Intelligent sensing and identifying compilation system for power grid communication maintenance risk

Abstract

The invention discloses a power grid communication maintenance risk intelligent perception and identification compilation system, and belongs to the technical field of power grid maintenance analysis. The method is used for solving the technical problem of poor flexibility and reliability of active implementation of power grid communication maintenance risk identification compilation in an existing scheme. Through space-time convolution and an attention mechanism, association risks of nodes, links and topologies of a power grid communication network can be captured, continuous learning is carried out in a digital twinning environment by utilizing a reinforcement learning agent, online evolution of a maintenance strategy is realized, and the maintenance risk of the power grid communication network can be determined through a generated Pareto optimal solution. A high-risk suppression, high-resource input or balanced strategy can be flexibly selected according to a power grid operation scene, an output risk propagation path and a decision process of reinforcement learning are visualized, a Pareto optimal solution is automatically converted into an execution strategy by using an intelligent contract, and through a real-time feedback and deviation adjustment mechanism, the resource utilization rate can be effectively improved, and the system performance is improved. And the deviation between the actual value and the predicted value of the risk suppression ratio is reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power grid maintenance analysis, and particularly relates to an intelligent sensing and knowledge identification system for power grid communication maintenance risk. BACKGROUND

[0002] With the continuous development of the power system and the expansion of the power grid scale, the stable operation of the power grid communication equipment is crucial to ensure the reliability and safety of the power system.

[0003] Currently, there may be safety risks in the maintenance operation of the power communication system, and how to identify, evaluate and develop control measures through a systematic method is the current research direction.

[0004] The existing technical solutions are mostly still based on static modeling and hybrid identification schemes, and cannot implement diversified risk identification sensing and active preparation, and there are defects of poor flexibility and reliability of active implementation of power grid communication maintenance risk identification and preparation. SUMMARY

[0005] The present application relates to the technical field of power grid maintenance analysis, and particularly relates to an intelligent sensing and knowledge identification system for power grid communication maintenance risk.

[0006] The purpose of the present application can be achieved by the following technical solutions: An intelligent sensing and knowledge identification system for power grid communication maintenance risk, comprising: a three-dimensional dynamic risk feature map processing module: a dynamic simulation model of the power grid communication system based on digital twinning is constructed, the running state of the physical equipment, the topological connection relationship and the environmental interference parameters are mapped in real time, and a three-dimensional dynamic risk feature map containing the equipment health degree, the link transmission quality and the network topology vulnerability is generated; a maintenance strategy dynamic deduction output module: based on the three-dimensional dynamic risk feature map obtained by processing, a risk decision model integrating graph neural network and reinforcement learning is constructed, the propagation path of the topological correlation risk is mined through the graph neural network, the risk evolution process of different maintenance strategies is dynamically deduced in the digital twinning environment by the reinforcement learning agent, and a Pareto optimal solution set of risk suppression rate and resource consumption rate is output; a cross-regional maintenance resource collaborative scheduling module: according to the Pareto optimal solution set obtained by processing, a distributed maintenance scheme consensus mechanism based on block chain is constructed, the risk decision result is converted into a smart contract and synchronized to each regional maintenance node, the contract execution parameters are dynamically adjusted through real-time state feedback, and the adaptive collaborative scheduling of cross-regional maintenance resources is realized.

[0007] Preferably, when constructing a dynamic simulation model of the power grid communication system based on the physical entity-virtual mapping-data interaction three-layer architecture, device-level twin modeling, link-level twin modeling, and topology-level twin modeling are included.

[0008] Preferably, when implementing device-level twin modeling, a multi-body dynamics equation is used to describe the device aging process; when implementing link-level twin modeling, a link attenuation model is constructed based on transmission line theory; and when implementing topology-level twin modeling, the power grid communication network is converted into a directed weighted graph G=(V,E,W), V is the node set, E is the edge set, and W is the edge weight.

[0009] Preferably, the device health degree is obtained by fusing temperature, power consumption, and runtime; the link transmission quality is obtained by fusing optical signal-to-noise ratio, packet loss rate, and time delay; and the network topology vulnerability is calculated based on betweenness centrality.

