A Dynamic Identification Method for Low-Voltage Line Transformers Based on Electrical Features and Real-Time Correlation Matrix

Abstract

This invention relates to a method for dynamic identification of low-voltage line-transformers based on electrical characteristics and a real-time correlation matrix. The method includes real-time acquisition of electrical characteristic data from the transformer side and the user side of a low-voltage distribution network; dimensionality reduction of the three-phase voltage data in the user-side electrical characteristic data to obtain a single-phase equivalent voltage sequence; reduction of the single-phase equivalent voltage sequence to obtain multiple virtual user clusters; calculation of the virtual active power and virtual daily electricity consumption time-series data for each virtual user cluster; wherein each virtual user cluster includes multiple real users; solving a joint optimization model based on the virtual active power and virtual daily electricity consumption of the virtual user clusters, and the electrical characteristic data from the transformer side, to obtain a real-time correlation matrix representing the optimal connection relationship between the transformer and the virtual user clusters; comparing the real-time correlation matrix with the historical ledger correlation matrix to perform dynamic identification and anomaly assessment of the line-transformer relationship, obtaining the dynamic identification and anomaly assessment results of the low-voltage line-transformer relationship.

Description

Technical Field

[0001] This invention relates to the field of power system distribution automation and intelligent technology, and in particular to a dynamic identification method for low-voltage line transformers based on electrical characteristics and real-time correlation matrices. Background Technology

[0002] With the continuous expansion of the power grid and the gradual improvement of distribution automation, identifying the wiring relationships between distribution transformers (hereinafter referred to as distribution transformers) and electricity users in low-voltage distribution networks has become a fundamental task for lean operation and maintenance and intelligent dispatching of the power grid. Currently, most power grid companies still rely on manually maintained equipment ledgers to record the connection information between distribution transformers and users. However, in actual operation, due to frequent adjustments to user connections, on-site construction modifications, and delays in information entry, there are numerous problems with incorrect connections, omissions, or outdated information in the distribution transformer ledgers. This not only affects the accuracy of core operations such as load forecasting and line power flow analysis but also poses a threat to the safe and stable operation of the distribution network.

[0003] In recent years, with the widespread adoption of Advanced Metering Infrastructure (AMI) and Supervisory Control and Data Acquisition (SCADA) systems, power systems have gained the ability to collect and record various electrical quantities such as voltage, current, and electricity consumption in real time. Building on this foundation, some research has begun to explore data-driven line-transformer relationship identification methods, hoping to infer the transformer node to which a user belongs by leveraging the statistical characteristics or similarities between electrical quantities. However, these methods still have significant limitations.

[0004] First, regarding data utilization, traditional methods often rely solely on single electrical parameters of users, such as electricity consumption or single-phase voltage data, neglecting the coupling relationships between multi-dimensional electrical information such as three-phase voltage fluctuations and active power characteristics. This results in poor adaptability of the identification model to complex scenarios. Second, in terms of model construction, many methods lack physical constraint mechanisms and fail to utilize the principle of energy conservation or line loss models, thus failing to ensure the physical rationality of the identified results relative to the power grid. Third, in terms of identification strategies, some methods simply use Euclidean distance or correlation coefficients to determine the similarity between users, ignoring the complex evolution of electrical characteristics over time, making it difficult to guarantee the stability and accuracy of the results.

[0005] Furthermore, because the structure and load of the distribution network are constantly changing, and the line-transformer relationship itself is also subject to real-time changes, traditional static models lack dynamic tracking and updating mechanisms, making it difficult to detect abnormal changes in the ledger information in a timely manner. In terms of evaluation and verification, many current studies only formally compare the identification results with historical ledgers, lacking scientific error measurement methods and quantifiable evaluation models, making it difficult to support large-scale deployment and automated anomaly identification in practical applications.

[0006] In summary, the current identification of line-transformer relationships in low-voltage distribution networks still faces many challenges, such as insufficient data dimensions, imperfect model mechanisms, coarse similarity discrimination, weak dynamic identification capabilities, and lack of evaluation mechanisms. These challenges result in low accuracy and poor adaptability in identifying line-transformer relationships in low-voltage distribution networks. Summary of the Invention

[0007] Based on the above analysis, the present invention aims to provide a dynamic identification method for low-voltage line transformers based on electrical characteristics and real-time correlation matrices, in order to solve the technical problems of inaccurate identification of line transformer relationships and inability to dynamically detect wiring anomalies in existing low-voltage distribution networks due to reliance on static ledgers and insufficient data dimensions.

[0008] This invention provides a method for dynamic identification of low-voltage line transformers based on electrical characteristics and a real-time correlation matrix, comprising the following steps:

[0009] Real-time acquisition of electrical characteristic data from the transformer side and user side of the low-voltage distribution network;

[0010] The three-phase voltage data in the electrical characteristic data of the user side is reduced in dimension to obtain a single-phase equivalent voltage sequence; the single-phase equivalent voltage sequence is reduced to obtain multiple virtual user clusters; the virtual active power and virtual daily electricity consumption time series data of each virtual user cluster are calculated; wherein, each virtual user cluster includes multiple real users;

[0011] Based on the virtual active power and virtual daily electricity consumption of each virtual user cluster, as well as the electrical characteristic data of the distribution transformer side, the joint optimization model is solved to obtain the real-time correlation matrix that characterizes the optimal connection relationship between the distribution transformer and each virtual user cluster.

[0012] The real-time correlation matrix is ​​compared with the historical ledger correlation matrix to perform dynamic identification and anomaly assessment of line change relationships, thereby obtaining the dynamic identification and anomaly assessment results of low-voltage line change relationships.

[0013] Furthermore, the electrical characteristic data on the user side includes the three-phase voltage, three-phase current, and daily electricity consumption time-series data of real users;

[0014] The electrical characteristic data of the distribution transformer side includes the active power and daily electricity consumption time-series data of the distribution transformer side;

[0015] By using the same sampling frequency, the electrical characteristic data on the user side and the electrical characteristic data on the distribution transformer side are aligned in time sequence.

[0016] Furthermore, the dimensionality reduction of the three-phase voltage data in the electrical characteristic data on the user side to obtain a single-phase equivalent voltage sequence includes:

[0017] For each sampling time The three-phase voltage measurements are used to solve the equivalent formula for the three-phase voltage equal phase angle. A numerical iteration method is then employed to solve this formula, yielding the first... The single-phase equivalent voltage value at each sampling time;

[0018] The single-phase equivalent voltages obtained at each sampling time are combined sequentially into a single-phase equivalent voltage sequence according to the sampling time.

[0019] The equivalent formula for the three-phase voltage equal phase angle is as follows:

[0020] ;

[0021] in, The first The phase voltages of phases A, B, and C at each sampling time. , Number of sampling times This represents the single-phase equivalent voltage after dimension reduction at the corresponding time.

[0022] Furthermore, based on the similarity of voltage fluctuations under the same distribution transformer, the single-phase equivalent voltage sequence is reduced to obtain multiple virtual user clusters, including:

[0023] Calculate the voltage fluctuation similarity between two users based on the single-phase equivalent voltage sequence of each user;

[0024] Iterate through all users and merge users whose voltage fluctuation similarity is higher than a preset similarity threshold into the same virtual user cluster to obtain multiple virtual user clusters.

[0025] Further, the calculation of the virtual active power and virtual daily electricity consumption time-series data for each virtual user cluster includes:

[0026] Based on the single-phase equivalent voltage sequence of all users in the virtual user cluster, the mean value corresponding to each sampling time is calculated to obtain the virtual voltage of the virtual user cluster at each sampling time.

[0027] Based on the three-phase current at each sampling time on the user side, the corresponding single-phase equivalent current value is calculated by normalization; based on the average of the single-phase equivalent current values ​​of all users in the virtual user cluster at each sampling time, the virtual current of the virtual user cluster at the corresponding sampling time is obtained.

