A reservoir intelligent inspection system based on knowledge graph and digital twin

Abstract

An intelligent reservoir inspection system based on knowledge graph and digital twin, which includes a perception and acquisition layer, a modeling and simulation layer, a data processing and decision support layer, a communication layer, and an interactive visualization layer; the perception and acquisition layer obtains image and video data, water level data, rainfall data, temperature and humidity data of the reservoir and its surrounding environment through cameras and sensors; the modeling and simulation layer maps the actual reservoir and its surrounding environment into a virtual model, and imports the data obtained by the perception and acquisition layer into the virtual model, and realizes dynamic monitoring and analysis through digital twin technology; the data processing and decision support layer processes, analyzes, mines data and provides decision support; the communication layer is responsible for the interconnection, remote control and automatic execution of tasks of the system; the interactive visualization layer provides an operation interface and data display.

Description

Technical Field

[0001] The present invention belongs to the technical field of reservoir inspection systems, and particularly relates to an intelligent reservoir inspection system based on a knowledge graph and digital twin. Background Art

[0002] Reservoirs are key facilities for water resource regulation, capable of effectively storing and distributing water sources to ensure the water use needs of agriculture, industry, and cities. They not only contribute to flood control and disaster reduction, promote economic development, but also have far-reaching significance for ecological protection and social stability. To ensure the safe operation of reservoirs and maximize their benefits, regular reservoir inspections are crucial.

[0003] Traditional reservoir inspections generally rely on manual inspections. Inspectors often need to visit each inspection point one by one, with a huge workload, long time consumption, and are easily affected by external factors such as weather and terrain. Subjective factors of inspectors may also lead to omissions, resulting in low efficiency. In addition, inspectors often need to enter high-risk areas (such as high dams, landslide areas, deep water areas, etc.) for inspections, posing significant safety hazards. Especially in bad weather or sudden natural disasters, the lives of personnel may be threatened. After each inspection, data still needs to be manually recorded and sorted, with information transmission lagging behind, making it impossible to quickly grasp the status and potential risks of the reservoir area. The results of manual inspections are often limited to written records and photos, making it difficult to intuitively present the severity or change trend of overall problems in the reservoir area, lacking timely professional decision-making support and being unable to make effective decisions in a timely manner.

[0004] Currently, with the development of technology, some modern inspection technologies, such as unmanned aerial vehicles, sensors, remote monitoring systems, etc., have been applied to reservoir inspections. These technologies can provide real-time data collection and monitoring, reduce the frequency of manual inspections, and can effectively avoid the entry of personnel into high-risk areas, improving inspection safety.

[0005] However, the existing technologies still have some deficiencies. Although a large amount of data can be obtained in real time, there is a lack of intelligent analysis and decision-making support for these data, and manual intervention is still required for problem identification and judgment. At the same time, there is a lack of intuitive visual display, making it difficult to quickly help managers identify potential risks or the severity of overall problems, resulting in the overall efficiency and response speed of the system not reaching the expected optimal level.

[0006] Therefore, a reservoir inspection system with a high degree of visualization and capable of providing intelligent decision-making support is needed, and this application has conducted further research in this direction. Summary of the Invention

[0007] In view of the deficiencies in the prior art, the present application provides a reservoir intelligent inspection system based on a knowledge graph and digital twin. This system can reflect the dynamic changes in the reservoir area in real time through digital twin technology, intelligently judge potential hazards through drone inspections, analyze problems based on the knowledge graph, query historical events and relevant data, and comprehensively evaluate problems by combining cross-domain (such as climate, environment, urban planning, etc.) knowledge graphs, helping decision-makers make more scientific decisions from a global perspective, improving reservoir safety and management efficiency, that is, being able to optimize the efficiency and accuracy of reservoir inspection work and solve the problems of low efficiency of manual inspection and lag in decision-making existing in the prior art.

[0008] The technical solution of the present application is as follows.