[0010] Preferably, the last layer of the improved spatio-temporal graph convolution network outputs the risk contribution degree of each node, the risk contribution degrees of the nodes are sorted, the key propagation nodes are identified, and a risk propagation path set is generated.

[0011] Preferably, according to the risk propagation path in the risk propagation path set, a reinforcement learning agent is trained in the digital twin environment to realize dynamic deduction of the maintenance strategy, and a state space, an action space, and a reward function are designed.

[0012] Preferably, based on the policy output of the reinforcement learning agent, the risk evolution process is deduced in the digital twin environment, and the Pareto optimal solution is generated through multi-objective optimization, the maintenance strategy output by the reinforcement learning agent is input, and the digital twin environment simulates the device state update, link quality recovery, and topology risk diffusion.

[0013] Preferably, when converting the Pareto optimal solution set into executable smart contract logic, the risk suppression rate and resource consumption rate in the Pareto optimal solution set are obtained through the risk decision module in the smart contract, and the target solution is selected through a preset rule; the resource scheduling module in the smart contract implements resource scheduling according to the defined resource allocation rule.

[0014] Preferably, when the risk level of any region in the three-dimensional risk feature map is greater than or equal to M, it is determined as high risk, and the smart contract is automatically called; M is a level threshold; When executing the smart contract, input the Pareto optimal solution set, the region importance weight, and the current total amount of resources; The target solution is selected through the risk decision module, and the resource scheduling module is called to allocate personnel and spare parts; A structured maintenance work order is generated and synchronized to the corresponding regional verification node.

[0015] Preferably, the smart contract calculates a deviation rate according to the feedback data, analyzes the deviation rate, and automatically adjusts the resource allocation parameters.

[0016] Compared with the prior art, the present application has the following beneficial effects: The present application can solve the hysteresis of traditional static risk assessment through mapping of the digital twin model and real-time data, and realize second-level updating of risk characteristics;The risk graph is constructed from three dimensions of equipment, link and topology, covering all elements of points, lines and surfaces of the power grid communication system, realizing multi-dimensional data fusion and avoiding one-sidedness of single index evaluation;By introducing the aging model and topology vulnerability calculation mechanism, the prediction error of equipment health degree can be effectively reduced, the link fault early warning accuracy can be improved, and high credibility input can be provided for subsequent risk decision-making.

[0017] The present application can capture the associated risks of power grid communication network nodes, links and topologies through spatio-temporal convolution and attention mechanism, thereby effectively improving the identification accuracy of propagation path;The reinforcement learning agent continuously learns in the digital twin environment, realizes online evolution of the maintenance strategy, and can effectively shorten the response speed when facing sudden failures;The Pareto optimal solution generated by the NSGA-III algorithm can flexibly select high-risk suppression, high-resource investment or balanced strategy according to the power grid operation scene, which can effectively improve resource utilization and reduce risk misjudgment rate;The risk propagation path and the decision-making process of reinforcement learning can be visualized, which can provide clear risk source, propagation chain and control node logic for maintenance personnel, and can effectively solve the decision-making trust problem of traditional black box model.

[0018] The present application can reduce manual intervention and effectively shorten the scheme generation time by automatically converting the Pareto optimal solution into an execution strategy through the smart contract;Through the real-time feedback and deviation adjustment mechanism, the resource utilization can be effectively improved, and the deviation between the actual value and the predicted value of the risk suppression rate can be reduced. BRIEF DESCRIPTION OF DRAWINGS

[0019] The present application will be further described below with reference to the accompanying drawings.

