A Method and System for Establishing Line Loss Model Based on K-Means Clustering and Optimized RVM

By using K-Means clustering and optimizing the RVM line loss model, and utilizing power supply indicators to calculate line loss evaluation indicators, the problem of low efficiency in calculating line loss in low-voltage distribution transformer areas is solved. This enables efficient analysis and prediction of line loss anomalies, thereby improving power management and economic benefits.

CN115081551BActive Publication Date: 2026-11-17STATE GRID JIANGSU ELECTRIC POWER CO LTD MARKETING SERVICE CENT
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202210893168.4
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-07-27
Publication Date
2026-11-17
Estimated Expiration
2042-07-27

AI Technical Summary

Technical Problem

Existing technologies are inefficient and inaccurate in calculating line losses in low-voltage distribution transformer areas, leading to delays and errors in line loss management. This is especially true in situations with complex network structures, a large number of users, diverse user characteristics, and massive amounts of data, making it difficult to effectively analyze and predict line loss anomalies.

Method used

We employ a line loss model based on K-Means clustering and an optimized RVM algorithm. We process power supply indicators through a noise-reducing autoencoder, calculate line loss evaluation indicators using these indicators, and construct a line loss model by combining the optimized RVM algorithm. This enables line loss anomaly analysis and prediction without the need for theoretical line loss calculation.

Benefits of technology

It improves the efficiency of line loss management, reduces the lag and errors caused by theoretical line loss calculation, enhances the universality and modularity of line loss models, and can effectively analyze and predict line loss anomalies, supporting the management level and economic benefits of the power sector.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115081551B_ABST
    Figure CN115081551B_ABST
Patent Text Reader

Abstract

This invention relates to a method for establishing a line loss model based on K-Means clustering and an optimized RVM algorithm. The method includes: collecting power supply indicators and historical line loss data for distribution transformer areas; constructing a power supply indicator processing model using a noise-reducing autoencoder to obtain power supply indicators related to line loss; calculating line loss evaluation indicators for distribution transformer areas with different grid structures using these power supply indicators, dividing the distribution transformer areas into K groups, and using the center of each group as the initial cluster center for the K-Means clustering algorithm; collecting historical line loss causes for distribution transformer areas and performing pre-classification; constructing a line loss model based on the optimized RVM algorithm; and inputting the clustering results into the line loss model for training to obtain classification results of historical line loss causes for distribution transformer areas with different grid structures; and collecting power supply indicators in real time and using the line loss model to detect line loss anomalies and their causes. This invention does not rely on theoretical line loss calculations and enables the analysis and prediction of line loss anomalies based on the power supply indicators of distribution transformer areas.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention belongs to the field of transformer area line loss detection technology, specifically involving a method and system for establishing a line loss model based on K-Means clustering and optimized RVM. Background Technology

[0002] Line loss refers to the energy loss and damage that occurs at various stages of the power transmission process from the power plant to the customer, including transmission, transformation, distribution, and marketing. In the power industry, the line loss rate is a crucial technical and economic indicator for power companies, and a key economic and technical indicator comprehensively reflecting the planning, design, operation, and management level of the power grid. Although my country's power utilization efficiency has gradually improved over the past 20 years, it still lags behind international advanced levels. Therefore, reducing line loss is a comprehensive and systematic project for power supply companies' own operation and management.

[0003] Current research on abnormal line loss mainly focuses on two aspects: theoretical line loss and management line loss. Furthermore, various studies on management line loss are often based on the calculation results of theoretical line loss.

[0004] In existing technologies, "a fast line loss calculation method based on K-Means clustering algorithm and multi-classification correlation vector machine" selects electrical indicators related to distribution network line loss as the model input. For cases where indicators are similar but network structure and load characteristics affect the line loss rate, this paper uses K-Means clustering algorithm to perform cluster analysis on transformer substations, simplifying data processing and conducting specific analysis for each substation category. Then, a model optimized by the fruit fly algorithm (MRVM) is built to calculate the line loss of the substations. This existing technology, for the study of theoretical line loss, uses a line loss prediction method based on a combination of K-Means clustering algorithm, particle swarm optimization algorithm, and support vector machine, along with deep learning LSTM (Long Short-Term Memory) network, to predict line loss, provide early warning for lines or substations with abnormal line loss, and can also repair abnormal line loss when it occurs.

