Artificial intelligence analysis method and system applied to traveling wave fault location

Abstract

The application discloses an artificial intelligence analysis method and system applied to traveling wave fault location, relates to the technical field of power fault location, and collects traveling wave data and standing wave data through current and voltage sensors and carries out pretreatment; a convolutional neural network and a long short-term memory network hybrid model are combined with a time-space attention mechanism to deeply mine time-space features and correlation complementarity of two data modes; based on fused features, fault types are divided into typical faults and complex faults, and identification and location are respectively carried out; the typical faults utilize traveling wave time difference positioning, and standing wave verification; the complex faults analyze multiple wave heads and standing wave reflection characteristics to comprehensively locate; when other fault points exist near the fault point, the time sequence and spatial correlation between multiple fault points are analyzed; combined with fault type, location and correlation information, operation and maintenance personnel are assisted to make decisions and diagnoses.

Description

Technical Field

[0001] This invention relates to the field of power fault location technology, specifically to an artificial intelligence analysis method and system for traveling wave fault location. Background Technology

[0002] Transmission lines are an important component of the power system, and their safe and stable operation is crucial for normal power supply. However, various types of faults are inevitable in transmission lines during long-term operation.

[0003] When a fault occurs, a traveling wave is generated at the fault point. These traveling waves propagate along the line and are reflected and refracted at non-uniform points on the line, forming a standing wave. Traditional fault location techniques mainly focus on capturing and analyzing the initial traveling wave and consider establishing a correlation model with the generated standing wave and comprehensively analyzing the correlation and complementarity. Traditional techniques often only perform unified analysis and calculation for fault type and location, without making judgments based on different types. At the same time, they lack in-depth analysis of the grounding characteristics of traveling waves and the zero-sequence component in standing waves, resulting in insufficient identification and inaccurate results. When there are other fault points near the fault location, existing techniques do not consider the correlation between adjacent faults, affecting the efficiency of fault diagnosis and handling. Summary of the Invention

[0004] The purpose of this invention is to provide an artificial intelligence analysis method and system for traveling wave fault location, so as to solve the problems raised in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] In a first aspect, the present invention provides an artificial intelligence analysis method for traveling wave fault location, comprising:

[0007] Traveling wave data and standing wave data are collected using current sensors and voltage sensors, and then preprocessed.

[0008] By using a hybrid model combining convolutional neural networks and long short-term memory networks, spatiotemporal features are extracted from preprocessed traveling wave and standing wave data. A spatiotemporal attention mechanism is incorporated to explore the correlation and complementarity between the two modal features, thereby achieving deep information fusion.

[0009] Based on the fused features, a hybrid model is used for fault type identification and fault location; by combining the characteristics of traveling wave grounding and the identification of zero-sequence components in standing waves, the fault types of typical faults are obtained; by fusing and analyzing the information dimensions provided by the combination of features of traveling wave and standing wave performance, the fault types of complex faults are obtained.

[0010] In terms of fault location, for typical faults, we focus on the time difference between the arrival of the initial traveling wave front at both measurement points, and calculate the fault distance by combining the line parameters, with standing wave information as an auxiliary verification; for complex faults, we analyze the time interval and energy distribution of multiple wave fronts, and rely on the reflection characteristics of the fault point reflected in the standing wave information to comprehensively determine the fault location.

[0011] When multiple fault points are detected near the fault location, a hybrid model is used to perform correlation analysis to evaluate whether there is a temporal sequence and spatial propagation relationship between the fault signals, and to obtain the fault correlation.

[0012] Based on the fault type, fault location, and fault correlation, it assists maintenance personnel in making decisions and diagnosing problems.

[0013] In conjunction with the first aspect, in the first implementation of the first aspect of this application, the step of acquiring traveling wave data and standing wave data through a current sensor and a voltage sensor, and performing preprocessing, includes:

[0014] Deploy current and voltage sensors at key nodes of transmission lines to measure traveling wave transient current and voltage signals at key nodes, and collect voltage and current signals under steady-state and quasi-steady-state conditions after a fault.

[0015] The collected raw data is preprocessed, and data alignment and synchronization verification are performed. GPS timestamps are used to adjust the time alignment of data collected by different sensor nodes to synchronize the data in the time dimension. Digital filtering technology is used to denoise and filter the data to remove high-frequency noise and power frequency interference. Wavelet transform is used to analyze the traveling wave signal, extract the instantaneous amplitude, and highlight the key features and instantaneous changes in the signal that are related to the fault. Data loss and outliers are handled and corrected using the mean filling method.

[0016] In conjunction with the first aspect, in the second implementation of the first aspect of this application, the method of using a hybrid model combining convolutional neural networks and long short-term memory networks to extract spatiotemporal features from preprocessed traveling wave and standing wave data, incorporating a spatiotemporal attention mechanism, and mining the correlation and complementarity between the two modal features to achieve deep information fusion includes:

[0017] The hybrid model comprises a Convolutional Neural Network (CNN) branch and a Long Short-Term Memory (LSTM) branch. Preprocessed data is used as input to the hybrid model. The CNN branch captures the structured information of the signal within local regions. Convolutional layers operate on the time series of the signal using sliding filters, automatically learning and extracting local spatial features. Pooling layers reduce feature dimensionality and enhance the translation invariance of the hybrid model. The LSTM branch captures the evolutionary features of the signal over time, processing temporal dynamics in sequence data. It learns and memorizes dependencies and dynamic patterns in long-term sequences through gating mechanisms, including input gates, forget gates, and output gates.

