Power grid fault correlation analysis method, device and system based on event evolution path

Abstract

The application relates to the technical field of power grid fault correlation analysis, in particular to a power grid fault correlation analysis method, device and system based on an event evolution path, which comprises the following steps: acquiring multi-source heterogeneous data related to power grid faults, preprocessing and constructing standardized event sequences; calculating the causal score between standardized events in the standardized event sequences based on multi-dimensional physical constraints, and constructing a causal edge weight matrix; performing dark event inference and completion on the edges with low causal scores in the causal edge weight matrix based on a power grid historical fault case library, to obtain the completed standardized event sequences; and searching and verifying the event evolution path of the power grid fault according to the completed standardized event sequences and the causal edge weight matrix. The application solves the problems of traditional methods lacking physical constraints, which may easily cause causal inversion, and dark event missing, which may cause the causal chain to break, and improves the accuracy and reliability of power grid fault correlation analysis.

Description

Technical Field

[0001] This application relates to the field of correlation analysis technology for power grid faults, specifically to a method, device, and system for power grid fault correlation analysis based on event evolution paths. Background Technology

[0002] With the development of large-scale and intelligent power grids, the massive, multi-source, and multi-modal heterogeneous monitoring data poses a severe challenge to the rapid and accurate response to power grid faults. Due to the complex topology of power grids and the close interrelationships between nodes, a single fault often follows a specific temporal sequence, causal relationship, and topological relationship to propagate step by step, forming a complete evolution path. Therefore, accurately analyzing the causal relationships between events and reconstructing the entire fault evolution path from massive amounts of data is not only a prerequisite for achieving accurate fault location and early warning, but also a core basis for smart grids to formulate prevention and control strategies and reduce economic losses.

[0003] Current research on power grid diagnosis has certain limitations. Traditional methods use the Pearson correlation coefficient to quantify event associations. This method relies solely on statistical correlation and lacks physical mechanism constraints. It is easily judged as a "spurious association" due to similar data distributions and cannot determine the direction of event evolution, leading to the defect of reversed causality. In addition, during the actual propagation of power grid faults, due to limitations in recording accuracy, some key intermediate events (such as manual intervention, protection failure to operate, or maloperation, also known as "dark events") are often not effectively recorded. The absence of these dark events can cause causal chain breaks in the event evolution path. If these breaks are not corrected, it can easily lead to serious misjudgments of the fault evolution trend and path, thereby reducing the accuracy and reliability of power grid fault association analysis. Summary of the Invention

[0004] In a first aspect, embodiments of this application provide a power grid fault correlation analysis method based on event evolution paths, the method comprising the following steps:

[0005] Acquire multi-source heterogeneous data related to power grid faults, perform preprocessing, and construct standardized event sequences;

[0006] Based on multidimensional physical constraints, causal scores between standardized events in a standardized event sequence are calculated, and a causal edge weight matrix is ​​constructed.

[0007] Based on the historical fault case library of the power grid, dark event inference and completion are performed on the edges with low causal scores in the causal edge weight matrix to obtain the standardized event sequence after completion;

[0008] Based on the completed standardized event sequence and causal edge weight matrix, the event evolution path of power grid faults is searched and verified.

[0009] Preferably, the step of acquiring multi-source heterogeneous data related to power grid faults and performing preprocessing to construct a standardized event sequence includes:

[0010] Acquire SCADA alarm logs, relay protection action records, manual operation logs, and electrical data from monitoring nodes of the power grid, and perform time-series alignment of various data from the power grid;

[0011] Based on the time-aligned data, event time, event location, event type, associated device, and feature value are extracted to construct standardized events;

[0012] Arrange all standardized events in chronological order to form a standardized event sequence.

[0013] Preferably, an N×N causal edge weight matrix is ​​constructed, where N represents the number of all standardized events in the standardized event sequence, and the element in the m-th row and n-th column is the causal score between the m-th standardized event and the n-th standardized event.

[0014] A preferred method for constructing causal scores is as follows:

[0015] For any two standardized events in a standardized event sequence, construct a physical constraint causal evaluation function by combining time dimension, spatial dimension, physical mechanism dimension and statistical significance dimension;

[0016] The calculation result of the physical constraint causal evaluation function is used as the causal score.

[0017] A preferred quantification process for indicators in the time dimension, spatial dimension, physical mechanism dimension, and statistical significance dimension is as follows:

[0018] The time dimension score, spatial dimension score, physical mechanism consistency score, and statistical significance score are calculated separately. The time dimension score is calculated based on the time interval between the occurrence times of standardized events and the time decay characteristics of fault propagation. The spatial dimension score is calculated based on the shortest path hop count of the devices associated with the standardized events in the power grid topology. The physical mechanism consistency score is calculated based on the matching judgment of causal logic between standardized events by the power grid fault propagation rule base. The statistical significance score is calculated based on the conditional probability statistics of events occurring in the power grid historical fault case database.

[0019] Preferably, the process of obtaining the completed standardized event sequence is as follows:

[0020] In the causal edge weight matrix, edges with causal scores lower than the preset chain break detection threshold and time intervals between standardized events greater than the preset minimum propagation delay are selected and denoted as low causal score edges.

