One-way grounding fault detection method based on target detection and time sequence analysis
By adopting a one-way grounding fault detection method with object detection and timing analysis in the power system distribution network, deep learning algorithms and PELT algorithms extract fault characteristics from wave recording data, the problems of misjudgment and misjudgment in the existing technology are solved, and efficient and accurate fault detection is achieved.
Patent Information
- Application Number
- CN202510538078.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-27
- Publication Date
- 2025-08-01
AI Technical Summary
The prior art fault detection method based on full wave recording data in the power system distribution network has misjudgment and misjudgment, and the algorithm is inefficient, mainly because a large amount of non-fault feature data interferes with the accurate extraction of fault features.
A one-way grounding fault detection method based on object detection and timing analysis is adopted, and fault characteristics are extracted from the wave recording data through deep learning algorithms, and mutation points are detected in combination with PELT algorithm to identify abnormal waveforms and normal waveforms.
It significantly improves the accuracy and efficiency of fault detection, reduces misjudgment and misjudgment, simplifies the system maintenance and update process, and improves the stability and reliability of the system.
Smart Images
Figure CN120405314A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of power system distribution network automation, and particularly to a single-phase grounding fault detection method based on object detection and time series analysis. Background Art
[0002] The prior art usually analyzes based on full-scale waveform recording data. Due to the huge number of sampling points, a large amount of non-fault feature data is included, and these invalid data often interfere with the accurate extraction of fault features, reducing the recognition efficiency and accuracy of the algorithm.
[0003] Based on this, we propose a single-phase grounding fault detection method based on object detection and time series analysis. Summary of the Invention
[0004] The purpose of the present invention is to provide a single-phase grounding fault detection method based on object detection and time series analysis. By adopting an advanced object detection deep learning algorithm, it automatically extracts and learns fault features from waveform recording data, improves the accuracy of fault detection, reduces misjudgment and missed judgment; by designing an extensible and adaptive algorithm framework, it simplifies the maintenance and update process of the fault detection system, reduces the long-term operation cost, and improves the stability and reliability of the system.
[0005] The purpose of the present invention can be achieved through the following technical solutions: A single-phase grounding fault detection method based on object detection and time series analysis, the fault detection method specifically includes the following steps: S1. Data preprocessing; S2. Waveform classification and abnormal half-wave position detection; S3. Fusing time series data to detect the mutation point of the half-wave; S4. Identifying abnormal waveforms and normal waveforms.
[0006] As a further solution of the present invention, the data preprocessing in step S1 includes the following steps: S11. Data acquisition: Using the waveform recording data actually collected in the distribution network and storing it in the COMTRADE format; S12. Data parsing: Parsing the collected data, extracting key electrical parameters, such as the time series change waveforms of zero-sequence voltage and zero-sequence current, and then performing data denoising preprocessing to clean the data set.
[0007] As a further solution of the present invention, the waveform classification and abnormal half-wave position detection in step S2 includes the following steps: S21. Extracting zero-sequence current IZ1 and zero-sequence voltage UZ1, and converting the waveform recording data into an image; S22. Marking the abnormal half-waves of zero-sequence current IZ1 and zero-sequence voltage UZ1 on the image; S23. Train a deep learning object detection algorithm based on the labeled data set to identify the waveform type, the position and orientation of the abnormal half-wave.
[0008] As a further solution of the present invention, the step of detecting the mutation point of the half-wave by fusing the time series data in step S3 includes the following steps: S31. Detect the position interval of the abnormal half-wave based on deep learning object detection. By analyzing the original zero-sequence current IZ1 and zero-sequence voltage UZ1 data, extract the time series data of this section; S32. Use the mutation point detection algorithm PELT algorithm to analyze the time series data to find the mutation point and the mutation direction.
[0009] As a further solution of the present invention, the identification of abnormal waveforms and normal waveforms in step S4 is specifically as follows: Study the global normal waveforms, and identify the judgment basis for abnormal waveforms and normal waveforms: within the first half-wave where the mutation occurs, the mutation directions of the zero-sequence voltage and the zero-sequence current are opposite for abnormal waveforms, and the mutation directions of the zero-sequence voltage and the zero-sequence current are the same for normal waveforms.
[0010] Advantages of the present invention: 1. By using the single-phase grounding fault recording and detection algorithm for distribution networks based on deep learning object detection neural networks, the accuracy and efficiency of fault detection can be significantly improved.
[0011] 2. Based on the object detection results, fuse the time series data of the half-wave data that may be abnormal, and accurately detect the mutation point and mutation direction in the time series through the PELT algorithm, further improving the recognition accuracy of the algorithm. Description of the Drawings
[0012] The present invention will be further described below with reference to the accompanying drawings.
[0013] Figure 1 It is the method flow block diagram of the present invention. Detailed Embodiments
[0014] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.
[0015] Please refer to Figure 1 As shown, the present invention is a single-phase grounding fault detection method based on object detection and time series analysis. The fault detection method specifically includes the following steps: Data preprocessing: Data acquisition: Use the actual recorded wave data collected in the distribution network and store it in the COMTRADE format; Data parsing: Parse the collected data, extract key electrical parameters, such as the time-series change waveforms of zero-sequence voltage and zero-sequence current, and then perform data denoising preprocessing to clean the data set; In this embodiment, the data used is the single-phase grounding recorded wave data in the COMTRADE format, and the COMTRADE format is a file format data for storing transient data of the electrical system, generally including four types: header file hdr, configuration file cfg, data file dat, and information file inf.
