Risk identification method based on low-dimensional vector calculation and power grid topology thought chain fusion

Abstract

The application discloses a risk identification method based on low-dimensional vector calculation and power grid topology thought chain fusion, relates to the power system early warning technical field, and comprises the following steps: acquiring multi-source heterogeneous sensing data of a power transmission channel area; the multi-source heterogeneous sensing data comprises time-stamped visible light or infrared images, and synchronously acquired power grid real-time topology structure snapshots and electrical operation state data; performing space-time alignment on the multi-source heterogeneous sensing data to form paired samples of risk perception events and power grid operation states; constructing a dynamic time series diagram based on power grid topology structure snapshots at continuous multiple time steps, and generating dynamic topology embedding vectors of power grid nodes through a time series diagram neural network; performing risk target detection on the visible light or infrared images, extracting visual semantic feature vectors, and mapping the detected targets to corresponding power grid nodes in the dynamic time series diagram; and fusing the visual semantic feature vectors with the dynamic topology embedding vectors of the mapped power grid nodes.

Description

Technical Field

[0001] This invention relates to the field of power system early warning technology, and in particular to a risk identification method based on low-dimensional vector computation and the fusion of power grid topology thinking chain. Background Technology

[0002] Power system early warning technology refers to a class of technical means that uses real-time monitoring and analysis of the power grid's operating status, equipment health status, and external environmental factors to identify potential faults or safety risks in advance and issue warning information before an accident occurs. Its core objective is to improve the power grid's observability, predictability, and proactive defense capabilities. Typical applications include transient stability early warning, voltage instability early warning, line overload early warning, and equipment defect early warning. It usually integrates methods such as state estimation, dynamic simulation, big data analysis, and artificial intelligence to provide decision support for dispatching and operation personnel, prevent the escalation of accidents, and ensure the safe and stable operation of the power system.

[0003] Existing methods are mostly based on static power grid topology or ignore real-time operation modes, which cannot reflect the impact of dynamic operating conditions such as switch operation, line maintenance, and N-1 on risk level. This leads to missed risk reporting under high load or weak grid structure, or false risk reporting on outage lines. In addition, the mainstream deep learning models are black box structures, only outputting risk probabilities, and cannot explain why the risk exists or what consequences it may cause. They are difficult for dispatchers to trust and cannot generate handling suggestions that comply with power regulations. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a risk identification method based on low-dimensional vector computation and the fusion of power grid topology thinking chain to solve the problems of existing methods that are mostly based on static power grid topology or ignore real-time operation mode, and cannot reflect the impact of dynamic operating conditions such as switching operation, line maintenance, and N-1 on risk level. This leads to missed risk reporting under high load or weak grid structure, or false risk reporting on outage lines. In addition, the mainstream deep learning model is a black box structure, which only outputs risk probability and cannot explain why the risk exists or what consequences it may cause. It is difficult for dispatchers to trust it and cannot generate handling suggestions that comply with power regulations.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a risk identification method based on low-dimensional vector computation and the fusion of power grid topology thinking chain, which includes:

[0008] Acquire multi-source heterogeneous sensing data of the power transmission channel area, including timestamped visible light or infrared images as well as synchronously acquired real-time power grid topology snapshots and electrical operation status data;

[0009] Spatiotemporal alignment of multi-source heterogeneous sensing data is performed to form paired samples of risk sensing events and power grid operation status;

[0010] A dynamic time-series graph is constructed based on snapshots of the power grid topology at multiple consecutive time steps, and dynamic topology embedding vectors of power grid nodes are generated through a time-series graph neural network.

[0011] Risk target detection is performed on visible light or infrared images, visual semantic feature vectors are extracted, and the detected targets are mapped to the corresponding power grid nodes in a dynamic time series graph;

[0012] The visual semantic feature vector is fused with the dynamic topological embedding vector of the mapped power grid node, and the causal rule constraint of the power system is introduced to generate a causal-enhanced low-dimensional joint risk vector.

[0013] The current risk level is predicted based on the low-dimensional joint risk vector with causal enhancement, and structured risk warning information is generated according to the preset risk level threshold.

[0014] Trajectory curvature analysis is performed on the causal enhanced low-dimensional joint risk vector sequence within a continuous time window. When the trajectory curvature exceeds the adaptive threshold determined based on the statistical distribution of historical normal operating conditions, a risk evolution mutation early warning signal is generated as a supplementary early warning to the structured risk warning information.

