A power distribution network multi-branch line fault accurate positioning method based on traveling wave positioning

Abstract

The application relates to a kind of distribution network multi-branch line fault accurate positioning method based on travelling wave positioning, comprising: 1, build double-end travelling wave synchronous acquisition device, collect line travelling wave signal and carry out denoising and time synchronization processing;2, carry out distribution network dynamic topology-travelling wave joint modeling, calculate and establish fault distance matrix using Dijkstra algorithm;3, the fault distance matrix is screened by density clustering algorithm, the data affected by time error is screened out, the data after screening is substituted into optimization model, and the basic data after error correction is calculated;The effective coordinate cluster obtained by clustering is calculated, the branch node coordinates are proposed, and the centroid and node euclidean distance are calculated, finally, the euclidean distance is compared with threshold value, less than, is branch fault, otherwise, is trunk fault, the double-end ranging formula and the trunk-to-branch node formula are used to calculate the results of trunk and branch fault respectively, and the correction of fault positioning result is realized.The application realizes the accuracy of fault positioning.

Description

Technical Field

[0001] This invention belongs to the field of power system fault detection and location technology, specifically relating to a method for accurate fault location of multi-branch lines in distribution networks based on traveling wave positioning. Background Technology

[0002] In power systems, rapid and accurate detection and location of distribution network faults are crucial for ensuring power supply reliability. Traditional traveling wave acquisition devices have many shortcomings. They have poor noise immunity, and the signals acquired in complex electromagnetic environments are easily interfered with, leading to signal distortion and affecting subsequent fault analysis. Furthermore, their synchronization accuracy is insufficient, making it impossible to accurately obtain the time difference of the traveling wave arriving at different acquisition terminals, resulting in large fault location errors.

[0003] Furthermore, existing distribution network fault location technologies face the problem of disconnection between dynamic topology changes and traveling wave propagation models. With the operation and development of the distribution network, line switching, load changes, and other factors cause dynamic changes in the topology. Traditional models are difficult to adapt to these changes and cannot accurately describe the propagation characteristics of traveling waves under dynamic topologies, thus affecting the accuracy of fault location.

[0004] Meanwhile, traditional ranging methods fail to locate faults when dealing with branch dead zones or multiple fault points. When a fault occurs in a specific area of ​​a branch line (branch dead zone) or when multiple fault points exist, traditional methods cannot effectively distinguish and locate the fault location, making it difficult to meet the needs of rapid fault location and repair in actual distribution networks. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention proposes a method for accurate fault location in multi-branch lines of a distribution network based on traveling wave positioning.

[0006] The above-mentioned objective of this invention is achieved through the following technical solution:

[0007] A method for accurate fault location in multi-branch lines of a distribution network based on traveling wave positioning includes the following steps:

[0008] Step 1: Build a dual-end traveling wave synchronous acquisition device, and use the built dual-end traveling wave synchronous acquisition device to acquire the traveling wave signal of the line and perform noise reduction and time synchronization processing.

[0009] Step 2: Based on the traveling wave signal processed in Step 1, perform dynamic topology-traveling wave joint modeling of the distribution network, and use Dijkstra's algorithm to calculate and establish the fault distance matrix;

[0010] Step 3: The fault distance matrix output in Step 2 is filtered using the density clustering algorithm DBSCAN to remove data affected by time errors. The filtered data is then substituted into the optimization model to calculate the basic data after error correction. The geometric centroids of the effective coordinate clusters obtained by DBSCAN clustering are calculated according to the formula to extract the coordinates of the branch nodes and calculate the Euclidean distance between the centroids and the nodes. Finally, the calculated Euclidean distance is compared with a threshold. If it is less than the threshold, it is a branch fault; otherwise, it is confirmed as a trunk fault. The trunk and branch faults are calculated using the double-end ranging formula and the trunk-to-branch node formula, respectively, to correct the fault location results.

[0011] Moreover, the dual-end traveling wave synchronous acquisition device built in step 1 includes a Rogowski coil, a capacitive voltage transformer (CVT), a 10MHz DC, a synchronization module, an FPGA processing chip, and a 4G communication module integrated on the same circuit board. The Rogowski coil is used to acquire current traveling wave signals, and the capacitive voltage transformer is used to acquire voltage traveling wave signals.