[0020] Figure 1 The flowchart of the operation of the power grid communication maintenance risk intelligent sensing and identification system of the present application is shown. DETAILED DESCRIPTION

[0021] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0022] As Figure 1 shown, the application is an intelligent risk identification and preparation system for power grid communication maintenance, comprising: A three-dimensional dynamic risk feature map processing module: a dynamic simulation model of the power grid communication system is constructed based on digital twinning, the running state, topological connection relationship and environmental interference parameters of the physical equipment are mapped in real time, and a three-dimensional dynamic risk feature map containing equipment health degree, link transmission quality and network topology vulnerability is generated; the specific steps include: Deploy multiple types of sensors on the core equipment of the power grid communication, including but not limited to optical transceiver, router and switch, collect electrical parameters, optical transmission parameters, environmental parameters and topological correlation data; Among them, the electrical parameters include device input voltage, working current and power consumption; The optical transmission parameters include optical module transmission power, receiving sensitivity and optical signal-to-noise ratio; The environmental parameters include the temperature, humidity and electromagnetic interference intensity in the equipment cabin; The topological correlation data is obtained in real time through the SNMP protocol, and records the node IP, port state and link bandwidth utilization; The collected multi-source data is preprocessed, including but not limited to filtering abnormal values using 3σ criterion, such as determining that the voltage exceeding the rated value ±10% is abnormal; missing values are filled using the spatiotemporal interpolation method, such as using the sliding average of the previous 5 sampling points in the time dimension and the weighted average of the same type parameters of adjacent devices in the spatial dimension; wavelet denoising is performed on the time series data, such as selecting db4 wavelet basis, 3 layers of decomposition, and soft threshold function for threshold function to suppress high-frequency noise caused by electromagnetic interference; The implemented preprocessing is an existing conventional technical solution, and the specific implementation steps are not described here; When constructing the dynamic simulation model of the power grid communication system based on the three-layer architecture of physical entity-virtual mapping-data interaction, it includes device-level twinning modeling, link-level twinning modeling and topology-level twinning modeling; Among them, when implementing device-level twinning modeling, the multi-body dynamics equation is used to describe the aging process of the device, and its life decay model is: ; wherein, is the remaining life of the device at time t; is the initial life, in hours; k is the aging coefficient, with a value range of 0.001~0.01, α is the temperature influence index, β is the power consumption influence index, α value range is 1.5~2.0, β value range is 0.8~1.2, which is obtained by fitting historical failure data; is the working temperature at time τ, in degrees Celsius; is the working power consumption at time τ, in watts; The 3D modeling tool used is SolidWorks. The equipment CAD drawings are imported and material properties are assigned. Material properties include, but are not limited to, the thermal conductivity of the outer shell and the dielectric constant of the circuit board, so as to achieve an accurate mapping between the physical appearance and the internal structure. When implementing link-level twin modeling, a link attenuation model is constructed based on transmission line theory. The formula for calculating the signal attenuation A of the optical fiber link is as follows: ;in, Fixed connector loss, unit is decibel; The fiber attenuation coefficient; Link length, in kilometers; This is the temperature influence coefficient, with a value ranging from 0.005 to 0.02. Average temperature; This is the bandwidth utilization impact coefficient, with a value range of 0.005 to 0.01; Link bandwidth utilization, expressed as a percentage. When implementing topology-level twin modeling, the power grid communication network is transformed into a directed