A dam closed-loop intelligent regulation and control method and system based on multiple agents

Abstract

The application discloses a kind of dam closed loop intelligent regulation and control method and system based on multiple agents, comprising: constructing the global multimodal perception system of multiple mobile agents and various fixed sensor nodes cooperation, and collecting the multi-source data of dam body;Data fusion, state identification and heat preservation thermal parameter inversion are carried out using edge computing;Real-time mapping of state to digital twin model for risk deduction;Based on real-time state and predicted trend, generate water regulation and control, heat preservation repair, collaborative inspection and early warning instructions autonomously;Through the execution layer, complete on-site regulation and feedback;And based on the whole process data, continuously optimize model and strategy.The application realizes the whole life cycle integration of dam from construction period temperature control crack prevention, winter heat preservation to operation period health monitoring, closed loop intelligent management and control, significantly improves the perception ability of safety state, the accuracy of risk analysis, the initiative of regulation and control measures and the self-adaptive level of system.

Description

Technical Field

[0001] This invention relates to the fields of intelligent dams, hydropower project safety monitoring, digital twins and artificial intelligence, and in particular to a closed-loop intelligent control method and system for dams based on multi-agent systems. Background Technology

[0002] With the increasing demands for intelligent dam construction and safe operation of hydropower projects, temperature control and crack prevention during dam construction, winter insulation, structural health monitoring during operation, and defect handling during maintenance have become crucial aspects of hydraulic structure safety management. Concrete dams are susceptible to hydration heat, sudden changes in ambient temperature, cold waves and strong winds, water level fluctuations, material aging, and changes in seepage conditions during construction and long-term service, leading to defects such as temperature cracks, leakage, abnormal deformation, and surface erosion. Among these, improper control of the temperature difference and temperature drop rate between the inside and outside of the concrete during construction is a significant cause of temperature cracks; the coupled effects of multiple factors such as water pressure, temperature, aging, and seepage during operation can affect the long-term safety of the dam.

[0003] Existing dam monitoring technologies typically rely on fixed-point sensors, manual inspections, drone aerial photography, and remote manual analysis. While these methods can acquire localized monitoring data, they still have significant shortcomings: First, the sensing range is limited, making it difficult to cover the dam surface, interior, foundation, underwater areas, and surrounding environment. Second, there is a lack of unified integration among images, infrared thermal maps, 3D point clouds, internal temperature data, meteorological data, and water flow parameters, easily leading to data silos. Third, the insulation layer coverage and thermal performance during construction rely heavily on manual judgment, making it difficult to promptly and quantitatively identify insulation deficiencies, damage, warping, abnormal equivalent heat transfer coefficients, and insufficient insulation thickness. Fourth, temperature control, insulation repair, defect verification, and risk warning depend on human experience, resulting in delayed responses. Fifth, the systems are fragmented across the construction, impoundment, operation, and maintenance phases, lacking a unified data, model, and strategy system spanning the entire lifecycle. Sixth, existing systems lack multi-agent collaborative operation, closed-loop execution, and continuous learning capabilities, making it difficult to support the upgrade of intelligent dams from "passive monitoring" to "active control."

[0004] Therefore, there is an urgent need for an intelligent system and method that can integrate autonomous mobile intelligent agents, fixed sensing nodes, edge recognition, thermal insulation inversion, digital twins, finite element behavior prediction, autonomous decision-making, control execution, and continuous learning to achieve continuous sensing, quantitative evaluation, risk simulation, proactive control, and adaptive optimization throughout the entire life cycle of a dam. Summary of the Invention

[0005] This invention addresses the shortcomings of existing technologies by providing a closed-loop intelligent control method and system for dams based on multiple agents.

[0006] To achieve the above-mentioned objectives, the technical solution adopted by the present invention is as follows:

[0007] A closed-loop intelligent control method for dams based on multi-agent systems, characterized by comprising:

[0008] Construct a full-domain multimodal perception system covering the dam surface, dam interior, dam foundation, underwater area and surrounding environment, and collect multi-source monitoring data through autonomous mobile intelligent agents and fixed sensing nodes in collaboration;

[0009] The multi-source monitoring data is synchronized in time, registered in space, fused in data and identified in state to obtain the insulation coverage state, apparent defect state, temperature field state, structural response state and environmental boundary state of the dam body;

[0010] The identified dam status is mapped in real time to a pre-built digital twin model of the dam, and status synchronization, trend inference and risk assessment are performed in the digital twin model;

[0011] Based on the real-time status, predicted trends, finite element analysis results, historical operating data, and preset safety thresholds reflected by the digital twin model, intelligent water supply control commands, insulation repair commands, multi-agent collaborative inspection commands, and risk warning commands are generated autonomously.

[0012] The instructions are sent to the control and execution layer to complete cooling water flow control, insulation repair work, intelligent agent path scheduling, on-site early warning and re-inspection feedback;

[0013] Based on monitoring feedback data, control implementation effects, repair and re-inspection results, and historical operation data, the insulation identification model, defect identification model, insulation thermal inversion model, temperature field reconstruction model, behavior prediction model, risk assessment model, and collaborative decision-making strategy are iteratively optimized.

[0014] Furthermore, the construction of a comprehensive multimodal perception system covering the dam surface, dam interior, dam foundation, underwater area, and surrounding environment includes:

[0015] Deploy at least one of the following: wall-climbing robots, drones, tracked robots, cable robots, and underwater robots to collect visible light images, infrared thermal images, hyperspectral data, three-dimensional point cloud data, and surface defect data of the dam surface;

[0016] Temperature sensors, strain sensors, seepage pressure sensors, uplift pressure sensors, crack gauges, smart patches, and water flow monitoring nodes are embedded or attached inside and on the surface of the dam body to collect data on temperature, strain, seepage pressure, uplift pressure, crack opening, water flow rate, and water temperature inside the dam body.

