Active power distribution network fault section positioning method based on voltage and current visualization

Through voltage and current visualization and three-dimensional convolutional neural network model, the problem of inefficient fault positioning of distribution networks is solved, efficient and accurate fault identification and positioning is achieved, and the operational safety and reliability of distribution networks are improved.

CN118914760AInactive Publication Date: 2024-11-08HANGZHOU ELECTRIC EQUIP MFG
View PDF 10 Cites 0 Cited by

Patent Information

Application Number
CN202411235839.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-04
Publication Date
2024-11-08
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The traditional distribution network fault location method is inefficient in distributed power supply environments, difficult to accurately and promptly detect faults, and is susceptible to abnormal data and unknown factors, affecting model performance and generalization capabilities.

Method used

Using a method based on voltage and current visualization, by establishing an active medium voltage distribution network model, collecting fault data, performing standardization and visualization processing, constructing a three-dimensional feature cube data set, and building a three-dimensional convolutional neural network model for training to extract deep layers Features and improves fault positioning accuracy.

Benefits of technology

It significantly improves the efficiency and response speed of fault handling, reduces the losses and impacts caused by faults, improves the operating safety and reliability of the distribution network, and adapts to distribution network environments of different scales and complexities.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118914760A_ABST
    Figure CN118914760A_ABST
Patent Text Reader

Abstract

The invention discloses an active power distribution network fault section positioning method based on voltage and current visualization, and the method comprises the following steps: S1, building an active medium-voltage power distribution network model, and collecting fault data; s2, carrying out standardization processing on the collected data; s3, performing visualization processing on the standardized data, compounding the standardized data into three-dimensional feature cube data, and establishing a three-dimensional feature cube data set; s4, building a three-dimensional convolutional neural network model; and S5, dividing the three-dimensional feature cube data set into a training set and a test set, inputting the training set into the three-dimensional convolutional neural network model for iterative training, and then inputting the test set into the three-dimensional convolutional neural network model to output a model positioning result. By adopting the active power distribution network fault section positioning method based on voltage and current visualization, the fault processing efficiency and the response speed can be greatly improved, the loss and the influence caused by the fault can be reduced, and the operation safety and the reliability of the power distribution network can be improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the technical field of electric power automation, and in particular to a method for locating a fault section of an active distribution network based on voltage and current visualization. Background Art

[0002] Distribution lines are an integral part of modern power systems, which directly affect the safety and stability of power supply. With the widespread access of distributed power sources, a series of challenges have been brought to the operation and management of traditional distribution networks, especially in terms of fault detection, location and isolation. This is because distributed power sources change the power flow direction and fault current distribution of the distribution network, affecting the effectiveness of traditional fault location methods. An effective power system protection scheme should be able to detect all faults as soon as possible. Accurate and timely fault diagnosis is of great significance to the safe operation and power supply reliability of the distribution system. In general, fault diagnosis methods can be divided into model-based, signal-based and data-driven methods. Unlike the previous two methods, data-driven fault diagnosis methods only need to use a large amount of historical data. It has the ability to self-learn and adjust weights and thresholds while using historical data for training. However, traditional intelligent methods limit the utilization of the physical structure and data information of the power network, and are easily affected by abnormal data input and unknowns, thereby affecting the performance and generalization ability of the model. Therefore, studying the characteristics of fault data and processing it as well as transforming the deep learning model so as to better extract more critical features and information from the image domain is of great significance to improving model performance and generalization ability, improving the accuracy of locating fault sections in distribution networks, and ensuring the stable and safe operation of distribution networks. Summary of the invention

[0003] The purpose of the present invention is to provide a method for locating fault sections of an active distribution network based on voltage and current visualization, which can greatly improve the efficiency and response speed of fault handling, reduce the losses and impacts caused by faults, and improve the operating safety and reliability of the distribution network.

[0004] To achieve the above object, the present invention provides a method for locating a fault section of an active distribution network based on voltage and current visualization, comprising the following steps:

[0005] S1. Establish an active medium voltage distribution network model and collect fault data;

[0006] S2, standardize the collected data;

[0007] S3, visualizing the standardized data and compounding them into three-dimensional feature cube data to establish a three-dimensional feature cube data set;

[0008] S4, build a three-dimensional convolutional neural network model;

[0009] S5. Divide the three-dimensional feature cube data set into a training set and a test set, input the training set into the three-dimensional convolutional neural network model for iterative training, then input the test set into the three-dimensional convolutional neural network model, and finally output the model positioning result.

[0010] Preferably, in step S1, an active medium voltage distribution network model is constructed using Simulink software, different types of faults are set in different line sections, and distribution network fault characteristic data, namely, the three-phase voltage and current of each node, are collected.

