Search

CN-121981562-A - Dam safety monitoring intelligent system based on multidimensional large model and digital twin

CN121981562ACN 121981562 ACN121981562 ACN 121981562ACN-121981562-A

Abstract

The invention discloses a dam safety monitoring intelligent body system based on a multidimensional large model and digital twinning, and belongs to the technical field of safety monitoring of water conservancy and hydropower engineering. The invention takes a cognitive decision kernel as a core innovation module, collects multi-source data by means of a sky land hydraulic integrated sensor network, realizes cross-modal data fusion and anomaly identification by a multi-dimensional large model cluster, completes multi-physical field coupling simulation by combining a digital twin body, generates an optimal combination scheme by case retrieval and measure unit adaptation calculation, outputs a control instruction after digital twin preview verification, and finally realizes closed-loop control of dam safety real-time early warning, autonomous decision and efficient disposal by an autonomous response mechanism. The method solves the problems of island data, passive response, expert dependence and the like of the traditional monitoring system, remarkably improves the intelligent level and emergency response capability of dam safety management, and has wide engineering application value.

Inventors

  • LI SHUANGPING
  • SHI BO
  • WANG ZHENG
  • YANG CHENYU
  • ZHANG BIN
  • LIU ZUQIANG
  • ZHENG JUNXING
  • GAO LIN
  • YE GUO
  • ZHANG XIN
  • Guan Linjie
  • SONG YANMIN

Assignees

  • 长江空间信息技术工程有限公司(武汉)
  • 华中科技大学

Dates

Publication Date
20260505
Application Date
20260408

Claims (12)