[0017] Connect to meteorological monitoring equipment to obtain data on ambient temperature, humidity, wind speed, wind direction, solar radiation, rainfall, and snowfall.

[0018] Furthermore, the state identification of multi-source monitoring data includes:

[0019] The coverage status of the insulation layer is identified based on the target detection model or semantic segmentation model. The coverage status includes normal coverage, missing, damaged, curled edges, hollow, snow cover, shadow interference, and construction cover.

[0020] Based on target detection models, instance segmentation models, or image recognition models, the cracks, seepage, erosion, misalignment, voids, and surface damage on the dam surface are located, quantified, graded, and their orientation is extracted.

[0021] The dam's temperature field was reconstructed based on infrared thermal images, internal temperature, contact surface temperature, meteorological boundaries, and water flow parameters. The internal and external temperature differences, temperature drop rate, temperature gradient, and time-varying temperature characteristics were also calculated.

[0022] Furthermore, the method also includes a step of inverting the equivalent heat release coefficient and equivalent thickness of the insulation:

[0023] Based on the surface temperature of the insulation layer, the temperature of the contact surface between the concrete and the insulation layer, the air temperature, wind speed, humidity and solar radiation data, the equivalent heat release coefficient and the equivalent insulation thickness are calculated.

[0024] Alternatively, based on the internal temperature at a predetermined depth within the dam body, infrared surface temperature, meteorological data, and material parameters, the equivalent heat release coefficient and equivalent insulation thickness can be indirectly calculated.

[0025] The inversion results based on the contact surface temperature are cross-checked with the inversion results based on the internal temperature. When the relative error between the two exceeds a preset threshold, a prompt is made to perform data verification.

[0026] Furthermore, the real-time mapping of the dam's state to a pre-constructed digital twin model of the dam includes:

[0027] The insulation coverage status, defect location, defect size, temperature field distribution, structural deformation, seepage status, water flow parameters, and intelligent agent location information are written into the digital twin model according to spatiotemporal coordinates.

[0028] The status of the digital twin is dynamically updated based on the monitoring time sequence, forming a traceable, replayable, and predictable digital twin entity;

[0029] Temperature field simulation, structural response simulation, risk area location, and trend prediction are performed based on the updated digital twin model.

[0030] Furthermore, the autonomously generated intelligent water flow control commands, insulation repair commands, multi-agent collaborative inspection commands, and risk warning commands include:

[0031] When the internal and external temperature difference, temperature drop rate, maximum temperature, temperature gradient, or predicted temperature change trend exceeds the preset threshold, control commands are generated for cooling water flow rate, inlet water temperature, valve opening, water flow duration, and pipeline start-up and shutdown sequence.

[0032] When the insulation layer is found to be missing, damaged, warped, not tightly overlapped, insufficiently covered, or with insufficient thermal insulation performance, task instructions are generated for the repair area, repair material, repair method, work path, and re-inspection sequence.

[0033] Based on risk level, task priority, agent location, remaining power, payload capacity, weather conditions, and construction area limitations, dynamically plan the verification tasks using drones, wall-climbing robots, ground robots, underwater robots, or humans.

[0034] Based on the level of risk, tiered early warning instructions are generated, along with corresponding handling suggestions or manual intervention prompts.

[0035] Furthermore, the iterative optimization includes:

[0036] The raw data from multi-source monitoring, preprocessed data, insulation identification results, infrared temperature field, insulation equivalent heat release coefficient inversion results, equivalent thickness inversion results, temperature field reconstruction results, digital twin status, water flow control records, insulation repair records, early warning records, execution feedback, meteorological data, and dam design data are classified and stored.

[0037] Update the insulation recognition model, defect recognition model, or image segmentation model based on the newly added labeled samples;

[0038] Based on the measured temperature, meteorological conditions, and the effect after repair, the wind speed correction factor, humidity correction factor, material thermal conductivity, and solar radiation absorption factor are corrected.

[0039] Based on the newly added monitoring data during the operation period, retrain or fine-tune the finite element-HST hybrid model, LSTM model, Informer model or PatchTST model used for dam behavior prediction.

[0040] Update control limits, low-probability early warning indicators, or confidence interval early warning indicators based on changes in the residual distribution of the prediction model.

[0041] This invention also discloses a multi-agent-based intelligent sensing and collaborative control system for the entire life cycle of a dam, used to implement the aforementioned multi-agent-based closed-loop intelligent control method for dams, comprising:

[0042] The multimodal sensing layer is used to collect multi-source monitoring data on the dam surface, dam interior, dam foundation area, underwater area, water supply system and surrounding environment through autonomous mobile intelligent agents and fixed sensing nodes;

[0043] The edge recognition and understanding layer is used to perform time synchronization, spatial registration, data fusion, insulation recognition, defect recognition, temperature field reconstruction, and insulation thermal parameter inversion on the multi-source monitoring data.

[0044] The digital twin platform is used to map the identified and inverted dam status information to the dam digital twin model, and to complete status updates, trend projections, risk location, and historical playback.

[0045] The autonomous decision-making and collaboration layer is used to generate instructions for water flow control, insulation repair, collaborative inspection, and risk warning based on real-time status, predicted trends, and safety thresholds.

[0046] The control and execution layer is used to perform cooling water flow control, intelligent agent scheduling, insulation repair, on-site early warning, and re-inspection feedback.