[0015] As a preferred embodiment of the risk identification method based on low-dimensional vector computation and power grid topology thinking chain fusion described in this invention, the specific steps for performing spatiotemporal alignment of multi-source heterogeneous sensing data to form paired samples of risk sensing events and power grid operating states are as follows:

[0016] Receive risk target detection results reported by the intelligent monitoring device, the risk target detection results including target category, geographic coordinates and first timestamp. ;

[0017] Obtain a sequence of power grid topology snapshots updated at fixed periods from the power grid dispatch automation system. Each snapshot contains a second timestamp and the corresponding switch status and electrical quantity.

[0018] Search for the first timestamp in the snapshot sequence of the power grid topology. The snapshot with the smallest absolute time difference is denoted by its timestamp. The topology is Electrical operating status is ;

[0019] The risk target detection results are compared with the topology. and electrical operating status Binding forms paired samples.

[0020] As a preferred embodiment of the risk identification method based on low-dimensional vector computation and power grid topology thinking chain fusion described in this invention, the specific steps of constructing a dynamic time series graph based on power grid topology snapshots of multiple consecutive time steps and generating dynamic topology embedding vectors of power grid nodes through a time series graph neural network are as follows:

[0021] Select by timestamp The endpoint is and the length is A sequence of snapshots of continuous power grid topology;

[0022] For each snapshot, a graph representation is constructed, where the set of vertices corresponds to the physical nodes in the transmission line, the set of edges is determined by the current switching state to determine whether they are connected, and the node feature matrix includes voltage magnitude, current intensity, and load rate.

[0023] The graph sequence is input into a temporal graph convolutional network, and the temporal information is propagated layer by layer through a gated graph update mechanism to output the time sequence. The dynamic topological embedding vector of each vertex;

[0024] Let the first The vertex at time The embedding is Its update process satisfies:

[0025] ;

[0026] in, Represents vertices At any moment The neighborhood group, and For learnable weight matrix, It is a non-linear activation function;

[0027] go through After the recursive steps, the final dynamic topological embedding vector is obtained. .

[0028] As a preferred embodiment of the risk identification method based on low-dimensional vector computation and power grid topology thinking chain fusion described in this invention, the specific steps of detecting risk targets in visible light or infrared images, extracting visual semantic feature vectors, and mapping the detected targets to the corresponding power grid nodes in the dynamic time series diagram are as follows:

[0029] Forward inference is performed on the input image, and the bounding boxes, class confidence scores, and deep feature maps of the risk targets are output through a lightweight convolutional neural network.

[0030] Visual semantic feature vectors are extracted from the deep feature map using global average pooling. ;

[0031] Calculate the Euclidean distance based on the geographical coordinates of the risk target and the pre-stored tower coordinate library;

[0032] Determine the tower index with the minimum distance, map the risk target to the corresponding vertex in the dynamic time series graph, and obtain the dynamic topological embedding vector of that vertex. .

[0033] As a preferred embodiment of the risk identification method based on low-dimensional vector computation and power grid topology thinking chain fusion described in this invention, the specific steps of fusing the visual semantic feature vector with the dynamic topology embedding vector of the mapped power grid node, and introducing power system causal rule constraints to generate a causally enhanced low-dimensional joint risk vector are as follows:

[0034] Visual semantic feature vectors With dynamic topological embedding vectors Concatenate them into a joint vector;

[0035] The joint vector is input into a multilayer perceptron, which outputs an initial low-dimensional risk representation vector. ;

[0036] Define the loss of causal intervention consistency, including:

[0037] Let the actual risk level be Risk causal variables are The model predicts the distribution as follows: The intervention distribution derived from the causal model is as follows The causal loss is:

[0038] ;

[0039] in, This represents the Kullback-Leibler divergence.

[0040] As a preferred embodiment of the risk identification method based on low-dimensional vector computation and power grid topology thinking chain fusion described in this invention, the following steps are taken: Trajectory curvature analysis is performed on the causally enhanced low-dimensional joint risk vector sequence within a continuous time window. When the trajectory curvature exceeds an adaptive threshold determined based on the statistical distribution of historical normal operating conditions, a risk evolution abrupt change warning signal is generated.

