Fault monitoring method, system, equipment and medium

Abstract

The invention discloses a fault monitoring method, system and device and a medium, and belongs to the field of power grids, and the method comprises the steps: obtaining a preset fault track signature library and a real-time track of power distribution equipment; performing matching degree calculation on the real-time track and each fault track signature to obtain a matching distance, and comparing a size relationship between a minimum value of the matching distance and a preset distance threshold value; if the matching distance is larger than or equal to the matching distance, matching confidence is obtained through calculation according to the matching distance, and grading early warning is executed according to the matching confidence; and if the matching distance is smaller than the minimum matching distance, determining a current fault state according to a target fault trajectory signature corresponding to the minimum matching distance, determining a plurality of first target fault types having a causal relationship according to the target fault trajectory signature and the fault causal graph, and determining first fault early warning information according to the current fault state and each first target fault type. According to the invention, the accuracy and reliability of fault monitoring can be improved.

Description

Technical Field

[0001] This invention relates to the field of power grids, and more particularly to a fault monitoring method, system, device, and medium. Background Technology

[0002] The power distribution network is a crucial link connecting users in the power system, and its operational reliability directly affects power supply quality and electricity safety. With the widespread integration of new loads such as distributed energy resources and electric vehicles, the operating status of the distribution network is becoming increasingly complex, significantly increasing the risk of sudden failures due to the slow accumulation and deterioration of potential equipment faults. Therefore, developing an intelligent method capable of real-time monitoring of equipment operating status and effective early warning before faults occur is an urgent practical need for achieving preventative power maintenance, avoiding unplanned power outages, and improving power supply reliability.

[0003] Currently, by collecting equipment operation data in real time and constructing real-time trajectories, and by calculating the matching degree between the real-time trajectory and each fault trajectory in the preset fault signature database, it is determined whether the equipment is currently undergoing a certain fault evolution process. If the matching degree exceeds the preset threshold, an early warning for the known fault is triggered, thereby realizing fault monitoring of power distribution equipment.

[0004] However, existing technologies rely entirely on the completeness of a pre-set fault signature database for fault monitoring, lacking the ability to perceive unknown fault risks. This reduces the accuracy of fault identification, resulting in low reliability of monitoring and early warning functions and low security of the power distribution network. Summary of the Invention

[0005] This invention provides a fault monitoring method, system, device, and medium that can solve the problem of limited early warning range caused by the closed and static signature library of existing methods, as well as the resulting lack of ability to identify unknown or new faults.

[0006] This invention provides a fault monitoring method, comprising: Obtain a preset fault trajectory signature library and the real-time trajectory of the power distribution equipment. The fault trajectory signature library includes multiple first fault trajectories and a fault cause-effect graph. Each first fault trajectory is associated with a preset first fault type. The fault cause-effect graph is used to characterize the causal relationship between each first fault trajectory. The matching degree of the real-time trajectory is calculated with each fault trajectory signature in the fault trajectory signature library to obtain the matching distance between the real-time trajectory and each fault trajectory signature, and the relationship between the minimum value of the matching distance and the preset distance threshold is compared. If the minimum matching distance is greater than or equal to the preset distance threshold, the matching confidence level is calculated based on each matching distance, and a graded warning is executed based on the matching confidence level. If the minimum matching distance is less than the preset distance threshold, the current fault state is determined according to the target fault trajectory signature corresponding to the minimum matching distance, and multiple first target fault types with causal relationships are determined according to the target fault trajectory signature and the fault cause-effect graph. First fault warning information is determined according to the current fault state and each of the first target fault types.

[0007] This invention establishes a complete monitoring data foundation by acquiring a preset fault trajectory signature library and the real-time trajectory of power distribution equipment, ensuring that the system has a signature library of known fault modes and current status data available for analysis. The fault trajectory signature library includes multiple first fault trajectories and a fault cause-effect graph, storing not only the morphological characteristics of the faults but also the logical relationships between faults, providing dual knowledge support for subsequent accurate matching and risk prediction. By quantitatively calculating the matching degree between the real-time trajectory and each fault trajectory signature, it objectively judges whether the current operating state deviates from normal or known fault modes, avoiding errors in subjective judgment. If the minimum matching distance... If the distance is greater than or equal to a preset threshold, the matching confidence level is calculated based on the matching distance, and a tiered early warning is executed. For cases where the difference from known fault modes is large (which may be new faults or complex states), the confidence level is calculated to assess the certainty of the judgment, and a tiered early warning is executed, which improves the reliability of fault monitoring, reduces false alarms and missed alarms, and thus improves power grid safety. If the minimum matching distance is less than the preset threshold, which corresponds to a high degree of matching with known fault modes, the specific fault type can be quickly identified, and the causal graph can be used to predict the possible chain faults. This achieves the improvement from diagnosing a single fault to warning of a fault chain, which greatly improves the predictability and comprehensiveness of the early warning.

[0008] Further, the step of calculating the matching confidence score based on each matching distance and executing a tiered early warning based on the matching confidence score specifically involves: When the maximum matching confidence is less than a preset confidence threshold, the real-time trajectory is marked as a trajectory to be verified. For multiple trajectories to be verified within a preset time interval, the trajectories to be verified are prioritized according to the maximum matching confidence to obtain a sequence to be verified. A fault prompt message is generated and responded based on the sequence to be verified. The second fault trajectory corresponding to the second fault type is obtained in sequence. The fault trajectory signature library is dynamically updated according to the second fault type and the second fault trajectory. The power distribution equipment is monitored for faults according to the updated fault trajectory signature library. When the maximum matching confidence level is greater than a preset confidence threshold, multiple second target fault types with causal relationships are determined based on the fault trajectory signature corresponding to the maximum matching confidence level and the fault causal relationship graph, and second fault warning information is determined based on each second target fault type.

[0009] When the maximum matching confidence level is less than a preset confidence threshold, the real-time trajectory is marked as a trajectory to be verified and prioritized. Samples whose fault modes cannot be accurately determined (low confidence level) are identified and listed as trajectories requiring manual intervention for verification. By prioritizing based on confidence level, cases most likely representing novel and unknown faults are processed first, thereby maximizing the value and learning efficiency of manual review. After a new fault type is confirmed by manual verification, the new knowledge (second fault trajectory) can be automatically integrated into the signature library, realizing the self-growth and self-evolution of the signature library. When the maximum matching confidence level is greater than a preset confidence threshold, a second fault warning is determined. For relatively certain matches (high confidence level), the causal graph is directly called to predict and warn of cascading risks, which realizes a rapid and automated response to high-certainty risks.