[0047] The continuous learning and knowledge evolution layer is used to store monitoring data, identification results, inversion results, prediction results, decision instructions, and execution feedback, and is used to train, update models, and optimize decision strategies.

[0048] Furthermore, the autonomous decision-making and collaboration layer includes at least one of the following: monitoring data management intelligent agent, dam behavior prediction intelligent agent, anomaly detection and early warning intelligent agent, engineering knowledge retrieval intelligent agent, temperature control and regulation intelligent agent, insulation inspection and repair intelligent agent, and insulation thermal performance evaluation intelligent agent.

[0049] Furthermore, the system supports integrated management and control during the construction phase, including pouring and curing, winter insulation, water storage and commissioning, operation monitoring, and maintenance and repair, through parameter configuration, model switching, and strategy optimization.

[0050] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor, when executing the computer program, implements a multi-agent-based closed-loop intelligent control method for dams.

[0051] A non-transitory computer-readable storage medium storing a computer program thereon, characterized in that the computer program, when executed by a processor, implements a multi-agent-based closed-loop intelligent control method for dams.

[0052] Compared with the prior art, the advantages of the present invention are as follows:

[0053] 1. Achieve full-area, multi-stage, and continuous perception of the dam: Through the collaborative work of various autonomous mobile intelligent agents and fixed sensing nodes, multi-dimensional and blind-spot-free perception of the dam surface, interior, foundation, underwater environment, and surrounding environment is achieved, significantly improving the completeness and continuity of status acquisition.

[0054] 2. Improve the ability to control temperature and prevent cracking during construction and to maintain insulation during winter: The system can not only identify whether the insulation layer is covered, but also quantitatively evaluate whether the insulation effect meets the requirements by inverting the insulation equivalent heat release coefficient and equivalent thickness, thereby improving the insulation management from qualitative judgment to quantitative and refined level.

[0055] 3. Achieve multi-source heterogeneous data fusion and unified representation of dam status: Through the edge computing layer, spatiotemporal registration and fusion analysis of multi-source data such as images, temperature, point clouds, sensors, meteorology, and water flow are performed and mapped to the digital twin platform, which solves the problem of data fragmentation in traditional monitoring and provides a unified and reliable data foundation for risk assessment.

[0056] 4. Improve the quantification and reliability of thermal insulation performance evaluation: Innovatively combine infrared, internal, contact surface temperature measurement and environmental data, and use two inversion paths to cross-calibrate and quantitatively calculate key insulation parameters, so that the identification of insulation defects, repair judgment and effect verification are transformed from relying on manual experience to objective evaluation based on data.

[0057] 5. Achieve deep synergy between digital twin and finite element behavior prediction: Map the real-time perceived state to the digital twin and couple finite element analysis, FEM-HST hybrid model, LSTM and other prediction models to perform state extrapolation and risk prediction, which greatly improves the intelligence level of structural health monitoring and risk warning during operation.

[0058] 6. Achieve closed-loop linkage of water flow regulation, insulation repair, collaborative inspection and risk warning: Based on real-time status and prediction results, autonomously generate regulation and disposal instructions, and drive field equipment to complete operation and feedback through the execution layer, forming a complete closed loop of "perception-identification-decision-execution-feedback-optimization", reducing the lag of manual intervention.

[0059] 7. Improve the efficiency of collaborative operation of multiple agents in complex environments: Through the task planning and collaborative scheduling module, the tasks of various agents such as drones and wall-climbing robots can be dynamically and efficiently arranged according to multiple constraints such as risk, location, power, and capability, reducing path conflicts and repeated inspections, and improving overall handling efficiency.

[0060] 8. Enhance proactive prevention and control capabilities under extreme weather and complex working conditions: The system can combine forecasts or real-time weather conditions such as cold waves, strong winds, and strong radiation to comprehensively assess the changing trend of thermal insulation performance, and generate preventive reinforcement, inspection, or control tasks in advance, realizing the transformation from post-event handling to pre-event prediction and proactive prevention and control.

[0061] 9. Achieve integrated, full lifecycle management of construction and operation phases: Through parameter configuration, model switching, and strategy optimization, the same system can adapt to the core needs of different stages such as temperature control, winter insulation, water storage and commissioning, operation monitoring, and maintenance during construction, avoiding system fragmentation and improving the reusability of equipment and data and management continuity.

[0062] 10. Possesses continuous model self-learning and policy adaptive optimization capabilities: By accumulating operational data, samples, and feedback results through continuous learning layers, it can continuously iterate and optimize the identification model, inversion model, prediction model, early warning threshold, and decision-making strategy, enabling the system to autonomously evolve with changes in environment, materials, and operating conditions, maintaining long-term adaptability and accuracy. Attached Figure Description

[0063] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0064] Figure 1 This is a schematic diagram of the overall system structure in an embodiment of the present invention;

[0065] Figure 2 This is a schematic diagram of the on-site deployment of multiple intelligent agents and fixed sensing nodes in an embodiment of the present invention;

[0066] Figure 3 This is a schematic diagram of the insulation equivalent heat release coefficient and thickness inversion process in an embodiment of the present invention;

[0067] Figure 4 This is a schematic diagram of YOLO thermal insulation recognition and infrared temperature field registration in an embodiment of the present invention;

[0068] Figure 5 This is a schematic diagram of the digital twin and state mapping of the dam in an embodiment of the present invention;

[0069] Figure 6 This is a schematic diagram of digital twin and finite element behavior prediction in an embodiment of the present invention;

[0070] Figure 7 This is a schematic diagram of autonomous decision-making and multi-agent collaborative scheduling in an embodiment of the present invention;

