Power grid topological structure extraction method, power grid fault prediction method and computer storage medium

Abstract

The invention relates to the technical field of power grid analysis, and discloses a power grid topological structure extraction method, a power grid fault prediction method and a computer storage medium, and the method comprises the steps: firstly obtaining the data of each node in a power grid, constructing a power incidence matrix through the obtained data, determining the edge weight between the nodes through the power incidence matrix, and obtaining the edge weight of each node; constructing a vertical edge weight matrix, and establishing a diagonal angle matrix by using the edge weight matrix; constructing a weighting matrix based on the edge weight matrix and the diagonal angle matrix; and finally, constructing an objective function of a weighting matrix based on the power injection amount, solving the objective function through a gradient descent method to obtain an optimized weighting matrix, constructing a distance matrix by using the optimized weighting matrix, clustering each node by using the distance matrix, and extracting a power grid topological structure according to a clustering result. According to the extracted power grid topological structure, the accuracy of the connection relation between the nodes is improved, and the precision of the power grid topological structure is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power grid analysis, and in particular to a power grid topology extraction method, a power grid fault prediction method and a computer storage medium. Background Art

[0002] With the rapid development of power systems, China's power grid has achieved widespread coverage, while also becoming increasingly complex. To reduce grid failures and avoid widespread power outages, existing technologies often utilize intelligent devices to monitor grid operation in real time. With the rapid development of modern power grids towards intelligentization, grid topology has become a mainstream research area in the field of grid monitoring. Using grid topology to implement real-time monitoring of grid operation not only allows for timely acquisition of grid operation data and monitoring of grid operation, but also enables the timely detection of grid anomalies and the precise location of anomalies based on the topology.

[0003] In actual operation, the more commonly used power grid topology analysis method is the synchronized phasor measurement unit, which infers the topological relationship of the power grid by directly measuring the voltage and current in the power grid lines. However, this method is too dependent on physical monitoring equipment. In actual power grids, due to the expensive equipment of the connected measurement unit (Phasor Measurement Unit PMU) and the limited deployment density, the traditional method has problems such as insufficient sampling points and low coverage in practical applications. Secondly, in the process of topology identification, the traditional topology method often only constructs the topology structure based on a single indicator, and does not comprehensively consider physical characteristics such as power flow, resulting in insufficient modeling accuracy. At the same time, the update of the power grid topology in the existing technology relies on the global reconstruction method, which has high computational complexity and cannot reflect the dynamic changes of the power grid topology in real time, further affecting the accuracy of the power grid topology structure. Summary of the Invention

[0004] The purpose of the present invention is to provide a power grid topology extraction method, a power grid fault prediction method and a computer storage medium to solve the technical problem of low accuracy of power grid topology determined in the prior art.

[0005] In order to solve the above technical problems, the present invention provides a method for extracting a power grid topology, which comprises the following steps:

[0006] 1) Acquire data from each node in the power grid and use the acquired data to construct a power correlation matrix. The elements in the power correlation matrix represent the similarity of power changes between two nodes. The power correlation matrix is used to determine the resistance and reactance of the line formed between any two nodes. The edge weights between the nodes are determined based on the resistance and reactance. This is used to construct an edge weight matrix. The elements in the edge weight matrix are used to calculate the sum of the edge weights of all lines connected to each node in the power grid to establish an angle matrix.

[0007] 2) Construct a weighting matrix based on the edge weight matrix and the diagonal matrix. The diagonal elements of the weighting matrix are the diagonal elements of the diagonal matrix, and the off-diagonal elements of the weighting matrix are the negatives of the elements at the corresponding positions in the edge weight matrix.

[0008] 3) Based on the power injection amount, an objective function of the weighted matrix is constructed. The objective function is solved by the gradient descent method to obtain the optimized weighted matrix. The optimized weighted matrix is used to construct a distance matrix. The elements in the distance matrix are the inverses of the absolute values of the non-zero elements in the optimized weighted matrix. The distance matrix is used to cluster the nodes, and the grid topology is extracted based on the clustering results.

