A remote intelligent monitoring system for water conservancy gate

Abstract

A kind of water conservancy gate remote intelligent monitoring system, it is characterized in that, including perception acquisition module, edge autonomous computing module, cloud global scheduling module, gate execution drive module;Gate station locally deployed edge autonomous computing module, integrated multi-modal fusion, adaptive control and instruction arbitration unit, supporting three-layer redundant communication link and partition cache mechanism.Edge module is built-in multi-modal fusion logic with hydraulic correction, fusion image and hydrological time series data, reduce environmental interference;Hydrology feedforward is combined with feedback control mode, superimposed mechanical, flood control, three constraints of energy consumption, automatically exclude out-of-limit opening instruction;Instruction arbitration unit sets three operation priority interlock, avoid multi-source instruction conflict. When communication is normal, cloud collaborative scheduling, when communication is interrupted, edge can independently complete closed-loop control, effectively improve the reliability of gate station flood season operation, reduce wireless transmission flow and equipment operation risk.

Description

Technical Field

[0001] This invention relates to the field of water conservancy and hydropower, and mainly to a remote intelligent monitoring system for water conservancy gates. Background Technology

[0002] Water conservancy gates are core control facilities for flood control in river basins and water resource allocation in irrigation areas. Currently, the water conservancy industry is widely promoting remote intelligent monitoring mode. By deploying image, hydrological, and mechanical condition sensing equipment at the gate station, the on-site monitoring data is uploaded to the remote server through wireless network or fiber optics. The gate opening is remotely and automatically adjusted based on automatic control algorithms, thereby reducing the workload of on-site manual duty and improving the efficiency of water resource allocation.

[0003] The invention patent CN119310837B, entitled "A Remote Control System for Hydraulic Gates," proposes a remote control system for gates based on multimodal visual data reconstruction. It consists of a series of modules for data acquisition, transmission, cloud processing, adaptive control, and gate execution. It relies on visible light and infrared cameras to acquire image data, performs multimodal data fusion through matrix decomposition, and solves control parameters using an optimization network with a forgetting elimination mechanism. However, this scheme centrally deploys all computations on a remote server, with no local computing power on-site. If communication is interrupted, the gate will completely lose its autonomous control capability. Furthermore, it only uses current-time data for delayed feedback control, without incorporating hydrological forecasting feedforwards or setting multi-dimensional safety constraints related to machinery and ecology. The data fusion is not corrected for hydraulic principles, leading to a significant decrease in control accuracy under conditions of heavy rain and sediment obstruction.

[0004] In summary, existing cloud-based centralized gate control systems suffer from problems such as loss of control due to network outages, delayed regulation, and data distortion in harsh environments, failing to meet the actual needs of flood control safety and long-term stable scheduling in water conservancy projects. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide a remote intelligent monitoring system for water conservancy gates.

[0006] The technical solution adopted by this invention to solve its technical problem is as follows: A remote intelligent monitoring system for water conservancy gates includes a sensing and acquisition module, an edge autonomous computing module, a cloud-based global scheduling module, and a gate execution drive module. The sensing and acquisition module is used to collect three types of raw monitoring data: visual images of the river channel, hydrological time series, and gate mechanical operating conditions. All raw monitoring data are transmitted to the edge autonomous computing module in real time via the bus. The edge autonomous computing module is deployed inside the gate control cabinet. It integrates a multi-link communication driver subunit, a link monitoring subunit, a local cache subunit, a hierarchical transmission scheduling subunit, a lightweight multimodal fusion subunit, a hydrological feedforward multi-constraint adaptive control subunit, and a command arbitration subunit. The edge autonomous computing module can independently complete lightweight fusion data, adaptive opening degree calculation, communication status discrimination, local data caching, transmission flow scheduling, and multi-source command arbitration based on its internal subunits. It can achieve local closed-loop autonomous control of the gate when communication is completely interrupted. The cloud-based global scheduling module is deployed on a remote data center server to receive lightweight fusion data processed and generated by the edge autonomous computing module. Based on this feature, it iterates the global control coefficient library, stores historical hydrological and equipment operating data of the entire basin, and carries a digital twin model to realize simulation and pre-drilling of multi-gate coordinated scheduling. A three-layer redundant communication link is established between the edge autonomous computing module and the cloud global scheduling module. The three-layer redundant communication link includes an industrial Ethernet main link, a 5G wireless backup link, and a Beidou short message emergency link.

[0007] The gate execution drive module is used to receive compliant opening instructions output by the edge autonomous computing module, drive the gate to rise and fall, and transmit the gate's operating status back to the edge autonomous computing module in real time, forming a complete control closed loop.

[0008] The multi-link communication driver subunit provides unified management of three-layer redundant communication links and has a built-in automatic link switching decision-making strategy.

[0009] The link monitoring subunit inside the edge autonomous computing module is equipped with a heartbeat verification mechanism to continuously determine the communication connectivity status between the edge autonomous computing module and the cloud global scheduling module.

[0010] The local cache subunit inside the edge autonomous computing module relies on the gateway's 16GB Flash storage to achieve data cyclic retention. It is configured with a 7-day cyclic overwrite storage strategy, sets differentiated retention periods for different types of data by dividing independent storage partitions, and sets the data upload priority for resuming transmission after communication recovery.

[0011] The hierarchical transmission scheduling subunit inside the edge autonomous computing module relies on three-layer redundant links to divide three levels of differentiated data upload priorities, and is equipped with fixed sampling and transmission cycles, thereby distinguishing the importance level of monitoring data, balancing the wireless transmission bandwidth load, preventing large-capacity video images from crowding out the emergency alarm transmission channel, and ensuring real-time reporting of flood control anomalies.

[0012] The lightweight multimodal fusion subunit inside the edge autonomous computing module receives all the raw monitoring data output by the sensing and acquisition module and executes a lightweight multimodal data correction and fusion strategy to eliminate the numerical distortion of river image acquisition caused by water mist and sediment obstruction and improve the reliability of multi-source monitoring data. The hydrological feedforward multi-constraint adaptive control subunit inside the edge autonomous computing module receives lightweight fusion data output by the lightweight multimodal fusion subunit and completes the optimal gate opening optimization strategy.