[0018] After extracting spatial and temporal features from the two branches respectively, the outputs of the two branches are fused and input into the spatiotemporal attention mechanism module. The weights are dynamically calculated and assigned according to the importance of features to fault type identification and localization. The importance of features on the time axis and the spatial dimension are evaluated, so that the hybrid model can automatically focus on the most relevant spatiotemporal region, suppress noise and redundant information, and effectively explore the correlation and complementarity between traveling wave signals and standing wave signals.

[0019] In conjunction with the first aspect, in the third implementation of the first aspect of this application, the method of combining traveling wave grounding characteristics and zero-sequence component identification in standing waves to obtain typical fault types includes:

[0020] The features fused through a spatiotemporal attention mechanism are combined with a hybrid model to identify and locate fault types. In the fault type identification stage, the hybrid model focuses on analyzing the transient characteristics of traveling wave grounding and the zero-sequence component information in standing waves reflected in the fused features to distinguish typical fault types. Typical fault types include single-phase grounding, two-phase short-circuit grounding, two-phase short circuit, and three-phase short circuit.

[0021] When identifying a single-phase ground fault, the initial transient signal of the traveling wave generated by the single-phase ground fault is captured. This is manifested as the voltage of the faulty phase being lower than the preset single-phase ground fault phase voltage threshold, while the voltage of the non-faulty phase is higher than the preset single-phase ground fault non-faulty phase voltage threshold. At the same time, the faulty phase generates an initial traveling wave current with an amplitude exceeding the preset single-phase threshold. The hybrid model can identify this single-phase asymmetric initial disturbance characteristic, including the difference in amplitude and polarity of the traveling waves between the faulty and non-faulty phases. During the steady-state or quasi-steady-state period after the fault, the single-phase ground fault generates a zero-sequence current and zero-sequence voltage higher than the preset single-phase ground zero-sequence threshold. The hybrid model analyzes and fuses the spectral or time-domain information of the standing wave component in the characteristics to detect the existence and characteristics of the zero-sequence component. Since the zero-sequence component has a specific amplitude and mode during a single-phase ground fault, when the single-phase asymmetric disturbance of the initial traveling wave is detected, and the presence of a zero-sequence component conforming to the single-phase ground fault mode in the standing wave is confirmed, it can be determined as a single-phase ground fault.

[0022] When identifying a two-phase short-circuit-to-ground fault, the system captures the initial transient signal of a traveling wave where both phases are simultaneously short-circuited and at least one phase is grounded. The traveling wave signal is characterized by the voltages of the two faulted phases being lower than their respective preset two-phase short-circuit-to-ground fault phase voltage thresholds, while the voltages of the non-faulted phases are higher than or equal to the preset two-phase short-circuit-to-ground non-faulted phase voltage thresholds. The hybrid model analyzes the initial transient signal of the traveling wave, learns and identifies the specific characteristics of the phase relationship between the currents of the two faulted phases under this fault, which is characterized by the initial phase difference being within the preset two-phase phase difference threshold range. The two phases generate initial traveling wave currents with amplitudes exceeding the preset two-phase thresholds, and there is a specific phase relationship. The two-phase short-circuit-to-ground fault generates zero-sequence current and zero-sequence voltage, but their magnitude and characteristics are different from those of a single-phase ground fault. Here, the zero-sequence component is caused by the simultaneous grounding of both phases or by one phase being grounded and short-circuited with the other phase, and has identifiable amplitude and phase characteristics. When a two-phase asymmetrical disturbance of the traveling wave signal is detected, and it is confirmed that there is a zero-sequence component in the standing wave that conforms to the two-phase ground fault mode, it is determined to be a two-phase short-circuit-to-ground fault.

[0023] When identifying a two-phase short-circuit fault, the hybrid model analyzes the initial transient characteristics of the traveling wave signal, which are manifested as the voltage and current of the two involved phases being lower than or deviating from the preset normal range of two-phase short-circuit fault voltage and current, while the voltage and current of the non-faulty phases remain within the normal range. Since there is no grounding, when the hybrid model analyzes the fusion characteristics of the standing wave component, it will detect a zero-sequence component with an amplitude lower than or equal to the preset two-phase short-circuit fault zero-sequence threshold, which is also a key feature for differentiation. When the hybrid model detects the initial transient characteristics of two phases being disturbed simultaneously in the traveling wave signal and confirms that the amplitude of the standing wave zero-sequence component is lower than the preset two-phase short-circuit fault zero-sequence threshold, it can be determined as a two-phase short-circuit fault.

[0024] When identifying three-phase short-circuit faults, the hybrid model analyzes the initial transient characteristics of simultaneous short circuits in the three phases reflected in the fusion features. The traveling wave signal is characterized by three-phase voltages simultaneously being lower than their respective preset three-phase short-circuit fault voltage thresholds, and three-phase currents also deviating from their respective preset normal ranges for three-phase short-circuit fault currents. Furthermore, there is a specific phase relationship between the three phases, manifested as the initial time difference of the initial changes in the three-phase voltage and current being less than the preset three-phase time threshold, and the similarity of the changing waveforms being higher than the preset three-phase similarity threshold. The hybrid model analyzes the fusion features of the standing wave component and detects zero-sequence components with amplitudes lower than or equal to the preset three-phase short-circuit fault zero-sequence threshold. When the hybrid model detects the initial transient characteristics of simultaneous disturbance in the three phases in the traveling wave signal and confirms that the amplitude of the zero-sequence component in the standing wave is lower than the preset three-phase short-circuit fault zero-sequence threshold, it can be determined as a three-phase short-circuit fault.

