Intelligent power distribution cabinet fault tracing system based on knowledge graph

Abstract

The invention relates to the field of knowledge graph and information retrieval, in particular to an intelligent power distribution cabinet fault traceability system based on a knowledge graph. The system comprises a data acquisition module used for acquiring static data and dynamic alarm data; the knowledge graph construction module is used for constructing a knowledge graph containing physical attributes based on the data acquired by the data acquisition module; the multi-fault cluster traceability module is used for calculating sequential logic causal indexes among the alarm events based on the knowledge graph; calculating a root asymmetry index of each alarm event based on a sequential logic causal index, wherein the alarm event with the highest root asymmetry index is a root fault alarm; according to the method, the knowledge graph containing the physical attributes is adopted, and the root asymmetry indexes are combined, so that accurate positioning of root fault alarms in massive alarm events is realized.

Description

Technical Field

[0001] This invention relates to the field of knowledge graphs and information retrieval, specifically to a knowledge graph-based intelligent power distribution cabinet fault tracing system. Background Technology

[0002] As a core node in modern power systems, intelligent distribution cabinets deeply integrate high-precision electrical components, complex mechanical transmission structures, and multi-level intelligent control software, making them crucial facilities for ensuring stable power grid operation. However, with the continuous improvement of equipment integration, the coupling between its internal subsystems is deepening, and the causes of faults often exhibit nonlinearity, multi-dimensional coupling, and temporal lag. Currently, the industry's fault diagnosis technology for intelligent distribution cabinets is still at a relatively rudimentary stage, relying excessively on real-time threshold alarms in data acquisition and monitoring control systems. This approach is typically based on preset physical quantity limits, such as temperature exceeding limits, current overload, and voltage drop, or simple expert rule bases for status determination, lacking analysis of the deep-seated operating mechanisms of the equipment.

[0003] In traditional fault tracing practices, this shallow rule-based approach has revealed significant limitations. Due to the highly instantaneous interconnected nature of power systems, when a fundamental physical fault occurs within a distribution cabinet, such as a short circuit or insulation breakdown at a critical node, it often triggers the interlocking actions of multiple upstream and downstream protection devices in the electrical topology network within a millisecond-level time window. This cascading effect leads to a massive, chaotic flood of alarm events received by the monitoring backend, creating an alarm storm. Among these hundreds or thousands of concurrent secondary alarm messages, the key alarms truly reflecting the root cause of the fault are often submerged and extremely difficult to identify quickly. Furthermore, existing diagnostic models heavily rely on fixed parameters set by engineers based on experience, such as fixed action delays and fixed threshold dead zones. These static parameters cannot adapt to dynamic changes caused by equipment aging or altered operating conditions, resulting in high false alarm and false negative rates. Ultimately, fault location work must revert to an inefficient manual mode: maintenance personnel need to retrieve discrete equipment static ledgers, production process documents, and historical maintenance records across systems for time-consuming and laborious manual comparison and analysis. This approach not only severely restricts the efficiency of fault repair and prolongs power outage time, but also fails to meet the urgent needs of smart grids for automated and intelligent operation and maintenance. Summary of the Invention

[0004] To address the problem that traditional methods cannot accurately identify key alarms with root causes when faced with massive and disorganized alarm events, this invention provides a knowledge graph-based intelligent power distribution cabinet fault tracing system. The system includes the following modules: The data acquisition module is used to acquire static data and dynamic alarm data of the intelligent power distribution cabinet; The knowledge graph construction module constructs a knowledge graph containing physical attributes based on the static data. The knowledge graph includes device nodes, relation edges, and edge attributes. The multi-fault cluster tracing module calculates the temporal logical causal index between each alarm event based on the dynamic alarm data and the knowledge graph, which is used to assess the probability that one alarm event will trigger another alarm event; based on the temporal logical causal index, it calculates the inbound causal entropy, outbound causal entropy, inbound total amplitude, and outbound total amplitude of each alarm event to determine the root cause asymmetry index of each alarm event, which is used to determine the root cause fault alarm.