[0028] The virtual voltage and virtual current at each sampling time are reduced to obtain the virtual active power of the virtual user cluster at the corresponding time; the virtual active power of each virtual user cluster at all sampling times is combined to form the virtual active power time series data of that virtual user cluster.

[0029] The virtual daily electricity consumption of the virtual user cluster is obtained by summing the daily electricity consumption of all users in the virtual user cluster at the corresponding time. The virtual daily electricity consumption of each virtual user cluster at all times constitutes the virtual daily electricity consumption sequence of that virtual user cluster.

[0030] Furthermore, a joint optimization model is constructed, including:

[0031] Based on the principle of energy conservation of active power and daily electricity consumption between distribution transformers and virtual user clusters, a loss coefficient model is constructed with the objective of minimizing the error between the total active power and daily electricity consumption of the distribution transformer side and the virtual user clusters. ;

[0032] The virtual voltage of each virtual user cluster is normalized to obtain the normalized virtual voltage; the Euclidean distance between any two virtual user clusters is calculated based on the normalized virtual voltage; and the profile coefficient of the virtual user cluster is calculated based on the Euclidean distance. ;

[0033] based on , The objective function of the joint optimization model is constructed as follows:

[0034] .

[0035] Furthermore, solving the joint optimization model includes:

[0036] Based on the objective function and constraints of the aforementioned loss coefficient model, and using the historical ledger correlation matrix as initial input, the CPLEX solver is employed to minimize the energy loss coefficient. To obtain the initial connection relationship between the distribution transformer and the virtual user cluster;

[0037] Starting from the initial connection relationship between the distribution transformer and the virtual user cluster, maximize the profile coefficient. Voltage similarity;

[0038] To obtain the objective function of the joint optimization model that is maximized ;

[0039] At this point, the connection relationship between the distribution transformer and the virtual user cluster is the optimal connection relationship, yielding the current real-time correlation matrix between the distribution transformer and the virtual user cluster. .

[0040] Furthermore, the real-time correlation matrix is ​​compared with the historical ledger correlation matrix to perform dynamic identification and anomaly assessment of line-change relationships, obtaining the results of dynamic identification and anomaly assessment of low-voltage line-change relationships, including:

[0041] The real-time correlation matrix is ​​compared with the historical ledger correlation matrix to obtain the difference matrix between the two.

[0042] Calculate the Frobenius norm of the difference matrix to measure the overall correlation offset.

[0043] Obtain the line loss change rate and voltage mean offset, and combine them with the Frobenius norm of the difference matrix to construct a multi-index weighted evaluation, and calculate the multi-index weighted evaluation value.

[0044] The weighted evaluation value of the multi-indicator is compared with the corresponding dynamic threshold. If the weighted evaluation value of the multi-indicator is greater than the dynamic threshold, there is an abnormality in the line change, which is marked as an anomaly point to assist in the correction of the ledger or the connection check.

[0045] Furthermore, the objective function and constraints of the loss coefficient model are as follows:

[0046] The objective function is:

[0047] ;

[0048] in, For the index of the distribution transformer, For the set of all distribution variables; For sampling time index, Total number of sampling time points; To be at the sampling time , No. Daily electricity consumption of each distribution transformer; This represents the total number of virtual user clusters. To be at the sampling time , No. Virtual daily electricity consumption of a virtual user cluster; For the first The first distribution transformer and the first The connection relationships between virtual user clusters; To be at the sampling time , No. The active power value of each transformer; To be at the sampling time , No. Virtual active power of each virtual user cluster; For the first A set of virtual user clusters with multiple associations;

[0049] The constraints are:

[0050] ;

[0051] in, This represents the total number of distribution transformers; This means that each virtual user cluster can only connect to one transformer. Number of virtual user clusters; This means that each distribution transformer is connected to at least one virtual user cluster.

[0052] Furthermore, the maximum similarity between virtual user clusters is evaluated based on the voltage similarity of the maximum profile coefficient;

[0053] The contour coefficient It is determined by both the density within virtual user clusters and the separation between virtual user clusters, as shown below:

[0054] ;

[0055] in, For the first Profile coefficients of a virtual user cluster; For the first The density of all real users within a virtual user cluster For virtual user clusters The user set in; For the first The degree of separation between a virtual user cluster and other virtual user clusters. For different Other virtual user clusters; For virtual user clusters and The Euclidean distance between them; For other virtual user clusters The total number of real users included.

[0056] Compared with the prior art, the present invention can achieve at least one of the following beneficial effects:

[0057] 1. This invention integrates multi-dimensional electrical features (three-phase voltage, current, active power, and electricity consumption) and constructs a joint optimization model. It combines the loss coefficient reflecting energy conservation with the contour coefficient reflecting curve similarity, making a comprehensive judgment from multiple physical dimensions. This significantly improves the accuracy of linear relationship identification and its robustness under different operating conditions. It solves the problems of low identification accuracy and poor adaptability caused by the single data dimension (such as only electricity consumption or single-phase voltage) and coarse similarity discrimination (such as simply using Euclidean distance) of traditional methods.

[0058] 2. This invention introduces the principle of energy conservation as the core physical constraint and constructs a differential comparison mechanism between the real-time correlation matrix and historical ledgers. By utilizing the Frobenius norm and the dynamic threshold of the sliding time window, it not only ensures that the identification results conform to the basic physical laws of the power grid, but also realizes real-time and automatic detection and tracking of abnormal changes in line-transformer relationships, enhancing the physical rationality and dynamic tracking capability of the results. It overcomes the technical problems of traditional methods lacking physical constraint mechanisms and static models being unable to reflect dynamic changes in wiring relationships in a timely manner.

[0059] 3. This invention establishes a quantifiable and calculable scientific evaluation system by introducing a multi-index weighted evaluation function (combining Frobenius norm, line loss change rate, and voltage mean offset) and dynamic threshold determination. This makes the identification results no longer dependent on a simple formal comparison with historical records, enabling automatic and accurate anomaly location, providing reliable technical support for large-scale automated operation and maintenance of power grid companies; it achieves quantifiable scientific evaluation and automated anomaly identification. It overcomes the shortcomings of existing methods in terms of evaluation mechanism deficiencies, making it difficult to support large-scale deployment and automated anomaly identification.

[0060] In this invention, the above-described technical solutions can be combined with each other to achieve more preferred combinations. Other features and advantages of this invention will be set forth in the following description, and some advantages may become apparent from the description or be learned by practicing the invention. The objects and other advantages of this invention can be realized and obtained from what is particularly pointed out in the description and drawings. Attached Figure Description

[0061] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts.

[0062] Figure 1 This is a flowchart of a low-voltage line transformer dynamic identification method based on electrical features and a real-time correlation matrix in an embodiment of the present invention;

[0063] Figure 2 This is a table analyzing the recognition performance of the joint optimization model and the traditional single recognition model in this embodiment of the invention.

[0064] Figure 3 This is a diagram showing the recognition effect after introducing a difference matrix and Frobenius norm to set a static threshold in an embodiment of the present invention.

[0065] Figure 4 The embodiment of the present invention shows the recognition effect after introducing the difference matrix and Frobenius norm to set a dynamic multi-index weighted threshold. Detailed Implementation

[0066] Preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, which form part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not intended to limit the scope of the present invention.

[0067] This invention proposes a dynamic identification method for low-voltage line transformers based on electrical features and a real-time correlation matrix. It constructs an electrical feature matrix by real-time collection of multi-dimensional data such as voltage, current, and electricity consumption of users and distribution transformers in the distribution network. Users are categorized using voltage similarity to generate a set of virtual user clusters. A joint optimization model is constructed by combining energy conservation and similarity indices to dynamically identify the connection relationship between distribution transformers and virtual user clusters. Furthermore, a historical ledger comparison mechanism is introduced, using a difference matrix, Frobenius norm, and a multi-index weighted evaluation function to achieve real-time detection of line transformer anomalies.

