A power transmission network fault location system based on edge agent collaborative decision-making

Abstract

The application discloses a power transmission network fault location system based on edge agent collaborative decision-making, and belongs to the technical field of power system fault diagnosis. The system comprises a token relay communication module, a consensus arbitration module and an unknown event cognition module. The token relay communication module performs ordered token directional relay transmission among edge agents according to the physical propagation direction of a fault energy wave front. The consensus arbitration module diagnoses and arbitrates the health degree of a preliminary positioning conclusion by jointly evaluating decision entropy and physical space correlation, so as to inhibit false consensus caused by systematic bias. When an unknown fault with a semantic gap existing in a local knowledge base is detected, the unknown event cognition module stops positioning and generates a structured feature sketch report. The application fundamentally improves the reliability, robustness and safety of fault location by constructing an ordered collaborative mechanism throughout the whole process.

Description

Technical Field

[0001] This invention relates to the field of power system fault diagnosis technology, and in particular to a power transmission network fault location system based on edge agent collaborative decision-making. Background Technology

[0002] Fault location in power transmission networks is fundamental to ensuring the safe operation of the power grid. In existing technologies, a common approach is to utilize multiple edge agents deployed within the transmission network for collaborative fault location. These edge agents exchange fault information they have collected via a communication network and jointly determine the fault location based on a pre-defined algorithm.

[0003] However, in practical applications, existing cooperative positioning systems of this type exhibit inherent disorder in their cooperative process, leading to unreliable positioning results under complex conditions. This disorder manifests at several levels: at the communication level, broadcast communication is commonly used, which can easily cause network data congestion during faults and is highly dependent on high-precision time synchronization among all agents. At the decision-making level, consensus is often reached through voting or weighted averaging mechanisms. These mechanisms struggle to effectively identify and suppress erroneous consensuses when faced with multiple agents simultaneously reporting systematically biased data due to localized common interference sources. At the cognitive level, existing systems are primarily designed for known fault modes. When faced with unknown fault types that do not match any patterns in the knowledge base, their handling methods lack clear definitions, potentially resulting in unreliable classification results. Summary of the Invention

[0004] This invention provides a power transmission network fault location system based on edge agent collaborative decision-making to solve the technical problem that the fault location is not reliable under complex operating conditions due to the inherent disorder in the collaborative process in the prior art.

[0005] In view of the above problems, the present invention provides a power transmission network fault location system based on edge agent collaborative decision-making, comprising:

[0006] The token relay communication module is configured to, when a power transmission network fault is detected, relay tokens between multiple edge agents according to the physical propagation direction of the fault energy wavefront, so as to form an ordered token chain containing local measurement information and corresponding timestamps of each agent.

[0007] The consensus arbitration module is configured to diagnose and arbitrate the health of the preliminary fault location conclusion by jointly evaluating the decision entropy and physical space correlation of the agents that have reached a consensus after obtaining a preliminary fault location conclusion based on the ordered token chain.

[0008] The unknown event recognition module is configured to suspend the localization task and generate a structured feature sketch report describing the peculiar features of the unknown fault event when it detects an unknown fault event that has a preset semantic gap with the local knowledge base.

[0009] The technical solution provided in this application has at least the following technical effects or advantages:

[0010] By employing token relay communication based on the physical propagation direction of the fault energy wavefront, the orderly transmission of fault information is achieved, reducing the dependence on communication network bandwidth and high-precision global time synchronization.

[0011] By introducing a joint assessment of the correlation between decision entropy and physical space after reaching preliminary conclusions, a diagnostic and arbitration mechanism for the health of consensus conclusions is established, which enhances the system's ability to suppress systematic biases and erroneous consensuses.