[0009] Furthermore, density clustering is adopted for clustering. The process of extracting the network topology structure using the density clustering algorithm includes: taking the median of the elements in the distance matrix as the domain radius, taking the nodes within the domain radius of each node as its adjacent nodes, and when the number of adjacent nodes of a node is greater than or equal to the minimum number of samples, classifying the node and the adjacent nodes into a cluster, performing density clustering on each node in the power grid to obtain a cluster set, setting the non-zero elements in the optimized weighted matrix as the adjacency matrix, and generating a power grid topology map based on the cluster set and the adjacency matrix.

[0010] Furthermore, the method also includes updating the obtained power grid topology structure: re-collecting data of each node based on the currently obtained power grid topology, calculating a new weighting matrix using the re-collected data, calculating the test statistic of each element based on the error between the new weighting matrix and each element in the weighting matrix described in step 3), and comparing the test statistic with the test threshold. When the test statistic of an element is greater than the test threshold, updating the connection relationship between the two nodes corresponding to the element, and using the updated connection relationship to update the power grid topology structure.

[0011] Furthermore, the method also includes optimizing the updated power grid topology: constructing an objective function with the goal of minimizing the difference between the measured power and the predicted power of the updated power grid topology, solving the objective function using a gradient descent iterative optimization algorithm to obtain updated resistance and reactance, obtaining an updated edge weight matrix based on the updated resistance and reactance, and then obtaining the corresponding updated diagonal angle matrix, weight matrix and optimized weight matrix, constructing an updated distance matrix using the updated optimized weight matrix, and clustering each node using the updated distance matrix, and obtaining the optimized power grid topology according to the clustering results.

[0012] Furthermore, the power correlation matrix is composed of the similarity values of the power grid nodes. The similarity value is the ratio of the sum of the active power correlation value and the reactive power value of the two nodes to the time series. The active power correlation value is the sum of the squares of the active power differences at all time points in the time series, and the reactive power value is the square of the reactive power difference between the two nodes at the current time point.

[0013] Furthermore, the elements in the edge weight matrix are the calculated impedance data values of the line between two nodes of the power grid. When there is a line between the two nodes, the calculated impedance data value is the inverse of the sum of the resistance and reactance of the line between the two nodes. When there is no line between the two nodes, the calculated impedance data value is 0.

[0014] Furthermore, the objective function of constructing the weighted matrix based on the power injection amount includes: constructing a fitting model of the linear power flow equation based on the power injection amount, and constructing the objective function by maximizing the likelihood function of the data using the MLE through the fitting model. The objective function is expressed as:

[0015]

[0016] Among them, σ 2 is the variance of the measurement noise, T1 is the total number of time points t, P i obs is the active power of node i, is the reactive power of node i.

[0017] A method for predicting power grid faults, the method comprising: extracting a power grid topology, collecting data on each power grid node according to the extracted power grid topology, determining that a potential power grid fault exists when the collected data on a power grid node is greater than a corresponding set threshold, and inputting the data on each power grid node and power grid topology information into a trained machine learning algorithm to predict the fault type; the power grid topology is obtained using the power grid topology extraction method.

[0018] Furthermore, the set threshold is set based on the mean and standard deviation of historical power grid data.

[0019] A computer storage medium stores a computer program for implementing the above-mentioned method for extracting a power grid topology.

[0020] The beneficial effects of the present invention are as follows: as an improved invention, the present invention utilizes the power changes between the nodes in the power grid to construct a power correlation matrix, determines the resistance and reactance between the two nodes through the power correlation matrix, and determines the edge weights between the nodes based on the resistance and reactance, establishes an edge weight matrix, utilizes each element in the edge weight matrix to establish an angle matrix, and utilizes the edge weight matrix and the angle matrix to construct a weighted matrix, comprehensively considering the total connection weight of the node and the connection relationship between each node, to form a matrix that can characterize the overall connectivity of the power grid. At the same time, the present invention utilizes a clustering algorithm based on the weighted matrix to extract the power grid topology, thereby improving the accuracy of the connection relationship between the nodes, and thus improving the accuracy of the power grid topology. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 is a schematic diagram of the process of obtaining the power grid topology of the present invention;

[0022] Figure 2 Schematic diagram of updating the power grid topology according to the present invention. DETAILED DESCRIPTION

[0023] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings.

