A low-voltage power distribution network phase sequence and topology clustering identification method and system

Abstract

The application discloses a low-voltage distribution network phase sequence and topology clustering identification method and system; three-phase voltage at a district head end and end user voltage and active power time sequence data are acquired and preprocessed; user voltage similarity is calculated, users are divided into three sets corresponding to each phase of a transformer through clustering, and phase sequence identification is completed; for a single-phase set, a local similarity matrix is constructed, two nodes with the largest similarity are found from bottom to top, a virtual node representing an implicit intermediate node is generated if a merging condition is met, voltage time sequence of the virtual node is reconstructed, and power is aggregated; node sets and the similarity matrix are iteratively updated until only a root node is left and is connected with a transformer corresponding phase bus, and a complete tree-shaped topology containing the implicit intermediate node is reconstructed; only end data is used to realize joint identification of the phase sequence and the topology, a restoration bottleneck of an unmeasured branch box is broken, error accumulation is inhibited, and engineering cost is reduced.

Description

Technical Field

[0001] This invention relates to the field of power system distribution network operation monitoring and digital sensing technology, specifically to a method and system for identifying phase sequence and topology clustering in low-voltage distribution networks. Background Technology

[0002] With the continuous advancement of new power system construction, the low-voltage distribution network, as the "last mile" connecting end users and the main grid, is becoming increasingly complex in its operation due to the widespread integration of new interactive devices such as distributed photovoltaics and electric vehicles. Accurate low-voltage distribution area phase sequence and network topology are fundamental for achieving precise line loss calculation, three-phase load imbalance management, and rapid fault location. However, due to frequent historical construction changes and a lack of effective records, the topology and phase sequence information in the distribution network GIS (Geographic Information System) archives often suffers from missing records or delayed updates. Currently, smart meters (AMIs) are widely used on the end-user side, providing massive amounts of measurement data for data-driven online identification of distribution area topology and phase sequence. However, most existing methods separate phase sequence identification from topology reconstruction and heavily rely on measurement information from intermediate nodes, making it difficult to achieve complete network reconstruction using only data from the beginning and end points.

[0003] In phase sequence identification, existing conventional methods mainly include manual on-site verification and characteristic signal injection. The former is time-consuming and labor-intensive, while the latter requires expensive hardware. In recent years, some studies have begun to use the similarity of user voltage time series for phase sequence division, but these methods do not fully consider the deep-seated interference of complex branch structures within the same phase on voltage fluctuation characteristics. For example, two users located deep within different branches of the same phase may have lower voltage fluctuations than two users in different phases but with close electrical distances due to differences in line impedance and load distribution, leading to errors in phase sequence division based solely on voltage similarity. When faced with measurement noise or high three-phase imbalance, existing methods are highly prone to phase sequence misjudgment.

[0004] In terms of topology reconstruction, existing technologies face even more severe challenges. Intermediate branch nodes in low-voltage distribution networks (such as cable branch boxes and overhead line branch joints) typically lack the conditions for installing smart measurement devices, resulting in large areas of "data blind spots" within the network. Existing topology identification algorithms based on correlation coefficients or distance matrices mostly only construct simplified "star" topologies from transformers directly to each user, or incorrectly connect users directly, ignoring the inherent "tree-like radial" physical characteristics of the distribution network and failing to effectively identify and generate intermediate branch nodes without measurement data. Especially in traditional clustering algorithms, the inability to accurately estimate the electrical state of virtual branch nodes lacking measurement leads to a continuous accumulation of state reconstruction errors as the clustering level increases, ultimately failing to reconstruct a complete tree-like topology that conforms to the true physical structure. Therefore, there is an urgent need in this field for a method that can simultaneously solve the problems of phase sequence verification and topology reconstruction with implicit intermediate nodes. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method and system for identifying phase sequence and topology clustering in low-voltage distribution networks.

