Power grid real-time event detection and positioning method and system based on synchronous phasor data

Abstract

The application discloses a power grid real-time event detection and positioning method and system based on synchronous phasor data. Firstly, linear regression and Chebyshev detector are used for abnormal detection and cleaning of PMU synchronous phasor data, and a DBSCAN clustering algorithm is used for detecting clustering change points to classify events. Then, PMU data is statistically analyzed from five dimensions, and PMU scores are calculated. Finally, subgraph construction and scanning are performed to obtain the event location. The simulation results prove the effectiveness of the application in event detection, classification and positioning. The results of industrial data prove the necessity of abnormal detection and the effectiveness of the event detection algorithm. The application can provide a very effective PMU-based power grid real-time event detection tool, and can help power system operators to improve decision-making efficiency.

Description

Technical Field

[0001] This invention relates to the field of power grid technology, specifically to a method and system for real-time power grid event detection and location based on synchronous phasor data. Background Technology

[0002] Power systems can become more resilient and less susceptible to large-scale blackouts if operators have a broad, real-time view of the system. Operators need to take control measures based on multiple alerts appearing in the Energy Management System (EMS). It is necessary to develop a tool using PMU (Synchronization Vector Measurement Unit) measurements to help enhance the decision-making capabilities of operators in the control room, complementing the EMS. The goal of this work is to detect, classify, and locate frequently occurring physical power system events. Examples of these events include line faults, load changes, capacitor bank switching, and generator outages. The process begins by detecting and mining current data for anomalous events, then classifying these events as active power events, reactive power events, or fault events. Generator outages fall under the category of active power events, while capacitor bank switching is a reactive power event, and short-circuit events can be classified as faults. In active power events, the active power flow changes. Reactive power events affect the reactive power flowing in the lines and the total voltage. If a fault occurs, the current increases sharply, and the voltage drops continuously. There are significant frequency variations near the fault location, and the rate of frequency variation increases to very high values. The first step in an event detection algorithm is to detect these changes. Changes in operating conditions are an indicator of power system events.

[0003] Existing real-time power grid event detection methods mainly include: research on event detection using moving averages, dimensionality reduction of synchronous phasor data based on linear principal component analysis, geographic visualization of PMU data, wavelet-based event detection, dynamic programming swing gate trend analysis (SPSSDT), and attempts to classify events using energy similarity and temporal location. Most existing work on event detection and classification methods cannot classify events in a comprehensive and automated manner, and therefore cannot provide the precise location of events. Summary of the Invention

[0004] Purpose of the invention: To address the shortcomings of existing technologies, the purpose of this invention is to provide a method and system for real-time power grid event detection and location based on synchronous phasor data. This system can comprehensively and automatically detect and classify events, provide the exact location of events, and assist operators in discovering possible data anomalies in PMU data, thereby enabling real-time and reasonable operational decisions.

[0005] Technical solution: To achieve the above objectives, the present invention adopts the following technical solution:

[0006] A real-time power grid event detection and location method based on synchronized phasor data includes the following steps:

[0007] Step 1: Check for anomalies in the PMU synchronization phasor data using linear regression detectors and Chebyshev detectors. After cleaning the abnormal data, calculate the active and reactive power flows based on the PMU data. Then, use the DBSCAN clustering algorithm to detect cluster change points based on the voltage, current, active power flow, reactive power flow, and frequency of all PMUs.

[0008] Step 2: Classify events based on cluster changes. Event types include active power events, reactive power events, and fault events. Active power events are those where cluster changes are detected in current and active power, but not in voltage and reactive power. Reactive power events are those where cluster changes are detected in voltage and reactive power, but not in current and active power. Fault events are those where cluster changes are detected in voltage, current, active power, reactive power, and frequency.