[0071] Figure 8 This is a schematic diagram of the intelligent water flow control closed loop in an embodiment of the present invention;

[0072] Figure 9 This is a schematic diagram of the closed-loop execution and re-inspection of thermal insulation repair in an embodiment of the present invention;

[0073] Figure 10 This is a schematic diagram of the continuous learning and iterative optimization mechanism in an embodiment of the present invention;

[0074] Figure 11 This is a schematic diagram of the intelligent sensing and collaborative control method for the entire life cycle of a dam in an embodiment of the present invention;

[0075] Figure 12 This is a schematic diagram of the closed-loop operation process of the system in an embodiment of the present invention;

[0076] Figure 13 This is a schematic diagram of multimodal sensing data fusion and temperature field reconstruction in an embodiment of the present invention;

[0077] Figure 14 This is a schematic diagram of the electronic device structure in an embodiment of the present invention. Detailed Implementation

[0078] 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, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0079] I. Overall System Deployment Concept

[0080] The system described in this invention adopts a layered deployment architecture of "global perception, edge recognition, twin mapping, collaborative decision-making, control and execution, and continuous learning." Its overall structure and the data flow relationships between each layer are as follows: Figure 1 As shown. The system is connected via a hybrid wired and wireless network (such as 5G, Wi-Fi, and fiber optics for large data transmission, and LoRa and NB-IoT for low-power sensing data), covering the entire life cycle of the dam, including pouring and curing during construction, winter insulation, water storage and commissioning, operation monitoring and maintenance, achieving an upgrade from "passive monitoring" to "active control".

[0081] 1. Deployment of the multimodal sensing layer

[0082] This layer aims to achieve blind-spot-free, continuous perception of the dam's surface, interior, foundation, underwater areas, and surrounding environment. It employs a collaborative and adaptive deployment model combining autonomous mobile intelligent agents and fixed sensing nodes. The on-site deployment concept is as follows: Figure 2 As shown.

[0083] Autonomous mobile intelligent agents:

[0084] Dam-climbing robot: Equipped with permanent magnets or vacuum adsorption devices, it can move vertically or inclined along the upstream and downstream faces of the dam. It integrates a 12-48 megapixel visible light camera, an infrared thermal imager with a temperature measurement range of -20℃ to 150℃, a LiDAR with centimeter-level ranging accuracy, an IMU, and an edge computing unit. The inspection speed is adjustable between 0.05m / s and 0.30m / s, with a path overlap rate set to 10%~20%, used for high-definition acquisition of images of the insulation layer, surface defects, infrared temperature fields, and high-precision 3D point clouds.

[0085] Autonomous inspection drones: used for rapid surveys of dam tops, dam shoulders, high-altitude facades, and areas with complex terrain. Flying at altitudes of 30m to 150m above the dam surface, they acquire ground-resolution images of 1cm to 5cm. The forward and lateral overlap rates are set to 60%–80% and 40%–70%, respectively, to support 3D modeling.

[0086] Other intelligent agents include tracked ground robots used for dam foundations and corridors, cable robots used for inspecting specific areas of the dam surface, and underwater robots used for inspecting underwater structures, which together constitute a three-dimensional mobile perception network.

[0087] Fixed sensing nodes:

[0088] Internal sensor network: Temperature sensors (2-10m spacing in key areas, 5-20m spacing in general areas), strain sensors, and seepage pressure sensors are embedded in the concrete; crack gauges and displacement gauges are installed at structural joints and critical sections.

[0089] Surface and contact surface sensing: Attach flexible smart patches to monitor surface strain and temperature; deploy an array of temperature sensors at the contact surface between concrete and insulation layer (e.g., deploy 9 contact surface thermometers on a key storage area).

[0090] Water system monitoring: High-precision water temperature sensors, electromagnetic flow meters, and pressure sensors are deployed at the inlet, outlet, and key branches of the cooling water pipe network to monitor water flow, water temperature, pressure, and valve opening.

[0091] Environmental meteorological monitoring: Meteorological stations are set up in and around the dam area to continuously collect data on temperature, relative humidity, wind speed, wind direction, total solar radiation, rainfall and air pressure, with a collection cycle of 30 seconds to 10 minutes.

[0092] 2. Deployment of Edge Recognition and Understanding Layer

[0093] This layer is deployed on edge servers in the dam area or on local computing units of intelligent agents, and is responsible for lightweight processing, fusion and deep understanding of massive and heterogeneous real-time sensing data.

[0094] Data preprocessing and spatiotemporal registration: Noise reduction and correction are performed on visible light images; emissivity correction and environmental reflection compensation are applied to infrared thermal images; filtering and downsampling are performed on lidar point clouds. Through a unified timestamp and spatial coordinate system (such as the dam construction coordinate system), images, point clouds, time-series sensor data, and meteorological data are precisely aligned and fused in time and space. Figure 13 As shown.

[0095] Visual state recognition based on deep learning:

[0096] Thermal insulation coverage recognition: A model trained on YOLO or Mask R-CNN is used to analyze visible light images in real time, identifying various states of the thermal insulation layer such as "normal coverage," "missing," "damaged," "curved edges," "hollow areas," and "poor overlap." The system outputs bounding boxes, pixel-level segmentation masks, coverage area, and recognition confidence scores. Figure 4 As shown. The confidence threshold is set to 0.80. Areas with a confidence level higher than this value are considered effective insulation areas and used to extract infrared temperature; areas with a confidence level lower than this value (which may be obscured by shadows or snow) are marked as "areas to be verified".