[0009] A reservoir intelligent inspection system based on a knowledge graph and digital twin, the system includes a perception and acquisition layer, a modeling and simulation layer, a data processing and decision support layer, a communication layer, and an interactive visualization layer; the perception and acquisition layer obtains image and video data, water level data, rainfall data, temperature and humidity data of the reservoir and its surrounding environment through cameras and sensors; the modeling and simulation layer maps the actual reservoir and its surrounding environment into a virtual model, and imports the data obtained by the perception and acquisition layer into the virtual model to achieve dynamic monitoring and analysis through digital twin technology; the data processing and decision support layer performs data processing, analysis, mining and provides decision support; the communication layer is responsible for the interconnection, remote control and automatic execution of tasks of the system; the interactive visualization layer provides an operation interface and data display.

[0010] Preferably, the data processing and decision support layer performs in-depth analysis and intelligent decision-making by integrating cross-domain knowledge graphs, combining historical data and real-time data, and generates targeted and global operation and maintenance suggestions.

[0011] Preferably, the cross-domain knowledge graph is integrated by the following method: S10: Construction of cross-domain knowledge graph: According to multiple domains to be integrated, construct knowledge graphs of the water conservancy field and other related fields, and the knowledge graph includes at least one of entities, attributes, relationships and instances; automatically extract and construct domain knowledge graphs from different data sources through data mining, natural language processing NLP, entity recognition NER technical means; S20: Standardized interface design: To ensure the interoperability of cross-domain data and knowledge, design standardized data interfaces and exchange protocols, and the interface supports data transmission, access and update between different domain knowledge graphs; S30: Construct a cross-domain graph neural network based on graph neural network and transfer learning to align and fuse multi-domain graph entities and construct a cross-domain unified graph; S40: Introduce blockchain technology: Use blockchain technology to ensure the data credibility, privacy protection and security of the cross-domain knowledge graph.

[0012] Preferably, the knowledge graph field includes: (1) Water conservancy engineering field: knowledge about the construction, design, operation, and maintenance of reservoirs, including design specifications, materials, and construction methods of water conservancy projects; the knowledge graph in this field can provide support for the construction, maintenance, and transformation of reservoirs, and help analyze the performance and durability of reservoir facilities; (2) Hydrometeorology field: relevant data involving precipitation, flow, water level, and climate change factors related to weather, climate, and rainfall; the knowledge graph in this field can assist in identifying potential threats brought by severe weather, such as heavy rain and floods, during reservoir scheduling and inspection; (3) Environmental protection field: knowledge about water quality monitoring and ecological protection, involving water pollutants (such as heavy metals, organic substances, nutrients, etc.), ecological environment changes, species diversity, etc.; the knowledge graph in this field can provide a scientific basis for ecological restoration or pollution prevention and control to ensure the water quality and ecological balance of reservoirs; (4) Urban planning field: planning issues related to land use, infrastructure construction, and urban development related to reservoirs; the knowledge graph in this field can help identify the potential impact of urbanization on water quality and water resources, and evaluate the long-term impact of urban expansion on reservoirs; (5) Agricultural irrigation field: involving water resource management, agricultural water use, and irrigation technology; including demand prediction for agricultural irrigation and effective utilization of water resources; the knowledge graph in this field can assist in reasonably scheduling water resources in reservoirs for crop growth; (6) Laws, regulations, and policies field: relevant laws, regulations, policy requirements, and compliance inspections regarding reservoir management; the knowledge graph in this field provides real-time queries of laws and regulations related to reservoir management to ensure that the inspection and operation processes comply with government and industry standards.

[0013] Preferably, the knowledge graph is fused based on graph neural network and transfer learning, and the fusion method includes the following steps:

[0014] S31: For the graph neural network of each field (assuming the knowledge graphs of source field A and target field B), represent the node representation of each layer through the update rule of the graph convolutional network GCN as follows:

[0015]

[0016] Where:

[0017] represents the node representation of the th layer, which is the initial node feature;

[0018] is the normalized adjacency matrix, including self-connection;

[0019] is the learning weight matrix of the th layer;

[0020] is an activation function, and the rectified linear unit ReLU can be used;

[0021] In the task of cross-domain alignment, the graphs between two domains can be respectively subjected to graph convolution operations in this way to obtain the node representations of the two graphs.

[0022] S32: The two domains are defined as the source domain A and the target domain B. After constructing the graph convolutional network, pre-training is performed on the source domain A: Train a GCN model on the source domain A so that it learns the structural and entity feature representations of the source domain. At the same time, a regularization strategy is introduced. Let the GCN training loss of the source domain A be , and the regularization term of the source domain A is , as follows:

[0023]

[0024] Among them, is the pre-training loss function, and γ is a hyperparameter used to balance the regularization term Ra and the loss La of the source domain in the loss function.