[0009] Step 3: Perform statistical analysis on PMU data using Shannon entropy (SE), standard deviation (σ), range of variation (Range), mean deviation (MD), and peak coefficient (CF). Calculate the PMU score (NS) = (SE + CF) * (σ + Range + MD). Select one or more PMUs with the highest scores to construct a subgraph. Scan the subgraph based on event category and score to obtain event locations.

[0010] Preferably, step 1 further includes: eliminating false positive data through transient window detection. If the abnormal data detected by the linear regression detector and the Chebyshev detector falls within the transient window, the abnormal label is removed and the data is regarded as normal data.

[0011] Preferably, in step 1, clustering is performed using the given cluster radius and minimum number of points of the DBSCAN algorithm. If a data point changes its cluster affiliation between two consecutive clustering operations, then a cluster change point is considered to have been detected.

[0012] Preferably, in step 3, Shannon entropy Standard deviation Range of variation = |max(X) i )-min(X i |; Mean difference MD = |μCluster1 - μCluster2|; Crest factor X i It is a data stream, where N is the length of the window, and the data window contains data before and after the event; μ is the average value of the data in the data window, and max and min represent the maximum and minimum values ​​of the data in the data window, respectively. μCluster1 and μCluster2 are the average values ​​of the data in the data window before and after the event, respectively.

[0013] In practice, the PMU score can be calculated using one of the following data from the PMU synchronization phasor data: voltage, current, active power flow, reactive power flow, or frequency. Alternatively, the PMU score can be calculated separately using multiple data and then weighted and fused to obtain the final score.

[0014] Preferably, in step 3, when there is only one PMU, a subgraph is constructed from the PMU and its adjacent buses. When the number of PMUs is greater than one, at least the first two PMUs are selected according to NS, and the shortest distance between the selected PMUs is calculated as the radius. A subgraph is constructed from the selected PMUs and the buses within the calculated radius.

[0015] Preferably, in step 3, the subgraph is scanned to obtain the event location. Specifically, all buses in the subgraph are obtained. If the event type is a reactive power event, the event location is on the reactive power bus. If the event type is an active power event, the event location is on the active power bus. If the event type is a fault event, the event location is on the PMU bus with the highest NS score.

[0016] A real-time power grid event detection and location system based on synchronized phasor data includes:

[0017] The event detection module is used to check for anomalies in the PMU synchronization phasor data of linear regression detectors and Chebyshev detectors. After cleaning the abnormal data, it calculates the active and reactive power flows based on the PMU data and uses the DBSCAN clustering algorithm to detect cluster change points based on the voltage, current, active power flow, reactive power flow and frequency of all PMUs.

[0018] The event classification module is used to classify events based on cluster changes. Event types include active power events, reactive power events, and fault events. Active power events are those where cluster changes are detected in current and active power, but not in voltage and reactive power. Reactive power events are those where cluster changes are detected in voltage and reactive power, but not in current and active power. Fault events are those where cluster changes are detected in voltage, current, active power, reactive power, and frequency.

[0019] The event location module is used to perform statistical analysis on PMU data from Shannon entropy (SE), standard deviation (σ), range of variation (Range), mean deviation (MD), and peak coefficient (CF), calculate the PMU score (NS = (SE + CF) * (σ + Range + MD), select one or more PMUs with the highest scores to construct a subgraph, and scan the subgraph based on the event category and score to obtain the event location.

[0020] A computer system includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is loaded onto the processor, it implements the steps of the power grid real-time event detection and location method based on synchronous phasor data.

[0021] A computer-readable storage medium storing a computer program, characterized in that, when executed by a processor, the computer program implements the steps of the power grid real-time event detection and location method based on synchronous phasor data.