[0013] The instruction arbitration subunit inside the edge autonomous computing module can simultaneously receive three types of opening control instructions. These three types of instructions are respectively remote scheduling instructions issued by the cloud global scheduling module, local autonomous instructions generated by the edge autonomous computing module, and manual operation instructions input by the local touch screen of the gate execution drive module. The instruction arbitration subunit presets three levels of interlocking operation priorities to prevent frequent gate reciprocating actions caused by multi-channel instruction conflicts. The priorities, from highest to lowest, are: local touchscreen manual operation, edge autonomous computing module local autonomous automatic control, and cloud global scheduling module remote scheduling control. When multiple channels issue instructions simultaneously, high-priority instructions directly lock the output of low-priority instructions to avoid water hammer impact and mechanical wear caused by the repeated opening and closing of the gate.

[0014] The cloud-based global scheduling module includes a remote control server, a global data repository, a global parameter iteration subunit, a multi-gateway collaborative scheduling subunit, and a visual monitoring client. The global parameter iteration subunit is used to receive lightweight melting data uploaded from the edge, run high-precision fusion and optimization algorithms, iterate the global optimal control coefficient library and synchronously send it to the edge autonomous computing module. The global data repository is used to store historical hydrological and equipment operation data of the entire basin's gate stations, and supports data playback and export of scheduling reports. The multi-gate coordinated scheduling subunit is equipped with a GIS map and a 3D digital twin simulation module for the gates, enabling pre-simulation of scheduling plans, joint water control between upstream and downstream gates, and coordinated regulation of graded flood discharge.

[0015] The gate actuator module includes a hydraulic hoist, an electric actuator, and a local operation touch screen; The on-site touchscreen is used for manual issuance of gate opening commands by on-site personnel. The hydraulic gate hoist and electric actuator receive the opening command output by the edge autonomous computing module to drive the gate to rise and fall, and transmit the gate opening, cylinder oil pressure and motor current operating data back to the edge autonomous computing module in real time to form a closed-loop control.

[0016] Due to the adoption of the above technical solution, the present invention has the following advantages: This invention employs a cloud-edge dual-layer computing architecture, relying on the edge autonomous computing module to achieve independent local control after network outages. Combined with three-layer redundant links, partitioned caching, and breakpoint resume functionality, it ensures complete retention of monitoring command data, improving operational reliability under gate communication failures. The edge autonomous computing module incorporates a hydraulic correction-multimodal data fusion algorithm to reduce acquisition errors caused by sediment and water mist; it also uses hydrological feedforward to predict inflow conditions, suppressing water level fluctuations and improving gate control response speed and water level control accuracy. The adaptive control algorithm sets triple operational constraints for mechanical, flood control, and energy consumption, filtering dangerous opening commands, avoiding gate overload, and ensuring equipment and flood season scheduling safety. The command arbitration unit sets a three-level command interlock priority to prevent multi-channel command conflicts and reduce wear on components caused by water hammer impact and frequent gate opening and closing. The cloud-based global scheduling module receives lightweight feature data from the edge autonomous computing module to update scheduling parameters, and uses digital twins to complete multi-gate scheduling simulation; the edge only uploads compressed features, reducing data traffic during field wireless transmission. Attached Figure Description

[0017] Figure 1 This is a block diagram of the overall architecture of the present invention; Figure 2 This is a data flow structure diagram of the internal sub-units of the edge autonomous computing module of the present invention; Figure 3 This is a schematic diagram of the three-layer redundant communication link, data storage partitioning, and hierarchical upload logic of the present invention; Figure 4 This is a simulation verification diagram of the gate control performance under the condition of communication interruption between the present invention and the traditional cloud-based centralized solution. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0019] Example 1 like Figure 1 The remote intelligent monitoring system for water conservancy gates shown includes a sensing and acquisition module, an edge autonomous computing module, a cloud-based global scheduling module, and a gate execution drive module. The sensing and acquisition module is used to collect three types of raw monitoring data: visual images of the river channel, hydrological time series, and gate mechanical operating conditions. All raw monitoring data are then transmitted to the edge autonomous computing module in real time via the bus.

[0020] Specifically, the perception and acquisition module includes visual acquisition equipment, hydrological acquisition equipment, and gate operation status acquisition equipment. The visual acquisition equipment includes visible light cameras and infrared thermal imaging cameras deployed at the same location to simultaneously acquire images of the river cross-section and output raw image frames with millisecond timestamps. The hydrological acquisition equipment includes radar level gauges, electromagnetic flow meters, and rainfall sensors to collect real-time data on upstream and downstream water levels, instantaneous flow through the gate, and cumulative rainfall. The gate operation status acquisition equipment includes a gate opening encoder, hydraulic pressure sensor, and motor current acquisition subunit to collect data on gate lifting stroke, hydraulic cylinder working pressure, and real-time load of the three-phase motor. The visual, hydrological, and operation status acquisition data are uniformly labeled with gate station coordinates and then output to the edge autonomous computing module via an RS485 serial bus.

[0021] The edge autonomous computing module is deployed inside the gate control cabinet, such as Figure 2 As shown, the edge autonomous computing module integrates a multi-link communication driving subunit, a link monitoring subunit, a local caching subunit, a hierarchical transmission scheduling subunit, a lightweight multimodal fusion subunit, a hydrological feedforward multi-constraint adaptive control subunit, and a command arbitration subunit. The edge autonomous computing module can independently complete lightweight fusion data, adaptive opening degree calculation, communication status discrimination, local data caching, transmission flow scheduling, and multi-source command arbitration based on its internal subunits, and achieve local closed-loop autonomous control of the gate station when communication is completely interrupted.

[0022] A three-layer redundant communication link is established between the edge autonomous computing module and the cloud global scheduling module. The three-layer redundant communication link includes an industrial Ethernet main link, a 5G wireless backup link, and a Beidou short message emergency link.

