major objectives of flood control safety, irrigation demand and ecological protection.

[0015] Compared with existing technologies, the advantages of the irrigation district hydrological management system based on integrated sky-ground sensing and AI of this invention are as follows: 1. This invention utilizes a multi-dimensional monitoring device integrating space, air, ground, water, and engineering elements within a space-ground integrated sensing layer, a hierarchical data acquisition strategy, and a multi-mode transmission scheme combining BeiDou short message service, NB-IoT, and ZMESH self-organizing network. Coupled with local storage devices at each monitoring point and a full-link encryption mechanism, it achieves comprehensive coverage, stable transmission, and secure storage of hydrological, water quality, and water conservancy facility operation data across the entire irrigation area. This solves the problems of traditional monitoring relying on single devices, which leads to numerous blind spots, easy data transmission interruptions, and easy information loss or leakage.

[0016] 2. By leveraging the dual-chip collaboration and AI acceleration module of the edge computing layer, combined with the "deep learning framework + big data processing platform + simulation engine" technology stack of the AI ​​analysis and decision-making layer, the CNN-LSTM-Transformer algorithm combination, the dual-engine architecture, and the digital twin platform, rapid data preprocessing, accurate initial screening of anomalies, in-depth mining of hydrological features, and pre-demonstration verification of decision-making schemes are achieved. This solves the problems of long data processing time, incomplete feature capture, and lack of scientific basis for decision-making in existing technologies, resulting in low decision-making efficiency and poor accuracy.

[0017] 3. Through core control equipment such as PLCs and intelligent gate controllers in the intelligent control and execution layer, dual-mode control design for on-site and remote operation, upstream and downstream linkage mechanism, and multi-channel alarm push function, coupled with standard interfaces for multi-protocol adaptation at each level and spatiotemporal synchronization calibration mechanism, the system achieves rapid response to decision-making instructions, precise linkage control of water conservancy facilities, and real-time feedback on abnormal handling status. This solves the problems of delayed response, poor linkage, and single control mode in traditional control and execution, which lead to untimely scheduling and large execution deviations. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the irrigation district hydrological management system based on integrated sky-ground perception and AI of the present invention. Detailed Implementation

[0019] The following examples are for illustrative purposes only and are not intended to limit the scope of the invention.

[0020] like Figure 1 As shown, this application proposes an irrigation district hydrological management and control system based on integrated sky-ground perception and AI. It adopts a layered distributed architecture of cloud-edge-device, including a sky-ground integrated perception layer, an edge computing layer, an AI analysis and decision-making layer, and an intelligent management and control execution layer, which are connected in sequence. Each layer interacts with data through a standard interface with a data transmission rate ≥100Mbps. The sky-ground integrated perception layer is a three-dimensional monitoring network covering sky, air, ground, water, and engineering, composed of satellites, drones, ground sensors, underwater sensors, and engineering sensors. All devices coordinate monitoring through a spatiotemporal synchronization calibration mechanism. The trigger condition for satellite and drone data completion is that the duration of missing ground sensor data is ≥ 5 min; The edge computing layer is built around dual-chip collaboration and AI acceleration to construct edge processing nodes and achieve lightweight data processing; The AI ​​analysis and decision-making layer is built on a technology stack, integrating a large water conservancy model, a combination of deep learning algorithms, a digital twin platform, and decision optimization algorithms. The physical scene mapping accuracy of the digital twin platform is ≥95%; The intelligent control and execution layer includes control equipment and visual monitoring equipment to realize command implementation and status feedback; Data transmission adopts MQTT and the national cryptographic SM4 encryption protocol to form a full-link control system of "full-domain perception - edge processing - intelligent decision-making - linkage control". When the absolute value of the deviation of hydrological parameters from the preset threshold exceeds 10%, an abnormal response is triggered.