[0041] Maintain a first-in-first-out queue to store the most recently accessed items. The causal enhancement low-dimensional joint risk vector at each time step is denoted as the sequence. ;

[0042] Fit local cubic spline curves to the sequence in chronological order. ,in, This is the normalized time parameter;

[0043] Calculate the curve at the latest moment Principal curvature at =1 Its expression is:

[0044] ;

[0045] in, Represents curve right The first derivative reflects the instantaneous direction of change in the risk trajectory; It represents the second derivative, reflecting the acceleration in the direction of change; This represents the cross product operation of vectors;

[0046] Curvature sequences of the same line segments during the same season and without risk events are extracted from historical databases, and an empirical cumulative distribution function is constructed. ;

[0047] Set adaptive threshold The 95th percentile of the empirical cumulative distribution function is expressed as follows:

[0048] ;

[0049] in, It represents the infimum, used to determine the upper confidence limit from the curvature distribution of historical normal operating conditions;

[0050] When the principal curvature is calculated in real time Greater than the adaptive threshold When a sudden change trend is detected in the risk evolution, a risk evolution change warning signal is generated.

[0051] As a preferred embodiment of the risk identification method based on low-dimensional vector computation and power grid topology thinking chain fusion described in this invention, the specific steps of predicting the current risk level using a low-dimensional joint risk vector based on causal enhancement and generating structured risk alarm information according to a preset risk level threshold are as follows:

[0052] Low-dimensional joint risk vector with enhanced causality Input a fully connected classification layer and output a risk level probability distribution;

[0053] Select the risk level with the highest probability in the risk level probability distribution as the current risk level;

[0054] When the current risk level reaches the preset level, the predefined causal rule base is traversed to retrieve all causal rules that are activated by the current visual semantic feature vector and dynamic topological embedding vector, forming a causal inference chain.

[0055] Based on the aforementioned causal inference chain and power grid topology Perform power flow disconnection simulation to determine the set of downstream substations affected and the number of associated users;

[0056] Based on the set of affected downstream substations and the number of associated users, a structured risk warning message is generated, which includes the risk location, causal reasoning chain text, a list of affected equipment, and disposal suggestions.

[0057] In a second aspect, the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, it implements the risk identification method based on low-dimensional vector calculation and power grid topology thinking chain fusion as described in the first aspect of the present invention.

[0058] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements the risk identification method based on low-dimensional vector calculation and power grid topology thinking chain fusion as described in the first aspect of the present invention.

[0059] The beneficial effects of this invention are as follows: By capturing real-time changes in the power grid topology through a dynamic time-series neural network, risk representation is strictly aligned with the operating mode. By embedding intervention consistency loss based on a structural causal model in low-dimensional vector training, risk reasoning is ensured to conform to the physical laws of the power system, avoiding reverse logic misjudgments. The invention introduces principal curvature analysis of risk trajectories, uses differential geometry methods to quantify the evolutionary nonlinearity intensity, and sets adaptive thresholds based on the statistical distribution of historical normal operating conditions, thereby achieving early warning of risk mutation trends. At the same time, a dual-track output mechanism of structured causal alarm + evolutionary mutation early warning is constructed, which provides accurate handling suggestions while supporting preventive intervention, effectively improving the active defense capability and decision-making intelligence level of transmission channels against complex threats such as external damage and wildfires. Attached Figure Description

[0060] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0061] Figure 1 This is a flowchart of the risk identification method based on low-dimensional vector calculation and the fusion of power grid topology thinking chain in Example 1. Detailed Implementation

[0062] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0063] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0064] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0065] Example 1, referring to Figure 1 This is one embodiment of the present invention, which provides a risk identification method based on low-dimensional vector computation and the fusion of power grid topology thinking chain, including the following steps:

[0066] S1. Acquire multi-source heterogeneous sensing data of the power transmission channel area. The multi-source heterogeneous sensing data includes timestamped visible light or infrared images as well as synchronously acquired real-time topology snapshots of the power grid and electrical operation status data.

[0067] Furthermore, intelligent monitoring devices deployed in key sections of transmission lines continuously collect visible light or infrared images at a preset sampling frequency. When they detect construction machinery, hanging foreign objects, wildfire smoke, or personnel intrusion, they trigger an event reporting mechanism, generating an event report containing the target category, geographic coordinates, and a first timestamp. Risk perception event data.