[0025] In conjunction with the first aspect, in the fourth implementation of the first aspect of this application, the information dimension provided by the fusion analysis of the characteristic combination of traveling wave and standing wave performance is used to obtain the fault type of complex faults, including:

[0026] Complex fault types include high-resistance grounding faults and intermittent faults;

[0027] When identifying high-resistance grounding faults, the hybrid model analyzes the following characteristics: the amplitude of the initial transient signal of the traveling wave is lower than the preset high-resistance grounding fault traveling wave amplitude threshold, the waveform change exceeds the preset normal range of the high-resistance grounding traveling wave waveform, and the transient signal is accompanied by noise or oscillation with an amplitude lower than the preset noise threshold and a frequency higher than the preset high-frequency threshold. At the same time, in the standing wave analysis, the amplitude of the zero-sequence component is lower than that of a typical grounding fault, the rate of change with time is lower than the preset zero-sequence component change threshold, and the fluctuation exceeds the preset normal range of high-resistance grounding zero-sequence component fluctuation. The hybrid model determines a high-resistance grounding fault by analyzing this combination of characteristics: the traveling wave transient is lower than the preset high-resistance grounding fault traveling wave amplitude threshold, and the standing wave zero-sequence component amplitude is low and unstable.

[0028] When identifying intermittent faults, the hybrid model analyzes the time-varying nature of the traveling wave signal in the fusion features. This is manifested in the following: the interval between the appearance of the traveling wave transient signals of the fault phase voltage and current exceeds the preset minimum interval time, the interval between their disappearance exceeds the preset maximum disappearance time, the amplitude exceeds the preset intermittent fault amplitude range, the polarity change exceeds the preset polarity regular change range, and the arrival time interval stability is lower than the preset intermittent fault interval stability threshold. At the same time, in the standing wave analysis, the existence time of the zero-sequence component is lower than the preset intermittent fault zero-sequence component persistence time threshold. The hybrid model determines the fault as intermittent by capturing this combination of features where both the traveling wave transient and the standing wave zero-sequence component exhibit obvious intermittent characteristics and do not meet the preset continuity or stability.

[0029] In conjunction with the first aspect, in the fifth implementation of the first aspect of this application, regarding fault location, for typical faults, attention is paid to the time difference between the arrival of the initial traveling wavefront at both measurement points, and the fault distance is calculated in conjunction with line parameters. Standing wave information is used as auxiliary verification, including:

[0030] After identifying the fault type as a typical fault, the time difference T between the initial traveling wave front and the measurement points at both ends of the line, extracted from the traveling wave data, is used. d The formula for calculating the fault distance is:

[0031] ;

[0032] Where D is the fault distance, representing the distance from one end of the line to the fault point, c is the wave velocity, and T is the wave velocity. d The time difference between the arrival of the initial traveling wavefront at the measurement points at both ends of the line;

[0033] In this process, the steady-state features extracted from the standing wave data are used as auxiliary verification information. By comparing them with the expected standing wave reflection characteristics corresponding to the fault distance D calculated based on the traveling wave, the traveling wave location results are verified, the confidence of the location results is improved, the fault distance D is output, and the mapping relationship between the fault distance D and the fault location coordinates is established by combining the geographical or topological information of the line. The specific coordinates of the fault location are obtained by linear interpolation calculation.

[0034] In conjunction with the first aspect, in the sixth implementation of the first aspect of this application, the step of analyzing the time intervals and energy distribution of multiple wavefronts for complex faults, relying on the reflection characteristics of the fault point reflected in the standing wave information, and comprehensively determining the fault location includes:

[0035] When the fault type is identified as a complex fault, the hybrid model analyzes the time interval sequence of multiple traveling wave fronts arriving at the measurement point and the corresponding energy distribution characteristics extracted from the traveling wave data; it also analyzes the steady-state characteristics extracted from the standing wave data, including the evolution of the standing wave ratio and reflection coefficient with frequency and time.

[0036] Based on a comprehensive analysis of multiple wavefront time intervals, energy distributions, and standing wave reflection characteristics, a unified multidimensional feature vector is formed. The hybrid model establishes a mapping relationship between the fault location and the multidimensional feature vector. Using historical fault data, it learns the feature vectors generated by high-resistance grounding and intermittent faults at different locations. Through training, the hybrid model automatically learns and fits the nonlinear relationship between the feature vector and the fault location. The currently extracted feature vector is input into the trained hybrid model, and through calculation, the specific fault location coordinates are output. These coordinates are the final determined fault location coordinates, realizing the location of high-resistance grounding and intermittent faults.

[0037] In conjunction with the first aspect, in the seventh implementation of the first aspect of this application, when multiple fault points are detected near the fault location, a hybrid model is used to perform correlation analysis to evaluate whether there is a temporal sequence and spatial propagation relationship between the fault signals, thereby obtaining fault correlation, including:

[0038] When multiple fault points exist, the hybrid model analyzes the arrival time of each wavefront to identify the temporal relationship between different fault points and determine the propagation sequence of the fault signal.

[0039] The hybrid model utilizes a spatiotemporal attention mechanism to evaluate the spatial propagation characteristics of fault signals and determine whether there are spatial propagation relationships between various fault points, including signal reflection, refraction, and scattering phenomena.

[0040] In terms of time analysis, the hybrid model analyzes the time difference of the fault wavefront, infers the order of occurrence of fault points, and helps to locate the sequential relationship between multiple fault points; in terms of spatial analysis, the hybrid model combines the location of sensor nodes to evaluate the signal propagation path, fault distance and fault location, and further confirms the spatial correlation of different fault points.

[0041] Based on the correlation analysis of time and space dimensions, the hybrid model comprehensively judges the mutual influence between multiple fault points and obtains the fault correlation.