[0021] It should be noted that the cloud-edge-device layered distributed architecture described in this embodiment can flexibly adjust the number of devices deployed and the computing power configuration at each level according to the scale of the irrigation area. It can not only meet the massive data processing needs of large irrigation areas, but also adapt to the lightweight application scenarios of small and medium-sized irrigation areas. The various types of devices in the integrated space-ground sensing layer can be selectively combined and deployed according to the terrain and climate characteristics of the irrigation area to achieve a reasonable allocation of monitoring resources. At the same time, the end-to-end encrypted transmission ensures the security of hydrological data during the collection, transmission and processing process, and avoids data leakage or tampering from affecting management and control decisions.

[0022] Specifically, the system achieves comprehensive collection of hydrological data in the irrigation area through an integrated space-ground sensing layer, solving the problem of incomplete coverage in traditional monitoring; it preprocesses massive amounts of data at the edge computing layer, reducing the interference of invalid data on subsequent decision-making; the AI ​​analysis and decision-making layer deeply mines the value of data with professional models and simulation platforms to generate scientific decision-making solutions; and the intelligent control and execution layer quickly implements instructions and provides feedback on the status, forming a complete control closed loop. The collaborative operation of each layer significantly reduces the time spent on the entire process from data collection to decision execution, effectively solving the core problem of insufficient decision-making efficiency in existing technologies.

[0023] Furthermore, the standard interface is compatible with MQTT, Modbus and HTTP protocols, with MQTT protocol being the preferred protocol. In case of communication failure, it automatically switches to Modbus protocol, and in case of further failure, it switches to HTTP protocol. The spatiotemporal synchronization calibration mechanism consists of "satellite time synchronization + timestamp alignment algorithm", with satellite time synchronization error ≤1ms and calibration cycle ≤30s.

[0024] It should be noted that the multi-protocol adaptation design described in this embodiment is compatible with monitoring equipment and control terminals from different manufacturers and of different types, reducing the cost of system upgrades or equipment replacements; the automatic protocol switching mechanism requires no manual intervention, ensuring the continuity of communication links, and is especially suitable for remote irrigation areas with unstable signals; the spatiotemporal synchronization calibration mechanism is not only applicable to new equipment, but can also be used to synchronously upgrade existing old equipment, improving the collaborative working capability of existing equipment.

[0025] Specifically, the standard interface, through multi-protocol adaptation and priority switching, ensures smooth data interaction between devices at different levels and of different types, avoiding data transmission interruptions due to protocol incompatibility. The spatiotemporal synchronization calibration mechanism unifies the time and spatial references of all monitoring devices, enabling spatiotemporal consistency of multi-source acquired data and providing a reliable foundation for subsequent data fusion and intelligent decision-making. This design solves the problem of decision lag caused by poor device coordination and data asynchrony in existing technologies, improving the timeliness of control and management response.

[0026] Furthermore, the AI ​​analysis and decision-making layer's technology stack consists of a "deep learning framework + big data processing platform + simulation engine." The collaborative logic of these three components is to output preprocessed data from the big data processing platform to the deep learning framework, and input the inference results of the deep learning framework into the simulation engine. The physical scene mapping accuracy of the digital twin platform is based on a weighted calculation of coordinate deviation and hydrological parameter error, with a fixed weighting coefficient. Both coordinate deviation and hydrological parameter error are determined by the degree of deviation between the actual value and the simulated value.

[0027] It should be noted that the technology stack described in this embodiment supports modular expansion, and special algorithm modules or data processing functions can be added according to the needs of irrigation district management, adapting to the decision-making needs of different scenarios such as flood control, irrigation, and ecological protection; the weighted calculation method of the digital twin platform can be flexibly adjusted according to the core management objectives of the irrigation district, highlighting the mapping accuracy of key parameters, so that the simulation results are more in line with the actual application needs.

[0028] Specifically, the big data processing platform cleans and integrates multi-source collected data to provide high-quality data input for the deep learning framework; the deep learning framework mines the spatiotemporal features and correlations in the data to generate preliminary decision-making basis; and the simulation engine uses a digital twin platform to pre-run the decision-making scheme and verify its feasibility. This collaborative process ensures that the decision-making scheme is based on real data and verified through simulation, solving the problems of low efficiency and poor accuracy caused by the lack of data support and pre-run verification in existing technologies.