[0021] Obtain a set of dark events whose co-occurrence probabilities of the first and last standardized events of low causal scoring edges satisfy a preset condition from the historical fault case database of the power grid;

[0022] Calculate the causal probability of each dark event in the dark event set existing between the first and last standardized events. Select the dark event with the highest causal probability that is greater than the preset reception threshold and insert it into the first and last standardized events. Traverse all standardized events in the standardized event sequence to obtain the completed standardized event sequence.

[0023] Preferably, the causal probability of a dark event existing between the first and last normalized events is calculated, including:

[0024] Calculate the historical conditional probability based on the frequency of the first and last standardized event evolution paths in the historical fault case library of the power grid.

[0025] By querying the historical fault case database of the power grid, the average time delay and standard deviation of the time delay for each dark event, as well as the time interval between the first and last standardized events, are obtained. Combined with the Gaussian normal distribution, a Gaussian time constraint factor is constructed.

[0026] The result of multiplying the historical conditional probability, the Gaussian time constraint factor, and the physical mechanism consistency score is used as the causal probability of the dark event existing between the first and last standardized events.

[0027] Preferably, the step of searching and verifying the event evolution path of the power grid fault based on the completed standardized event sequence and causal edge weight matrix includes:

[0028] The causal edge weight matrix is ​​updated based on the completed standardized event sequence. A directed weighted graph is constructed based on the updated causal edge weight matrix. The optimal evolution path of the power grid fault is searched using Dijkstra's shortest path algorithm.

[0029] Electrical connectivity is verified for the optimal evolution path, and physically unreasonable paths are eliminated to obtain the event evolution path of power grid faults.

[0030] Secondly, embodiments of this application provide a power grid fault correlation analysis device based on event evolution paths, wherein the device stores a computer program, and when the computer program is executed by a processor, it implements the power grid fault correlation analysis method based on event evolution paths described above.

[0031] Thirdly, embodiments of this application also provide a power grid fault correlation analysis system based on event evolution paths, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of any of the above-described power grid fault correlation analysis methods based on event evolution paths.

[0032] As can be seen from the above embodiments, the power grid fault correlation analysis method based on event evolution path provided in this application has at least the following beneficial effects:

[0033] This application effectively overcomes the shortcomings of traditional single statistical correlation analysis, which cannot determine causal direction and is prone to false associations, by constructing a multidimensional physical constraint causal evaluation function based on time, space, physical mechanism and statistical significance. The scheme takes hard physical rule constraints as a premise, combines the prior features of historical big data, adaptively quantifies the causal edge weights between events and generates a matrix, and ensures the rigor of causal evolution logic and the reliability of data support from the source, providing a solid and error-free quantitative input foundation for subsequent accurate identification and completion of dark event chain breaks;

[0034] Furthermore, this application establishes a dual triggering mechanism of chain break detection and time gap to accurately locate low-score false chain break edges caused by missing intermediate events. Based on this, a multidimensional product probability model is constructed by integrating historical prior probability, Gaussian time constraints, and physical mechanism consistency. This model adaptively infers and completes the unrecorded "dark events." At the same time, it combines a multi-layer iterative mechanism to solve the problem of continuous breakage of multiple hidden events, effectively repairing the integrity of the fault causal evolution chain and providing a continuous and complete input foundation for the subsequent construction of high-confidence event evolution paths.

[0035] Ultimately, this application transforms the completed multidimensional causal score into path cost, constructs a directed graph, and uses Dijkstra's algorithm to search for the optimal evolutionary path with the minimum cumulative cost. Subsequently, it combines the real topology of the power grid EMS to perform hard electrical connectivity dual verification and false dark event elimination on the path, and calculates the comprehensive confidence based on the final total path cost. Thus, under the dual guarantee of graph theory optimization and physical mechanism verification, the accuracy and reliability of power grid fault correlation analysis are improved. Attached Figure Description

[0036] Figure 1 A flowchart illustrating the steps of a power grid fault correlation analysis method based on event evolution paths provided in one embodiment of this application;

[0037] Figure 2 This is a schematic diagram of a causal scoring extraction process provided in one embodiment of this application. Detailed Implementation

[0038] The following description, in conjunction with the accompanying drawings, details the specific scheme of the power grid fault correlation analysis method, equipment, and system based on event evolution path provided in this application.

[0039] Please see Figure 1 The diagram illustrates a flowchart of a power grid fault correlation analysis method based on event evolution paths provided in an embodiment of this application. The method includes the following steps:

[0040] S1: Acquire multi-source heterogeneous data related to power grid faults, perform preprocessing, and construct standardized event sequences;

[0041] S2: Calculate the causal scores between standardized events in a standardized event sequence based on multidimensional physical constraints, and construct a causal edge weight matrix;

[0042] S3: Based on the historical fault case library of the power grid, perform dark event inference and completion on the edges with low causal scores in the causal edge weight matrix to obtain the standardized event sequence after completion;

[0043] S4: Based on the completed standardized event sequence and causal edge weight matrix, search and verify the event evolution path of power grid faults to perform fault analysis on the power grid.

[0044] In this embodiment, the specific process of constructing the standardized event sequence in step S1 is as follows:

[0045] To support the processing of multi-source heterogeneous big data and subsequent causal analysis in a smart grid environment, this step first acquires multi-source heterogeneous data related to power grid faults, followed by standardized preprocessing. The specific implementation process is as follows:

[0046] S1.1 Multi-source heterogeneous data acquisition stage:

[0047] This embodiment obtains four types of key data sources from the power grid dispatch center and substation sites through different dedicated power interfaces and protocols. The specific acquisition methods and data contents are as follows:

[0048] 1. SCADA Alarm Log:

[0049] Acquisition method: Read alarm information logs through the internal DL / T 860 MMS (Manufacturing Message Specification) interface of the SCADA server of the power dispatch center station.