[0023] Specifically, such as Figure 3 As shown, the multi-link communication driver subunit uniformly manages three-layer redundant communication links and has a built-in automatic link switching determination strategy; the execution steps of the automatic link switching strategy are as follows: Step 1: Collect communication status indicators of three links every 1 second: for industrial Ethernet and 5G wireless links, collect real-time signal strength, transmission delay, and packet loss rate; for Beidou short message links, collect signal strength, single round-trip delay of message, and message transmission success rate.

[0024] Step 2, Hierarchical Switching Threshold and Degradation Logic: When the industrial Ethernet latency is <50ms and the packet loss rate is <1%, it serves as the primary priority transmission channel; when the Ethernet transmission latency continuously exceeds 200ms and there is no response after 3 consecutive heartbeat checks, it automatically degrades and switches to the 5G wireless backup link; if both the industrial Ethernet and 5G wireless links are lost, only the BeiDou short message emergency link is used for transmission; when the low-priority link is working, the status of the high-priority link is continuously monitored, and after the high-priority link indicators are continuously met, it automatically switches back to the higher-priority link step by step.

[0025] Step 3: The maximum length of a single BeiDou short message transmission is 128 bytes. It only carries simplified alarm messages such as exceeding water level limits, gate jamming, and motor overload, and does not transmit large-capacity data such as images, features, and logs. The BeiDou link does not perform bandwidth allocation scheduling. Heartbeat messages and alarm messages are sent at staggered intervals, strictly following the BeiDou terminal's 30-second minimum transmission interval specification to avoid exceeding the message transmission frequency limit and being rejected by the satellite. Emergency alarm information is only uploaded according to a fixed period.

[0026] Specifically, the link monitoring subunit is equipped with a heartbeat verification mechanism to continuously determine the communication connectivity status between the edge autonomous computing module and the cloud global scheduling module; The heartbeat verification mechanism is executed as follows: For the industrial Ethernet main link, the edge autonomous computing module sends a heartbeat data packet carrying the gate number to the cloud global scheduling module every 1 second. The cloud response timeout threshold is set to 2 seconds. If no response is received for 3 consecutive times, the main link is determined to be faulty, and the data transmission is automatically switched to the 5G wireless backup link.

[0027] For 5G wireless backup links, the heartbeat sending cycle is set to 5 seconds and the response timeout threshold is set to 10 seconds; if no response is received for 3 consecutive times, the backup link is determined to be faulty and automatically switches to the BeiDou short message emergency link.

[0028] For the BeiDou short message emergency link, the heartbeat sending cycle is set to 30 seconds and the response timeout threshold is set to 60 seconds. If no response is received for two consecutive times, the entire link communication is deemed to have failed, and the edge autonomous computing module automatically cuts off the command receiving channel of the cloud global scheduling module and switches to the local autonomous closed-loop operation mode.

[0029] When a link is restored, it is switched back step by step from high to low priority. After a high-priority link is detected to have normal heartbeat responses for multiple consecutive cycles, the corresponding link is automatically switched back. After the link is restored and stabilized, a cache index message is automatically generated to trigger the breakpoint resume transmission process of the partition cache data.

[0030] Specifically, such as Figure 3 As shown, the local caching subunit relies on the gateway's 16GB Flash storage to achieve data cyclic retention, configured with a 7-day cyclic overwrite storage strategy. Different retention periods are set for different types of data by dividing the storage into independent partitions, and the data upload priority for resuming interrupted transmissions after communication recovery is set. The system program and algorithm firmware reside in 4GB, with the remaining storage allocated to each partition. Quantitative calculations show that sensor data, lightweight fusion features, and operation logs occupy less than 6GB in total within the 7-day period, indicating sufficient storage margin. Original video is only cached briefly and does not occupy flash memory for extended periods.

[0031] The 7-day cyclic overwrite storage strategy specifies the storage partitioning and retention rules: 1) The storage space is divided into a real-time sensing partition, an image feature partition, and a gate adjustment log partition; 2) Real-time hydrological and operational condition sensor data are retained cyclically for 7 days, and automatically overwritten when the time window expires; original video images are cached for only 30 minutes, and lightweight fusion features are retained cyclically for 7 days; logs of each gate adjustment command and execution result are retained cyclically for 7 days. 3) When communication is restored and transmission resumes from a breakpoint, the data upload priority is: gate adjustment log > first-level alarm sensor data > lightweight image features.

[0032] Specifically, the hierarchical transmission scheduling subunit implements differentiated scheduling strategies for different links: For the industrial Ethernet main link and the 5G wireless backup link, a bandwidth priority allocation mechanism is adopted, setting three levels of transmission priority: Level 1 priority is alarm data such as exceeding water level limits, gate jamming, and motor overload, which is uploaded first after anomaly detection, and can temporarily occupy 80% of the communication bandwidth in Ethernet and 5G links; the Beidou short message link has inherent transmission latency, and only simplified alarm messages are reported periodically and off-peak, without bandwidth allocation scheduling; Level 2 priority is lightweight fusion features and operating condition data; Level 3 priority is historical logs and non-real-time statistical data. When high-priority data is transmitted, low-priority data automatically yields. For the Beidou short message emergency link, following the constraint that the length of a single message does not exceed 128 bytes, only Level 1 alarm data is transmitted, and simplified alarm messages are sent at fixed time intervals, without bandwidth allocation scheduling; other levels of data are temporarily stored in local cache and retransmitted through a breakpoint resumption mechanism after the high-priority link is restored.

[0033] Figure 3 In the system, the local cache subunit is divided into three independent storage partitions to store monitoring data, lightweight image fusion data, and gate operation logs according to the storage cycle. The link monitoring subunit detects the connectivity status of the three communication links in real time, and sets the data upload priority in conjunction with the hierarchical transmission scheduling subunit. Then, the multi-link communication drive subunit switches between industrial Ethernet, 5G wireless, and Beidou short message communication links to complete data transmission. When all link communication is interrupted, the system suspends data upload, and all monitoring logs are stored in the local cache partition. After communication is restored, the interrupted transmission is resumed according to the priority, realizing reliable data storage and transmission scheduling at the gate edge.