[0005] The existing technology (CN111123039B), "A Method for Diagnosing Abnormal Line Losses in Distribution Networks Based on Synchronous Characteristics and Improved K-means Clustering," analyzes the characteristics of historical synchronous line losses. Based on the research results of theoretical line losses, it constructs real-time line loss rate, average line loss rate over the past 24 hours, and line loss distortion rate as key indicators of abnormal line losses, and correlates them with the causes of anomalies to obtain a preliminary diagnostic model. Furthermore, based on these indicators and anomaly types, an improved K-Means clustering method is established, solving the problem of difficulty in determining the number of clusters. At the same time, data density and Euclidean distance are combined, along with the average similarity evaluation index between clusters, making the clusters more independent and the data between clusters more cohesive, thus improving the clustering effect. Finally, the cluster centers and cluster performance are mapped to the causes of anomalies. By finding the cluster center with the smallest distance, the corresponding cause of anomaly can be found, making the diagnosis simpler and faster.

[0006] Existing technologies analyze and predict line loss anomalies based on the difference between the theoretically expected line loss of electrical components and the actual line loss. For low-voltage distribution transformer areas, in situations with complex network structures, a large number of users, complex characteristics, and massive amounts of data, there will be problems with low calculation efficiency and low accuracy of theoretical line loss, which directly leads to lag and errors in line loss management and causes unnecessary losses. Summary of the Invention

[0007] To address the shortcomings of existing technologies, the present invention aims to provide a method and system for establishing a line loss model based on K-Means clustering and optimized RVM, which does not rely on the calculation results of theoretical line loss and enables the analysis and prediction of line loss anomalies based on the power supply indicators of distribution transformer areas.

[0008] The present invention adopts the following technical solution.

[0009] This invention proposes a method for establishing a line loss model based on K-Means clustering and optimized RVM, including:

[0010] Step 1: Collect power supply indicators and historical line loss data for the distribution transformer area; use a power supply indicator processing model built with a noise-reducing autoencoder to obtain power supply indicators related to line loss;

[0011] Step 2: Calculate the line loss evaluation index for different grid structure transformer substations using power supply indicators related to line loss; divide the different grid structure transformer substations into K groups using the line loss evaluation index, and use the center of each group as the initial cluster center of the K-Means clustering algorithm.

[0012] Step 3: Collect historical causes of line losses in the distribution transformer area and perform pre-classification;

[0013] Step 4: Using the pre-classification results of historical line loss causes, construct a line loss model based on the optimized RVM algorithm, and input the clustering results of Step 2 into the line loss model for training, so as to obtain the classification results of historical line loss causes for distribution transformer areas with different grid structures;

[0014] Step 5: Real-time collection of power supply indicators related to line loss, after processing in steps 2 and 3, input into the line loss model trained in step 4 for detection of line loss anomalies and causes.

[0015] Preferably, in step 1, the power supply indicators related to line loss include: power supply radius, total length of low-voltage lines, power consumption type and proportion, and load factor;

[0016] The nature and proportion of electricity use include: the electrical characteristics of the electricity load, the importance of the electricity load, the time, occasion, purpose, and permitted power outage time of the electricity load.

[0017] Preferably, in step 1, when training the power supply index processing model, power supply indices unrelated to line loss are added to the training set as noise, and an unsupervised method is used to train the weights to obtain a trained power supply index processing model; wherein, the training set includes: power supply indices related to line loss and historical line loss data.

[0018] Preferably, in step 2, the line loss evaluation index for distribution transformer areas with different grid structures is calculated using the power supply indexes related to line loss, according to the following formula:

[0019]

[0020] In the formula,

[0021] P Ei Let be the line loss evaluation index for the distribution transformer area of ​​the i-th type of grid structure.

[0022] Z ij Let j be the power supply index related to line loss in the distribution transformer area with the i-th grid structure.

[0023] ω j Let be the weight corresponding to the j-th power supply index related to line loss in the distribution transformer area of ​​the i-th grid structure.

[0024] j = 1, 2, ..., m, where m is the number of power supply indicators related to line loss;

[0025]

[0026] In the formula, Z jmin Let be the minimum value of the j-th power supply index related to line loss in various distribution transformer areas with different grid structures, i = 1, 2, ..., N.

[0027] Furthermore, using P E The distribution transformer areas with different grid structures are divided into K groups. The distribution transformer area corresponding to the shortest distance to the average value of the line loss evaluation index of each distribution transformer area within the group is taken as the center of the group, which is the initial cluster center of the K-Means clustering algorithm.