[0006] To achieve the above objectives, the present invention provides the following technical solution: This application provides a method for phase sequence and topology clustering identification of low-voltage distribution networks, including the following steps: Acquire the three-phase voltage timing data at the beginning of the transformer area and the voltage amplitude of the end users. and active power Time series data, and preprocessed it; Based on the preprocessed user voltage time series data, the similarity between users is calculated. Then, all users are divided into three sets corresponding to the transformers by a clustering algorithm to complete the phase sequence identification. For any single-phase user set, each user is defined as an initial node, the voltage similarity between nodes is calculated, and a local similarity matrix is ​​constructed. Bottom-up agglomerative clustering includes: Find the two nodes with the highest similarity in the current local similarity matrix, and denote them as nodes. and nodes ; If the similarity meets the preset merging conditions, a virtual node representing the unmeasured hidden intermediate node is generated. ; The virtual nodes are reconstructed using the following weighted calculation formula. Voltage timing : in, For time cross-section, , They are nodes , active power, , They are nodes and nodes The voltage, and ; The active power of the two nodes is aggregated as a virtual node. Equivalent load power ; Node and nodes Replace with virtual nodes Update the set of active nodes and utilize the reconstructed voltage. Recalculate Update the local similarity matrix based on the similarity with other nodes in the set; Repeat the above agglomerative clustering steps until only one root node remains in the set of activated nodes. Connect the root node to the corresponding phase bus of the transformer to reconstruct a tree topology containing hidden intermediate nodes.

[0007] Optionally, the preprocessing includes: outlier removal, missing value imputation, multi-source time series alignment, and per-unit processing of the time series data.

[0008] Optionally, the user similarity is calculated based on the Pearson correlation coefficient; the clustering algorithm adopts the K-Means algorithm and uses the measured voltage sequence of each phase of the transformer as a reference for phase matching.

[0009] Optionally, the preset merging condition is: the similarity between two nodes is greater than a preset topology merging threshold; if the maximum similarity in the current local similarity matrix is ​​not greater than the threshold, then merging is stopped, and all currently active nodes are directly connected to the corresponding phase bus of the transformer.

[0010] Optionally, the physical basis of the weighted calculation formula is as follows: In low-voltage radial lines, when the impedance difference between branch lines and trunk lines is ignored, the voltage drop of the upper node is mainly affected by the total load current, and the active power of the lower node approximately reflects the current weight of each branch. The active power weighted average can approximate the voltage of the real hidden node.

[0011] Optionally, the method for updating the local similarity matrix is ​​as follows: delete the rows and columns corresponding to the two nodes being merged in the original matrix, add virtual nodes, and calculate the similarity between the virtual node and all other nodes in the current active set, while maintaining the symmetry of the matrix.

[0012] Optionally, the method independently performs bottom-up agglomerative clustering and reconstruction on each of the three phases, and combines the tree topologies obtained from the reconstruction of each phase with the low-voltage side bus of the transformer as the common connection point to output a complete three-phase tree topology containing the phase relationships and all implicit nodes.

[0013] Optionally, the equivalent load power of the virtual node u It is stored and used in subsequent higher-level clustering and merging, participating as a new active power weight in the voltage reconstruction of the hidden nodes of the previous layer.

[0014] On the other hand, this application provides a low-voltage distribution network phase sequence and topology clustering identification system employing the above method, comprising: The data acquisition and preprocessing module is used to acquire voltage and active power time-series data from the transformer head end and user smart meters and perform preprocessing. The phase sequence identification module is used to calculate the voltage similarity between users and divide users into three-phase independent sets through clustering. A single-phase topology reconfiguration module is used to perform bottom-up agglomerative clustering within each phase, generate virtual hidden nodes, and reconstruct their voltage timing based on the weighted calculation formula described in claim 1. The dynamic update and loop control module is used to update the set of active nodes and the local similarity matrix, and to control the termination of the loop. The topology output module is used to connect the tree topology obtained by reconstructing each phase to the transformer bus and output a complete transformer substation topology map.

[0015] Optionally, the single-phase topology reconfiguration module further includes a physical constraint unit, which is used to replace the weighted calculation formula with the arithmetic mean method to avoid division by zero error when the sum of the active power of the merged nodes is zero; otherwise, the weighted calculation formula is used.

[0016] Compared with the prior art, this application has the following beneficial effects: This application utilizes only the three-phase voltage at the beginning of the transformer substation and the voltage and active power time-series data of the end users, without the need for any additional hardware. It first identifies the phase sequence based on coarse clustering of correlation, then performs bottom-up clustering within each phase and reconstructs the hidden node voltage based on the active power weighting formula. This overcomes the data blind spot of the unmeasuring branch box and accurately restores the complete tree-like radial topology containing hidden intermediate nodes. At the same time, this scheme decouples phase sequence verification from topology reconstruction, fundamentally avoiding cross-phase mismatches. Furthermore, by introducing the physical law of power weighting, it effectively suppresses the accumulation of errors in multi-layer clustering. This solves the technical problems of easy misjudgment of phase sequence, inability to identify intermediate nodes, and star simplified topology not conforming to physical reality in the background technology, and realizes low-cost, high-precision online verification of low-voltage transformer substation archives. Attached Figure Description

[0017] Figure 1 This is a diagram illustrating the household relationship identification process of the present invention.