[0029] Furthermore, the preset thresholds for hydrological parameters are set based on the industry standards for irrigation district design and the 95th percentile of historical data statistically analyzed by irrigation district over the past 10 years; the AI ​​analysis and decision-making layer and the intelligent control and execution layer achieve command interaction through the MQTT extended protocol, and the simulation error of the digital twin platform is <5%.

[0030] It should be noted that the preset thresholds for hydrological parameters described in this embodiment are not fixed and can be dynamically updated based on the actual hydrological conditions each year and the adjustment of crop planting structure in the irrigation area to ensure the rationality of the threshold settings. The MQTT extended protocol supports custom command formats and can flexibly configure command content according to the control needs of different water conservancy facilities, thereby improving the adaptability and flexibility of command interaction.

[0031] Specifically, the preset thresholds for hydrological parameters are set based on industry standards and historical data, conforming to the general patterns of long-term irrigation district operation while adapting to the individual circumstances of different zones, making anomaly response triggering more precise. The MQTT extended protocol ensures the efficiency and accuracy of command transmission between the AI ​​analysis and decision-making layer and the intelligent control and execution layer, while the low-error simulation of the digital twin platform makes the decision-making scheme more reliable. This design solves the problems of unreasonable threshold settings, delayed anomaly identification caused by poor command transmission, and decision execution deviations in existing technologies, improving the accuracy and efficiency of control.

[0032] Furthermore, the integrated space-ground sensing layer includes navigation satellites and high-resolution satellites at the space level, drones and their onboard lidar and multispectral cameras at the air level, water level gauges and flow meters at the ground level, water quality sensors and Doppler current meters at the water level, and gate opening sensors and pump station vibration sensors at the engineering level. Data fusion adopts a Kalman filter algorithm based on spatiotemporal weight adaptive adjustment of filter gain, and the monitoring accuracy after data fusion is ≤±2%.

[0033] It should be noted that the device combination of the integrated space-ground sensing layer described in this embodiment supports redundant deployment. The number of devices can be increased in key monitoring areas to improve the reliability of data acquisition. The combination of lidar and multispectral camera can not only capture physical features such as terrain and water level, but also help to judge crop water requirements, water pollution status, etc., enriching the monitoring dimensions. The data fusion algorithm can dynamically adjust the weights according to the acquisition accuracy of different devices to further improve the accuracy of fused data.

[0034] Specifically, the collaborative data collection from multiple dimensions—air, ground, water, and engineering—achieved comprehensive coverage of data on irrigation district hydrology, water quality, and facility status, solving the problem of numerous blind spots in traditional monitoring. Data fusion algorithms integrated multi-source data, eliminating the errors and limitations of single-device data collection and providing high-quality data support for subsequent decision-making. This design ensures the integrity and accuracy of information from the source of data collection, laying a data foundation for addressing the inefficiencies of existing technology in decision-making.

[0035] Furthermore, the core hardware of the edge computing layer includes a main processor, an auxiliary computing chip, and integrates a 5G module, an NB-IoT module, and an AI acceleration module; the ResNet-50 model is distilled into a lightweight version of MobileNet-FP8, with an inference latency of ≤50ms for the lightweight model, and the recognition response time is the sum of the data preprocessing latency and inference latency of the edge computing layer, with an overall recognition response time of ≤100ms, achieving millisecond-level recognition of hazards such as piping and landslides.

[0036] It should be noted that the edge computing layer hardware described in this embodiment supports computing power expansion. Data processing capabilities can be improved by increasing the number of AI acceleration modules to meet the growing data volume requirements of irrigation areas. The lightweight model not only reduces the hardware resource consumption of edge nodes, but also has good migration capabilities, which can quickly adapt to different types of irrigation area hazard identification scenarios without the need for repeated model training.

[0037] Specifically, the edge computing layer enhances data processing speed through dual-chip collaboration and an AI acceleration module, completing data preprocessing and initial anomaly identification locally, reducing the amount of data transmitted to the cloud and the transmission latency; the lightweight model quickly mines abnormal features in the data, enabling real-time identification of hazards. This design solves the problem of untimely hazard identification caused by large data transmission volumes and lagging remote processing in existing technologies, buying time for subsequent rapid decision-making.