[0022] Beneficial Effects: This invention provides a real-time power grid event detection, classification, and localization method based on synchronized phasor data, offering the following beneficial effects: This invention utilizes PMU data and a set of mathematical and statistical techniques to develop novel, computationally efficient algorithms for anomaly and event detection, classification, and localization, including linear regression and Chebyshev basic detection, DBSCAN-based cluster change detection, Shannon entropy, mean difference, crest coefficient, and graph theory. Simulation results demonstrate the effectiveness of this invention's method for event detection, classification, and localization. This invention provides a highly effective PMU-based real-time power grid event detection tool, helping power system operators improve decision-making efficiency. Attached Figure Description

[0023] Figure 1 This is a technical roadmap of the method according to an embodiment of the present invention.

[0024] Figure 2 The test case diagram is generated using RTDS-Case1 in the simulation example of this invention.

[0025] Figure 3 The test case diagram is generated using RTDS-Case3 in the simulation example of this invention. Detailed Implementation

[0026] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] like Figure 1As shown in the figure, this invention discloses a method for real-time power grid event detection and location based on synchronization phasor data. First, it uses a linear regression detector and a Chebyshev detector to check for anomalies in the PMU synchronization phasor data. After cleaning the abnormal data, it uses the DBSCAN clustering algorithm to detect cluster change points for event classification. Then, it performs statistical analysis on the PMU data from five dimensions to calculate the PMU score. Finally, it constructs and scans a subgraph to obtain the event location. The detailed implementation process of this invention is described below.

[0028] ① Synchronous Phasor Data Anomaly Detection: Linear regression and Chebyshev's algorithm are used as the basic detectors based on statistics and clustering. Voltage (V), current (I), active power (P), reactive power (Q), and frequency (Fz) are input into the linear regression and Chebyshev detectors to determine if there are anomalies in the PMU data. Event data points can also be detected as anomalies. To improve the accuracy of anomaly detection, we can use Prony analysis to detect transient windows and separate event points from anomalies to obtain bad data. Once a transient window is detected, all data points considered to be anomalous will be de-marked and treated as normal data points. Finally, depending on the application, data can be marked as bad data or replaced by interpolation.

[0029] (A) Basic Detectors: The basic detectors used are linear regression and Chebyshev detectors; details are as follows:

[0030] (A1) Linear Regression-Based Detector: By minimizing the sum of squared residuals (i.e., the vertical distance between the window data points and the regression line), a regression line model is obtained for the selected data in the window (usually 1 second). The regression line is represented as follows:

[0031] Regession Line = βx + α (1)

[0032] Where β is the slope, α is the y-intercept of the regression line, and x is the point on the regression line closest to the actual data point. Based on this regression line, data points outside the low or high threshold are considered potentially bad data. The high and low thresholds are set according to equations (2) and (3):

[0033] High Threshold=βx+α+k*dev (2)

[0034] Low Threshold=βx+α-k*dev (3)

[0035] Where k is the number of standard deviations, a preset number that determines the high and low thresholds, and dev is the root mean square value of the y-distance from the regression line, as shown in equation (4):

[0036]

[0037] Where N is the number of points in the window, and x(i) and regression(i) are the actual and regression line data points, respectively;

[0038] (A2) Chebyshev-based detector: This detector is typically used when the distribution of the dataset is unknown. It is a two-step process. In the first step, a strict threshold is applied. In the next step, data points exceeding this threshold are omitted. In the second step, the threshold is recalculated, but k is greater than in the first step. Now, it is checked whether the omitted data points are within this relaxed threshold. If they still fail to fall within this wider threshold, they are considered outliers. The Chebyshev equation is shown in (5).

[0039]

[0040] Where X represents the input PMU data, μ represents the average value of the data within the window, σ represents the standard deviation of the data within the window, and k represents the standard deviation of the average value.