[0042] In conjunction with the first aspect, in the eighth implementation of the first aspect of this application, the step of assisting maintenance personnel in decision-making and diagnosis based on fault type, fault location, and fault correlation includes:

[0043] Based on the in-depth analysis of traveling wave and standing wave signals using a hybrid model, fault types are accurately identified, including single-phase grounding, two-phase short-circuit grounding, two-phase short circuit, three-phase short circuit, high-resistance grounding, and intermittent faults. By feeding back the identified fault types to maintenance personnel, the nature of the fault is provided, and corresponding handling measures are determined.

[0044] By combining fault type and signal propagation analysis, the hybrid model provides fault location information to determine the specific location of the fault; by analyzing the propagation time difference of the traveling wave signal, the zero-sequence component in the standing wave, and the wavefront arrival time, it helps maintenance personnel locate the fault point in the power line.

[0045] When multiple fault points exist, spatiotemporal correlation analysis is used to identify the temporal and spatial relationships between the fault points; hybrid model analysis is used to analyze the mutual influence between the fault points to determine whether the fault occurs alone or has a linkage effect, and the fault correlation is used to assist maintenance personnel in making decisions.

[0046] Based on fault type, location, and correlation, it provides decision support and maintenance strategies for operations and maintenance personnel.

[0047] Secondly, the present invention provides an artificial intelligence analysis system for traveling wave fault location, comprising:

[0048] Data acquisition and preprocessing module: includes a data acquisition unit and a data preprocessing unit; the data acquisition unit acquires traveling wave data and standing wave data through current sensors and voltage sensors; the data preprocessing unit performs alignment and synchronization, noise reduction filtering, feature extraction, and correction of missing and outlier values ​​on the data;

[0049] The deep feature extraction and fusion module includes a hybrid model structure unit, a feature extraction unit, and a feature fusion and attention mechanism unit. The hybrid model structure unit constructs a hybrid network structure comprising a Convolutional Neural Network (CNN) and a Long Short-Term Memory (LSTM) network, used to process spatial features and temporal series features, respectively. The feature extraction unit uses CNN branches to extract spatial structural features from the input data and LSTM branches to extract temporal evolution features. The feature fusion and attention mechanism unit fuses spatial and temporal features into spatiotemporal features and applies an attention mechanism to dynamically emphasize important features and suppress irrelevant information.

[0050] Fault type identification module: includes a typical fault identification unit and a complex fault identification unit; the typical fault identification unit, based on the features fused by the spatiotemporal attention mechanism, focuses on analyzing the transient characteristics of traveling wave grounding and the zero-sequence component information of standing wave, and identifies typical fault types including single-phase grounding, two-phase short-circuit grounding, two-phase short circuit, and three-phase short circuit; the complex fault identification unit, based on the features fused by the spatiotemporal attention mechanism, integrates and analyzes the characteristic combination of traveling wave and standing wave behavior to identify high-resistance grounding faults and intermittent faults.

[0051] The fault location module includes a typical fault location unit and a complex fault location unit. The typical fault location unit calculates the fault distance based on the traveling wave time difference, verifies it with standing wave data, and locates the location coordinates of typical faults by combining line information. The complex fault location unit integrates and analyzes the traveling wave multi-wavehead and standing wave steady-state characteristics, inputs a trained hybrid model, and outputs the location coordinates of complex faults.

[0052] The multi-fault correlation analysis module includes a time relationship analysis unit, a spatial relationship analysis unit, and a correlation comprehensive judgment unit. The time relationship analysis unit analyzes the arrival time of each wavefront to identify the temporal sequence between different fault points. The spatial relationship analysis unit utilizes spatiotemporal attention mechanisms and sensor location information to evaluate the spatial propagation characteristics of fault signals and determine whether there are spatial propagation or influence relationships between various fault points. The correlation comprehensive judgment unit integrates the analysis results from both time and spatial dimensions to determine the mutual influence between multiple fault points and draw conclusions about fault correlation.

[0053] The fault integration and decision support module includes a fault information integration unit and a decision support unit. The fault information integration unit integrates fault type, fault location, and correlation information obtained from the fault type identification module, fault location module, and multi-fault correlation analysis module. Based on the integrated fault information, the decision support unit provides maintenance personnel with complete fault information and proposes preliminary maintenance strategy suggestions.

[0054] Compared with the prior art, the beneficial effects of the present invention are:

[0055] 1. This invention employs a hybrid model of convolutional neural networks and long short-term memory networks, along with a spatiotemporal attention mechanism, to deeply integrate the spatiotemporal characteristics of traveling wave and standing wave data.

[0056] 2. Based on deep fusion features, fault type identification and localization are performed. Typical faults are identified by combining traveling wave time difference and standing wave zero sequence features. Localization depends on traveling wave time difference and is supplemented by standing wave verification. Complex faults are identified and located by fusion analysis of multi-wavehead and standing wave reflection characteristics.

[0057] 3. For scenarios with multiple fault points, a hybrid model is used to analyze the temporal sequence and spatial propagation correlation of fault signals. Attached Figure Description

[0058] Figure 1 This is a schematic diagram illustrating the steps of the artificial intelligence analysis method for traveling wave fault location according to the present invention;

[0059] Figure 2 This is a system structure diagram of the artificial intelligence analysis system for traveling wave fault location according to the present invention. Detailed Implementation

[0060] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0061] Example: Figures 1-2 As shown, the present invention provides a technical solution.

[0062] like Figure 1 The schematic diagram illustrates the steps of an artificial intelligence analysis method applied to traveling wave fault location. This invention provides an artificial intelligence analysis method for traveling wave fault location, comprising:

[0063] Step S100: Acquire traveling wave data and standing wave data using current and voltage sensors, and perform preprocessing;

[0064] Specifically, current and voltage sensors are deployed at key nodes of transmission lines to measure traveling wave transient current and voltage signals at key nodes, and to collect voltage and current signals under steady-state and quasi-steady-state conditions after a fault.