[0011] Preferably, the collected distribution network fault characteristic data includes fault characteristic data of the fault period and the non-fault period.

[0012] Preferably, in step S2, the collected data is first filtered and denoised, and then the fault data is standardized according to the maximum and minimum standardization criteria, and the maximum and minimum standardization criteria formula is defined as follows:

[0013]

[0014] Among them, d is the collected original data, is the processed data, d max and d min are the maximum and minimum values ​​under normal conditions respectively.

[0015] Preferably, in step S3, the standardized voltage and current data are converted into a visualized heat map using a signal-to-image algorithm, and the heat map corresponding to each node is embedded and composited into three-dimensional feature cube data, specifically:

[0016] A fault trigger is set in each fault section. The heat map in each three-dimensional feature cube data is converted from the fault data collected by the distribution network under the same fault type and fault location. The heat map data corresponding to each node is embedded along the Z axis to form a three-dimensional feature cube data.

[0017] For each 3D feature cube data, a label is manually set according to its corresponding fault location, and a 3D feature cube data set is established; wherein the label setting follows a consistent classification system, and the label exists in the form of one-hot encoding.

[0018] Preferably, in step S4, a structure of an input layer, a three-dimensional convolution layer, a pooling layer, and a fully connected layer is constructed; a plurality of three-dimensional convolution layers are provided, each of which is followed by an activation layer and a pooling layer; the calculation formula of each three-dimensional convolution layer is:

[0019] f(x,y,z)=(K*V)(x,y,z)=∑ i,j,kK(i,j,k)·V(xi,yj,zk)

[0020] Among them, K is the convolution kernel, V is the input volume, x, y, z are the positions in the input volume, and i, j, k are the relative positions in the convolution kernel;

[0021] The loss function used in the training of the 3D convolutional neural network model is the cross entropy loss, which is expressed as:

[0022]

[0023] Among them, y o,c is the real one-hot encoding, P o,c is the predicted probability of the model for each category, M is the total number of categories; the optimization algorithm selects Adam optimizer.

[0024] Preferably, a batch normalization layer and a discard layer are also introduced into the three-dimensional convolutional neural network model.

[0025] Preferably, in step S5, an independent validation set is also provided, and testing is performed on the validation set to verify the effectiveness and accuracy of the three-dimensional convolutional neural network model by calculating the loss rate and the final positioning accuracy.

[0026] Therefore, the beneficial effects of the present invention using the above-mentioned active distribution network fault section location method based on voltage and current visualization are as follows:

[0027] (1) The present invention constructs a three-dimensional feature cube and effectively embeds time series data on the Z axis, so that the network can extract richer and deeper features in the three-dimensional space, thereby improving the accuracy of fault identification.

[0028] (2) Through deep learning of three-dimensional data, this network model can quickly identify the specific location and type of fault. Compared with traditional methods, it greatly improves the efficiency and response speed of fault handling and reduces the losses and impacts caused by faults.

[0029] (3) The present invention is not limited by the scale and structure of the distribution network, and can adapt to distribution network environments of different scales and complexities, thereby improving the operational safety and reliability of the distribution network and providing strong technical support for the intelligent management of the distribution network.

[0030] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 It is a step diagram of an embodiment of a method for locating a fault section of an active distribution network based on voltage and current visualization of the present invention;

[0032] Figure 2 It is a visualization heat map conversion process diagram of an embodiment of a method for locating a fault section of an active distribution network based on voltage and current visualization of the present invention. DETAILED DESCRIPTION

[0033] The technical solution of the present invention is further described below through the accompanying drawings and embodiments.

[0034] Embodiment 1

[0035] like Figure 1 As shown, the present invention provides a method for locating a fault section of an active distribution network based on voltage and current visualization, comprising the following steps:

[0036] S1. Establish an active medium-voltage distribution network model and collect fault data.

[0037] A detailed IEEE 33-node active medium-voltage distribution network model was constructed using Simulink software. The model accurately simulates the electrical characteristics and topology of the actual distribution network, including various distribution equipment (such as transformers, lines, branches, etc.) and control systems. Through this model, various normal and fault conditions can be simulated in a virtual environment, providing an experimental basis for further fault analysis and detection.

[0038] A systematic approach is adopted to simulate various possible distribution network faults, and different types of faults are set in different line sections, such as A-phase single-phase grounding fault, AB-phase short-circuit grounding fault, ABC short-circuit fault, etc. These faults are precisely set in the model and introduced at specific line sections and times (the total simulation time is 0.5s, the fault occurrence period is 0.02s to 0.1s, and the sampling frequency is 20kHz) to simulate the distribution network fault scenarios in the real world. The distribution network fault characteristic data, that is, the three-phase voltage and current of each node, are collected. These data not only include the instant readings when the fault occurs, but also the data before and after the fault, thus forming a time series data set, which is of great significance for analyzing the impact and propagation of faults.