[0024] The present invention extracts the power grid topology structure using a clustering algorithm based on a weighted matrix, thereby improving the accuracy of the connection relationship between nodes and further improving the accuracy of the power grid topology structure.

[0025] Grid topology structure extraction method embodiment

[0026] like Figure 1 As shown, the specific steps of the power grid topology extraction method proposed in the present invention include:

[0027] 1) Collect power grid data and construct a weighted matrix based on the power grid data.

[0028] Specifically, the collected power grid data includes voltage, current, power, and the corresponding timestamps of the above power grid data. The power grid data is all collected by smart meters. To ensure the accuracy of the data, the collected power grid data needs to be standardized through the edge computing nodes of the smart meters. The specific method of standardization is to first use the voltage, current, power, and the timestamp corresponding to each data to construct a time series data matrix of size n×T. Here, n is the number of nodes and T is the length of the time series. Subsequently, principal component analysis (PCA) is used to extract features from the time series data matrix, retaining the extracted main components and discarding the secondary components to remove noise in the time series data matrix.

[0029] By standardizing the power grid data, the burden of data transmission to the central server can be reduced, thereby reducing communication delays and data transmission costs. The time series data matrix can record the changes in the voltage, current, and power of each node over time, which can effectively capture the dynamic characteristics of power grid operation and the mutual influence between nodes. The original data can be decomposed into principal components and secondary components through PCA. By retaining the principal component (the energy concentration part), the random noise in the measurement data can be effectively removed, and the accuracy of the data can be improved.

[0030] Then, based on the denoised time series data, a power correlation matrix between nodes is constructed. The power correlation matrix is composed of the similarity values of pairs of grid nodes. The similarity value is the ratio of the sum of the active power correlation value and the reactive power value of two nodes to the time series. The active power correlation value is the sum of the squared active power difference at all time points in the time series, and the reactive power value is the squared reactive power difference between the two nodes at the current time point. The active power correlation value and reactive power value are both collected by smart meters in the grid.

[0031] Specifically, the similarity value formula is expressed as:

[0032]

[0033] Among them, a ij represents the similarity of power changes between nodes i and j at different times, p i (t) and q i (t) represent the active power and reactive power of node i at time t, respectively, p j (t) and q j (t) represent the active power and reactive power of node j at time t, respectively, and T is the length of the time series.

[0034] The similarity values are used as elements to construct a power correlation matrix, and the elements of the power correlation matrix are normalized.

[0035] The elements of the power correlation matrix directly reflect the correlation of power changes between nodes by calculating the similarity of the active and reactive power changes of two nodes at different times. The higher the similarity between node pairs, the closer their power behavior is, which helps to accurately identify strong connections in the power grid topology. Furthermore, the calculation of power change similarity based on time series data can explore the potential patterns of dynamic behavior between nodes, so that the analysis is not limited to static connections but can reflect the dynamic power relationships between nodes. Through normalization, the similarity values in the power correlation matrix have consistent statistical significance across different times and node pairs, providing greater robustness for subsequent analysis and model training.

[0036] Based on the power correlation matrix, the resistance and reactance data of the lines formed between the node pairs are further counted, and the weight matrix W is constructed according to the impedance data. The elements of the matrix are expressed as:

[0037]

[0038] Among them, w ij is the edge weight between node i and node j, r ij and x ij are the resistance and reactance of the lines at node i and node j, respectively.

[0039] The angle matrix is constructed based on the edge weight calculation. The elements of the angle matrix are expressed as:

[0040]

[0041] Among them, d ii is the total connection weight of node i, and n is the total number of nodes.

[0042] The elements of the weight matrix represent the resistance and reactance of the lines between each node pair, directly reflecting the strength of the physical connection between nodes. The lower the impedance, the higher the weight, indicating a greater contribution to power flow. The angle matrix accurately captures the electrical characteristics of lines in the power grid and reflects the strength of the physical connection between a node and other nodes.

[0043] Then, a weighted matrix is constructed based on the weight matrix and the diagonal matrix. The weighted matrix is a weighted Laplacian matrix. ii is the diagonal element of the diagonal matrix, expressed as:

[0044]

[0045] Among them, L ii represents the element of the weighted Laplacian matrix in row i and column i, D ii represents the element in the i-th row and i-th column of the diagonal matrix, Indicates calculating the sum of weights of all nodes j connected to node i.