[0010] Furthermore, the first fault trajectory is obtained through cluster analysis of historical fault trajectory data, specifically as follows: Obtain the historical fault trajectory data corresponding to each of the first fault types from the historical operation data; Cluster analysis is performed on the historical fault trajectory data to obtain at least one trajectory cluster; For each of the trajectory clusters, the target center trajectory of the trajectory cluster is calculated using the dynamic time warping centroid averaging algorithm to obtain the first fault trajectory corresponding to the first fault type.

[0011] By combining clustering and DBA center trajectory calculation, a clean, robust, and truthfully reflective standard signature of fault evolution can be automatically extracted from historical data containing noise and individual differences, ensuring the accuracy and reliability of subsequent online matching. It supports multiple trajectory clusters (i.e., multiple signatures) for one fault type, enabling more refined modeling and identification of different manifestations of the same fault, thereby improving the early warning sensitivity for complex and variable faults. The final generated target center trajectory (i.e., the first fault trajectory) serves as a standardized, high-fidelity signature for that fault type, enhancing representativeness and noise resistance.

[0012] Furthermore, the historical fault trajectory data includes multiple historical fault trajectories, and the clustering analysis of the historical fault trajectory data to obtain at least one trajectory cluster specifically involves: A dynamic time warping algorithm is used to calculate the distance between each pair of the historical fault trajectories mentioned in the middle, and a distance matrix is ​​obtained. Based on the distance matrix, each trajectory cluster is obtained through manifold learning algorithm and clustering algorithm.

[0013] By combining DTW (Precise Morphological Measurement), manifold learning (structural dimensionality reduction and denoising), and clustering algorithms (automatic grouping), a complete and advanced time series clustering technology solution is constructed. This solution can more effectively address the high-dimensionality, nonlinearity, and time distortion characteristics of power fault trajectory data, ensuring high consistency within the generated trajectory clusters and clear distinction between clusters. This results in more accurate and robust trajectory clustering, laying a more solid foundation for generating high-quality signatures. Furthermore, it reduces the reliance on manual prior knowledge and parameter tuning in the clustering process, enabling the system to discover patterns from the data itself more objectively and automatically, thus improving the automation and intelligence level of the method.

[0014] Furthermore, the trajectory cluster includes multiple first historical trajectories, and for each trajectory cluster, the target center trajectory of the trajectory cluster is calculated using a dynamic time warping centroid averaging algorithm, specifically as follows: Randomly select one of the first historical trajectories from each trajectory cluster as the initial center trajectory; The initial center trajectory is iteratively optimized. In each iteration, for each first historical trajectory, a dynamic time warping algorithm is used to calculate the target alignment path between the first historical trajectory and the current center trajectory. The current center trajectory is updated according to all the target alignment paths to obtain the updated center trajectory. The iteration stops when the preset convergence condition is met, and the target center trajectory is obtained.

[0015] This approach breaks down the complex DBA algorithm into four clear steps: initialization, alignment, update, and convergence judgment, generating high-quality signatures. Through iterative optimization, the final target center trajectory is the centroid or average of the trajectory cluster in the DTW sense. It maximizes the generalization and preservation of the common morphological features of all trajectories within the cluster, while smoothing out individual differences and noise. The resulting signature is mathematically optimal or near-optimal, greatly enhancing the discriminativeness and accuracy of subsequent matching. The convergence condition ensures that the algorithm outputs a stable, optimal (or suboptimal) center trajectory, preventing infinite loops or oscillations. It achieves a balance between computational cost and result accuracy, ensuring the effectiveness and efficiency of the algorithm.

[0016] Further, the calculation of the matching confidence score based on each matching distance specifically involves: Convert each of the aforementioned matching distances into a corresponding similarity score; The matching confidence level between the real-time trajectory and each fault trajectory signature is obtained by calculating all the similarity scores using a probability normalization function.

[0017] This process converts each matching distance into a similarity score, transforming the distance values ​​representing differences into scores that better align with the intuitive concept of similarity, thus laying the foundation for subsequent probabilistic interpretation. By calculating the matching confidence score through a probability normalization function, the similarity score is transformed into a probabilistic confidence score, providing a standardized and comparable quantitative indicator for assessing the degree of certainty regarding different matching results. This is a crucial prerequisite for achieving tiered early warning and proactive learning ranking.

[0018] Furthermore, the fault cause-effect graph is obtained by performing time-series analysis on historical fault events or by analyzing them according to preset fault rules, specifically as follows: Each of the first fault type and each of the second fault types are used as nodes; By analyzing the temporal correlation between various fault types in historical fault events, or by determining the causal relationship between nodes according to preset fault rules; Based on the causal relationships described, the directed edges between the nodes are determined to obtain the fault causal graph.

[0019] This clarifies two reliable sources of causal graph knowledge: data-driven and knowledge-driven, ensuring the effectiveness and practicality of causal relationships; by determining the causal relationships between nodes and obtaining directed edges, implicit fault-related knowledge is made explicit into a structured, computable, and reasonable graph, thereby enabling causal logical reasoning about faults.

[0020] Another embodiment of the present invention provides a fault monitoring system, including: an acquisition module, a matching module, a first monitoring module, and a second monitoring module; The acquisition module is used to acquire a preset fault trajectory signature library and the real-time trajectory of the power distribution equipment. The fault trajectory signature library includes multiple first fault trajectories and a fault cause-effect graph. Each first fault trajectory is associated with a preset first fault type. The fault cause-effect graph is used to characterize the causal relationship between each first fault trajectory. The matching module is used to calculate the matching degree between the real-time trajectory and each fault trajectory signature in the fault trajectory signature library, obtain the matching distance between the real-time trajectory and each fault trajectory signature, and compare the minimum value of the matching distance with a preset distance threshold. The first monitoring module is used to calculate the matching confidence level based on each matching distance if the minimum matching distance is greater than or equal to the preset distance threshold, and to execute a graded early warning based on the matching confidence level; The second monitoring module is configured to determine the current fault state based on the target fault trajectory signature corresponding to the minimum matching distance if the minimum matching distance is less than the preset distance threshold, determine multiple first target fault types with causal relationships based on the target fault trajectory signature and the fault cause-effect graph, and determine first fault warning information based on the current fault state and each of the first target fault types.