[0012] Furthermore, in step 1, the synchronization module adopts a dual-backup structure of a GPS module and a temperature-controlled crystal oscillator. During normal operation, the GPS module outputs 1PPS pulse and NMEA-0183 data per second, which is transmitted to the MCU via a UART interface with a baud rate of 115200. The MCU uses the 1PPS pulse to calibrate the temperature-controlled crystal oscillator with an accuracy of ±1ppm, so that the system clock and UTC synchronization error is ≤1μs. When no satellite signal is received for 30 consecutive seconds, it automatically switches to the crystal oscillator timekeeping mode. The crystal oscillator maintains the reference with a drift rate of ≤1μs / hour. At the same time, it cross-calibrates the clock with the adjacent acquisition device through the IEEE1588 time synchronization protocol to ensure that the cumulative error over 24 hours is ≤24μs.

[0013] Furthermore, step 2 includes:

[0014] Step 2.1: Combine circuit breaker status information and relay protection action signals to collect complete topology information of the distribution network, including node locations, line connection relationships, line lengths and surge impedances. Abstract the collected information into a complex plane vector diagram and establish a two-dimensional coordinate system with the main control substation as the origin, the X-axis as east, and the Y-axis as north. Substation nodes are located using GPS, and branch and terminal nodes are extracted using laser ranging or CAD. The node coordinates, line lengths, and surge impedance parameters are then bound to the data.

[0015] Step 2.2: Synchronously collect the arrival time information of the initial traveling wave of the fault in all substations and establish an initial traveling wave arrival time matrix. Let be the arrival time of the traveling wave at the i-th substation;

[0016] Step 2.3: Identify the validity of the elements in the initial traveling wave arrival time matrix, and then combine the breaker status information and relay protection action signals of each substation to accurately determine the faulty line section;

[0017] Step 2.4: Construct a traveling wave positioning network based on the power grid structure, use the Dijkstra algorithm to search for the shortest paths between all substations in the power grid, and complete the construction of the shortest path distance matrix;

[0018] Step 2.5: Based on the initial traveling wave arrival time matrix established in Step 2.2 and the shortest path distance matrix established in Step 2.4, establish a fault distance matrix D.

[0019] Moreover, in Step 2.4, the construction process of the shortest path distance matrix is as follows:

[0020] First, assume that the power grid topology has n nodes, where the nodes represent substations. Select the line length as the weight and construct a weighted matrix:

[0021] ;

[0022] where, satisfies when node i and j are connected 、when not connected 、when i = j = 0; is the line length between connected nodes;

[0023] Then, use the Dijkstra algorithm to obtain the shortest paths between any two nodes. Let represent the total length of the lines passed by the shortest path between node i and j, and construct the shortest path distance matrix L:

[0024] .

[0025] Moreover, in Step 2.5, the specific correction process includes:

[0026] The specific correction steps are as follows:

[0027] Step 2.5.1: Based on the fact that matrix L is a symmetric matrix, when i < j, let = 0;

[0028] Step 2.5.2: Substitute the corresponding elements of the shortest path distance matrix L and the initial traveling wave arrival time matrix T into the double-ended positioning formula: , and calculate the preliminary fault distance. Among them, d ij is the distance from the fault point to node i on the shortest path line between node i and j, and L ijLet v be the total length of the shortest path between node i and node j, and v be the speed of the traveling wave. The arrival time of the two-ended traveling wave;

[0029] Step 2.5.3: Select the node at one end of the faulty line as the reference node r, and apply the formula... Calculate the fault distance; among which, Let the distance from the fault point to node i be the shortest path between node i and reference node r; finally, use all the calculated fault distances. Establish the fault distance matrix D.

[0030] Furthermore, step 3 includes:

[0031] Step 3.1: Perform DBSCAN clustering on the fault distance matrix D, label the noisy data, and obtain the cluster partition C={ , ,⋯, };

[0032] Step 3.2, Threshold algorithm based on the distance between the centroid of the coordinate cluster and the branch node: For the effective clusters obtained by DBSCAN clustering... Perform coordinate family centroid calculation. ,in Let be the number of coordinate points in the k-th cluster. They are the kth clusters. The x and y coordinates of the i-th specific coordinate point within the range;

[0033] Step 3.3: Extract the coordinates of all branch nodes from the topology matrix. Among them, a branch node refers to a pivotal point connecting the main trunk and branch lines;

[0034] Step 3.4: Perform distance calculation and threshold determination: Calculate the Euclidean distance between the centroid of the coordinate cluster and each branch node. Set distance threshold This is 1 / 5 to 1 / 3 of the average length of the corresponding branch line;

[0035] If it exists The fault was determined to be a branch line fault, and the fault point was located on the branch line to which branch node b belonged.

[0036] If all The fault was determined to be a main line fault, and the fault point was located on the main line segment.