[0068] "Line-to-transformer" is an abbreviation in the power industry for "the connection relationship between distribution lines and distribution transformers". This invention can be used to improve the accuracy of distribution network wiring relationship identification and the level of intelligent operation. It is applicable to scenarios such as wiring review, ledger verification, and power dispatch auxiliary decision-making in low-voltage distribution areas for power grid companies, so as to support the digital transformation and intelligent development of low-voltage distribution networks.

[0069] A specific embodiment of the present invention discloses a dynamic identification method for low-voltage line transformers based on electrical characteristics and a real-time correlation matrix, such as... Figure 1 As shown, it includes the following steps:

[0070] Step S1: Real-time acquisition of electrical characteristic data from the transformer side and user side of the low-voltage distribution network;

[0071] Step S2: Dimensionally reduce the three-phase voltage data in the electrical characteristic data of the user side to obtain a single-phase equivalent voltage sequence; perform reduction on the single-phase equivalent voltage sequence to obtain multiple virtual user clusters; calculate the virtual active power and virtual daily electricity consumption time series data of each virtual user cluster; wherein, each virtual user cluster includes multiple real users;

[0072] Step S3: Based on the virtual active power and virtual daily electricity consumption of each virtual user cluster, as well as the electrical characteristic data of the distribution transformer side, solve the joint optimization model to obtain the real-time correlation matrix that characterizes the optimal connection relationship between the distribution transformer and each virtual user cluster.

[0073] Step S4: Compare the real-time correlation matrix with the historical ledger correlation matrix to perform dynamic identification and anomaly assessment of the line change relationship, and obtain the dynamic identification and anomaly assessment results of the low-voltage line change relationship.

[0074] Step S1, specifically.

[0075] The electrical characteristic data on the user side includes the three-phase voltage, three-phase current, and daily electricity consumption time-series data of real users;

[0076] The electrical characteristic data of the distribution transformer side includes the active power and daily electricity consumption time-series data of the distribution transformer side;

[0077] By using the same sampling frequency, the electrical characteristic data on the user side and the electrical characteristic data on the distribution transformer side are aligned in time sequence.

[0078] For example, this invention collects electrical characteristic data from the distribution transformer side and the user side every 15 minutes, totaling 96 sampling data points per day. Three-phase voltage, three-phase current, and active power are collected every 15 minutes, for a total of 96 data points, so n=96. Subsequent data point numbers are all replaced by n. The first sampling point is 00:00, and sampling occurs every 15 minutes, with the next sampling time being 00:15, and so on, corresponding to the same time points.

[0079] For example, the electricity consumption is collected once a day, with only one data point. The data point collected each day corresponds to the three-phase voltage, three-phase current, and active power data at the last sampling time of the day.

[0080] Based on the planning and design of the power supply area, the number of distribution transformers and electricity users within the low-voltage distribution network is clearly defined. Within the power supply area, there are... Each distribution transformer and Each user constitutes a distribution transformer set. Aggregating electricity users .

[0081] For example, real-time electrical characteristic data of electricity users are obtained through an electricity consumption information collection system, including the three-phase voltage, three-phase current, and daily electricity consumption on the user side, and a user three-phase voltage matrix is ​​constructed. User current matrix and user electricity consumption matrix .

[0082] The three-phase voltage timing data for each user is as follows:

[0083] ;

[0084] Each element The data includes the voltage data for phases A, B, and C. For the first Individual electricity users; This represents the number of sampling points.

[0085] The three-phase current timing data for each user are as follows:

[0086] ;

[0087] Each element The data includes the current data for phases A, B, and C.

[0088] The daily electricity consumption time-series data for each user is as follows:

[0089] ;

[0090] Thus, the voltage, current, and daily electricity consumption matrix for each user is obtained, as shown in formulas (4)-(6):

[0091] ;

[0092] ;

[0093] ;

[0094] in, For the first A user voltage matrix; For the first One user, Three-phase voltage data at each sampling point; For the first A user current matrix; For the first One user, Three-phase current data at each sampling point; For the first Electricity consumption matrix of individual users; For the first Daily electricity consumption per user. (Top right corner icon) It is represented as a transpose matrix.

[0095] The advantage of matrix transpose is that, and In the matrix, each column represents a complete data sequence for a user across all time points. This encapsulates each user's data in a separate column vector, resulting in a very clear data structure that facilitates subsequent calculations.

[0096] Real-time electrical characteristic data of the distribution transformer side is obtained through a scheduling data acquisition and monitoring system, including active power and daily electricity consumption information of the distribution transformer side, and an active power matrix of the distribution transformer is constructed. and daily electricity consumption matrix .

[0097] The active power and daily electricity consumption data sequence on the distribution transformer side are shown in formulas (7)-(8):

[0098] ; ;

[0099] in, For the first Time-series data of active power of each distribution transformer; For the first Individual distribution, Active power data at each sampling point; For the first Time-series data of daily electricity consumption of each distribution transformer; For the first Daily electricity consumption data for each distribution transformer.

[0100] For example, the active power of the distribution transformer is sampled every 15 minutes, and the daily workload is sampled once a day. Thus, the active power and daily electricity consumption matrix for each distribution transformer is obtained, as shown in formulas (9)-(10):

[0101] ;

[0102] ;

[0103] in, For the first The active power matrix of each distribution transformer; For the first The daily electricity consumption matrix of each distribution transformer.

[0104] The purpose of step S1 is to synchronously collect multi-dimensional electrical characteristic time-series data from the distribution transformer side and the user side, construct the corresponding electrical characteristic data matrix, and provide a standardized data foundation for subsequent line-transformer relationship identification.

[0105] Step S2 includes steps S21-S23.

[0106] Step S21: Dimensionally reduce the three-phase voltage data in the electrical characteristic data of the user side to obtain a single-phase equivalent voltage sequence.

[0107] The step of dimensionality reduction of the three-phase voltage data in the electrical characteristic data of the user side to obtain a single-phase equivalent voltage sequence includes:

[0108] For each sampling time The three-phase voltage measurements are used to solve the equivalent formula for the three-phase voltage equal phase angle. A numerical iteration method is then employed to solve this formula, yielding the first... The single-phase equivalent voltage value at each sampling time;

[0109] The single-phase equivalent voltages obtained at each sampling time are combined sequentially into a single-phase equivalent voltage sequence according to the sampling time.

[0110] The equivalent formula for the three-phase voltage equal phase angle is as follows:

[0111] ;

[0112] in, The first The phase voltages of phases A, B, and C at each sampling time. , Number of sampling times This represents the single-phase equivalent voltage after dimension reduction at the corresponding time.

[0113] The user voltages for phases A, B, and C are respectively denoted as:

[0114] Phase A voltage value ;

[0115] Phase B voltage value ;

[0116] C-phase voltage value .

[0117] To convert the three-phase voltage vector into a single-phase equivalent voltage value, dimensionality reduction is performed using formula (11):

[0118] This represents the single-phase equivalent voltage value after dimension reduction at the corresponding sampling time. It was obtained through numerical iteration.

[0119] Formula (11) is about Nonlinear equations, where These are the three-phase voltage values ​​from the known electrical characteristic data collected from the user side. For the variable to be solved, since it cannot be solved directly, a numerical iteration method is used to solve it. The steps are as follows:

[0120] Step 1, set Initial values ​​are as follows:

[0121] ;

[0122] in, for The initial value.

[0123] The second step is to rewrite formula (11) as an error function. As shown below:

[0124] ;

[0125] The goal is to optimize the error function using a numerical iterative algorithm. .