[0050] Data content: This type of data is presented as structured text, mainly including: device ID (Dev_ID), alarm type (Alarm_Type, enumeration type, such as "overcurrent", "grounding", "open circuit", "frequency limit exceeded", etc.), alarm level (Level, with a value range of {1,2,3,4}, the larger the value, the higher the alarm level), and the timestamp of the alarm event.

[0051] 2. Relay protection operation record:

[0052] Acquisition method: The relay protection action records are obtained through the microcomputer protection device in the substation relay protection panel using the IEC 61850 GOOSE / SV communication protocol in conjunction with COMTRADE (General Format for Transient Data Exchange in Power Systems) files.

[0053] Data content: This type of data is presented as structured text, mainly including: Action type (Action_Type, enumeration type, such as "trip", "reclosing", "alarm", etc.), fault phase (Phase, enumeration type, such as phase A, phase B, phase C, phase ABC), fault location value (Fault_Location) and the timestamp of the fault occurrence.

[0054] 3. Manual Operation Log:

[0055] Acquisition method: Manual operation logs are obtained through the OMS workstation (power outage management system) in the dispatch center using the database structured query interface.

[0056] Data content: This type of data is presented as semi-structured text, mainly including: operation object (Op_Object), operation type (Op_Type, such as "opening", "closing", "mode switching", etc.), operation time (Op_Time), and operation result (Result, enumeration type, such as "success" and "failure").

[0057] 4. Real-time monitoring and topology data:

[0058] Node electrical data: Phasor measurement units (PMUs) are deployed in the PMU cabinets of substations to acquire real-time electrical data such as voltage amplitude and frequency of the corresponding monitoring nodes; for other power nodes such as transformers and circuit breaker mechanism boxes, operating data are acquired through equipment status monitoring devices.

[0059] Limit Exceedance Event Generation: Based on the GB / T 12325 power quality standard, real-time electrical data is judged, limit exceeding events are identified and recorded, and continuous time-series monitoring data is converted into discrete event records.

[0060] Topology data: Based on the EMS (Energy Management System) server of the dispatch center, obtain the topology connection data between various nodes of the power grid.

[0061] S1.2 Data Preprocessing Stage:

[0062] To address the issues of multi-source differences, time-scale discrepancies, and non-standard formats in the aforementioned multi-source heterogeneous big data, the following preprocessing operations are performed:

[0063] 1. Timing alignment processing:

[0064] Using the GPS precision clock of the phasor measurement unit (PMU) as a unified time reference, the PMU, following the IEEE 1588PTP protocol, achieves a time accuracy of ≤1μs, meeting the accuracy requirements for fault analysis. The timestamps of all events acquired, including SCADA alarms, protection actions, and manual operations, are uniformly mapped onto the PMU's timeline, eliminating time discrepancies between multi-source data.

[0065] 2. Text data structuring:

[0066] The BERT-BiLSTM-CRF (BERT-based bidirectional long short-term memory network conditional random field) named entity recognition model is used to perform semantic parsing on unstructured text such as manual operation logs.

[0067] This model automatically extracts key information fields from text and transforms them into a standardized event format. Specifically, the BERT layer is used to extract contextual semantic features, the BiLSTM layer is used to capture long-distance dependencies, and the CRF layer is used to decode the optimal label sequence (this model architecture is a well-known technique, and the specific training process will not be elaborated upon).

[0068] 3. Construct a standardized event sequence:

[0069] Sequence construction logic:

[0070] After time-series alignment and structuring, the multi-source heterogeneous event data is arranged according to the chronological order of event occurrence to construct a standardized event sequence. The sequence is defined as .

[0071] in, This represents the total number of standardized events within the preset acquisition period, i.e., the total number of all standardized events in the standardized event sequence. The standardized events refer to power grid fault-related events, which consist of five elements. In this embodiment, the length of the preset acquisition period is 20ms. In practical applications, as other implementation methods, the implementer can set it according to specific circumstances. (Subscript) This represents the index of the event in the sequence, with a value range of 1. Used to uniquely identify the first element in the sequence. A standardized event.

[0072] Standardized event structure definition:

[0073] The first in the standardized event sequence A standardized event Each contains five core data elements, and its data structure is defined as a quintuple: ;

[0074] The meaning of each character and the corresponding data acquisition process are as follows:

[0075] (Timestamp): Represents the th event in a normalized event sequence. The moment when a standardized event occurs;

[0076] Acquisition process: If the event originates from the SCADA alarm log, extract the alarm occurrence time; if it originates from the relay protection action record, extract the protection device action time; if it originates from the manual operation log, extract the operation time; if it originates from the PMU monitoring data, extract the over-limit judgment time. All time data must be unified to the PMU reference time.

[0077] (Event Position): Represents the position of the event in the normalized event sequence. The location of the power grid topology where a standardized event occurs.

[0078] Acquisition process: based on associated device field The system queries the power grid topology data obtained from the EMS server to locate the substation, line, or node number where the associated device is located, which serves as the spatial coordinates of the event.

