A visual reasoning approach for progressive fault detection in large power grids

Abstract

The present invention discloses a visual reasoning method for progressively detecting large-scale power grid faults, comprising the following steps: extracting voltage and frequency characteristic data; clustering the voltage data in; displaying the clustering results in a clustering curve graph; selecting a curve graph with obvious changes to obtain a corresponding bus number; drawing a colored graph and a geographic wiring diagram to help users understand the scope of fault impact; extracting voltage data and frequency data; calculating the mean and variance of the voltage data and frequency data and organizing them into four-tuple time series data; defining a multivariate similarity measurement function and performing secondary clustering; expressing the clustering results as a characteristic symbol matrix graph; selecting buses with significant differences from the characteristic matrix graph to further narrow the scope of potential fault sources; obtaining bus voltage and frequency time series data, and highlighting the data to further confirm whether the bus is the real fault source. The present invention can be used in the aspects of fault source detection and impact range analysis in large-scale power systems.

Description

Technical Field

[0001] The present invention relates to the technical field of power system visual analysis, and in particular to a visual reasoning method for progressively detecting large-scale power grid faults. Background Art

[0002] China's power system is undergoing rapid expansion, with tens of thousands of busbars and equipment commonly found throughout the country. Most end users are highly sensitive to power outages and therefore demand reliable, uninterrupted power. One factor hindering this continuous power supply is power system faults. Having a detection system that can quickly report faults and accurately locate them within the power system is crucial.

[0003] Fault detection methods are primarily categorized into three types: signal-based, model-based, and knowledge-based. With the development of system identification technology, the first two types of methods have received considerable attention in recent years. Signal-based methods require high-frequency sampling on live power systems. Some literature uses wavelet transforms, high-frequency components of traveling waves, and moving window averages of current signals to detect faults. For example, patent application publication number CN108732465A proposes a distribution network fault location method based on wavelet transforms and CNNs. This method uses wavelet transform multi-scale analysis to decompose fault current data, sequentially calibrating the modulus maximum moments in a parallel coordinate system. These points are then connected to form a line graph, which is ultimately processed into a grayscale image as input to the CNN. The CNN's powerful feature extraction capabilities are then used to extract topological features hidden in the data, enabling automatic machine recognition of traveling wave heads. Finally, fault location is achieved using a B-type traveling wave ranging method. Model-based technologies detect faults in industrial systems by comparing residuals with predefined thresholds. Residuals can be generated by comparing the calculated output of a mathematical model with the output measured by a sensor. For example, patent application publication number CN113075498A provides a distribution network traveling wave fault location method and system based on residual clustering. This method incorporates data that may be affected by time errors into the residuals of the fault section, then uses the DBSCAN algorithm to cluster and group the residuals of the fault section, identifying bad data affected by time errors. The remaining normal data is used to reconstruct the optimization model and calculate the optimal solution to correct the fault location results.

[0004] In systems with tens of thousands of buses, it's difficult to assess the power system's status intuitively and quickly. Visualizing fault-related quantities such as generator rotor angle, bus voltage, and system frequency can enhance engineers' situational awareness and enable a better understanding and interpretation of system faults. However, current fault detection methods offer only a few basic visualization tools (e.g., scatter plots, line charts, bar charts, etc.), making it difficult to help engineers detect faults from millions of variables. The lack of practical and efficient visualization tools is a significant challenge facing engineers today. Contour lines are a common technique for visualizing bus voltages, but they cannot independently address fault detection. Consequently, most engineers still rely on tabular data displays to analyze large amounts of bus-related information, which limits their operational scope and prevents fault analysis within seconds or minutes. Researchers at Zhejiang University have designed a time-series multivariate visualization method, WaveLines, to analyze power system stability, but it can only be used to analyze data from power systems with thousands of buses. Summary of the Invention

[0005] The present invention provides a visual reasoning method, particularly related to a progressive detection of large-scale power grid faults, which can detect faults in a power system with tens of thousands of buses, gradually narrow the scope of the faulty bus through multiple visual analysis methods, and ultimately locate the source of the fault.