[0018] Figure 2 This is a flowchart of the topology identification process performed after the household relationship is realized in this invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Furthermore, in this invention, an element referred to as fixed to or disposed on another element may be directly disposed on the other element, or there may be an intermediate element. When an element is considered to be connected to another element, it may be directly connected to the other element, or there may be an intermediate element present simultaneously. The terms vertical, horizontal, left, right, and similar expressions used herein are for illustrative purposes only and do not represent the only possible implementation.

[0021] Example 1 This embodiment, in conjunction with the accompanying drawings, provides a detailed description of the technical solution of this application: like Figures 1-2 As shown: This application provides a method for phase sequence and topology clustering identification of low-voltage distribution networks, characterized by the following steps: Acquire the three-phase voltage timing data at the beginning of the transformer area and the voltage amplitude of the end users. and active power Time series data, and preprocessed it; Based on the preprocessed user voltage time series data, the similarity between users is calculated. Then, all users are divided into three sets corresponding to the transformers by a clustering algorithm to complete the phase sequence identification. For any single-phase user set, each user is defined as an initial node, the voltage similarity between nodes is calculated, and a local similarity matrix is ​​constructed. Bottom-up agglomerative clustering includes: Find the two nodes with the highest similarity in the current local similarity matrix, and denote them as nodes. and nodes ; If the similarity meets the preset merging conditions, a virtual node representing the unmeasured hidden intermediate node is generated. ; The virtual nodes are reconstructed using the following weighted calculation formula. Voltage timing : in, For time cross-section, , They are nodes , active power, , They are nodes and nodes The voltage, and ; The active power of the two nodes is aggregated as a virtual node. Equivalent load power ; Node and nodes Replace with virtual nodes Update the set of active nodes and utilize the reconstructed voltage. Recalculate Update the local similarity matrix based on the similarity with other nodes in the set; Repeat the above agglomerative clustering steps until only one root node remains in the set of activated nodes. Connect the root node to the corresponding phase bus of the transformer to reconstruct a tree topology containing hidden intermediate nodes.

[0022] In this embodiment, the voltage time series of phases A, B, and C are first collected from the low-voltage side of the transformer in the target low-voltage distribution area, and the voltage amplitude is simultaneously read from the smart meters of all end users. and active power The time series data was processed. Preprocessing included removing anomalous jump data using a 3σ sliding window method, filling missing values ​​using cubic spline interpolation, aligning time series offsets using dynamic time warping, and dividing all voltage data by the system reference voltage of 220V to obtain a per-unit value sequence. Then, the Pearson correlation coefficient between user voltage sequences was calculated to generate a global correlation coefficient matrix. The K-Means clustering algorithm was used to forcibly divide all users into three large clusters, and the center of each cluster was matched with the three-phase voltage sequence of the transformer to determine the phase sequence of each user, splitting the transformer area users into three independent single-phase sets: A, B, and C. For phase A (phases B and C were processed similarly), each user was defined as an initial leaf node, and the voltage correlation coefficient between nodes was calculated to construct a local similarity matrix. During bottom-up agglomerative clustering, the two nodes with the highest similarity were searched each time. If their similarity was greater than a preset threshold γ (e.g., 0.95), a virtual node representing a hidden branch box was generated. And using the active power weighted formula Reconstruct its voltage timing while aggregating power. Then delete the node. , Add node Update the set of active nodes and utilize Recalculate The correlation coefficients with the remaining nodes are used to update the local similarity matrix. The above process is repeated until only one root node remains in the set. Then, the root node is connected to the corresponding phase bus of the transformer to obtain the complete tree topology containing hidden nodes.

[0023] In a preferred embodiment, the preprocessing includes: outlier removal, missing value imputation, multi-source time series alignment, and per-unit processing of the time series data.

[0024] In this embodiment, the preprocessing is specifically implemented as follows: For the collected voltage and active power time series, firstly, reasonable upper and lower thresholds for voltage and power are set, and then... The sliding window method is used to remove bad data with abnormal jumps; for short-term missing data caused by communication packet loss, cubic spline interpolation is used to fill in the gaps; to address the issue of inconsistent acquisition frequencies or time offsets between the transformer terminal and each user's meter, the Dynamic Time Warping (DTW) algorithm is used to ensure that all data are strictly synchronized at the same time segment t; finally, all voltage time series data are divided by the system reference voltage. (e.g., 220V) is converted into a per-unit (pu) sequence to eliminate the absolute difference in base voltage amplitude between different nodes.