[0046] The off-diagonal elements of the weighted Laplacian matrix L ij is the negative of the element in row i and column j in the edge weight matrix, expressed as:

[0047] L ij =-w ij , (i≠j);

[0048] Among them L ij Represents the element in the i-th row and j-th column of the matrix L.

[0049] The sum of each row (or column) of the weighted Laplacian matrix is zero. The weighted Laplacian matrix forms a matrix that can characterize the overall connectivity of the power grid by comprehensively considering the total connection weight of the node (that is, the sum of the connection weights of all adjacent nodes) and the connection relationship between each node.

[0050] 2) Based on the power injection amount, the objective function of the weighted Laplacian matrix is constructed, and the grid topology is extracted using a clustering algorithm.

[0051] Specifically, firstly, a fitting model based on the linear power flow equation is constructed based on the power injection amount. The fitting model is expressed as:

[0052]

[0053] Among them, P i is the active power injection amount of node i, Q i is the reactive power injection amount of node i, V i and V j are the voltage amplitudes at node i and node j respectively.

[0054] The linear power flow equation uses the physical properties of voltage amplitude, resistance and reactance between nodes to accurately describe the distribution of active power and reactive power in the power grid and truly reflect the operation mechanism of the power grid.

[0055] Based on the fitting model, the objective function is constructed by using MLE to maximize the likelihood function of the data. The objective function is expressed as:

[0056]

[0057] Among them, σ 2 is the variance of the measurement noise, P i obs is the active power of node i, is the reactive power of node i. By maximizing the likelihood function of the observed data through MLE, the prediction accuracy of the fitting model is significantly improved.

[0058] The objective function is updated and optimized by the gradient descent method. The update formula is expressed as:

[0059]

[0060] Where η represents the learning rate and k represents the number of iterations. The number of iterations refers to the number of times the gradient descent optimization is performed. The learning rate can be set based on historical experience. During the continuous iteration process, when the objective function no longer changes significantly, the iteration is stopped and the optimized matrix L is output. Extract the non-zero elements L in the optimized matrix L ij Form the adjacency matrix A, and As the distance between node i and node j. The distance between each pair of nodes is used to construct the distance matrix D, which is a global distance matrix shared by the entire network and has a size of n×n.

[0061] Then, based on the distance matrix D, the density clustering algorithm is used to extract the network topology. First, the domain radius and the minimum number of samples are set. The minimum number of samples is the minimum number of points in the neighborhood. The minimum number of samples can be set based on historical data. In the power grid, at least two nodes are generally set to form the minimum topological structure unit. In this embodiment, the median in the distance matrix D is set as the domain radius of the node. Then, a node that has not yet been assigned to any cluster is selected, and the distance D between the node and all domain nodes is calculated. ij , filter all distances D between the node and the node ij Nodes within the domain radius are considered neighboring nodes. When the number of neighboring nodes is greater than or equal to the minimum sample number, the node and its neighboring nodes are grouped together in a cluster. By comparing the number of neighboring nodes with the minimum sample number, isolated nodes can be screened out and not assigned to any cluster. In the power grid topology, isolated nodes are used as possible fault points and load points to promptly alert maintenance personnel to perform maintenance inspections and prevent signs of possible interruptions or failures of isolated nodes. By detecting isolated nodes, it is possible to promptly determine whether the power grid has experienced physical disconnection, equipment failure, or other types of system anomalies. Density clustering is performed on each node in the power grid to obtain a cluster set. The non-zero elements in the optimized matrix are set as the adjacency matrix A. A power grid topology map is generated based on the cluster set and the adjacency matrix A, where the cluster set represents all node groups with strong connection relationships in the power grid. These clusters represent the connected areas of the power grid, where the nodes are tightly connected through power grid lines. A node set is formed based on the nodes in all clusters. The adjacency matrix A only retains the valid connections within the cluster, and eliminates weak connections across clusters to determine the edge set, forming a power grid topology map, where each cluster corresponds to a subgraph, and the nodes and connection relationships of all subgraphs are determined by the adjacency matrix.