[0126] The third step is to use Newton's iteration method, leveraging derivatives to accelerate convergence, as shown below:

[0127] ;

[0128] in, In the first During the nth iteration The estimated value of the single-phase equivalent voltage at each sampling time; In the first During the nth iteration, the 1st The updated value of the single-phase equivalent voltage at each sampling time; In the estimated value The error function value calculated at the location; Error function In the estimated value The derivative value calculated at that point.

[0129] For example, the maximum number of iterations is set to 100, and the three-phase voltage data at each sampling time are... Perform the above iterations independently to obtain the single-phase equivalent voltage at each sampling time. Combine the single-phase equivalent voltages at all sampling times to obtain the single-phase equivalent voltage sequence for each user.

[0130] Step S22: Calculate the single-phase equivalent voltage sequence of all real users to obtain multiple virtual user clusters.

[0131] Based on the similarity of voltage fluctuations under the same distribution transformer, the single-phase equivalent voltage sequence is reduced to obtain multiple virtual user clusters, including:

[0132] Calculate the voltage fluctuation similarity between two users based on the single-phase equivalent voltage sequence of each user;

[0133] Iterate through all users and merge users whose voltage fluctuation similarity is higher than a preset similarity threshold into the same virtual user cluster to obtain multiple virtual user clusters.

[0134] Single-phase equivalent voltage Real users with highly similar fluctuation trends are grouped together to reduce the dimensionality of electrical feature data, thereby constructing a highly representative virtual user cluster. The similarity between any two real users... The calculation is as follows:

[0135] ;

[0136] in, For user A's single-phase equivalent voltage sequence, the first... The value at each sampling time; The first phase in the single-phase equivalent voltage sequence of user B The value at each sampling time; This represents the number of sampling times for the user's three-phase voltage sequence.

[0137] For example, the preset similarity threshold is set to 0.95; in specific applications, it can be changed according to specific needs.

[0138] By using statistical distribution analysis, a histogram of similarity distribution for all real user pairs was calculated. The inflection point of the high similarity interval was observed. When the similarity threshold was greater than 0.95, the proportion of user pairs decreased significantly (from 94% to 6%). Therefore, a similarity threshold of 0.95 was approximated. When the similarity between any two users was greater than this preset similarity threshold, their voltage fluctuation trends were considered to be consistent, and the real users were assigned to the same virtual user cluster. This resulted in multiple virtual user clusters, denoted as […]. Virtual user clusters have One real user, that is .

[0139] The meaning of virtual user clusters can be illustrated by an example: a distribution transformer has 20 real electricity users, and the clusters are formed by calculating a preset similarity. , Compared with the preset similarity threshold, 16 users are merged into one virtual user cluster, and the other 4 users are merged into one virtual user cluster, thus obtaining two virtual user clusters under one distribution.

[0140] Each virtual user cluster is treated as a single electricity user for subsequent calculations, resulting in only two users under one distribution transformer, thus reducing the dimensionality of the data. At this point, the entire low-voltage distribution network has multiple virtual user clusters.

[0141] In theory, all users under a distribution transformer can be grouped into a single virtual user cluster. However, in reality, some users under the same distribution transformer will not be grouped into the same virtual user cluster. Therefore, this step merges real users into multiple virtual user clusters, reducing the data input dimensions and the amount of computation in this method.

[0142] Through the above steps, assuming a low-voltage distribution network has obtained... A virtual user cluster is constructed, and a virtual user cluster set is created. .

[0143] Step S23: Calculate the virtual active power and virtual daily electricity consumption time-series data for each virtual user cluster.

[0144] The calculation of virtual active power and virtual daily electricity consumption time-series data for each virtual user cluster includes:

[0145] Based on the single-phase equivalent voltage sequence of all users in the virtual user cluster, the mean value corresponding to each sampling time is calculated to obtain the virtual voltage of the virtual user cluster at each sampling time.

[0146] Based on the three-phase current at each sampling time on the user side, the corresponding single-phase equivalent current value is calculated by normalization; based on the average of the single-phase equivalent current values ​​of all users in the virtual user cluster at each sampling time, the virtual current of the virtual user cluster at the corresponding sampling time is obtained.

[0147] The virtual voltage and virtual current at each sampling time are reduced to obtain the virtual active power of the virtual user cluster at the corresponding time; the virtual active power of each virtual user cluster at all sampling times is combined to form the virtual active power time series data of that virtual user cluster.

[0148] The virtual daily electricity consumption of the virtual user cluster is obtained by summing the daily electricity consumption of all users in the virtual user cluster at the corresponding time. The virtual daily electricity consumption of each virtual user cluster at all times constitutes the virtual daily electricity consumption sequence of that virtual user cluster.

[0149] (1) Calculate the virtual voltage of the virtual user cluster at each sampling time.

[0150] The virtual voltage sequence of a virtual user cluster is denoted as , by virtual user cluster The voltage data of all real users within the system are averaged at the sampling time, as shown below:

[0151] ;

[0152] ;

[0153] in, For virtual user clusters All users in The average value of the single-phase equivalent voltage at each sampling time point; Indicates at the sampling time Virtual user clusters The average of the single-phase equivalent voltages for all users in the system; Indicates at the sampling time ,user Normalized single-phase equivalent voltage.

[0154] Thus, the virtual voltage of each sampling moment of the virtual user cluster is obtained, and the virtual voltage of each virtual user cluster at all sampling moments is formed into the virtual voltage sequence of the virtual user cluster.

[0155] (2) Calculate the virtual current of the virtual user cluster at the corresponding time.

[0156] First, the three-phase currents on the user side are normalized. Let the three-phase currents on the user side be A-phase current. Phase B current C-phase current The root mean square (RMS) method is used for dimensionality reduction to calculate the single-phase equivalent current. As shown below:

[0157] ;

[0158] Based on the single-phase equivalent current values ​​of all real users in the virtual user cluster, the average value is calculated according to the sampling time to obtain the virtual current of the virtual user cluster at the corresponding sampling time. The virtual currents at all sampling times form the virtual current sequence of the virtual user cluster.

[0159] The virtual current sequence of a virtual user cluster is denoted as As shown below:

[0160] ;

[0161] ;

[0162] in, For virtual user clusters All users in The average of the single-phase equivalent current at each sampling time; Indicates at the sampling time Virtual User Cluster The average current of all users in the system; Indicates at the sampling time Below, user Normalized single-phase equivalent current.

[0163] (3) Obtain the virtual active power time series data and virtual daily electricity consumption sequence of the virtual user cluster, and obtain the virtual active power and virtual daily electricity consumption of the virtual user cluster.

[0164] Virtual voltage and virtual current combined power factor Calculate the virtual active power As shown below:

[0165] ;

[0166] in, Power factor Characterizing the degree of phase synchronization of voltage and current, for example, residential and general industrial and commercial users default to... That is, approximately 85% of the apparent power is converted into active power.

[0167] Active power sequence of virtual user clusters and daily electricity consumption sequence As shown below:

[0168] ;

[0169] ;

[0170] in, This represents the total daily electricity consumption of all users in the set. For all users in the set The sum of active power at each moment.

[0171] For user sets whose similarity meets the threshold condition, virtual user clusters are formed. The virtual active power and virtual daily electricity consumption of each virtual user cluster are calculated as follows:

[0172] Virtual User Cluster Daily electricity consumption , and the Total active power at each sampling time , respectively equal to set The sum of the corresponding values ​​of all users at that moment is calculated as follows:

[0173] ;

[0174] ;

[0175] in, ; For virtual user clusters The total daily electricity consumption of all users within the system; Let t be the total active power of all users in the virtual user cluster. Represents virtual user clusters Any user in the list; For virtual user clusters Daily electricity consumption of one of the users When sampling time t, the set The active power of user e.