[0038] Furthermore, the AI ​​analysis and decision-making layer adopts a dual-engine architecture consisting of a "model training engine + inference acceleration engine", which improves inference efficiency by more than 30% compared to a single-engine architecture. The deep learning algorithm combination is CNN-LSTM-Transformer, and the decision optimization algorithm is an improved NSGA-II multi-objective optimization algorithm, which is applied to irrigation district water resource allocation and emergency response scheduling scenarios. The algorithm output is directly mapped to scheduling instructions.

[0039] It should be noted that the dual-engine architecture described in this embodiment supports load balancing and can dynamically allocate the workload of the two engines according to the amount of data processed, avoiding the decrease in processing efficiency caused by overloading a single engine; the combination of deep learning algorithms can flexibly adjust the weight ratio of each algorithm according to different data characteristics, and the decision optimization algorithm can adjust the optimization direction according to the core objectives of the irrigation area at different times (such as prioritizing irrigation during the irrigation period and prioritizing flood control during the flood season).

[0040] Specifically, the model training engine continuously optimizes the performance of the large-scale hydrological model, while the inference acceleration engine improves the speed of decision-making inference. The CNN-LSTM-Transformer combination comprehensively captures the spatial, temporal, and global correlation features of hydrological data. The improved decision optimization algorithm balances the needs of multiple objectives to generate the optimal solution, and the algorithm output is directly mapped to scheduling instructions, reducing intermediate conversion steps. This design solves the problems of slow model inference, incomplete feature capture, and cumbersome decision solution generation in existing technologies, resulting in low decision-making efficiency and improving the speed and scientific rigor of decision-making.

[0041] Furthermore, the core control equipment of the intelligent management and control execution layer includes PLC, frequency converter, and intelligent gate controller. The auxiliary monitoring equipment includes alarm system and video surveillance, supporting on-site and remote dual-mode control. When an anomaly is triggered, audible and visual alarms, SMS and APP push notifications are activated simultaneously, with an alarm response delay of ≤3s. A hierarchical decision-making architecture is adopted, with an upstream and downstream linkage instruction transmission delay of <100ms, water distribution accuracy of ±5%, and a scheduling scheme update cycle of ≤5 minutes.

[0042] It should be noted that the dual-mode control design described in this embodiment allows switching to on-site control when remote communication is interrupted, ensuring the continuity of management and control; alarm information push supports multi-channel hierarchical push, and can push alarm information of corresponding level of detail according to the responsibilities and permissions of the receiving object, thereby improving the efficiency of handling; the hierarchical decision-making architecture can adjust the time cycle of decision-making at each level according to the needs of irrigation district management, adapting to management and control needs of different scales.

[0043] Specifically, the core control equipment responds quickly to decision commands, enabling precise control of water conservancy facilities; dual-mode control and multi-channel alarms ensure timely handling of abnormal events; and the hierarchical decision-making architecture and upstream-downstream linkage mechanism allow the scheduling scheme to both conform to long-term planning and respond quickly to real-time changes. This design solves the scheduling lag problems caused by slow execution response, single control mode, and poor linkage in existing technologies, improving the efficiency and flexibility of control execution.

[0044] Furthermore, the integrated space-ground sensing layer adopts a hierarchical acquisition strategy: key monitoring points acquire data at a high frequency of once per minute, general monitoring points acquire data at a standard frequency of once per 5 minutes to once per 15 minutes, and auxiliary monitoring points acquire data at a low frequency of once per 30 minutes. Data transmission adopts a multi-mode solution of "BeiDou short message + NB-IoT + ZMESH self-organizing network". Each monitoring point is equipped with a storage device with a storage capacity of ≥1TB, and the data transmission success rate is ≥99.5%, which is the daily average statistical value after 24 hours of continuous operation.