[0016] Waveform classification and abnormal half-wave position detection: Extract the zero-sequence current IZ1 and zero-sequence voltage UZ1, and convert the recorded wave data into an image; Mark the abnormal half-waves of the zero-sequence current IZ1 and zero-sequence voltage UZ1 on the image; Train a deep learning object detection algorithm based on the marked data set to identify the waveform type, the position and orientation of the abnormal half-wave; Specifically, first analyze the recorded wave format in the COMTRADE format, extract the waveform data of the zero-sequence current IZ1 and zero-sequence voltage UZ1, and batch-convert the data into visual waveform image data; label the converted marked data, and based on the labeled data, use the yolov5 algorithm to train a high-precision detection model, and classify and identify the position, category and orientation of the possible abnormal half-wave of the waveform based on the trained detection algorithm model.
[0017] In this embodiment, the half-wave types are labeled as: abnormal upward zero-sequence current half-wave, abnormal downward zero-sequence current half-wave, abnormal upward zero-sequence voltage half-wave, abnormal downward zero-sequence voltage half-wave; the waveform types are labeled as: global normal waveform and global interference waveform.
[0018] Fuse time-series data to detect the mutation points of half-waves: Based on deep learning object detection, detect the position interval of the abnormal half-wave, analyze the original zero-sequence current IZ1 and zero-sequence voltage UZ1 data, and extract the time-series data of this section; use the PELT algorithm of the mutation point detection algorithm to analyze this time-series data to find the mutation point and the mutation direction.
[0019] Identify abnormal waveforms and normal waveforms: Study the global normal waveforms, and identify the judgment basis for abnormal waveforms and normal waveforms: In the first half-wave of the mutation, the mutation directions of the zero-sequence voltage and zero-sequence current are opposite for abnormal waveforms, and the mutation directions of the zero-sequence voltage and zero-sequence current are the same for normal waveforms.
[0020] The present invention can significantly improve the accuracy and efficiency of fault detection by converting the distribution network recording wave data into visual images and classifying the waveforms and detecting abnormal half waves based on the object detection algorithm; when detecting the position where abnormal waveforms may occur, by analyzing the timing data at this position, the mutation points are detected to further confirm the mutation points and mutation positions of the waveforms in this area, thus improving the recognition accuracy of the algorithm.
[0021] The above has described in detail an embodiment of the present invention, but the content described is only the preferred embodiment of the present invention and cannot be considered as limiting the scope of implementation of the present invention. All equivalent changes and improvements made according to the scope of the application of the present invention should still fall within the scope covered by the patent of the present invention.
Claims
1. A single-phase grounding fault detection method based on object detection and time series analysis, characterized in that The described fault detection method specifically includes the following steps: S1. Data preprocessing; S2. Waveform classification and abnormal half-wave position detection; S3. Detecting the mutation points of half-waves by fusing time-series data; S4. Identifying abnormal waveforms and normal waveforms.
2. The unidirectional grounding fault detection method based on object detection and timing analysis according to claim 1, wherein, The data preprocessing described in step S1 includes the following steps: S11. Data acquisition: Using the actual recorded wave data collected in the distribution network and storing it in the COMTRADE format; S12. Data parsing: Parsing the collected data, extracting key electrical parameters, such as the time-series change waveforms of zero-sequence voltage and zero-sequence current, and then performing data denoising preprocessing to clean the data set.
3. The single-phase grounding fault detection method based on object detection and time series analysis according to claim 1, wherein The waveform classification and abnormal half-wave position detection described in step S2 includes the following steps: S21. Extracting the zero-sequence current IZ1 and zero-sequence voltage UZ1 and converting the recorded wave data into an image; S22. Marking the abnormal half-waves of the zero-sequence current IZ1 and zero-sequence voltage UZ1 on the image; S23. Training a deep learning object detection algorithm based on the marked data set to identify the waveform type, the position and orientation of the abnormal half-waves.
4. The unidirectional ground fault detection method based on object detection and time series analysis according to claim 1, characterized in that, The detection of the mutation points of half-waves by fusing time-series data described in step S3 includes the following steps: S31. Detecting the position interval of the abnormal half-waves based on deep learning object detection, and extracting the time-series data of this section by analyzing the original zero-sequence current IZ1 and zero-sequence voltage UZ1 data; S32. Using the PELT algorithm, a mutation point detection algorithm, to analyze this time-series data and find the mutation points and mutation directions.
5. The single-phase grounding fault detection method based on object detection and time series analysis according to claim 1, wherein The identification of abnormal waveforms and normal waveforms described in step S4 is specifically: Studying the global normal waveforms to identify the judgment basis for abnormal waveforms and normal waveforms: In the first half-wave where the mutation occurs, the mutation directions of the zero-sequence voltage and zero-sequence current are opposite for abnormal waveforms, and the mutation directions of the zero-sequence voltage and zero-sequence current are the same for normal waveforms.