[0079] (Event Type): Represents the event type in the normalized event sequence. Category attributes of a standardized event.

[0080] Acquisition process: If it is a SCADA alarm log, directly read the alarm type field (such as "overcurrent" or "grounding"); if it is protection action data, read the action type field (such as "trip" or "reclosing"); if it is an operation log, extract the operation type (such as "opening" or "closing") through BERT-BiLSTM-CRF model parsing; if it is monitoring data, mark the type according to the over-limit situation (such as "voltage over-limit").

[0081] (Associated device): Represents the first device in a normalized event sequence. The specific power equipment number involved in each standardized event.

[0082] Acquisition process: Read the device ID directly from the structured fields of SCADA alarm logs and relay protection records, or extract the operation object name from manual operation logs through the named entity recognition model.

[0083] (Eigenvalue): Represents the th element in the standardized event sequence. A standardized event corresponds to a quantified numerical value or state value.

[0084] Acquisition process: If it is a protection action event, read the fault distance data (unit: kilometers); if it is a SCADA alarm, read the alarm level value (e.g., 1-4); if it is monitoring data, read the voltage amplitude or frequency deviation value at the time of exceeding the limit; if it is an operation event, record the operation result status (e.g., "success" or "failure").

[0085] To aid understanding, an example is provided below:

[0086] Original scene description of the event:

[0087] After the aforementioned relay protection device issued a trip command, circuit breaker No. 2051 in substation A immediately performed a tripping operation to isolate the faulty line. At this time, the SCADA system monitored the change in the position status of the circuit breaker (from "closed" to "open") in real time through the station-side measurement and control device. The system then generated an alarm action event of "circuit breaker change" and sent the structured alarm message to the master station through the DL / T 860 MMS interface.

[0088] The five elements of the standardized events mentioned above are extracted as follows:

[0089] (Extract the alarm upload time recorded in the SCADA server log);

[0090] (Based on device) (Spatial coordinates obtained by querying EMS topology positioning).

[0091] (The alarm type enumeration value read from the MMS interface represents the device status change).

[0092] (Device ID read from the structured fields of the SCADA alarm log);

[0093] (Read the alarm level value corresponding to this circuit breaker alarm. Level 4 indicates the highest severity of the emergency change.)

[0094] In this embodiment, the specific process of constructing the causal edge weight matrix in step S2 is as follows:

[0095] S2.1 Background and Defect Analysis:

[0096] In smart grids, there are certain causal relationships between various events. Traditional technical means often use single-dimensional statistical correlation analysis (such as Pearson correlation coefficient) to determine the causal relationship between events. However, in practice, "correlation" and "causation" are not completely equivalent. Statistical correlation analysis mainly measures the degree of synchronous change between two events, but it cannot determine the direction of event evolution. At the same time, it lacks physical mechanism constraints and is prone to "false correlation" phenomenon where data is judged to be related due to similar data distribution.

[0097] In the actual evolution of power events, fault propagation is a semi-random process constrained by physical mechanisms (possessing a certain degree of randomness but also exhibiting correlation). Therefore, the causal directionality between two events can be determined based on multiple dimensions. This scheme utilizes a multi-dimensional joint judgment based on time, space, physical mechanism, and statistical significance dimensions. Implementers can select any one or more of these dimensions to construct a causal evaluation function constrained by physical constraints.

[0098] S2.2 Construction of Multidimensional Causal Evaluation Function:

[0099] In a preferred embodiment, a causal evaluation function for physical constraints is constructed by combining the above four dimensions. For any two standardized events in a standardized event sequence... and Its causal scoring function is defined as follows: ;

[0100] Character explanation:

[0101] : Represents the causal score of the m-th standardized event on the n-th standardized event;

[0102] : Represents the time dimension score between the m-th standardized event and the n-th standardized event;

[0103] : Represents the spatial dimension score between the m-th standardized event and the n-th standardized event;

[0104] : Represents the physical mechanism consistency score between the m-th standardized event and the n-th standardized event, i.e., the quantitative result of the physical mechanism dimension;

[0105] : Represents the statistical significance score between the m-th standardized event and the n-th standardized event;

[0106] : These represent the preset first weight coefficient, preset second weight coefficient, preset third weight coefficient, and preset fourth weight coefficient, respectively. ;

[0107] : Represents the binary indicator function for physical mechanism constraints, taking a value of 0 or 1 (if and only if the physical mechanism consistency score is 0). ; all other cases are 1).

[0108] Preferably, the schematic diagram of the causal scoring extraction process provided in this embodiment is as follows: Figure 2 As shown.

[0109] It should be noted that in this embodiment, the preset first weight coefficient, preset second weight coefficient, preset third weight coefficient and preset fourth weight coefficient are all set to 0.25. In actual application, as other implementation methods, implementers can also set them according to specific circumstances. This embodiment does not impose any special restrictions.

[0110] S2.3 Detailed Explanation of Scoring Functions for Each Dimension:

[0111] 1. Time dimension scoring :

[0112] The impact of a time-based scoring quantifies the decay of the failure event's influence over time, as expressed below: ;

[0113] Character explanation and principle description:

[0114] exp(): Represents an exponential function with the natural constant as the base;

[0115] : Represents the time interval between the timestamp of the m-th normalized event and the timestamp of the n-th normalized event;

[0116] : Represents the time constant for fault propagation, and its value range is typically 100%. This value is obtained by statistically analyzing the time differences of event pairs with causal chains already marked in the historical fault case database of the power grid. The specific process is as follows:

[0117] (1) Historical causal pair extraction: From the historical fault case database of the power grid, all event pairs with clear causal labels are retrieved, where the causal event is denoted as... The resulting event is recorded as .

