A new power system grid fault feature analysis method

Abstract

The application relates to a novel power system grid fault feature analysis method, first, four-dimensional data of alternating current side electrical quantity, direct current side conversion parameter, equipment operation state and environmental meteorological parameter are integrated, and the integrated data is subjected to noise reduction, normalization, missing value completion and abnormal value elimination treatment, so that a standardized feature sample set is formed, then, an improved wavelet packet transform is adopted to decompose a transient signal, time domain, frequency domain and time-frequency domain features are extracted, so that a high-dimensional fault feature matrix is constructed, then, a fault feature secondary verification is realized in a mode of combination of cosine similarity and dynamic time warping; meanwhile, a graph convolution network is introduced, fault contribution degrees of each node are calculated in combination of a power grid topological structure and a power flow distribution, a fault source is accurately positioned, a fault propagation path is combed, and a fault influence range is quantified; and the application has the advantages of comprehensiveness, high efficiency, accuracy and strong adaptability.

Description

Technical Field

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

[0002] With the advancement of new power system construction, the power grid as a whole exhibits characteristics of high proportion of new energy sources, high density of power electronic equipment, and deep coupling of AC and DC power grids. Currently, traditional fault analysis methods rely on synchronous motor power frequency models, which can only analyze fundamental components and simple transients, and have the following shortcomings: 1. Incomplete feature extraction, unable to capture new fault features such as broadband harmonics, non-power frequency components, and short-term abrupt changes; 2. Single analysis method, mainly relying on single electrical quantities and threshold criteria, without integrating multi-source spatiotemporal data, making it difficult to adapt to complex fault scenarios; 3. Insufficient intelligence, traditional models have weak generalization ability and poor adaptability to feature variations caused by new energy fluctuations and control strategy switching; 4. Difficult fault tracing, lacking spatiotemporal correlation analysis, unable to quickly locate fault sources and clarify propagation paths, affecting fault isolation and system recovery. Therefore, to solve the above problems, it is necessary to develop a comprehensive, efficient, accurate, and highly adaptable new power system power grid fault feature analysis method. Summary of the Invention

[0003] The purpose of this invention is to overcome the shortcomings of the prior art and provide a comprehensive, efficient, accurate and highly adaptable new method for analyzing the characteristics of power grid faults in power systems.

[0004] The objective of this invention is achieved as follows: a novel method for analyzing the characteristics of power grid faults in a power system, comprising the following steps: Step 1: Synchronously collect three-phase voltage, three-phase current, zero-sequence voltage, zero-sequence current, harmonic components, power, frequency, DC components, and switch status of all nodes in the power grid, as well as ambient temperature, humidity, wind speed, and light intensity data, and construct a spatiotemporal coupled fault dataset based on the collected data. Step 2: The dataset collected in Step 1 is subjected to noise reduction, normalization, missing value completion, and outlier removal. Then, the time sequence alignment of all measurement point data is completed based on the network-wide time synchronization protocol to eliminate the time difference of multi-device acquisition. Finally, a dynamic sliding window is used to extract transient data before and after the fault to form a standardized feature sample set. Step 3: The transient signal is decomposed using an improved wavelet packet transform to extract the time-domain amplitude, phase, steepness, and mutation rate; the frequency-domain fundamental wave, harmonics, interharmonics, and spectral entropy; the time-frequency domain wavelet energy, wavelet entropy, and modal components; as well as topological correlation features and temporal correlation features, thereby constructing a high-dimensional fault feature matrix. Step 4: Using principal component analysis combined with an improved genetic algorithm, redundant features are eliminated, and the optimal feature subset that is strongly correlated with the fault type and location is selected; Step 5: Construct a hybrid model that integrates local spatial features, temporal correlation, and key feature weights. The optimal feature subset from Step 4 is used as input, and the classified fault is used as output. This enables accurate location and intelligent identification of fault types (short circuit, open circuit, high-resistance grounding, arc grounding), fault phases, and fault sections. Then, the identified faults are classified by hazard level. Step 6: Based on the fault feature matching algorithm, compare the real-time features with the historical fault database to verify the analysis results. Then, construct a power grid topology map and combine the power grid topology and power flow direction to calculate the node fault contribution using graph convolutional networks, thereby tracing the fault origin, propagation path and impact range. Step 7: Output the fault type, location, time, characteristic parameters, and risk level, and generate the fault trend curve, topology location map, and characteristic spectrum map.