  1. 1. Dam safety monitoring agent system based on multidimensional big model and digital twin, characterized in that it comprises: the data acquisition layer is used for acquiring multi-source time sequence monitoring data through a plurality of types of sensors distributed on the dam structure and the periphery of the dam structure; the data processing and fusing module is used for fusing the multi-source data of the data acquisition layer and outputting structured fused data; The digital twin module is used for constructing a virtual mapping model based on the structured fusion data and outputting a dam running state prediction result; the knowledge graph module is used for constructing a dam safety related structured knowledge system and outputting risk assessment results and expert knowledge support; The cognitive decision kernel is used for receiving output data of the data processing and fusing module, the digital twin module and the knowledge graph module and generating a physical control instruction for safety disposal of the dam; The autonomous response mechanism is used for receiving the control instruction of the cognitive decision core and executing early warning, resource scheduling and decision process evidence storage operation; and the communication and interaction module is used for realizing data interaction among the modules and bidirectional data transmission with an external terminal.
  2. 2. The multi-dimensional large model and digital twinning-based dam safety monitoring agent system according to claim 1, wherein the cognitive decision kernel generating decision scheme comprises the steps of: ① Semantic matching, namely calculating a current abnormal vector q and a historical case vector by adopting a cosine similarity algorithm Similarity of (2); ② Case retrieval, namely selecting the first k historical cases with highest similarity; ③ The decision scheme is generated by dividing the treatment measures in the cases into independent measure units, calculating the adaptation degree of each unit, generating an optimal combination sequence, and outputting physical control instructions after digital twin preview verification.
  3. 3. The dam safety monitoring agent system based on multidimensional large model and digital twin according to claim 2, wherein the cosine similarity algorithm is adopted to calculate current abnormal vector q and historical case vector The similarity method comprises the following steps: ; Wherein: the similarity of the current anomaly vector q to the historical case vector ci is represented.
  4. 4. The dam safety monitoring agent system based on a multidimensional large model and digital twin according to claim 1, wherein determining an optimal decision scheme by calculating the fitness of each unit respectively comprises the following steps: a. Measure vectorization expression For all treatment measures in the historical case library, constructing a unified measure vector: storing the knowledge graph in action nodes of the knowledge graph; b. Measure level adaptation calculation Semantic vector based on current exception event Calculating a dynamic adaptation score for each candidate measure in combination with the real-time monitoring value: dynamic adaptation score @ ) For screening the physical execution units; Wherein, the Semantic similarity of the current anomaly and the measure is represented; Representing an independent measure unit; Representing the success rate weight of the measure in the historical case; representing the risk reduction effect obtained by digital twin simulation; representing the degree of matching of the measure with the current resource availability; , , , weight coefficients representing four indexes; c. multi-measure combination and optimization Generating an optimal combined sequence from a plurality of measures with highest scores: Wherein, the Representing any combination of candidate measure sets; representing a complete set of all optional measures; Representing the value of a parameter that maximizes the objective function; the decision objective function is: max F (A)= E(A)- R(A)- T(A) A represents a measure combination subset, E (A) represents the comprehensive risk reduction effect of the combination measure, R (A) represents the total resource consumption of the combination measure, T (A) represents the total execution time of the combination measure; , , representing each target weight; Generating a device action time sequence in an optimal mode through drainage, inspection, reinforcement and scheme verification; d. digital twin previewing and closed loop correction Simulating candidate combinations in a digital twin model, and simulating the effects of the combinations on a seepage field, a displacement field and a stress field: Wherein, the Representing the risk reduction effect obtained by digital twin simulation; Representing a function that converts the simulated output into a composite index; if the simulation result does not meet the safety constraint, the scheme is automatically eliminated or corrected; e. The optimal decision scheme meeting the constraint of resources and ageing and subjected to digital twin verification is converted into a hardware control signal, and the hardware control signal is sent to the autonomous response mechanism so as to physically start a drainage pump station of the dam, adjust the opening of a flood discharge gate or schedule an unmanned aerial vehicle inspection path, and therefore the running state of the dam is adjusted to be within a safe range.
  5. 5. The multi-dimensional large model and digital twinning-based dam safety monitoring agent system according to claim 1, wherein: The sensor in the data acquisition layer comprises a distributed optical fiber strain gauge, an osmometer, an inclinometer, a GNSS displacement monitoring station, a temperature and humidity sensor, a rain gauge and a water level gauge, wherein the video monitoring system adopts a high-definition low-illumination camera combined with infrared imaging equipment to support all-weather operation, the unmanned aerial vehicle inspection system is provided with a three-dimensional laser radar and a high-resolution camera for acquiring three-dimensional point clouds and texture images of dam appearance structures and surrounding environments, and the multi-mode data are transmitted to a central processing platform after being subjected to preliminary processing and compression encoding through an edge gateway node so as to reduce network bandwidth occupation and improve data instantaneity.
  6. 6. The dam safety monitoring intelligent system based on the multi-dimensional large model and the digital twin is characterized in that the multi-dimensional large model adopts a pre-training-fine-tuning double-stage training strategy, the pre-training stage utilizes a plurality of pieces of historical operation data, simulation data and multi-mode image data of a plurality of dam types and a plurality of climate areas to initialize parameters, an Adam optimizer is adopted, the learning rate initial value is 1e-4 to conduct multi-round training, the fine-tuning stage combines a plurality of pieces of recent actual measurement data of a target dam to conduct parameter optimization, the migration learning adopts a parameter freezing and incremental training strategy to freeze 90% of bottom layer parameters of the pre-training model, only fine-tunes 10% of top layer parameters and full-connection layers to migrate training results of deployed dam bodies to a new dam, and the large model realizes unified modeling and fusion of monitoring data through time sequence mode capturing, spatial feature analysis, cross-mode semantic understanding and abnormal mode identification.