[0118] (2) Time interval calculation: For each pair of known causal events retrieved, calculate the difference in their occurrence timestamps to obtain the time interval. Because historical data contains a large number of different types of faults (such as main protection actions and backup protection actions), a result will be obtained consisting of multiple A set of time interval samples.

[0119] Statistical distribution estimation: In power systems, the propagation time of similar faults typically follows an exponential distribution or a truncated normal distribution. In this embodiment, the mean of the aforementioned time interval sample set is used as the fault propagation time constant. Furthermore, to avoid circular dependencies, it is important to emphasize that the sample data relied upon in the above statistical calculations all originate from offline labeled data confirmed manually or by relay protection devices in the historical fault case database of the power grid. This process belongs to the offline parameter calibration stage before model deployment and is completely independent of the subsequent real-time causal scoring. This calculation fundamentally avoids the circular dependency problem of "using real-time calculated results to define real-time parameters".

[0120] 2. Spatial Dimension Scoring :

[0121] Spatial dimension scoring quantifies the electrical coupling strength of physical nodes corresponding to events. The greater the distance, the lower the probability of causal propagation. The expression is as follows: ;

[0122] Character explanation:

[0123] : Indicates the device associated with the m-th normalized event to the n-th normalized event. The shortest path hop count of the associated device in the power grid topology (obtained by the EMS system), when two events occur at the same device node. At this point, the maximum score for the spatial dimension is 1.

[0124] 3. Physical Mechanism Consistency Score :

[0125] The physical mechanism consistency score uses a power grid fault propagation rule base to determine the rationality of causal logic, and the expression is as follows: ;

[0126] Character explanation and principle description:

[0127] : Indicates that the power grid fault propagation rule base supports standardized events. The number of rules governing causality;

[0128] : Indicates the standardized event pairs applicable to the power grid fault propagation rule base. The total number of rules;

[0129] No prior information mechanism: when (i.e., there are no rules in the rule base applicable to standardized event pairs) When (according to the rules), Setting it to the default value of 0.5 indicates uncertainty in the physical mechanism. The overall score is determined by three dimensions: time, space, and statistics.

[0130] It should be noted that the rule base is constructed based on:

[0131] The rules were set by experts in the power grid field based on the following physical mechanisms:

[0132] Protection action logic: According to the "Technical Specification for Relay Protection and Safety Automatic Devices", if the "main protection of the line trips", it will inevitably lead to the "circuit breaker change".

[0133] Physical causal exclusion: According to power topology and electrical common sense, such as "circuit breaker tripping" (cause) cannot cause "line short circuit fault" (effect) because the circuit breaker has cut off the current.

[0134] Sequence logic: Operational events must meet certain constraints in time to avoid reversing cause and effect.

[0135] Example of rule-based determination:

[0136] Assuming standardized events "Line protection tripping", a standardized event This refers to "circuit breaker displacement".

[0137] The power grid fault propagation rule base contains rule R1: IF cause.type=="protection trip" AND result.type=="circuit breaker change" AND result.time-cause.time < 100ms THEN, which supports causality.

[0138] The rule base contains rule R2: IF cause.type=="protection trip" AND result.type=="circuit breaker change" AND result.val=="failure" THEN negate causality.

[0139] If standardized events are If R1 is satisfied but R2 is not satisfied, then , (Assuming that only these two rules for this event pair are configured in the system), at this time If the power grid fault propagation rule base contains no rules configured for this event pair type, the aforementioned no-information prior mechanism is triggered. It is recorded as 0.5.

[0140] 4. Statistical significance score :

[0141] The statistical significance score is based on statistical evidence provided by the historical fault case database of the power grid. Its essence is conditional probability, and the expression is as follows: ;

[0142] Character explanation and principle description:

[0143] : Indicates the first standardized event to occur in the power grid's historical fault case database. And then a standardized event occurred. The number of times it was recorded;

[0144] : Represents an event in the power grid historical fault case database. Total number of occurrences.

[0145] Because the power grid's historical fault case database contains a large amount of data, the events are highly likely to be recorded. However, for faults in newly installed equipment, there may be... In such cases, setting the statistical significance score to zero indicates that historical data cannot provide any significant evidence.

[0146] S2.4 Generation of Causal Edge Weight Matrix:

[0147] This embodiment analyzes the causal relationship of events from four dimensions: the time dimension ensures the causal sequence, the spatial dimension ensures electrical connectivity, the physical dimension ensures the rationality of the mechanism, and the statistical dimension provides big data evidence. By additively combining the evaluation values ​​of each dimension, it ensures that weak evidence in one dimension can be compensated by strong evidence in other dimensions.

[0148] Finally, for the standardized event sequence Normalized event pairs are formed by any pairwise combination of all normalized events in the set. Calculate causal scores Construct a causal edge weight matrix The causal edge weight matrix is ​​an N×N matrix, where the element in the i-th row and j-th column is... This matrix will be used as input data for subsequent steps of dark event inference and completion.