[0039] S2. Standardize the collected data.

[0040] In this embodiment, data enhancement technology is used to generate more diverse fault data by adjusting the fault type and fault parameters (fault resistance) and setting different fault locations to enhance the generalization ability of model training.

[0041] First, the collected data is filtered and denoised, and then the fault data is standardized according to the maximum and minimum standardization criteria to ensure that the data quality meets the requirements of subsequent processing.

[0042] The maximum and minimum normalization criterion formula is defined as follows:

[0043]

[0044] Among them, d is the collected original data, is the processed data, d max and d min are the maximum and minimum values ​​under normal conditions respectively.

[0045] S3. Visualize the standardized data and compound them into three-dimensional feature cube data to establish a three-dimensional feature cube data set.

[0046] The signal-to-image algorithm is used to transform the standardized voltage and current data from the time domain to the image domain and convert it into a visual heat map. In this embodiment, although the entire simulation time is 0.5s, it only needs to include fault and non-fault waveform data. Therefore, it is only necessary to intercept the waveform data from the simulation step length of 0 to 2000 (0 to 0.1s, 0 to 0.02s is non-fault data, 0.02 to 0.1s is fault data). The horizontal axis corresponds to the simulation step length, and the vertical axis is divided into 6 areas, representing the three-phase current and three-phase voltage of a certain node. Therefore, the dimension size of each heat map is (6, 2000). Each heat map can be regarded as a two-dimensional matrix. The specific steps are as follows Figure 2 shown.

[0047] The heat map corresponding to each node is embedded and compounded into a three-dimensional feature cube data, specifically:

[0048] A fault trigger is set in each fault section. The heat map in each three-dimensional feature cube data is converted from the fault data collected by the distribution network under the same fault type and fault location. The heat map data corresponding to each node is embedded along the Z axis to form a three-dimensional feature cube data. The three-dimensional feature cube data formed not only contains node information in space, but also contains dynamic changes in the time dimension, which significantly improves the accuracy and efficiency of fault location.

[0049] For each 3D feature cube data, a label is manually set according to its corresponding fault location, and a 3D feature cube data set is established. The label setting follows a consistent classification system, such as the line section where the specific fault occurs. These labels exist in the form of one-hot encoding, which ensures the clarity and accuracy of the classification task during training.

[0050] S4. Build a three-dimensional convolutional neural network model.

[0051] In this embodiment, a three-dimensional convolutional neural network (3D CNN) model designed for processing three-dimensional data is constructed to train the three-dimensional feature cube data generated by the above steps. Considering that the classification problem is relatively simple, a lightweight network structure is selected. The hyperparameters of the three-dimensional convolutional neural network (3D-CNN) are shown in the following table:

[0052] Table 1 Hyperparameters of 3D convolutional neural network

[0053]

[0054]

[0055] The 3D convolutional neural network model uses a 3D convolutional layer to process the input 3D feature cube and fully mine the correlation and feature information of the data in spatial and temporal dimensions.

[0056] The three-dimensional convolutional neural network model in this embodiment includes an input layer, a three-dimensional convolutional layer, a pooling layer, and a fully connected layer, wherein a plurality of three-dimensional convolutional layers are provided, each of which is followed by an activation layer (usually using a ReLU activation function) and a pooling layer (using maximum pooling or average pooling). The core function of the three-dimensional convolutional layer is to slide the convolution kernel on the data cube, calculate the convolution of the kernel and the data, and thus extract features. The calculation formula for each three-dimensional convolutional layer is:

[0057] f(x,y,z)=(K*V)(x,y,z)=∑ i,j,k K(i,j,k)·V(xi,yj,zk)

[0058] Among them, K is the convolution kernel, V is the input volume, x, y, z are the positions in the input volume, and i, j, k are the relative positions in the convolution kernel.

[0059] In order to improve the generalization ability of the 3D convolutional neural network model and prevent overfitting, batch normalization layer and dropout layer are introduced into the 3D convolutional neural network model. The batch normalization layer is used to adjust the mean and variance of each layer output to ensure the stability of network training. The dropout layer randomly discards a certain proportion of neurons during the training process to reduce the overfitting of the 3D convolutional neural network model to the training data.

[0060] The loss function used in the training of the 3D convolutional neural network model is the cross entropy loss, which is suitable for classification problems and is expressed as:

[0061]

[0062] Among them, y o,c is the real one-hot encoding, P o,cis the predicted probability of the model for each category, and M is the total number of categories.

[0063] In this embodiment, the optimization algorithm of the three-dimensional convolutional neural network model selects the Adam optimizer, which combines the advantages of momentum and adaptive learning rate, can automatically adjust the learning rate in the learning of different parameters, accelerate the convergence speed and improve the training efficiency.