[0005] This invention constructs a knowledge graph containing physical attributes and combines it with the causal asymmetry assessment of alarm events. This enables the identification of the true root cause of fault alarms in chaotic alarm events, reducing reliance on fixed thresholds and simple rules, and improving the automation level and accuracy of fault tracing.

[0006] Furthermore, the time-series logical causality index satisfies: ; in, It is a time-series logical causality indicator used to determine the probability that alarm event A is the direct cause of alarm event B. , These are the source devices for alarm events A and B, respectively. for arrive The set of all paths For one of the paths, For path Upper Dynamic health indicators of each intermediate component The observation time difference between alarm event B and alarm event A. For path Expected time indicators For path The tolerance variance index.

[0007] This invention evaluates multiple physical paths between devices and combines dynamic health indicators with time matching to obtain the quantitative probability of causal relationships between alarm events. Compared with traditional timing judgment, this indicator incorporates electrical logic and physical latency, making its judgment of causal relationships more reliable.

[0008] Furthermore, the dynamic health index satisfies: ; in, For the first Real-time communication error count of each intermediate component This is the maximum tolerance for packet loss. For the first The first intermediate component Real-time monitoring indicators for each parameter As a health benchmark, This represents the maximum allowable deviation.

[0009] The dynamic health index evaluated in this invention comprehensively considers the communication status and physical parameter deviations of components, and can reflect the real health status of intermediate links in the alarm event propagation path in real time, ensuring the effectiveness of causal chain reasoning.

[0010] Furthermore, the expected time indicator is determined by the path The expected action times of all intermediate components are summed to obtain the tolerance variance index, which is derived from the path. The time variance of the action response of all intermediate components is summed to obtain the result.

[0011] Furthermore, the root cause asymmetry index satisfies: ; in, The root cause asymmetry indicator for alarm event A. Let A be the total outgoing amplitude of alarm event A. Let A be the outward causal entropy of alarm event A. Let A be the total incoming amplitude of alarm event A. Let A be the inward causal entropy of alarm event A. To prevent extremely small positive numbers with a denominator of zero.

[0012] The root cause asymmetry index applied in this invention evaluates the causal characteristics of alarm events from two dimensions: inbound and outbound. A true root cause fault alarm has an extremely low inbound index and an extremely high outbound index. This model accurately captures this core characteristic of root cause fault alarms.

[0013] Furthermore, the outgoing total amplitude is the sum of the temporal logical causal indicators of all other alarm events triggered by alarm event A, and the incoming total amplitude is the sum of the temporal logical causal indicators of all other alarm events triggered by alarm event A.

[0014] Furthermore, the outward causal entropy is the ratio of the temporal logical causal index that alarm event A triggers any other alarm event to the total outward amplitude of alarm event A, obtained through the calculation principle of standard information entropy.

[0015] This invention assesses the complexity of the causes and consequences of alarm events by introducing information entropy. An event that triggers multiple different downstream alarm events has high outward entropy, while an event without any preceding cause has low inward entropy. This assessment of complexity enhances the model's ability to distinguish between root cause fault alarms and intermediate alarm events.

[0016] Furthermore, the incoming causal entropy is derived from the ratio of the temporal logical causal index of alarm event A caused by any other alarm event to the total incoming magnitude of alarm event A, using the calculation principle of standard information entropy.

[0017] Furthermore, the static data includes electrical schematics, equipment ledgers, product specifications, and maintenance records.

[0018] Furthermore, the dynamic alarm data includes alarm event ID, timestamp, and device ID.

[0019] The present invention has the following technical effects: This invention addresses the severe challenges faced by traditional fault diagnosis methods when dealing with complex faults in intelligent power distribution cabinets, particularly when cascading trips cause a massive number of secondary alarm events, leading to data feature ambiguity and distortion. It proposes a knowledge graph-based fault tracing system for intelligent power distribution cabinets.

[0020] This invention abandons the static estimation parameters or simple rule-matching modes that rely heavily on manual experience in traditional technologies. Instead, it adopts a data-driven strategy. In the process of constructing a knowledge graph that includes physical attributes, it creatively embeds the inherent physical characteristics of the device, including the precise value of the expected action time and the time variance of the action response, directly into the nodes and relationship attributes of the graph. Through this physical modeling, the system can accurately quantify the temporal logical causal strength between different alarm events and effectively filter noise interference caused by network latency or signal jitter.