[0062] Density clustering ensures that all closely related nodes are grouped together, ensuring that critical nodes in the grid are not missed, and allowing the cluster set of the network topology to be properly defined. Since density clustering itself does not change the connections between nodes, and the adjacency matrix A records the direct connection information of all nodes, only valid connections within the same cluster need to be retained, effectively eliminating redundant connections and ensuring the accuracy of locally connected areas. At the same time, extracting the grid topology through the adjacency matrix ensures that the topology retains the grid's original connection structure and does not lose critical connectivity information. This not only improves the accuracy of topological analysis but also dynamically adapts to changes in the grid structure, thereby enhancing the monitoring, scheduling, and optimization capabilities of smart grids.

[0063] like Figure 2 As shown in FIG, in order to ensure the accuracy of the grid topology, it is necessary to regularly collect data from the grid to update the grid topology. The regularly collected grid data includes the voltage, active power, and reactive power of each node. The new weighted Laplacian matrix is calculated according to step 1) using the circuit data collected again. The test statistic of each element is calculated based on the error between the new weighted Laplacian matrix and the weighted Laplacian matrix of step 3). The formula for calculating the test statistic of each element is:

[0064]

[0065] Among them, h L is the test statistic, ΔL ij is the change in the edge weights of nodes i and j between the current topology matrix and the new topology matrix, and σ(ΔL) is ΔL ij The standard deviation of .

[0066] The test statistic h L Compared with the test threshold, when there is a test statistic h of the element L When the value is greater than the test threshold, the node pair corresponding to the element is marked. After the test statistics of all elements are compared, the connection relationship between the marked node pairs is updated, and the updated weighted Laplacian matrix is output. Subsequently, a new power correlation matrix is calculated based on the updated connection relationship. A new edge weight matrix is constructed using the new power correlation matrix. A new diagonal matrix is constructed using each element in the new edge weight matrix. A new weighted Laplacian matrix is constructed using the new edge weight matrix and the new diagonal matrix. The objective function of the new weighted Laplacian matrix is constructed based on the power injection amount. The objective function is optimized using the gradient descent method to obtain the optimized new weighted Laplacian matrix. A new distance matrix is constructed using the optimized new weighted Laplacian matrix, and the nodes are clustered using the new distance matrix. The new power grid topology is extracted based on the clustering results.

[0067] By recollecting data from the power grid and calculating a new weighted Laplacian matrix, the present invention enables real-time reflection of changes in node connectivity within the network topology. Furthermore, by calculating the test statistic for each element based on the error between the new weighted Laplacian matrix and the original, the new element can distinguish between normal fluctuations and abnormal changes, enhancing the robustness of the model.

[0068] After the grid topology is updated, an objective function is constructed to minimize the difference between the measured power and the power predicted by the grid topology, and the updated grid topology is optimized. The power prediction values of all nodes are calculated one by one by optimizing the objective. The formula for the power prediction value is:

[0069]

[0070] Among them, P i m and Represent the active power value and reactive power value of node i respectively, N i represents all neighbor nodes of node i, θ i and θ j are the voltage phase angles at node i and node j respectively.

[0071] Based on the optimization objective, the gradient descent iterative optimization is performed. When the norm of the gradient no longer changes significantly during the continuous iteration process, the iteration is stopped to obtain the corrected admittance parameter r ij 、x ij Based on the updated resistance and reactance, an updated edge weight matrix is obtained, and an updated diagonal matrix is constructed using each element in the updated edge weight matrix. An updated weight matrix is constructed using the updated edge weight matrix and the updated diagonal matrix. An objective function of the updated weight matrix is constructed based on the power injection amount. The objective function is optimized by the gradient descent method to obtain an optimized updated weight matrix. An updated distance matrix is constructed using the optimized updated weight matrix, and the updated distance matrix is used to cluster the nodes. A further updated power grid topology is extracted based on the clustering results.

[0072] This method accurately reflects the actual electrical characteristics of the line through the corrected admittance parameters (resistance and reactance), preventing the impact of parameter errors on power flow calculations and topology analysis. The updated network topology and admittance parameters can be used in multiple scenarios, including power flow analysis, load scheduling optimization, new energy access planning, fault diagnosis, and grid simulation, providing comprehensive support for intelligent grid management.