[0149] In this embodiment, the process of completing the standardized event sequence in step S3 is specifically described as follows:

[0150] S3.1 Dark Event Background and Triggering Mechanism:

[0151] The causal edge weight matrix can be obtained through step S2. , the elements It can reflect the degree of causal relationship between two events corresponding to coordinates. This is due to the standardized event sequence length. The values ​​of these values ​​are often large, resulting in a large number of low-scoring causal edges in the causal edge weight matrix. These low-scoring causal edges may be two events that truly have no causal relationship, or they are very likely due to unrecorded intermediate events (i.e., "dark events") between the two events, which leads to a low direct causal score between the two events. Among these, dark events often include protection refusal to act / not acting, additional manual operations, etc. Therefore, it is necessary to analyze the hidden events to complete the causal chain.

[0152] Dark event completion trigger conditions:

[0153] For causal edges in the causal edge weight matrix that have low scores and satisfy the time interval, the following criteria are used for determination:

[0154] 1. Numerical conditions for chain breakage: ;

[0155] Character explanation: This indicates the preset chain break detection threshold, and its value range is typically [value range missing]. In this embodiment, The first quantile of all elements in the causal edge weight matrix is ​​selected to adaptively filter out broken edges with scores at the bottom.

[0156] 2. Time gap condition: ;

[0157] Character explanation: Represents the i-th normalized event The timestamp of the occurrence and j standardized events The time interval between the timestamps that occurred; This represents the preset minimum propagation delay. In this embodiment, the preset minimum propagation delay value is... .

[0158] An edge between standardized events that simultaneously meet the above conditions is denoted as a low causality score edge, and the standardized event is determined accordingly. and There are certain hidden events, which trigger the hidden event inference process. The main triggering logic in this embodiment is the fault propagation statistics accumulated in the power grid historical fault case database, combined with the causal scoring and physical constraints in step S2, to realize the standardization of events. The timestamp of the occurrence and Analysis of the probability of dark events within the time interval between occurrence timestamps, through probabilistic completion of normalized events. and The causal break.

[0159] S3.2 Dark Event Completion Probabilistic Model Construction:

[0160] A pair of standardized events that meet the conditions for triggering a dark event and Taking this as an example, firstly, based on the historical fault case database of the power grid, we retrieve and filter out standardized events. and The top 10 historical standardized events with the highest co-occurrence probabilities are grouped into a candidate dark event set to be filled; for the d-th dark event in this candidate dark event set... It truly exists in standardized events. and events The causal probability evaluation model between them is defined as follows: ;

[0161] Character explanation:

[0162] : Represents the causal probability that the d-th dark event exists between the i-th and j-th normalized events;

[0163] : Represents the prior probability that the d-th dark event exists after the i-th normalized event;

[0164] : Represents the Gaussian time constraint factor.

[0165] : Represents the physical mechanism consistency score between the d-th dark event and the j-th normalized event;

[0166] Model principle explanation:

[0167] Finally, multiplication is used to combine the three factors, mainly because the three represent the three necessary conditions for the existence of dark events: historical prior probability, temporal consistency, and physical rationality. None of them can be missing. If any factor is zero, the causal probability of the dark event is zero. Thus, multiplication naturally satisfies the logical semantic constraint that all necessary conditions are satisfied at the same time.

[0168] Detailed explanation of each indicator in the above causal probability evaluation model:

[0169] 1. Historical conditional probability:

[0170] The historical conditional probability of the d-th dark event under the i-th standardized event. The expression is: ;

[0171] Character explanation and principle description:

[0172] : This represents the frequency of historical cases in the power grid historical fault case database where the d-th dark event occurs between the occurrence of the i-th standardized event and the occurrence of the j-th standardized event;

[0173] : Represents the total number of arbitrary fault record events (without restriction on the type and number of intermediate record events) between the i-th and j-th standardized events in the power grid historical fault case database. (Total number of statistics)

[0174] : Represents the total number of dark event types in the power grid historical fault case database.

[0175] : Represents the Laplace smoothing parameter, which is set to 0.01 in this embodiment to prevent zero probability.

[0176] 2. Gaussian time constraint factor:

[0177] Gaussian time constraint factor between the i-th and j-th standardized events The specific expression is: ;

[0178] Character explanation and principle description:

[0179] and : indicates that the d-th dark event occurs in the i-th normalized event. With the j-th standardized event The average time delay and standard deviation of the time delay; these are obtained by selecting cases from the historical power grid fault case database that meet the criteria. Samples of the pattern, calculate the number of samples in these samples The timestamp of the occurrence and The average interval between events occurring between timestamps is used as And calculate the standard deviation of the time interval as ;

[0180] exp[ ] represents an exponential function with the natural constant as the base.

[0181] In Gaussian time constraints, a Gaussian function characterizes the time delay of "previous event → intermediate event → subsequent event" in power system fault propagation, which is influenced by the superposition of multiple random factors (such as the dispersion of protection device operation time, circuit breaker tripping time deviation, communication delay, etc.). According to the central limit theorem, the time delay caused by the superposition of multiple random factors follows a normal distribution. (When the observation gap...) close to At this time, the value of the Gaussian function is close to 1, indicating a high degree of time coincidence; the further the dark event deviates, the lower the corresponding probability.

[0182] S3.3 Dark Event Filling and Iteration Mechanism:

[0183] For standardized events Calculate the causal probability of each dark event in the corresponding candidate dark event set. And select the dark event with the highest causal probability, denoted as .