[0034] This invention employs a cloud-edge dual-layer computing architecture, deploying an edge autonomous computing module locally at the sluice gate, integrating data processing, control calculation, and instruction execution units throughout the entire process. When communication is normal, it coordinates with the cloud for scheduling; when communication is interrupted, it can independently complete sensing data correction, gate opening calculation, and closed-loop control, effectively improving the control failure problem caused by communication interruptions in a purely remote centralized management architecture, and enhancing the reliability of sluice gate operation during the flood season. It is equipped with a three-layer redundant communication link, partitioned cyclic caching, and a breakpoint resume mechanism, which can continuously retain monitoring and operational data during communication interruptions and retransmit data according to priority after communication is restored, ensuring the integrity of the data sequence throughout the entire cycle and providing data support for the traceability of water conservancy operation safety.

[0035] The edge autonomous computing module incorporates multimodal fusion processing logic with hydraulic correction, fusing image and hydrological time-series multi-source data. This reduces interference from water mist, water sediment, and other conditions on image monitoring, improving the stability and accuracy of multi-source sensing data. Simultaneously, it introduces hydrological feedforward information to predict water inflow trends, forming a feedforward and feedback control mechanism. Compared to traditional pure feedback control, this accelerates dispatch response speed, suppresses water level overshoot and oscillation, and improves the accuracy of water level control during the flood season.

[0036] Specifically, the lightweight multimodal fusion subunit receives all the raw monitoring data output by the sensing and acquisition module and executes the data correction and fusion strategy; The steps of the data correction and fusion strategy are as follows: Step 1: Perform grayscale compression and noise reduction preprocessing on the original visible light and infrared image frames, and extract image texture feature vectors. Hydrological time series feature vector By splicing together, a high-dimensional original feature set is constructed. Principal Component Analysis (PCA) algorithm is used to analyze the high-dimensional original feature set. Dimensionality reduction was performed, retaining the top k principal components with a cumulative contribution rate of over 85%, and removing redundant interference features caused by water surface reflection, silt, and water mist. The core feature vector after decorrelation was then extracted. ; Step 2: Based on the real-time upstream and downstream water level difference Define the hydraulic correction coefficient based on the bottom elevation Z of the gate. (Dimensionless, value range 0.8~1.2): ;in, To design the standard water level difference, The empirical fitting coefficients are typically taken as 0.1 to 0.3. After calculation, a threshold clamp is applied to λ, forcibly constraining it to the range of 0.8 to 1.2. Step 3: Extract the core feature vector With correction factor Multiply the data to output lightweight fused data after eliminating environmental interferences such as water level difference and water turbidity: In the formula: This is the final lightweight fused feature vector output.

[0037] Specifically, the hydrological feedforward multi-constraint adaptive control subunit receives lightweight fusion data output by the lightweight multimodal fusion subunit, completes the optimal gate opening optimization strategy, and adjusts the gate in advance in conjunction with the upstream water inflow prediction, while avoiding multiple problems such as equipment over-limit, ecological water shortage, and frequent opening and closing energy consumption. The steps of the optimal gate opening optimization strategy are as follows: Step 1, Hydrological Feedforward Expansion: When cloud communication is normal, the edge autonomous computing module periodically downloads short-term inflow forecast parameters issued by the cloud and stores them locally; it aggregates locally cached historical rainfall records, continuous upstream water level change data, and offline saved inflow forecast information to form feedforward reference data; when communication is completely interrupted and the local autonomous mode is entered, it directly retrieves the locally pre-stored forecast parameters to complete offline feedforward calculation; the feedforward reference data is integrated with lightweight fusion data to form a comprehensive analysis dataset.

[0038] Step 2: Construct multi-dimensional constraint control objectives: Let the gate opening be denoted as u, and the target water level as denoted as . The actual river level is denoted as H; Constructing a comprehensive cost function The control system finds the minimum value of the substitution function and outputs the current optimal gate opening: ; In the formula: J—Comprehensive cost function, the minimum value of the cost function corresponds to the optimal gate opening; H—Real-time actual water level of the river channel; —Target water level for river channel regulation; —Estimated operating current of the motor corresponding to the current gate opening; —Rated operating current of the motor; —River flood control warning water level; These are the weighted coefficients for the three constraints: water level deviation, motor overload, and flood control safety. —Energy consumption penalty item, used to suppress frequent opening and closing of gates.

[0039] Calculation logic for each item: Water level deviation term Used to ensure that the actual water level in the river channel is close to the target water level; Mechanical safety penalties When the estimated motor operating current corresponding to the opening command exceeds 90% of the rated current, the penalty value increases by the square of the over-limit ratio; when the current is within the safe range, this value is equal to 0. Flood control safety penalties Penalties are activated once the river level exceeds the warning level; the closer the water level is to the upper limit of the warning level, the higher the penalty weight. Energy consumption constraints : Count the number of times the gate opens and closes within the last 10 minutes, n; when A fixed penalty value of 0.5 is applied to suppress the frequent reciprocating oscillation of the gate and reduce wear and tear on the components.

[0040] For adjustable weighting coefficients, satisfying Based on the three operating conditions of flood discharge, daily irrigation, and equipment maintenance, refer to the table and change the parameters; During the flood season, the weight of flood control items is prioritized for flood discharge operations; the weight of water level deviation items is increased for routine irrigation operations; and the weight of mechanical safety penalties is increased for maintenance operations to lock the safe gate opening range. A gradient optimization algorithm is used to search for the optimal gate opening corresponding to the minimum cost function, and this information is then sent to the gate execution drive unit. The system incorporates three sets of weighting coefficients for flood season flood discharge, routine irrigation, and gate maintenance. Automatic switching logic: When the real-time water level of the river reaches the flood control warning level, the flood discharge weight configuration is automatically switched to increase the weight of the flood control safety penalty item; when the water level is stable in the normal water storage range for a long time and there is no rainfall warning, the daily irrigation weight configuration is activated to prioritize the accuracy of water level control; when the system receives a manual maintenance instruction and the gate is continuously stopped without opening or closing, the equipment maintenance weight configuration is automatically switched to increase the mechanical overload constraint weight.