[0012] By setting a mechanism to suspend localization and generate a structured feature report when faced with unknown fault events, a safe handling mode outside the cognitive boundary is defined for the system, avoiding misjudgments caused by forced classification. Attached Figure Description

[0013] Figure 1 This is a schematic diagram of the overall architecture of a power transmission network fault location system based on collaborative decision-making by edge agents according to the present invention. Detailed Implementation

[0014] The above technical solutions will now be described in detail with reference to the accompanying drawings and specific embodiments to provide a better understanding of them. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. It should be understood that the present invention is not limited to the exemplary embodiments used only to explain the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention. Furthermore, it should be noted that, for ease of description, only the parts related to the present invention are shown in the drawings, not all of them.

[0015] See Figure 1 One embodiment of the present invention provides a power transmission network fault location system based on edge agent collaborative decision-making.

[0016] This system is deployed within the physical environment of the power transmission network, which includes facilities such as transmission lines, towers, and substations. The core of the system consists of multiple distributed Edge Intelligent Agents (EIAs), physically deployed at key nodes along the transmission network, such as those installed on towers or integrated into the protection and control devices of substations. Each EIA is an independent hardware unit, containing at least a processor for high-speed data processing, a data acquisition interface for connecting current transformers or transient traveling wave sensors, a storage unit for storing local knowledge bases and operational data, and a communication unit for interacting with other EIAs.

[0017] Logically, the system comprises multiple functional modules, which, as software entities, are deployed within each of the distributed edge agents. Among them, the token relay communication module organizes an ordered, single-line information relay among relevant agents based on the physical propagation direction of the fault energy wavefront, replacing traditional broadcast communication. The consensus arbitration module, after the system reaches a preliminary fault location conclusion, initiates a metacognitive review of the decision quality, diagnosing and arbitrating the health of the conclusion by quantitatively evaluating the internal consistency and external spatial distribution characteristics of the consensus. The unknown event cognition module, when the system encounters an unknown fault mode that its local knowledge base cannot recognize, can safely suspend the regular processing flow and instead generate a structured intelligence report containing key singular features for reporting. Furthermore, the system includes a fault location module, which receives and parses the ordered information chain generated by the token relay communication module and performs specific location calculations.

[0018] To ensure the normal operation of the system, basic data pre-configuration is required during the deployment phase. Each edge agent's local knowledge base needs to pre-store a set of feature vectors for known fault modes; these vectors are used for subsequent fault identification and matching. Simultaneously, the system needs to pre-store the topology information of the entire transmission network, particularly the relationship between each agent and its physically or electrically adjacent agents. In a specific embodiment, the feature vectors are constructed through the following steps: First, a large amount of historical fault waveform data of known types (e.g., single-phase grounding, phase-to-phase short circuit, etc.) is collected; then, for the initial transient stage of each waveform data, multi-resolution wavelet transform is used to decompose it, obtaining detail coefficients at different scales; next, from these detail coefficients, a set of statistical features that can stably characterize the fault type is calculated, such as the energy value of the first-scale detail coefficient, the kurtosis value of the third-scale detail coefficient, and the modulus maxima of the initial traveling wave front; finally, these calculated statistical feature values ​​are combined into a fixed-dimensional numerical vector according to a predefined order, and this vector is stored in the knowledge base as a feature vector for the fault mode.

[0019] In one embodiment, the fault location workflow is characterized by orderly and collaborative processes throughout:

[0020] The fault location process described in this invention begins with the generation and transmission of ordered information, which is executed by a token relay communication module deployed in an edge agent.

[0021] The process is initiated by the initial fault detection and the establishment of the chair agent. When a fault occurs in the power transmission network, transient electrical signals such as traveling waves are generated at the fault point and propagate to both sides of the line. Edge agents deployed along the line continuously monitor the current or voltage signals on the line at a high sampling rate through their data acquisition interfaces. When the amplitude or rate of change of a signal detected by an agent exceeds a first preset threshold within a preset time window, a potential fault event is triggered. To confirm the fault, the agent further extracts the waveform features of the signal and matches them with typical fault feature vectors stored in its local knowledge base. When the confidence level of the match, i.e., the value obtained by calculating the cosine similarity between feature vectors, is higher than a second preset threshold (e.g., 0.95), the agent determines that the fault has occurred with high confidence and immediately establishes itself as the chair agent for this fault handling process.