[0097] Appearance defect identification: Using instance segmentation models (such as Mask R-CNN) or image recognition algorithms, defects such as cracks, seepage traces, concrete erosion, misalignment, and voids on the dam surface are automatically identified, located at the pixel level, quantified (calculated in terms of length, width, and area), and preliminarily classified.

[0098] Quantitative inversion of thermal insulation performance: This is one of the core innovative steps of this invention, and its inversion and verification process is as follows: Figure 3 As shown.

[0099] Path 1 (Direct Inversion Based on Contact Surface Temperature): Utilizing the outer surface temperature of the insulation layer obtained by an infrared thermal imager ( The concrete-insulation layer interface temperature measured by the contact surface temperature sensor. ), temperature collected by the weather station ( ), wind speed ( ), humidity (RH) and solar radiation ( Based on the data, using a one-dimensional steady-state heat conduction model and the third type of convection-radiation composite boundary conditions, the equivalent heat transfer coefficient of the insulation layer is directly solved. Furthermore, considering the thermal conductivity of the insulation material (… ), to inversely calculate the equivalent insulation thickness in the current state ( The model incorporates a wind speed correction factor (k1, which can reach 3.60 under strong wind conditions without a windproof layer, and can be reduced to about 1.73 with a windproof and waterproof layer), a humidity correction factor (k2, which is 1.3 when the humidity is >80%), and a solar radiation absorption coefficient (e.g., 0.85-0.9) for dynamic correction.

[0100] Path 2 (Indirect Inversion Based on Internal Temperature): Temperature sensors are placed at a specific burial depth (e.g., 1.5m) inside the concrete to obtain the internal temperature. ), combined with infrared surface temperature ( Based on meteorological boundary conditions, the overall thermal resistance is inferred from a one-dimensional unsteady-state heat conduction model, and then indirectly calculated. and .

[0101] Dual-path cross-correction mechanism: the inversion of the above two methods to obtain Value and Real-time comparison is performed. When the relative error between the two exceeds a preset threshold (e.g., 15%), the system automatically issues an early warning, indicating possible abnormal temperature measurement data, improper boundary condition settings, or abnormal insulation layer construction (e.g., internal dampness, uneven compaction), and generates a verification task for that area.

[0102] 3. Deployment of Digital Twin Platform

[0103] This platform is the core connecting the physical world and the information world. It constructs a digital twin model that corresponds one-to-one with the physical dam, is synchronized in real time, is computable, and can be extrapolated. Its state mapping relationship is as follows: Figure 5 As shown.

[0104] Model Construction and Data Injection: Based on the dam design BIM model, geological model, and construction progress information, a refined 3D digital twin was constructed, including dam geometry, material zoning, cooling water network, and sensor placement information. All structured information output from the edge recognition layer—including the vector boundary of the insulation coverage, the location and quantification parameters of defects, the reconstructed full-field temperature distribution, and the inverted data—was incorporated. Isosurfaces, real-time sensor data, agent positions, etc., are all dynamically written into and updated in the digital twin according to a unified spatiotemporal coordinate system.

[0105] Coupled Analysis and Predictive Inference: The digital twin platform has a built-in or interface-integrated professional finite element analysis (FEM) kernel and data-driven predictive models, such as... Figure 6 As shown.

[0106] Finite element analysis: Using a finite element model with updated real-time temperature field and boundary conditions, we simulated the temperature stress during construction and deduced the structural response (displacement and stress) under the coupled action of water pressure, temperature and time-effect loads during operation to locate potential high-risk areas.

[0107] Behavioral prediction models: Integrating time-series prediction models such as the FEM-HST hybrid model, Long Short-Term Memory (LSTM) network, and Informer. For example, the FEM-HST model decomposes water pressure, temperature, and time-dependent components in displacement; the LSTM model uses reservoir water level, multi-point and multi-source temperature, and time series as inputs to predict the deformation or seepage trends of key points in the dam body in future periods, achieving forward-looking early warning.

[0108] 4. Autonomous Decision-Making and Collaborative Deployment

[0109] This layer is the system's "intelligent brain." Based on the real-time status reflected by the digital twin, the inferred future trends, and the built-in engineering knowledge base and security thresholds, it autonomously generates precise control, inspection, repair, and early warning commands. Its internal multi-agent collaborative scheduling logic is as follows: Figure 7 As shown.

[0110] Functional intelligent agent composition: This layer consists of multiple specialized intelligent agents working collaboratively, including a monitoring data management intelligent agent, a dam behavior prediction intelligent agent, an anomaly detection and early warning intelligent agent, an engineering knowledge retrieval intelligent agent, a temperature control and regulation intelligent agent, an insulation inspection and repair intelligent agent, and an insulation thermal performance evaluation intelligent agent, etc.

[0111] Decision logic and instruction generation:

[0112] Temperature control and regulation decision: When the thermal performance evaluation agent determines a certain area The value exceeds the control value (e.g., 0.68 kJ / (m³)). 2 When the thickness of the water supply is insufficient or the equivalent thickness is insufficient, and the digital twin simulation shows that there is a risk of excessive internal and external temperature difference and temperature drop rate, the temperature control intelligent agent will generate detailed water supply control instructions, including target flow rate, inlet water temperature, opening adjustment of specific valves, water supply duration and pipeline start-stop sequence.

[0113] Inspection and Repair Decision-Making: When insulation deficiencies, damage, or insufficient thermal performance are identified, the insulation inspection and repair agent generates a repair task order, specifying repair materials, processes (such as overlaying, spraying, compaction), and work area coordinates. The collaborative scheduling module is simultaneously activated, performing multi-objective optimization calculations based on a comprehensive risk assessment level, the real-time location of each mobile agent, remaining power, load capacity, weather conditions, and air / ground traffic control areas, dynamically planning the most efficient collaborative inspection or work path (e.g., ...). Figure 7 As shown, the nearest wall-climbing robot or drone with matching capabilities is assigned to perform the task.