weighted graph G=(V,E,W), where V is the node set, E is the edge set, and W is the edge weight, where: Node set V: contains devices and terminals. Node attributes include device type, IP address, and health status. Devices include, but are not limited to, routers and switches. Terminals include, but are not limited to, protection devices and monitoring and control units. Edge set E: represents a physical link, such as an optical fiber or cable; The edge weight W represents the link transmission quality, a normalized value that combines packet loss rate, latency, and jitter, ranging from 0 to 1. The Neo4j graph database is used to store topology relationships, and it supports dynamic addition and deletion of nodes and edges. For example, when a new link is added, the adjacency matrix is ​​automatically updated. When performing multi-source data fusion, the preprocessed real-time data and historical fault data are spatiotemporally aligned. A sliding time window is used to extract time-varying features. The window size is 10 and the step size is 5, which includes the equipment temperature fluctuation rate and the link bandwidth fluctuation coefficient. Among them, the equipment temperature fluctuation rate: ;in, These are the maximum operating temperature and the minimum operating temperature, respectively. Link bandwidth fluctuation coefficient: ;in, The standard deviation of bandwidth utilization; Instantaneous bandwidth utilization; This represents the time average of bandwidth utilization. Furthermore, dynamic weights are assigned to device, link, and environmental data based on an attention mechanism, with the weight calculation formula as follows: ;in, For the first Types of data, such as temperature and optical power; This is a sequence of historical fault labels; λ is the Pearson correlation coefficient; λ is the adjustment factor, ranging from 5 to 10, used to enhance the weight of key features. Index the target element; n is the index of the element being iterated over; n is the total number of elements. When performing dynamic feature extraction, the device health score H is obtained by fusing temperature, power consumption, and runtime, using the following formula: ;in, These are the maximum and minimum allowable temperatures for the equipment, respectively. The rated power of the equipment; This refers to the continuous runtime. By fusing the optical signal-to-noise ratio, packet loss rate L, and latency D, the link transmission quality Q is obtained, as shown in the formula: ;in, These are the maximum and minimum signal-to-noise ratios allowed by the device, respectively. This represents the actual signal-to-noise ratio. The network topological vulnerability V is calculated based on betweenness centrality, using the following formula: Where N is the total number of network nodes; These are the source node and the target node, respectively; v represents a node. This represents the distance from the source node s to the target node. The number of paths that must pass through node v among all the shortest paths; From source node s to target node The total number of all shortest paths, regardless of whether they pass through node v; When generating a three-dimensional dynamic risk feature map, the X-axis represents the equipment dimension, including the equipment health level H, with a value range of 0 to 1. It is divided into four levels—healthy, attention, warning, and dangerous—at intervals of 0.2, with corresponding value ranges of 0.8 to 1, 0.6 to 0.8, 0.4 to 0.6, and 0 to 0.4, respectively. The Y-axis represents the link dimension, including the link transmission quality Q, with a value ranging from 0 to 1. It is weighted in conjunction with the link importance, for example, the backbone link weight is 1.2, and the branch link weight is 0.8. The Z-axis represents the topology dimension, including the network topology vulnerability V, which ranges from 0 to 1, and is superimposed with the cascading impact coefficient of node failures, for example, the impact range of a core node failure × 1.5. A 3D visualization interface is built using WebGL technology, with color intensity representing risk level. For example, red to green indicates high risk level to low risk level, and node size indicates equipment importance. New data is received every 5 minutes, and the feature map is updated through incremental learning algorithms, such as Finetune to fine-tune model parameters, with an update delay of ≤2 seconds, to ensure synchronization with the physical system state.