[0037] Step 3.5: Accurately locate the fault point:

[0038] For branch line faults, the distance from the fault point to the branch node is calculated using the two-end distance measurement formula, with branch node b as the starting point:

[0039] in, Let v be the arrival time of the two-ended traveling wave, and v be the wave velocity. This is the corrected shortest path length. This represents the distance from the main branch node to the branch node in the topological matrix;

[0040] For faults on the main line: the location of the fault point on the main line can be directly calculated using the two-end distance measurement formula.

[0041] Furthermore, the specific steps of step 3.1 are as follows:

[0042] Step 3.1.1: Select any data point Y in the fault distance matrix D that has not been searched. Then search for the neighborhood distance threshold of data point Y. Define the set of data points found within the neighborhood distance threshold of data point Y as Ne. If the number of data points in Ne is less than the minimum number of sample points within the neighborhood distance threshold of data point Y, then mark all data in Ne as noise; otherwise, mark all data in Ne as the current cluster. And update the cluster partition C={ , ,⋯, };

[0043] Step 3.1.2: Update dataset Ne and cluster C. Search for the neighborhood distance threshold of any data point y in dataset Ne. Define the dataset NE that searches within the neighborhood distance threshold of data point y. If the number of data points in NE is greater than the minimum number of sample points within the neighborhood distance threshold of data point y, merge datasets NE and Ne and redefine them as Ne. If y does not belong to any cluster C, add y to cluster C.

[0044] Step 3.1.3: Repeat step 3.1.2 until Ne = {0};

[0045] Step 3.1.4: Repeat steps 3.1.1 to 3.1.3 to traverse the fault distance matrix D and divide the data into clusters or noise; finally, output the cluster division result: C={ , ,⋯, }

[0046] The advantages and positive effects of this invention are as follows:

[0047] 1. The method of this invention employs a dual-end traveling wave acquisition system to acquire traveling wave signals. By integrating a wideband sensor, a high-speed ADC, and a high-precision synchronization module, and utilizing advanced communication and denoising technologies, the noise immunity and synchronization accuracy of the acquisition device are significantly improved, providing high-quality signal data for subsequent fault analysis. In dual-end traveling wave acquisition systems, the acquired traveling wave signals are often subject to interference from complex electromagnetic environments, and traditional acquisition devices have poor noise immunity. The variational mode decomposition (VMD) hardware accelerator (FPGA implementation) can perform real-time denoising on the acquired traveling wave signals. It utilizes parallel computing characteristics to quickly perform variational mode decomposition on the signal, separating and removing noise components while retaining effective signal components. This not only reduces data transmission volume and lowers the pressure on the communication module but also improves signal quality, providing a more accurate data foundation for subsequent fault analysis and location based on traveling wave signals, and solving the problem of poor noise immunity of traditional acquisition devices affecting fault analysis.

[0048] 2. The method of this invention achieves an organic combination of topology structure and traveling wave propagation model through the distribution network topology-traveling wave joint modeling method. It can adapt to the dynamic changes of distribution network topology, accurately describe the traveling wave propagation characteristics, and provide a reliable model basis for fault location.

[0049] 3. The multi-branch fault location decision method adopted in this invention effectively solves the problem of location failure of traditional ranging methods in complex fault scenarios. Through a two-layer location mechanism and threshold algorithm, it realizes accurate identification of fault branches and precise location of fault points, improves the accuracy and efficiency of fault location in distribution networks, significantly shortens fault repair time, reduces power outage losses, and improves power supply reliability. Attached Figure Description

[0050] Figure 1 This is a system structure block diagram of the present invention;

[0051] Figure 2 This is a flowchart of data acquisition and processing in the dual-end traveling wave acquisition device of the present invention;

[0052] Figure 3 This is a timing diagram of the ping-pong caching mechanism involved in this invention;

[0053] Figure 4 This is a flowchart of the present invention that uses Dijkstra's method to determine whether the shortest path includes the route where the fault occurred.

[0054] Figure 5 This is a schematic diagram of the clustering algorithm of this invention;

[0055] Figure 6 This is a flowchart illustrating the automatic differentiation logic for trunk / branch faults in this invention. Detailed Implementation

[0056] The structure of the present invention will be further described below with reference to the accompanying drawings and embodiments. It should be noted that these embodiments are descriptive and not limiting.