[0025] In a preferred embodiment, the similarity between users is calculated based on the Pearson correlation coefficient; the clustering algorithm uses the K-Means++ algorithm and phase matching is performed with the measured voltage sequence of each phase of the transformer as a reference.

[0026] In this embodiment, the Pearson correlation coefficients between all preprocessed user voltage time series in the entire transformer area are calculated to generate an N×N global correlation coefficient matrix. The initial cluster centers are optimized using the K-Means++ algorithm to accelerate model convergence and avoid getting trapped in local optima. The cluster centers of the three large clusters are calculated; that is, for each time segment, the arithmetic mean of all user voltage per-unit values ​​within the cluster is calculated to generate an average voltage time series representing the characteristics of that cluster. The similarity (Pearson correlation coefficient) is calculated with the measured voltage series of phases A, B, and C on the low-voltage side of the transformer. Each large cluster is then bound to the corresponding phase sequence with the highest correlation coefficient. The maximum correlation coefficient between each user voltage series and the measured voltage series of its bound phase sequence is extracted. If the maximum correlation coefficient is greater than a preset phase sequence confidence threshold (e.g., 0.8), the verification passes, and the user is determined to belong to that phase. If it is less than or equal to the threshold, the user's phase sequence is determined to be abnormal or in an unidentifiable state, and an alarm is output. After the above clustering and verification, the phase sequence of each normal user is accurately identified, and the users in the entire station area are divided into three independent single-phase user sets: A, B, and C.

[0027] In a preferred embodiment, the preset merging condition is: the similarity between two nodes is greater than a preset topology merging threshold; if the maximum similarity in the current local similarity matrix is ​​not greater than the threshold, then merging is stopped, and all currently active nodes are directly connected to the corresponding phase bus of the transformer.

[0028] In this embodiment, a stringent topology merging threshold is preset. For example, we can set the value to 0.95. In each round of agglomerative clustering, we traverse the local similarity matrix S and find the two nodes with the highest current similarity. and nodes If the similarity is greater than Then determine the node and nodes Physically connected to the same parent branch box, merging and generating virtual nodes are allowed; if the maximum similarity in the current local similarity matrix is ​​less than or equal to... If the merging process stops, all currently active nodes (i.e., unmerged leaf nodes or generated virtual nodes) are directly connected to the low-voltage side busbar of the corresponding phase of the transformer. This process corresponds to the physical scenario in a real distribution network where there may be multiple independent radial sub-branches, ensuring that the algorithm can terminate correctly and output a reasonable topology.

[0029] In a preferred embodiment, the physical basis of the weighted calculation formula is as follows: In a low-voltage radial line, when the impedance difference between the branch line and the main line is ignored, the voltage drop of the upper node is mainly affected by the total load current, and the active power of the lower node approximately reflects the current weight of each branch. The active power weighted average can approximate the voltage of the real hidden node.

[0030] In this embodiment, an active power weighted formula is used when reconstructing the virtual node voltage. The physical basis for this is that low-voltage distribution networks typically have a radial structure, and the impedance of branch lines and main lines is relatively small, so their difference can be ignored. The voltage drop of an upstream node (such as a branch box) mainly depends on the total load current flowing through it. The active power of downstream nodes i and j... and It approximately reflects the magnitude of the load current on each branch. Therefore, by using active power as the weight to perform a weighted average of the voltage of the lower-level nodes, it can more accurately approximate the voltage amplitude of the real hidden nodes (branch boxes). Compared with a simple arithmetic average, this reconstruction method that combines physical laws can effectively suppress the accumulation of errors during subsequent multi-level clustering.

[0031] In a preferred embodiment, the method for updating the local similarity matrix is ​​as follows: delete the rows and columns corresponding to the two merged nodes in the original matrix, add the similarity calculation results of the virtual node and all other nodes in the current active set, and maintain the symmetry of the matrix.

[0032] In this embodiment, each time a virtual node u is generated and its voltage timing is reconstructed, the process is repeated. Then, the local similarity matrix S needs to be dynamically updated; the specific operation is as follows: first, delete the nodes in the original matrix that are similar to the nodes. and nodes The corresponding rows and columns; then calculate the virtual nodes. The voltage correlation coefficient with each of the remaining nodes in the currently active node set is used to add these newly calculated similarities as new rows and columns to the matrix; finally, the matrix is ​​ensured to remain symmetric (i.e., ...). The updated matrix is ​​one row and one column smaller than the original matrix because two rows and two columns were deleted and one row and one column were added, resulting in a net reduction of one row and one column, which is consistent with the total number of currently active nodes, thus providing the correct input for the next round of clustering.