[0006] A visual reasoning method for progressively detecting large-scale power grid faults includes the following steps:

[0007] (1) Reading grid geographic wiring diagram data, transient stability simulation multivariable time series data, and map tile data;

[0008] (2) extracting the voltage and frequency data from the transient stability simulation multivariate time series data in step (1), and organizing the voltage and frequency time series data of each bus into an n-dimensional vector of equal length;

[0009] (3) Clustering all voltage n-dimensional vectors organized in step (2) using the k-means algorithm;

[0010] (4) Divide the clustering results in step (3) into k sub-views and display them as clustering curves;

[0011] (5) Select the curve graph with obvious changes from the view in step (4) to obtain m bus voltage time series data and the corresponding bus numbers, which are the potential fault buses;

[0012] (6) Add a new layer to the map, display the voltage time series data in the form of a colored map and visualize the geographical wiring diagram in a network diagram structure to help users understand the impact range of the fault bus;

[0013] (7) extracting voltage data and frequency data from the transient stability simulation multivariate time series data in step (1) according to the bus number obtained in step (5);

[0014] (8) Taking t as the time window size, calculate the mean and variance of the voltage data and frequency data in step (7) and organize them into four-tuple variable time series data;

[0015] (9) defining a multivariate similarity measurement function, and performing secondary clustering on the multivariate time series data in step (8) according to the function;

[0016] (10) The clustering results are represented as a feature symbol matrix, where the rows of the matrix are arranged according to the similarity measurement value and the columns are arranged according to the time window order;

[0017] (11) Select busbars with significant differences (tone and shape) from the feature matrix diagram in step (10) to further narrow the scope of potential fault sources;

[0018] (12) Obtain the bus voltage and frequency time series data in step (11), highlight the bus location in the geographic wiring diagram, and display its voltage and frequency in a curve graph, and further confirm whether the bus is the real fault source by observing the changes in the curve graph;

[0019] (13) Repeat steps (5)-(12) until the real fault source is found.

[0020] Several optional methods are also provided below, but they are not intended to be additional limitations on the above-mentioned overall solution. They are merely further supplements or optimizations. Under the premise that there are no technical or logical contradictions, each optional method can be combined separately for the above-mentioned overall solution, or multiple optional methods can be combined.

[0021] Preferably, in step (3), the specific method of clustering all n-dimensional vectors organized in step (2) using the k-means algorithm is as follows:

[0022] Define n-dimensional vector as x= <x1,x2,…,x n >, the root mean square distance between any two vectors p and q is defined as:

[0023]

[0024] Where t is the subscript value of the n-dimensional vector. Based on this, the k-means algorithm is applied to cluster all n-dimensional vectors. The distance between the vector and the center point in the clustering process is the root mean square distance. The k value is determined according to the elbow curve.

[0025] Preferably, in step (4), the specific method for visualizing the clustering curve graph in step (3) is as follows:

[0026] In general, the clustering curve graph is divided into k sub-views; each sub-view contains a class of clustering results, where the n-dimensional vector is displayed as a curve graph with a line width of 0.1 and a light gray color, and the cluster center selects k contrasting colors from the colorbrewer color library with a line width of 1.

[0027] Preferably, in step (8), the specific steps of calculating the mean and extreme values of the voltage data and the frequency data are:

[0028] The voltage data can be represented as an n-dimensional vector v, the frequency data can be represented as an n-dimensional vector f, and the time window size is t. The voltage mean can be represented as an n / t-dimensional vector v m , the voltage variance is v r , the frequency mean is f m , the frequency variance is f r .

[0029]

[0030]

[0031] where v m i is the mean voltage of the i-th dimension, v r i is the voltage variance of the i-th dimension, and the frequency mean and frequency extreme value can be calculated similarly.

[0032] Preferably, in step (9), the specific steps of the multivariate similarity measurement function are:

[0033] The four-tuple variables are the mean voltage v m , voltage variance v r , frequency mean f m , frequency variance f r Since the values and units of each variable in the quadruple are different, in order to obtain a consistent distance metric, the present invention designs a distance function that is independent of the variable type.