[0025] S33: Transfer the pre-trained model to the target domain B, introduce an adversarial training mechanism to reduce the distribution difference between the source domain and the target domain, and perform fine-tuning and adversarial training simultaneously on the target domain B: During fine-tuning, the GCN network structure of the source domain A can be fixed, and the node features and relationship mappings of the target domain can be adjusted; Let the fine-tuning loss of the target domain B be , and the adversarial training loss is , The regularization term of the target domain B, and the total loss of the training process can be expressed as:

[0026]

[0027] Among them, , are important hyperparameters for regulating transfer learning, used to balance the losses of fine-tuning and adversarial training, is the adversarial training loss;

[0028] The goal of adversarial training is to achieve transfer learning by maximizing the similarity between the source domain and the target domain, which is achieved through an adversarial loss function, as follows:

[0029]

[0030] Among them:

[0031] is the discriminator;

[0032] is the node representation of the source domain and the target domain;

[0033] is the adversarial loss.

[0034] Preferably, the specific steps to introduce blockchain technology are as follows:

[0035] S41: Data on-chain: When new data (such as new nodes or relationships in the knowledge graph) is added to the knowledge graph, its data hash value is written into the blockchain; each data has a unique hash value, representing the "fingerprint" of the data. Once the data is written into the blockchain, it cannot be modified or deleted, ensuring the immutability and historical traceability of the data.

[0036] S42: Verification mechanism: All updates or modifications to the cross-domain knowledge graph need to be verified by consensus among multiple participating parties; for example, the update of certain data may require verification by multiple nodes (such as academic experts, data source providers, etc.). Only through the consensus mechanism can the data be confirmed and added to the knowledge graph. This can ensure that the data is widely verified and avoid the addition of false or misleading data.

[0037] S43: Encryption technology: Blockchain uses encryption algorithms (such as public-private key encryption) to ensure that data can only be accessed and decrypted by authorized users. Even if the data is written into the blockchain, only authorized users can view the detailed content of the data, rather than being exposed to everyone.

[0038] S44: Zero-knowledge proof: Zero-knowledge proof (ZKP) is a cryptographic technique that allows one party (the prover) to prove the authenticity of a certain piece of information to another party (the verifier) without revealing the specific content of the information;

[0039] S45: Distributed storage: Data is not stored on a single server or central institution, but is distributed across multiple nodes; even if one or more nodes fail or are attacked, the copies on other nodes can still ensure the integrity and availability of the data.

[0040] S46: Smart contract: A smart contract is a program that automatically executes an agreement and can be automatically executed when specific conditions are met. Through smart contracts, data access and operations in the cross-domain knowledge graph can be automatically managed and monitored according to predefined rules, ensuring that only operations that meet specific conditions can be performed and avoiding unsafe operations.

[0041] S47: Permission management: Through the permission management system of the blockchain, a strict access control mechanism can be set to ensure that only authorized users can modify or access certain sensitive data.

[0042] In the prior art, the conventional reservoir inspection system mainly has the following problems:

[0043] (1) Limited data analysis ability: Existing sensors and monitoring devices can collect a large amount of data in real time, but most inspection systems lack data processing and intelligent analysis functions, unable to deeply analyze the data, automatically identify potential problems or predict risks, resulting in the failure to fully utilize the data value.

[0044] (2) Low level of intelligence: Existing reservoir inspection systems are unable to automatically extract knowledge from multi-source data. Despite a large amount of real-time data, managers need to analyze and judge manually, lacking intelligent decision support, which affects the response speed and efficiency.

[0045] (3) Unable to provide a global perspective: Existing reservoir inspections mainly focus on the water conservancy field itself, lacking the integration of knowledge in fields such as the environment, climate change, and urban planning. When dealing with problems, it may not be able to provide a global perspective, resulting in narrow decision-making, lacking pertinence and long-term nature.