[0033] In a preferred embodiment, the method independently performs bottom-up agglomerative clustering and reconstruction on each of the three phases, and combines the tree topologies obtained from the reconstruction of each phase with the low-voltage side bus of the transformer as the common connection point, outputting a complete three-phase tree topology that includes the phase relationships and all implicit nodes.

[0034] In this embodiment, after completing the phase sequence division of the entire distribution area in step 2, three independent single-phase user sets, A, B, and C, are obtained. For phase A, the bottom-up agglomerative clustering and reconstruction described in steps 3 to 5 yields a tree topology containing all explicit users and implicit intermediate nodes within phase A, with its root node connected to the low-voltage side busbar of phase A of the transformer. Similarly, the same process is executed independently for phases B and C, respectively, yielding their respective tree topologies. Finally, using the low-voltage side busbar of the transformer as a common connection point, the tree topologies reconstructed by the three phases are combined to output a complete three-phase tree topology diagram of the low-voltage distribution area. This topology diagram not only contains the accurate phase sequence information of each user but also includes all implicit branch box nodes derived by the algorithm.

[0035] In a preferred embodiment, the virtual node Equivalent load power It is stored and used in subsequent higher-level clustering and merging, participating as a new active power weight in the voltage reconstruction of the hidden nodes of the previous layer.

[0036] In this embodiment, the calculated virtual node Equivalent load power It is not for one-time use; when the virtual node When a node is merged again in a subsequent higher-level cluster (e.g., with another dummy node or leaf node), the system stores and uses its power timing data. Specifically, in higher-level mergers, if With another node If the merging conditions are met, then reconstruct the upper-level virtual nodes. When the voltage is [value], the formula will be used. Even using The aggregated power is used as the weight; this mechanism allows power information to be passed up the clustering level, ensuring that the entire reconstruction process conforms to Kirchhoff's current law and the principle of power conservation.

[0037] On the other hand, this application provides a low-voltage distribution network phase sequence and topology clustering identification system employing the above method, comprising: The data acquisition and preprocessing module is used to acquire voltage and active power time-series data from the transformer head end and user smart meters and perform preprocessing. The phase sequence identification module is used to calculate the voltage similarity between users and divide users into three-phase independent sets through clustering. A single-phase topology reconfiguration module is used to perform bottom-up agglomerative clustering within each phase, generate virtual hidden nodes, and reconstruct their voltage timing based on the weighted calculation formula described in claim 1. The dynamic update and loop control module is used to update the set of active nodes and the local similarity matrix, and to control the termination of the loop. The topology output module is used to connect the tree topology obtained by reconstructing each phase to the transformer bus and output a complete transformer substation topology map.

[0038] In this embodiment, a complete system is constructed. The data acquisition and preprocessing module is responsible for obtaining raw time-series data from the low-voltage side of the transformer and the user's smart meter, and performing anomaly removal, missing data filling, time-series alignment and per-unit processing. The phase sequence identification module receives the preprocessed user voltage data, calculates the global Pearson correlation coefficient matrix, runs the K-Means clustering algorithm to divide the users into three large clusters, matches them with the three-phase voltage sequence of the transformer, and outputs the phase sequence label of each user. The single-phase topology reconstruction module performs bottom-up agglomerative hierarchical clustering for each phase's user set: it is responsible for finding the optimal merging pair, determining whether the threshold condition is met, generating virtual nodes, and reconstructing the voltage timing of the virtual nodes based on the active power weighted formula; the dynamic update and loop control module is responsible for updating the active node set and local similarity matrix after each merging, and determining whether the termination condition is met (only one root node remains in the set or the maximum similarity is not greater than the threshold); the topology output module combines the tree topologies obtained by each phase independently reconstructing with the transformer low-voltage side bus as the common connection point, and finally outputs a complete three-phase tree topology diagram of the low-voltage distribution area containing complete user phase relationships and all implicit branch box nodes.

[0039] In a preferred embodiment, the single-phase topology reconfiguration module further includes a physical constraint unit, which is used to replace the weighted calculation formula with the arithmetic mean method to avoid division by zero error when the sum of the active power of the merged nodes is zero; otherwise, the weighted calculation formula is used.