[0045] It should be noted that the hierarchical data acquisition strategy described in this embodiment can dynamically adjust the acquisition frequency according to the importance of the monitoring points. For example, the acquisition frequency of key monitoring points can be increased during the flood season, and the acquisition frequency of general monitoring points can be appropriately reduced during the non-irrigation season to achieve rational use of resources. The multi-mode transmission scheme can automatically select the optimal transmission method according to the communication conditions of different regions. The storage device supports local data caching and breakpoint resume transmission to avoid data loss.

[0046] Specifically, the tiered data acquisition strategy ensures the real-time nature of critical data while reducing unnecessary resource consumption; the multi-mode transmission scheme achieves stable data transmission across the entire irrigation area, particularly solving communication challenges in remote areas; local storage and breakpoint resume functionality ensure data integrity. This design addresses the problems of insufficient decision-making data support caused by unreasonable data acquisition, incomplete transmission coverage, and easy data loss in existing technologies, providing comprehensive and continuous data support for efficient decision-making.

[0047] A method for hydrological management and control in irrigation districts based on integrated sky-ground perception and AI, using the system of any one of claims 1-9, includes the following steps: S1. Collecting hydrological, water quality, and water conservancy facility operation data of the irrigation district according to a hierarchical acquisition strategy, and transmitting the data to the edge computing layer via multi-mode transmission and SM4 encryption; S2. Preprocessing the data and lightweighting AI inference at the edge computing layer, initially screening for abnormal data with an accuracy of ≥85%, and then uploading the data to the AI ​​analysis and decision-making layer; S3. Realizing anomaly identification, hydrological prediction, and scheme simulation through deep learning algorithms and a digital twin platform, and generating the optimal scheme using an improved NSGA-II algorithm; S4. Receiving instructions and linking control facilities at the intelligent management and control execution layer, pushing alarm information and handling status, with facility control response latency ≤500ms; The improved NSGA-II algorithm outputs a Pareto optimal solution set through 100-200 iterations, balancing the three major objectives of flood control safety, irrigation demand, and ecological protection.

[0048] It should be noted that the control method described in this embodiment supports parallel processing and dynamic adjustment in each step. For example, data acquisition in S1 and edge preprocessing in S2 can be partially synchronized, and the scheme simulation in S3 can be optimized in reverse based on the execution feedback in S4. The initial screening of abnormal data supports custom screening rules, and the screening logic can be adjusted according to common abnormal types in the irrigation area to improve the accuracy of the initial screening.

[0049] Specifically, S1 achieves comprehensive data collection and secure transmission across the entire domain, resolving issues of incomplete data coverage and insecure transmission; S2 rapidly preprocesses and screens for anomalies, reducing the transmission and processing time of invalid data; S3 generates scientific decision-making solutions through professional algorithms and simulation platforms, addressing the problems of lack of evidence and low efficiency in decision-making; and S4 quickly executes instructions and provides status feedback, forming a closed-loop control system. The entire methodology is interconnected, significantly reducing the time from data collection to decision execution, effectively solving the core problem of insufficient decision-making efficiency in existing technologies.