[0021] Ultimately, this invention can quickly and accurately extract the root cause fault alarms that trigger a chain reaction from the complex alarm data stream, greatly improving the accuracy and robustness of fault location, thereby significantly reducing the troubleshooting workload of maintenance personnel and greatly improving the fault response speed and overall maintenance efficiency of the power system. Attached Figure Description

[0022] Figure 1 This is a block diagram of a knowledge graph-based intelligent power distribution cabinet fault tracing system according to an embodiment of the present invention.

[0023] Figure 2 This is a comparison diagram of the effects of the embodiments of the present invention and the traditional system. Detailed Implementation

[0024] This invention discloses a knowledge graph-based intelligent power distribution cabinet fault tracing system. By transforming discrete electrical information into a reasonable knowledge network and combining it with advanced graph algorithms, it can identify the true root cause fault alarms from a large number of millisecond-level alarm events. (Refer to...) Figure 1 This includes modules 101-103, which are described in detail below.

[0025] Data acquisition module 101: Used to collect all the raw information needed to build the knowledge graph and standardize it.

[0026] 1. Static data collection: Collect relevant information about the intelligent power distribution cabinet from multiple departments such as design and operation and maintenance, including but not limited to electrical schematic diagrams, BOM lists, secondary wiring diagrams, product specifications, skill manuals, equipment ledgers, network design procedures, and maintenance and repair records.

[0027] 2. Dynamic data collection: Real-time collection of dynamic alarm data from the intelligent operation and maintenance platform, mainly including alarm event ID, timestamp, and device ID.

[0028] Knowledge Graph Construction Module 102: Used to construct a knowledge graph of physical attributes.

[0029] It should be noted that traditional knowledge graphs only store relationships, lacking physical attributes and logical strength information. This module defines weighted, attributed entities and relationships to directly encode physical information such as the physical time attributes of devices, action uncertainties, and logical costs of relationships into the graph structure, thus realizing the physicalization of knowledge and providing a foundation for subsequent computation.

[0030] 1. Preprocessing: ETL tools are used to clean and standardize the format of data collected from different sources. Vector analysis is used to parse electrical schematics and secondary wiring diagrams to obtain electrical equipment and the connection relationships between them. Finally, a standardized, structured dataset is obtained, defined as the electrical connection table, to be imported into the knowledge graph.

[0031] 2. Extract data: Extract all electrical devices from the electrical connection table, such as intelligent switch cabinets, circuit breakers, protection relays, etc., and create a device list.

[0032] 3. Attribute Preset: Static physical parameters are extracted from the equipment's technical specifications, production process documents, and electrical design standards, serving as attributes for nodes or relationships in the knowledge graph. These static physical parameters include: expected action time (the standard time required for the equipment to respond) and the time variance of the action response (the uncertainty of the standard time).

[0033] 4. Knowledge Graph Construction: Using the Neo4j graph database, the device list is traversed, creating a node for each device as a node in the knowledge graph, and extracting the corresponding attributes. Combined with the electrical connection table, the relationships between corresponding nodes are used as relationships in the knowledge graph, and the attributes of these relationships are also stored in the graph.

[0034] Finally, the output is a knowledge graph containing complete electrical topology, device information, and key attributes attached to nodes and relationships, including physical attributes.

[0035] Multi-fault cluster tracing module 103: Used to find the true root cause of fault alarms.

[0036] It should be noted that the core of this module is to identify and separate the true root cause fault alarms by analyzing the characteristics of alarm events.

[0037] In a complex electrical system, a true root cause fault alarm is rarely triggered by upstream alarm events, but it can trigger a series of downstream cascading effects—that is, a single root cause fault alarm can trigger a multitude of alarm events. For example, a short circuit in a critical transformer often generates multiple alarm events. This module, based on the characteristics of root cause fault alarms, assesses the probability that an alarm event is a true root cause fault alarm, thereby accurately locating the root cause fault alarm within complex alarm event data.