[0040] In this embodiment, to enhance the numerical stability and robustness of the algorithm, a physical constraint unit is embedded within the single-phase topology reconfiguration module. This unit checks the two nodes to be merged before each virtual node voltage reconfiguration. and nodes The sum of active power If for a certain time segment If the sum is zero (e.g., during nighttime when there is no load or when meter data is abnormal), directly using the weighted formula will result in a division by zero error. In this case, the physical constraint unit automatically switches the reconstruction strategy, using the arithmetic mean method instead of the weighted formula. This process ensures that the algorithm can run stably under any data conditions, while still prioritizing the use of the weighted formula to maintain physical accuracy on time segments where the actual active power is non-zero.

[0041] Example 2 To make the technical solution of the present invention clearer, based on the above embodiment 1, this embodiment will be described in detail below using a typical low-voltage distribution substation as an example, but the scope of protection of the present invention is not limited to this specific example.

[0042] In this embodiment, the rated voltage of the low-voltage side of the target low-voltage distribution transformer is 380V / 220V, and the system reference voltage is... =220; There are a total of 9 terminal users in the transformer area. In the actual physical topology, 5 users are connected to phase A, and there is an unknown secondary branch box (i.e., the intermediate node has no measurement data). User 1 and user 2 are connected to this branch box; users 3, 4, and 5 are connected through other branches; phases B and C each have 2 users connected, and the structure is relatively simple.

[0043] First, multi-source measurement data acquisition and time-series preprocessing are performed; then, a series of effective voltage values ​​for three phases (A, B, and C) of 96 consecutive time sections (e.g., one point every 15 minutes within a day) are collected from the low-voltage side of the transformer in the distribution area. , , Simultaneously, voltage amplitude values ​​for the same time period are read from all nine users' smart meters. and active power ,in =1,2,…,9, Representing time segments; cleaning the collected raw data: using... The sliding window method (or the isolation forest criterion) identifies and removes abnormal jump points exceeding a reasonable threshold in each time series. For the removed points and short-term missing data caused by communication packet loss, cubic spline interpolation is used to fill in the gaps. Since there may be inconsistencies in the acquisition frequency or slight time offsets between the transformer acquisition terminal and the user's electricity meter, the Dynamic Time Warping (DTW) algorithm is used to align each user's voltage series with the transformer's A-phase voltage series, ensuring that all data are on the same time segment. Strict synchronization is required; finally, all voltage data are divided by the system reference voltage. =220, converted to a per-unit value sequence This is to eliminate the absolute difference in the base voltage amplitude between different nodes.

[0044] After preprocessing, precise identification of user relationships is performed based on coarse correlation clustering. Based on the preprocessed per-unit voltage value sequence, Pearson correlation coefficients are calculated between each pair of the nine users in the entire distribution area, generating a 9×9 global correlation coefficient matrix. The K-Means++ clustering algorithm is then employed (setting the number of clusters). =3, randomly initialized multiple times and the optimal result is taken) all 9 users are forcibly divided into 3 large clusters; calculate the cluster center of each cluster, that is, the average value of the voltage sequence of all users in the cluster, and calculate the similarity with the measured voltage sequences of phase A, phase B, and phase C on the low voltage side of the transformer, and bind each large cluster to the corresponding phase sequence of the transformer; for example, if the correlation coefficient between a cluster center and the phase A voltage sequence is the highest, then all users in that cluster are determined to be phase A; in this way, the phase sequence of each user is accurately identified, and the users in the entire distribution area are divided into three independent single-phase user sets: A, B, and C; in this embodiment, users 1, 2, 3, 4, and 5 are divided into phase A, users 6 and 7 are divided into phase B, and users 8 and 9 are divided into phase C.

[0045] After phase sequence division, topology reconstruction is performed independently for each single phase. The following details the process using phase A as an example. Each end user within the phase A set is defined as the initial leaf node of the topology tree. Currently, there are five active nodes: user 1 to user 5. For these five nodes only, the Pearson correlation coefficient of voltage fluctuations between them is calculated, constructing a 5×5 symmetric local similarity matrix. ,in diagonal elements =0.