[0176] The purpose of step S2 is to construct a virtual user cluster by normalizing and reducing the dimensionality of the three-phase electrical data on the user side and performing clustering calculations, so as to reduce the data dimensionality and improve the stability of features, and provide optimized input data for subsequent line-transformer relationship identification.

[0177] Step S3 includes steps S31-S33.

[0178] Step S31: Obtain the historical ledger association matrix .

[0179] The historical ledger association matrix is ​​used to represent the known and stable topological connections between distribution transformers and virtual user clusters during the period of correct historical ledgers.

[0180] The historical ledger association matrix is ​​obtained through the following steps, including:

[0181] Obtain historical time periods of stable wiring in low-voltage distribution networks, and obtain historical electrical characteristic data of the distribution transformer side and user side in low-voltage distribution networks during these historical time periods;

[0182] Based on historical electrical characteristic data from the distribution transformer side and the user side, step S2 is executed to obtain multiple virtual user clusters;

[0183] Each virtual user cluster is uniquely mapped to a distribution transformer to obtain the corresponding mapping relationship; one distribution transformer maps to one or more virtual user clusters, and one virtual user cluster maps to only one distribution transformer.

[0184] Based on the obtained mapping relationship, a two-dimensional historical ledger association matrix is ​​constructed. Among them, if virtual user clusters Belongs to distribution transformer ,but matrix elements It is 1 if it is true, otherwise it is 0.

[0185] For example, the historical time period is a one-year time cycle.

[0186] To characterize the first The first distribution transformer and the first The connection relationships between virtual user clusters are defined in the matrix elements. ,as follows:

[0187] Formula (26)

[0188] in, For the index of the distribution transformer, For indexing virtual user clusters, Historical ledger association matrix Matrix elements in; Indicates the first The set of virtual user clusters associated with each distribution transformer is used to reflect the topological connection relationship between distribution transformers and virtual user clusters in the low-voltage distribution network.

[0189] =1, indicating the first The virtual user cluster is connected to the first One variant;

[0190] =0, indicating the first The virtual user cluster is not connected to the first Each variable.

[0191] Formula (27)

[0192] common A virtual user cluster, Columns, total One distribution change, OK, for A 3D matrix; For matrix The Middle The column vector of the column; express All column vectors are arranged horizontally.

[0193] Data on distribution transformers and electricity users with a static acquisition rate close to 100% are selected. Then, multiple virtual user clusters are obtained through step S2. A mapping relationship between distribution transformers and virtual user clusters is constructed to obtain the historical ledger association matrix. .

[0194] Step S32: Construct a joint optimization model.

[0195] Construct a joint optimization model, including:

[0196] Based on the principle of energy conservation of active power and daily electricity consumption between distribution transformers and virtual user clusters, a loss coefficient model is constructed with the objective of minimizing the error between the total active power and daily electricity consumption of the distribution transformer side and the virtual user clusters. ;

[0197] The virtual voltage of each virtual user cluster is normalized to obtain the normalized virtual voltage; the Euclidean distance between any two virtual user clusters is calculated based on the normalized virtual voltage; and the profile coefficient of the virtual user cluster is calculated based on the Euclidean distance. ;

[0198] based on , The objective function of the joint optimization model is constructed as follows:

[0199] ;

[0200] (1) Assessment of linear energy difference based on loss coefficient:

[0201] Based on the principle of energy conservation between distribution transformers and virtual user clusters, at any sampling time The distribution transformer and its virtual user clusters satisfy the conservation relationship of electricity consumption and active power, as shown below:

[0202] ;

[0203] in, Line loss, errors generated during data acquisition and transmission, and They are respectively Time of the first The total electricity consumption and total active power of each distribution transformer and They are respectively Time of the first The virtual daily electricity consumption and virtual active power of each virtual user cluster.

[0204] To maintain the energy balance of the low-voltage distribution network, the optimization objective is to minimize the error between the output of the distribution transformer and the actual reception of the virtual user cluster. A loss coefficient optimization function is constructed as follows:

[0205] ;

[0206] in, It is a loss coefficient optimization function; To minimize the overall difference between the total output energy of the distribution transformer and the sum of the energies of all virtual user clusters;

[0207] To be at the sampling time No. The daily electricity consumption measurement value of each distribution transformer is known data, which is collected in real time from the dispatch and monitoring system and represents the total output of the distribution transformer at a specific moment. To be at the sampling time No. The virtual daily electricity consumption of each virtual user cluster is calculated from the electricity consumption data on the user side in step S2.

[0208] Let be a binary decision variable, representing the first... The first distribution transformer and the first The connection relationships between virtual user clusters; this variable is the final output that the joint optimization model needs to solve, all This forms a real-time correlation matrix describing the entire low-voltage distribution network topology.

[0209] To be at the sampling time No. The active power measurement value of each distribution transformer; and Similarly, this refers to known data collected in real time from the scheduling and monitoring system;

[0210] The virtual active power sequence for all virtual user clusters;

[0211] For the daily electricity consumption time series of all virtual user clusters;

[0212] This refers to the virtual user cluster affiliation weight to be optimized;

[0213] “ " indicates multiplication, that is, summing the element-wise products of the input matrix and the associated vector;

[0214] Using the historical correlation matrix X as the initial input to the joint optimization model reduces the number of iterations. Each iteration of the real-time correlation matrix generates a virtual user cluster classification result, which can be considered as the expected set. Solving the real-time correlation matrix is ​​essentially an optimization process of the expected set. The final output is the optimal real-time correlation matrix X*, reflecting the optimal connection relationship between the distribution transformer and the virtual user cluster, and has the same dimension as the historical correlation matrix X.

[0215] The objective function The value range of is [0,1]. The parameter update is achieved by using the gradient descent method to minimize the parameter and ensure the dynamic balance of energy supply and demand of the line.

[0216] The objective function and constraints of the loss coefficient model are as follows:

[0217] The objective function is:

[0218] ;

[0219] in, For the index of the distribution transformer, For the set of all distribution variables; For sampling time index, Total number of sampling time points; To be at the sampling time , No. Daily electricity consumption of each distribution transformer; This represents the total number of virtual user clusters. To be at the sampling time , No. Virtual daily electricity consumption of a virtual user cluster; For the first The first distribution transformer and the first The connection relationships between virtual user clusters; To be at the sampling time , No. The active power value of each transformer; To be at the sampling time , No. Virtual active power of each virtual user cluster; For the first A set of virtual user clusters with multiple associations;

[0220] The constraints are:

[0221] ;

[0222] in, This represents the total number of distribution transformers; This means that each virtual user cluster can only connect to one transformer. Number of virtual user clusters; This means that each distribution transformer is connected to at least one virtual user cluster.

[0223] Each virtual user cluster belongs exclusively to a single distribution transformer. The non-empty constraint for distribution transformers means that each distribution transformer is connected to at least one virtual user to prevent unloaded distribution transformers from occurring.

[0224] The physical meaning of formula (31) is to transform the physical laws of the power grid into a computable mathematical optimization problem through the loss coefficient model. The problem of identifying the line-transformer relationship between the distribution transformer and the virtual user cluster is essentially the problem of solving the correlation matrix that characterizes the connection relationship between the distribution transformer and the user.

[0225] (2) Voltage similarity assessment based on profile coefficient.

[0226] Profile coefficients optimize the attribution relationship between virtual users and distribution transformers by adjusting the connectivity matrix. Cohesion is used to characterize... Characterizing virtual user clusters Within the same distribution transformer, the degree of voltage similarity among other virtual user clusters; utilizing separation degree Capable of representing virtual user clusters The degree of voltage similarity between virtual user clusters and those under different distribution transformers.