[0028] Preferably, in step 3, the causes of historical line loss include: incorrect household-transformer relationship, file error, metering error, metering failure, electricity use without meter, electricity theft or over-capacity electricity use, data acquisition failure, and abnormal line loss.

[0029] The causes of historical line loss are pre-classified according to region, user type, average capacity per user, and load rate.

[0030] Preferably, step 4 includes:

[0031] Step 4.1: Construct the kernel function for optimizing the RVM algorithm, satisfying the following relationship:

[0032]

[0033] In the formula,

[0034] K(μ,μ u To optimize the kernel function of the RVM algorithm,

[0035] μ is the input value of the kernel function.

[0036] μ u The output value of the kernel function.

[0037] γ is the width factor;

[0038] Step 4.2: Use a multi-population genetic algorithm to find the optimal solution γ for the width factor. * ;

[0039] Step 4.3, optimize the solution γ using the width factor. * Based on the kernel function K(μ, μ) of the optimized RVM algorithm u ), thus obtaining the line loss detection model;

[0040] Step 4.4: Input the clustering results from Step 2 into the line loss detection model, and the line loss detection model will output the historical line loss causes corresponding to the distribution transformer areas with different grid structures.

[0041] Furthermore, step 4.2 includes:

[0042] Step 4.2.1: Initialize the population size G, maximum number of iterations D, the value of the historical largest individual L, and the initial x-axis coordinate X of the population. axis and the initial coordinates of the y-axis Y axisAmong them, the value L of the largest individual in history is negative infinity, and the other parameters are all random values;

[0043] Step 4.2.2: Assign a random distance R to each individual in the population. t ;

[0044] Step 4.2.3: Utilize the distances between each individual's x-axis and y-axis coordinates and the random distance R. i The fitness function value for each individual can be expressed by the following formula:

[0045]

[0046] In the formula,

[0047] S t This represents the fitness function value for the t-th individual.

[0048] R t This represents the random distance of the t-th individual.

[0049] x axis_t Let x be the current x-axis coordinate of the t-th individual. In the first iteration, x... axis_t =X axis ,

[0050] y axis_t Let y be the current y-axis coordinate of the t-th individual. In the first iteration, y axis_t =Y axis ;

[0051] Step 4.2.4: Input the fitness function value of each individual into the fitness function, and use the fitness function value as the value of the largest individual. The fitness function satisfies the following relationship:

[0052] f=(1-β)S t +β(x axis_t +R t )+β 2 (y axis_t -R t )

[0053] In the formula,

[0054] f is the fitness function.

[0055] β is a random value in [0, 1];

[0056] The x-axis and y-axis coordinates of the largest individual satisfy the following relationships:

[0057] x axis_t =x axis+R f

[0058] y axis_t =y axis +R t

[0059] Step 4.2.5, set the value of the largest individual. Compared with the historical maximum value L, if Then let Otherwise proceed to step 5.2.6;

[0060] Step 4.2.6: Determine if the current iteration count has reached the maximum iteration count D. If it has, output the historical maximum individual value L as the width factor optimization solution γ. * If the number of iterations has not reached the maximum number of iterations D, then return to step 4.2.2.

[0061] Another aspect of the present invention proposes a system for establishing a line loss model based on K-Means clustering and optimized RVM, including a memory, a processor, and a wireless communication module; the memory stores a computer program, and the processor calls the computer program to execute each step of the method for establishing a line loss model based on K-Means clustering and optimized RVM.

[0062] The beneficial effects of this invention are compared with those of the prior art:

[0063] 1) The line loss model established by the method proposed in this invention does not require the calculation of theoretical line loss and dependent line loss indicators. Through unsupervised learning of the noise reduction autoencoder, it can automatically remove power supply indicators that are not related to line loss, and can also automatically add and extract power supply indicators related to line loss. Power supply indicators are objective indicators of distribution transformer areas and have stable and reliable characteristics. Based on the power supply indicators, the analysis and prediction of line loss anomalies can be realized, which can effectively avoid the lag or even error in line loss management caused by changes in theoretical line loss or operating data, and significantly improve the efficiency of line loss management.