[0064] S5. Divide the 3D feature cube data set into a training set and a test set, input the training set into the 3D convolutional neural network model for multiple iterations of training, calculate the loss function and update the network weights in each iteration. Then input the test set into the 3D convolutional neural network model, and finally output the model positioning result.

[0065] In this embodiment, an independent validation set is also provided, and testing is performed on the validation set to verify the effectiveness and accuracy of the three-dimensional convolutional neural network model by calculating the loss rate and the final positioning accuracy.

[0066] Therefore, the present invention adopts the above-mentioned active distribution network fault section locating method based on voltage and current visualization, which can greatly improve the efficiency and response speed of fault handling, reduce the losses and impacts caused by faults, and improve the operation safety and reliability of the distribution network.

[0067] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A method for locating fault sections in active distribution networks based on voltage and current visualization. It is characterized in that The following steps are involved: S1. Establish an active medium voltage distribution network model and collect fault data; S2, standardize the collected data; S3, visualizing the standardized data and compounding them into three-dimensional feature cube data to establish a three-dimensional feature cube data set; S4, build a three-dimensional convolutional neural network model; S5. Divide the three-dimensional feature cube data set into a training set and a test set, input the training set into the three-dimensional convolutional neural network model for iterative training, then input the test set into the three-dimensional convolutional neural network model, and finally output the model positioning result.

2. According to claim 1, a method for locating fault sections of an active distribution network based on voltage and current visualization, Features: In step S1, the active medium voltage distribution network model is constructed using Simulink software, different types of faults are set in different line sections, and the distribution network fault characteristic data, namely the three-phase voltage and current of each node, are collected.

3. A method for locating fault sections of an active distribution network based on voltage and current visualization according to claim 2, Features: The collected distribution network fault characteristic data includes fault characteristic data of the fault period and the non-fault period.

4. The method for locating fault sections of an active distribution network based on voltage and current visualization according to claim 1, Features: In step S2, the collected data is first filtered and denoised, and then the fault data is standardized according to the maximum and minimum standardization criteria. The maximum and minimum standardization criteria formula is defined as follows: Among them, d is the collected original data, is the processed data, d max and d min are the maximum and minimum values ​​under normal conditions respectively.

5. The method for locating fault sections of an active distribution network based on voltage and current visualization according to claim 1, Features: In step S3, the standardized voltage and current data are converted into a visual heat map using a signal-to-image algorithm, and the heat map corresponding to each node is embedded and composited into a three-dimensional feature cube data, specifically: A fault trigger is set in each fault section. The heat map in each three-dimensional feature cube data is converted from the fault data collected by the distribution network under the same fault type and fault location. The heat map data corresponding to each node is embedded along the Z axis to form a three-dimensional feature cube data. For each 3D feature cube data, a label is manually set according to its corresponding fault location, and a 3D feature cube data set is established; wherein the label setting follows a consistent classification system, and the label exists in the form of one-hot encoding.

6. A method for locating fault sections in an active distribution network based on voltage and current visualization according to claim 1, Features: In step S4, the structure of input layer, three-dimensional convolution layer, pooling layer and full connection layer is constructed; there are several three-dimensional convolution layers, each of which is followed by an activation layer and a pooling layer; the calculation formula of each three-dimensional convolution layer is: f(x,y,z)=(K*V)(x,y,z)=∑ i,j,k K(i,j,k)·V(x-i,y-j,z-k) Among them, K is the convolution kernel, V is the input volume, x, y, z are the positions in the input volume, and i, j, k are the relative positions in the convolution kernel; The loss function used in the training of the 3D convolutional neural network model is the cross entropy loss, which is expressed as: Among them, y o,c is the true one-hot encoding, P o,c is the predicted probability of the model for each category, M is the total number of categories; the optimization algorithm selects Adam optimizer.

7. A method for locating fault sections of an active distribution network based on voltage and current visualization according to claim 6, Features: Batch normalization layers and dropout layers are also introduced in the 3D convolutional neural network model.

8. The method for locating fault sections of an active distribution network based on voltage and current visualization according to claim 1, Features: In step S5, an independent validation set is also provided, and testing is performed on the validation set to verify the effectiveness and accuracy of the three-dimensional convolutional neural network model by calculating the loss rate and the final positioning accuracy.

Citation Information

Patent Citations

  • Lithium ion power battery estimation method based on time convolution network

    CN113671381A

  • Lithium ion battery SOC estimation method based on 3DCNN

    CN114062948A

  • Electric energy meter overheating fault discrimination method based on thermal imaging

    CN114359146A

  • Transmission line fault detection positioning method based on convolutional neural network

    CN115372754A

  • Single-phase earth fault positioning method

    CN115980515A