[0046] Then, bottom-up clustering and reconstruction of hidden branch nodes begin; the local similarity matrix is ​​traversed. Find the two nodes with the highest current similarity; assuming the similarity between user 1 and user 2 is 0.97; then combine this similarity with a preset topology merging threshold. In comparison, this embodiment takes =0.95; This threshold is selected based on the physical characteristics and empirical statistics of low-voltage distribution networks: Physically, the terminal nodes on the same branch line in a low-voltage distribution area share most of the upstream trunk impedance, and their voltage fluctuations are mainly dominated by the upstream total load current. Therefore, adjacent nodes in the same local physical branch will exhibit extremely high synchronicity in their voltage timing. In terms of empirical verification, statistical analysis of a large number of distribution network simulation tests and actual measurement data shows that the voltage Pearson correlation coefficient between nodes sharing the same non-measuring branch box is usually stable between 0.95 and 0.99, while the similarity between cross-branch nodes is usually lower than this lower limit due to the influence of their respective independent local load disturbances. Therefore, setting it to 0.95 can effectively tolerate the measurement noise of smart meters while accurately identifying node combinations with the same source connection characteristics. Since 0.97 > 0.95, it is determined that User 1 and User 2 are physically connected to the same upstream branch box; therefore, a new virtual node is generated between Node 1 and Node 2. This node represents a hidden branch box in the actual physical power grid where no measurement data is available. Next, the voltage state of this virtual node needs to be reconstructed; the active power time-series data of node 1 and node 2 need to be extracted. , and voltage timing data , Using power as a weight, virtual nodes voltage After weighted derivation and estimation, the reconstructed formula is: in It is a time section, and the denominator is... >0; The physical basis of this formula is as follows: In low-voltage radial lines, when the impedance difference between branch lines and main lines is ignored, the voltage drop of the upper-level node is mainly affected by the total load current, and the active power of the lower-level node approximately reflects the current weight of its respective branch. Therefore, using an active power weighted average can more accurately approximate the voltage of the real hidden node; at the same time, according to Kirchhoff's current law, the voltage of the virtual node is calculated by accumulating the values. Equivalent load power This polymerization power will be used for upward propagation to the next level.

[0047] After a merge is completed, the set of active nodes and the similarity matrix need to be updated. Remove the merged nodes 1 and 2 from the current set of active nodes, and add the newly generated virtual nodes. Add it to the set; at this point, the active node becomes { There are 4 nodes: User 3, User 4, User 5; the reconstructed virtual node voltage timing is used. Recalculate its correlation coefficient with all remaining nodes in the set (users 3, 4, and 5), and replace the original similarity matrix. By analyzing the rows and columns related to nodes 1 and 2, a new 4×4 local similarity matrix is ​​obtained.

[0048] Check the number of currently active nodes. If it is not equal to 1, repeat the steps above: searching for the optimal merging pair, determining the topology physical connection, generating virtual nodes, reconstructing voltage and power, and updating the set and matrix. In the next round of this embodiment, assuming that the similarity between user 4 and user 5 is 0.96, which is greater than 0.95, then merge and generate virtual nodes. Similarly, its voltage is reconstructed using a power-weighted formula. and aggregate power The updated set of active nodes is { User 3 There are 3 nodes in total; the similarity matrix is ​​calculated again, assuming... The similarity with user 3 is 0.93. and The similarity is 0.91, and user 3 and The similarity is 0.72; the current maximum similarity is 0.93, which is not greater than the threshold. Therefore, the aggregation process is stopped, and all currently active nodes (i.e., User 3 It is directly connected to the low-voltage side busbar of phase A of the transformer; at this point, the topology restoration of all visible users and hidden branch boxes within phase A is completed; for phases B and C, the same bottom-up agglomerative clustering and reconstruction process as above is performed completely independently, and finally the tree topology obtained by the independent reconstruction of the three phases is combined with the low-voltage side busbar of the transformer as the common connection point, and the output is a complete three-phase tree topology diagram of the low-voltage distribution area containing complete user-phase relationships and all hidden branch box nodes.

[0049] In this embodiment, the topology merging threshold The initial value is 0.95, but in practical applications, it can be adjusted between 0.93 and 0.98 based on the noise level measured in the transformer area. If, in a certain round of clustering, the maximum similarity in the current local similarity matrix is ​​consistently no greater than... If the active power of two merged nodes is less than or equal to the sum of the active power of the two merged nodes, the algorithm will terminate prematurely, and all currently active nodes will be directly connected to the bus. This corresponds to a physical scenario where multiple independent radial sub-branches exist within the transformer area. Furthermore, when the sum of the active power of two merged nodes is less than or equal to the sum of the active power of the two merged nodes, the algorithm will terminate prematurely. When a certain time section is zero (e.g., during a long period of no load), the arithmetic mean method can be used to avoid division by zero errors. Voltage reconstruction is performed using an alternative weighted formula.

[0050] As can be seen from the above specific implementation methods, the present invention can accurately identify the user phase sequence and reconstruct the sequence including hidden intermediate nodes (such as...) using only the measurement data at the beginning and end of the transformer area. , The complete tree topology effectively solves the technical problems mentioned in the background technology, such as easy misjudgment of phase sequence, inability to identify intermediate nodes, and star simplified topology not conforming to physical reality, and has significant engineering practicality.