[0041] The system uses the actual river water level approaching the desired dispatch level as the basic control objective, simultaneously superimposed with three types of constraint and penalty mechanisms: the first is mechanical safety constraints, limiting the gate's lifting and lowering stroke, maximum opening and closing speed, and ensuring the motor's operating current does not exceed the equipment's rated range; the second is flood control and ecological constraints, controlling the river's highest warning water level and ensuring the minimum ecological flow downstream; the third is energy consumption constraints, suppressing meaningless small-amplitude reciprocating movements of the gate. The system pre-stores three different constraint weight configuration schemes for flood discharge during the flood season, daily irrigation, and gate maintenance, which can automatically switch according to real-time water conditions; the adaptive control logic superimposes three constraints: mechanical limit, flood control safety, and operational energy consumption, and automatically filters out over-limit opening commands during the calculation process. This reduces the risk of overloaded operation of the hoist, while ensuring stable operation of the gate within the constraint boundaries, balancing equipment operation safety and flood control dispatch safety.

[0042] Step 3: Optimization of Forgetting Factor Nodes: Establish a candidate library of multiple gate control parameters and introduce the forgetting coefficient. (Values ​​range from 0.9 to 0.98) Historical parameters are updated by decay; a sliding time window is used to store historical operating condition samples. The weight of historical samples that are further away from the current time decays exponentially with the forgetting coefficient; after each round of gate opening calculation, the evaluation value of the parameter node is updated according to the calculation result of the current comprehensive cost function J, and failed parameter nodes that have been inactive for a long time, have excessively high cost function values, or have low reference value are periodically removed; the candidate library of control parameters is continuously updated iteratively using hydrological and image real-time monitoring samples, and finally, compliant gate opening values ​​that fully meet the multiple constraints of mechanical, flood control, and energy consumption are selected and output.

[0043] Specifically, the instruction arbitration subunit can simultaneously receive three types of opening control instructions. These three types of instructions are respectively remote scheduling instructions issued by the cloud global scheduling module, local autonomous instructions generated by the edge autonomous computing module, and manual operation instructions input by the local touch screen of the gate execution drive module. The instruction arbitration subunit presets three levels of interlocking operation priorities to prevent frequent reciprocating actions of the gate caused by conflicts between multiple channels of instructions. The priorities, from highest to lowest, are: local touchscreen manual operation, edge autonomous computing module local autonomous automatic control, and cloud global scheduling module remote scheduling control. When multiple channels issue instructions simultaneously, high-priority instructions directly lock the output of low-priority instructions to avoid water hammer impact and mechanical wear caused by the repeated opening and closing of the gate.

[0044] The cloud-based global scheduling module is deployed on a remote data center server to receive lightweight fusion data processed and generated by the edge autonomous computing module. Based on this feature, it iterates the global control coefficient library, stores historical hydrological and equipment operating data of the entire basin, and carries a digital twin model to realize multi-gate coordinated scheduling simulation and pre-play, assisting in the prediction of flood discharge risks and optimizing the water control effect of the entire basin.

[0045] The cloud-based global scheduling module consists of a remote control server, a global data repository, a global parameter iteration subunit, a multi-gate collaborative scheduling subunit, and a visual monitoring client. The global parameter iteration subunit receives lightweight feature data uploaded from the edge, runs a high-precision complete fusion and optimization algorithm, iterates the optimal control coefficient library across the entire domain, and synchronously distributes it to the edge gateways of each gate. The global data repository stores long-term hydrological and equipment operation data of gates throughout the entire basin, supports historical curve playback, and exports scheduling reports. The multi-gate collaborative scheduling subunit is equipped with a GIS layer and a 3D digital twin simulation model of the gates, which can simulate changes in river water level under different opening degrees, and realize joint graded flood discharge and water distribution scheduling of upstream and downstream gates.

[0046] The simulation uses basic data of the entire basin's rivers and sluice gates as the model base. After inputting the preset scheduling target, it virtually simulates the changes in river water level and flow under the linkage of multiple sluice gates, identifies scheduling risks in advance, and outputs safe and feasible water distribution and flood discharge scheduling schemes for the entire basin.

[0047] The gate execution drive module is used to receive compliant opening instructions output by the edge autonomous computing module, drive the gate to rise and fall, and transmit the gate's operating status back to the edge autonomous computing module in real time, forming a complete control closed loop.

[0048] The gate execution drive module includes a local operation touch screen, a hydraulic hoist, and an electric actuator. The local operation touch screen is used for on-site manual direct issuance of opening commands and is the highest priority operation source. The hydraulic hoist and electric actuator receive compliant opening commands after edge arbitration to drive the gate to rise and fall, and simultaneously transmit the gate opening, cylinder oil pressure, and motor real-time load back to the edge autonomous computing module to form a complete control closed loop.