[0114] Tiered early warning: Based on the severity and urgency of the risk, tiered early warning instructions ranging from "attention" to "danger" are generated and issued through multiple channels such as the platform interface, SMS, and audible and visual alarms, along with handling suggestions.

[0115] 5. Deployment of the control and execution layer

[0116] This layer is the system's "hands and feet," responsible for translating instructions from the digital world into precise actions in the physical world, and collecting on-site feedback to form an execution loop.

[0117] Intelligent water supply execution: Control commands are sent to the local control unit, driving equipment such as water pumps, frequency converters, electric regulating valves, and mixing and temperature control devices to dynamically adjust the flow, temperature, and pressure of the cooling water system. Its closed-loop control logic of "command-execution-monitoring-readjustment" is as follows: Figure 8 As shown. The valve response time can be controlled within 5-60 seconds, and the flow adjustment response is completed within 10-180 seconds.

[0118] Intelligent agent-driven insulation repair: A dispatched wall-climbing robot or drone, equipped with a work module, arrives at the designated location. The repair module may include an automated insulation board grasping and laying mechanism, a spraying robotic arm, or compaction rollers to automatically repair defective areas. The repair process and results are recorded by sensors integrated into the intelligent agent.

[0119] Re-inspection and feedback: After any adjustment or repair action is completed, the system automatically triggers a re-inspection process. For example, after insulation repair, the agent immediately performs visible light and infrared re-inspection on the repaired area (e.g., ...). Figure 9 (As shown), the re-examined image and data are transmitted back. The edge recognition layer performs recognition and inversion again to verify. Whether the value has been restored to the acceptable range and whether the equivalent thickness meets the standard are fed back to the decision-making level with the results of "acceptable" or "requires secondary treatment", forming a closed loop.

[0120] 6. Deployment of Continuous Learning and Knowledge Evolution Layer

[0121] This layer is crucial for the system to achieve "continuous evolution," and its operating mechanism is as follows: Figure 10 As shown, it constructs a dam operation and maintenance database covering the entire lifecycle and multiple modes, and drives iterative optimization of models and strategies based on this database.

[0122] Data accumulation and knowledge base construction: Categorize and store all raw monitoring data, preprocessed data, identification results, inversion results, digital twin status snapshots, every decision instruction, execution feedback records, meteorological data, and design information.

[0123] Iterative optimization of the model:

[0124] Visual recognition model optimization: After accumulating a sufficient number of newly labeled samples (such as adding 100-1000 new insulation layer samples containing various abnormal states), start incremental training or fine-tuning of YOLO and segmentation models to improve their recognition accuracy and robustness in complex scenes (shadows, reflections, snow).

[0125] Inversion and Prediction Model Calibration: Based on long-term accumulated measured temperature data and corresponding meteorological conditions, empirical parameters such as wind speed correction coefficient k1 and humidity correction coefficient k2 are continuously regressed and calibrated. Utilizing newly generated monitoring sequences during operation, behavioral prediction models such as LSTM and Informer are periodically retrained or fine-tuned online to adapt to the impacts of material aging and environmental changes.

[0126] Decision-making strategy optimization: Analyze historical early warning events, response measures and their final effects, and use methods such as reinforcement learning to optimize temperature control rules, inspection cycle configuration and risk warning thresholds (such as control limits and confidence intervals), so that the system's decision-making becomes more and more accurate and efficient.

[0127] II. System Operation Flow

[0128] The system operates according to a strict intelligent closed loop of "perception-recognition-mapping-decision-execution-feedback-optimization," with its top-level methodology and process running throughout the entire lifecycle as follows: Figure 11 As shown, the core operating closed loop is as follows: Figure 12 As shown.

[0129] 1. System Initialization and Parameter Configuration: After deployment, the system loads all algorithm models and sets safety thresholds for each stage (e.g., temperature difference threshold during construction period: 15-20℃; temperature drop rate threshold: 0.5-2℃ / day; β control value: 0.68 kJ / (m³)). 2 ·h·K)), complete network communication integration and full equipment self-test.

[0130] 2. Task-driven collaborative perception: The multimodal perception layer collaboratively collects data based on pre-set routine inspection plans (such as wall-climbing robots inspecting every 4 hours) or emergency tasks dynamically issued by the decision-making layer (such as increased inspections before a cold wave). Mobile intelligent agents work synchronously with fixed nodes to ensure the spatiotemporal consistency of data collection.

[0131] 3. Real-time understanding and diagnosis at the edge: The acquired data stream is processed rapidly at the edge layer. This is achieved through methods such as... Figure 4 and Figure 13 The process shown completes the visual identification of insulation and defects, and performs the following: Figure 3 The core inversion calculation shown outputs a quantitative diagnostic report on the "health status" of the dam body in real time.

[0132] 4. Digital Twin Mapping and Prospective Inference: Diagnostic reports drive real-time updates to the digital twin's status (e.g., Figure 5 The platform utilizes integrated simulation and prediction models (such as...). Figure 6 This involves projecting the temperature and stress fields under the current conditions to predict future trends and identify potential risks.

[0133] 5. Autonomous Collaborative Decision-Making and Task Orchestration: The decision-making level integrates real-time diagnostics and forward-looking simulation results to conduct a comprehensive risk assessment. Once a risk is detected, a response instruction package including control, remediation, and review is immediately generated and implemented through methods such as... Figure 7 The collaborative scheduling mechanism shown efficiently decomposes instruction packets and allocates them to the most suitable execution resources.