[0068] The power grid operation data stream is received from the power grid dispatch automation system at fixed intervals. The power grid operation data stream contains multiple snapshots of the power grid topology at various time points and corresponding electrical operation status data. Each snapshot of the power grid topology records the current open / closed status of all circuit breakers and disconnectors, and constructs the connectivity between nodes accordingly. The electrical operation status data includes the bus voltage amplitude of each substation, line active power, reactive power, and load rate.

[0069] Risk perception event data and power grid operation data streams are cached separately in the synchronization buffer of edge computing nodes to ensure that the two types of data have an alignment basis in the time dimension. Through the timestamp matching mechanism, each risk perception event data is associated with its most recent power grid topology snapshot and electrical operation status data to form a complete multi-source heterogeneous perception dataset for subsequent risk identification processes.

[0070] It should be noted that by using intelligent monitoring devices to trigger event reporting only when a risky target is detected, and simultaneously caching power grid dispatch data, the communication bandwidth and storage overhead on the edge side are effectively reduced, and the continuous back transmission of the original video stream is avoided. At the same time, by caching and aligning the two types of heterogeneous data locally on the edge nodes, the processing latency in the cloud is reduced, providing a high-quality, low-latency data foundation for subsequent real-time risk identification.

[0071] In particular, by using an event-driven data acquisition mechanism, the high-cost continuous video streaming transmission is transformed into low-frequency risk event triggering and reporting, fundamentally solving the industry pain points of limited bandwidth for edge devices and high cloud storage pressure. At the same time, local caching of power grid operation data on the edge side and initial alignment are achieved, avoiding the latency uncertainty caused by repeated interaction of multi-source data in the wide area network, laying a reliable data synchronization foundation for subsequent high real-time risk identification. This design fully considers the actual constraints of the power Internet of Things edge-cloud collaborative architecture.

[0072] S2. Perform spatiotemporal alignment on multi-source heterogeneous sensing data to form paired samples of risk sensing events and power grid operation status.

[0073] Furthermore, it receives risk target detection results reported by intelligent monitoring devices, including target category, geographic coordinates, and first timestamp. Obtain a sequence of power grid topology snapshots updated at fixed intervals from the power grid dispatch automation system. Each snapshot contains a second timestamp and the corresponding switch status and electrical quantity. Search for the first timestamp in the power grid topology snapshot sequence. The snapshot with the smallest absolute time difference is denoted by its timestamp. The topology is Electrical operating status is ; Integrate risk target detection results with topology and electrical operating status Binding forms paired samples.

[0074] It should be noted that the recent timestamp matching strategy is used to pair visual events with power grid status, which not only ensures spatiotemporal consistency but also avoids mismatch problems caused by minor deviations in the system clock. This alignment mechanism does not require high-precision hardware synchronization, has low engineering implementation cost, is suitable for existing power communication network environments, and improves the deployment adaptability and robustness of the method.

[0075] In particular, the pairing strategy based on minimizing the absolute difference of timestamps is adopted. Its underlying logic is to acknowledge the objective fact that power system state updates are quasi-periodic while visual events are sudden. It does not require strict synchronization, but uses the most recent valid state as the event context, which is both in line with engineering practice and ensures semantic rationality. This method does not require the deployment of high-precision timing equipment or modification of existing scheduling communication protocols. It can achieve cross-system data fusion without increasing hardware investment, effectively improving the feasibility and promotion value of the solution.

[0076] S3. Construct a dynamic time series graph based on grid topology snapshots of multiple consecutive time steps, and generate dynamic topology embedding vectors for grid nodes through a time series graph neural network.

[0077] Furthermore, select based on timestamps The endpoint is and the length is A continuous sequence of snapshots of the power grid topology is generated. For each snapshot, a graph representation is constructed, where the set of vertices corresponds to physical nodes in the transmission lines, the set of edges is determined by the current switching state to indicate connectivity, and the node feature matrix includes voltage magnitude, current intensity, and load rate. The graph sequence is then input into a time-series graph convolutional network, which propagates the time-series information layer by layer through a gated graph update mechanism, outputting the time-series data. The dynamic topological embedding vector of each vertex; let the first vertex be... The vertex at time The embedding is Its update process satisfies:

[0078] ;

[0079] in, Represents vertices At any moment The neighborhood group, and For learnable weight matrix, It is a non-linear activation function; after After the recursive steps, the final dynamic topological embedding vector is obtained. .