[0021] Another embodiment of the present invention provides a terminal device, including: a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, it implements the steps of the fault monitoring method of the present invention.

[0022] Another embodiment of the present invention provides a computer-readable storage medium item, including: a stored computer program, which, when the computer program is running, controls the device where the computer-readable storage medium is located to perform the steps of the fault monitoring method of the present invention. Attached Figure Description

[0023] To more clearly illustrate the technical solution of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0024] Figure 1 This is a flowchart illustrating a fault monitoring method provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of a fault monitoring system provided in an embodiment of the present invention. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0026] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the application; the terms “comprising” and “having”, and any variations thereof, in the specification, claims, and foregoing description of the drawings are intended to cover non-exclusive inclusion.

[0027] In the description of the embodiments of this application, technical terms such as "first" and "second" are used only to distinguish different objects and should not be construed as indicating or implying relative importance or implicitly specifying the number, specific order, or primary and secondary relationship of the indicated technical features. In the description of the embodiments of this application, "multiple" means two or more, unless otherwise explicitly defined.

[0028] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0029] In the description of the embodiments in this application, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.

[0030] In the description of the embodiments of this application, the term "multiple" refers to two or more (including two), similarly, "multiple sets" refers to two or more (including two sets), and "multiple pieces" refers to two or more (including two pieces).

[0031] In the description of the embodiments of this application, unless otherwise expressly specified and limited, technical terms such as "installation," "connection," "joining," and "fixing" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. For those skilled in the art, the specific meaning of the above terms in the embodiments of this application can be understood according to the specific circumstances.

[0032] See Figure 1 To address the limitations of existing technologies in terms of early warning scope due to the closed and static nature of signature libraries, and the resulting lack of ability to identify unknown or novel faults, an embodiment of the present invention provides a fault monitoring method, comprising: Step S101: Obtain a preset fault trajectory signature library and the real-time trajectory of the power distribution equipment. The fault trajectory signature library includes multiple first fault trajectories and a fault cause-effect graph. Each first fault trajectory is associated with a preset first fault type. The fault cause-effect graph is used to characterize the causal relationship between each first fault trajectory.

[0033] In this embodiment, a pre-built and stored fault trajectory signature library is invoked, and the real-time monitoring data stream of the power distribution equipment is accessed. The signature library includes two core knowledge entities: first, fault trajectory signatures (i.e., the first fault trajectory), each signature having been mapped to a defined or identified fault type via offline or online methods; second, a fault cause-effect graph, which encodes in a structured manner the potential or known causal logical relationships, such as initiation and propagation, between the fault types represented by different fault trajectory signatures within the library. The specific steps for constructing the real-time trajectory are as follows: At each sampling time point, multi-source heterogeneous data is collected and fused to form a dynamic state vector. This multi-source heterogeneous data is acquired in real time from intelligent electronic devices (IEDs), feeder terminal units (FTUs), distribution terminal units (DTUs), and other devices installed on power distribution lines or equipment (e.g., transformers, switchgear) through data acquisition interfaces conforming to specific communication protocols (e.g., IEC 61850, Modbus, or CDT). The data specifically includes: 1) telemetry data, such as the effective values ​​of three-phase voltages (A, B, and C), effective values ​​of three-phase currents, active power, reactive power, and power factor; 2) sensor data, such as oil temperature and winding temperature installed on transformers, ambient temperature and humidity and sulfur hexafluoride (SF6) gas pressure installed in switchgear, or discharge quantities collected by partial discharge monitoring devices; 3) status data, such as the open / closed position of circuit breakers (which can be encoded as values ​​0 or 1), remote / local control status, and alarm signals from protection devices. The fusion process of multi-source heterogeneous data is specifically as follows: The aforementioned data, aligned to the same sampling time, are combined into a high-dimensional numerical vector according to a preset dimensional order. This vector numerically constitutes a snapshot of the comprehensive operating state of the power distribution equipment at that specific moment, serving as a dynamic state vector. Based on the continuous generation of dynamic state vectors, the method of this invention performs the following sub-steps to construct a time series for real-time analysis: ensuring that the scale of the real-time trajectory is consistent with the data in the signature library, for example, applying the same standardization transformation as the fault trajectory signature library to the currently acquired real-time trajectory; and combining the latest dynamic state vectors arranged in chronological order through a sliding time window to form a real-time trajectory.

[0034] In one embodiment, the sliding time window has a fixed length, which can be preset according to the typical evolution cycle of the fault to be detected. For example, it can be set to a length that can accommodate all data collected in the most recent 15 minutes, or it can be set to a fixed number of vectors (e.g., containing the latest 100 dynamic state vectors). The sliding time window is updated in a first-in, first-out (FIFO) manner: whenever a new dynamic state vector appears at the current time... The generated vector is added to the end of the window, while the oldest dynamic state vector at the front of the window is removed. Through this update method, the window slides forward on the timeline, always containing a fixed number of the latest continuous dynamic state vectors. At any given moment, the set of time-ordered dynamic state vectors contained within this sliding time window constitutes a real-time trajectory. It can be represented as: ; in, This represents the real-time trajectory used for analysis at the current moment; It is the preset length of the sliding time window (in terms of vector quantity); Indicates the index of the current sampling time; Indicates at time The generated dynamic state vector. This real-time trajectory. As a complete data unit, it is sent to the subsequent real-time analysis module for matching with the signature database.

[0035] Step S102: Calculate the matching degree between the real-time trajectory and each fault trajectory signature in the fault trajectory signature library to obtain the matching distance between the real-time trajectory and each fault trajectory signature, and compare the minimum value of the matching distance with the preset distance threshold.