  7. 7. The multi-dimensional large model and digital twinning-based dam safety monitoring agent system according to claim 1, wherein: The digital twin module comprises a geometric model, a physical model, a behavior model and a rule model, wherein after the model predicts risks, the model instructs the physical sensor to encrypt and sample, the geometric model is constructed according to a design drawing, a BIM model and a real-time point cloud, the physical model is coupled with seepage, stress and temperature multi-physical-field calculation based on a finite element method, the behavior model is constructed cooperatively by adopting machine learning and a mathematical model, response behaviors of a dam under different working conditions are simulated, and the rule model integrates operation rules, early warning thresholds and control strategies to realize linkage prediction and optimization control of virtual simulation and real-time operation.
  8. 8. The multi-dimensional large model and digital twinning-based dam safety monitoring agent system according to claim 1, wherein: the knowledge graph module comprises an entity layer, a relation layer and an association analysis layer, wherein the entity layer comprises monitoring equipment, monitoring indexes, risk types, event records and emergency resource entities, the relation layer defines causal, time sequence, space and function multi-type association among the entities, and the association analysis layer maps monitoring data to corresponding fault mode nodes based on a graph neural network.
  9. 9. The multi-dimensional large model and digital twinning-based dam safety monitoring agent system according to claim 1, wherein: The resource scheduling algorithm is based on a Q-learning reinforcement learning strategy, a state space is defined as { emergency equipment state, dam safety risk state and resource availability }, a start-stop sequence of automatic equipment is generated based on the Q-learning algorithm, an action space is defined as { equipment start-stop, resource allocation and sampling frequency }, a reward function is comprehensively designed according to a risk reduction rate, a resource utilization rate and response time, the system is subjected to offline training in a simulation environment and online fine adjustment in actual operation, and finally a scheduling strategy is formed so as to achieve a double optimization target of minimizing response time and resource consumption in dangerous case response.
  10. 10. The multi-dimensional large model and digital twinning-based dam safety monitoring agent system according to claim 1, wherein: The decision process certification is completed through a blockchain certification module, the blockchain certification module is used as a safety verification layer in a control loop, a alliance chain architecture is adopted, decision basis, execution process and feedback result of a cognitive decision core are written into a blockchain account book according to time stamp sequences, blockchain nodes are deployed at a dam operation control center, a supervision mechanism and part of edge nodes to realize multiparty consensus and authority control, certification information comprises a data abstract, decision basis, a digital twin simulation result, a knowledge map inference chain, an execution result and a responsibility main body, so that the non-tamper modification and traceability of the decision process are ensured, the data integrity and tamper resistance of an automatic control instruction output by the cognitive decision core are ensured, and a consensus mechanism adopts a Raft-BFT algorithm based on PBFT improvement, and adapts to the real-time requirement of dam safety monitoring through node role dynamic allocation and consensus process layering optimization.
  11. 11. The multi-dimensional large model and digital twinning-based dam safety monitoring agent system according to claim 1, wherein: the communication and interaction module supports a multi-terminal access mode of a man-machine control interface, a mobile terminal and a large-screen visual terminal, a self-adaptive data visual framework is adopted to realize three-dimensional digital twin scene display, real-time monitoring of data curves, risk distribution thermodynamic diagrams and early warning information panels, voice interaction and multi-language text interaction are supported, a user inquires a dam running state, a historical trend and an emergency plan through natural language, and decision suggestions and simulation prediction results generated by a system are received.
  12. 12. The multi-dimensional large model and digital twinning-based dam safety monitoring agent system according to claim 1, wherein: The system adopts a layered distributed architecture of bottom layer support-intermediate processing-top layer application, each layer of function coupling and data instruction bidirectional linkage, and simultaneously ensures the system operation stability and scene suitability through a software and hardware integrated scheme, and specifically comprises the following steps: the bottom layer is an edge perception and data preprocessing layer, is directly in butt joint with the data acquisition layer and is responsible for receiving multi-mode original data acquired by multi-type sensors, performing edge side preliminary preprocessing, wherein the preprocessing comprises, but is not limited to, data denoising, format conversion and compression coding, transmitting the processed data to the middle layer as required, and simultaneously receiving a sampling frequency adjustment instruction issued by the middle layer to provide real-time and high-quality basic data input for the whole system; The middle layer is a cloud multidimensional large model and an associated analysis layer, carries core calculation tasks of the data processing and fusing module, the digital twin module, the knowledge graph module and the cognitive decision kernel, receives preprocessing data transmitted by the bottom layer, realizes data depth fusion, digital twin multi-physical field simulation and knowledge graph reasoning through the multidimensional large model, generates a preliminary decision suggestion, transmits the preliminary decision suggestion to the top layer, and simultaneously receives decision execution effect data fed back by the top layer for model parameter fine adjustment and reasoning logic optimization; The top layer is an application decision and execution control layer, is linked with the autonomous response mechanism and the communication and interaction module, receives an optimized decision scheme output by the middle layer, triggers a three-level early warning strategy, emergency resource scheduling and blockchain evidence storage operation, and simultaneously realizes monitoring information visualization, early warning pushing and emergency instruction issuing through the communication and interaction module, receives an instruction execution result fed back by the terminal and returns the instruction execution result to the middle layer; Each layer carries out data exchange and instruction transmission through a standardized interface including a sensor-containing data interface, a model calling interface, an instruction transmission interface and an encryption security communication protocol, so that the real-time synchronization of data and the reliable transmission of instructions between the bottom layer and the middle layer and between the middle layer and the top layer are realized; The system comprises edge computing nodes which are deployed on the site of the dam and support the pretreatment of the bottom layer, sensor networks which are deployed on the cloud end and bear the central server cluster of the middle layer computing task to form the bottom layer data acquisition terminal, and a butt joint top layer and a mobile operation and maintenance terminal which is used for receiving early warning and executing instructions on site.