[0046] (4) Lack of dynamic monitoring ability: Existing reservoir monitoring mainly relies on the data of fixed sensors, which can often only reflect the state of the reservoir at a certain moment and cannot provide comprehensive and real-time dynamic information of the reservoir. This makes it impossible for managers to dynamically observe the overall state of the reservoir, equipment failures, and potential risks and make real-time decisions.

[0047] In contrast, this application provides a reservoir intelligent inspection system based on knowledge graphs and digital twins. This system can reflect the dynamic changes in the reservoir area in real time through digital twin technology, intelligently judge potential hidden dangers through UAV inspections, analyze problems based on knowledge graphs, query historical events and relevant data, and comprehensively evaluate problems by combining cross-domain (such as climate, environment, urban planning, etc.) knowledge graphs, helping decision-makers make more scientific decisions from a global perspective, improving reservoir safety and management efficiency, that is, being able to optimize the efficiency and accuracy of reservoir inspection work and solve the problems of low efficiency of manual inspection and lag in decision-making existing in the prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 Overall flowchart of the reservoir intelligent inspection system based on knowledge graphs and digital twins.

[0049] Figure 2 Construction of a cross-domain graph neural network graph fusion method based on graph neural networks and transfer learning.

[0050] Figure 3 Flowchart of blockchain technology applied to the dynamic update scenario of knowledge graphs. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0051] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0052] In the following embodiments, the same or similar reference numerals throughout denote the same or similar elements or elements having the same or similar functions. The embodiments described below by reference to the accompanying drawings are exemplary only for explaining the present invention and should not be construed as limiting the present invention.

[0053] Referring to the accompanying drawings, a reservoir intelligent inspection system based on a knowledge graph and digital twin involved in the present application is mainly divided into a perception and acquisition layer, a modeling and simulation layer, a data processing and decision support layer, a communication layer, and an interactive visualization layer.

[0054] Perception and acquisition layer: This layer is mainly responsible for the real-time monitoring and data acquisition of the reservoir environment, facilities, and equipment, using tools such as sensors, monitoring cameras, and drones. Monitoring cameras are installed in the reservoir area, and various sensors such as water level, rainfall, temperature and humidity, and liquid level are equipped on the physical dam to collect data on the operation status, structural health, and surrounding environment of the dam in real time. The drone is responsible for aerial inspections to obtain images and video information of the dam and its surrounding areas, and all data is transmitted to the central system in real time.

[0055] Modeling and simulation layer: This layer maps the actual physical dam and surrounding environment into a virtual model, simulates the state of the reservoir area according to the data obtained by the perception and acquisition layer, and realizes dynamic monitoring and analysis through digital twin technology.

[0056] Data processing and decision support layer: This layer mainly performs data processing, analysis, mining, and provides decision support. The raw data collected by the sensors is cleaned, filtered, and verified, and converted into various data formats that meet business requirements according to predetermined rules. Through the integration of cross-domain knowledge graphs, combined with historical data and real-time data, in-depth analysis and intelligent decision-making are carried out to generate targeted and overall operation and maintenance suggestions.

[0057] Internet of Things communication layer: This layer is mainly responsible for the interconnection, remote control, and automatic execution of tasks of the system. Through wireless communication technology, data transmission between sensors, control devices, and monitoring centers is ensured. When an abnormality is detected or a set threshold is reached, the system can automatically issue an alarm or execute preset automated operations (such as enabling standby equipment, adjusting water flow, etc.).

[0058] Interactive visualization layer: This layer is the inspection and scheduling center, providing a friendly operation interface and data display for users to ensure that relevant personnel can intuitively and effectively understand the operation status of the reservoir. The real-time status, early warning information, and inspection results of the reservoir are displayed through 3D models and charts. Functions such as inspection task scheduling, equipment status monitoring, and fault alarm are provided.

[0059] The following is a specific embodiment in the present application.

[0060] An embodiment of the present invention provides a reservoir intelligent inspection system based on a knowledge graph and digital twin. The various module layers and corresponding methods in this system are as follows.