[0065] The collected raw data is preprocessed, and data alignment and synchronization verification are performed. GPS timestamps are used to adjust the time alignment of data collected by different sensor nodes to synchronize the data in the time dimension. Digital filtering technology is used to denoise and filter the data to remove high-frequency noise and power frequency interference. Wavelet transform is used to analyze the traveling wave signal, extract the instantaneous amplitude, and highlight the key features and instantaneous changes in the signal that are related to the fault. Data loss and outliers are handled and corrected using the mean filling method.

[0066] In one specific embodiment, traveling wave and standing wave data are collected using current and voltage sensors and preprocessed. Specifically, current and voltage sensors are deployed at critical nodes of the transmission line, at distances of 5 km, 15 km, and 25 km from the substation, to measure the transient current and voltage signals of the traveling wave at the critical nodes and to collect the voltage and current signals under steady-state and quasi-steady-state conditions after the fault. In a single line fault, the sensor at 5 km location collected an initial traveling wave current peak of approximately 10 kA and a voltage peak of approximately 200 kV, followed by a series of damped oscillating traveling wave signals within 5 milliseconds. Simultaneously, under steady-state conditions, the sensor recorded an effective current value of approximately 200 A and an effective voltage value of approximately 500 kV. The collected raw data is preprocessed, and data alignment and synchronization verification are performed. GPS timestamps are used to adjust the time alignment of data collected by different sensor nodes to synchronize the data in the time dimension. Digital filtering technology is used to denoise and filter the data, removing high-frequency noise above 1MHz and power frequency interference of 50Hz. The traveling wave signal is analyzed using the db4 wavelet basis function, with a decomposition layer of 5, and the instantaneous amplitude is extracted, with a 10kA peak detected at 5 milliseconds. If a node has missing data within a specific time window, the average current of 200A of normal nodes within that time window is used to fill the gap.

[0067] Step S200: Using a hybrid model combining convolutional neural networks and long short-term memory networks, spatiotemporal features are extracted from the preprocessed traveling wave and standing wave data. A spatiotemporal attention mechanism is incorporated to explore the correlation and complementarity between the two modal features, thereby achieving deep information fusion.

[0068] Specifically, the hybrid model comprises a Convolutional Neural Network (CNN) branch and a Long Short-Term Memory (LSTM) branch. Preprocessed data is used as input to the hybrid model. The CNN branch captures the structured information of the signal within local regions. Convolutional layers operate on the time series of the signal using sliding filters, automatically learning and extracting local spatial features. Pooling layers reduce feature dimensionality and enhance the translation invariance of the hybrid model. The LSTM branch captures the evolutionary features of the signal over time, processing temporal dynamics in sequence data. It learns and memorizes dependencies and dynamic patterns in long-term sequences through gating mechanisms, including input gates, forget gates, and output gates.

[0069] After extracting spatial and temporal features from the two branches respectively, the outputs of the two branches are fused and input into the spatiotemporal attention mechanism module. The weights are dynamically calculated and assigned according to the importance of features to fault type identification and localization. The importance of features on the time axis and the spatial dimension are evaluated, so that the hybrid model can automatically focus on the most relevant spatiotemporal region, suppress noise and redundant information, and effectively explore the correlation and complementarity between traveling wave signals and standing wave signals.

[0070] In one specific embodiment, the hybrid model comprises a Convolutional Neural Network (CNN) branch and a Long Short-Term Memory (LSTM) branch. Preprocessed data is used as input to the hybrid model. Each sample contains 1000 time steps and 8 features: 4 from traveling waves and 4 from standing waves. The CNN branch captures the structured information of the signal within local regions. Convolutional layers operate on the time series of the signal using a sliding filter of size 5 and a stride of 1, extracting peak features reflecting the instant of the fault occurrence. Max pooling with a pooling window size of 2 is used to reduce feature dimensionality and enhance the translation invariance of the hybrid model. The LSTM branch captures the evolutionary features of the signal over time, processing the temporal dynamics in the sequence data. Through its input, forget, and output gates, it learns and memorizes dependencies and dynamic patterns over 100 time steps. The number of LSTM units is set to 64.

[0071] After extracting spatial and temporal features from the two branches respectively, the outputs of the two branches are concatenated to obtain the fused feature dimension. The concatenated features are then input into the spatiotemporal attention mechanism module, which contains a query, key, and value generation network with shared weights. A fully connected layer maps the features to a 64-dimensional space. By calculating the dot product of the query and key and passing it through a softmax function, the attention weights for each time step and spatial location are obtained. This evaluates the importance of features on the time axis and the spatial dimension, enabling the hybrid model to automatically focus on the most relevant spatiotemporal regions, including the fault occurrence time and its subsequent time steps, as well as features corresponding to the region near the fault point. This suppresses noise and redundant information, effectively uncovering the correlation and complementarity between the initial traveling wavefront features and the steady-state reflected wave features.

[0072] Step S300: Based on the fused features, use a hybrid model to identify fault types and locate faults; combine traveling wave grounding features and zero-sequence component identification in standing waves to obtain the fault types of typical faults; fuse and analyze the information dimensions provided by the feature combination of traveling wave and standing wave performance to obtain the fault types of complex faults.

[0073] Specifically, the features fused through a spatiotemporal attention mechanism are combined with a hybrid model to identify and locate fault types. In the fault type identification stage, the hybrid model focuses on analyzing the transient characteristics of traveling wave grounding and the zero-sequence component information in standing waves reflected in the fused features to distinguish typical fault types. Typical fault types include single-phase grounding, two-phase short-circuit grounding, two-phase short circuit, and three-phase short circuit.