[0022] In a preferred embodiment, to handle situations where multiple agents trigger faults approximately simultaneously or at multiple points, the system employs a lightweight chairman election rule. Each candidate agent that detects a high-confidence fault calculates a weighted candidate score based on locally measured signal energy, matching confidence with the knowledge base, and background noise levels. Subsequently, within a very short, pre-defined election window (e.g., 5 milliseconds), the agent broadcasts an election message containing the event ID, candidate score, and local timestamp to its one-hop neighbor. Within this window, if an agent receives an election message from another candidate for the same event ID but with a higher score, it automatically relinquishes its chairmanship and becomes a follower. If no message with a higher score is received within the window, the agent officially becomes the chairman. This rule ensures that, in most cases, only one optimal agent becomes the chairman and initiates token passing.

[0023] After establishing the chairman agent, the initial token generation step will be executed. The chairman agent creates a structured data object, the initial token, in its internal storage. This token's data structure includes at least the following fields: a globally unique event ID; a chairman agent ID field, filled with its unique identifier; an initial fault feature field, containing key time-frequency domain feature vectors extracted after wavelet transform of the original traveling wave signal that triggered the fault; and a high-precision local timestamp field, which precisely marks the moment the fault traveling wave front arrived at the agent's sensor.

[0024] In a preferred embodiment, to prevent token forgery or replay attacks during transmission, the system employs a message authentication code (HMAC)-based security mechanism. When generating or updating a token, the sending agent uses a shared key with the receiver to calculate the message authentication code (HMAC) for the entire token message and appends this code as a security field to the end of the message. The message also includes a monotonically increasing sequence number. Upon receiving the token, the receiving agent independently calculates the HMAC of the received message using the same shared key and compares it with the HMAC carried in the message. The token is accepted only if both match perfectly and the message sequence number is valid. This security mechanism ensures the integrity and authenticity of the token.

[0025] After the initial token is generated, a dynamic determination step for the relay path is performed. In a specific embodiment, the chairman agent determines the propagation direction using the following algorithm: Assuming the agent connects to two lines, port A and port B, it compares the precise timestamps of the arrival of the fault traveling wave front recorded by its sensors at ports A and B. If the timestamp at port A is earlier than the timestamp at port B, and the time difference between the two is greater than a minimum identifiable time interval (e.g., 5 microseconds) determined by the line parameters, then it is determined that the fault energy is coming from the direction of port A and propagating towards port B. Therefore, the next adjacent agent connected to port B is selected as the target for token transmission.

[0026] After determining the next-hop target, the single-line relay and information appending process of the token begins. The chairman agent sends the initial token to the determined next-hop neighboring agent via a point-to-point secure communication protocol. Upon receiving the token, the neighboring agent first verifies the event ID, and then performs an information appending operation, that is, appending a new data block containing its own ID, local sensor measurement information, and a locally recorded arrival timestamp of the traveling wave front to the data payload of the token. After completing the information appending, the agent repeats the relay path determination steps described above to determine the direction of continued energy propagation and continues to pass on its updated token. This single-line relay process, which strictly follows the physical propagation path of the fault, continues until a preset termination condition is met. Finally, an ordered token chain data object that aggregates the measurement information and precise timing information of multiple agents along the fault propagation path is formed.

[0027] In a preferred embodiment, to address the possibility of token loss or link breakage during transmission due to communication interruption, the system employs a layered fault tolerance and fallback strategy. When an agent issues a token, if it fails to receive confirmation from the next-hop node within a preset timeout period (e.g., 200 milliseconds), it initiates a recovery process. This process first involves a limited number of retries (e.g., 3). If the retries still fail, the agent queries its pre-stored adjacency topology table, selects a logically valid alternative next-hop node, and attempts to redirect the token to that node. If all alternative paths fail, the agent triggers a limited short-term broadcast, sending a distress message to all its one-hop neighbors to collect surrounding observation data as supplementary information. If all automatic recovery measures fail, the agent ultimately encapsulates the current token snapshot with detailed local sampling data, reports it as a high-priority anomaly to the cloud master station, and triggers manual review.

