Search

CN-121981312-A - Emergency repair-oriented power optical cable space-time trajectory backtracking and impact analysis method

CN121981312ACN 121981312 ACN121981312 ACN 121981312ACN-121981312-A

Abstract

The invention relates to the technical field of safety and fault analysis of electric power systems, and discloses an emergency repair-oriented electric power optical cable space-time track backtracking and influence analysis method, which comprises the steps of constructing a heterogeneous optical cable database integrated with optical cable static attribute, dynamic operation data, equipment association relation and text description information; the method comprises the steps of receiving a query request input by a user through natural language, carrying out intelligent information retrieval from an optical cable database, returning a structured query result, constructing a dynamic graph structure, modeling the dynamic graph by utilizing a space-time graph neural network to predict the influence range of optical cable faults, the influenced probability of each device and the estimated interruption time length, and dynamically generating a priority sequence and a quantization index of a rush-repair task based on the prediction result, wherein the physical nodes and edges are defined, and reflective ghost nodes are generated based on physical boundaries. The invention obviously improves the response speed, decision scientificity and overall security and guarantee capability of the power communication network of the emergency repair.

Inventors

  • SHI XUETAO
  • WANG TAO
  • WANG ZHEN
  • XU ZHI
  • PENG FEI

Assignees

  • 国网江苏省电力有限公司淮安供电分公司

Dates

Publication Date
20260505
Application Date
20251217

Claims (10)

  1. 1. The power optical cable space-time trajectory backtracking and influence analysis method for emergency repair is characterized by comprising the following steps of: constructing a heterogeneous optical cable database integrated with optical cable static attribute, dynamic operation data, equipment association relation and text description information; Receiving a query request input by a user through natural language, performing intelligent information retrieval from the heterogeneous optical cable database after semantic understanding, intention recognition and entity linking, and returning a structured query result comprising an optical cable routing topology, an associated equipment list and a history fault record; defining physical nodes and edges, generating reflective ghost nodes based on physical boundaries, and converting space constraint into graph interaction relation; Carrying out space topology modeling and time dynamic modeling on the dynamic graph structure by utilizing a space-time graph neural network, and fusing space topology features and time dynamic features to predict the influence range of optical cable faults, the influenced probability of each device and the estimated interruption time length; based on the prediction result, a deep reinforcement learning framework is adopted, and a priority sequence and a quantization index of the rush repair task are dynamically generated with the aim of minimizing communication interruption time and preferentially guaranteeing core equipment.
  2. 2. The method of claim 1, wherein the process of generating reflective ghost nodes based on physical boundaries comprises: Identifying a physical boundary along an optical cable route from an electric power GIS system, and acquiring geometrical parameters and space orientation of the boundary; For the physical nodes of the optical cable section close to the physical boundary, generating corresponding ghost nodes according to the geometric reflection principle and the external normal vector reflection of the boundary, wherein the ghost nodes inherit the physical characteristics of the boundary; and establishing a unidirectional connection edge between the physical node and the ghost node meeting the distance condition so as to encode a space constraint relation in the dynamic graph structure.
  3. 3. The method of claim 2, wherein modeling the dynamic map using a space-time-map neural network comprises: space topology modeling is performed by using a space-time diagram neural network, the method specifically comprises the following steps: The method comprises the steps of adopting a graph neural network based on an attention mechanism, using LeakyReLU to activate a function to enhance nonlinearity when calculating attention scores, applying punishment coefficients to ghost nodes to reduce the weight of virtual nodes in feature aggregation and avoid excessive influence on real physical topology, and aggregating the features of the nodes and neighbor nodes thereof through a multi-layer perceptron so as to update the spatial embedded representation of the nodes.
  4. 4. A method according to claim 3, wherein the attention score is calculated using the formula: In the formula, Non-normalized attention scores between node v and neighbor node u; leakyReLU (35) a linear rectification activation function with leakage coefficient; node type penalty coefficient, if the node is a physical node, then =1, If the node is a ghost node =0.3; Query vector of node v; Key vector of node u; The characteristics of the node and the neighboring nodes thereof are aggregated through the multi-layer perceptron, so that the space embedded representation of the node is updated, and the formula is as follows: In the formula, Node v is at The updated feature vector in the layer space block comprises a space representation after neighbor feature aggregation and nonlinear mapping; Node v is at An input feature vector of the layer; Neighbor node u is at An input feature vector of the layer; a neighbor node set of a node v; Node v versus neighbor node Is a weight of attention of (2); the attention weight of the node v; multilayer perceptron.
  5. 5. A method according to claim 1 or 3, wherein the space-time diagram neural network adopts a gating fusion mechanism to balance contributions of spatial topological features and temporal dynamic features, the gating mechanism comprising a reset gate and an update gate for adaptively controlling the degree of retention of historical spatial features and the fusion ratio of temporal new features.
  6. 6. The method of claim 1, wherein when predicting the range of influence of the cable faults, performing level pooling on the node characteristics after time-space fusion to obtain global state vectors, and calculating the affected probability of each device through a classifier, and simultaneously introducing a constraint checking mechanism, and when the predicted fault propagation direction conflicts with the physical boundary constraint represented by the ghost node, performing downward correction on the affected probability.
  7. 7. The method of claim 4, wherein the reward function in the deep reinforcement learning framework is in a multi-objective weighted form, the formula being: Wherein, the For deciding moment of time Is a real-time prize value for (1), And As the weight coefficient of the light-emitting diode, For the maximum allowable interrupt duration, In order to estimate the remaining interrupt duration, The method comprises the steps of recovering an indication function for Core equipment, wherein Core is a Core equipment node set.
  8. 8. The method of claim 1, wherein the deep reinforcement learning trains a strategy network using a near-end strategy optimization algorithm, the strategy network taking as input features output by the spatio-temporal modeling step, outputting probabilities of selecting each rush-repair action by a Softmax function, and generating a priority list.
  9. 9. The method of claim 1, wherein receiving the query request entered by the user in natural language, performing intelligent information retrieval from the heterogeneous optical cable database after semantic understanding, intent recognition and entity linking, and returning a structured query result including an optical cable routing topology, a list of associated devices and a history of failure, comprises: converting the query text into high-dimensional semantic vectors using a pre-trained language model; Identifying the user query intention through a fine-tuned intention classifier; and correlating the named entity in the query with the optical cable and the equipment identifier in the database.
  10. 10. The method of claim 1, wherein the heterogeneous fiber optic cable database is stored using a distributed database and establishes a data quality assurance mechanism comprising cleaning and normalizing text data, processing unbalanced historical failure data using an oversampling technique, and performing incremental updates periodically to preserve data timeliness.