[0080] It should be noted that the gated normalized neighborhood aggregation mechanism can adaptively suppress the embedding oscillations caused by frequent topology switching, so that the dynamic topology embedding vector can still maintain semantic stability under operating conditions such as switching operations and line maintenance. At the same time, by introducing the geometric mean normalization factor of the number of neighbors, the feature submersion problem caused by the large degree difference between hub nodes and end nodes in the power grid is effectively alleviated, and the fairness and discriminative power of the embedding representation are improved.

[0081] In particular, the dynamic time-series diagram constructed in this step is not a simple stacking of static snapshots, but rather treats the power grid topology as a signal that evolves over time. Its core lies in capturing the information about the migration of operating modes implied by changes in the topology. By introducing a normalized neighborhood aggregation mechanism, the feature propagation bias caused by differences in node degrees is effectively suppressed, enabling hub substations and end towers to obtain fair representation in the embedded space. This ensures that risk assessment will not be systematically offset by network location, which is a key technical guarantee for achieving consistent risk perception across the entire network.

[0082] S4. Detect risk targets in visible light or infrared images, extract visual semantic feature vectors, and map the detected targets to the corresponding power grid nodes in the dynamic time series diagram.

[0083] Furthermore, forward inference is performed on the input image, outputting the bounding box, class confidence score, and deep feature map of the risk target through a lightweight convolutional neural network; visual semantic feature vectors are extracted from the deep feature map using global average pooling. Calculate the Euclidean distance between the geographical coordinates of the risk target and the pre-stored tower coordinate library; determine the tower index with the smallest distance, map the risk target to the corresponding vertex in the dynamic time series graph, and obtain the dynamic topological embedding vector of that vertex. .

[0084] It should be noted that by mapping geographic coordinates to the Euclidean distance of the tower library, the visual perception space and the power grid topology logic space are accurately anchored, avoiding the topological misalignment caused by the traditional rough division of image regions. The mapping method does not rely on complex 3D reconstruction or SLAM technology, but only requires standard GIS data, and has good engineering feasibility and cross-line generalization ability.

[0085] In particular, this step maps the visual detection results from the image pixel space to the power grid topology logical space. Essentially, it establishes a semantic bridge between the physical world and the electrical model. By using geographic coordinate matching instead of image region division, it avoids positioning drift caused by changes in viewpoint, occlusion, or lighting. This ensures that the same physical target always corresponds to the same power grid node under different cameras, thereby guaranteeing the consistency and traceability of risk information at the topology level and providing accurate spatial anchors for subsequent causal inference.

[0086] S5. The visual semantic feature vector is fused with the dynamic topology embedding vector of the mapped power grid node, and the causal rule constraint of the power system is introduced to generate a causal-enhanced low-dimensional joint risk vector.

[0087] Furthermore, the visual semantic feature vector With dynamic topological embedding vectors The vectors are concatenated into a joint vector; this joint vector is then input into a multilayer perceptron, which outputs an initial low-dimensional risk representation vector. Define the loss of consistency between causal intervention and actual risk, including: assuming the true risk level is... Risk causal variables are The model predicts the distribution as follows: The intervention distribution derived from the causal model is as follows The causal loss is:

[0088] ;

[0089] in, Represents the Kullback-Leibler divergence; during the training phase, Incorporating the total loss function causes the optimized model to output a causal-enhanced low-dimensional joint risk vector. This ensures that risk characterization conforms to the physical causal logic of the power system.

[0090] It should be noted that by introducing the intervention distribution derived from the structural causal model as a supervision signal into the training process, the low-dimensional risk vector is forced to internalize the physical causal chain of external damage behavior → equipment status → system consequences during the learning process. This effectively improves the model's generalization reasoning ability in unseen scenarios and effectively overcomes the defect of pure data-driven models being susceptible to spurious correlation interference.

[0091] In particular, this step introduces causal rule constraints in the vector fusion stage. The underlying idea is to encode the long-term accumulated safe operation knowledge of the power system into learnable regular terms, guiding the neural network to follow physical laws while being data-driven. The hybrid paradigm of knowledge guidance + data-driven enables the model not only to learn what it is, but also to understand why it is so. As a result, it can still make judgments that conform to common sense in engineering when faced with new scenarios not covered by training, effectively enhancing the robustness and credibility of the system.

[0092] S6. Based on the causal enhancement of the low-dimensional joint risk vector, predict the current risk level and generate structured risk alarm information according to the preset risk level threshold.