[0028] After the ordered token chain is formed and aggregated, the system will initiate a preliminary fault location process, executed by the fault location module. This process begins with the aggregation and parsing of the ordered token chain. In one embodiment, when the token relay terminates, the edge agent holding the final version of the token chain will send the complete ordered token chain data object to two pre-designated boundary agents located at opposite ends of the fault region. Upon receiving the complete token chain, the fault location modules of these two boundary agents (hereinafter referred to as Agent M and Agent N) first parse it, extracting the ID sequence of all agents recorded in the token chain and the local timestamp corresponding to each ID.

[0029] After the analysis is completed, the fault location module will execute the core fault segment calculation method. First, the module identifies two adjacent agents whose fault traveling wave signal intensity changes significantly, thereby determining that the fault point is located on the transmission line between these two agents (denoted as agent i and agent i+1).

[0030] Next, the module uses a two-end ranging principle for calculation. The specific implementation steps of this calculation are as follows: First, calculate the arrival time difference between the two ends, that is, subtract the timestamp recorded by agent M from the timestamp recorded by agent N to obtain a time difference value T_diff. Second, obtain the lengths of four key lines from the pre-stored line topology data: the total line length L_Mi from agent M to agent i, the line length L_ii1 from agent i to agent i+1, and the total line length L_i1N from agent i+1 to agent N. Third, set an unknown x to represent the distance of the fault point from agent i. Based on this, the distance from the fault point to the two boundary agents is expressed using text and known parameters: the distance to agent M can be expressed as the sum of L_Mi and x; the distance to agent i+1 can be expressed as the difference between L_ii1 and x; the distance to agent N can be expressed as the sum of L_i1N and (L_ii1 minus x). The fourth step is to calculate the distance difference D_diff from the fault point to the two boundary agents. This difference can be expressed as (L_Mi plus x) minus (L_i1N plus L_ii1 minus x). The fifth step is to establish and solve the localization equation. The left side of the equation is the distance difference D_diff calculated in the previous step divided by the known traveling wave propagation speed v, and the right side is the time difference T_diff calculated in the first step. Since this equation contains only one unknown x, the value of x can be solved through simple algebraic rearrangement and simplification, thus determining the precise location of the fault point. Agent N will also independently execute the exact same calculation process for cross-validation. After each boundary agent independently calculates the fault point location, they will encapsulate the calculation results into a preliminary fault localization conclusion data object.

[0031] After the fault location module outputs a preliminary fault location conclusion, to ensure the reliability of this conclusion under complex operating conditions and to suppress potential systematic biases, the system will initiate a consensus health diagnosis and arbitration process. This process is executed by the consensus arbitration module, and its trigger condition is that one or more agents reach a preliminary fault location conclusion and form a preliminary consensus.

[0032] The consensus arbitration module reviews the preliminary consensus by parallel computing two diagnostic indicators. The first indicator is decision entropy. In a specific embodiment, the decision entropy is calculated as follows: First, the target segment of the transmission network is discretized into multiple consecutive small segments with unique numbers. Next, the fault location conclusions output by all participating agents are collected, and each location is mapped to its corresponding discrete small segment number. Then, the number of agent judgments obtained for each small segment is counted, thus forming a probability distribution describing the fault location judgment. Finally, based on this probability distribution, the decision entropy value of this consensus is obtained using the information entropy calculation formula.

[0033] The second metric used in parallel computation is physical spatial correlation. In a specific embodiment, this physical spatial correlation is calculated as follows: First, the pre-defined geographical coordinates of all agents participating in the current valid consensus are obtained. Then, a spatial statistical algorithm, such as the average nearest neighbor distance index, is used to quantify the degree of clustering of these agents geographically. The calculation process involves finding the next agent with the closest geographical distance to each agent and recording that nearest neighbor distance; then, the average of all these nearest neighbor distances is calculated. This average value serves as the quantitative metric for physical spatial correlation.