[0227] Virtual voltage of normalized virtual user clusters:

[0228] ;

[0229] Number of virtual user clusters; This is the normalized virtual voltage; This is the single-phase equivalent voltage value; This represents the maximum virtual voltage value, which is the global maximum value. The minimum value of the virtual voltage is denoted as , and the global minimum value is denoted as .

[0230] The physical meaning of formula (33) is to eliminate the difference in dimensions and scale the voltage value to the range of [0,1].

[0231] Calculate the Euclidean distance between any two virtual user clusters as follows:

[0232] ;

[0233] in, For virtual user clusters and Between ; , Do not use virtual user clusters and The normalized virtual voltage. The calculated Euclidean distance. It serves as the direct input for subsequent calculations of profile coefficients and voltage similarity analysis.

[0234] In a high-dimensional space composed of normalized virtual voltage sequences, the Euclidean distance between the voltage curves of two virtual user clusters is calculated. The smaller the Euclidean distance, the more similar their fluctuation trends are.

[0235] The physical meaning of formula (34) is the geometric distance between two virtual user clusters in the space formed by the normalized voltage sequence throughout the time period, representing the difference in the overall shape of their voltage curves.

[0236] Voltage similarity based on maximizing profile coefficients is used to evaluate the maximum similarity between virtual user clusters;

[0237] The contour coefficient It is determined by both the density within virtual user clusters and the separation between virtual user clusters, as shown below:

[0238] ;

[0239] in, For the first Profile coefficients of a virtual user cluster; For the first The density of all real users within a virtual user cluster For virtual user clusters The user set in; For the first The degree of separation between a virtual user cluster and other virtual user clusters. For different Other virtual user clusters; For virtual user clusters and The Euclidean distance between them; For other virtual user clusters The total number of real users included.

[0240] Similarity score Based on cohesion index and separation index calculate.

[0241] Similarity scores incorporate intra-cluster compactness and inter-cluster separation To avoid the one-sidedness of a single indicator. Similarity scoring The positive / negative value directly reflects the correctness of clustering (e.g., negative values ​​warn of misclassification), and the denominator... Resistance to differences in the dimensions of distance.

[0242] The specific steps are as follows:

[0243] (1) Initial screening of candidate users within the same cluster.

[0244] (2) To test whether virtual users under the same distribution transformer are sufficiently similar (high cohesion).

[0245] (3) Verify whether the virtual user is independent (sufficiently distant from other virtual users not under the same distribution transformer).

[0246] (4) Final similarity score:

[0247] If a certain virtual user cluster It can be confirmed that the virtual user cluster belongs to this distribution variant;

[0248] like This virtual user cluster may have been mistakenly assigned to another distribution transformer area.

[0249] Among them, the cohesion index :

[0250] Virtual User Cluster The degree of similarity between the voltage fluctuations of virtual user cluster i and other users in the same cluster is used to quantify the average difference in voltage characteristics of virtual user cluster i under the same transformer.

[0251] (Minimum value) represents a virtual user cluster. The voltage curves of virtual user clusters under the same distribution transformer are highly similar, indicating a high degree of similarity.

[0252] Larger value → Virtual user cluster The voltage curves of the virtual user clusters under the same distribution transformer differ significantly from those of the virtual user clusters, indicating low density.

[0253] Virtual User Cluster The average difference in voltage fluctuation patterns among all virtual user clusters under the same transformer directly reflects the consistency strength of the voltage behavior between the virtual user and the virtual user clusters under the same transformer.

[0254] Separation index :

[0255] Virtual User Cluster The degree of voltage fluctuation difference between virtual user clusters and other virtual user clusters, i.e., virtual user clusters not under the same transformer.

[0256] like ,Pick This is to avoid the denominator of the contour coefficient being zero.

[0257] The final maximum overall similarity score is shown below:

[0258] ;

[0259] Where h is the number of virtual user clusters used to evaluate overall similarity. The range of its value is [−1, 1].

[0260] The closer the value is to 1, the better the clustering effect of the virtual user clusters. That is, the voltage of users within the same virtual user cluster is very similar, while the voltage curves of users in different virtual user clusters are significantly different.

[0261] Simply put, It is used to divide the virtual user clusters into the most reasonable ones.

[0262] This indicates a clustering method that maximizes the average profile coefficient of all virtual user clusters.

[0263] To ensure the joint optimization objective function has a clear optimization direction and that the denominator is not zero, we will... Take the reciprocal; at the same time, to avoid the objective function having multiple optimization directions, [the following is done]... Add 1, so that The value range is 0 to 2.

[0264] Establish a joint optimization model based on electrical characteristic indices. .

[0265] Step S33: Solve the joint optimization model.

[0266] Solving the joint optimization model includes:

[0267] Based on the objective function and constraints of the aforementioned loss coefficient model, and using the historical ledger correlation matrix as initial input, the CPLEX solver is employed to minimize the energy loss coefficient. To obtain the initial connection relationship between the distribution transformer and the virtual user cluster;

[0268] Starting from the initial connection relationship between the distribution transformer and the virtual user cluster, maximize the profile coefficient. Voltage similarity;

[0269] To obtain the objective function of the joint optimization model that is maximized ;

[0270] At this point, the connection relationship between the distribution transformer and the virtual user cluster is the optimal connection relationship, yielding the current real-time correlation matrix between the distribution transformer and the virtual user cluster. .

[0271] To avoid the optimization result converging to a suboptimal solution due to a large deviation in the starting point of random optimization, and because... The correspondence between distribution transformers and users was not considered, therefore, in the optimization... First, use A good initial solution (initial connection relationship) is obtained, and then the initial solution is used as the input of the profile coefficient for optimization to obtain the maximum profile coefficient. The output is a real-time correlation matrix that can characterize the connection relationship between the distribution transformer and the virtual user.

[0272] Simultaneously satisfy and The connection relationship between the distribution transformer and the virtual user cluster is the current optimal real-time correlation matrix.

[0273] Step S3 aims to dynamically identify the optimal connection relationship between the distribution transformer and the virtual user cluster by constructing and solving a joint optimization model that integrates energy conservation and voltage similarity, and to generate a real-time correlation matrix.

[0274] Step S4, specifically.

[0275] The real-time correlation matrix is ​​compared with the historical ledger correlation matrix to perform dynamic identification and anomaly assessment of line change relationships, resulting in the dynamic identification and anomaly assessment results of low-voltage line change relationships, including:

[0276] The real-time correlation matrix is ​​compared with the historical ledger correlation matrix to obtain the difference matrix between the two.

[0277] Calculate the Frobenius norm of the difference matrix to measure the overall correlation offset.

[0278] Obtain the line loss change rate and voltage mean offset, and combine them with the Frobenius norm of the difference matrix to construct a multi-index weighted evaluation, and calculate the multi-index weighted evaluation value.

[0279] The weighted evaluation value of the multi-indicator is compared with the corresponding dynamic threshold. If the weighted evaluation value of the multi-indicator is greater than the dynamic threshold, there is an abnormality in the line change, which is marked as an anomaly point to assist in the correction of the ledger or the connection check.

[0280] The dynamic recognition process mainly includes the following steps:

[0281] Step 1: Calculate the difference matrix between the real-time correlation matrix and the historical ledger correlation matrix. As shown below:

[0282] ;

[0283] in, For real-time correlation matrix, This is a historical ledger association matrix.

[0284] The second step is to calculate the Frobenius norm of the overall correlation offset based on the difference matrix, as shown below:

[0285] ;

[0286] in, It is the Frobenius norm; Difference matrix Each element; They are difference matrices Middle elements Row indexes and column indexes.

[0287] In this invention, the Frobenius norm is used to calculate the overall offset strength between the real-time correlation matrix and the historical ledger correlation matrix. If there is a user wiring abnormality (such as incorrect connection, missing connection, ledger error, etc.) under a certain transformer, the difference at the corresponding position will change from 0 to 1, thereby causing the Frobenius norm to increase significantly.