[0005] Furthermore, in step 2, a dynamic window with a length of 200ms can be set to capture transient data 50ms before and 150ms after the fault, thereby forming a standardized sample set.

[0006] Furthermore, the noise reduction process in step 2 specifically involves: using wavelet threshold noise reduction to suppress electromagnetic interference; the normalization process specifically involves: using minimum-maximum normalization to unify the dimensions; the missing value completion operation specifically involves: using linear interpolation to complete the missing data; and the outlier removal operation specifically involves: removing outliers based on the Laida criterion.

[0007] Furthermore, in step 5, the fault severity levels are divided into four levels: minor distortion fault, general fault, severe short circuit fault, and destructive fault.

[0008] Furthermore, the fault feature matching algorithm in step 6 adopts a combination of cosine similarity and dynamic time warping.

[0009] The beneficial effects of this invention are as follows: This invention synchronously collects data on three-phase voltage, three-phase current, zero-sequence voltage, zero-sequence current, harmonic components, power, frequency, DC components, and switching status of all nodes in the power grid, as well as ambient temperature, humidity, wind speed, and light intensity. It innovatively integrates four-dimensional data of AC side electrical quantities, DC side converter parameters, equipment operating status, and environmental meteorological parameters, which solves the shortcomings of traditional methods that are only suitable for traditional synchronous machine power grids and cannot adapt to complex fault scenarios in new power systems. This greatly broadens the applicable scenarios and versatility of fault analysis methods. This invention solves problems such as electromagnetic interference, missing data points, abnormal jumps, inconsistent dimensions, and time sequence misalignment in signals by integrating multiple preprocessing methods, including wavelet threshold denoising, Raida criterion outlier removal, linear interpolation completion, maximum-minimum normalization, and dynamic sliding window truncation. This significantly improves the quality of sample data and ensures the accuracy and effectiveness of subsequent fault feature extraction. This invention achieves secondary verification of fault features by combining cosine similarity with dynamic time warping, thus avoiding model misjudgment. At the same time, by introducing graph convolutional networks and combining the power grid topology and power flow distribution to calculate the fault contribution of each node, it accurately locates the fault source, sorts out the fault propagation path, and quantifies the fault impact range, realizing the upgrade from "simple fault identification" to a full-chain fault analysis capability of "identification-verification-source tracing-evaluation". In summary, this invention overcomes the shortcomings of traditional fault analysis methods, which rely on power frequency steady state, have single characteristics, poor anti-interference capabilities, lack of traceability, and weak adaptability. Targeting typical operating conditions of new power systems characterized by mixed disturbances from new energy sources, weak fault characteristics, complex transient distortions, and severe AC / DC coupling, it achieves a high-precision, highly sensitive, real-time, traceable, and self-updating intelligent fault analysis system. This effectively reduces the probability of power grid protection maloperation and failure to operate, and improves the operational stability and fault handling efficiency of new power systems. Overall, this invention has the advantages of being comprehensive, efficient, accurate, and highly adaptable. Detailed Implementation