[0034] After calculating the decision entropy and physical space correlation, the system executes the logic for determining unhealthy consensus. The consensus arbitration module compares the calculated decision entropy value with a preset first threshold, and the calculated physical space correlation index with a preset second threshold. When the decision entropy is lower than the first threshold and the physical space correlation is higher than the second threshold, an unhealthy consensus event is determined. This logic aims to identify situations where multiple neighboring agents collectively output highly consistent but erroneous conclusions, potentially caused by local or regional common interference sources.

[0035] Once an unhealthy consensus is determined, the arbitration mechanism will immediately execute pre-set response actions. In one embodiment, the actionable actions include at least one or more of the following: First, reject the conclusion, that is, mark the preliminary fault location conclusion as invalid and generate an alarm log; Second, request supplementary data, that is, the system sends instructions to other intelligent agents located outside the aggregation area, requiring them to perform independent analysis and calculation of the fault signal for cross-validation; Third, escalate the event, that is, report the location event, along with the original token chain data, the preliminary conclusion, and the diagnostic report, to the cloud main station or a human operation and maintenance expert for review.

[0036] At any stage of the fault handling process, if the fault signal captured by the edge agent cannot be effectively identified by its local knowledge base, the system will initiate the unknown event cognition and handling process. This process is executed by the unknown event cognition module, which aims to safely manage the cognitive boundaries of the system.

[0037] The process begins with the detection of a semantic gap. The unknown event recognition module first extracts the feature vector of the currently captured fault signal. In a specific embodiment, this module calculates the matching degree between the feature vector of the current fault signal and the feature vectors of all known fault patterns preset in the local knowledge base. This matching degree can be quantified by calculating the cosine similarity or the reciprocal of the Euclidean distance between the two vectors. The module compares all the calculated matching degree values ​​with a preset confidence threshold. When all matching degree values ​​are lower than the confidence threshold, the system determines that there is a preset semantic gap between the current fault event and all known patterns, and identifies it as an unknown event.

[0038] Upon detecting a semantic gap, the system will execute a pre-defined behavior pattern transition. The unknown event recognition module will immediately suspend all routine fault localization and classification tasks currently being performed by the agent. The system's role at this point changes from fault decision-maker to on-site intelligence gatherer and reporter.

[0039] After the role switch, the core task of the Unknown Event Recognition Module is to generate a structured feature sketch report. In a specific embodiment, this structured feature sketch report includes at least the following explicitly defined fields: First, the time-domain information of the signal's singularities, which records the rise time, amplitude, and precise timestamps of the first and subsequent zero-crossings of the steepest part of the fault signal wavefront. Second, the time-frequency domain energy distribution characteristics of the signal, which stores a two-dimensional matrix obtained by performing a short-time Fourier transform or wavelet transform on the original signal. Third, the differential feature analysis with the most similar known fault mode, which records the ID of the known mode with the highest matching degree to the current signal in the local knowledge base, and lists the indices and specific differences of the top few dimensions of the feature vector of the current signal with the largest numerical differences from the most similar mode. After the feature sketch report is generated, the system will initiate a reporting protocol. The Unknown Event Recognition Module marks the report as the highest priority and sends it to the cloud master station or the designated operation and maintenance expert terminal through the communication unit.

[0040] It is important to note that the multiple preset thresholds mentioned in this embodiment are determined using a method designed to ensure a balance between the system's sensitivity and reliability. In a specific embodiment, these thresholds can be determined as follows: First, a simulation test platform containing a large amount of historical real fault data and interference signal data under normal operating conditions is constructed. Then, for each key threshold (e.g., the first threshold for decision entropy used for consensus arbitration and the second threshold for spatial correlation), parameter scanning tests of the system are performed on this platform. That is, while keeping other parameters constant, the threshold is adjusted within a reasonable range, and the fault detection accuracy, false alarm rate, and false negative rate of the system for all samples in the test dataset are recorded at different thresholds. Finally, by analyzing the curves of these performance indicators changing with the thresholds, the numerical point that optimizes the system's overall performance indicators (e.g., the weighted score of accuracy and false alarm rate) is selected as the final configuration of the threshold. For the confidence threshold for semantic gap detection, an additional set of known external interference signal (non-fault) samples can be introduced, and the threshold point that maximizes the distinction between known faults and external interference can be selected.