[0036] In this embodiment, the Dynamic Time Warping (DTW) algorithm is used to calculate the distance between the real-time trajectory and each fault trajectory signature in the signature database. This DTW distance is used to quantify the matching degree. Specifically, the core recursive formula for this calculation process is: ; in, Subsequence representing real-time trajectory Subsequence signed with a certain fault trajectory The minimum cumulative distance between them; It is a vector and The local cost between them can be calculated using Euclidean distance; This indicates taking the minimum of the three values ​​within the parentheses.

[0037] This step outputs a set of DTW distance values. Each distance value corresponds to the degree of matching between the real-time trajectory and a fault trajectory signature in the signature database. The smaller the distance value, the more similar the real-time trajectory and the signature are in form, and the higher the matching degree. The minimum value is determined from this set of distance values. and its corresponding fault trajectory signature, and the minimum distance With a preset warning distance threshold Comparison. Specifically, the minimum distance judgment formula is used for judgment. The minimum distance judgment formula is as follows: ; in, It is a real-time trajectory; It is a pre-defined fault trajectory signature library; It is the first in the library Signature of each fault trajectory; This indicates the calculation of the DTW distance between two trajectories; The algorithm finds the signature that minimizes the DTW distance. This signature is the best-match signature. The corresponding minimum distance is .

[0038] Step S103: If the minimum matching distance is greater than or equal to the preset distance threshold, the matching confidence is calculated based on each matching distance, and a graded warning is executed based on the matching confidence.

[0039] In this embodiment, if To achieve tiered early warning and proactive learning, this embodiment not only uses the minimum matching distance It also introduces a matching confidence level. To assess the uncertainty of the matching results, this confidence level is determined by evaluating all DTW distance values ​​corresponding to the real-time trajectory. ,in This is calculated using the Softmax function (the total number of signatures in the signature database). Considering... The closer the value is to 1, the higher the uniqueness and certainty of the match (i.e., (much smaller than all other distances) The lower the value (e.g., compared to other values), the better. The closer the values ​​are, the more ambiguous the matching result, and the higher the uncertainty (i.e., the closer the values ​​are). and the second-best distance (Very close). Based on this, the following tiered early warning rules are designed: 1. High-confidence warning (regular warning): If Greater than a preset high confidence threshold (For example A score of 0.8 indicates that the match is not only close in distance but also highly unique (i.e., significantly closer to other signatures). In this case, a standard known fault warning is generated and associated with it. Fault labels; 2. Low confidence warning (priority review warning): If Less than a preset low confidence threshold (For example This corresponds to a reasonably good match. However, in cases with very low confidence, it indicates that the real-time trajectory may fall between two or more known fault signatures in terms of morphology, exhibiting high ambiguity. For real-time trajectories corresponding to low-confidence warnings, in addition to triggering regular warnings, they will be marked as trajectories to be verified. This marking will allow them to be verified in the human-machine collaborative verification interface for their corresponding second fault type. This is because such ambiguous samples are of extremely high value for model diagnosis (is it fault A or fault B?) or model iteration (is it possible that it is a new type between A and B?), requiring human intervention to learn and improve the model's autonomous learning ability.

[0040] Step S104: If the minimum matching distance is less than the preset distance threshold, the current fault state is determined according to the target fault trajectory signature corresponding to the minimum matching distance, and multiple first target fault types with causal relationships are determined according to the target fault trajectory signature and the fault cause-effect graph. First fault warning information is determined according to the current fault state and each of the first target fault types.

[0041] In this embodiment, if If the current operating status of the device is highly matched with a known fault, an early warning will be triggered, where the early warning distance threshold is specified. It can be set based on the statistical distribution of the distance between the signatures of historical normal operation trajectories and each fault trajectory. When an alert is triggered, the best matching signature is used. The current failure's evolution progress is determined by calculating the minimum matching distance. The optimal regularized path obtained through concurrent processing is calculated, and this path records the real-time trajectory. Best matching signature The optimal correspondence between all data points is determined. The evolutionary progress calculation formula is used for calculation: ; in, Indicates the current progress of the fault evolution; It is the best matching signature The total length represents the standard length of a complete fault evolution; It is the optimal regularized path, and the real-time trajectory The last dynamic state vector matches in the signature. The index of the dynamic state vector in the system. This index value is obtained directly by finding the endpoint of the optimal regularized path.

[0042] Furthermore, once the best matching signature is determined... And meet the early warning conditions ( When the best matching signature S is reached, bestThe associated first fault type label is identified as the first target fault type, and a prognostic analysis of cascading fault risk is performed based on the fault cause-effect graph.

[0043] In one embodiment, triggering a node query will The associated first fault type (e.g., transformer A overload fault) is used as the parent node for the query; causal relationship retrieval is performed in the fault causal graph to find all child nodes pointed to by the parent node A (e.g., line insulation degradation fault B, abnormal temperature rise fault of switch contacts C, etc.); first fault warning information is generated, which includes... The first fault warning information includes the directly associated primary target fault type, one or more cascading faults associated with the primary target fault type, and the current fault status (i.e., fault evolution progress) of the primary target fault type. For example, since fault A (overload of transformer A) has been detected, according to the fault cause-effect graph, this event will significantly increase the risk of future faults B (insulation degradation of line B) and C (abnormal temperature rise of switch contacts C), and the primary fault warning information also includes the calculated current evolution progress of fault A. And the causal association weights (e.g., P(fault B|fault A)) obtained from the causal graph, so that operations personnel can assess the urgency and relevance of risks.

[0044] Through this step, the matching and early warning mechanism of the present invention is no longer limited to analyzing a single trajectory, but extends to analyzing the temporal and logical relationships between multiple trajectories (different fault types), realizing the improvement from diagnosis to prognosis, and can effectively predict and prevent potential cascading fault risks.

[0045] As an example of an embodiment of the present invention, the step of calculating the matching confidence level based on each matching distance and performing hierarchical early warning based on the matching confidence level specifically includes: when the maximum matching confidence level is less than a preset confidence threshold, the real-time trajectory is marked as a trajectory to be verified; for multiple trajectories to be verified within a preset time interval, the trajectories to be verified are prioritized according to the maximum matching confidence level to obtain a sequence to be verified; fault prompt information is generated and responded based on the sequence to be verified; the second fault trajectory corresponding to the second fault type is obtained sequentially; the fault trajectory signature library is dynamically updated according to the second fault type and the second fault trajectory; and the power distribution equipment is monitored for faults according to the updated fault trajectory signature library; when the maximum matching confidence level is greater than the preset confidence threshold, multiple second target fault types with causal relationships are determined according to the fault trajectory signature corresponding to the maximum matching confidence level and the fault causal relationship graph; and second fault early warning information is determined according to each second target fault type.