[0074] When identifying a single-phase ground fault, the initial transient signal of the traveling wave generated by the single-phase ground fault is captured. This is manifested as the voltage of the faulty phase being lower than the preset single-phase ground fault phase voltage threshold, while the voltage of the non-faulty phase is higher than the preset single-phase ground fault non-faulty phase voltage threshold. At the same time, the faulty phase generates an initial traveling wave current with an amplitude exceeding the preset single-phase threshold. The hybrid model can identify this single-phase asymmetric initial disturbance characteristic, including the difference in amplitude and polarity of the traveling waves between the faulty and non-faulty phases. During the steady-state or quasi-steady-state period after the fault, the single-phase ground fault generates a zero-sequence current and zero-sequence voltage higher than the preset single-phase ground zero-sequence threshold. The hybrid model analyzes and fuses the spectral or time-domain information of the standing wave component in the characteristics to detect the existence and characteristics of the zero-sequence component. Since the zero-sequence component has a specific amplitude and mode during a single-phase ground fault, when the single-phase asymmetric disturbance of the initial traveling wave is detected, and the presence of a zero-sequence component conforming to the single-phase ground fault mode in the standing wave is confirmed, it can be determined as a single-phase ground fault.

[0075] When identifying a two-phase short-circuit-to-ground fault, the system captures the initial transient signal of a traveling wave where both phases are simultaneously short-circuited and at least one phase is grounded. The traveling wave signal is characterized by the voltages of the two faulted phases being lower than their respective preset two-phase short-circuit-to-ground fault phase voltage thresholds, while the voltages of the non-faulted phases are higher than or equal to the preset two-phase short-circuit-to-ground non-faulted phase voltage thresholds. The hybrid model analyzes the initial transient signal of the traveling wave, learns and identifies the specific characteristics of the phase relationship between the currents of the two faulted phases under this fault, which is characterized by the initial phase difference being within the preset two-phase phase difference threshold range. The two phases generate initial traveling wave currents with amplitudes exceeding the preset two-phase thresholds, and there is a specific phase relationship. The two-phase short-circuit-to-ground fault generates zero-sequence current and zero-sequence voltage, but their magnitude and characteristics are different from those of a single-phase ground fault. Here, the zero-sequence component is caused by the simultaneous grounding of both phases or by one phase being grounded and short-circuited with the other phase, and has identifiable amplitude and phase characteristics. When a two-phase asymmetrical disturbance of the traveling wave signal is detected, and it is confirmed that there is a zero-sequence component in the standing wave that conforms to the two-phase ground fault mode, it is determined to be a two-phase short-circuit-to-ground fault.

[0076] When identifying a two-phase short-circuit fault, the hybrid model analyzes the initial transient characteristics of the traveling wave signal, which are manifested as the voltage and current of the two involved phases being lower than or deviating from the preset normal range of two-phase short-circuit fault voltage and current, while the voltage and current of the non-faulty phases remain within the normal range. Since there is no grounding, when the hybrid model analyzes the fusion characteristics of the standing wave component, it will detect a zero-sequence component with an amplitude lower than or equal to the preset two-phase short-circuit fault zero-sequence threshold, which is also a key feature for differentiation. When the hybrid model detects the initial transient characteristics of two phases being disturbed simultaneously in the traveling wave signal and confirms that the amplitude of the standing wave zero-sequence component is lower than the preset two-phase short-circuit fault zero-sequence threshold, it can be determined as a two-phase short-circuit fault.

[0077] When identifying three-phase short-circuit faults, the hybrid model analyzes the initial transient characteristics of simultaneous short circuits in the three phases reflected in the fusion features. The traveling wave signal is characterized by three-phase voltages simultaneously being lower than their respective preset three-phase short-circuit fault voltage thresholds, and three-phase currents also deviating from their respective preset normal ranges for three-phase short-circuit fault currents. Furthermore, there is a specific phase relationship between the three phases, manifested as the initial time difference of the initial changes in the three-phase voltage and current being less than the preset three-phase time threshold, and the similarity of the changing waveforms being higher than the preset three-phase similarity threshold. The hybrid model analyzes the fusion features of the standing wave component and detects zero-sequence components with amplitudes lower than or equal to the preset three-phase short-circuit fault zero-sequence threshold. When the hybrid model detects the initial transient characteristics of simultaneous disturbance in the three phases in the traveling wave signal and confirms that the amplitude of the zero-sequence component in the standing wave is lower than the preset three-phase short-circuit fault zero-sequence threshold, it can be determined as a three-phase short-circuit fault.

[0078] Complex fault types include high-resistance grounding faults and intermittent faults;

[0079] When identifying high-resistance grounding faults, the hybrid model analyzes the following characteristics: the amplitude of the initial transient signal of the traveling wave is lower than the preset high-resistance grounding fault traveling wave amplitude threshold, the waveform change exceeds the preset normal range of the high-resistance grounding traveling wave waveform, and the transient signal is accompanied by noise or oscillation with an amplitude lower than the preset noise threshold and a frequency higher than the preset high-frequency threshold. At the same time, in the standing wave analysis, the amplitude of the zero-sequence component is lower than that of a typical grounding fault, the rate of change with time is lower than the preset zero-sequence component change threshold, and the fluctuation exceeds the preset normal range of high-resistance grounding zero-sequence component fluctuation. The hybrid model determines a high-resistance grounding fault by analyzing this combination of characteristics: the traveling wave transient is lower than the preset high-resistance grounding fault traveling wave amplitude threshold, and the standing wave zero-sequence component amplitude is low and unstable.