[0064] 2) This invention also considers the impact of different grid structures of distribution transformer areas on line loss. It uses power supply indicators to calculate the line loss evaluation index of distribution transformer areas to reflect the line loss of each area. The line loss evaluation index has the same trend as the actual line loss of the area. Based on the line loss evaluation index, it is possible to group distribution transformer areas and obtain the center of each group. Using the group center of the distribution transformer area as the initial cluster center of the K-Means clustering algorithm, it effectively avoids the influence of external factors such as grid structure on the clustering results, and improves the versatility and modularity of the line loss model proposed in this invention.

[0065] 3) This invention pre-classifies the causes of historical line loss in accordance with the management requirements of the power sector, so as to improve the efficiency of classification and ensure the relevance of the classification results. At the same time, the use of the optimized RVM algorithm can avoid the defects of human experience in selecting parameters and failing to achieve the ideal classification effect, further improving the versatility and modularity of the line loss model proposed in this invention.

[0066] 4) The line loss model system proposed in this invention can not only analyze the line loss anomalies of existing distribution transformer areas, but also predict the line loss of newly built or renovated distribution transformer areas. This is conducive to improving the management level and economic benefits of the power sector, promoting the scientific and rational construction and renovation of the power grid, and effectively supporting the improvement of the quality and efficiency of related businesses in the power industry. Attached Figure Description

[0067] Figure 1 This is a flowchart illustrating the steps of the method for establishing a line loss model based on K-Means clustering and optimization of RVM in this invention;

[0068] Figure 2 This is a case library of abnormal line loss causes in distribution transformer areas obtained in the embodiments of the present invention. Detailed Implementation

[0069] The present application will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention, and should not be construed as limiting the scope of protection of the present application.

[0070] Figure 1 The steps of the method for optimizing the RVM line loss model based on K-Means clustering proposed in this invention include steps 1 to 5.

[0071] Step 1: Collect power supply indicators and historical line loss data for the distribution transformer area; use a power supply indicator processing model built with a noise-reducing autoencoder to obtain power supply indicators related to line loss.

[0072] Preferably, in step 1, the power supply indicators related to line loss include: power supply radius, total length of low-voltage lines, power consumption type and proportion, and load factor;

[0073] The nature and proportion of electricity use include: the electrical characteristics of the electricity load, the importance of the electricity load, the time, occasion, purpose, and permitted power outage time of the electricity load.

[0074] Preferably, in step 1, when training the power supply index processing model, power supply indices unrelated to line loss are added to the training set as noise, and an unsupervised method is used to train the weights to obtain a trained power supply index processing model; wherein, the training set includes: power supply indices related to line loss and historical line loss data.

[0075] In this embodiment, a DAE noise reduction autoencoder is used, which consists of three layers of neurons: an input layer, a hidden layer, and an output layer. The number of neurons in the input layer and the output layer are the same, and the structure can be divided into two parts: an encoder and a decoder.

[0076] The original training set x, containing power supply indicators related to line loss and historical line loss data, is used as the input information of the encoder. The input information x is encoded into hidden layer information h after passing through the input layer. The hidden layer information h is then decoded and remapped into output information y. The hidden layer and the output layer satisfy the following relationship:

[0077] h = n(W1x + b1)

[0078] y = n(W2h + b2)

[0079] In the formula, W1 and b1 are the weight parameter and bias parameter matrices of the encoder, respectively; W2 and b2 are the weight parameter and bias parameter matrices of the decoder, respectively; η is the activation function between neurons, which is the SELU activation function in this embodiment.

[0080] To enable the encoder to perform dimensionality reduction, the number of neurons in the hidden layer is set to be less than the number of neurons in the input and output layers.

[0081] An autoencoder iteratively seeks to minimize the error y between the input information x and the output information. It solves for and updates the encoder and decoder parameters by minimizing the loss function, which can be characterized as:

[0082] J(W2,b2)=∑||y,x|| 2

[0083] To improve the robustness of the encoder, this invention treats power supply indicators unrelated to line loss as noise. The original training set x, which contains power supply indicators related to line loss and historical line loss data, is artificially noisy to obtain a noisy training set z. The noisy hidden layer information h' and the output layer information y' satisfy the following relationship:

[0084] h′=n(W l z+b1)

[0085] y′=n(W2h′+b2)

[0086] Step 2: Calculate the line loss evaluation index for different grid structure transformer substations using power supply indicators related to line loss; divide the different grid structure transformer substations into K groups using the line loss evaluation index, and use the center of each group as the initial cluster center of the K-Means clustering algorithm.

[0087] The K-Means clustering algorithm has the following main drawbacks.