[0046] In this embodiment, if the calculated minimum distance The warning triggering conditions are not met, i.e. If the current real-time trajectory differs significantly in form from all known fault trajectory signatures in the signature database, it represents a novel, unrecorded anomaly. The marking process involves taking the complete real-time trajectory data, along with its timestamp and the calculated minimum distance, and marking it as a trajectory to be verified. and the nearest best-matching signature The identifiers are stored together in a dedicated data storage area and assigned a status label to be verified. To reduce the manual burden of subsequent human-machine collaborative verification, the maximum matching confidence of multiple trajectories to be verified within a preset time interval is used. The priorities of each trajectory to be verified are sorted, and the specific steps are as follows: When At that time, obtain the maximum matching confidence score for each trajectory to be verified. In this scenario, it reflects the model's confidence in the nearest signature, a relatively low... (For example This indicates that the real-time trajectory does not match any known signature. Furthermore, the model is highly uncertain about its worst match (e.g., it is equidistant from multiple signatures). The most uncertain samples of this model are the most valuable samples for active learning; therefore, when marked as samples to be verified, they are assigned a priority, which in one embodiment can be related to... Negative correlation (i.e., the lower the confidence level, the higher the uncertainty, and the higher the priority).

[0047] This verification tag and verification process establishes a feedback mechanism for the dynamic updating and expansion of the signature database. This mechanism specifically includes: Manual verification: The marked traces to be verified are provided to technicians for analysis. If the trace is confirmed to correspond to a meaningful, previously undefined equipment fault, the trace is identified as a new second fault trace.

[0048] Signature database update: After accumulating a sufficient number of manually confirmed new second fault trajectories of the same type, these trajectories are used as new input data to re-execute the cluster analysis and center trajectory calculation process, thereby generating a new fault trajectory signature for the new historical fault data.

[0049] Signature library supplement: The newly generated signature is added to the preset fault trajectory signature library and the causal relationship graph is updated, thereby realizing the discovery, modeling and future identification of unknown fault modes.

[0050] In the human-machine collaborative verification process, the process presents analytical information to assist decision-making through a visual interactive interface. In one embodiment, when an alert is triggered or a trajectory to be verified is marked, the interface displays at least the following: Trajectory pattern visualization: Comparing the current real-time trajectory with the calculated best matching signature Parallel or superimposed visualizations are drawn in the same coordinate system to facilitate an intuitive comparison of the morphological similarity between two trajectories; Quantitative analysis results: Showing the calculated minimum DTW distance And the warning distance threshold used as the basis for judgment. ; Fault diagnosis conclusion: Clearly give the following conclusion: The associated fault type label, or for the trajectory to be verified, clearly indicates that it is an unknown anomaly; Context metadata: Provides the timestamp of the trajectory occurrence, the associated device identifier, and other relevant telemetry or sensor data snapshots.

[0051] Causal association analysis; in a preferred embodiment, when the labeled track is a trajectory to be verified (i.e. When the current trajectory occurs, the interactive interface will automatically search for the device's "upstream causal events." It will query whether, within a preset time window before the current trajectory occurred (e.g., the past week), a known fault warning that served as its "cause" occurred on its associated devices (determined based on the fault cause-effect diagram and power grid topology). If such upstream causal events are found, the interface will highlight them: "Note; the occurrence of this 'unknown anomaly' has a potential causal relationship with fault A occurring in [device X] at [time Y]." Confidence and Priority Assessment: In a preferred embodiment, the interface also displays the calculated match confidence level. For known fault warnings: This confidence level is used to assist in the judgment. For example, if a "low confidence warning" is triggered, "fuzzy match, priority review is recommended" will be highlighted here.

[0052] For trajectories to be verified: the interface will display the verification priority of the trajectory, and the worklist will by default arrange the high-priority (i.e. the least uncertain samples in the model) at the top to guide the verification of the most valuable samples.

[0053] Based on the information presented above, this visual interactive interface provides a set of interactive elements for performing verification operations. The verification operations include at least: Confirm: If you agree with the fault type assessment given by the system, you can perform the confirmation operation; Correction: If the trajectory is considered abnormal but does not fall under the type determined by the system, a correction operation can be performed. This operation allows you to select a more accurate label from a preset fault list or enter a new fault type label. Rejection: If the trajectory is determined to be a normal operational disturbance or data noise, and not a real fault, a rejection operation can be performed to mark it as non-abnormal.

[0054] The results of manually performed verification operations (including confirmed, corrected labels, or rejection statuses) will be recorded by the system and stored in association with the corresponding trajectory data. This manually verified and labeled trajectory data constitutes a set of confirmed sample data. In particular, trajectories that are corrected or confirmed as novel anomalies will serve as the direct basis for executing the dynamic update process of the signature library, thereby enabling the self-learning and model iteration of the method of this invention.

[0055] As an example of an embodiment of the present invention, the first fault trajectory is obtained by clustering analysis of historical fault trajectory data. Specifically, the historical fault trajectory data corresponding to each first fault type is obtained from historical operating data; the historical fault trajectory data is clustered to obtain at least one trajectory cluster; for each trajectory cluster, the target center trajectory of the trajectory cluster is calculated using the dynamic time warping centroid averaging algorithm to obtain the first fault trajectory corresponding to the first fault type.

[0056] In this embodiment, historical operational data can originate from various operating systems of the distribution network, such as Supervisory Control and Data Acquisition (SCADA), Distribution Management System (DMS), or Operation and Maintenance Management System (OMS). The extraction process begins with historical fault event records in the OMS, which include confirmed fault types, fault occurrence times, fault clearing times, and associated equipment identifiers. Based on the equipment identifiers, fault occurrence and clearing times in the records, continuous operational data of the corresponding equipment within a specified time window (e.g., from one hour before the fault occurred to the time of fault clearing) is queried and extracted from the historical databases of systems such as SCADA, forming a historical fault trajectory.