[0049] Example 2 A remote intelligent monitoring method for water conservancy gates, implemented based on a remote intelligent monitoring system for water conservancy gates, includes the following steps: S1. The sensor acquisition module synchronously collects three types of raw monitoring data: river visual images, hydrological time series, and gate mechanical conditions. After adding time and gate location identifiers to the collected data, the data is uniformly transmitted to the edge autonomous computing module. S2. The multi-link communication driver subunit inside the edge autonomous computing module continuously monitors the connectivity status of the three-layer redundant communication links and uses the heartbeat verification mechanism to determine the effectiveness of communication. When the communication is normal, the system runs in cloud-edge collaborative mode. If the heartbeat verification fails continuously, it switches to local autonomous operation mode. S3, the lightweight multimodal fusion subunit inside the edge autonomous computing module performs noise reduction and hydraulic correction processing on the original monitoring data to generate lightweight fusion data after eliminating image interference; the hydrological feedforward multi-constraint adaptive control subunit combines hydrological prediction reference data and lightweight fusion data for comprehensive analysis, and solves the compliant gate opening based on multiple safety constraints; S4. The instruction arbitration subunit inside the edge autonomous computing module filters valid opening control instructions according to the preset operation priority and sends the valid instructions to the gate execution drive module. After the gate execution drive module completes the gate opening adjustment, it sends the equipment status back to the edge autonomous computing module in real time. S5. After the communication link is restored, the local cache subunit inside the edge autonomous computing module will transmit the various monitoring and adjustment data cached during the network outage to the cloud global scheduling module according to the preset data priority. The cloud global scheduling module updates the global control parameters and simultaneously sends them to the edge autonomous computing modules corresponding to each gate station. S6, the multi-gate-station collaborative scheduling subunit within the cloud-based global scheduling module relies on a digital twin simulation model to complete the whole-basin water resources scheduling simulation and the joint scheduling simulation of upstream and downstream gate stations.

[0050] It should be noted that the cloud-edge collaboration mode in step S2 refers to the normal operation mode in which the three-layer redundant communication links between the edge autonomous computing module and the cloud global scheduling module are normally connected, and the two can transmit data bidirectionally and synchronously adjust parameters.

[0051] In the cloud-edge collaborative mode, the edge autonomous computing module processes the sensing data on-site and uploads lightweight feature data to the cloud global scheduling module. The cloud global scheduling module combines the historical operating conditions of all gates and pumping stations in the entire basin with global data to iteratively optimize the global control parameter library and sends the updated parameters back to the edge autonomous computing module. The edge autonomous computing module integrates local real-time monitoring data with the global optimization parameters of the cloud global scheduling module to calculate the gate opening, taking into account both precise control of a single gate and joint scheduling of upstream and downstream areas of the entire basin. This is the system's daily standard scheduling mode.

[0052] The local autonomous mode refers to an emergency backup mode in which the edge autonomous computing module and the cloud global scheduling module completely lose communication (Ethernet and 5G are completely disconnected), and the edge autonomous computing module operates independently in a closed loop, detached from the cloud global scheduling module.

[0053] In the local autonomous mode, the edge autonomous computing module no longer receives scheduling instructions and global optimization parameters issued by the cloud-based global scheduling module; it relies solely on data collected in real time by local cameras and hydrological sensors at the gate station, and calls three sets of pre-stored operating condition weight parameter libraries to independently complete multimodal data correction, hydrological feedforward analysis, and multi-constraint opening calculation, and autonomously issues instructions to control the gate; all monitoring and adjustment data generated by network outages are stored in local cache, and uploaded in batches to the cloud-based global scheduling module to complete the ledger after communication is restored, specifically to deal with flood control emergency scenarios such as no network in mountainous areas and base station outages due to heavy rain.

[0054] Example 3: Simulation Verification of Gate Control Performance under Communication Interruption Conditions This embodiment builds a small-to-medium-sized river channel control gate model based on the MATLAB simulation platform. Under the extreme condition of complete communication link interruption during the flood season, the water level control effect of the traditional cloud-based centralized management and control scheme and the edge autonomous scheme of this invention are compared to verify the effectiveness of the invention.

[0055] Simulation object: Single-gate river control gate, normal water level (target water level) 2.0m, warning water level 2.8m, minimum water level 1.6m; equivalent channel storage area 25000㎡, initial steady-state gate opening 0.3, normal inflow 600m³ / h 3 / min.

[0056] Operating conditions: Total simulation duration 300 minutes, simulation step size 0.1 minutes; three key operating conditions are set: upstream water flow begins to rise at 80 minutes, all communication links between the sluice gate and the remote control center are interrupted at 100 minutes, and the flood peak arrives at the sluice gate at 130 minutes, with a peak flow of 1500 m³ / min. 3 / min (2.5 times the normal flow rate).

[0057] Traditional cloud-based centralized solution: The gate station is only equipped with a remote execution terminal, without local autonomous control logic. All gate control commands rely entirely on cloud-based distribution. After communication is interrupted, the on-site terminal cannot receive new commands, and the gate opening remains constant at the state before the network outage.

[0058] The edge autonomy scheme of this invention: an edge autonomous computing module is deployed on-site at the gate station, with built-in closed-loop control algorithm and gate mechanical rate constraint. The entire process of water level monitoring, control calculation and opening execution can be completed independently locally, without being affected by the status of the communication link; after communication is interrupted, it seamlessly switches to the local autonomous operation mode.

[0059] like Figure 4 As shown, the simulation results are divided into two sub-graphs: a water level response comparison curve and a gate opening change curve, as detailed below: During the normal and stable phase (0~80min): the water level curves of the two schemes completely overlapped and remained stable at the target water level of 2.0m without significant fluctuations, verifying that the initial steady state of the two systems was consistent and the water balance state was the same.

[0060] During the rising water flow phase (80-100 min): the upstream water flow gradually increases. In traditional schemes, the water level rises continuously as the inflow increases due to the fixed gate opening. The edge module of this invention senses the water level change trend in real time and actively and gradually opens the gate to release water and empty the reservoir, resulting in a water level rise rate that is significantly lower than that of traditional schemes.

[0061] Communication interruption and flood peak passage phase (after 100 minutes): After the complete communication interruption at 100 minutes, the traditional solution loses cloud control capability, the gate opening is locked, the inflow continues to increase, the water level shows an accelerated upward trend, and it exceeds the 2.8m warning level at about 140 minutes, and continues to rise thereafter, posing a risk of flooding.

[0062] The present invention allows for seamless switching to local autonomous mode, operating independently from the cloud, and dynamically adjusting the flood discharge opening based on real-time water levels. When the flood peak arrives at 130 minutes, the peak water level is approximately 2.35 meters, far below the warning level. After the flood peak, the water level gradually recedes and converges, stabilizing near the target water level after 250 minutes, and entering steady-state operation.