[0057] A method for accurate fault location in multi-branch lines of a distribution network based on traveling wave localization; please refer to... Figures 1-6 Its inventive point is: including the following steps:

[0058] Step 1: Construct a dual-end traveling wave synchronous acquisition device. Use this device to acquire the traveling wave signal from the line and perform noise reduction and time synchronization processing. Specifically, this includes:

[0059] Step 1.1 Hardware Setup:

[0060] First, the hardware architecture of the dual-end traveling wave acquisition device was built: the Rogowski coil responsible for current traveling wave acquisition, the CVT (capacitive voltage transformer) responsible for voltage traveling wave acquisition, the 10MHz high-speed ADC, the synchronization module, the FPGA processing chip, and the 4G communication module were integrated on the same circuit board, forming an integrated hardware platform for "sensing-sampling-processing-transmission". During acquisition, the zero-sequence current was first extracted from the three-phase instantaneous current as the core analysis object, and the calculation formula is as follows:

[0061]

[0062] in, , , These are the three-phase instantaneous currents, It is the zero-sequence current.

[0063] Subsequently, a second-order low-pass filter and an INA128 instrumentation amplifier were configured at the signal input to preprocess the original traveling wave signal, filtering out high-frequency interference and conditioning the signal amplitude. This ensured that the voltage input to the FPGA remained stable between 0 and 3.3V, meeting the level requirements of the chip's I / O interface and laying the hardware foundation for subsequent signal processing. The cutoff frequency of the second-order low-pass filter was set to 5MHz, and the adjustable gain of the INA128 instrumentation amplifier was 0–40dB.

[0064] Step 1.2, Signal Processing: Real-time noise reduction is achieved through FPGA-based VMD.

[0065] After the hardware platform is built, signal preprocessing is performed first. Before the signal enters the FPGA, the zero-sequence current is de-trended to eliminate DC bias, resulting in the processed signal. After normalization to eliminate the influence of dimensions, the formula is:

[0066]

[0067] in, The signal to be decomposed is normalized; max(⋅) and min(⋅) are the maximum and minimum values ​​of the signal, respectively; the normalized signal is filtered by a second-order low-pass filter to eliminate high-frequency noise, and an instrumentation amplifier is used to achieve signal gain conditioning to ensure that the voltage range of the input FPGA is 0~3.3V, which meets the requirements of the chip's IO interface.

[0068] The variational model was then constructed, and the core variational problem of VMD was solved in the FPGA using the hardware description language (Verilog). The modal components uk and Lagrange multipliers were defined. Iterative update equation:

[0069]

[0070] uk(t) is the k-th modal component, ωk is its center frequency, δ(t) is the Dirac function, and ∗ is the convolution operation.

[0071] The frequency domain segmentation of the signal is achieved by a 64th-order FIR filter through a pipeline structure, processing 16 sampling points per clock cycle.

[0072] Then, modal decomposition is performed. Based on the characteristics of the traveling wave signal (frequency range 10kHz~10MHz), the preset number of decomposed modes K=8 is dynamically adjusted through register configuration. Utilizing the parallel characteristics of the FPGA, the update iteration of 8 modal components is run simultaneously, with each iteration cycle being 256 clock cycles (@100MHz clock), and the time taken for a single decomposition is ≤25.6μs. By calculating the spectral energy ratio of each mode, a threshold is set (e.g., energy ratio <5%) to automatically identify noise modes, and the corresponding modal components are set to zero through combinational logic circuits.

[0073] Finally, to ensure timing consistency, the VMD accelerator and GPS synchronization module share a 10MHz reference clock, which is multiplied to a 100MHz operating clock via a phase-locked loop (PLL) to ensure that the synchronization error between the timestamp of the decomposed signal and the sampling clock is ≤10ns. Then, a ping-pong buffer mechanism (dual-port RAM) is adopted. When the ADC completes sampling of a frame of data (1024 points), it triggers the VMD accelerator to read the data, and at the same time, the ADC starts sampling the next frame, realizing a pipelined operation of acquisition-decomposition with a data processing delay of ≤1ms.

[0074] Step 1.3, Synchronization and Communication Configuration:

[0075] After signal denoising, the data needs to be accurately time-stamped and reliably transmitted through synchronization and communication configuration. The synchronization module adopts a dual-backup structure of "GPS + temperature-controlled crystal oscillator": During normal operation, the GPS module outputs 1PPS pulse and NMEA-0183 data per second, which is transmitted to the MCU via a UART interface with a baud rate of 115200. The MCU uses the 1PPS pulse to calibrate the temperature-controlled crystal oscillator with an accuracy of ±1ppm, ensuring that the system clock and UTC synchronization error is ≤1μs. When no satellite signal is received for 30 consecutive seconds, it automatically switches to crystal oscillator timekeeping mode. The crystal oscillator maintains the reference with a drift rate of ≤1μs / hour. At the same time, it cross-calibrates the clock with adjacent acquisition devices through the IEEE1588 time synchronization protocol to ensure that the cumulative error over 24 hours is ≤24μs. In terms of communication, a dual redundant fiber optic + 4G link is configured, each connected to the corresponding communication interface. The edge computing module (ARM Cortex-A53 processor) performs wavefront detection on the VMD-denoised signal: it uses db4 wavelet for 3-level decomposition, extracts high-frequency coefficients of 1~5MHz, and identifies the arrival time of the first wavefront of the traveling wave using the modulus maxima method, with an accuracy of ≤1 sampling point (80ns). When the wavefront is detected, metadata containing a timestamp of "GPSUTC time + sampling point offset", signal amplitude and frequency characteristics is generated and stored in a 1MB buffer FIFO, waiting to be uploaded by the communication link.

[0076] Step 2: Based on the traveling wave signal processed in Step 1, perform dynamic topology-traveling wave joint modeling of the distribution network, and calculate and establish the fault distance matrix using Dijkstra's algorithm; specifically including:

[0077] During the daily operation of the distribution network, the SCADA system first acquires real-time power grid operation data. When the system detects changes in the topology, such as line switching or the addition of new branches, it immediately triggers a dynamic update mechanism for the topology matrix, automatically correcting the branch parameters. The correction of the wave impedance parameter in the topology matrix follows a specific formula: In the formula t measured t is the measured traveling wave propagation time. modeled For model prediction time, the correction step size is ≤5%; for newly added nodes, their initial coordinates are first entered from construction data and marked "to be verified," and then, within one week of operation, the model's prediction time is adjusted based on actual traveling wave propagation time. The inversion correction is performed, and the corrected data is marked as "valid" to ensure that the topology matrix is ​​consistent with the actual power grid state.

[0078] When a fault occurs on a certain line in the traveling wave positioning network, the fault traveling wave will spread from the fault point along the line to the entire power transmission network from near to far. At this time, it is necessary to first collect the complete topological structure information of the distribution network, including parameters such as node positions, line connection relationships, line lengths, and wave impedances. Abstract these information into a complex plane vector diagram, and establish a two-dimensional coordinate system (X-axis east, Y-axis north) with the main control substation as the origin - where the substation nodes are located by GPS (accuracy ≤ 0.5m), the branch and terminal nodes are extracted by laser ranging or CAD (accuracy ≤ 2m), and the node coordinates, line lengths, and wave impedance parameters are bound to construct an initial topological matrix; then synchronously collect the time information of the initial traveling waves of the entire network reaching all substations, and establish an initial traveling wave arrival time matrix is the time when the initial traveling wave reaches the i-th substation, and the preliminary data preparation is completed.

[0079] Subsequently, the elements in the initial traveling wave arrival time matrix are identified for validity, and combined with the breaker status information and relay protection action signals of each substation, the fault line is accurately discriminated; on this basis, a traveling wave positioning network is constructed according to the power grid structure, and the Dijkstra algorithm (which is applicable to edges with non-negative weights and has advantages in calculation speed, stability, and engineering implementation ability) is used to search for the shortest paths between all substations in the power grid: First, assume that the power grid topology has n nodes (the nodes represent substations), and select the line length as the weight to construct a weighted matrix , where satisfies "when node i and j are connected is the line length between the connected nodes), when not connected , when i = j = 0"; then use this algorithm to find the shortest path between any two nodes, and use to represent the total length of the lines passed by the shortest path between node i and j, and construct a shortest path distance matrix

[0080] After the shortest path distance matrix L is constructed, the fault distance matrix D is constructed. Specifically:

[0081] Step1: Since the shortest path distance matrix L is a symmetric matrix ( ), to avoid repeated calculations, when i < j, let = 0;

[0082] Step: Substitute the corresponding elements of the shortest path distance matrix L and the initial traveling wave arrival time matrix T into the double-end positioning formula: , and calculate the preliminary fault distance. Among them, d ij is the distance of the fault point from node i on the shortest path line between node i and j, L ijLet v be the total length of the shortest path between node i and node j, and v be the speed of the traveling wave. The arrival time of the two-ended traveling wave;

[0083] Step 3: Select a node at one end of the faulty line as the reference node r, and apply the formula... Calculate the fault distance; among which, Let the distance from the fault point to node i be the shortest path between node i and reference node r; finally, use all the calculated fault distances. Establish the fault distance matrix D;