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

[0011] Example: A novel power grid fault characteristic analysis method for power systems, comprising the following steps: Step 1: Synchronously collect three-phase voltage, three-phase current, zero-sequence voltage, zero-sequence current, harmonic components, power, frequency, DC components, and switch status of all nodes in the power grid, as well as ambient temperature, humidity, wind speed, and light intensity data, and construct a spatiotemporal coupled fault dataset based on the collected data. Step 2: The dataset collected in Step 1 is subjected to noise reduction, normalization, missing value completion, and outlier removal. Specifically, noise reduction uses wavelet thresholding to suppress electromagnetic interference; normalization uses minimum-maximum normalization to unify dimensions; missing value completion uses linear interpolation; and outlier removal uses the Laida criterion to remove outliers. Then, time alignment of all measurement point data is completed based on the network-wide time synchronization protocol to eliminate time differences between multiple devices. Finally, a dynamic sliding window is used to extract transient data before and after the fault, forming a standardized feature sample set. Specifically, a dynamic window of 200ms is set to extract transient data 50ms before and 150ms after the fault, thus forming a standardized sample set. Step 3: The transient signal is decomposed using an improved wavelet packet transform to extract the time-domain amplitude, phase, steepness, and mutation rate; the frequency-domain fundamental wave, harmonics, interharmonics, and spectral entropy; the time-frequency domain wavelet energy, wavelet entropy, and modal components; as well as topological correlation features and temporal correlation features, thereby constructing a high-dimensional fault feature matrix. Step 4: Using principal component analysis combined with an improved genetic algorithm, redundant features are eliminated, and the optimal feature subset that is strongly correlated with the fault type and location is selected; Step 5: Construct a hybrid model that integrates local spatial features, temporal correlation, and key feature weights. The optimal feature subset from Step 4 is used as input, and the classified fault is used as output. This enables accurate location and intelligent identification of fault types (short circuit, open circuit, high-resistance grounding, arc grounding), fault phases, and fault sections. Then, the identified faults are classified into four levels of hazard: minor distortion fault, general fault, severe short circuit fault, and destructive fault. Step 6: Based on the combination of cosine similarity and dynamic time warping, compare real-time features with the historical fault database to verify the analysis results. Then, construct a power grid topology map and combine the power grid topology and power flow direction to calculate the node fault contribution using graph convolutional networks, thereby tracing the fault origin, propagation path and impact range. Step 7: Output the fault type, location, time, characteristic parameters, and risk level, and generate the fault trend curve, topology location map, and characteristic spectrum map.

[0012] In its application, this invention first synchronously collects data on three-phase voltage, three-phase current, zero-sequence voltage, zero-sequence current, harmonic components, power, frequency, DC components, and switching status of all nodes in the power grid, as well as ambient temperature, humidity, wind speed, and light intensity. Based on this collected data, a spatiotemporal coupled fault dataset is constructed. In this process, the invention innovatively integrates four-dimensional data: AC-side electrical quantities, DC-side converter parameters, equipment operating status, and environmental meteorological parameters. This overcomes the limitations of traditional methods, which are only suitable for traditional synchronous machine power grids and cannot adapt to complex fault scenarios in new power systems, greatly expanding the applicability and versatility of fault analysis methods. Then, the collected dataset is subjected to noise reduction, normalization, and other processing. The system employs missing value completion and outlier removal to address issues such as electromagnetic interference, data loss, abrupt changes, inconsistencies in measurement units, and timing misalignment, significantly improving the quality of sample data and ensuring the accuracy and effectiveness of subsequent fault feature extraction. Next, time alignment of all measurement point data is achieved based on a network-wide time synchronization protocol, eliminating time differences between multiple devices. A dynamic sliding window is used to extract transient data before and after a fault, forming a standardized feature sample set. Finally, an improved wavelet packet transform is used to decompose the transient signal, extracting time-domain amplitude, phase, steepness, and abrupt change rate; frequency-domain fundamental, harmonic, interharmonic, and spectral entropy; time-frequency domain wavelet energy, wavelet entropy, and modal components; as well as topological and temporal correlation features. This process constructs a high-dimensional fault feature matrix. Finally, principal component analysis combined with an improved genetic algorithm is used to eliminate redundant features and select the optimal feature subset strongly correlated with fault type and location. Subsequently, a hybrid model integrating local spatial features, temporal correlations, and key feature weights is constructed. Using the optimal feature subset as input and fault classification as output, this model achieves accurate location and intelligent identification of fault types (short circuit, open circuit, high-resistance grounding, arc grounding), fault phases, and fault sections. After completing these operations, the analysis results are verified by comparing real-time features with a historical fault database using a combination of cosine similarity and dynamic time warping. Finally, a power grid topology map is constructed, and combined with the power grid topology and power flow direction, the model is further refined. Graph convolutional networks are used to calculate the fault contribution of nodes, thereby tracing the origin, propagation path, and impact range of faults. In this process, this invention achieves secondary verification of fault features by combining cosine similarity and dynamic time warping, avoiding model misjudgment. At the same time, by introducing graph convolutional networks and combining the power grid topology and power flow distribution to calculate the fault contribution of each node, the fault source is accurately located, the fault propagation path is sorted out, and the fault impact range is quantified. This realizes the upgrade from "simple fault identification" to a full-chain fault analysis capability of "identification-verification-source tracing-assessment". Finally, the fault type, location, time, feature parameters, and risk level are output, and fault trend curves, topology location maps, and feature spectrum maps are generated.