Description

Dam safety monitoring intelligent system based on multidimensional large model and digital twin Technical Field The application relates to the technical field of safety monitoring and intelligent control of water conservancy and hydropower engineering, in particular to a dam safety monitoring intelligent system based on a multidimensional large model and digital twinning. Background In the field of hydraulic engineering, dam safety monitoring is a key link for guaranteeing life and property safety of people and stable social development. However, the existing dam monitoring system has the defects that firstly, the data island phenomenon is serious, various sensors such as a displacement sensor, a seepage sensor and the like work independently, collected data lacks a unified collaborative analysis mechanism, potential information behind the data cannot be fully mined, and a monitoring result is on one side, secondly, a passive response mode is adopted, alarm can be sent out only when a safety index exceeds a fixed threshold value due to overdependence on preset threshold value alarm, hidden risks which do not reach the threshold value but have developed trend are difficult to predict, and prevention cannot be achieved, thirdly, expert dependence is high, professional personnel are required to read and analyze the monitoring data, manual reading and decision making processes take long time when emergency conditions are met, emergency response delay is serious, and requirements of dam safety monitoring on timeliness and accuracy are difficult to meet. In addition, the current dam safety monitoring system has the problems of lack of scientific planning of sensor layout, low system architecture integration level and the like in the construction process, so that dead zones exist in data acquisition and the equipment compatibility is poor. In the operation stage, the traditional monitoring system depends on fixed threshold early warning, cannot adapt to dynamic changes of dam operation conditions, has insufficient generalization capability of an intelligent analysis model, is difficult to accurately identify potential safety hazards in complex environments, has most of decision support functions in post analysis, lacks instantaneity and initiative, and is difficult to meet the safety refined control requirements of modern dams. Therefore, a dam safety monitoring system capable of realizing multi-source data fusion and autonomous intelligent decision is needed, full life cycle monitoring and management of a dam safety state are realized by constructing an intelligent body system integrating the internet of things, artificial intelligence and edge computing technology, real-time performance, accuracy and initiative of monitoring are improved, dependence on artificial intervention is reduced, self-adaption and self-optimization capacity of the system are enhanced, and accordingly intelligent level and emergency response capacity of dam safety operation are comprehensively improved. Disclosure of Invention The embodiment of the application provides a dam safety monitoring intelligent body system based on a multidimensional large model and digital twinning, which aims to solve the technical problems. The dam safety monitoring intelligent system based on the multidimensional large model and the digital twin comprises: The following modules are sequentially cooperated and functionally coupled, and real-time monitoring and intelligent disposal of dam safety are realized through closed loop logic of data acquisition-processing fusion-cognitive decision-autonomous response-interaction feedback: The data acquisition layer is used for acquiring multisource time sequence monitoring data such as deformation, seepage pressure, temperature, stress strain, weather, hydrology and the like through an integrated multisystem sensor network distributed on a dam structure and the periphery, and forming a multi-mode data set by combining video monitoring, three-dimensional laser point cloud, geological radar and unmanned aerial vehicle inspection images; The data processing and fusion module is used for receiving the multi-mode data set output by the data acquisition layer, adopting a multi-dimensional large model with time sequence mode capturing, spatial feature analysis and cross-modal semantic understanding capability, carrying out feature extraction, semantic understanding and cross-modal alignment on the multi-mode data, and simultaneously combining data cleaning, outlier detection and time sequence complementation technology to realize unified modeling and deep fusion of monitoring data and output structured fusion data, wherein the structured fusion data are respectively transmitted to the digital twin module, the knowledge graph module and the cognitive decision kernel to provide high-quality data for virtual simulation, knowledge construction and decision analysis; The digital twin module is used for receiving the structured fu