[0063] The horizontal axis of the gate opening change curve represents the simulation time, and the vertical axis represents the relative gate opening (0 for fully closed, 1 for fully open).

[0064] During the normal and stable phase (0~80min): the gate opening of both schemes is stable at the initial equilibrium opening of 0.3, with slight fluctuations in the curve, which matches the sensor noise characteristics and conforms to the actual engineering operation state.

[0065] During the change of operating conditions (after 80 minutes): the opening curve of the traditional solution remains a horizontal straight line throughout the entire process, with no change before and after the communication interruption. This intuitively demonstrates the technical defects of the pure cloud architecture, which has no local control capability and whose opening is locked after the network is disconnected.

[0066] The opening of the gate in this invention increases smoothly from the 80th minute onwards as the water level rises. Constrained by the mechanical speed of the gate hoist, it exhibits a continuous and gradual upward trend without sudden changes or high-frequency oscillations. After the flood peak arrives, it continuously makes minor adjustments to adapt to the flow rate changes, eventually stabilizing at a balanced opening of approximately 0.75, which corresponds one-to-one with the steady-state water level results. The adjustment process is stable and reliable, conforming to the mechanical operation law of the gate equipment.

[0067] This invention integrates an edge autonomous computing module with complete local intelligent control capabilities, constructs a cloud-edge dual-layer collaborative architecture, and breaks through the shortcomings of existing technologies that rely entirely on the cloud for computation and cause gates to lose control after communication interruption. It integrates a lightweight multimodal fusion unit, a hydrological feedforward multi-constraint adaptive control unit, and a communication cache and command arbitration unit locally at the edge, so as to achieve independent multi-source data processing, adaptive opening degree solution, and gate safety closed-loop control without the need for the cloud in the case of network outage.

[0068] The parts not detailed above are existing technologies and therefore have not been described in detail.

Claims

1. A remote intelligent monitoring system for hydraulic gates, characterized in that, It includes a sensing and acquisition module, an edge autonomous computing module, a cloud-based global scheduling module, and a gate execution driver module; The sensing and acquisition module is used to collect three types of raw monitoring data: river visual images, hydrological time series, and gate mechanical operating conditions, and transmits all raw monitoring data to the edge autonomous computing module in real time via bus. The edge autonomous computing module is deployed inside the gate control cabinet and integrates a multi-link communication driver subunit, a link monitoring subunit, a local cache subunit, a hierarchical transmission scheduling subunit, a lightweight multimodal fusion subunit, a hydrological feedforward multi-constraint adaptive control subunit, and a command arbitration subunit. The edge autonomous computing module can independently complete lightweight fusion data, adaptive opening degree calculation, communication status discrimination, local data caching, transmission flow scheduling, and multi-source command arbitration based on its internal subunits, and achieve local closed-loop autonomous control of the gate when communication is completely interrupted. The cloud-based global scheduling module is deployed on a remote computer room server. It is used to receive lightweight fusion data generated by the edge autonomous computing module, and relies on this feature to iterate the global control coefficient library, store historical hydrological and equipment operating data of the entire basin, and carry a digital twin model to realize multi-gate station collaborative scheduling simulation and pre-drilling. The edge autonomous computing module and the cloud global scheduling module establish a three-layer redundant communication link, which includes an industrial Ethernet main link, a 5G wireless backup link, and a Beidou short message emergency link. The gate execution drive module is used to receive the compliant opening command output by the edge autonomous computing module, drive the gate to rise and fall, and transmit the gate's operating status back to the edge autonomous computing module in real time, forming a complete control closed loop.

2. The remote intelligent monitoring system for hydraulic gates according to claim 1, characterized in that, The multi-link communication driver subunit inside the edge autonomous computing module uniformly manages the three-layer redundant communication links and has a built-in automatic link switching determination strategy. The automatic link handover strategy execution steps are as follows: Step 1: Collect communication status indicators for three links every 1 second: for industrial Ethernet and 5G wireless links, collect real-time signal strength, transmission delay, and packet loss rate; for Beidou short message links, collect signal strength, single round-trip delay of messages, and message transmission success rate. Step 2: Hierarchical Switching Threshold and Degradation Logic: When the industrial Ethernet latency is <50ms and the packet loss rate is <1%, it serves as the primary priority transmission channel; when the Ethernet transmission latency continuously exceeds 200ms, the link is physically disconnected, or there is no response after three consecutive heartbeat checks, it automatically degrades to the 5G wireless backup link; if both the industrial Ethernet and 5G wireless links are lost simultaneously, only the BeiDou short message emergency link is used for transmission; when a low-priority link is working, the status of the high-priority link is continuously monitored, and after the high-priority link indicators are continuously met, it automatically switches back to a higher-priority link step by step. Step 3: The maximum length of a single BeiDou short message transmission is 128 bytes. It only carries simplified alarm messages such as exceeding water level limits, gate jamming, and motor overload, and does not transmit large-capacity data such as images, features, and logs. The BeiDou link does not perform bandwidth allocation scheduling and only uploads emergency alarm information according to a fixed period.

3. The remote intelligent monitoring system for hydraulic gates according to claim 2, characterized in that, The link monitoring subunit inside the edge autonomous computing module is equipped with a heartbeat verification mechanism to continuously determine the communication connectivity status between the edge autonomous computing module and the cloud global scheduling module. The execution steps of the heartbeat verification mechanism are as follows: For the industrial Ethernet main link, the edge autonomous computing module sends a heartbeat data packet carrying the gate number to the cloud global scheduling module every 1 second, and the cloud response timeout threshold is set to 2 seconds; If no response is received for three consecutive times, the main link is determined to be faulty, and the data transmission is automatically switched to the 5G wireless backup link. For 5G wireless backup links, the heartbeat sending cycle is set to 5 seconds, and the response timeout threshold is set to 10 seconds. If no response is received for three consecutive times, the backup link is determined to be faulty, and the system will automatically switch to the BeiDou short message emergency link. For the BeiDou short message emergency link, the heartbeat sending cycle is set to 30 seconds, and the response timeout threshold is set to 60 seconds. If no response is received twice in a row, the end-to-end communication is deemed to have failed. The edge autonomous computing module automatically cuts off the command receiving channel of the cloud global scheduling module and switches to the local autonomous closed-loop operation mode. When a link is restored, it is switched back step by step from high to low priority. After a high-priority link is detected to have normal heartbeat responses for multiple consecutive cycles, it is automatically switched back to the corresponding link. Once the link stabilizes, a cache index message is automatically generated, triggering the breakpoint resume transmission process for the partition cache data.