[0034] (9-1) Define the amplitude distance d mag It is used to characterize the deviation of the quaternion amplitude and is calculated as:

[0035]

[0036] (9-2) Define the cosine distance d cos It is used to characterize the deviation in the direction of the quadruple and can be calculated as:

[0037]

[0038] (9-3) Define the distance metric function as:

[0039]

[0040] (9-4) The distances between different busbars are calculated using the k-means algorithm for clustering, and the k value is determined based on the elbow curve.

[0041] Preferably, in step (10), the clustering results are represented as a feature matrix diagram, and the specific steps are:

[0042] (10-1) Encode the quadruple of each time window into a donut symbol (such as Figure 3 As shown in the figure, the x-axis direction of the center line of the circle (black dotted line) is the frequency mean, the y-axis direction is the voltage mean, the x-axis width of the circle is the frequency variance, and the y-axis width of the circle is the voltage variance.

[0043] (10-2) The circular symbols are arranged in rows according to the order of the time windows

[0044] (10-3) The busbars of the same category are arranged adjacent to each other along the y-axis, the busbar name is marked at the beginning of the row, and each category is assigned a different color to form a characteristic symbol matrix.

[0045] Compared with the prior art, the present invention has the following beneficial effects:

[0046] The visual reasoning method for progressively detecting large-scale power grid faults of the present invention can add step-by-step detection and reasoning processes through visual interaction as system input, which is significantly effective in detecting and tracing faults in 10,000-level busbar power systems and is an effective supplement to current fault detection methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 A schematic diagram of the clustering curve diagram.

[0048] Figure 2 Schematic diagram of the feature symbol matrix diagram.

[0049] Figure 3 A schematic diagram of potential fault sources.

[0050] Figure 4 A schematic diagram showing the busbar location highlighted in the geographic wiring diagram.

[0051] Figure 5 A schematic diagram showing the voltage and frequency curves. DETAILED DESCRIPTION

[0052] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0053] The following describes embodiments of the present invention in detail, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and are not to be construed as limiting the present invention.

[0054] The visual reasoning method for progressively detecting large-scale power grid faults in this embodiment includes a data processing part, a visual encoding part, and a visual reasoning part.

[0055] Data processing part:

[0056] (1) Reading grid geographic wiring diagram data, transient stability simulation multivariable time series data, and map tile data;

[0057] (2) extracting the voltage and frequency data from the transient stability simulation multivariate time series data in step (1), and organizing the voltage and frequency time series data of each bus into an n-dimensional vector of equal length;

[0058] (3) Define all the voltage n-dimensional vectors organized in step (2) as x= <x1,x2,…,x n >, the RMS distance between any two voltage vectors p and q is defined as:

[0059]

[0060] Where t is the subscript value of the n-dimensional vector. Based on this, the k-means algorithm is applied to cluster all n-dimensional vectors. The distance measurement between the vector and the center point in the clustering process is the root mean square distance. The k value is determined according to the elbow curve.

[0061] (4) The clustering results in step (3) are divided into k sub-views and displayed as cluster curve graphs. The specific method of visualizing the cluster curve graph is as follows: the cluster curve graph is divided into k sub-views, each sub-view contains a class of clustering results, wherein the n-dimensional vector is displayed as a curve graph with a line width of 0.1 and a light gray color, and the cluster center selects k contrasting colors from the colorbrewer color library with a line width of 1.

[0062] (5) Select a curve chart with obvious changes (tone and shape) from the view in step (4) to obtain m bus voltage time series data and the corresponding bus numbers;

[0063] (6) With t as the time window size, calculate the mean and variance of the voltage data and frequency data of the m buses obtained in step (5). Assuming that the voltage data is represented by an n-dimensional vector v, the frequency data is represented by an n-dimensional vector f, and the time window size is t, the voltage mean can be represented by an n / t-dimensional vector vm , the voltage variance is v r , the frequency mean is f m , the frequency variance is f r , the four-tuple can be obtained as <v m ,v r ,f m , f r >. The calculation formula is as follows:

[0064]

[0065]

[0066] where v m i is the mean voltage of the i-th dimension, v r i is the voltage variance of the i-th dimension, and the frequency mean and frequency extreme value can be calculated similarly.