[0057] Each extracted historical fault trajectory is a multi-dimensional time series, consisting of multiple dynamic state vectors arranged in chronological order. Each dynamic state vector is a fusion of various operational data of the power distribution equipment at a given point in time. This operational data includes: telemetry data, such as line voltage, three-phase current, active power, and reactive power; sensor data, such as transformer temperature, ambient humidity, and partial discharge; and state data, such as switch opening / closing positions and protection device operation signals. By repeating this extraction process for multiple historical fault events, multiple historical fault trajectories corresponding to these events can be obtained.

[0058] Because the data in each dimension of the dynamic state vector have different physical dimensions and numerical ranges (e.g., voltage is in kV, temperature is in °C, and switch states are 0 / 1 Boolean values), direct numerical calculations will result in the dimension with the larger numerical range dominating the calculation results. To eliminate this influence and ensure the accuracy of subsequent trajectory morphology similarity calculations, the data needs to be standardized.

[0059] In one embodiment, the standardization process employs the Z-score standardization method. The standardized vector dimension is calculated using the Z-score standardization formula, which is: ; in, The first digit represents the standardized dynamic state vector. Values ​​in each dimension; The first element representing the original dynamic state vector is... Values ​​in each dimension; and The first The mean and standard deviation of each dimension on the training set. The training set consists of all or part of the historical fault trajectories.

[0060] After this step, all dynamic state vectors in each historical fault trajectory are converted into standardized vectors, with each dimension containing dimensionless values. Furthermore, data preprocessing may include data cleaning steps, such as interpolating and filling missing values ​​in the trajectory or smoothing outliers. The specific implementation of this data cleaning is well-known in the field and will not be elaborated upon here.

[0061] After preprocessing the historical fault trajectories, the method of this invention continues to perform cluster analysis on multiple historical fault trajectories to obtain at least one trajectory cluster. This cluster analysis process may include the following sub-steps: A dynamic time warping algorithm is used to calculate the pairwise distances between multiple historical fault trajectories, resulting in a distance matrix. Based on the distance matrix, at least one trajectory cluster is obtained through a manifold learning algorithm and a clustering algorithm. For each generated trajectory cluster, a representative central trajectory is calculated using a dynamic time warping centroid averaging algorithm; this representative central trajectory constitutes a fault trajectory signature.

[0062] As an example of an embodiment of the present invention, the historical fault trajectory data includes multiple historical fault trajectories. The step of performing cluster analysis on the historical fault trajectory data to obtain at least one trajectory cluster specifically involves: using a dynamic time warping algorithm to calculate the distance between each pair of the historical fault trajectories to obtain a distance matrix; and based on the distance matrix, obtaining each trajectory cluster through a manifold learning algorithm and a clustering algorithm.

[0063] In this embodiment, since the historical fault trajectories of the same type of fault are stretched or compressed on the time axis, the Dynamic Time Warping (DTW) algorithm can effectively calculate their morphological similarity by finding the best nonlinear alignment path between two sequences, regardless of the difference in evolution rate.

[0064] In one embodiment, the algorithm first defines the local cost between any two normalized dynamic state vectors in two trajectories. The local cost is calculated using the Euclidean distance formula, which is: ; in, and These represent the first two different trajectories. The and the first A standardized dynamic state vector; It is the total dimension of the dynamic state vector; and They are vectors and The Values ​​in each dimension; This represents the Euclidean distance between the two vectors, i.e., the local alignment cost.

[0065] Based on local cost, the cumulative distance matrix is ​​calculated using the DTW core recursive formula. The core recursive formula for DTW is: ; in, Subsequence representing the first trajectory Subsequence of the second trajectory The minimum cumulative distance between them; It is the local cost of aligning the two vectors. This indicates taking the minimum of the three values ​​within the parentheses, corresponding to selecting the path with the lowest cost from the three preceding paths.

[0066] To ensure the correct execution of this recursive calculation, its boundary conditions are set as follows: ; (for ); (for For two lengths respectively and The complete trajectory, and its final DTW distance is By performing pairwise DTW distance calculations on all preprocessed historical fault trajectories, a symmetric distance matrix can be obtained.

[0067] To improve the effectiveness of subsequent clustering algorithms, in one embodiment, a manifold learning algorithm can be first used to reduce the dimensionality of the geometric structure implied by the distance matrix. The manifold learning algorithm can be the Uniform Manifold Approximation and Projection (UMAP) algorithm. In the dimensionality-reduced feature space, or directly on the original distance matrix, a clustering algorithm is used to divide the trajectories. The clustering algorithm can be the Density-Based Noise Applied Spatial Clustering (DBSCAN) algorithm. Key parameters of the DBSCAN algorithm, such as the neighborhood radius and the minimum number of samples required to form a cluster, can be set based on experiments or the experience of those skilled in the art. Through this step, historical fault trajectories with similar morphologies are divided into at least one trajectory cluster. After obtaining at least one trajectory cluster, the method of the present invention continues to perform subsequent steps to calculate representative fault trajectory signatures and construct a signature library.

[0068] As an example of an embodiment of the present invention, the trajectory cluster includes multiple first historical trajectories. For each trajectory cluster, the target center trajectory of the trajectory cluster is calculated using a dynamic time warping centroid averaging algorithm. Specifically, one first historical trajectory is randomly selected from each trajectory cluster as the initial center trajectory; the initial center trajectory is iteratively optimized. In each iteration, for each first historical trajectory, the target alignment path between the first historical trajectory and the current center trajectory is calculated using a dynamic time warping algorithm. The current center trajectory is updated according to all the target alignment paths to obtain the updated center trajectory. The iteration stops when a preset convergence condition is met, and the target center trajectory is obtained.

[0069] In this embodiment, to eliminate individual differences and noise effects among trajectories within a trajectory cluster and to extract their common evolutionary patterns, a representative central trajectory needs to be calculated for each cluster. The Dynamic Time Warped Centroid Average (DBA) algorithm performs this calculation through an iterative process.

[0070] In one embodiment, the execution process of the DBA algorithm is as follows: Initialization: Randomly select a historical fault trajectory from a trajectory cluster as the initial center trajectory. .