[0084] Step 3: First, the fault distance matrix D output in Step 2 is filtered using the DBSCAN density clustering algorithm to identify data affected by time errors. Then, the filtered data is substituted into the optimization model to calculate the error-corrected baseline data. Finally, the effective coordinate clusters obtained from DBSCAN clustering are clustered according to the formula... Calculate the geometric centroid, extract the coordinates of the branch nodes, and calculate the Euclidean distance between the centroid and the nodes. Finally, compare the calculated Euclidean distance with the threshold. If the value is less than a certain threshold, it indicates a branch fault; otherwise, it is confirmed as a trunk fault. Trunk and branch faults are calculated using the double-ended ranging formula and the trunk-to-branch node formula, respectively, to correct the fault location results. The specific process includes:

[0085] DBSCAN is a density-based clustering algorithm that clusters samples based on the density of their distribution. It does not require a pre-defined number of clusters and effectively handles noise. Specifically, the algorithm uses a set of parameters [Eps, MinPts] to describe the density of the sample distribution. Here, Eps is the neighborhood distance threshold, and MinPts is the minimum number of sample points within the neighborhood distance threshold. This type of algorithm overcomes the shortcomings of distance-based algorithms (such as K-Means), which are easily affected by outliers and cannot accurately cluster non-spherical samples. The DBSCAN used in this invention can discover clusters of arbitrary shapes and is insensitive to noisy data. The calculation process of DBSCAN clustering is as follows:

[0086] Step 1: Select any data point Y in the fault distance matrix D that has not been searched, and then search for the neighborhood distance threshold (Eps) of data point Y. Define the set of data points searched within the neighborhood distance threshold (Eps) of data point Y as Ne; if the number of data points in Ne is less than the minimum number of sample points (MinPts) within the neighborhood distance threshold of data point Y, then mark all data points in Ne as noise; otherwise, mark all data points in Ne as the current cluster. And update the cluster partition C={ , ,⋯,}

[0087] Step 2: Update dataset Ne and cluster C. Search within dataset Ne for the neighborhood distance threshold (Eps) of any data point y. Define the dataset searched within the neighborhood distance threshold (Eps) of data point y as NE. If the number of data points in NE is greater than the minimum number of samples (MinPts) within the neighborhood distance threshold of data point y, merge datasets NE and Ne and redefine them as Ne. If y does not belong to any cluster C, add y to cluster C.

[0088] Step 3: Repeat step 2 until Ne = {0}.

[0089] Step 4: Repeat steps 1-3 to iterate through the fault distance matrix D, and divide the data into clusters or noise. Finally, output the clustering results: C={ , ,⋯, }

[0090] For determining the neighborhood distance threshold (Eps) and the minimum number of sample points within the neighborhood distance threshold (MinPts), based on the analysis of the DBSCAN algorithm principle, the neighborhood distance threshold (Eps) and the minimum number of sample points within the neighborhood distance threshold (MinPts) work together to describe the density of the neighborhood sample distribution. Therefore, considering the actual situation of power grid fault section judgment, the parameter MinPts is calculated using the formula: MinPts=(n-1) / 2, n≥10, and rounded down, where n is the number of detectors. After extensive simulation experiments, the corresponding parameters are determined to be Eps=0.0065 and MinPts=7. This identifies the bad data affected by time errors, and the remaining normal data is used to repeatedly construct the optimization model and calculate its optimal solution to obtain the accurate location result after error correction. Finally, based on the analysis of the fault area, a threshold algorithm based on the distance between the centroid of the coordinate cluster and the branch node is used to automatically distinguish between main trunk / branch faults and perform fault location.

[0091] Threshold algorithm based on the distance between the centroid of the coordinate cluster and the branch node: Effective clusters obtained by DBSCAN clustering Perform coordinate family centroid calculation. ,in Let be the number of coordinate points in the k-th cluster. They are the kth clusters. The x and y coordinates of the i-th specific coordinate point.

[0092] Branch node coordinate extraction: Extract the coordinates of all branch nodes from the topology matrix. A branch node refers to a pivotal point connecting the main trunk and branch lines, such as a pole or junction box.

[0093] Distance Calculation and Threshold Determination: Calculate the Euclidean distance between the centroid of the coordinate cluster and each branch node. Set distance threshold (This is 1 / 5 to 1 / 3 of the average length of the corresponding branch line, with a default of 50 meters.)

[0094] If it exists The fault was determined to be a branch line fault, and the fault point was located on the branch line to which branch node b belongs.

[0095] If all The fault was determined to be a main line fault, and the fault point was located on the main line segment.