[0288] Step 3: Multi-indicator weighted evaluation:

[0289] Combining the Frobenius norm with other electrical metrics (such as the rate of change of line loss) Voltage mean offset Construct a comprehensive evaluation function (etc.) :

[0290] ;

[0291] in, These are the weights for the Frobenius norm, the rate of change of line loss, and the voltage mean offset, respectively; for example, the weighting coefficients are initially set to... , , To balance , , right . contributions.

[0292] The physical meaning of the comprehensive evaluation function is to comprehensively reflect the stability of the low-voltage distribution area by quantifying the abnormal changes in three key dimensions of power grid operation.

[0293] Among them, the rate of change of line loss at time t This indicates the degree of deviation of real-time line loss from historical baseline line loss. A larger value indicates a more severe energy imbalance in the current system and a higher probability of wiring abnormalities; as shown below:

[0294] ;

[0295] in, This is for real-time line loss; the total loss in the current low-voltage distribution area is calculated based on real-time collected data. =Total power supply on the distribution transformer side - Total power consumption of all users, directly reflecting the actual energy loss of the current low-voltage distribution network; For historical line loss records, and when the wiring relationships of low-voltage distribution areas are known to be correct, the theoretical or baseline line loss value serves as a stable reference benchmark. For a very small positive number, when When it is very small or zero, add To ensure the denominator is non-zero, the mathematical stability of the formula is guaranteed.

[0296] Voltage mean offset , where is the voltage mean offset at time t, representing the average level of the absolute values ​​of the real-time virtual voltage of all virtual user clusters compared to the historical reference voltage; the larger this value, the more significant the overall voltage level on the user side has changed, potentially indicating anomalies; as shown below:

[0297] ;

[0298] Where h is the number of virtual user clusters, This represents the real-time virtual voltage of virtual user cluster k at time t; The historical record voltage for virtual user cluster k is the voltage reference value for the same virtual user cluster during the correct historical record period. This is usually data collected and stored by the low-voltage distribution area during its initial normal operation.

[0299] Adjustments can be made based on actual data and project requirements, using specific methods such as data fitting or optimization analysis.

[0300] Define the current time as Select the previous time window Comprehensive evaluation indicators within As a sample set; time window length It can be set to 24 hours (i.e., 96 15-minute samples) or 48 hours, depending on the actual operation.

[0301] For example, if the current t is 14:00, It is a 24-hour period, with the time window being from 14:00 yesterday to 14:00 today.

[0302] Let the evaluation sequence within this window be:

[0303] ;

[0304] For time window The overall evaluation value at all times. The data set arranged in chronological order is the basis for calculating dynamic thresholds.

[0305] Step 4: Calculate the statistics, including the mean and standard deviation of the evaluation series within the window.

[0306] The mean (expected value) is calculated as follows:

[0307] ;

[0308] Standard deviation (range of fluctuation), as shown below:

[0309] ;

[0310] Step 5: Dynamic Threshold The definition is as follows:

[0311] ;

[0312] in, This is the sensitivity coefficient, which, for example, takes a value of 1.5-2.5.

[0313] Smaller values, such as 1.5, result in a lower dynamic threshold, making detection more sensitive and easier to report anomalies, but may also increase false alarms.

[0314] Larger values, such as 2.5, result in a higher dynamic threshold, leading to more conservative detection, less likelihood of reporting anomalies, and an increased risk of missed detections, but also more reliable results.

[0315] Step 6: Perform dynamic identification and anomaly assessment based on dynamic thresholds.

[0316] like If an anomaly is detected at that moment, the anomaly point is marked to assist in the correction of the ledger or the verification of the wiring.

[0317] like Figure 2 As shown, the joint optimization model exhibits higher recognition accuracy and robustness under various typical operating conditions compared to traditional methods based on a single indicator (such as electricity consumption or voltage). Specifically, by introducing a loss coefficient to reflect energy deviation, using a profile coefficient to quantify voltage aggregation effect, and using the product of the two as the optimization objective function, the physical rationality of the identification results can be improved while ensuring model convergence.

[0318] like Figure 3 As shown, by introducing a difference matrix and Frobenius norm and setting a static threshold, some typical wiring anomalies can be identified. However, when facing operating environments with frequent load fluctuations or significant seasonal changes, the static threshold may pose a risk of misjudgment or missed judgment, and the sensitivity is difficult to set uniformly. To improve adaptability and identification stability, this invention further introduces a dynamic multi-index weighted evaluation mechanism.

[0319] like Figure 4 As shown, a comprehensive evaluation function is constructed by combining the Frobenius norm, the rate of change of line loss, and the voltage mean offset. A dynamic threshold is adaptively set within the sliding time window to achieve dynamic discrimination and anomaly detection of the line loss relationship.

[0320] and Figure 3 Compared to the static threshold method shown, Figure 4 The method shown exhibits better anomaly identification and fault tolerance under dynamic operating conditions, effectively reducing false alarms caused by data disturbances, while improving the response sensitivity to sudden wiring changes.

[0321] The purpose of step S4 is to calculate dynamic thresholds and make anomaly judgments on the weighted evaluation values ​​of multiple indicators, thereby realizing the dynamic identification and anomaly assessment of the relationship between distribution transformers and users, and improving the accuracy of identifying the relationship between distribution transformers and users.

[0322] In summary, the low-voltage line transformer dynamic identification method based on electrical characteristics and real-time matrix of this invention has the following beneficial effects:

[0323] 1. This invention integrates multi-dimensional electrical features (three-phase voltage, current, active power, and electricity consumption) and constructs a joint optimization model. It combines the loss coefficient reflecting energy conservation with the contour coefficient reflecting curve similarity, making a comprehensive judgment from multiple physical dimensions. This significantly improves the accuracy of linear relationship identification and its robustness under different operating conditions. It solves the problems of low identification accuracy and poor adaptability caused by the single data dimension (such as only electricity consumption or single-phase voltage) and coarse similarity discrimination (such as simply using Euclidean distance) of traditional methods.

[0324] 2. This invention introduces the principle of energy conservation as the core physical constraint and constructs a differential comparison mechanism between the real-time correlation matrix and historical ledgers. By utilizing the Frobenius norm and the dynamic threshold of the sliding time window, it not only ensures that the identification results conform to the basic physical laws of the power grid, but also realizes real-time and automatic detection and tracking of abnormal changes in line-transformer relationships, enhancing the physical rationality and dynamic tracking capability of the results. It overcomes the technical problems of traditional methods lacking physical constraint mechanisms and static models being unable to reflect dynamic changes in wiring relationships in a timely manner.

[0325] 3. This invention establishes a quantifiable and calculable scientific evaluation system by introducing a multi-index weighted evaluation function (combining Frobenius norm, line loss change rate, and voltage mean offset) and dynamic threshold determination. This makes the identification results no longer dependent on a simple formal comparison with historical records, enabling automatic and accurate anomaly location, providing reliable technical support for large-scale automated operation and maintenance of power grid companies; it achieves quantifiable scientific evaluation and automated anomaly identification. It overcomes the shortcomings of existing methods in terms of evaluation mechanism deficiencies, making it difficult to support large-scale deployment and automated anomaly identification.

[0326] Those skilled in the art will understand that all or part of the processes of the methods described in the above embodiments can be implemented by a computer program instructing related hardware, and the program can be stored in a computer-readable storage medium. The computer-readable storage medium may be a disk, optical disk, read-only memory, or random access memory, etc.