[0023] In this embodiment of the invention, by mapping digital twin models with real-time data, the lag in traditional static risk assessment can be solved, enabling second-level updates of risk characteristics. A risk map is constructed from three dimensions: equipment, links, and topology, covering all elements of the power grid communication system, including points, lines, and surfaces, achieving multi-dimensional data fusion and avoiding the one-sidedness of single-indicator assessment. By introducing aging models and topology vulnerability calculation mechanisms, the prediction error of equipment health can be effectively reduced, and the accuracy of link fault early warning can be improved, providing highly reliable input for subsequent risk decisions.

[0024] The maintenance strategy dynamic simulation output module: Based on the processed and acquired 3D dynamic risk feature map, a risk decision model integrating graph neural networks and reinforcement learning is constructed. The graph neural network mines the propagation paths of topologically related risks, and the reinforcement learning agent dynamically simulates the risk evolution process of different maintenance strategies in a digital twin environment, outputting the Pareto optimal solution set for risk suppression rate and resource consumption rate. Specific steps include: When constructing a risk decision-making model that integrates graph neural networks and reinforcement learning, the three-dimensional risk feature map is transformed into node feature vectors X=[H,Q,V], and the edge weight E is defined as the risk propagation coefficient. The relevant expressions are as follows: ;in, For link bandwidth utilization, Let be the physical distance between nodes i and j. These are all weighting coefficients, with values ​​of 0.8 and 0.2 respectively; ω is the backbone link enhancement factor, with a default value of 1.5, and the default value of the enhancement factor for ordinary links is 1; The spatial convolutional layer of the improved spatiotemporal graph convolutional network uses an attention mechanism to dynamically adjust the influence weights of adjacent nodes, as shown in the formula: ;in, Let v be the feature vector of node v in the l-th layer; Let be the set of adjacent nodes; 'a' be the attention vector. Here, σ is the learnable weight matrix; σ is the ReLU activation function. The temporal convolutional layer uses dilated causal convolution to capture long temporal dependencies, with a kernel size of 3×3 and a dilation rate of [missing information]. , As a time step, it covers the risk evolution sequence over the past 24 hours; By analyzing the risk contribution of the output nodes of the last layer of the improved spatiotemporal graph convolutional network, the PageRank algorithm is used to rank the node risk contribution, identify key propagation nodes such as core routers and aggregation switches, and generate a set of risk propagation paths. m represents the total number of risk propagation paths, and each risk propagation path includes a node sequence and a propagation probability; Among them, the output of the last layer of the improved spatiotemporal graph convolutional network is a node embedding vector, which contains the risk propagation characteristics of the node in the spatiotemporal dimension; The high-dimensional embedding vector of the last layer is passed through a fully connected layer. Mapped to scalar values For all nodes Normalization is performed, for example, Min-Max normalization to [0,1], to obtain the final node risk contribution. The larger the node risk contribution value, the stronger the node's contribution to risk propagation; PageRank is a graph theory-based algorithm that treats the internet as a directed graph, with web pages as nodes and hyperlinks as edges. It measures the importance of each node by calculating its steady-state probability. Based on network topology and risk propagation direction, a directed risk propagation graph is constructed. The risk contribution of each node is used as the initial importance score for PageRank. PageRank iterative calculations are then performed on the directed risk propagation graph, involving the following expressions: ;in, The PageRank score for the target node v; This is the damping coefficient, with a default value of 0.85, indicating that the node... The probability of risk propagating along the edge, with 1- Probability-based random redirection; Let v be the set of incoming neighbors of the incoming neighbor node v; Let k be the set of outgoing neighbors of node u; k is the set of outgoing neighbor nodes. For all nodes Sort in descending order and select the top K% of nodes as key propagation nodes; for example, select the top 5% of nodes as key propagation nodes. The risk propagation path set consists of a node sequence and a propagation probability. The node sequence is the propagation path, which is generated through path search and probability calculation. Among them, the generation of node sequences is based on key propagation nodes and a directed graph of risk propagation, and possible propagation paths are generated through source node selection, path search, deduplication and pruning. When implementing source node selection, the initial risk trigger node, such as a node that has already failed, or one of the top 1% of critical propagation nodes, should be used as the starting point of the path. ; When implementing path search, it can be based on greedy search and / or breadth-first search, which are existing conventional technical solutions. The specific implementation steps will not be elaborated here. When performing deduplication and pruning, duplicate paths and