[0041] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details and should not be construed as limiting the invention to these specific implementations. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A power transmission network fault location system based on edge agent collaborative decision-making, characterized in that, include: The token relay communication module is configured to, when a power transmission network fault is detected, relay tokens between multiple edge agents according to the physical propagation direction of the fault energy wavefront, so as to form an ordered token chain containing local measurement information and corresponding timestamps of each agent. The consensus arbitration module is configured to diagnose and arbitrate the health of the preliminary fault location conclusion by jointly evaluating the decision entropy and physical space correlation of the agents that have reached a consensus after obtaining a preliminary fault location conclusion based on the ordered token chain. The unknown event cognition module is configured to suspend the localization task and generate a structured feature sketch report describing the peculiar features of the unknown fault event when it detects an unknown fault event that has a preset semantic gap with the local knowledge base. The token relay communication module is configured such that: the first agent that detects the fault with high confidence generates an initial token, and passes the initial token to the next neighboring agent according to the physical propagation direction; each agent that receives the token continues to pass the token downstream after adding its own local measurement information and timestamp. The system also includes a fault location module, which is configured to use the relative relay order and timestamp information of the agents contained in the ordered token chain to perform calculations to determine the fault segment without the need for a high-precision global time synchronization signal. The calculation of the decision entropy includes: discretizing the target section of the power transmission network into multiple consecutive small segments with unique numbers; collecting the fault location conclusions output by all agents participating in the consensus and mapping each fault location to its corresponding small segment number; counting the number of agent judgments obtained for each small segment to form a probability distribution of the fault location; and calculating the decision entropy value based on the probability distribution using the information entropy formula. The calculation of physical spatial correlation includes: obtaining the preset geographical coordinates of all agents participating in the consensus; calculating the nearest neighbor distance between each agent and the geographically closest agent; calculating the average of all nearest neighbor distances, and using the average as a quantitative indicator of physical spatial correlation.

2. The power transmission network fault location system based on edge agent collaborative decision-making as described in claim 1, characterized in that, The consensus arbitration module is configured to: when the decision entropy is lower than a first preset threshold and the physical space correlation is higher than a second preset threshold, determine the preliminary fault location conclusion as an unhealthy consensus.

3. The power transmission network fault location system based on edge agent collaborative decision-making as described in claim 2, characterized in that, After determining that the preliminary fault location conclusion is an unhealthy consensus, the consensus arbitration module is configured to perform at least one of the following operations: reject the preliminary fault location conclusion, request supplementary data collection for the relevant area, or escalate the decision event for manual review.

4. The power transmission network fault location system based on edge agent collaborative decision-making as described in claim 1, characterized in that, The unknown event cognition module is configured to: match the current fault signal features with all known fault patterns in the local knowledge base, and determine that the preset semantic gap exists when the confidence of all matches is lower than a preset confidence threshold.

5. A power transmission network fault location system based on edge agent collaborative decision-making as described in claim 1, characterized in that, The structured feature sketch report includes at least one of the following: the time-domain information of the singularity of the unknown fault event signal, the time-frequency energy distribution characteristics of the signal, or the differential feature analysis of the signal and the most similar known fault mode.

6. The power transmission network fault location system based on edge agent collaborative decision-making as described in claim 1, characterized in that, The system includes multiple distributed edge agents deployed in the power transmission network, wherein the token relay communication module, the consensus arbitration module, and the unknown event recognition module are all deployed in each of the distributed edge agents.

Images