[0051] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0052] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A method for phase sequence and topology clustering identification of low-voltage distribution networks, characterized in that, Includes the following steps: Acquire the three-phase voltage timing data at the beginning of the transformer area and the voltage amplitude of the end users. and active power Time series data, and preprocessed it; Based on the preprocessed user voltage time series data, the similarity between users is calculated. Then, all users are divided into three sets corresponding to the transformers by a clustering algorithm to complete the phase sequence identification. For any single-phase user set, each user is defined as an initial node, the voltage similarity between nodes is calculated, and a local similarity matrix is ​​constructed. Bottom-up agglomerative clustering includes: Find the two nodes with the highest similarity in the current local similarity matrix, and denote them as nodes. and nodes ; If the similarity meets the preset merging conditions, a virtual node representing the unmeasured hidden intermediate node is generated. ; The virtual nodes are reconstructed using the following weighted calculation formula. Voltage timing : in, For time cross-section, , They are nodes , active power, , They are nodes and nodes The voltage, and ; The active power of the two nodes is aggregated as a virtual node. Equivalent load power ; Node and nodes Replace with virtual nodes Update the set of active nodes and utilize the reconstructed voltage. Recalculate Update the local similarity matrix based on the similarity with other nodes in the set; Repeat the above agglomerative clustering steps until only one root node remains in the set of activated nodes. Connect the root node to the corresponding phase bus of the transformer to reconstruct a tree topology containing hidden intermediate nodes.

2. The method according to claim 1, characterized in that, The preprocessing includes: outlier removal, missing value imputation, multi-source time series alignment, and per-unit processing of the time series data.

3. The method according to claim 1, characterized in that, The similarity between users is calculated based on the Pearson correlation coefficient; the clustering algorithm adopts the K-Means algorithm and uses the measured voltage sequence of each phase of the transformer as a reference for phase matching.

4. The method according to claim 1, characterized in that, The preset merging condition is: the similarity between two nodes is greater than the preset topology merging threshold; if the maximum similarity in the current local similarity matrix is ​​not greater than the threshold, then merging is stopped, and all currently active nodes are directly connected to the corresponding phase bus of the transformer.

5. The method according to claim 1, characterized in that, The physical basis of the weighted calculation formula is as follows: In low-voltage radial lines, when the impedance difference between branch lines and trunk lines is ignored, the voltage drop of the upper node is mainly affected by the total load current, and the active power of the lower node approximately reflects the current weight of each branch. The active power weighted average can approximate the voltage of the real hidden node.

6. The method according to claim 1, characterized in that, The method for updating the local similarity matrix is ​​as follows: delete the rows and columns corresponding to the two merged nodes in the original matrix, add the similarity calculation results of the virtual node and all other nodes in the current active set, and maintain the symmetry of the matrix.

7. The method according to claim 1, characterized in that, The method independently performs bottom-up agglomerative clustering and reconstruction on each of the three phases, and combines the tree topologies obtained from the reconstruction of each phase with the low-voltage side bus of the transformer as the common connection point, outputting a complete three-phase tree topology that includes the phase relationships and all implicit nodes.

8. The method according to claim 1, characterized in that, The equivalent load power of the virtual node u It is stored and used in subsequent higher-level clustering and merging, participating as a new active power weight in the voltage reconstruction of the hidden nodes of the previous layer.

9. A low-voltage distribution network phase sequence and topology clustering identification system employing the method described in any one of claims 1 to 8, characterized in that, include: The data acquisition and preprocessing module is used to acquire voltage and active power time-series data from the transformer head end and user smart meters and perform preprocessing. The phase sequence identification module is used to calculate the voltage similarity between users and divide users into three-phase independent sets through clustering. A single-phase topology reconfiguration module is used to perform bottom-up agglomerative clustering within each phase, generate virtual hidden nodes, and reconstruct their voltage timing based on the weighted calculation formula described in claim 1. The dynamic update and loop control module is used to update the set of active nodes and the local similarity matrix, and to control the termination of the loop. The topology output module is used to connect the tree topology obtained by reconstructing each phase to the transformer bus and output a complete transformer substation topology map.

10. The system according to claim 9, characterized in that, The single-phase topology reconfiguration module also includes a physical constraint unit, which is used to replace the weighted calculation formula with the arithmetic mean method to perform voltage reconfiguration when the sum of the active power of the merged nodes is zero, so as to avoid division by zero error; otherwise, the weighted calculation formula is used.

Images