low-probability paths with a propagation probability lower than the propagation threshold are removed. The propagation threshold is set to 0.1 by default, resulting in the final node sequence. When calculating the propagation probability, the propagation probability of each path is the product of the single-step propagation probabilities of each edge on the path; Among them, when calculating the single-step propagation probability, the side The propagation probability is obtained from the node dependencies learned by the improved spatiotemporal graph convolutional network or from historical data statistics; for example, based on historical fault propagation records, the edge propagation probability is statistically analyzed. After the malfunction occurred Conditional probability of failure ; Based on the risk propagation paths in the risk propagation path set, a reinforcement learning agent is trained in a digital twin environment to realize the dynamic deduction of maintenance strategies. The state space, action space and reward function are designed. The state space is: ; These are the device health vector, link transmission quality vector, topology vulnerability vector, and resource remaining vector, respectively. The dimensions corresponding to the device health vector, link transmission quality vector, and topology vulnerability vector are the total number of devices, the total number of links, and the total number of nodes, respectively. The resource remaining vector includes the number of personnel, spare parts, and tools. Action space This is a combination of maintenance strategies, including the selection of maintenance targets and resource allocation schemes; Among them, the selection of maintenance targets: select the top b nodes with the highest priority from the risk propagation path set P, where b≤5, to avoid resource dispersion; Resource allocation plan: Allocate a number of personnel, a number of spare parts, and a time window to each selected node; the number of personnel can be 1 to 2, the number of spare parts can be 0 to 1 set, and the time window can be 1 to 4 hours; reward function A multi-objective weighted reward system is adopted to balance risk mitigation and resource consumption. The relevant expression is: ;in, These are all reward weighting coefficients, with values ​​of 1, 0.3, and 0.2 respectively; Risk mitigation rate is the percentage or absolute amount by which the risk level is reduced after implementing risk mitigation measures. The total value of resource consumption includes the economic costs, hardware resources, and / or human resources consumed in implementing risk mitigation measures. Before calculation, it needs to be standardized to a dimensionless value in the range [0,1]. For time delays, the system response delays or business interruption times caused by the implementation of measures, the learning also needs to be standardized to dimensionless values ​​in the [0,1] interval; A deep deterministic strategy is used to process the continuous action space with gradients. The Actor network outputs the action probability distribution, and the Critic network evaluates the state-action value. Experience replay pool storage The sample size is 1 million records, and priority sampling is used. Simulate 1,000 fault scenarios in a digital twin environment, such as single node failure, link congestion, and regional power outage, and train iteratively for 5 million steps to ensure policy generalization. Based on the policy output of the reinforcement learning agent, the risk evolution process is simulated in the digital twin environment. When generating the Pareto optimal solution through multi-objective optimization, the maintenance policy output by the reinforcement learning agent is input, and the digital twin environment simulates equipment status updates, link quality recovery and topology risk diffusion. Among them, when simulating equipment status updates, the health status of the equipment after maintenance is obtained. Wherein, ΔH is the change in health, which is positively correlated with the maintenance duration; for example, ΔH = 0.3 for a 2-hour maintenance. When simulating link quality recovery, obtain transmission quality. ; When simulating topological risk diffusion, if critical nodes are not inspected, the risk spreads along the risk propagation path set P, and the diffusion speed is... ; Multi-objective optimization includes maximizing risk mitigation rate and minimizing resource consumption rate; The NSGA-III algorithm is used to solve the problem, with a population size of 200 and 100 generations. The diversity of the solution set is maintained by crowding distance. NSGA-III is an advanced multi-objective evolutionary algorithm. By introducing a reference point mechanism, it solves the problem of insufficient diversity of NSGA-II in high-dimensional objective space. It is particularly suitable for handling complex optimization problems with three or more objectives and is one of the mainstream algorithms in the field of high-dimensional multi-objective optimization. Output Pareto optimal solution set Each solution corresponds to a combination of risk inhibition rate and resource consumption rate; When implementing decision recommendations, the optimal solution is dynamically selected based on the power grid operation scenario: In emergency scenarios, such as extremely high risks, solutions with a risk mitigation rate ≥ 0.9 are preferred, while resource consumption rate ≤ 0.8 is allowed. In typical scenarios, an equilibrium solution with a risk suppression rate ≥ 0.7 and a resource consumption rate ≤ 0.5 is selected, and decision recommendations are displayed through a visual interface.