[0067] (7) Define a multivariate similarity metric function and perform secondary clustering on the quadruple time series data in step (6) based on the function. Define the amplitude distance d mag It is used to characterize the deviation of the quaternion amplitude, which can be calculated as:

[0068]

[0069] Define the cosine distance d cos It is used to characterize the deviation in the direction of the quadruple and can be calculated as:

[0070]

[0071] The distance metric function is defined as

[0072]

[0073] The distance between different busbars is calculated by the k-means algorithm for clustering, and the k value is determined according to the elbow curve.

[0074] Visual coding part:

[0075] (1) The clustering results of bus voltage time series data are visualized using a clustering curve graph. The specific method is as follows:

[0076] In general, the clustering graph is divided into k sub-views (such as Figure 1Each subview contains a clustering result, where n-dimensional vectors are displayed as a graph with a line width of 0.1 and a light gray color. The cluster centers are represented by k contrasting colors selected from the Colorbrewer color library and have a line width of 1. Cluster graphs are used to display the temporal variation of power system voltage, allowing engineers to quickly and easily identify faulty buses based on the trend differences between the curve clusters.

[0077] (2) Add a new layer to the map to visualize voltage time series data and the geographical wiring diagram of the power system.

[0078] (3) Encode the quadruple data into a circular symbol (such as Figure 2 As shown in the figure, the x-axis direction of the center line of the circle (black dashed line) is the frequency mean, the y-axis direction is the voltage mean, the x-axis width of the circle is the frequency variance, and the y-axis width of the circle is the voltage variance; the circle symbols are arranged in rows according to the sequence of the time windows; buses of the same category are arranged adjacently along the y-axis, with the bus name marked at the beginning of the row, and each category is assigned a different color to form a feature symbol matrix.

[0079] Visual reasoning part:

[0080] (1) Select the inference starting point from the original data: select the geographic wiring diagram data of a specific region, as well as the map tiles and temporary stable simulation multivariate time series data of the region. In this example, data for the entire country are selected.

[0081] (2) Select the curve graph with obvious changes from the k views of the clustering curve graph (see Figure 1 ), obtain several bus voltage time series data and corresponding bus numbers;

[0082] (3) The data after secondary clustering is encoded into a primitive feature matrix, from which several busbars with significant differences (tone and shape) are selected. These busbars are potential fault sources ( Figure 3 );

[0083] (4) Highlight the busbar location in the geographic wiring diagram ( Figure 4 ), and display its voltage and frequency in a graph, such as Figure 5 As shown, by observing the changes in the curve graph, it is further confirmed whether the bus is the real fault source.

[0084] The method of this embodiment uses visualization methods and interactive methods, combined with machine intelligence and user intelligence, to solve the problem of difficult fault tracing in large-scale power systems. For the power system simulation calculation data set case, this method automatically extracts the bus voltage and frequency data characteristics, automatically calculates the difference in characteristic data between the faulty bus and the normal bus, and allows users to gradually narrow down the range of potential fault sources based on their own professional knowledge through a visual interactive interface, and finally determine the fault source. The reasoning steps are simple (most of which are automatically implemented by the system), the reasoning process is intuitive and easy to operate, and can effectively improve the efficiency and accuracy of user fault tracing.

[0085] The above description is only an example of a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A visual reasoning method for progressively detecting large-scale power grid faults, characterized in that: The following steps are involved: (1) Reading grid geographic wiring diagram data, transient stability simulation multivariable time series data, and map tile data; (2) extracting the voltage and frequency data from the transient stability simulation multivariate time series data in step (1), and organizing the voltage and frequency time series data of each bus into an n-dimensional vector of equal length; (3) Clustering all voltage n-dimensional vectors organized in step (2) using the k-means algorithm; (4) Divide the clustering results in step (3) into k sub-views and display them as clustering curves; (5) Selecting a curve graph with obvious changes from the view in step (4) to obtain m bus voltage time series data and corresponding bus numbers; (6) Display voltage time series data in the form of a colored graph and visualize the geographical wiring diagram in a network graph structure; (7) extracting voltage data and frequency data from the transient stability simulation multivariate time series data in step (1) according to the bus number obtained in step (5); (8) Taking t as the time window size, calculate the mean and variance of the voltage data and frequency data in step (7) and organize them into four-tuple variable time series data; (9) Perform secondary clustering on the multivariate time series data in step (8); (10) The clustering results are represented as a feature symbol matrix, where the rows of the matrix are arranged according to the similarity measurement value and the columns are arranged according to the time window order; (11) Selecting buses with significant differences from the characteristic matrix diagram in step (10) to further narrow the scope of potential fault sources; (12) Obtain the bus voltage and frequency time series data in step (11), highlight the bus location in the geographic wiring diagram, and display its voltage and frequency in a curve graph, and further confirm whether the bus is the real fault source by observing the changes in the curve graph; (13) Repeat steps (5) to (12) until the real fault source is found.