[0038] 1. Timing logic causality index For any two alarm events, let's name them alarm event A and alarm event B. We need to determine the credibility of the hypothesis that alarm event A is the direct cause of alarm event B. On one hand, logically, using electrical principles, we need to determine whether it's reasonable for alarm event A to trigger alarm event B. On the other hand, temporally, we need to determine whether the time of occurrence of alarm event B precisely matches the expected action time required for alarm event A to trigger alarm event B. Only when both conditions are met simultaneously will the hypothesis that alarm event A is the direct cause of alarm event B have a high degree of credibility.

[0039] It should be noted that each alarm event corresponds to a source device that triggered the alarm event. The source device is represented as a node in the knowledge graph. There may be multiple paths between any two nodes. For each path, in addition to the nodes corresponding to the two source devices, the other nodes on the path are named intermediate components.

[0040] It should be noted that, to improve system feasibility, in the design of power systems, there are multiple paths between two related devices, such as a protective relay and a circuit breaker. These paths are constructed in the knowledge graph building module and contain physical attribute knowledge graph paths. For example, the protective relay sends a signal to the circuit breaker's intelligent controller via optical fiber, commanding it to trip. However, if the optical fiber is severed or the controller software malfunctions, this path becomes invalid. In this case, a backup circuit is needed, connected to the circuit breaker's backup trip coil via a traditional copper core cable. Although slightly slower, this is more stable. With this design, tripping is guaranteed regardless of which path is used. Therefore, it is necessary to analyze each path separately and comprehensively evaluate the probability that alarm event A is the direct cause of alarm event B.

[0041] The causal index of time sequence logic satisfies the following relation:

[0042] in: It is a time-series logic causal indicator used to determine the probability that alarm event A is the direct cause of alarm event B; , It is the source device that triggered alarm events A and B; It is a set of paths, representing the path sequence within a knowledge graph that includes physical attributes, from... arrive All paths; yes The Path; It is the first The first on the path One intermediate component; It is a dynamic health indicator; It is the observation time difference, which is obtained by subtracting the timestamp of alarm event A from the timestamp of alarm event B; It is an expected time indicator, derived from the first The expected action times of all intermediate components along the path are summed to obtain the result. It is the tolerance variance index, which is composed of each... The sum of the time variances of the action responses of all intermediate components along the path is used to obtain the value representing the first path. The allowable error range for each path.

[0043] It should be noted that when If a system crash occurs, then , At this time, the path This does not prove the necessity of designing multiple paths for two related devices.

[0044] When path All intermediate components are running normally; at this point, in addition to considering health status... We also need to consider the matching degree of timestamps: This item analyzes the difference in timestamps between two alarm events. Compared with expected time indicators Deviation and tolerance variance index To evaluate the path The time uncertainty; if alarm event A is the direct cause of alarm event B, from From the time alarm event A was generated until The time difference between the generation of alarm event B and the time difference between the generation of alarm event B should be approximately equal to the time difference between the generation of alarm event B and ... arrive The total time of all intermediate components along the path; if the deviation is large, it means that alarm event A is not the direct cause of alarm event B.

[0045] It can be seen that, through analysis from arrive The health status of each intermediate component along the path is used to determine the connectivity of the path. The difference between the timestamps of two detected alarm events and the deviation from the physical standard time of the path are calculated. The two are combined to comprehensively analyze the probability that alarm event A is the direct cause of alarm event B.

[0046] Dynamic health indicators meet the following requirements:

[0047] Let the first Each intermediate component has Each parameter has a real-time monitoring index. Combined with the health benchmark value of this parameter specified in the equipment's product specifications. and the maximum allowable deviation This yields the degree of deviation of the parameter; then, the first [parameter] is read from the network log. Real-time communication error count of each intermediate component This example uses the number of packet losses in the past minute, combined with the packet loss tolerance limit within one minute specified in the network design specifications. This yields the degree of communication deviation for that parameter; ultimately, as long as the first... If any physical parameter of an intermediate component, such as temperature or voltage, deviates from its normal value, or if packet loss occurs during communication, If they all become smaller, it indicates that the health of the intermediate component has decreased.