[0061] Step S1: Install monitoring cameras in the reservoir area, and equip the physical dam with various sensors such as water level, rainfall, temperature and humidity, and liquid level to collect data on the operation status, structural health, and surrounding environment of the dam in real time, and transmit the data to the Internet of Things central platform in real time. Collect the domain knowledge required to construct the knowledge graph, including but not limited to fields such as water conservancy projects, hydrometeorology, environmental protection, urban planning, agricultural irrigation, laws, regulations, and policies. Focus on collecting materials on the inspection and operation and maintenance services in the field of water conservancy projects. Among them, the domain knowledge and its functions are as follows:

[0062] Water conservancy projects, knowledge about the construction, design, operation, and maintenance of reservoirs, including design specifications, materials, construction methods, etc. of water conservancy projects. The knowledge graph in this field can provide support for the construction, maintenance, and renovation of reservoirs, and help analyze the performance, durability, etc. of reservoir facilities.

[0063] Hydrometeorology, relevant data on factors such as precipitation, flow, water level, and climate change, involving weather, climate, rainfall, etc. The knowledge graph in this field can help identify potential threats brought by severe weather, such as heavy rain and floods, during the scheduling and inspection of reservoirs.

[0064] Environmental protection, knowledge about water quality monitoring, ecological protection, etc., involving water pollutants (such as heavy metals, organic matter, nutrient salts, etc.), ecological environment changes, species diversity, etc. The knowledge graph in this field can provide a scientific basis for ecological restoration or pollution prevention and control to ensure the water quality and ecological balance of the reservoir.

[0065] Urban planning, involving planning issues related to land use, infrastructure construction, urban development, and the reservoir. The knowledge graph in this field can help identify the potential impacts of urbanization on water quality and water resources, and evaluate the long-term impacts of urban expansion on the reservoir.

[0066] Agricultural irrigation, involving aspects such as water resource management, agricultural water use, and irrigation technology. It includes the demand prediction of agricultural irrigation and the effective utilization of water resources, etc. The knowledge graph in this field can help rationally schedule the water resources in the reservoir for crop growth.

[0067] Laws, regulations, and policies, relevant laws, regulations, policy requirements, and compliance inspections regarding reservoir management. The knowledge graph in this field provides real-time queries on the laws and regulations related to reservoir management to ensure that the inspection and operation processes comply with government and industry standards.

[0068] Step S2: Clean and verify the original data, and convert it into various data formats that meet business requirements according to the rules. Construct knowledge graphs for each field based on knowledge graph construction technology, and fuse cross-field knowledge graphs. Among them, a cross-field graph neural network is constructed based on graph neural network and transfer learning for graph fusion method, please refer to Figure 2 :

[0069] Step S2.1.1, construct knowledge graphs for each field based on knowledge graph construction technology, such as field A, field B, field C, etc.

[0070] Step S2.1.2, for the graph neural network of each field (assuming the graphs of field A and field B), the node representation of each layer (graph convolutional network GCN) can be represented by the following update rules:

[0071]

[0072] Where:

[0073] represents the node representation of the th layer, is the initial node feature;

[0074] is the normalized adjacency matrix (including self-connection);

[0075] is the th layer's learning weight matrix;

[0076] is the activation function, usually using the rectified linear unit (ReLU).

[0077] In the task of cross-field alignment, the graphs between fields can be respectively subjected to graph convolution operations in this way to obtain the node representations of the two graphs.

[0078] Step S2.1.3, after constructing the graph convolutional network, pre-train the source field A. Train a GCN model on the source field A so that it learns the structure and entity feature representations of the source field, and at the same time introduce a regularization strategy. Let the GCN training loss of the source field A be , and the regularization term of the source field A be .

[0079]

[0080] Where, is the pre-training loss function, and γ is a hyperparameter used to balance the regularization term Ra and the loss La of the source field in the loss function.

[0081] Step S2.1.4, transfer the pre-trained model to the target domain B, introduce an adversarial training mechanism to reduce the distribution difference between the source domain and the target domain, and perform fine-tuning and adversarial training simultaneously on domain B. During fine-tuning, the GCN network structure of the source domain can be fixed, but the node features and relationship mappings of the target domain can be adjusted. Let the fine-tuning loss of the target domain B be , and the adversarial training loss be .

[0082]

[0083] where , are important hyperparameters for adjusting transfer learning, used to balance the losses of fine-tuning and adversarial training, is the adversarial training loss. The goal of adversarial training is to achieve transfer learning by maximizing the similarity between the source domain and the target domain. It is achieved through an adversarial loss function, as follows:

[0084]

[0085] where:

[0086] is the discriminator;

[0087] are the node representations of the source domain and the target domain;

[0088] is the adversarial loss.