[0184] Filling judgment and execution:

[0185] For satisfying Fill in the dark events.

[0186] Character explanation: This represents the preset receiving threshold, which is set to 0.25 in this embodiment.

[0187] The specific filling operation is as follows: [The rest of the text is missing.] Insert into standardized events and standardized events between.

[0188] Multi-level iterative completion:

[0189] Further traversal of the causal edge weight matrix All low-causal scoring edges are filled with dark events based on the power grid historical fault case library to obtain a complete and standardized event sequence. In step S3, this embodiment sets the maximum completion depth to 3 layers, that is, iterates three times in step S3. The purpose of this design is to avoid the existence of standardized events. and standardized events There are multiple hidden events, and a single completion cannot form a complete causal chain (i.e., (in the case of...); finally, the completed standardized event sequence is used as the input for the event evolution path in step S4.

[0190] In this embodiment, the process of obtaining the event evolution path of the power grid fault in step S4 is specifically described as follows:

[0191] After step S3 outputs the complete event sequence, it needs to be further substituted into the physical constraint causal evaluation function of step S2 above to recalculate the completed causal edge weight matrix. Based on the updated causal edge weight matrix, this step aims to search for the optimal fault evolution path and verify its physical feasibility in the power system using a real power grid topology. The specific implementation process is as follows:

[0192] S4.1 Graph Model Construction and Shortest Path Search:

[0193] This embodiment constructs a directed graph model based on the completed standardized event sequence and causal edge weight matrix, and uses Dijkstra's shortest path algorithm to search for the optimal fault evolution path. The specific implementation logic is as follows:

[0194] 1. Directed Graph Model Construction:

[0195] Node set ( ): Use all normalized events in the completed normalized event sequence as nodes of the directed graph.

[0196] Edge set and weight setting ( ): Directed edges are constructed based on the updated causal edge weight matrix. To adapt to the logic of Dijkstra's algorithm of "finding the shortest path (i.e., minimum cost)," the cause-effect score (a probabilistic indicator, the larger the better) needs to be transformed into path cost (a resistance indicator, the smaller the better). Therefore, the weights of the directed edges are set to the complementary values ​​of the corresponding elements of the causal edge weight matrix, i.e. .

[0197] Character explanation: This represents the weight of the directed edge from node i to node j. This represents the value of the element in the i-th row and j-th column of the updated causal edge weight matrix.

[0198] Disconnection setting: If the value of the element in row i and column j is 0, it means that there is no causal relationship between the i-th node and the j-th node. Then the corresponding edge is directly set as "disconnected" (i.e., the directed edge is not generated in the graph).

[0199] 2. Shortest path search:

[0200] Based on the constructed directed graph, the earliest event in the completed standardized event sequence is first extracted and fixed as the starting node for the optimal path search. Then, graph flow calculation is triggered based on the starting point, and Dijkstra's shortest path algorithm is used for global search to obtain the fault evolution path. .

[0201] Explanation of the principle: Causal scoring The higher the value, the stronger the causal relationship between the two events, and the higher the transformed edge weight. The smaller the value, the better. Dijkstra's algorithm calculates the path with the minimum cumulative cost from the failure initiation event to the final termination event by accumulating the weighted costs along the path, and uses this as the optimal evolutionary path. .

[0202] The process of using Dijkstra's algorithm to obtain the shortest path is a well-known technique and will not be elaborated further.

[0203] S4.2 Dual verification and confidence output of electrical connectivity:

[0204] Since the dark event inference in step S3 relies on statistical probability and a physical rule base, there may still be a few "pseudo-dark events" that do not conform to the actual physical connections of the power grid topology. Therefore, the searched paths must be... Perform hard electrical connectivity verification in the spatial dimension.

[0205] 1. Electrical connectivity verification mechanism:

[0206] First, the optimal evolution path of the inferred power grid fault is extracted. The associated device identifiers (i.e., those in the aforementioned normalized event quintuple) corresponding to all node events (especially inferred dark events and their head-to-tail normalized events) are included. elements).

[0207] Subsequently, the macroscopic network topology tree of the power grid management terminal EMS (Energy Management System) is retrieved, and in the real physical space map provided by EMS, it is searched to see if there is a physical electrical conduction path loop between the corresponding related devices.

[0208] Judgment and Removal Logic: If, after retrieval, it is determined that there is no direct line connection (i.e., electrical disconnection) between the devices corresponding to a certain adjacent node event in the real physical power grid, then this path segment is marked as "physically unreasonable". Physically unreasonable evolution nodes and their associated pseudo-dark events are directly removed from the candidate fault propagation graph (directed graph model).

[0209] 2. Final path determination and confidence calculation:

[0210] After completing the above elimination operation, Dijkstra's algorithm is re-executed in the directed graph model, and finally the physical feasibility-verified event evolution path of the power grid fault is output. .

[0211] Finally, the path confidence of the event evolution path of the power grid fault is calculated using the total path cost output by Dijkstra's algorithm. .

[0212] Confidence calculation explanation: If the total cost of the verified path is The path confidence can then be expressed as The higher the path confidence, the closer the comprehensive causal relationship of the power grid fault event evolution path in multiple dimensions such as time, space, physical mechanism and statistical history, providing an intuitive quantitative basis for subsequent power grid operation and maintenance personnel to trace the source and classify faults.