[0041] (B) Synchronous Phase Anomaly Detection (SyncAD): The base detector independently evaluates the anomaly detection of the PMU data, then normalizes the scores using the expectation-maximization (EM) algorithm, and inputs them into the transient window estimation process;

[0042] (C) Transient window estimation: Due to the uncertainty of PMU measurement, the steady-state window of voltage amplitude may have small oscillations; these small oscillations are modeled as noise by the window selection filter; the window selection filter is designed by arranging the sampled values ​​of voltage amplitude measurement in the Hankel matrix Y, as shown in (6);

[0043] The method was rigorously tuned on analog and industry data (PMU) to obtain a transient window; a 2.5-second window of voltage measurement data was used (this window size selection provides a trade-off between speed and accuracy), with a PMU reporting rate of 120 frames per second and a total sample size of J=300 for the 2.5-second window;

[0044]

[0045] Eigenvalue decomposition of the sample correlation matrix yields the rank of the Hankel matrix whose elements, such as y(0), represent the first element of the 2.5-second measurement window.

[0046] QSQ′=YY′=R YY (7)

[0047] Where Q is an orthogonal matrix with columns R YYThe eigenvectors of S; S is a diagonal matrix, the superscript ' denotes transpose, and the singular values ​​of the sample correlation matrix are arranged in descending order of order of magnitude, which can be represented as:

[0048]

[0049] Singularity The logarithm of the first singular value divided by the logarithm of the first singular value is expressed as:

[0050]

[0051] From d1 to d p The p singular values ​​correspond to the complex sine waves presented in the signal, and the rest... From d p+1 arrive The singular values ​​correspond to noise, and σ for k = 2, 3, ..., p k The value depends on the amplitude rating and damping of each component; when the singular point is calculated, σ k If the value is less than the set threshold, the data point is determined to be within the transient window.

[0052] The singular values ​​of matrix Y consist of signal singular values ​​and noise singular values. For a perfectly noise-free signal, the noise singular value is zero. Therefore, even a DC signal will have a singular value with a finite real part and a zero imaginary part. Signals with more variations will have many important singular values. Offline analysis is performed by randomly generating matrices from continuous voltage measurements to determine if a matrix is ​​singular. It is impossible to find a matrix whose determinant is zero and which is singular for continuous PMU measurements. However, as an additional step, the algorithm can be updated to check for singularities. If a singularity is found, the matrix can be formed by moving the measured values ​​by 10 measurement points. This does not affect the performance of the algorithm because only the principal modes are of interest, and it is desired to determine whether a given window is a transient window or a quasi-steady-state window. Here, only the principal modes (one or two modes) smaller than the matrix dimension are of interest.

[0053] Due to transients and oscillations, the transient window has more frequency modes; σ k The threshold was adjusted to -4.27 using known events obtained from industry data simulated and transmitted from a Real-Time Digital Simulator (RTDS) to the PMU. Data points within the transient window were marked as normal data, resulting in higher accuracy.

[0054] ② Event Detection, Classification, and Location: Detects changes in operating point of voltage, current, active power flow, reactive power flow, and frequency; once events are classified, a top-level PMU is selected based on statistical parameters and normalized scores; a subgraph is formed using the top PMU, and a graph scanning algorithm is used to detect the location of events. The event detection algorithm architecture is as follows: Figure 1As shown.

[0055] First, an anomaly detection algorithm is used to check whether there are abnormal data in the PMU data. The abnormal data is replaced with the average value of the data points before and after. After obtaining clean data, the active and reactive power flows are calculated using the PMU data given by formulas (10) and (11). Then the parameters are input into the DBSCAN algorithm.

[0056]

[0057]

[0058] Among them, P ij and Q ij It refers to the injection of three-phase active and reactive power (pu) into lines i and j, I ij It refers to the line currents i and j (pu). It is the line voltage (pu), V angle and I angle It refers to the voltage and current angles.

[0059] (A) Event detection using DBSCAN: Relevant data such as voltage (V), current (I), active power flow (P), reactive power flow (Q), and frequency (Fz) from all PMUs are input into the DBSCAN algorithm; the DBSCAN algorithm uses two parameters ∈ (cluster radius) and minimum number of points (MinPts); data points located within the cluster radius become part of cluster 1; once an event occurs, the operating point changes, and when the next steady state is reached, a new cluster with the new operating point is formed; our interest is in detecting the boundary point instances of the cluster, i.e., cluster change points.