[0025] In this embodiment of the invention, spatiotemporal convolution and attention mechanisms can capture the associated risks of power grid communication network nodes, links, and topologies, thereby effectively improving the accuracy of propagation path identification. Utilizing reinforcement learning agents to continuously learn in a digital twin environment enables online evolution of maintenance strategies, effectively shortening the response time to sudden faults. The Pareto optimal solution generated by the NSGA-III algorithm allows for flexible selection of high-risk suppression, high-resource investment, or balanced strategies based on the power grid operation scenario, effectively improving resource utilization and reducing risk misjudgment rate. The visualization of the output risk propagation path and reinforcement learning decision-making process provides maintenance personnel with clear risk sources, propagation chains, and control node logic, effectively solving the decision trust problem of traditional black-box models.

[0026] Cross-regional maintenance resource collaborative scheduling module: Based on the Pareto optimal solution set obtained through processing, a blockchain-based distributed maintenance solution consensus mechanism is constructed. Risk decision results are transformed into smart contracts and synchronized to maintenance nodes in each region. Contract execution parameters are dynamically adjusted through real-time status feedback to achieve adaptive collaborative scheduling of cross-regional maintenance resources. Specific steps include: When transforming the Pareto optimal solution set into executable smart contract logic, the risk mitigation rate in the Pareto optimal solution set is obtained through the risk decision module in the smart contract. With resource consumption rate The target solution is selected according to preset rules, and the relevant expression is: ;in, is the scenario coefficient, with a value of 0.8 for emergency scenarios and 0.5 for normal scenarios; S is the Pareto optimal solution set of the output. The resource scheduling module in a smart contract performs resource scheduling according to defined resource allocation rules, involving the following expressions: ;in, The importance weight of a region is determined by the topological vulnerability V of the power grid, for example, 1.2 for a core region and 0.8 for a general region. For the area's number index; For traversal index of the region; This represents the current total amount of resources; When the risk level of any region in the 3D risk feature map is ≥ M, it is judged as high risk and the smart contract is automatically invoked; M is the level threshold, and the default value is 0.7; When executing a smart contract, input the Pareto optimal solution set, the regional importance weights, and the current total resources; The target solution is selected through the risk decision-making module, and the resource scheduling module is invoked to allocate personnel and spare parts. Generate structured maintenance work orders, including equipment ID, maintenance time window, and resource list, and synchronize them to the corresponding regional verification nodes; Furthermore, it adopts an off-chain data storage plus on-chain hash anchoring mode, with maintenance work order details stored in the InterPlanetary File System, and only the hash value recorded on the chain, reducing the storage pressure on the ledger; Smart contract code undergoes formal verification, such as with tools like Coq, to ​​prevent logical vulnerabilities, and must be authorized by digital signatures from all verification nodes before deployment; The maintenance scheme is based on an improved practical Byzantine fault-tolerant algorithm to achieve distributed consensus. This is an existing conventional technical solution, and the specific implementation steps will not be elaborated here. The smart contract calculates the deviation rate based on feedback data, analyzes the deviation rate, and automatically adjusts resource allocation parameters. The relevant expression is: ;in, The deviation rate; These are the actual risk reduction rate and the predicted risk reduction rate, respectively. like This will trigger resource addition: ;in, These are the resources after the addition and the resources before the addition, respectively. like This will trigger resource reclamation: .

[0027] In this embodiment of the invention, smart contracts are used to automatically transform Pareto optimal solutions into execution strategies, which can reduce manual intervention and effectively shorten the solution generation time. Through real-time feedback and deviation adjustment mechanisms, resource utilization can be effectively improved and the deviation between the actual and predicted values ​​of risk suppression rates can be reduced. Through the coordinated efforts of the above steps, dynamic simulation and distributed decision-making are achieved. The technical approach covers innovative technologies such as digital twins, graph neural networks, and blockchain, and deeply integrates the core characteristics of power grid communication, such as topological correlation, real-time performance, and distributed management, which can effectively solve the problems of flexibility and reliability in proactive implementation.

[0028] In the several embodiments provided by this invention, it should be understood that the disclosed system can be implemented in other ways. For example, the embodiments of the invention described above are merely illustrative; for example, the division of modules is only a logical functional division, and there may be other division methods in actual implementation.

[0029] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0030] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated module can be implemented in hardware or in the form of hardware plus software functional modules.

[0031] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the essential characteristics of the present invention.