[0073] Power grid fault prediction method embodiment

[0074] The method for predicting power grid faults proposed by the present invention is as follows:

[0075] Data from each grid node is collected based on the extracted grid topology. When the actual value of the collected data from each grid node exceeds a set threshold, a potential grid fault is determined. The data from each grid node and the grid topology information are then input into a trained support vector machine algorithm to predict the type of fault. The data from each grid node can be collected using smart meters.

[0076] By mapping real-time monitoring data into the grid topology model, the specific area (such as a node or line) where anomalies occur can be quickly located, improving the efficiency of anomaly detection. By comparing real-time collected data with the predicted values calculated in the topology model, the deviation between the two can be clearly quantified, thereby more accurately identifying possible anomalies.

[0077] To ensure maintenance personnel receive timely fault warnings, when a machine learning algorithm predicts a grid fault, it issues an alert through audio and visual alarms. Furthermore, while maintenance personnel are performing grid maintenance, an industrial tablet terminal records their maintenance operation data and the real-time grid operation data at that moment, aligning the timestamps of both data. This data is then wirelessly transmitted to the cloud and remote terminals for backup. Reporting tools generate a table of maintenance personnel's maintenance operation data, making maintenance record management more standardized and systematic, and facilitating the archiving and statistics of maintenance work by the grid operation and maintenance department.

[0078] The specific implementation process has been described in detail in the embodiment of the power grid topology structure extraction method, and will not be repeated here.

[0079] Computer Storage Medium Embodiments

[0080] The computer storage medium proposed in the present invention stores a computer program to implement the power grid topology extraction method as described above.

[0081] The computer program includes a data collection and processing module, a storage medium, a calculation module, and a fault prediction module. The data collection and processing module is primarily used to obtain line information from each node of the smart meter and perform standardized processing on the information. The storage medium is primarily used to store the data collected and processed by the data collection and processing module. The calculation module is mainly used to: 1. Construct a power correlation matrix based on line information, and further construct an edge weight matrix, an angle matrix, and a weight matrix; construct an objective function of the weight matrix based on the power injection amount; solve the objective function using the gradient descent method to obtain an optimized weight matrix; use the optimized weight matrix to construct a distance matrix; cluster each node using the distance matrix, and extract the power grid topology based on the clustering results; 2. Calculate a new weight matrix using re-collected data; calculate the test statistic of each element based on the error between the new weight matrix and each element in the weight matrix described in step 3); when the test statistic of an element is greater than the test threshold, update the connection relationship between the two nodes corresponding to the element, and use the updated connection relationship to update the power grid topology; 3. Construct an objective function with the goal of minimizing the difference between the measured power and the power predicted by the power grid topology, and optimize the updated power grid topology. The fault prediction module includes a machine learning algorithm and a fault judgment algorithm. The machine learning algorithm is mainly used to predict the fault type based on the data of each node in the power grid and the power grid topology information; the fault judgment algorithm is mainly used to determine whether the data of each node in the power grid is greater than a set threshold.

[0082] The storage medium proposed in this embodiment stores a computer program, which, when executed by a processor, implements the method for extracting the power grid topology based on a smart meter as proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0083] The specific implementation process has been described in detail in the embodiment of the power grid topology structure extraction method, and will not be repeated here.

Claims

1. A method for extracting a power grid topology, characterized in that the steps include: 1) Acquire data from each node in the power grid and use the acquired data to construct a power correlation matrix. The elements in the power correlation matrix represent the similarity of power changes between two nodes. The power correlation matrix is used to determine the resistance and reactance of the line formed between any two nodes. The edge weights between the nodes are determined based on the resistance and reactance. This is used to construct an edge weight matrix. The elements in the edge weight matrix are used to calculate the sum of the edge weights of all lines connected to each node in the power grid to establish an angle matrix. 2) Construct a weighting matrix based on the edge weight matrix and the diagonal matrix. The diagonal elements of the weighting matrix are the diagonal elements of the diagonal matrix, and the off-diagonal elements of the weighting matrix are the negatives of the elements at the corresponding positions in the edge weight matrix. 3) Based on the power injection amount, an objective function of the weighted matrix is constructed. The objective function is solved by the gradient descent method to obtain the optimized weighted matrix. The optimized weighted matrix is used to construct a distance matrix. The elements in the distance matrix are the inverses of the absolute values of the non-zero elements in the optimized weighted matrix. The distance matrix is used to cluster the nodes, and the grid topology is extracted based on the clustering results.

