A new method for power system fault characteristic analysis

Abstract

This invention relates to a novel method for analyzing fault characteristics in power systems. First, it collects and preprocesses three-phase voltage, three-phase current, zero-sequence electrical quantities, protection device operation information, and fault indicator information at the grid connection point and lines. Then, it performs multi-domain feature extraction to simultaneously acquire transient change features, steady-state sequence component features, frequency domain wavelet packet energy entropy features, and harmonic distortion features, thereby constructing a multi-dimensional fault feature vector. Next, it uses an improved attention mechanism fusion model to fuse the multi-dimensional fault feature vector, and based on the fused fault feature vector, constructs a deep neural network model. Finally, it employs a robust least squares algorithm to calculate fault impedance and locate the fault, outputting a complete analysis result of the fault characteristics. This result is then uploaded to the dispatch master station to generate fault waveform reports, characteristic curves, and maintenance suggestions. This invention has the advantages of multi-domain fusion, strong anti-interference capability, and accurate fault identification.

Description

Technical Field

[0001] This invention belongs to the field of power system fault analysis technology, and specifically relates to a novel method for analyzing the fault characteristics of power systems. Background Technology

[0002] Fault characteristic analysis of power systems is a key technology for ensuring the safe and stable operation of the power grid. Currently, traditional methods mainly rely on single information sources such as protection device operation information and fault recording data, which suffer from problems such as insufficient information utilization, limited fault feature extraction capabilities, and low accuracy in identifying complex faults. With the advancement of smart grid construction, power system monitoring data exhibits multi-source, massive, and high-dimensional characteristics. How to effectively integrate multi-source information and achieve accurate fault characteristic analysis has become a technical challenge. Furthermore, existing fault analysis methods are mostly designed for traditional synchronous machine power grids, which are difficult to handle high-dimensional data and cannot adapt to the nonlinear and highly disturbed fault characteristics of new power systems. Therefore, to solve the above problems, it is necessary to develop a new power system fault characteristic analysis method that integrates multiple domains, has strong anti-interference capabilities, and provides accurate fault identification. Summary of the Invention

[0003] The purpose of this invention is to overcome the shortcomings of the prior art and provide a novel power system fault characteristic analysis method that integrates multiple domains, has strong anti-interference capabilities, and provides accurate fault identification.

[0004] The objective of this invention is achieved as follows: a novel method for analyzing fault characteristics in power systems, comprising the following steps: Step 1: Synchronously collect multi-source monitoring data before and after the occurrence of faults in the new power system, and obtain full waveform data of faults including disturbances from new energy sources and power electronic equipment; Step 2: Preprocess the data collected in Step 1 to form a multi-source data matrix with a unified time scale. Then, through anti-aliasing filtering, DC bias removal and adaptive notch filtering, a clean fault electrical signal is obtained. Step 3: Perform multi-domain feature extraction on the pure fault electrical signal to obtain transient change features, steady-state sequence component features, frequency domain wavelet packet energy entropy features and harmonic distortion features, thereby constructing a multi-dimensional fault feature vector; Step 4: An improved attention mechanism fusion model is used to fuse the multi-dimensional fault feature vector. Specifically, this includes: first, constructing a multi-source data feature extraction network to extract time domain, frequency domain, and time-frequency domain features from each data source; then, introducing an adaptive weight allocation module to dynamically adjust the weights according to the contribution of each data source to fault identification; and finally, generating a fault fusion feature vector through feature concatenation and dimensionality reduction operations. Step 5: Based on the fault fusion feature vector, construct a deep neural network model, and then train the network model by inputting a historical fault dataset to obtain a fault analysis model. The fault analysis model can identify the fault type by analyzing the fault fusion feature vector and synthesizing a trust function by allocating dynamic feature weights. Step 6: Based on the fault type, use the robust least squares algorithm to calculate the fault impedance and locate the fault, and then output the complete fault characteristic analysis results; Step 7: Store the fault characteristic analysis results in the local database and upload them to the dispatch master station to generate fault waveform reports, characteristic curves, and operation and maintenance suggestions.

[0005] Furthermore, the preprocessing operations in step 2 specifically include data alignment, outlier removal, missing value imputation, and data standardization.

[0006] Furthermore, the deep neural network model includes multi-scale convolutional layers, bidirectional LSTM layers, attention mechanism layers, and fully connected layers.

[0007] Furthermore, the fault types in step 5 include at least: single-phase grounding, two-phase short circuit, two-phase grounding, three-phase short circuit, high-resistance grounding, and intermittent arcing fault.