[0060] (B) Event Classification: Classify these instances according to the following rules:

[0061] (B1) Active power event: If there is a cluster change in current measurement and calculated active power, but no change in flow in voltage measurement and calculated reactive power cluster;

[0062] (B2) Reactive power event: If there is a cluster of changes in voltage measurements and calculated reactive power measurements, but no cluster of changes in current measurements;

[0063] (B3) Fault event: A fault event is defined as a change observed in the voltage, current, active power, reactive power, and frequency of the cluster.

[0064] Active power, reactive power, and fault events are classified as follows: Figure 1 As shown, other event types can also be extended.

[0065] (C) Event Localization: The location of the PMUs plays an important role in better observation and event localization. If the PMUs are near the event location, they can detect the event, and in some cases, multiple PMUs can detect the event. Fault events have a greater impact, and their characteristics can be detected by multiple PMUs. However, events such as load changes and transformer tag changes are local events, and their characteristics are not sufficient for the PMU network to pick up the event from a distance. Therefore, we calculate statistical parameters and obtain a comprehensive score to determine the PMU with the strongest signal for a particular event.

[0066] (C1) PMU score: Once an event point instance is detected, calculate five statistical parameters that define the event characteristics;

[0067] (1) Shannon entropy: An event alters the quasi-steady-state properties of a power system; the PMU captures these events, which contain information about the magnitude and properties of the events; in information theory, Shannon entropy is defined as a measure of the information that can be captured; a window of 30 samples is formed around the event instances at the 15 points before and after the event instances; Shannon entropy is calculated according to equation (12):

[0068]

[0069] Where X is the data stream, such as voltage, and i is a point in a window of length N;

[0070] (2) Standard deviation: One of the best measures for defining data distribution. In this embodiment, it is defined as the magnitude of disturbance caused by events in the PMU data. The standard deviation is calculated according to formula (13):

[0071]

[0072] Where X is the data stream, i is a point in a window of length N, and μ is the average value of the data window N;

[0073] (3) Range: This is an important measure that includes changes in PMU data during the event. It is calculated using formula (14):

[0074] Range = |max(X) i )-min(X i (14)

[0075] Where the range of data X is from i=1 to i=N, and N is the length of the window;

[0076] (4) Average difference: Calculate the average difference using data points before and after the event instance; once the transient stabilizes and reaches a steady state, the average difference between the two steady states contains information about the operating point offset; calculate using formula (15):

[0077] MD=|μCluster1-μCluster2| (15)

[0078] Cluster1 is the data window before the event, and Cluster2 is the data window after the event. Each window contains 15 samples.

[0079] (5) Crest factor: This is a metric that defines the severity of the peak during an event; PMUs closer to the event will have a higher crest factor than PMUs farther away from the event; the crest factor is calculated using formula (16):

[0080]

[0081] The data length is the same as that used to calculate all other factors except MD; in terms of electrical distance, the PMU closest to the event location will see more significant event characteristics; the response values ​​of the above statistical indicators will be higher; it is difficult to compare the scores of each metric calculated above between PMUs, so standardized scores are useful; normalization helps to identify prominent PMUs for a specific event by comparing single-point scores, and the normalized score is calculated according to formula (17):

[0082] NS=(SE+CF)*(σ+Range+MD) (17)

[0083] When calculating the five statistical characteristic parameters above, you can use one type of data such as voltage and current from the PMU data for measurement, or you can calculate the NS of several types of data such as voltage and current separately, and then perform weighted fusion to obtain the final measurement score.