[0213] Based on the same inventive concept as the above method, this application also provides a power grid fault correlation analysis device based on event evolution path. The device stores a computer program, which, when executed by a processor, implements the power grid fault correlation analysis method based on event evolution path described above.

[0214] Based on the same inventive concept as the above methods, this application also provides a power grid fault correlation analysis system based on event evolution paths, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of any one of the above-described power grid fault correlation analysis methods based on event evolution paths.

Claims

1. A power grid fault correlation analysis method based on event evolution paths, characterized in that, The method includes the following steps: Acquire multi-source heterogeneous data related to power grid faults, perform preprocessing, and construct standardized event sequences; Based on multidimensional physical constraints, causal scores between standardized events in a standardized event sequence are calculated, and a causal edge weight matrix is ​​constructed. Based on the historical fault case library of the power grid, dark event inference and completion are performed on the edges with low causal scores in the causal edge weight matrix to obtain the standardized event sequence after completion; Based on the completed standardized event sequence and causal edge weight matrix, the event evolution path of power grid faults is searched and verified.

2. The power grid fault correlation analysis method based on event evolution path as described in claim 1, characterized in that, The process of acquiring multi-source heterogeneous data related to power grid faults and preprocessing it to construct standardized event sequences includes: Acquire SCADA alarm logs, relay protection action records, manual operation logs, and electrical data from monitoring nodes of the power grid, and perform time-series alignment of various data from the power grid; Based on the time-aligned data, event time, event location, event type, associated device, and feature value are extracted to construct standardized events; Arrange all standardized events in chronological order to form a standardized event sequence.

3. The power grid fault correlation analysis method based on event evolution path as described in claim 2, characterized in that, Construct an N×N causal edge weight matrix, where N represents the number of all standardized events in the standardized event sequence, and the element in the m-th row and n-th column is the causal score between the m-th and n-th standardized events.

4. The power grid fault correlation analysis method based on event evolution path as described in claim 3, characterized in that, The process of constructing the causal score is as follows: For any two standardized events in a standardized event sequence, construct a physical constraint causal evaluation function by combining time dimension, spatial dimension, physical mechanism dimension and statistical significance dimension; The calculation result of the physical constraint causal evaluation function is used as the causal score.

5. The power grid fault correlation analysis method based on event evolution path as described in claim 4, characterized in that, The quantification process for indicators in the time dimension, spatial dimension, physical mechanism dimension, and statistical significance dimension is as follows: The time dimension score, spatial dimension score, physical mechanism consistency score, and statistical significance score are calculated separately. The time dimension score is calculated based on the time interval between the occurrence times of standardized events and the time decay characteristics of fault propagation. The spatial dimension score is calculated based on the shortest path hop count of the devices associated with the standardized events in the power grid topology. The physical mechanism consistency score is calculated based on the matching judgment of causal logic between standardized events by the power grid fault propagation rule base. The statistical significance score is calculated based on the conditional probability statistics of events occurring in the power grid historical fault case database.

6. The power grid fault correlation analysis method based on event evolution path as described in claim 1, characterized in that, The process of obtaining the completed standardized event sequence is as follows: In the causal edge weight matrix, edges with causal scores lower than the preset chain break detection threshold and time intervals between standardized events greater than the preset minimum propagation delay are selected and denoted as low causal score edges. Obtain a set of dark events whose co-occurrence probabilities of the first and last standardized events of low causal scoring edges satisfy a preset condition from the historical fault case database of the power grid; Calculate the causal probability of each dark event in the dark event set existing between the first and last standardized events. Select the dark event with the highest causal probability that is greater than the preset reception threshold and insert it into the first and last standardized events. Traverse all standardized events in the standardized event sequence to obtain the completed standardized event sequence.

7. The power grid fault correlation analysis method based on event evolution path as described in claim 5, characterized in that, Calculating the causal probability of a dark event existing between the first and last normalized events includes: Calculate the historical conditional probability based on the frequency of the first and last standardized event evolution paths in the historical fault case library of the power grid. By querying the historical fault case database of the power grid, the average time delay and standard deviation of the time delay for each dark event, as well as the time interval between the first and last standardized events, are obtained. Combined with the Gaussian normal distribution, a Gaussian time constraint factor is constructed. The result of multiplying the historical conditional probability, the Gaussian time constraint factor, and the physical mechanism consistency score is used as the causal probability of the dark event existing between the first and last standardized events.

8. The power grid fault correlation analysis method based on event evolution path as described in claim 1, characterized in that, The steps of searching and verifying the event evolution path of power grid faults based on the completed standardized event sequence and causal edge weight matrix include: The causal edge weight matrix is ​​updated based on the completed standardized event sequence. A directed weighted graph is constructed based on the updated causal edge weight matrix. The optimal evolution path of the power grid fault is searched using Dijkstra's shortest path algorithm. Electrical connectivity is verified for the optimal evolution path, and physically unreasonable paths are eliminated to obtain the event evolution path of power grid faults.

9. A power grid fault correlation analysis device based on event evolution paths, wherein the device stores a computer program, characterized in that, When the computer program is executed by the processor, it implements the power grid fault correlation analysis method based on event evolution path as described in any one of claims 1-8.

10. A power grid fault correlation analysis system based on event evolution paths, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the power grid fault correlation analysis method based on event evolution path as described in any one of claims 1-8.

Images