Description

Emergency repair-oriented power optical cable space-time trajectory backtracking and impact analysis method Technical Field The invention relates to the technical field of safety and fault analysis of electric power systems, in particular to an emergency repair-oriented power optical cable space-time trajectory backtracking and influence analysis method. Background The power communication network is a key infrastructure for dispatching, protecting and automatizing operation of a power system, and a core transmission medium, namely a power optical cable, bears high-reliability data transmission tasks among a multi-stage transformer substation, a dispatching center and a communication hub. With the improvement of urban underground space utilization rate and the complicacy of an optical cable laying environment, an optical cable path is often constrained by physical boundaries (such as building walls, underground pipeline outer walls, road isolation belts and the like), so that a topological structure is increased in a nonlinear manner, and space connectivity is complicated. In an emergency repair scene, three significant problems exist in the traditional optical cable operation and maintenance mode: 1. The existing optical cable topology analysis method generally assumes that an optical cable path is continuous and can pass through space barriers, and ignores the limitation of a physical boundary on a fault propagation direction and a repair path, so that the space prediction deviation is large and the simulation result is distorted; 2. The analysis precision is limited because the analysis means based on static topology or two-dimensional plane model can not capture the space-time dynamic characteristics (such as load fluctuation, boundary change and fault time sequence dependence) of the operation of the optical cable, and the response mechanism of the high-dimensional complex power communication network is lack of depicting capability; 3. the intelligence of the rush-repair decision is insufficient, the current decision process mainly depends on manual experience and fixed rules, a quantifiable multi-objective weighing mechanism is lacked, and the cooperative optimization of 'quick response and critical equipment priority guarantee' is difficult to realize. In view of the above, conventional Graph Neural Networks (GNNs) can deal with spatial dependence in part, but cannot cope with reflection constraints and nonlinear propagation paths of optical cables at physical boundaries. In addition, reinforcement Learning (RL) has potential in decision optimization, but its state features lack physical semantic support, and cannot fully utilize space-time coupling information of the optical cable. Therefore, a new method for integrating space geometric modeling, time sequence feature analysis and decision optimization is urgently needed to achieve space-time track backtracking, influence range prediction and automatic generation of rush repair priority of the power optical cable in a complex environment, and intelligent emergency response of a power communication network is truly supported. Disclosure of Invention Aiming at the problems of unmodeled space constraint, delayed influence analysis, insufficient decision optimization and the like in the conventional power cable emergency repair, the invention provides a power cable space-time trajectory backtracking and influence analysis method based on a reflection type ghost node and a space-time diagram neural network (STGNN), and the dynamic optimization of a repair strategy is realized through reinforcement learning. The embodiment of the invention provides a power optical cable space-time trajectory backtracking and influence analysis method for emergency repair, which comprises the following steps: constructing a heterogeneous optical cable database integrated with optical cable static attribute, dynamic operation data, equipment association relation and text description information; Receiving a query request input by a user through natural language, performing intelligent information retrieval from the heterogeneous optical cable database after semantic understanding, intention recognition and entity linking, and returning a structured query result comprising an optical cable routing topology, an associated equipment list and a history fault record; defining physical nodes and edges, generating reflective ghost nodes based on physical boundaries, and converting space constraint into graph interaction relation; Carrying out space topology modeling and time dynamic modeling on the dynamic graph structure by utilizing a space-time graph neural network, and fusing space topology features and time dynamic features to predict the influence range of optical cable faults, the influenced probability of each device and the estimated interruption time length; based on the prediction result, a deep reinforcement learning framework is adopted, and a priority sequence and a q