[0008] Furthermore, the multi-source monitoring data in step 1 includes the three-phase voltage, three-phase current, zero-sequence electrical quantity of the grid connection point and line, as well as the operation information of the protection device and the fault indicator information.

[0009] The beneficial effects of this invention are as follows: By collecting multi-source monitoring data such as three-phase voltage, three-phase current, zero-sequence electrical quantities, and the action information of protection devices and fault indicators at the grid connection point and line, this invention overcomes the limitations of single-source information analysis. By extracting multi-domain features from electrical signals, it simultaneously acquires transient change features, steady-state sequence component features, frequency domain wavelet packet energy entropy features, and harmonic distortion features to construct multi-dimensional fault feature vectors, thereby improving its overall anti-interference capability. By using an improved attention mechanism fusion model to fuse multi-dimensional fault feature vectors and constructing a deep neural network model based on the fused fault feature vectors, it increases the accuracy and efficiency of fault type identification. By employing a robust least squares algorithm for fault impedance calculation and fault location, it outputs complete fault characteristic analysis results, stores the fault characteristic analysis results in a local database, and uploads them to the dispatch master station to generate fault waveform reports, characteristic curves, and maintenance suggestions, thereby increasing the convenience of subsequent fault handling operations. In summary, this invention has the advantages of multi-domain fusion, strong anti-interference capability, and accurate fault identification. Detailed Implementation

[0010] The present invention will now be further described.

[0011] An embodiment of a novel power system fault characteristic analysis method includes the following steps: Step 1: Synchronously collect multi-source monitoring data before and after the occurrence of faults in the new power system, and obtain full waveform data of faults including disturbances from new energy sources and power electronic equipment; among which, multi-source monitoring data include three-phase voltage, three-phase current, zero-sequence electrical quantity at the grid connection point and line, as well as the action information of protection devices and fault indicator information. Step 2: Perform preprocessing operations on the data collected in Step 1 to form a multi-source data matrix with a unified time scale. Then, obtain a clean fault electrical signal through anti-aliasing filtering, DC bias removal, and adaptive notch filtering. In this process, the data preprocessing operations specifically include data alignment, outlier removal, missing value imputation, and data standardization. Step 3: Perform multi-domain feature extraction on the pure fault electrical signal to obtain transient change features, steady-state sequence component features, frequency domain wavelet packet energy entropy features and harmonic distortion features, thereby constructing a multi-dimensional fault feature vector; Step 4: An improved attention mechanism fusion model is used to fuse the multi-dimensional fault feature vector. Specifically, this includes: first, constructing a multi-source data feature extraction network to extract time domain, frequency domain, and time-frequency domain features from each data source; then, introducing an adaptive weight allocation module to dynamically adjust the weights according to the contribution of each data source to fault identification; and finally, generating a fault fusion feature vector through feature concatenation and dimensionality reduction operations. Step 5: Based on the fault fusion feature vector, construct a deep neural network model, which includes multi-scale convolutional layers, bidirectional LSTM layers, attention mechanism layers, and fully connected layers. Then, input the historical fault dataset to train the network model to obtain a fault analysis model. The fault analysis model can identify fault types by analyzing the fault fusion feature vector and synthesizing a trust function by assigning dynamic feature weights. The fault types include at least: single-phase grounding, two-phase short circuit, two-phase grounding, three-phase short circuit, high-resistance grounding, and intermittent arcing faults. Step 6: Based on the fault type, use the robust least squares algorithm to calculate the fault impedance and locate the fault, and then output the complete fault characteristic analysis results; Step 7: Store the fault characteristic analysis results in the local database and upload them to the dispatch master station to generate fault waveform reports, characteristic curves, and operation and maintenance suggestions.