[0032] Finally, it should be noted that the above 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 preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A power grid communication maintenance risk intelligent perception and identification system, characterized in that, include: The 3D dynamic risk feature map processing module is used to construct a dynamic simulation model of the power grid communication system based on digital twins. By mapping the operating status of physical equipment, topological connection relationship and environmental interference parameters in real time, it generates a 3D dynamic risk feature map that includes equipment health, link transmission quality and network topology vulnerability. The maintenance strategy dynamic deduction output module is used to construct a risk decision model that integrates graph neural networks and reinforcement learning based on the three-dimensional dynamic risk feature map obtained through processing. The graph neural network is used to mine the propagation path of topologically related risks, and the reinforcement learning agent is used to dynamically deduce the risk evolution process of different maintenance strategies in a digital twin environment, and output the Pareto optimal solution set of risk suppression rate and resource consumption rate. The cross-regional maintenance resource collaborative scheduling module is used to construct a blockchain-based distributed maintenance solution consensus mechanism based on the Pareto optimal solution set obtained through processing. It transforms the risk decision results into smart contracts and synchronizes them to maintenance nodes in each region. Through real-time status feedback, it dynamically adjusts the contract execution parameters to achieve adaptive collaborative scheduling of cross-regional maintenance resources.

2. The intelligent perception and identification system for power grid communication maintenance risks according to claim 1, characterized in that, When constructing a dynamic simulation model of a power grid communication system based on a three-layer architecture of physical entity-virtual mapping-data interaction, it includes equipment-level twin modeling, link-level twin modeling, and topology-level twin modeling.

3. The intelligent perception and identification system for power grid communication maintenance risks according to claim 2, characterized in that, When implementing equipment-level twin modeling, multibody dynamics equations are used to describe the equipment aging process; when implementing link-level twin modeling, a link attenuation model is constructed based on transmission line theory; when implementing topology-level twin modeling, the power grid communication network is transformed into a directed weighted graph G=(V,E,W), where V is the set of nodes, E is the set of edges, and W is the edge weight.

4. The intelligent perception and identification system for power grid communication maintenance risks according to claim 1, characterized in that, Device health is obtained by integrating temperature, power consumption, and runtime; link transmission quality is obtained by integrating optical signal-to-noise ratio, packet loss rate, and latency; and network topology vulnerability is calculated based on betweenness centrality.

5. The intelligent perception and identification system for power grid communication maintenance risks according to claim 4, characterized in that, By ranking the risk contribution of the last layer output nodes of the improved spatiotemporal graph convolutional network, key propagation nodes are identified, and a set of risk propagation paths is generated.

6. The intelligent sensing and identification system for power grid communication maintenance risks according to claim 5, characterized in that, Based on the risk propagation paths in the risk propagation path set, a reinforcement learning agent is trained in a digital twin environment to realize the dynamic deduction of maintenance strategies. The state space, action space and reward function are designed.

7. The intelligent perception and identification system for power grid communication maintenance risks according to claim 6, characterized in that, Based on the policy output of the reinforcement learning agent, the risk evolution process is simulated in the digital twin environment. When generating the Pareto optimal solution through multi-objective optimization, the maintenance policy output by the reinforcement learning agent is input, and the digital twin environment simulates equipment status updates, link quality recovery, and topology risk diffusion.

8. The intelligent sensing and identification system for power grid communication maintenance risks according to claim 7, characterized in that, When transforming the Pareto optimal solution set into executable smart contract logic, the risk suppression rate and resource consumption rate in the Pareto optimal solution set are obtained through the risk decision module in the smart contract, and the target solution is selected according to preset rules; the resource scheduling module in the smart contract implements resource scheduling according to the defined resource allocation rules.

9. The intelligent sensing and identification system for power grid communication maintenance risks according to claim 8, characterized in that, When the risk level of any region in the three-dimensional risk feature map is ≥ M, it is judged as high risk and the smart contract is automatically invoked; M is the level threshold. When executing a smart contract, input the Pareto optimal solution set, the regional importance weights, and the current total resources; The target solution is selected through the risk decision-making module, and the resource scheduling module is invoked to allocate personnel and spare parts. Generate structured maintenance work orders and synchronize them to the corresponding regional verification nodes.

10. The intelligent sensing and identification system for power grid communication maintenance risks according to claim 9, characterized in that, The smart contract calculates the deviation rate based on feedback data, analyzes the deviation rate, and automatically adjusts the resource allocation parameters.

Images