[0089] Meanwhile, blockchain technology is adopted to ensure the data credibility, privacy protection and security of the cross-domain knowledge graph. Please refer to Figure 3 :

[0090] Step S2.2.1, add knowledge graph data.

[0091] Step S2.2.2, perform data preprocessing. Generate a unique hash value for the newly added data (such as entity relationships, attributes) through algorithms such as SHA-256 as the "fingerprint" of the data. Sensitive data is encrypted using asymmetric encryption (such as RSA) or homomorphic encryption to ensure that only authorized parties can decrypt it; verify the logical validity of the data through zero-knowledge proof technology (such as zk-SNARKs) without exposing the original data.

[0092] Step S2.2.3, perform multi-party consensus verification. Nodes verify the authenticity of the data through the PBFT or PoA consensus mechanism.

[0093] Step S2.2.4, Smart contract permission verification. The smart contract checks the permissions of the requester and only allows operations that meet the conditions.

[0094] Step S2.2.5, Distributed storage. The original data is stored in IPFS or a private database, and only the hash value and storage address are stored on the chain.

[0095] Step S2.2.6, Blockchain on-chain. The signed transaction is broadcast to the nodes, and a new block is generated through consensus such as PoW / PoS to form a hash chain. The block hash chain structure ensures that historical data cannot be modified.

[0096] Step S2.2.7, Trigger smart contract rules. The on-chain event triggers the contract logic to automatically update the knowledge graph state tree (such as adding entity link relationships). If the execution fails (such as insufficient permissions), the contract rolls back to the previous state, but the transaction record is still retained.

[0097] Step S2.2.8, Update the knowledge graph state tree. Write the new entity relationship into the Merkle tree to support fast retrieval and historical traceability.

[0098] Step S3: Use 3D GIS and drone oblique photography technology to construct a 3D digital scene model of the reservoir area.

[0099] Step S3.1, Collect the basic geographical data of the reservoir, plan the drone flight area, set the flight altitude, forward overlap rate and side overlap rate, and ensure the calibration of the equipment (multi-lens oblique camera, RTK positioning).

[0100] Step S3.2, Obtain high-resolution images (including ortho, front, back, left, and right five perspectives) through multi-angle oblique photography of the drone, synchronously record the POS data (position and attitude), and fly in blocks for complex terrain.

[0101] Step S3.3, Use software such as Pix4D and ContextCapture for aerial triangulation to generate dense point clouds; construct a TIN triangulation network through the point clouds, generate a white model and automatically map the texture, and output a 3D model in OSGB or 3D Tiles format.

[0102] Step S3.4, Repair defects such as water surface holes and vegetation noise, simplify redundant triangular faces; import the model into a 3D GIS platform (such as ArcGIS, SuperMap), overlay vector data such as terrain, hydrology, and engineering facilities, and construct a spatio-temporal integrated scene.

[0103] Step S4, Overlay the constructed reservoir area model data on the GIS map for display, simulate the reservoir area state according to the collected data, and display the environmental attribute information; implement functions such as recommended maintenance plans and intelligent Q&A during the inspection process, and construct a reservoir intelligent inspection system based on knowledge graph and digital twin.

[0104] The constructed system includes a cockpit, a three-dimensional visual monitoring subsystem, a safety inspection subsystem, an intelligent question-answering and knowledge retrieval subsystem, and a comprehensive early warning and plan deduction subsystem.

[0105] (1) The cockpit uses visual charts to display comprehensive data of multiple reservoirs, including macro indicators such as safety scores, flood control capabilities, and economic benefit analysis. It can integrate business data and generate reports with one click.

[0106] (2) The 3D visualization monitoring subsystem uses digital twin technology to build a holographic reservoir scene, dynamically displaying data such as topography, equipment distribution, and real-time water level / seepage. The 3D model is updated through sensor and drone inspection data, and key parameters such as water level, rainfall, and reservoir capacity are displayed simultaneously. Click on the monitoring point to view real-time monitoring.