[0071] Iterative alignment: in the first In the next iteration, for each trajectory within the cluster All of them use the Dynamic Time Warping (DTW) algorithm to calculate their relationship with the current center trajectory. Optimal alignment path .

[0072] Iterative update: Based on all the alignment paths obtained in the previous step, calculate the new center trajectory. The centroid average formula is used to calculate each vector of the new center trajectory. The centroid average formula is: ; in, It is in the The new center trajectory generated in the next iteration The One dynamic state vector; It is a set of indexed pairs containing all trajectories within the cluster that are related to the old center trajectory. The The indices of all vectors aligned with each vector; It is the first Trajectory The One dynamic state vector; It is a set The number of elements in the middle.

[0073] This formula indicates that by applying the set The components of all the dynamic state vectors of the index are summed, and then each component is averaged to calculate the first component of the new center trajectory. vectors .

[0074] Termination criteria: Iterate through alignment and updates until the center trajectory converges. The convergence criterion can be set as: the newly calculated center trajectory... The center trajectory of the previous one The DTW distance between them is less than a preset minimum threshold. Or, the number of iterations reaches the preset upper limit.

[0075] The central trajectory obtained after the iterative process converges is the representative central trajectory of the trajectory cluster and serves as a fault trajectory signature.

[0076] As an example of an embodiment of the present invention, the step of calculating the matching confidence based on each of the matching distances specifically involves: converting each of the matching distances into a corresponding similarity score; and calculating the matching confidence between the real-time trajectory and each of the fault trajectory signatures by using a probability normalization function.

[0077] In this embodiment, firstly, the distance is converted into a similarity score, for example, using negative distance. Or its scaling value. Then, the Softmax formula is used to calculate each signature. Matching probability : ; in: It is real-time trajectory and signature DTW distance; It is a temperature coefficient used to adjust the smoothness of the probability distribution; The smaller the value, the sharper the probability distribution and the more obvious the difference in confidence level. That is, a signature is matched. The probability of.

[0078] Finally, with minimum distance Corresponding best matching signature The corresponding probability This is the matching confidence score output in this step.

[0079] As an example of an embodiment of the present invention, the fault causal graph is obtained by performing time-series analysis on historical fault events or by analyzing them according to preset fault rules. Specifically, it involves taking each of the first fault types and each of the second fault types as nodes; analyzing the time-series correlation between each fault type in historical fault events or by determining the causal relationship between each node according to preset fault rules; and determining the directed edges between each node according to each causal relationship to obtain the fault causal graph.

[0080] In this embodiment, the fault trajectory signature calculated for each trajectory cluster is associated with the known fault type label corresponding to that cluster (e.g., insufficient oil pressure fault of a certain type of circuit breaker). Finally, all fault trajectory signatures with labeled fault types are stored in the database to form a preset fault trajectory signature library for subsequent real-time analysis steps.

[0081] This fault cause-effect graph uses the fault types represented by the generated fault trajectory signatures as nodes in the graph. The directed edges in the graph represent the causal relationships between nodes (i.e., between faults).

[0082] The construction of directed edges and their causal relationships can be achieved through one or a combination of the following two methods: Based on a manual signature database: Import predefined fault association rules. For example, predefined rules state that long-term overload of transformer A is one of the reasons for accelerated insulation degradation of line B.

[0083] Time-series data mining: By analyzing historical fault events, statistical analysis is performed using association rule mining or time-series causal inference algorithms (such as Granger causality test or convergent cross-mapping). For example, if the system finds in historical data that type A fault events (such as transformer overload) occur within a preset time window T... w If a node A is likely to precede a Class B fault event (such as insulation degradation) within a certain period (e.g., several days or weeks), then a directed edge from node A to node B is constructed in the graph.

[0084] Furthermore, each directed edge can be assigned a causal association weight, such as the conditional probability P(fault B|fault A) of the occurrence of the temporal association.

[0085] Ultimately, the generated fault trajectory signature set (as the basic template for matching) and the generated fault cause-effect graph (as the basis for prognostic analysis) together constitute the preset fault trajectory signature library of the present invention.

[0086] like Figure 2 As shown, based on the above-described method embodiments, an embodiment of the present invention provides a fault monitoring system 200, including: an acquisition module 201, a matching module 202, a first monitoring module 203, and a second monitoring module 204; The acquisition module 201 is used to acquire a preset fault trajectory signature library and the real-time trajectory of the power distribution equipment. The fault trajectory signature library includes multiple first fault trajectories and a fault cause-effect graph. Each first fault trajectory is associated with a preset first fault type. The fault cause-effect graph is used to characterize the causal relationship between each first fault trajectory. The matching module 202 is used to calculate the matching degree between the real-time trajectory and each fault trajectory signature in the fault trajectory signature library, to obtain the matching distance between the real-time trajectory and each fault trajectory signature, and to compare the minimum value of the matching distance with a preset distance threshold. The first monitoring module 203 is used to calculate the matching confidence level based on each matching distance if the minimum matching distance is greater than or equal to the preset distance threshold, and to execute a graded early warning based on the matching confidence level; The second monitoring module 204 is used to determine the current fault state based on the target fault trajectory signature corresponding to the minimum matching distance if the minimum matching distance is less than the preset distance threshold, and to determine multiple first target fault types with causal relationships based on the target fault trajectory signature and the fault cause-effect graph, and to determine first fault warning information based on the current fault state and each of the first target fault types.

[0087] It is understood that the above system embodiments correspond to the method embodiments of the present invention, and can implement the fault monitoring method provided by any of the above method embodiments of the present invention.

[0088] It should be noted that the system embodiments described above are merely illustrative, and some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the system embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0089] For ease of description and brevity, the system embodiments of the present invention include all the implementation methods described in the above-described fault monitoring method embodiments, and will not be repeated here.

[0090] Based on the above-described embodiments of the fault monitoring method, another embodiment of the present invention provides a terminal device, which includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the fault monitoring method of any embodiment of the present invention.

[0091] For example, in this embodiment, the computer program can be divided into one or more modules, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the terminal device.

[0092] The terminal device may be a desktop computer, laptop, handheld computer, or cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0093] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the terminal device, connecting all parts of the terminal device via various interfaces and lines.