[0080] When identifying intermittent faults, the hybrid model analyzes the time-varying nature of the traveling wave signal in the fusion features. This is manifested in the following: the interval between the appearance of the traveling wave transient signals of the fault phase voltage and current exceeds the preset minimum interval time, the interval between their disappearance exceeds the preset maximum disappearance time, the amplitude exceeds the preset intermittent fault amplitude range, the polarity change exceeds the preset polarity regular change range, and the arrival time interval stability is lower than the preset intermittent fault interval stability threshold. At the same time, in the standing wave analysis, the existence time of the zero-sequence component is lower than the preset intermittent fault zero-sequence component persistence time threshold. The hybrid model determines the fault as intermittent by capturing this combination of features where both the traveling wave transient and the standing wave zero-sequence component exhibit obvious intermittent characteristics and do not meet the preset continuity or stability.

[0081] In one specific embodiment, the hybrid model first analyzes the initial transient signal of the traveling wave. The transient signal amplitude of 4.5kA is significantly higher than the preset ground fault traveling wave amplitude threshold of 2kA. Its waveform includes a main spike with an amplitude exceeding the preset noise threshold of 100A and a frequency exceeding the preset high-frequency threshold of 5kHz. This spike manifests as a pulse with a frequency of approximately 10kHz and an amplitude between 3kA in the time domain waveform. Simultaneously, in the standing wave analysis, the hybrid model observes that the maximum amplitude reaches 250A, far exceeding the preset typical ground fault zero-sequence component threshold of 200A; its rate of change over time is also higher than the preset zero-sequence component change threshold of 20A / s, with an actual rate of change as high as 35A / s; furthermore, during the fault duration of 0.1 to 0.5 seconds, the zero-sequence current fluctuates relatively stably between 200A and 300A, with the fluctuation amplitude not exceeding the preset 50A. The hybrid model, through comprehensive analysis of this unique combination of characteristics—namely, the high transient amplitude and dramatic waveform of the traveling wave, and the high amplitude, high rate of change, and moderate fluctuation of the zero-sequence component of the standing wave—determines the fault type as a single-phase ground fault.

[0082] The hybrid model first analyzes the initial transient signal of the traveling wave. The amplitude of 800A is lower than the preset high-resistivity ground fault traveling wave amplitude threshold of 1kA. Its waveform contains multiple noise or oscillatory components with amplitudes exceeding the preset noise threshold of 50A but frequencies exceeding the preset high-frequency threshold of 1kHz. These components manifest as spikes in the time-domain waveform with frequencies around 2kHz and amplitudes fluctuating between ±30A. Simultaneously, in the standing wave analysis, the hybrid model observes that the maximum amplitude is only 100A, far below the preset typical ground fault zero-sequence component threshold of 200A; its rate of change over time, 5A / s, is also lower than the preset zero-sequence component change threshold of 10A / s. Furthermore, this zero-sequence component fluctuation occurs irregularly between 70A and 130A during the fault duration of 0.1 to 0.3 seconds, with fluctuation amplitudes exceeding the preset 20A. By comprehensively analyzing this unique combination of characteristics—low transient wave amplitude and abnormal waveform, and low amplitude, low rate of change, and large fluctuations in the standing wave zero-sequence component—the hybrid model determines the fault type to be a high-resistivity ground fault.

[0083] Step S400: In terms of fault location, for typical faults, focus on the time difference between the arrival of the initial traveling wave front at both measurement points, calculate the fault distance in combination with line parameters, and use standing wave information as an auxiliary verification; for complex faults, analyze the time interval and energy distribution of multiple wave fronts, rely on the reflection characteristics of the fault point reflected in the standing wave information, and comprehensively judge the fault location.

[0084] Specifically, after identifying the fault type as a typical fault, the time difference T between the initial traveling wave front and the measurement points at both ends of the line, extracted from the traveling wave data, is used. d The formula for calculating the fault distance is:

[0085] ;

[0086] Where D is the fault distance, representing the distance from one end of the line to the fault point, c is the wave velocity, and T is the wave velocity. d The time difference between the arrival of the initial traveling wavefront at the measurement points at both ends of the line;

[0087] In this process, the steady-state features extracted from the standing wave data are used as auxiliary verification information. By comparing them with the expected standing wave reflection characteristics corresponding to the fault distance D calculated based on the traveling wave, the traveling wave location results are verified, the confidence of the location results is improved, the fault distance D is output, and the mapping relationship between the fault distance D and the fault location coordinates is established by combining the geographical or topological information of the line. The specific coordinates of the fault location are obtained by linear interpolation calculation.

[0088] When the fault type is identified as a complex fault, the hybrid model analyzes the time interval sequence of multiple traveling wave fronts arriving at the measurement point and the corresponding energy distribution characteristics extracted from the traveling wave data; it also analyzes the steady-state characteristics extracted from the standing wave data, including the evolution of the standing wave ratio and reflection coefficient with frequency and time.

[0089] Based on a comprehensive analysis of multiple wavefront time intervals, energy distributions, and standing wave reflection characteristics, a unified multidimensional feature vector is formed. The hybrid model establishes a mapping relationship between the fault location and the multidimensional feature vector. Using historical fault data, it learns the feature vectors generated by high-resistance grounding and intermittent faults at different locations. Through training, the hybrid model automatically learns and fits the nonlinear relationship between the feature vector and the fault location. The currently extracted feature vector is input into the trained hybrid model, and through calculation, the specific fault location coordinates are output. These coordinates are the final determined fault location coordinates, realizing the location of high-resistance grounding and intermittent faults.