[0093] Furthermore, this involves causally enhanced low-dimensional joint risk vectors. Input a fully connected classification layer and output a risk level probability distribution; select the risk level corresponding to the highest probability in the risk level probability distribution as the current risk level; when the current risk level reaches a preset level, traverse a predefined causal rule base, retrieve all causal rules whose preconditions are activated by the current visual semantic feature vector and dynamic topological embedding vector, and form a causal inference chain; based on the causal inference chain and the power grid topology... Perform power flow disconnection simulation to determine the set of affected downstream substations and the number of associated users; generate structured risk alarm information containing risk location, causal reasoning chain text, list of affected equipment, and disposal suggestions based on the set of affected downstream substations and the number of associated users.

[0094] It should be noted that the structured risk alarm information not only includes the risk level, but also integrates the causal reasoning chain and power flow simulation results, enabling dispatchers to intuitively understand why an alarm is triggered and how significant the impact is, effectively shortening the decision-making process. This alarm format complies with the requirements of power dispatching regulations and can be directly connected to the existing OMS (Operation Management System) work order process to achieve closed-loop linkage from identification to handling.

[0095] In particular, the structured alarms generated in this step are not simply level labels, but decision support packages that integrate causal chains, scope of impact, and handling suggestions. Their design logic is derived from the actual workflow of dispatchers—they need to know the causes, consequences, and countermeasures of risks. By automatically performing power flow simulation to deduce the impact of faults, the abstract risk vector is transformed into specific power outage ranges and the number of users, giving alarm information direct action guidance value and effectively shortening the decision chain from identification to response.

[0096] S7. Perform trajectory curvature analysis on the causal enhanced low-dimensional joint risk vector sequence within a continuous time window. When the trajectory curvature exceeds the adaptive threshold determined based on the statistical distribution of historical normal operating conditions, generate a risk evolution mutation warning signal as a supplementary warning to the structured risk warning information.

[0097] Furthermore, maintain a first-in-first-out queue to store the most recently accessed... The causal enhancement low-dimensional joint risk vector at each time step is denoted as the sequence. Fit local cubic spline curves to the sequence in chronological order. ,in Normalized time parameters; calculation curve at the latest time. Principal curvature at =1 Its expression is:

[0098] ;

[0099] in, Represents curve right The first derivative reflects the instantaneous direction of change in the risk trajectory; It represents the second derivative, reflecting the acceleration in the direction of change; This represents the vector cross product operation; it extracts curvature sequences of the same line segments during the same season and at the same time of no-risk events from historical databases, and constructs an empirical cumulative distribution function. Set adaptive threshold The 95th percentile of the empirical cumulative distribution function is expressed as follows:

[0100] ;

[0101] in, It represents the infimum, used to determine the upper confidence limit from the curvature distribution of historical normal operating conditions;

[0102] When the principal curvature is calculated in real time Greater than the adaptive threshold When a sudden change in risk evolution is detected, a risk evolution change warning signal is generated; a separate flashing yellow icon is displayed next to the corresponding transmission line element in the visualization interface of the dispatch monitoring system; a new record is added to the alarm log, stating that the risk evolution is accelerating and suggesting early intervention, along with the current principal curvature. With adaptive threshold The ratio; the risk evolution mutation early warning signal only occurs when the structured risk warning information is not triggered and the principal curvature Greater than the adaptive threshold Activated in time to prompt maintenance personnel to implement preventative interventions before the risk reaches a high or emergency level; principal curvature The calculation relies on a sequence of causal-enhanced low-dimensional joint risk vectors within a continuous time window.

[0103] It should be noted that the mutation early warning mechanism based on the principal curvature of the risk trajectory does not rely on a preset risk threshold, but rather adaptively learns the abnormal boundary from historical normal evolution patterns, and has a unique sensitivity to slowly accumulating risks; this early warning, as a supplement to traditional threshold warnings, forms a two-layer defense system of fast response and early detection.

[0104] In particular, the trajectory curvature analysis method proposed in this step is innovative in that it treats risk as a continuous evolutionary process rather than a discrete state, and captures its inherent dynamic characteristics through differential geometry tools. This mechanism can identify critical precursors where the risk level has not yet exceeded the limit but the evolution speed is rapidly accelerating. It makes up for the shortcomings of the traditional threshold method in responding to slow-accumulating threats, and realizes a leap from post-event warning to pre-event prediction, providing a brand-new technical path for the power grid active defense system.