[0327] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A low-voltage line dynamic identification method based on electrical characteristics and real-time association matrix, characterized in that, The application relates to a low-voltage power distribution network real-time connection relationship identification method and device. Real-time acquisition of electrical characteristic data of a distribution transformer side and a user side in a low-voltage power distribution network; Dimension reduction is performed on three-phase voltage data in the electrical characteristic data of the user side to obtain a single-phase equivalent voltage sequence; the single-phase equivalent voltage sequence is calculated to obtain a plurality of virtual user clusters; virtual active power and virtual daily power consumption time sequence data of each virtual user cluster are calculated; each virtual user cluster comprises a plurality of real users; Based on the virtual active power and virtual daily power consumption of each virtual user cluster and the electrical characteristic data of the distribution transformer side, a joint optimization model is solved to obtain a real-time correlation matrix representing the optimal connection relationship between the distribution transformer and each virtual user cluster; The joint optimization model is constructed, comprising: Based on the active power and daily electricity energy conservation principle between the distribution transformer and the virtual user cluster, a loss coefficient model is constructed to minimize the error of the total active power and daily electricity of the distribution transformer and the virtual user cluster ; normalizing the virtual voltage of each virtual user cluster to obtain a normalized virtual voltage; calculating a Euclidean distance between any two virtual user clusters based on the normalized virtual voltage; and calculating a silhouette coefficient of the virtual user cluster based on the Euclidean distance ; Based on , , the objective function of the joint optimization model is constructed as follows: ； The objective function and the constraint condition of the loss coefficient model are as follows: The objective function is as follows: ； Wherein, is the index of the distribution transformer, is the set of all distribution transformers; is the index of the sampling time, is the total number of sampling time points; is the daily electricity consumption of the th distribution transformer at the sampling time ; is the total number of virtual user clusters; is the virtual daily electricity consumption of the th virtual user cluster at the sampling time ; is the connection relationship between the th distribution transformer and the th virtual user cluster; is the active power value of the th distribution transformer at the sampling time ; is the virtual active power of the th virtual user cluster at the sampling time ; is the set of virtual user clusters associated with the th distribution transformer; The constraint condition is as follows: ； wherein, is the total number of allocations; indicates that each virtual user cluster can be connected to only one allocation; indicates that each allocation is connected to at least one virtual user cluster; The real-time correlation matrix is compared with a historical account correlation matrix to perform dynamic identification and abnormal evaluation of the line transformer relationship, and a dynamic identification and abnormal evaluation result of the low-voltage line transformer relationship is obtained.

2. The low voltage line variation dynamic identification method based on electrical characteristics and real-time correlation matrix according to claim 1, characterized in that, The electrical characteristic data of the user side comprises three-phase voltage, three-phase current and daily power consumption time sequence data of real users; The electrical characteristic data of the distribution transformer side comprises active power and daily power consumption time sequence data of the distribution transformer side; The electrical characteristic data of the user side and the electrical characteristic data of the distribution transformer side are aligned in time sequence by using the same sampling frequency.

3. The low voltage line variation dynamic identification method based on electrical characteristics and real-time correlation matrix according to claim 2, characterized in that, The dimension reduction of the three-phase voltage data in the electrical characteristic data of the user side comprises the following steps: For each sampling time of the three-phase voltage measurement value, a three-phase voltage equal phase angle equivalent formula is solved, a numerical iteration method is used to solve the three-phase voltage equal phase angle equivalent formula, and a single-phase equivalent voltage value at the first sampling time is obtained; The single-phase equivalent voltage obtained at each sampling time is combined into a single-phase equivalent voltage sequence in sequence according to the sampling time; The three-phase voltage equivalent formula is as follows: ； wherein, are the phase voltages of phase A, B, C at the i-th sampling time, respectively, are the phase voltages of phase A, B, C at the i-th sampling time, respectively, , is the number of sampling times, is the single-phase equivalent voltage after dimension reduction at the corresponding time.

4. The low voltage line variation dynamic identification method based on electrical characteristics and real-time correlation matrix according to claim 3, characterized in that, Based on the similarity of voltage fluctuation under the same distribution transformer, the single-phase equivalent voltage sequence is calculated to obtain a plurality of virtual user clusters, comprising: Based on the single-phase equivalent voltage sequence of the user, the voltage fluctuation similarity between two users is calculated; All users are traversed, and users with a voltage fluctuation similarity higher than a preset similarity threshold are merged into the same virtual user cluster to obtain a plurality of virtual user clusters.

5. The low voltage line variation dynamic identification method based on electrical characteristics and real-time correlation matrix according to claim 4, characterized in that, The calculation of the virtual active power and the virtual daily power consumption time sequence data of each virtual user cluster comprises the following steps: Based on the single-phase equivalent voltage sequence of all users in the virtual user cluster, the mean value corresponding to each sampling time is calculated to obtain the virtual voltage of the virtual user cluster at each sampling time; Based on the three-phase current of each sampling time of the user side, the corresponding single-phase equivalent current value is calculated by normalization; based on the single-phase equivalent current value of all users in the virtual user cluster, the mean value is calculated according to the sampling time to obtain the virtual current of the virtual user cluster corresponding to the sampling time; The virtual voltage and the virtual current of each sampling time are calculated to obtain the virtual active power of the virtual user cluster at the corresponding time; the virtual active power of each virtual user cluster at all sampling times is combined into the virtual active power time sequence data of the virtual user cluster; The virtual daily power consumption of the virtual user cluster at the corresponding time is obtained by summing the daily power consumption of all users in the virtual user cluster; the virtual daily power consumption of each virtual user cluster at all times is combined into the virtual daily power consumption sequence of the virtual user cluster.

6. The low voltage line variation dynamic identification method based on electrical characteristics and real-time correlation matrix according to claim 1, characterized in that, The joint optimization model is solved, comprising: Based on the loss coefficient model of the objective function and constraint conditions, with the historical account associated matrix as the initial input, using CPLEX solver to solve, to minimize the energy loss coefficient , obtain the initial connection relationship of the distribution transformer and the virtual user cluster; maximizing the profile coefficient with the initial connection relationship of the distribution and the virtual user cluster as the starting point voltage similarity; Objective function of the joint optimization model is maximized ; The connection relationship between the corresponding distribution variable and the virtual user cluster at this time is the optimal connection relationship, and a current real-time association matrix of the distribution variable and the virtual user cluster is obtained .

7. The method of claim 1, wherein the method further comprises: The real-time correlation matrix is compared with the historical account correlation matrix to perform dynamic identification and abnormal evaluation of line and transformer relationship, and obtain dynamic identification and abnormal evaluation results of low-voltage line and transformer relationship, including: The real-time correlation matrix is compared with the historical account correlation matrix to obtain a difference matrix of the two; The Frobenius norm of the difference matrix is calculated to measure the overall correlation offset degree; The line loss change rate and voltage mean offset are obtained, and combined with the Frobenius norm of the difference matrix, a multi-index weighted evaluation is constructed to calculate a multi-index weighted evaluation value; The multi-index weighted evaluation value is compared with the corresponding dynamic threshold value; if the multi-index weighted evaluation value is greater than the dynamic threshold value, there is a line and transformer abnormality, which is marked as an abnormal point to assist in account correction or wiring inspection.

8. The low voltage line variation dynamic identification method based on electrical characteristics and real-time correlation matrix according to claim 6, characterized in that, The voltage similarity based on the maximum profile coefficient is used to evaluate the maximum similarity between virtual user clusters. The profile coefficient The closeness within a virtual user cluster and the separation between virtual user clusters are jointly determined as follows: ； wherein, is a profile coefficient of the th virtual user cluster; is a closeness of all real users in the th virtual user cluster, is a set of users in the virtual user cluster ; is a separation of the th virtual user cluster from other virtual user clusters, is other virtual user clusters different from ; is a Euclidean distance between the virtual user cluster and ; is a total number of real users contained in other virtual user clusters .

Images