[0096] Precise fault location: Branch line fault: Taking branch node b as the starting point, calculate the distance from the fault point to the branch node using the double-ended distance measurement formula:

[0097] in, Let v be the arrival time of the two-ended traveling wave, and v be the wave velocity. This is the corrected shortest path length. This represents the distance from the main branch node to the branch node in the topology matrix.

[0098] Main line fault: The location of the fault point on the main line is directly calculated using the two-end distance measurement formula.

[0099] Although embodiments and drawings of the present invention have been disclosed for illustrative purposes, those skilled in the art will understand that various substitutions, variations and modifications are possible without departing from the spirit and scope of the present invention and the appended claims. Therefore, the scope of the present invention is not limited to the contents disclosed in the embodiments and drawings.

Claims

1. A method for accurate fault location of multi-branch lines in power distribution networks based on traveling wave positioning, characterized in that: Comprise the following steps: Step 1, build a double-end traveling wave synchronous acquisition device, and utilize the double-end traveling wave synchronous acquisition device to collect the traveling wave signals of the line and perform denoising and time synchronization processing; Step 2, according to the traveling wave signals processed in step 1, perform dynamic topology-traveling wave joint modeling of the distribution network, and utilize Dijkstra algorithm to calculate and establish a fault distance matrix; Step 3, the fault distance matrix finally output in step 2 is filtered through density clustering algorithm DBSCAN, and the data affected by time error is filtered out, then the filtered data is substituted into an optimization model to calculate the basic data after error correction; then the effective coordinate clusters obtained through DBSCAN clustering are calculated according to a formula to obtain branch node coordinates, and the centroid and node Euclidean distance are calculated, finally the calculated Euclidean distance is compared with a threshold value, if the Euclidean distance is less than the threshold value, it is a branch fault, otherwise, it is confirmed as a trunk fault, the trunk and branch faults are calculated by using a double-end ranging formula and a trunk-to-branch node formula respectively to obtain results, and the correction of the fault location result is realized; Step 3 specifically comprises: Step 3.1, DBSCAN clustering is performed on the fault distance matrix, and noise data is marked to obtain cluster division C={ , ,⋯, }; Step 3.2, threshold algorithm based on the distance between the coordinate cluster centroid and the branch node: the effective clustering cluster obtained by DBSCAN clustering Carrying out coordinate cluster centroid calculation, calculating the coordinate cluster centroid , wherein is the number of coordinate points of the kth cluster, is the horizontal and vertical coordinate values of the i-th specific coordinate point in the kth clustering cluster , respectively. Step 3.3, extracting all branch node coordinates from the topology matrix wherein a branch node refers to a hub point connecting a trunk with a branch line; Step 3.4, distance calculation and threshold determination: calculate the Euclidean distance between the coordinate cluster centroid and each branch node , set the distance threshold as 1 / 5~1 / 3 of the average length of the corresponding branch line If there is , it is determined that the branch line is faulty, and the fault point is located on the branch line to which the branch node b belongs; If all , determine that the backbone line is faulty, and the fault point is located in the backbone line segment; Step 3.5, accurate positioning of the fault point: For branch line fault, the distance from fault point to branch node b is calculated by combining double-end ranging formula: ; wherein, is the two-terminal traveling wave arrival time, v is the traveling wave wave speed, is the corrected shortest path length, is the distance from the backbone node to the branch node in the topology matrix; For trunk line fault: directly calculate the position of the fault point on the trunk line through the double-end ranging formula.

2. The method for precise fault location of multi-branch line of power distribution network based on traveling wave positioning as claimed in claim 1, characterized in that: The double-end traveling wave synchronous acquisition device built in step 1 comprises a Rogowski coil, a capacitive voltage transformer CVT, a 10MHz DC, a synchronization module, an FPGA processing chip and a 4G communication module integrated on the same circuit board, wherein the Rogowski coil is used to collect current traveling wave signals, and the capacitive voltage transformer CVT is used to collect voltage traveling wave signals.

3. The method for precise fault location of multi-branch line of distribution network based on traveling wave positioning as claimed in claim 2, characterized in that: In step 1, the synchronization module adopts a double-backup structure form of GPS module and constant temperature crystal oscillator; in normal operation, the GPS module outputs 1PPS pulse and NMEA-0183 data every second, which are transmitted to the MCU through the UART interface with a baud rate of 115200, and the MCU calibrates the constant temperature crystal oscillator with an accuracy of ±1ppm by using the 1PPS pulse, so that the synchronization error of the system clock and UTC is ≤1μs; when satellite signals are not received for 30 seconds continuously, the system automatically switches to the crystal oscillator time keeping mode, and the crystal oscillator maintains the reference with a drift rate of ≤1μs / hour, and at the same time, the clocks are mutually calibrated through the IEEE1588 time synchronization protocol with adjacent acquisition devices, so that the 24-hour cumulative error is ≤24μs.