[0107] (3) The safety inspection subsystem realizes inspection task management, hidden danger photo uploading and trajectory tracking, supports the real-time fusion of drone video and twin scenes, and superimposes the drone footage onto the three-dimensional scene in real time to assist in the precise positioning of leakage points or illegal intrusion areas.

[0108] (4) Intelligent question-answering and knowledge retrieval subsystem, which supports users to quickly obtain relevant knowledge, emergency plans and real-time data reports through text or voice commands. Based on multi-domain knowledge graphs, dynamic reasoning is performed in combination with the user's question context to generate cross-domain collaborative decision-making recommendations and risk warnings.

[0109] (5) The comprehensive early warning and emergency plan simulation subsystem, based on the "four predictions" (forecast, warning, rehearsal, and emergency plan) capabilities, dynamically marks risk areas through red / orange / yellow three-level warning signs.

[0110] Step S5: The user initiates a drone inspection task at the dispatch center and checks the equipment status monitoring and fault alarm information.

[0111] Step S6: For abnormal inspection results and fault alarm information, intelligent question and answer are used to provide suggestions; based on the suggestions, business personnel rectify the relevant problems.

[0112] Step S7: export business reports regularly based on historical data.

[0113] As can be seen from the above description, the present application provides a reservoir intelligent inspection system based on a knowledge graph and digital twin. This system can reflect the dynamic changes in the reservoir area in real time through digital twin technology, intelligently judge potential hidden dangers through drone inspections, analyze problems based on the knowledge graph, query historical events and relevant data, and comprehensively evaluate problems in combination with cross-domain (such as climate, environment, urban planning, etc.) knowledge graphs, helping decision-makers make more scientific decisions from a global perspective, improving reservoir safety and management efficiency, that is, being able to optimize the efficiency and accuracy of reservoir inspection work and solve the problems of low efficiency of manual inspection and lag in decision-making existing in the prior art.

[0114] The protection scope of the present invention includes but is not limited to the above embodiments. The protection scope of the present invention is subject to the claims, and any substitutions, deformations, and improvements that are easily conceivable by those skilled in the art to this technology fall within the protection scope of the present invention.

Claims

1. A reservoir intelligent inspection system based on a knowledge graph and digital twin, characterized in that, The system includes a perception and acquisition layer, a modeling and simulation layer, a data processing and decision support layer, a communication layer, and an interactive visualization layer; The perception and acquisition layer obtains image and video data, water level data, rainfall data, temperature and humidity data of the reservoir and its surrounding environment through cameras and sensors; The modeling and simulation layer maps the actual reservoir and its surrounding environment into a virtual model, and imports the data obtained by the perception and acquisition layer into the virtual model to achieve dynamic monitoring and analysis through digital twin technology; The data processing and decision support layer processes, analyzes, mines data and provides decision support; The communication layer is responsible for the interconnection, remote control and automatic execution of tasks of the system; The interactive visualization layer provides an operation interface and data display; The data processing and decision support layer conducts in-depth analysis and intelligent decision-making by integrating cross-domain knowledge graphs, combining historical data and real-time data, and generates targeted and overall operation and maintenance suggestions; The cross-domain knowledge graph is fused through the following methods: S10: Construction of cross-domain knowledge graph: According to multiple domains to be fused, construct knowledge graphs of the water conservancy field and other related fields, where the knowledge graph includes at least one of entities, attributes, relationships and instances; through data mining, natural language processing NLP, entity recognition NER technical means, automatically extract and construct domain knowledge graphs from different data sources; S20: Standardized interface design: To ensure the interoperability of cross-domain data and knowledge, design standardized data interfaces and exchange protocols, and the interfaces support data transmission, access and update between different domain knowledge graphs; S30: Construct a cross-domain graph neural network based on graph neural network and transfer learning, align and fuse multi-domain graph entities, and construct a cross-domain unified graph; S40: Introduce blockchain technology: Use blockchain technology to ensure the data credibility, privacy protection and security of cross-domain knowledge graphs; The knowledge graph is fused based on graph neural network and transfer learning, and this fusion method includes the following steps: S31: For the graph neural network of each domain, represent the node representation of each layer through the update rule of the graph convolutional network GCN, as follows: ； Where: Indicates the representation of the nodes in the ith layer, which is the initial node feature; is a normalized adjacency matrix, including self-connections; is the learning weight matrix of the is an activation function, and the rectified linear unit (ReLU) can be used; In the task of cross-domain alignment, the graphs of the two domains are respectively subjected to graph convolution operations in this way to obtain the node representations of the two graphs; S32: Two domains are defined as the source domain A and the target domain B. After constructing the graph convolutional network, pre-training is performed on the source domain A: Train a GCN model on the source domain A so that it learns the structural and entity feature representations of the source domain. At the same time, a regularization strategy is introduced. Let the GCN training loss of the source domain A be , and the regularization term of the source domain A is as follows: ； Among them, is the pre-training loss function, and γ is a hyperparameter used to balance the regularization term Ra and the source domain loss La in the loss function; S33: Transfer the pre-trained model to the target domain B, introduce an adversarial training mechanism to reduce the distribution difference between the source domain and the target domain, and perform fine-tuning and adversarial training on the target domain B simultaneously: During fine-tuning, the GCN network structure of the source domain A can be fixed, and the node features and relationship mappings of the target domain can be adjusted; Let the fine-tuning loss of the target domain B be , the adversarial training loss be , the regularization term of the target domain B, and the total loss of the training process can be expressed as: ； Among them, and are important hyperparameters for adjusting transfer learning, used to balance the losses of fine-tuning and adversarial training, is the adversarial training loss; The goal of adversarial training is to achieve transfer learning by maximizing the similarity between the source domain and the target domain, and it is achieved through an adversarial loss function, as shown below: ； Where: is a discriminator; is the node representation of the source domain and the target domain; It is the adversarial loss.