[0084] (C2) Subgraph Formation and Graph Scan: Some events are detected by multiple PMUs because they have a greater system-level impact; fault events will be detected by multiple PMUs, and transformer tap changes or load changes will be locally sensed by these buses, adjacent buses, or PMUs on which they are placed, which are close to the event bus in terms of electrical distance; we use Dijkstra's algorithm to find the shortest path and store the shortest path for a specific system offline; Algorithm 1 explains the subgraph selection process; it must be noted here that subgraph formation can be an offline process; depending on the PMU of interest, a subgraph can be selected for graph scan analysis and updated only when the network changes.

[0085]

[0086] ③ Test platform and test cases

[0087] (A) Test Platform Architecture: This architecture consists of four layers. The first layer is the physical layer, which is the first phase of building the power grid model. The second layer is the sensor layer, including all measurement devices (PMUs). The third layer, the real-time data archiving layer, collects all data from each substation and is located centrally in the control center. OpenPDC software is used to archive data from laboratory test cases. The final layer is the application layer, which handles PMU data cleaning and the execution of event detection algorithms. It passes the output to an action layer that is not part of the test setup. However, during field implementation, operators can have better situational awareness or take control measures to operate the power system in a resilient and reliable manner.

[0088] (B) Test Cases: Modeling IEEE 14 and IEEE 39 bus systems in RTDS and simulating different events. Five PMUs are placed on different buses. The sampling rate of each PMU is set to 30 samples per second. Three cases are modeled using different events and PMU placements, as described in the following test cases:

[0089] (B1) Case 1: For example Figure 2 As shown, in the IEEE 14 bus system, five PMUs are placed on buses 2, 6, 8, 9, and 10 to monitor the current in 2-Gen, 6-11, 8-Load, 9-7, and 10-9, respectively. A capacitor bank is placed on bus 9, which operates at different times, while a load is placed on bus 8, and a three-phase fault is modeled on bus 10.

[0090] (B2) Case 2: The location and events of the PMU have changed in this case. In the IEEE 14 bus system, the PMU is placed on buses 2, 6, 7, 9, and 13, monitoring the current flowing into 2-Gen, 6-11, 7-4, 9-7, and 13-14, respectively. The capacitor bank is placed on buses 9 and 3, fault modeling is on bus 13, the transformer tap between 6 and 5 is running, and the load is placed on bus 8.

[0091] (B3) Case 3: For example Figure 3 As shown, the IEEE 39 bus system has five PMU subbuses: 4, 12, 17, 22, and 39, which monitor the current in lines 4-3, 12-11, 17-16, 22-21, and 39-9, respectively. Capacitor banks are placed on buses 4 and 15. Adjustable loads are placed on buses 13, 18, and 39. Faults are simulated on buses 8 and 19. During the simulation, generators on buses 38 and 32 stop operating at different points in time.

[0092] ④ Simulation and Results

[0093] (A) Anomaly Detection: Anomalies were manually inserted into two test cases, and an anomaly detection algorithm was applied. The results are expressed as precision and recall. The recall or true positive given by formula (18) is the proportion of inserted erroneous data correctly identified by the algorithm. Precision is the proportion of correctly identified bad data inserted out of the total number of times the algorithm identified data as bad data using formula (19).

[0094]

[0095]

[0096] The first five values ​​were used for the five PMUs in Case 1, and the last five values ​​were used for the five PMUs in Case 2. As can be seen from Table 1, in most cases, the ability to detect inserted bad data through recall is above 98%, while the ability to detect it accurately (i.e., error-free detection) is around 95%.

[0097] Table 1 Anomaly Detection Results

[0098]

[0099] (B) Score Calculation, Subgraph, and Graph Scan: Calculate the normalized score of the PMUs that detected a specific event, select the top two PMUs, and form a subgraph. If only one PMU detected the event, the subgraph consists of the PMU bus and adjacent buses. Subgraphs can also be calculated and stored offline to speed up processing. The formation of subgraphs can be modified based on the power system model.