[0088] 1) The K value needs to be given in advance; however, the K value is difficult to estimate and is very sensitive to noise and outliers.

[0089] 2) The K-Means algorithm is very sensitive to the initial centroid; different centroids will result in different clustering results.

[0090] 3) The algorithm may get stuck in local optima.

[0091] To address the aforementioned shortcomings, the following formula is used to calculate the line loss evaluation index for distribution transformer areas with different grid structures, based on power supply indicators related to line loss:

[0092]

[0093] In the formula,

[0094] P Ei Let be the line loss evaluation index for the distribution transformer area of ​​the i-th type of grid structure.

[0095] Z ij Let j be the power supply index related to line loss in the distribution transformer area with the i-th grid structure.

[0096] ω j Let be the weight corresponding to the j-th power supply index related to line loss in the distribution transformer area of ​​the i-th grid structure.

[0097] j = 1, 2, ..., m, where m is the number of power supply indicators related to line loss;

[0098]

[0099] In the formula, Z jmin Let be the minimum value of the j-th power supply index related to line loss in various distribution transformer areas with different grid structures, i = 1, 2, ..., N.

[0100] Furthermore, using P E The distribution transformer areas with different grid structures are divided into K groups. The distribution transformer area corresponding to the shortest distance to the average value of the line loss evaluation index of each distribution transformer area within the group is taken as the center of the group, which is the initial cluster center of the K-Means clustering algorithm.

[0101] And through the silhouette coefficient S of the clustering results t The size of constraint K.

[0102] Step 3: Collect historical causes of line loss in the distribution transformer area and perform pre-classification.

[0103] In step 3, the causes of historical line loss include: incorrect household-transformer relationship, file error, metering error, metering failure, electricity use without meter, electricity theft or over-capacity electricity use, data acquisition failure, and abnormal line loss.

[0104] The causes of historical line loss are pre-classified according to region, user type, average capacity per user, and load rate.

[0105] In this embodiment, the causes of historical line loss are classified according to four aspects: region, user type, average capacity per user, and load rate.

[0106] Specifically, according to geographical classification, it refers to the thirteen prefecture-level cities in Jiangsu Province, including: Nanjing, Suzhou, Wuxi, Changzhou, Zhenjiang, Yangzhou, Taizhou, Nantong, Huai'an, Lianyungang, Yancheng, Xuzhou, and Suqian.

[0107] According to user nature, the classification includes low-voltage users, low-voltage non-residential users, and high-voltage users, etc.

[0108] Classifying by average capacity per household refers to classifying the capacity according to the nature of the user.

[0109] Classification by load factor refers to classifying transformers according to the actual load they bear and their capacity.

[0110] Step 4: Using the pre-classification results of historical line loss causes, construct a line loss model based on the optimized RVM algorithm, and input the clustering results from Step 2 into the line loss model for training, so as to obtain the classification results of historical line loss causes for distribution transformer areas with different grid structures.

[0111] Preferably, step 4 includes:

[0112] Step 4.1: Construct the kernel function for optimizing the RVM algorithm, satisfying the following relationship:

[0113]

[0114] In the formula,

[0115] K(μ,μ u To optimize the kernel function of the RVM algorithm,

[0116] μ is the input value of the kernel function.

[0117] μ u The output value of the kernel function.

[0118] γ is the width factor;

[0119] Step 4.2: Use a multi-population genetic algorithm to find the optimal solution γ for the width factor. * .

[0120] Furthermore, step 4.2 includes:

[0121] Step 4.2.1: Initialize the population size G, maximum number of iterations D, the value of the historical largest individual L, and the initial x-axis coordinate X of the population. axis and the initial coordinates of the y-axis Y axis Among them, the value L of the largest individual in history is negative infinity, and the other parameters are all random values;

[0122] Step 4.2.2: Assign a random distance R to each individual in the population. t ;

[0123] Step 4.2.3: Utilize the distances between each individual's x-axis and y-axis coordinates and the random distance R. i The fitness function value for each individual can be expressed by the following formula:

[0124]

[0125] In the formula,

[0126] S t This represents the fitness function value for the t-th individual.

[0127] R t This represents the random distance of the t-th individual.