4. The remote intelligent monitoring system for hydraulic gates according to claim 3, characterized in that, The local cache subunit inside the edge autonomous computing module relies on the gateway's 16GB Flash storage to achieve data cyclic retention. It is configured with a 7-day cyclic overwrite storage strategy, sets differentiated retention periods for different types of data by dividing independent storage partitions, and sets the data upload priority for communication recovery interruption resume transmission.

5. The remote intelligent monitoring system for hydraulic gates according to claim 4, characterized in that, The hierarchical transmission scheduling subunit inside the edge autonomous computing module relies on three-layer redundant links to divide three levels of differentiated data upload priorities, and is equipped with fixed sampling and transmission cycles, thereby distinguishing the importance level of monitoring data and balancing the wireless transmission bandwidth load.

6. The remote intelligent monitoring system for hydraulic gates according to claim 2, characterized in that, The lightweight multimodal fusion subunit inside the edge autonomous computing module receives all the raw monitoring data output by the sensing and acquisition module and executes the data correction and fusion strategy. The steps of the data correction and fusion strategy are as follows: Step 1: Perform grayscale compression and noise reduction preprocessing on the original visible light and infrared image frames, extract image texture feature vectors and concatenate them with hydrological time series feature vectors to form a high-dimensional original feature set; use principal component analysis (PCA) algorithm to reduce the dimensionality of the high-dimensional original feature set, retain the top k principal components with a cumulative contribution rate of more than 85%, remove redundant interference features caused by water surface reflection, silt and water mist, and extract the core feature vector after decorrelation. Step 2: Define the hydraulic correction coefficient based on the real-time upstream and downstream water level difference and the bottom elevation of the gate; Step 3: Multiply the core feature vector with the correction coefficient to output lightweight fusion data after eliminating environmental interferences such as water level difference and water turbidity.

7. The remote intelligent monitoring system for hydraulic gates according to claim 6, characterized in that, The hydrological feedforward multi-constraint adaptive control subunit inside the edge autonomous computing module receives the lightweight fusion data output by the lightweight multimodal fusion subunit and completes the optimal gate opening optimization strategy. The steps of the optimal gate opening optimization strategy are as follows: Step 1: Hydrological Feedforward Expansion: The feedforward reference data is composed of locally cached historical rainfall records, continuous upstream water level change data, and short-term inflow forecast information sent from the cloud and stored offline. When communication is interrupted and the local autonomous mode is entered, the locally pre-stored forecast data is directly retrieved. The feedforward reference data is integrated with the lightweight fusion data to form a comprehensive analysis dataset. Step 2: Constructing Multi-Dimensional Constraint Control Objectives: Based on the actual river water level approaching the desired dispatch water level as the fundamental control objective, three types of constraint penalty mechanisms are simultaneously superimposed: The first type is mechanical safety constraints, limiting the gate lifting and lowering stroke, maximum opening and closing speed, and motor operating current to not exceed the rated range of the equipment; the second type is flood control and ecological constraints, controlling the highest warning water level of the river and ensuring the minimum ecological flow downstream; the third type is energy consumption constraints, suppressing meaningless small reciprocating movements of the gate; the system pre-stores three sets of constraint weight configuration schemes for flood discharge during the flood season, daily irrigation, and gate maintenance, and automatically switches according to real-time water level and manual maintenance instructions; Step 3, Forgotten Node Optimization: Build multiple sets of gate control parameter candidate libraries, set forgetting update rules for each parameter node, comprehensively evaluate the overall compliance of multi-dimensional constraint targets, eliminate control parameters that trigger high penalties or pose safety risks, and output compliant gate opening values ​​that fully meet all constraint conditions.

8. The remote intelligent monitoring system for hydraulic gates according to claim 7, characterized in that, The instruction arbitration subunit inside the edge autonomous computing module can simultaneously receive three types of opening control instructions. These three types of instructions are respectively remote scheduling instructions issued by the cloud global scheduling module, local autonomous instructions generated by the edge autonomous computing module, and manual operation instructions input by the local touch screen of the gate execution drive module. The instruction arbitration subunit presets three levels of interlocking operation priorities to prevent frequent reciprocating actions of the gate caused by multi-channel instruction conflicts. The priorities, from highest to lowest, are: local touchscreen manual operation, edge autonomous computing module local autonomous automatic control, and cloud global scheduling module remote scheduling control. When multiple channels issue instructions simultaneously, high-priority instructions directly lock the output of low-priority instructions to avoid water hammer impact and mechanical wear caused by the repeated opening and closing of the gate.

9. The remote intelligent monitoring system for hydraulic gates according to claim 1, characterized in that, The cloud-based global scheduling module includes a remote control server, a global data repository, a global parameter iteration subunit, a multi-gateway collaborative scheduling subunit, and a visual monitoring client. The global parameter iteration subunit is used to receive lightweight fusion data uploaded by the edge autonomous computing module, run high-precision fusion and optimization algorithms, iterate the global optimal control coefficient library and synchronously send it to the edge autonomous computing module. The global data repository is used to store historical hydrological and equipment operation data of the entire basin's gate stations, and supports data playback and export of scheduling reports. The multi-gate coordinated scheduling subunit is equipped with a GIS map and a 3D digital twin simulation module for the gates, enabling pre-simulation of scheduling schemes, joint water control between upstream and downstream gates, and coordinated regulation of graded flood discharge.

Images