4. The method for precise fault location of multi-branch line of distribution network based on traveling wave positioning as claimed in claim 1, characterized in that: Step 2 comprises: Step 2.1, combine the breaker state information and the relay protection action signal to collect the complete topology structure information of the distribution network, including node position, line connection relationship, line length and wave impedance; abstract the collected information into a complex plane vector diagram, and establish a two-dimensional coordinate system with the main control substation as the origin, the X axis as east and the Y axis as north; wherein the substation nodes are positioned by GPS, the branch and terminal nodes are positioned by laser ranging or CAD extraction, and the node coordinates, line length and wave impedance parameters are bound; Step 2.2, synchronously collect the time information of the initial traveling wave of the whole network fault reaching all substations, and establish an initial traveling wave arrival time matrix is the traveling wave arrival time for the i-th substation. Step 2.3, identify the effectiveness of the elements in the initial traveling wave arrival time matrix, and then accurately identify the line section where the fault occurs in combination with the breaker state information and the relay protection action signal of each substation; Step 2.

4. Construct the traveling wave positioning network according to the power grid structure, search the shortest path between all substations in the power grid by using Dijkstra algorithm, and complete the construction of the shortest path distance matrix; Step 2.

5. Establishing a fault distance matrix D based on the initial matrix of wave arrival times established in step 2.2 and the shortest path distance matrix established in step 2.

4. and the shortest path distance matrix established in step 2.

4.

5. The method for precise fault location of multi-branch line of power distribution network based on traveling wave positioning as claimed in claim 4, characterized in that: In step 2.4, the construction process of the shortest path distance matrix is as follows: First, set the power grid topology to have n nodes, the node represents a substation, select the line length as the weight, and construct the weighted matrix: ； wherein when nodes i and j are connected , when nodes i and j are not connected , when i = j = 0; is the length of the line between connected nodes; Then the shortest path between any two nodes is calculated by Dijkstra algorithm, and the shortest path distance matrix L is constructed as follows: where Lij represents the total length of the shortest path between node i and j. 。 6. The method for precise fault location of multi-branch line of power distribution network based on traveling wave positioning as claimed in claim 5, characterized in that: In step 2.5, the construction process of the fault distance matrix D is as follows: Step 2.5.

1. Let = 0; Step 2.5.2, substitute the shortest path distance matrix L and the corresponding element of the initial line wave arrival time matrix T into the double-end positioning formula: , calculate the preliminary fault distance, wherein d ij is the distance from the fault point on the shortest path line between node i and j to node i, L ij is the total length of the line on which the shortest path between node i and node j passes, and v is the line wave propagation speed, is the double-end line wave arrival time; Step 2.5.3, select the node at one end of the fault line as the reference node r, and calculate the fault distance according to the formula Convert the fault distance; wherein, is the distance from the fault point to node i on the shortest path line between node i and reference node r; finally, use all the converted fault distances to establish the fault distance matrix D.

7. The method for precise fault location of a multi-branch line of a power distribution network based on traveling wave positioning as claimed in claim 1, wherein: Step 3.1.2, update the data set Ne and the clustering cluster C, search the neighborhood distance threshold of any data point y in the data set Ne; define the data set searched in the neighborhood distance threshold of the data point y as NE, if the number of data in NE is greater than the minimum sample point number in the neighborhood distance threshold of the data point y, then merge the data sets NE and Ne, and redefine Ne; if y does not belong to any clustering cluster C, then add y to the clustering cluster C; Step 3.1.

1. If one data point Y in the optional fault distance matrix D is not searched, then search the neighborhood distance threshold of the data point Y, and define the data point set searched in the neighborhood distance threshold of the data point Y as Ne; if the number of data in Ne is less than the minimum sample point number in the neighborhood distance threshold of the data point Y, then mark all data in Ne as noise; otherwise, mark all data in Ne as the current clustering cluster , and update the clustering cluster division C={ , ,⋯, }. Step 3.1.3, repeat step 3.1.2 until Ne={0}; ​ Step 3.1.4: Repeat steps 3.1.1 to 3.1.3 to traverse the fault distance matrix D and divide the data into clusters or noise; finally, output the cluster division result: C={ , ,⋯, }

Images