[0050] In summary, a layered distributed architecture of cloud-edge-device is adopted. The integrated space-air-ground-water-engineering sensing layer collects hydrological, water quality, and facility operation data of the irrigation area according to a hierarchical strategy. After multi-mode transmission and encryption, the data is sent to the edge computing layer. Dual-chip collaboration and AI acceleration modules complete data preprocessing and lightweight AI inference. After initial screening of abnormal data, it is uploaded to the AI ​​analysis and decision-making layer. The AI ​​analysis and decision-making layer relies on a technology stack consisting of a deep learning framework, a big data processing platform, and a simulation engine. It captures the spatiotemporal characteristics and global correlations of the data through a combination of CNN-LSTM-Transformer algorithms. Combined with a digital twin platform pre-simulation scheme, the improved NSGA-II algorithm generates optimal scheduling instructions. Finally, the intelligent control and execution layer controls the water conservancy facilities in conjunction with the control equipment, simultaneously pushing alarm information and handling status through multiple channels. Each layer ensures smooth data interaction and spatiotemporal consistency through multi-protocol adapted standard interfaces and a spatiotemporal synchronization calibration mechanism, forming a closed-loop system of "full-domain perception - edge processing - intelligent decision-making - coordinated control". This working principle specifically addresses the core issue of insufficient decision-making efficiency raised by the background technologies: the integrated space-ground sensing layer eliminates the coverage blind spots of traditional monitoring, and multi-mode transmission and local storage ensure data integrity and security, laying a data foundation for efficient decision-making; the edge computing layer preprocesses data and screens for anomalies, reducing the transmission and processing time of invalid data; the AI ​​analysis and decision-making layer, through professional algorithm combinations and digital twin pre-simulation, ensures that decision-making solutions are both data-supported and feasibility-verified, and the dual-engine architecture and algorithm-directly mapped instruction design significantly improve the speed of inference and solution generation; the intelligent control and execution layer's dual-mode control, upstream and downstream linkage mechanism, and rapid response capability ensure that decision-making instructions are implemented in a timely manner, and the collaborative operation of all links in the entire chain significantly compresses the entire process time from data collection to decision execution. At the same time, through spatiotemporal synchronization calibration, dynamic threshold adjustment, and multi-protocol adaptation, it solves derivative problems such as poor equipment coordination, data asynchrony, and unreasonable thresholds, ultimately achieving efficient and accurate response in irrigation area hydrological management.

[0051] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A hydrological management and control system for irrigation districts based on integrated sky-ground sensing and AI, characterized in that, The system adopts a layered distributed architecture of cloud-edge-device, comprising a space-ground integrated sensing layer, an edge computing layer, an AI analysis and decision-making layer, and an intelligent control and execution layer, all interconnected in sequence. Each layer interacts with data via a standard interface with a data transmission rate ≥100Mbps. The space-ground integrated sensing layer is a comprehensive, three-dimensional monitoring network encompassing air, ground, water, and engineering sensors, composed of satellites, drones, ground sensors, underwater sensors, and engineering sensors. These devices coordinate monitoring through a spatiotemporal synchronization calibration mechanism. Data completion for satellites and drones is triggered when ground sensor data is missing for ≥5 minutes. The edge computing layer utilizes dual-chip collaboration. The system utilizes AI acceleration to build edge processing nodes, enabling lightweight data processing. The AI ​​analysis and decision-making layer is built upon a technology stack, integrating a large-scale water conservancy model, deep learning algorithms, a digital twin platform, and decision optimization algorithms. The digital twin platform achieves a physical scene mapping accuracy of ≥95%. The intelligent control and execution layer includes control equipment and visual monitoring equipment to implement command execution and status feedback. Data transmission adopts MQTT and the national cryptographic SM4 encryption protocol, forming a full-link control system of "full-domain perception - edge processing - intelligent decision-making - coordinated control". An abnormal response is triggered when the absolute value of the deviation of hydrological parameters from the preset threshold exceeds 10%.

2. The system according to claim 1, characterized in that, The standard interface is compatible with MQTT, Modbus and HTTP protocols. The priority of the compatibility is to use MQTT protocol first, and automatically switch to Modbus protocol when communication is abnormal, and switch to HTTP protocol when abnormal again. The spatiotemporal synchronization calibration mechanism consists of "satellite time synchronization + timestamp alignment algorithm", with satellite time synchronization error ≤1ms and calibration cycle ≤30s.

3. The system according to claim 1, characterized in that, The AI ​​analysis and decision-making layer's technology stack consists of a "deep learning framework + big data processing platform + simulation engine." The collaborative logic of these three components is that the big data processing platform outputs preprocessed data to the deep learning framework, and the inference results of the deep learning framework are input into the simulation engine. The physical scene mapping accuracy of the digital twin platform is based on a weighted calculation of coordinate deviation and hydrological parameter error, with a fixed weighting coefficient. Both coordinate deviation and hydrological parameter error are determined by the degree of deviation between the actual values ​​and the simulated values.