[0013] In summary, this invention overcomes the shortcomings of traditional fault analysis methods, which rely on power frequency steady state, have single characteristics, poor anti-interference capabilities, lack of traceability, and weak adaptability. Targeting typical operating conditions of new power systems characterized by mixed disturbances from new energy sources, weak fault characteristics, complex transient distortions, and severe AC / DC coupling, it achieves a high-precision, highly sensitive, real-time, traceable, and self-updating intelligent fault analysis system. This effectively reduces the probability of power grid protection maloperation and failure to operate, and improves the operational stability and fault handling efficiency of new power systems. Overall, this invention has the advantages of being comprehensive, efficient, accurate, and highly adaptable.

[0014] The above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Those skilled in the art should understand that any modifications or equivalent substitutions to the specific embodiments of the present invention without departing 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 the characteristics of power grid faults in a power system, characterized in that, Includes the following steps: Step 1: Synchronously collect three-phase voltage, three-phase current, zero-sequence voltage, zero-sequence current, harmonic components, power, frequency, DC components, and switch status of all nodes in the power grid, as well as ambient temperature, humidity, wind speed, and light intensity data, and construct a spatiotemporal coupled fault dataset based on the collected data. Step 2: The dataset collected in Step 1 is subjected to noise reduction, normalization, missing value completion, and outlier removal. Then, the time sequence alignment of all measurement point data is completed based on the network-wide time synchronization protocol to eliminate the time difference of multi-device acquisition. Finally, a dynamic sliding window is used to extract transient data before and after the fault to form a standardized feature sample set. Step 3: The transient signal is decomposed using an improved wavelet packet transform to extract the time-domain amplitude, phase, steepness, and mutation rate; the frequency-domain fundamental wave, harmonics, interharmonics, and spectral entropy; the time-frequency domain wavelet energy, wavelet entropy, and modal components; as well as topological correlation features and temporal correlation features, thereby constructing a high-dimensional fault feature matrix. Step 4: Using principal component analysis combined with an improved genetic algorithm, redundant features are eliminated, and the optimal feature subset that is strongly correlated with the fault type and location is selected; Step 5: Construct a hybrid model that integrates local spatial features, temporal correlation, and key feature weights. The optimal feature subset from Step 4 is used as input, and the classified fault is used as output. This enables accurate location and intelligent identification of fault types (short circuit, open circuit, high-resistance grounding, arc grounding), fault phases, and fault sections. Then, the identified faults are classified by hazard level. Step 6: Based on the fault feature matching algorithm, compare the real-time features with the historical fault database to verify the analysis results. Then, construct a power grid topology map and combine the power grid topology and power flow direction to calculate the node fault contribution using graph convolutional networks, thereby tracing the fault origin, propagation path and impact range. Step 7: Output the fault type, location, time, characteristic parameters, and risk level, and generate the fault trend curve, topology location map, and characteristic spectrum map.

2. The novel power grid fault characteristic analysis method for a power system as described in claim 1, characterized in that: In step 2, a standardized sample set can be formed by setting a dynamic window with a length of 200ms to capture transient data 50ms before and 150ms after the fault.

3. The novel power grid fault characteristic analysis method for a power system as described in claim 1, characterized in that: The noise reduction process in step 2 specifically involves: using wavelet threshold noise reduction to suppress electromagnetic interference; the normalization process specifically involves: using minimum-maximum normalization to unify the dimensions; the missing value completion operation specifically involves: using linear interpolation to complete missing data; and the outlier removal operation specifically involves: removing outliers based on the Laida criterion.

4. The novel power grid fault characteristic analysis method for a power system as described in claim 1, characterized in that: In step 5, the fault severity levels are divided into four levels: minor distortion fault, general fault, severe short circuit fault, and destructive fault.

5. The power grid fault characteristic analysis method for a power system as described in claim 1, characterized in that: In step 6, the fault feature matching algorithm adopts a combination of cosine similarity and dynamic time warping.