[0100] Once a subgraph is formed and the type of event is known, a graph scan algorithm is run to query the functionality of each bus. In Case 1, at 109 seconds, the PMUs at buses 9 and 10 detected the event, but bus 9 scored higher, indicating that only bus 9 has the capability to trigger a reactive power event. Therefore, it can be easily concluded that the capacitor bank at bus 9 was manipulated. In Case 3, a generator drop event was detected at 235 seconds and classified as an active power event, but it could not be located. Bus 32 is not adjacent to bus 12, nor is it in the current subgraph formation scheme, i.e., the subgraph formed around bus 12. The solution for this situation is to increase the number of PMUs so that adjacent PMUs cover more buses, or to expand the subgraph to multiple adjacent buses. In this case, if three adjacent buses are selected to form the subgraph, the generator drop event can be located. Based on the same inventive concept, this invention discloses a real-time power grid event detection and location system based on synchronous phasor data, comprising: an event detection module, used to check for anomalies in PMU synchronous phasor data using linear regression detectors and Chebyshev detectors, clean the abnormal data, calculate active and reactive power flows based on PMU data, and detect cluster change points based on voltage, current, active power flow, reactive power flow, and frequency of all PMUs using the DBSCAN clustering algorithm; an event classification module, used to classify events according to cluster changes, including active power events, reactive power events, and fault events; and an event location module, used to perform statistical analysis on PMU data using Shannon entropy (SE), standard deviation (σ), range of variation (Range), mean difference (MD), and crest coefficient (CF), calculate the PMU score (NS) = (SE + CF) * (σ + Range + MD), select one or more PMUs with the highest scores to construct a subgraph, and scan the subgraph based on the event category and score to obtain the event location.

[0101] The specific working processes of each module described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. The division of modules is only a logical functional division, and there may be other division methods in actual implementation. For example, multiple modules may be combined or integrated into another system.

[0102] Based on the same inventive concept, an embodiment of the present invention discloses a computer system, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is loaded onto the processor, it implements the steps of the power grid real-time event detection and location method based on synchronous phasor data.

[0103] Based on the same inventive concept, this invention discloses a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the power grid real-time event detection and location method based on synchronous phasor data.

[0104] Those skilled in the art will understand that the technical solution of this invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer system (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in the embodiments of this invention. The storage medium includes various media capable of storing computer programs, such as a USB flash drive, portable hard drive, read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk.

[0105] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for power grid real-time event detection and localization based on synchronized phasor data, characterized in that, It comprises the following steps: Step 1, using linear regression detector and Chebyshev detector to detect whether there is abnormal PMU synchronous phasor data, cleaning the abnormal data, calculating active and reactive power flow according to PMU data, and detecting clustering change points based on voltage, current, active power flow, reactive power flow and frequency of all PMUs through DBSCAN clustering algorithm; Step 2, classifying events according to clustering changes, and the event types include active power event, reactive power event and fault event; Among them, the active power event is to detect that the current and active power have clustering changes, and the voltage and reactive power have no clustering changes; the reactive power event is to detect that the voltage and reactive power have clustering changes, and the current and active power have no clustering changes; The fault event is to detect that the voltage, current, active power, reactive power and frequency have clustering changes; Step 3: Perform statistical analysis on the PMU data using Shannon entropy (SE), standard deviation (σ), range (Range), mean deviation (MD), and peak coefficient (CF). Calculate the PMU score (NS = (SE + CF) * (σ + Range + MD). Select one or more PMUs with the highest scores to construct a subgraph. Scan the subgraph based on event category and score to obtain event locations. Wherein, Shannon entropy... Standard deviation Range of variation = |max(X) i )-min(X i |; Mean difference MD = |μCluster1 - μCluster2|; Crest factor X i It is a data stream, where N is the length of the window, and the data window contains data before and after the event; μ is the average value of the data in the data window, and max and min represent the maximum and minimum values ​​of the data in the data window, respectively. μCluster1 and μCluster2 are the average values ​​of the data in the data window before and after the event, respectively.