[0134] 6. Precise Execution and Closed-Loop Verification: Each unit in the execution layer receives and executes instructions precisely. This applies to water flow control (such as...). Figure 8 ) or insulation repair (such as Figure 9 The execution action itself and the resulting changes in the dam's state will be captured again by the perception layer and input into the system as feedback data to verify the execution effect and form a closed loop.

[0135] 7. Data-driven continuous evolution: Data generated from all the above stages is fed into the continuous learning layer (e.g., Figure 10 The system uses this data to periodically retrain the model, recalibrate the parameters, and optimize the strategy, so that the perception, cognition, decision-making, and execution capabilities of the entire system continuously improve as the running time increases.

[0136] III. Adaptation throughout the entire lifecycle

[0137] This system is highly configurable. By switching the focus of perception tasks, the weight of analysis models, the decision rule base, and the safety threshold, it can seamlessly adapt to the differentiated core needs of different life stages of the dam, achieving "one system, full-process control":

[0138] Construction-phase pouring and curing: The core objective is "cracking prevention." Sensing focuses on the temperature of the pouring surface, water flow parameters, and early insulation; identification emphasizes the quality inspection of insulation installation; the core decision-making is water flow control and insulation assurance; the model mainly consists of temperature field simulation and early strength development model.

[0139] Winter insulation phase: The core objective is "verification of insulation effectiveness." Sensing requires continuous monitoring of the surface, contact surfaces, and internal temperature of the insulation layer, along with enhanced wind speed and radiation monitoring; identification and inversion rely on continuous calculation. And equivalent thickness; the decision focuses on early warning of thermal insulation performance degradation and preventive reinforcement.

[0140] During the water storage and commissioning phase, the core objective is to "track the initial load response." Building upon continued temperature monitoring, the focus is on integrating and merging data on reservoir water level, uplift pressure, seepage pressure, and deformation monitoring. The digital twin will be used to compare and analyze the simulated and measured values ​​of the initial displacement and stress response under water pressure load.

[0141] Operational health monitoring: The core objective is "long-term safety early warning." The sensing system operates routinely, with defect identification, seepage analysis, and deformation prediction becoming key focuses; the digital twin runs predictive models such as FEM-HST and LSTM for extended periods; and the decision-making layer provides intelligent early warnings based on residual analysis and low-probability events.

[0142] Maintenance and repair phase: The core objective is "precise repair and verification". Based on the defect database identified during operation, the system generates detailed repair task orders; during the repair process, an intelligent agent records the repair process; after repair, a special re-inspection process is initiated, the defect status in the digital twin is updated, and the repair effect data is incorporated into the knowledge base to optimize relevant models.

[0143] IV. Electronic Equipment

[0144] An electronic device for implementing the above method has a typical hardware structure as follows: Figure 14 As shown, the device includes at least one processor (e.g., a central processing unit, CPU), memory (e.g., RAM, ROM, flash memory), input / output interfaces (for connecting display devices, input devices, and networks), and a communication bus connecting these components. The memory stores a computer program that, when executed by the processor, controls the device to implement the multi-agent collaborative sensing, decision-making, and closed-loop control method for smart dams described in this invention. The system can be deployed on a local server in the dam area, a cloud server, or an edge computing node cluster.

[0145] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A closed-loop intelligent control method for dams based on multi-agent systems, characterized in that, include: Construct a full-domain multimodal perception system covering the dam surface, dam interior, dam foundation, underwater area and surrounding environment, and collect multi-source monitoring data through autonomous mobile intelligent agents and fixed sensing nodes in collaboration; The multi-source monitoring data is synchronized in time, registered in space, fused in data and identified in state to obtain the insulation coverage state, apparent defect state, temperature field state, structural response state and environmental boundary state of the dam body; The identified dam status is mapped in real time to a pre-built digital twin model of the dam, and status synchronization, trend extrapolation and risk assessment are performed in the digital twin model; Based on the real-time status, predicted trends, finite element analysis results, historical operating data, and preset safety thresholds reflected by the digital twin model, intelligent water supply control commands, insulation repair commands, multi-agent collaborative inspection commands, and risk warning commands are generated autonomously. The instructions are sent to the control and execution layer to complete cooling water flow control, insulation repair work, intelligent agent path scheduling, on-site early warning and re-inspection feedback; Based on monitoring feedback data, control implementation effects, repair and re-inspection results, and historical operation data, the insulation identification model, defect identification model, insulation thermal inversion model, temperature field reconstruction model, behavior prediction model, risk assessment model, and collaborative decision-making strategy are iteratively optimized.

2. The method according to claim 1, characterized in that, The construction of a comprehensive multimodal perception system covering the dam surface, dam interior, dam foundation, underwater area, and surrounding environment includes: Deploy at least one of the following: wall-climbing robots, drones, tracked robots, cable robots, and underwater robots to collect visible light images, infrared thermal images, hyperspectral data, three-dimensional point cloud data, and surface defect data of the dam surface; Temperature sensors, strain sensors, seepage pressure sensors, uplift pressure sensors, crack gauges, smart patches, and water flow monitoring nodes are embedded or attached inside and on the surface of the dam body to collect data on temperature, strain, seepage pressure, uplift pressure, crack opening, water flow rate, and water temperature inside the dam body. Connect to meteorological monitoring equipment to obtain data on ambient temperature, humidity, wind speed, wind direction, solar radiation, rainfall, and snowfall.