2. The visual reasoning method for progressively detecting large-scale power grid faults according to claim 1, characterized in that: In step (3), the specific method of clustering all n-dimensional vectors organized in step (2) using the k-means algorithm is as follows: Define n-dimensional vector as x= <x1,x2,…,x n >, the root mean square distance between any two vectors p and q is defined as: Where t is the vector subscript value; On this basis, the k-means algorithm is applied to cluster all n-dimensional vectors. The distance measurement between the vector and the center point in the clustering process is the root mean square distance; the k value is determined according to the elbow curve.

3. The visual reasoning method for progressively detecting large-scale power grid faults according to claim 1, characterized in that: In step (4), the specific method for visualizing the clustering curve graph in step (3) is as follows: The clustering curve graph is divided into k sub-views; each sub-view contains a class of clustering results, where the n-dimensional vector is displayed as a curve graph with a line width of 0.1 and a light gray color, and the cluster center is selected from k contrasting colors in the Colorbrewer color library with a line width of 1.

4. The visual reasoning method for progressively detecting large-scale power grid faults according to claim 1, characterized in that: In step (8), the specific steps for calculating the mean and extreme values of voltage data and frequency data are as follows: The voltage data can be represented as an n-dimensional vector v, the frequency data as an n-dimensional vector f, and the time window size is t, then the voltage mean is represented as an n / t-dimensional vector v m , the voltage variance is v r , the frequency mean is f m , the frequency variance is f r ; where v m i is the mean voltage of the i-th dimension, v r i is the voltage variance of the i-th dimension, and the frequency mean and frequency extreme values are calculated.

5. The visual reasoning method for progressively detecting large-scale power grid faults according to claim 1, characterized in that: The multivariate time series data in step (9) is specifically a four-tuple variable, which is the voltage mean v m , voltage variance v r , frequency mean f m , frequency variance f r .

6. The visual reasoning method for progressively detecting large-scale power grid faults according to claim 5, characterized in that: Since the values and units of each variable in the quadruple variables are different, in order to obtain a consistent distance measurement, a multivariate similarity measurement function that is independent of the variable type needs to be designed.

7. The visual reasoning method for progressively detecting large-scale power grid faults according to claim 6, characterized in that: The specific steps of the multivariate similarity measurement function are: (9-1) Define the amplitude distance d mag It is used to characterize the deviation of the quaternion amplitude and is calculated as: (9-2) Define the cosine distance d cos It is used to characterize the deviation in the direction of the quadruple and can be calculated as: (9-3) Define the distance metric function as: (9-4) The distances between different busbars are calculated using the k-means algorithm for clustering, and the k value is determined based on the elbow curve.

8. The visual reasoning method for progressively detecting large-scale power grid faults according to claim 1, characterized in that: In step (10), the clustering results are represented as a feature matrix diagram. The specific steps are: (10-1) Encode the quaternion of each time window into a donut symbol. The x-axis direction of the center line of the donut is the frequency mean, the y-axis direction is the voltage mean, the x-axis width of the donut is the frequency variance, and the y-axis width of the donut is the voltage variance. (10-2) The circular icons are arranged in rows according to the order of the time windows; (10-3) The busbars of the same category are arranged adjacent to each other along the y-axis, the busbar name is marked at the beginning of the row, and each category is assigned a different color to form a characteristic symbol matrix.

Images