2. The intelligent reservoir inspection system based on knowledge graph and digital twin according to claim 1, characterized in that, The knowledge graph domains include: (1) Water conservancy engineering field: Knowledge about the construction, design, operation and maintenance of reservoirs, including design specifications, materials and construction methods of water conservancy projects; the knowledge graph in this field can provide support for the construction, maintenance and transformation of reservoirs, and help analyze the performance and durability of reservoir facilities; (2) Hydrometeorological field: Relevant data involving precipitation, flow, water level, and climate change factors related to weather, climate, and rainfall; the knowledge graph in this field can help identify potential threats brought by bad weather during reservoir scheduling and inspection; (3) Environmental protection field: including knowledge of water quality monitoring and ecological protection, covering water pollutants, ecological environment changes, and species diversity; the domain knowledge graph can provide a scientific basis for ecological restoration or pollution prevention and control to ensure the water quality and ecological balance of the reservoir. (4) Urban planning field: including planning issues related to land use, infrastructure construction, urban development, and the reservoir; the domain knowledge graph in this field can help identify the potential impacts of urbanization on water quality and water resources and evaluate the long-term impacts of urban expansion on the reservoir. (5) Agricultural irrigation field: involving water resource management, agricultural water use, and irrigation technology; including demand forecasting for agricultural irrigation and the efficient use of water resources; the domain knowledge graph in this field can help rationally allocate the water resources in the reservoir for crop growth. (6) Laws, regulations, and policies field: relevant laws, regulations, policy requirements, and compliance inspections for reservoir management; the domain knowledge graph in this field provides real-time queries of laws and policies related to reservoir management to ensure that the inspection and operation processes comply with government and industry standards.

3. A reservoir intelligent inspection system based on a knowledge graph and digital twin according to claim 1, characterized in that The specific steps for introducing blockchain technology are as follows: S41: Data on-chain: When new data is added to the knowledge graph, its data hash value is written into the blockchain. S42: Verification mechanism: All updates or modifications to the cross-domain knowledge graph require consensus verification by multiple parties. S43: Encryption technology: The blockchain uses encryption algorithms to ensure that data can only be accessed and decrypted by authorized users. S44: Zero-knowledge proof: Allows one party to prove the authenticity of a certain piece of information to another party without disclosing the specific content of the information. S45: Distributed storage: Data is not stored on a single server or central institution but is distributed across multiple nodes. S46: Smart contract: Through smart contracts, data access and operations in the cross-domain knowledge graph are automatically managed and monitored according to predefined rules. S47: Permission management: Through the permission management system of the blockchain, a strict access control mechanism is set up to ensure that only authorized users can modify or access certain sensitive data.

Images