[0128] x axis_t Let x be the current x-axis coordinate of the t-th individual. In the first iteration, x... axis_t =X axis ,

[0129] y axis_t Let y be the current y-axis coordinate of the t-th individual. In the first iteration, y axis_t =Y axis ;

[0130] Step 4.2.4: Input the fitness function value of each individual into the fitness function, and use the fitness function value as the value of the largest individual. The fitness function satisfies the following relationship:

[0131] f=(1-β)S t +β(x axis_t +R t )+β 2 (y axis_t -R t )

[0132] In the formula,

[0133] f is the fitness function.

[0134] β is a random value in [0, 1];

[0135] The x-axis and y-axis coordinates of the largest individual satisfy the following relationships:

[0136] x axis_t =x axis +R f

[0137] y axis_t =y axis +R t

[0138] Step 4.2.5, set the value of the largest individual. Compared with the historical maximum value L, if Then let Otherwise proceed to step 5.2.6;

[0139] Step 4.2.6: Determine if the current iteration count has reached the maximum iteration count D. If it has, output the historical maximum individual value L as the width factor optimization solution γ. * If the number of iterations has not reached the maximum number of iterations D, then return to step 4.2.2.

[0140] Step 4.3, optimize the solution γ using the width factor. * Based on the kernel function K(μ, μ) of the optimized RVM algorithm u ), thus obtaining the line loss detection model;

[0141] Step 4.4: Input the clustering results from Step 2 into the line loss detection model, and the line loss detection model will output the historical line loss causes corresponding to the distribution transformer areas with different grid structures.

[0142] Step 5: Real-time collection of power supply indicators related to line loss, after processing in steps 2 and 3, input into the line loss model trained in step 4 for detection of line loss anomalies and causes.

[0143] Obtaining such by applying the method proposed in this invention Figure 2 The diagram shows a database of cases indicating abnormal line losses in distribution transformer areas. Numerical values ​​represent the causes of line losses, including: 0 for incorrect transformer-household relationship, 1 for record error, 2 for metering error, 3 for metering fault, 4 for electricity usage without a meter, 5 for electricity theft or overcapacity usage, 6 for data acquisition fault, and 7 for abnormal line loss.

[0144] Another aspect of the present invention proposes a system for establishing a line loss model based on K-Means clustering and optimized RVM, including a memory, a processor, and a wireless communication module; the memory stores a computer program, and the processor calls the computer program to execute each step of the method for establishing a line loss model based on K-Means clustering and optimized RVM.

[0145] The applicant of this invention has provided a detailed description of the embodiments of the invention in conjunction with the accompanying drawings. However, those skilled in the art should understand that the above embodiments are merely preferred embodiments of the invention. The detailed description is only intended to help readers better understand the spirit of the invention and is not intended to limit the scope of protection of the invention. On the contrary, any improvements or modifications made based on the inventive spirit of the invention should fall within the scope of protection of the invention.

Claims

1. A method for establishing a line loss model based on K-Means clustering and optimized RVM, characterized in that, The establishment method includes: Step 1: Collect power supply indicators and historical line loss data for the distribution transformer area; use a power supply indicator processing model built with a noise-reducing autoencoder to obtain power supply indicators related to line loss; Step 2: Calculate the line loss evaluation index for different grid structure transformer substations using power supply indicators related to line loss; divide the different grid structure transformer substations into K groups using the line loss evaluation index, and use the center of each group as the initial cluster center of the K-Means clustering algorithm. Step 3: Collect historical causes of line losses in the distribution transformer area and perform pre-classification; Step 4: Using the pre-classification results of historical line loss causes, construct a line loss model based on the optimized RVM algorithm, and input the clustering results of Step 2 into the line loss model for training, so as to obtain the line loss causes of distribution transformer areas with different grid structures. Step 4 includes: Step 4.1: Construct the kernel function for optimizing the RVM algorithm, satisfying the following relationship: In the formula, To optimize the kernel function of the RVM algorithm, , These are all input values ​​for the kernel function. It is the width factor; Step 4.2, use a multi-population genetic algorithm to adjust the width factor. Optimization is performed to obtain the width factor optimization solution. ; Step 4.2 includes: Step 4.2.1, Initialize the number of individuals in the population. Maximum number of iterations The largest individual value in history and the initial x-axis coordinates of the population and the initial coordinates of the y-axis Among them, the value of the largest individual in history. The value is negative infinity, and all other parameters are random values. Step 4.2.2: Assign a random distance to each individual in the population. ; Step 4.2.3: Utilize each individual's x-axis coordinates, y-axis coordinates, and random distance. The fitness function value for each individual can be expressed by the following formula: In the formula, Indicates the first The determination value of the fitness function for each individual. Indicates the first The random distance between individuals For the first The current x-axis coordinate of each individual, in the first iteration. , For the first The current y-axis coordinate of each individual, in the first iteration. ; Step 4.2.4: Input the fitness function value of each individual into the fitness function, and use the fitness function value as the value of the largest individual. The fitness function satisfies the following relationship: In the formula, For the fitness function, A random value in [0, 1]; x-coordinate of the largest individual and y-axis coordinates They respectively satisfy the following relations: Step 4.2.5, set the value of the largest individual. The value of the largest individual in history Compare, if Then let ;if Then let Keep it unchanged; then proceed to step 4.2.6; Step 4.2.6: Determine whether the current iteration count has reached the maximum iteration count. If the condition is met, output the value of the largest individual in history. As a solution for the width factor optimization If the number of iterations has not reached the maximum number of iterations. If so, return to step 4.2.2; Step 4.3, optimize the solution using the width factor. Kernel function based on optimized RVM algorithm Thus, the line loss model is obtained; Step 4.4: Input the clustering results from Step 2 into the line loss model, and the line loss model will output the causes of line loss corresponding to the distribution transformer areas with different grid structures. Step 5: Real-time collection of power supply indicators related to line loss, after processing in Step 2, input into the line loss model trained in Step 4 for detection of line loss anomalies and causes.