2. The method for extracting the power grid topology according to claim 1, characterized in that: Clustering adopts density clustering. The process of extracting network topology structure using density clustering algorithm includes: taking the median of the elements in the distance matrix as the domain radius, taking the nodes within the domain radius of each node as its adjacent nodes, when the number of adjacent nodes of a node is greater than or equal to the minimum sample number, the node and the adjacent nodes are grouped into a cluster, and density clustering is performed on each node in the power grid to obtain a cluster set. The non-zero elements in the optimized weighted matrix are set as the adjacency matrix, and the power grid topology map is generated according to the cluster set and the adjacency matrix.

3. The method for extracting the power grid topology according to claim 1, wherein: The method also includes updating the obtained power grid topology structure: re-collecting data of each node based on the currently obtained power grid topology, calculating a new weighting matrix using the re-collected data, calculating a test statistic of each element based on the error between the new weighting matrix and each element in the weighting matrix described in step 3), and comparing the test statistic with a test threshold. When the test statistic of an element is greater than the test threshold, updating the connection relationship between the two nodes corresponding to the element, and using the updated connection relationship to update the power grid topology structure.

4. The method for extracting the power grid topology according to claim 3, characterized in that: The method also includes optimizing the updated power grid topology structure: constructing an objective function with the goal of minimizing the difference between the measured power and the predicted power of the updated power grid topology structure, solving the objective function using a gradient descent iterative optimization algorithm to obtain updated resistance and reactance, obtaining an updated edge weight matrix based on the updated resistance and reactance, and then obtaining the corresponding updated angle matrix, weight matrix and optimized weight matrix, constructing an updated distance matrix using the updated optimized weight matrix, and clustering each node using the updated distance matrix, and obtaining the optimized power grid topology structure based on the clustering results.

5. The method for extracting the power grid topology according to claim 1, wherein: The power correlation matrix is composed of the similarity values of the power grid nodes. The similarity value is the ratio of the sum of the active power correlation value and the reactive power value of the two nodes to the time series. The active power correlation value is the sum of the squares of the active power differences at all time points in the time series, and the reactive power value is the square of the reactive power difference between the two nodes at the current time.

6. The method for extracting the power grid topology according to claim 1, characterized in that: The elements in the edge weight matrix are the calculated impedance data values of the line between two nodes in the power grid. When there is a line between the two nodes, the calculated impedance data value is the inverse of the sum of the resistance and reactance of the line between the two nodes. When there is no line between the two nodes, the calculated impedance data value is 0.

7. The method for extracting the power grid topology according to claim 1, characterized in that: The objective function of constructing the weighted matrix based on the power injection amount includes: constructing a fitting model of the linear power flow equation based on the power injection amount, and constructing the objective function by maximizing the likelihood function of the data using the MLE through the fitting model. The objective function is expressed as: Among them, σ 2 is the variance of the measurement noise, T1 is the total number of time points t, P i obs is the active power of node i, is the reactive power of node i.

8. A method for predicting power grid faults, characterized in that: The method includes: extracting a power grid topology, collecting data of each node in the power grid according to the extracted power grid topology, determining that a potential fault exists in the power grid when the collected data of a node in the power grid is greater than a corresponding set threshold, and inputting the data of each node in the power grid and the power grid topology information into a trained machine learning algorithm to predict the type of fault; the power grid topology is obtained using the power grid topology extraction method described in any one of claims 1 to 7.

9. The power grid fault prediction method according to claim 8, characterized in that: The set threshold is set based on the mean and standard deviation of historical power grid data.

10. A computer storage medium, characterized in that The computer storage medium stores a computer program for implementing the power grid topology extraction method as described in claims 1-7 above.

Images