[0012] In use, this invention first synchronously collects multi-source monitoring data before and after a fault in the new power system, acquiring full waveform data of the fault including disturbances from new energy sources and power electronic equipment. The multi-source monitoring data includes three-phase voltage, three-phase current, zero-sequence electrical quantities at the grid connection point and line, as well as the action information of protection devices and fault indicator information. Then, the collected data undergoes preprocessing operations, namely, data alignment, outlier removal, missing value interpolation, and standardization, thereby forming a multi-source data matrix with a unified time scale. Subsequently, through anti-aliasing filtering, DC bias removal, and adaptive notch filtering, a clean fault electrical signal is obtained. Finally, the clean fault electrical signal is further processed... Multi-domain feature extraction is performed to obtain transient mutation features, steady-state order component features, frequency domain wavelet packet energy entropy features, and harmonic distortion features, thereby constructing a multi-dimensional fault feature vector. Finally, an improved attention mechanism fusion model is used to fuse the multi-dimensional fault feature vector. Specifically, a multi-source data feature extraction network is first constructed to extract time-domain, frequency-domain, and time-frequency-domain features from each data source. Then, an adaptive weight allocation module is introduced to dynamically adjust the weights according to the contribution of each data source to fault identification. Finally, a fault fusion feature vector is generated through feature concatenation and dimensionality reduction. After completing the above operations, a deep neural network model is constructed based on the fault fusion feature vector, and... Subsequently, the network model is trained using a historical fault dataset to obtain a fault analysis model. This model analyzes fault fusion feature vectors and synthesizes a trust function by assigning dynamic feature weights, thereby identifying fault types. Finally, based on the fault type, a robust least squares algorithm is used to calculate fault impedance and locate the fault, outputting a complete fault characteristic analysis result. This result is stored in a local database and uploaded to the dispatch master station, generating fault waveform reports, characteristic curves, and maintenance suggestions. This invention employs this method to analyze the three-phase voltage, three-phase current, zero-sequence electrical quantities at the grid connection point and line, as well as the actions of protection devices. The acquisition of multi-source monitoring data, including information and fault indicator information, overcomes the limitations of single-source analysis. By extracting multi-domain features from electrical signals, and simultaneously acquiring transient change features, steady-state sequence component features, frequency domain wavelet packet energy entropy features, and harmonic distortion features, a multi-dimensional fault feature vector can be constructed, improving its overall anti-interference capability. By using an improved attention mechanism fusion model to fuse the multi-dimensional fault feature vector, and constructing a deep neural network model based on the fused fault feature vector, the accuracy and efficiency of fault type identification can be increased. In summary, this invention has the advantages of multi-domain fusion, strong anti-interference capability, and accurate fault identification.

[0013] The above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of the claims of the present invention.

Claims

1. A novel method for analyzing fault characteristics in power systems, characterized in that, Includes the following steps: Step 1: Synchronously collect multi-source monitoring data before and after the occurrence of faults in the new power system, and obtain full waveform data of faults including disturbances from new energy sources and power electronic equipment; Step 2: Preprocess the data collected in Step 1 to form a multi-source data matrix with a unified time scale. Then, through anti-aliasing filtering, DC bias removal and adaptive notch filtering, a clean fault electrical signal is obtained. Step 3: Perform multi-domain feature extraction on the pure fault electrical signal to obtain transient change features, steady-state sequence component features, frequency domain wavelet packet energy entropy features and harmonic distortion features, thereby constructing a multi-dimensional fault feature vector; Step 4: An improved attention mechanism fusion model is used to fuse the multi-dimensional fault feature vector. Specifically, this includes: first, constructing a multi-source data feature extraction network to extract time domain, frequency domain, and time-frequency domain features from each data source; then, introducing an adaptive weight allocation module to dynamically adjust the weights according to the contribution of each data source to fault identification; and finally, generating a fault fusion feature vector through feature concatenation and dimensionality reduction operations. Step 5: Based on the fault fusion feature vector, construct a deep neural network model, and then train the network model by inputting a historical fault dataset to obtain a fault analysis model. The fault analysis model can identify the fault type by analyzing the fault fusion feature vector and synthesizing a trust function by allocating dynamic feature weights. Step 6: Based on the fault type, use the robust least squares algorithm to calculate the fault impedance and locate the fault, and then output the complete fault characteristic analysis results; Step 7: Store the fault characteristic analysis results in the local database and upload them to the dispatch master station to generate fault waveform reports, characteristic curves, and operation and maintenance suggestions.

2. The novel power system fault characteristic analysis method as described in claim 1, characterized in that: The preprocessing operations in step 2 specifically include data alignment, outlier removal, missing value imputation, and data standardization.

3. The novel power system fault characteristic analysis method as described in claim 1, characterized in that: The deep neural network model includes multi-scale convolutional layers, bidirectional LSTM layers, attention mechanism layers, and fully connected layers.

4. The novel power system fault characteristic analysis method as described in claim 1, characterized in that: The fault types in step 5 include at least: single-phase grounding, two-phase short circuit, two-phase grounding, three-phase short circuit, high-resistance grounding, and intermittent arcing fault.

5. The novel power system fault characteristic analysis method as described in claim 1, characterized in that: The multi-source monitoring data in step 1 includes the three-phase voltage, three-phase current, zero-sequence electrical quantity of the grid connection point and line, as well as the operation information of the protection device and the fault indicator information.