2. The method for power grid real-time event detection and location based on synchronized phasor data according to claim 1, characterized in that, In step 1, false positive data is removed through transient window detection, and if the abnormal data detected by the linear regression detector and the Chebyshev detector falls within the transient window, the abnormal mark is cancelled and it is considered as normal data.

3. The method for power grid real-time event detection and location based on synchronized phasor data according to claim 1, characterized in that, In step 1, the cluster radius and minimum point number of DBSCAN algorithm are given for clustering, and if the class cluster of a data point changes in the previous and subsequent clustering, it is considered that a clustering change point is detected.

4. The method for power grid real-time event detection and location based on synchronized phasor data according to claim 1, characterized in that, One of the voltage, current, active power flow, reactive power flow or frequency in the PMU synchronous phasor data is used to calculate the PMU score, or multiple data are used to calculate the PMU score and then weighted fusion is used as the final score.

5. The method for power grid real-time event detection and location based on synchronized phasor data according to claim 1, characterized in that, In step 3, when there is only one PMU, a subgraph is constructed by the PMU and its adjacent buses, and when the number of PMUs is greater than one, at least the first two PMUs are selected according to NS, and the shortest distance between the selected PMUs is calculated as the radius, and a subgraph is constructed by the selected PMUs and the buses within the calculated radius.

6. The method of power grid real-time event detection and location based on synchronized phasor data according to claim 5, characterized in that, In step 3, the subgraph is scanned to obtain the event position, specifically: all buses in the subgraph are obtained, if the event type is a reactive power event, the event position is in the reactive power bus, if the event type is an active power event, the event position is in the active power bus, and if the event type is a fault event, the event position is in the PMU bus with the highest NS score.

7. A power grid real-time event detection and localization system based on synchrophasor data, characterized in that, It comprises: An event detection module for detecting whether there is abnormal PMU synchronous phasor data by using a linear regression detector and a Chebyshev detector, cleaning the abnormal data, calculating active and reactive power flow according to PMU data, and detecting clustering change points based on voltage, current, active power flow, reactive power flow and frequency of all PMUs through DBSCAN clustering algorithm; An event classification module for classifying events according to clustering changes, and the event types include active power event, reactive power event and fault event; Among them, the active power event is to detect that the current and active power have clustering changes, and the voltage and reactive power have no clustering changes; the reactive power event is to detect that the voltage and reactive power have clustering changes, and the current and active power have no clustering changes; The fault event is to detect that the voltage, current, active power, reactive power and frequency have clustering changes; and an event positioning module, configured to perform statistical analysis on the PMU data from Shannon entropy SE, standard deviation σ, variation range Range, mean difference MD and wave crest factor CF, calculate a PMU score NS=(SE+CF)*(σ+Range+MD), select one or more PMUs with the highest score to construct a subgraph, and scan the subgraph based on event categories and scores to obtain an event location; wherein the Shannon entropy SE=H(X)=-∑p standard deviation σ=σ(X) = sqrt(∑(X variation range Range=|max(X i )-min(X i )|; mean difference MD=|μCluster1-μCluster2|; wave crest factor CF=CF(X) = max(X X i is a data stream, N is the length of a window, the data window contains data before and after an event; μ is the average value of data in the data window, max and min represent the maximum and minimum values of data in the data window, and μCluster1 and μCluster2 are the average values of data before and after the event in the data window, respectively.

8. A computer system comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The computer program, when loaded into the processor, implements the steps of the method for real-time event detection and location in a power grid based on synchronized phasor data according to any one of claims 1-6.

9. A computer-readable storage medium storing a computer program, the computer program comprising instructions that, when executed by a computer, cause the computer to perform the method of any one of claims 1 to 8. The computer program, when executed by the processor, implements the steps of the method for real-time event detection and location in a power grid based on synchronized phasor data according to any one of claims 1-6.

Images