4. The system according to claim 1, characterized in that, The preset thresholds for the hydrological parameters are set based on the industry standards for irrigation district design and the 95th percentile of historical data statistically analyzed by irrigation district over the past 10 years. The AI ​​analysis and decision-making layer and the intelligent control and execution layer achieve command interaction through the MQTT extended protocol, and the simulation error of the digital twin platform is <5%.

5. The system according to claim 1, characterized in that, The integrated space-ground sensing layer includes navigation satellites and high-resolution satellites at the sky level, drones and their onboard lidar and multispectral cameras at the air level, water level gauges and flow meters at the ground level, water quality sensors and Doppler current meters at the water level, and gate opening sensors and pump station vibration sensors at the engineering level. Data fusion adopts a Kalman filter algorithm based on spatiotemporal weight adaptive adjustment of filter gain, and the monitoring accuracy after data fusion is ≤±2%.

6. The system according to claim 1, characterized in that, The core hardware of the edge computing layer includes a main processor, an auxiliary computing chip, and integrates a 5G module, an NB-IoT module, and an AI acceleration module. The ResNet-50 model is distilled into a lightweight version of MobileNet-FP8. The inference latency of the lightweight model is ≤50ms, and the recognition response time is the sum of the edge computing layer data preprocessing latency and inference latency. The overall recognition response time is ≤100ms, enabling millisecond-level recognition of hazards such as piping and landslides.

7. The system according to claim 1, characterized in that, The AI ​​analysis and decision-making layer adopts a dual-engine architecture consisting of a "model training engine + inference acceleration engine", which improves inference efficiency by more than 30% compared with the single-engine architecture. The deep learning algorithm combination is CNN-LSTM-Transformer, and the decision optimization algorithm is an improved NSGA-II multi-objective optimization algorithm, which is applied to the irrigation area water resource allocation and emergency response scheduling scenario. The algorithm output is directly mapped to scheduling instructions.

8. The system according to claim 1, characterized in that, The core control equipment of the intelligent management and control execution layer includes PLC, frequency converter, and intelligent gate controller. The auxiliary monitoring equipment includes alarm system and video monitoring. It supports dual-mode control of on-site and remote. When an abnormality is triggered, it will simultaneously start audible and visual alarm, SMS and APP push. The alarm response delay is ≤3s. The hierarchical decision-making architecture is adopted, with a transmission delay of less than 100ms for upstream and downstream linkage instructions, a water distribution accuracy of ±5%, and a scheduling scheme update cycle of ≤5 minutes.

9. The system according to claim 1, characterized in that, The integrated space-ground sensing layer adopts a hierarchical acquisition strategy: high-frequency acquisition once per minute for key monitoring points, standard acquisition once per 5 minutes to once per 15 minutes for general monitoring points, and low-frequency acquisition once per 30 minutes for auxiliary monitoring points; data transmission adopts a multi-mode solution of "BeiDou short message + NB-IoT + ZMESH self-organizing network", each monitoring point is equipped with a storage device with a storage capacity of ≥1TB, the data retention period is ≥90 days, and the data transmission success rate is ≥99.5%, which is the daily average statistical value after 24 hours of continuous operation.

10. A method for hydrological management and control in irrigation districts based on integrated sky-ground sensing and AI, characterized in that, The system according to any one of claims 1-9 is characterized by the following steps: S1. Collect hydrological, water quality and water conservancy facility operation data of irrigation area according to hierarchical collection strategy, and transmit them to the edge computing layer through multi-mode transmission and national cryptographic SM4 encryption. S2. The edge computing layer preprocesses the data and lightweights AI inference, initially screening out abnormal data with an accuracy of ≥85% before uploading it to the AI ​​analysis and decision-making layer; S3. Anomaly identification, hydrological prediction, and scheme simulation are achieved through deep learning algorithms and a digital twin platform, and the optimal scheme is generated using an improved NSGA-II algorithm; S4. The intelligent control execution layer receives instructions to link and control facilities, pushes alarm information and handling status, and the facility control response delay is ≤500ms; the improved NSGA-II algorithm outputs Pareto optimal solution set through 100-200 iterations, balancing the three major objectives of flood control safety, irrigation demand and ecological protection.

Images