2. The method for establishing a line loss model based on K-Means clustering and optimized RVM according to claim 1, characterized in that, In step 1, the power supply indicators related to line loss include: power supply radius, total length of low-voltage lines, power consumption type and proportion, and load factor; Among them, the nature and proportion of electricity use include: the electrical characteristics of the electricity load, the importance of the electricity load, the time, occasion, purpose of electricity use, and the permitted power outage time.

3. The method for establishing a line loss model based on K-Means clustering and optimized RVM according to claim 2, characterized in that, In step 1, when training the power supply index processing model, power supply indices unrelated to line loss are added to the training set as noise, and an unsupervised method is used to train the weights to obtain a trained power supply index processing model. The training set includes: power supply indices related to line loss and historical line loss data.

4. The method for establishing a line loss model based on K-Means clustering and optimized RVM according to claim 2, characterized in that, In step 2, the line loss evaluation index for distribution transformer areas with different grid structures is calculated using the power supply indexes related to line loss, according to the following formula: In the formula, For the first Line loss evaluation indicators for distribution transformer areas with grid structure. For the first The first in the distribution transformer area with a grid structure A power supply indicator related to line loss, For the first The first in the distribution transformer area with a grid structure The weights corresponding to the power supply indicators related to line loss , The number of power supply indicators related to line loss; In the formula, For various distribution transformer areas with different grid structures, the first The minimum value of a power supply indicator related to line loss. .

5. The method for establishing a line loss model based on K-Means clustering and optimized RVM according to claim 4, characterized in that, use The distribution transformer substations with different grid structures are divided into K groups. First, the average value of the line loss evaluation index of each distribution transformer substation area within each group is calculated. The distribution transformer substation area with the shortest distance from the average value within the group is taken as the center of the group, which is the initial cluster center of the K-Means clustering algorithm.

6. The method for establishing a line loss model based on K-Means clustering and optimized RVM according to claim 1, characterized in that, In step 3, the causes of historical line loss include: incorrect household-transformer relationship, file error, metering error, metering failure, electricity use without meter, electricity theft or over-capacity electricity use, data acquisition failure, and abnormal line loss.

7. The method for establishing a line loss model based on K-Means clustering and optimized RVM according to claim 6, characterized in that, The causes of historical line loss are pre-classified according to region, user type, average capacity per user, and load rate.

8. A system for establishing a line loss model based on K-Means clustering and optimized RVM, used to implement the method for establishing a line loss model based on K-Means clustering and optimized RVM as described in any one of claims 1 to 7, characterized in that, The line loss model establishment system includes a memory, a processor, and a wireless communication module; the memory stores a computer program, and the processor calls the computer program to execute the method described in any one of claims 1-7.

Citation Information

Patent Citations

  • A Method for Diagnosing Abnormal Line Losses in Distribution Networks Based on Concurrent Characteristics and Improved K-means Clustering

    CN111123039B

  • Transformer area reasonable line loss prediction method based on principal component analysis and neural network

    CN105160416A