[0048] In intelligent power distribution cabinet fault tracing scenarios, this time-series logic causal indicator relationship is used to determine whether an upstream alarm event, such as a relay protection action, is a direct cause of a downstream alarm event, such as a circuit breaker tripping. For example, if the system finds two paths—fiber optic and cable—from the relay to the circuit breaker in a knowledge graph containing physical attributes, the relationship will evaluate these two paths separately. If the dynamic health index of the fiber optic path... Very high, and the observed time difference , and the expected time index of the fiber optic path In the tolerance variance index If the path matches highly, it will contribute a high confidence value.

[0049] In traditional systems, maintenance personnel need to make manual judgments. They might check logs and find that the relay action time is 10.00 seconds and the circuit breaker tripping time is 10.01 seconds. Because the time interval is as long as one second, far exceeding the expected action time, they might incorrectly rule out a causal relationship between the two. Alternatively, they might rely solely on empirical rules, such as the assumption that a relay action will always trigger a circuit breaker, ignoring timing mismatches or intermediate links, such as the actual health status of fiber optic cables. The system of this invention, through a quantitative model, integrates topology logic, dynamic health indicators, and physical latency, and its judgment results are more accurate and automated than manual experience or simple rule bases.

[0050] 2. Root cause asymmetry indicators By combining the temporal logic causal indicators between alarm events, the probability of each alarm event being a root cause fault alarm is assessed by judging how each alarm event is triggered by other alarm events and how it triggers other alarm events.

[0051] The root cause asymmetry index satisfies:

[0052] in: It is the root cause asymmetry indicator for alarm event A. The larger the value, the higher the probability that alarm event A is a root cause fault alarm. It is the total outgoing amplitude of alarm event A, representing the sum of probabilities that alarm event A directly triggers all other alarm events, i.e., all... The sum of A and B represents all alarm events except A. It is the outward causal entropy of alarm event A, representing the complexity of the consequences directly caused by alarm event A; It is the total incoming amplitude of alarm event A, representing the sum of the total probabilities of all other alarm events triggering alarm event A; It is the inward causal entropy of alarm event A, representing the complexity of the cause that triggered alarm event A; It is a very small positive number, such as To prevent the denominator from being 0.

[0053] It can be seen that when an alarm event is a root cause fault alarm, it will trigger many alarm events. The item is very large, and other alarm events will not trigger root cause fault alarms. The value is very small; only alarm events that simultaneously meet both of these characteristics will have a large root cause asymmetry index.

[0054] It should be noted that outward causal entropy and incoming causal entropy It is calculated using the standard information entropy formula, namely:

[0055] It can be seen that if alarm event A only triggers B, then The consequences are simple: a single root cause fault often generates multiple alarm events, corresponding to... It will be very high, and the final result will be... It will also be very large;

[0056] Similarly, a true root cause fault alarm is often one that cannot be physically triggered by any alarm event; its corresponding It approaches 0, and finally gets It will be very big.

[0057] When a short-circuit fault occurs in the distribution cabinet, this root cause asymmetry index formula is used to filter the true source from a massive number of alarm events. Suppose alarm event D, such as a transformer short circuit, triggers alarm events A, B, and C, such as a downstream switch tripping. At this time, the total incoming amplitude of alarm event D... and incoming causal entropy All values ​​are close to zero, meaning no alarm event triggered it; and its total outgoing amplitude is close to zero. outward causal entropy The score is very high, meaning it triggered multiple downstream alarm events. Conversely, alarm event A, such as a circuit breaker tripping, has a very high incoming index, meaning it was triggered by D, while the outgoing index is very low, meaning it did not trigger any other alarm events. By calculating the root cause asymmetry index, alarm event D will receive an extremely high score, while alarm events A, B, and C will receive extremely low scores.