3. The method according to claim 1, characterized in that, The status identification of multi-source monitoring data includes: The coverage status of the insulation layer is identified based on the target detection model or semantic segmentation model. The coverage status includes normal coverage, missing, damaged, curled edges, hollow, snow cover, shadow interference, and construction cover. Based on target detection models, instance segmentation models, or image recognition models, the cracks, seepage, erosion, misalignment, voids, and surface damage on the dam surface are located, quantified, graded, and their orientation is extracted. The dam's temperature field was reconstructed based on infrared thermal images, internal temperature, contact surface temperature, meteorological boundaries, and water flow parameters. The internal and external temperature differences, temperature drop rate, temperature gradient, and time-varying temperature characteristics were also calculated.

4. The method according to claim 3, characterized in that, The method also includes a step of inverting the equivalent heat release coefficient and equivalent thickness of the insulation: Based on the surface temperature of the insulation layer, the temperature of the contact surface between the concrete and the insulation layer, the air temperature, wind speed, humidity and solar radiation data, the equivalent heat release coefficient and the equivalent insulation thickness are calculated. Alternatively, based on the internal temperature at a predetermined depth within the dam body, infrared surface temperature, meteorological data, and material parameters, the equivalent heat release coefficient and equivalent insulation thickness can be indirectly calculated. The inversion results based on the contact surface temperature are cross-checked with the inversion results based on the internal temperature. When the relative error between the two exceeds a preset threshold, a prompt is made to perform data verification.

5. The method according to claim 1, characterized in that, The real-time mapping of the dam's state to a pre-constructed digital twin model of the dam includes: The insulation coverage status, defect location, defect size, temperature field distribution, structural deformation, seepage status, water flow parameters, and intelligent agent location information are written into the digital twin model according to spatiotemporal coordinates. The status of the digital twin is dynamically updated based on the monitoring time sequence, forming a traceable, replayable, and predictable digital twin entity; Temperature field simulation, structural response simulation, risk area location, and trend prediction are performed based on the updated digital twin model.

6. The method according to claim 1, characterized in that, The autonomously generated intelligent water flow control commands, insulation repair commands, multi-agent collaborative inspection commands, and risk warning commands include: When the internal and external temperature difference, temperature drop rate, maximum temperature, temperature gradient, or predicted temperature change trend exceeds the preset threshold, control commands are generated for cooling water flow rate, inlet water temperature, valve opening, water flow duration, and pipeline start-up and shutdown sequence. When the insulation layer is found to be missing, damaged, warped, not tightly overlapped, insufficiently covered, or with insufficient thermal insulation performance, task instructions are generated for the repair area, repair material, repair method, work path, and re-inspection sequence. Based on risk level, task priority, agent location, remaining power, payload capacity, weather conditions, and construction area limitations, dynamically plan the verification tasks using drones, wall-climbing robots, ground robots, underwater robots, or humans. Based on the level of risk, tiered early warning instructions are generated, along with corresponding handling suggestions or manual intervention prompts.

7. The method according to claim 1, characterized in that, The iterative optimization includes: The raw data from multi-source monitoring, preprocessed data, insulation identification results, infrared temperature field, insulation equivalent heat release coefficient inversion results, equivalent thickness inversion results, temperature field reconstruction results, digital twin status, water flow control records, insulation repair records, early warning records, execution feedback, meteorological data, and dam design data are classified and stored. Update the insulation recognition model, defect recognition model, or image segmentation model based on the newly added labeled samples; Based on the measured temperature, meteorological conditions, and the effect after repair, the wind speed correction factor, humidity correction factor, material thermal conductivity, and solar radiation absorption factor are corrected. Based on the newly added monitoring data during the operation period, retrain or fine-tune the finite element-HST hybrid model, LSTM model, Informer model or PatchTST model used for dam behavior prediction. Update control limits, low-probability early warning indicators, or confidence interval early warning indicators based on changes in the residual distribution of the prediction model.

8. A multi-agent-based intelligent sensing and collaborative control system for the entire life cycle of a dam, used to implement the method described in any one of claims 1 to 7, characterized in that, include: The multimodal sensing layer is used to collect multi-source monitoring data on the dam surface, dam interior, dam foundation area, underwater area, water supply system and surrounding environment through autonomous mobile intelligent agents and fixed sensing nodes; The edge recognition and understanding layer is used to perform time synchronization, spatial registration, data fusion, insulation recognition, defect recognition, temperature field reconstruction, and insulation thermal parameter inversion on the multi-source monitoring data. The digital twin platform is used to map the identified and inverted dam status information to the dam digital twin model, and to complete status updates, trend projections, risk location, and historical playback. The autonomous decision-making and collaboration layer is used to generate instructions for water flow control, insulation repair, collaborative inspection, and risk warning based on real-time status, predicted trends, and safety thresholds. The control and execution layer is used to perform cooling water flow control, intelligent agent scheduling, insulation repair, on-site early warning, and re-inspection feedback. The continuous learning and knowledge evolution layer is used to store monitoring data, identification results, inversion results, prediction results, decision instructions, and execution feedback, and is used to train, update models, and optimize decision strategies.

9. The system according to claim 8, characterized in that, The autonomous decision-making and collaboration layer includes at least one of the following: monitoring data management intelligent agent, dam behavior prediction intelligent agent, anomaly detection and early warning intelligent agent, engineering knowledge retrieval intelligent agent, temperature control and regulation intelligent agent, insulation inspection and repair intelligent agent, and insulation thermal performance evaluation intelligent agent.

10. The system according to claim 8, characterized in that, The system supports integrated management and control during the construction phase, including pouring and curing, winter insulation, water storage and commissioning, operation monitoring, and maintenance and repair, through parameter configuration, model switching, and strategy optimization.

Images