[0094] Based on the above-described method embodiments, another embodiment of the present invention provides a computer-readable storage medium including a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to execute the fault monitoring method described in any of the above-described method embodiments of the present invention.

[0095] The modules / units integrated in the device / terminal equipment, if implemented as software functional units and sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0096] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method of fault monitoring, characterized by, The method comprises the following steps: obtaining a preset fault trajectory signature library and a real-time trajectory of a power distribution device, wherein the fault trajectory signature library comprises a plurality of first fault trajectories and a fault causal graph, each first fault trajectory is associated with a preset first fault type, and the fault causal graph is used to represent the causal relationship between each first fault trajectory; calculating the matching degree of the real-time trajectory and each fault trajectory signature in the fault trajectory signature library to obtain the matching distance between the real-time trajectory and each fault trajectory signature, and comparing the size relationship between the minimum value of the matching distance and a preset distance threshold; if the minimum matching distance is greater than or equal to the preset distance threshold, calculating a matching confidence based on each matching distance, and performing hierarchical early warning based on the matching confidence; if the minimum matching distance is less than the preset distance threshold, determining a current fault state based on a target fault trajectory signature corresponding to the minimum matching distance, determining a plurality of first target fault types having a causal relationship based on the target fault trajectory signature and the fault causal graph, and determining first fault early warning information based on the current fault state and each first target fault type.

2. The failure monitoring method according to claim 1, wherein The matching confidence is calculated based on each matching distance, and hierarchical early warning is performed based on the matching confidence. Specifically: when the maximum matching confidence is less than a preset confidence threshold, the real-time trajectory is marked as a to-be-verified trajectory, a plurality of to-be-verified trajectories in a preset time interval are prioritized based on each maximum matching confidence to obtain a to-be-verified sequence, fault prompt information is generated based on the to-be-verified sequence and a response is given, a second fault trajectory corresponding to a second fault type is obtained in sequence, the fault trajectory signature library is dynamically updated based on the second fault type and the second fault trajectory, and the power distribution device is monitored for faults based on the updated fault trajectory signature library; when the maximum matching confidence is greater than the preset confidence threshold, a plurality of second target fault types having a causal relationship are determined based on the fault trajectory signature corresponding to the maximum matching confidence and the fault causal graph, and second fault early warning information is determined based on each second target fault type.

3. The failure monitoring method according to claim 1, wherein The first fault trajectory is obtained by clustering and analyzing historical fault trajectory data. Specifically: obtain the historical fault trajectory data corresponding to each first fault type from historical operation data; cluster and analyze the historical fault trajectory data to obtain at least one trajectory cluster; for each trajectory cluster, calculate a target center trajectory of the trajectory cluster using a dynamic time warping centroid average algorithm to obtain the first fault trajectory corresponding to the first fault type.

4. The failure monitoring method according to claim 3, characterized by, Wherein, the historical fault trajectory data comprises a plurality of historical fault trajectories, and the clustering and analysis of the historical fault trajectory data to obtain at least one trajectory cluster is specifically as follows: using a dynamic time warping algorithm, calculate the distance between each two historical fault trajectories to obtain a distance matrix; based on the distance matrix, obtain each trajectory cluster through a manifold learning algorithm and a clustering algorithm.

5. The failure monitoring method according to claim 3, wherein The trajectory cluster includes a plurality of first historical trajectories, and the target center trajectory of the trajectory cluster is calculated by using a dynamic time warping and centroid average algorithm for each trajectory cluster, specifically as follows: A first historical trajectory is randomly selected from each trajectory cluster as an initial center trajectory; The initial center trajectory is iteratively optimized, and in each iteration process, a target alignment path of each first historical trajectory and the current center trajectory is calculated by using a dynamic time warping algorithm, the current center trajectory is updated according to all target alignment paths, and an updated center trajectory is obtained, until a preset convergence condition is met, the iteration is stopped, and the target center trajectory is obtained.

6. The failure monitoring method according to Claim 1, wherein The matching confidence is calculated according to each matching distance, specifically as follows: Each matching distance is converted into a corresponding similarity score; The matching confidence of the real-time trajectory and each fault trajectory signature is obtained by calculating all similarity scores through a probability normalization function.

7. The failure monitoring method according to Claim 1, wherein The fault causal graph is obtained by performing time sequence analysis on historical fault events or according to a preset fault rule, specifically as follows: Each first fault type and each second fault type is taken as a node; The causal relationship between each node is determined by analyzing the time sequence correlation between each fault type in the historical fault events or according to the preset fault rule; A directed edge between each node is determined according to each causal relationship, and the fault causal graph is obtained.

8. A fault monitoring system characterized by, It includes: An acquisition module, a matching module, a first monitoring module and a second monitoring module; The acquisition module is configured to acquire a preset fault trajectory signature library and a real-time trajectory of a power distribution device, wherein the fault trajectory signature library includes a plurality of first fault trajectories and a fault causal graph, each first fault trajectory is associated with a preset first fault type, and the fault causal graph is used to represent the causal relationship between each first fault trajectory; The matching module is configured to calculate the matching degree of the real-time trajectory and each fault trajectory signature in the fault trajectory signature library to obtain the matching distance between the real-time trajectory and each fault trajectory signature, and compare the size relationship between the minimum value of the matching distance and a preset distance threshold; The first monitoring module is configured to, if the minimum matching distance is greater than or equal to the preset distance threshold, calculate a matching confidence according to each matching distance, and perform hierarchical early warning according to the matching confidence; The second monitoring module is configured to, if the minimum matching distance is less than the preset distance threshold, determine a current fault state according to a target fault trajectory signature corresponding to the minimum matching distance, determine a plurality of first target fault types having a causal relationship according to the target fault trajectory signature and the fault causal graph, and determine first fault early warning information according to the current fault state and each first target fault type.

9. A terminal device, comprising: It includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, and when the processor executes the computer program, the fault monitoring method of any one of claims 1-7 is implemented.

10. A computer-readable storage medium, characterized in that, It includes: a stored computer program, wherein the computer program, when executed, controls a device in which the computer-readable storage medium is located to perform the fault monitoring method according to any one of claims 1-7.

Images