[0058] When a large number of alarm events are generated, the system iterates through the alarm event list, performs the above calculations on each alarm event, and filters out the alarm event with the largest value of the root cause asymmetry index that meets the criteria of having no upstream alarm events but having a series of downstream alarm events. This alarm event is the true root cause fault alarm. Figure 2This is a comparison diagram of the effects of the embodiment of the present invention and the traditional system. It can be seen that the alarm event E obtained by tracing the fault source of the intelligent power distribution cabinet through the traditional method is only the earliest alarm event in terms of time, and is not the real root cause fault alarm. However, the alarm event D selected by the present invention through comprehensive evaluation of the characteristics of the root cause fault alarm is the real root cause fault alarm of the intelligent power distribution cabinet.

Claims

1. A knowledge graph-based intelligent power distribution cabinet fault tracing system, characterized in that, Includes the following modules: The data acquisition module is used to acquire static data and dynamic alarm data of the intelligent power distribution cabinet; The knowledge graph construction module constructs a knowledge graph containing physical attributes based on the static data. The knowledge graph includes device nodes, relation edges, and edge attributes. The multi-fault cluster tracing module calculates the temporal logical causal index between each alarm event based on the dynamic alarm data and the knowledge graph, which is used to assess the probability that one alarm event will trigger another alarm event. Based on the aforementioned time-series logic causal index, the incoming causal entropy, outgoing causal entropy, total incoming amplitude, and total outgoing amplitude of each alarm event are calculated to determine the root cause asymmetry index of each alarm event, which is used to determine the root cause fault alarm.

2. The intelligent power distribution cabinet fault tracing system based on knowledge graphs according to claim 1, characterized in that, The time-series logical causality index satisfies: ； in, It is a time-series logical causality indicator used to determine the probability that alarm event A is the direct cause of alarm event B. , These are the source devices that triggered alarm events A and B, respectively. for arrive The set of all paths For one of the paths, For path Upper Dynamic health indicators of intermediate components The observation time difference between alarm event B and alarm event A. For path Expected time indicators For path The tolerance variance index.

3. The intelligent power distribution cabinet fault tracing system based on knowledge graph as described in claim 2, characterized in that, The dynamic health index satisfies: ； in, For the first Real-time communication error count of each intermediate component This is the maximum tolerance for packet loss. For the first The first intermediate component Real-time monitoring indicators for each parameter As a health benchmark, This represents the maximum allowable deviation.

4. The intelligent power distribution cabinet fault tracing system based on knowledge graph as described in claim 2, characterized in that, The expected time indicator is determined by the path The expected action times of all intermediate components are summed to obtain the tolerance variance index, which is derived from the path. The result is obtained by summing the time variances of the action responses of all intermediate components.

5. The intelligent power distribution cabinet fault tracing system based on knowledge graphs according to claim 1, characterized in that, The root cause asymmetry index satisfies: ； in, This is an indicator of the root cause asymmetry of alarm event A. Let A be the total outgoing amplitude of alarm event A. Let A be the outward causal entropy of alarm event A. Let A be the total incoming amplitude of alarm event A. Let A be the inward causal entropy of alarm event A. To prevent extremely small positive numbers with a denominator of zero.

6. The intelligent power distribution cabinet fault tracing system based on knowledge graph as described in claim 5, characterized in that, The outgoing total amplitude is the sum of the temporal logic causal indicators of all other alarm events triggered by alarm event A, and the incoming total amplitude is the sum of the temporal logic causal indicators of all other alarm events triggered by alarm event A.

7. The intelligent power distribution cabinet fault tracing system based on knowledge graphs according to claim 5, characterized in that, The outward causal entropy is the ratio of the temporal logical causal index that alarm event A triggers any other alarm event to the total outward amplitude of alarm event A, derived using the calculation principle of standard information entropy.

8. The intelligent power distribution cabinet fault tracing system based on knowledge graph as described in claim 5, characterized in that, The incoming causal entropy is the ratio of the temporal logical causal index of alarm event A caused by any other alarm event to the total incoming magnitude of alarm event A, and is derived using the calculation principle of standard information entropy.

9. The intelligent power distribution cabinet fault tracing system based on knowledge graph as described in claim 1, characterized in that, The static data includes electrical schematics, equipment ledgers, product specifications, and maintenance records.

10. A knowledge graph-based intelligent power distribution cabinet fault tracing system according to claim 1, characterized in that, The dynamic alarm data includes alarm event ID, timestamp, and device ID.

Images