[0105] This embodiment also provides a computer device applicable to the risk identification method based on low-dimensional vector computation and power grid topology thinking chain fusion, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the risk identification method based on low-dimensional vector computation and power grid topology thinking chain fusion as proposed in the above embodiment.

[0106] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0107] This embodiment also provides a storage medium on which a computer program is stored. When executed by a processor, the program implements the risk identification method based on low-dimensional vector computation and the fusion of power grid topology thinking chains as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0108] In summary, this invention captures real-time changes in power grid topology through a dynamic time-series graph neural network, ensuring strict alignment between risk representation and operational mode. By embedding intervention consistency loss based on a structural causal model in low-dimensional vector training, it ensures that risk reasoning conforms to the physical laws of the power system, avoiding reverse logic misjudgments. It introduces principal curvature analysis of risk trajectories, quantifies the evolutionary nonlinearity intensity using differential geometry methods, and sets adaptive thresholds based on historical normal operating condition statistical distributions to achieve early warning of risk mutation trends. Simultaneously, it constructs a dual-track output mechanism of structured causal alarm + evolutionary mutation early warning, providing precise handling suggestions while supporting preventive intervention, effectively improving the transmission channel's proactive defense capabilities and decision-making intelligence level against complex threats such as external damage and wildfires.

[0109] 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, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A risk identification method based on low-dimensional vector computation and the fusion of power grid topology thinking chain, characterized by: include: Acquire multi-source heterogeneous sensing data of the power transmission channel area, including timestamped visible light or infrared images as well as synchronously acquired real-time power grid topology snapshots and electrical operation status data; Spatiotemporal alignment of multi-source heterogeneous sensing data is performed to form paired samples of risk sensing events and power grid operation status; A dynamic time-series graph is constructed based on snapshots of the power grid topology at multiple consecutive time steps, and dynamic topology embedding vectors of power grid nodes are generated through a time-series graph neural network. Risk target detection is performed on visible light or infrared images, visual semantic feature vectors are extracted, and the detected targets are mapped to the corresponding power grid nodes in a dynamic time series graph; The visual semantic feature vector is fused with the dynamic topological embedding vector of the mapped power grid node, and the causal rule constraint of the power system is introduced to generate a causal-enhanced low-dimensional joint risk vector. The current risk level is predicted based on the low-dimensional joint risk vector with causal enhancement, and structured risk warning information is generated according to the preset risk level threshold. Trajectory curvature analysis is performed on the causal enhanced low-dimensional joint risk vector sequence within a continuous time window. When the trajectory curvature exceeds the adaptive threshold determined based on the statistical distribution of historical normal operating conditions, a risk evolution mutation early warning signal is generated as a supplementary early warning to the structured risk warning information.

2. The risk identification method based on low-dimensional vector computation and power grid topology thinking chain fusion as described in claim 1, characterized in that: The specific steps for spatiotemporal alignment of multi-source heterogeneous sensing data to form paired samples of risk sensing events and power grid operating status are as follows: Receive risk target detection results reported by the intelligent monitoring device, the risk target detection results including target category, geographic coordinates and first timestamp. ; Obtain a sequence of power grid topology snapshots updated at fixed periods from the power grid dispatch automation system. Each snapshot contains a second timestamp and the corresponding switch status and electrical quantity. Search for the first timestamp in the snapshot sequence of the power grid topology. The snapshot with the smallest absolute time difference is denoted by its timestamp. The topology is Electrical operating status is ; The risk target detection results are compared with the topology. and electrical operating status Binding forms paired samples.

3. The risk identification method based on low-dimensional vector computation and power grid topology thinking chain fusion as described in claim 2, characterized in that: The process of constructing a dynamic time-series graph based on snapshots of the power grid topology at multiple consecutive time steps, and generating dynamic topology embedding vectors for power grid nodes using a time-series graph neural network, involves the following steps: Select by timestamp The endpoint is and the length is A sequence of snapshots of continuous power grid topology; For each snapshot, a graph representation is constructed, where the set of vertices corresponds to the physical nodes in the transmission line, the set of edges is determined by the current switching state to determine whether they are connected, and the node feature matrix includes voltage magnitude, current intensity, and load rate. The graph sequence is input into a temporal graph convolutional network, and the temporal information is propagated layer by layer through a gated graph update mechanism to output the time sequence. The dynamic topological embedding vector of each vertex; Let the first The vertex at time The embedding is Its update process satisfies: ; in, Represents vertices At any moment The neighborhood group, and For learnable weight matrix, It is a non-linear activation function; go through After the recursive steps, the final dynamic topological embedding vector is obtained. .

4. The risk identification method based on low-dimensional vector computation and power grid topology thinking chain fusion as described in claim 3, characterized in that: The specific steps for detecting risk targets in visible light or infrared images, extracting visual semantic feature vectors, and mapping the detected targets to corresponding power grid nodes in a dynamic time-series graph are as follows: Forward inference is performed on the input image, and the bounding boxes, class confidence scores, and deep feature maps of the risk targets are output through a lightweight convolutional neural network. Visual semantic feature vectors are extracted from the deep feature map using global average pooling. ; Calculate the Euclidean distance based on the geographical coordinates of the risk target and the pre-stored tower coordinate library; Determine the tower index with the minimum distance, map the risk target to the corresponding vertex in the dynamic time series graph, and obtain the dynamic topological embedding vector of that vertex. .

5. The risk identification method based on low-dimensional vector computation and power grid topology thinking chain fusion as described in claim 4, characterized in that: The specific steps for fusing the visual semantic feature vector with the dynamic topology embedding vector of the mapped power grid node, and introducing power system causal rule constraints to generate a causal-enhanced low-dimensional joint risk vector are as follows: Visual semantic feature vectors With dynamic topological embedding vectors Concatenate them into a joint vector; The joint vector is input into a multilayer perceptron, which outputs an initial low-dimensional risk representation vector. ; Define the loss of causal intervention consistency, including: Let the actual risk level be Risk causal variables are The model predicts the distribution as follows: The intervention distribution derived from the causal model is as follows The causal loss is: ; in, This represents the Kullback-Leibler divergence.

6. The risk identification method based on low-dimensional vector computation and power grid topology thinking chain fusion as described in claim 5, characterized in that: The process involves performing trajectory curvature analysis on the causally enhanced low-dimensional joint risk vector sequence within a continuous time window. When the trajectory curvature exceeds an adaptive threshold determined based on the statistical distribution of historical normal operating conditions, a risk evolution mutation early warning signal is generated. The specific steps are as follows: Maintain a first-in-first-out queue to store the most recently accessed items. The causal enhancement low-dimensional joint risk vector at each time step is denoted as the sequence. ; Fit local cubic spline curves to the sequence in chronological order. ,in, This is the normalized time parameter; Calculate the curve at the latest moment Principal curvature at =1 Its expression is: ; in, Represents curve right The first derivative reflects the instantaneous direction of change in the risk trajectory; It represents the second derivative, reflecting the acceleration in the direction of change; This represents the cross product operation of vectors; Curvature sequences of the same line segments during the same season and without risk events are extracted from historical databases, and an empirical cumulative distribution function is constructed. ; Set adaptive threshold The 95th percentile of the empirical cumulative distribution function is expressed as follows: ; in, Indicates the infimum; When the principal curvature is calculated in real time Greater than the adaptive threshold When a sudden change trend is detected in the risk evolution, a risk evolution change warning signal is generated.

7. The risk identification method based on low-dimensional vector computation and power grid topology thinking chain fusion as described in claim 6, characterized in that: The specific steps for predicting the current risk level using a low-dimensional joint risk vector based on causal enhancement and generating structured risk alarm information according to a preset risk level threshold are as follows: Low-dimensional joint risk vector with enhanced causality Input a fully connected classification layer and output a risk level probability distribution; Select the risk level with the highest probability in the risk level probability distribution as the current risk level; When the current risk level reaches the preset level, the predefined causal rule base is traversed to retrieve all causal rules that are activated by the current visual semantic feature vector and dynamic topological embedding vector, forming a causal inference chain. Based on the aforementioned causal inference chain and power grid topology Perform power flow disconnection simulation to determine the set of downstream substations affected and the number of associated users; Based on the set of affected downstream substations and the number of associated users, a structured risk warning message is generated, which includes the risk location, causal reasoning chain text, a list of affected equipment, and disposal suggestions.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the risk identification method based on low-dimensional vector calculation and power grid topology thinking chain fusion as described in any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the risk identification method based on low-dimensional vector calculation and power grid topology thinking chain fusion as described in any one of claims 1 to 7.

Images