[0090] In one specific embodiment, after identifying the fault type as single-phase grounding, the time difference T between the initial traveling wave front extracted from the traveling wave data and the measurement points at both ends A and B of the line is used. d With a time interval of 2.4 ms and a line wave velocity c of 285 m / s, the calculated fault distance D is 342 m. During this process, steady-state features extracted from the standing wave data indicate that the expected standing wave ratio at a fault distance of 342 m is greater than 1.5, and the amplitude of the reflection coefficient is greater than 0.2. These features are used as auxiliary verification information. The actual extracted standing wave features are compared with the expected standing wave reflection characteristics corresponding to the fault distance of 342 m calculated based on the traveling wave. The results show a good match, thus verifying the accuracy of the traveling wave location result and improving the confidence level of the location result. After outputting the fault distance D = 342 m, and considering the known total line length of 10 km, the starting point coordinates (0, 0), and the ending point coordinates (10 km, 0), a mapping relationship between the fault distance D and the fault location coordinates is established. Through linear interpolation, the fault point is located 3.42 km from the starting point, with coordinates (3.42 km, 0).

[0091] When the fault type is identified as a high-resistance grounding fault, the hybrid model analyzes the time interval sequence of multiple traveling wave fronts arriving at the measurement point extracted from the traveling wave data and the corresponding energy distribution characteristics. A second wave front appears 0.8 ms after the first, with 30% of the energy of the first; a third wave front appears 1.2 ms after the second, with 20% of the energy of the first. Steady-state characteristics extracted from the standing wave data are analyzed, showing that the reflection coefficient amplitude reaches a peak of 0.35 at 100 Hz and fluctuates periodically over time. Based on a comprehensive analysis of the time intervals, energy distribution, and standing wave reflection characteristics of multiple wave fronts, a unified multi-dimensional feature vector is formed. The hybrid model uses historical fault data to learn the feature vectors generated by high-resistance grounding and intermittent faults at different locations. Through training, it automatically learns and fits the nonlinear relationship between the feature vector and the fault location. The currently extracted feature vector is input into the trained hybrid model, and the coordinates (1.2 km, 3.5 km) are calculated. These coordinates are the final determined fault location coordinates.

[0092] Step S500: When multiple fault points are detected near the fault location, a hybrid model is used to perform correlation analysis to evaluate whether there is a temporal sequence and spatial propagation relationship between the fault signals, and to obtain the fault correlation.

[0093] Specifically, when multiple fault points exist, the hybrid model analyzes the arrival time of each wavefront to identify the temporal relationship between different fault points and determine the propagation sequence of the fault signal.

[0094] The hybrid model utilizes a spatiotemporal attention mechanism to evaluate the spatial propagation characteristics of fault signals and determine whether there are spatial propagation relationships between various fault points, including signal reflection, refraction, and scattering phenomena.

[0095] In terms of time analysis, the hybrid model analyzes the time difference of the fault wavefront, infers the order of occurrence of fault points, and helps to locate the sequential relationship between multiple fault points; in terms of spatial analysis, the hybrid model combines the location of sensor nodes to evaluate the signal propagation path, fault distance and fault location, and further confirms the spatial correlation of different fault points.

[0096] Based on the correlation analysis of time and space dimensions, the hybrid model comprehensively judges the mutual influence between multiple fault points and obtains the fault correlation.

[0097] In one specific embodiment, in a complex scenario containing three fault points F1, F2, and F3, sensor A detects the first wavefront at timestamp T0 = 0.5 ms, sensor B detects the second wavefront at T1 = 1.2 ms, and sensor C detects the third wavefront at T2 = 2.0 ms. The hybrid model analyzes these arrival times and infers the possible sequence of fault occurrence: F1 occurs first, its traveling wave propagating to A, B, and C; then F2 occurs, its wavefront detected by B and C; and finally F3 occurs, its wavefront primarily detected by C. The hybrid model utilizes a spatiotemporal attention mechanism to evaluate the spatial propagation characteristics of the fault signal, determining whether there is a spatial propagation relationship between the various fault points, including signal reflection, refraction, and scattering phenomena. In terms of time analysis, the hybrid model analyzes the time differences of the fault wavefronts: T1-T0 = 0.7 ms, T2-T1 = 0.8 ms. Combining this with the line length and wave velocity c = 285 m / s, it infers the occurrence sequence and relative distance of the fault points, helping to locate the sequential relationship between multiple fault points. In terms of spatial analysis, the hybrid model, combined with the known locations of sensor nodes A (0,0), B (5km,0), and C (10km,0), determines that F1 is closest to point A, F2 is near point B, and F3 is near point C. Furthermore, the signal of F2 may be affected by the reflected wave from F1. The model then determines whether the occurrence of F2 is related to the continued existence of F1, or whether F3 is a secondary fault caused by F1 and F2. Ultimately, it concludes that "F1 is the primary fault, and F2 and F3 are secondary faults affected by it."

[0098] Step S600: Based on the fault type, fault location, and fault correlation, assist maintenance personnel in making decisions and diagnosing.

[0099] Specifically, based on the in-depth analysis of traveling wave and standing wave signals using a hybrid model, fault types are accurately identified, including single-phase grounding, two-phase short-circuit grounding, two-phase short circuit, three-phase short circuit, high-resistance grounding, and intermittent faults. By feeding back the identified fault types to maintenance personnel, the nature of the fault is provided, and corresponding handling measures are determined.

[0100] By combining fault type and signal propagation analysis, the hybrid model provides fault location information to determine the specific location of the fault; by analyzing the propagation time difference of the traveling wave signal, the zero-sequence component in the standing wave, and the wavefront arrival time, it helps maintenance personnel locate the fault point in the power line.

[0101] When multiple fault points exist, spatiotemporal correlation analysis is used to identify the temporal and spatial relationships between the fault points; hybrid model analysis is used to analyze the mutual influence between the fault points to determine whether the fault occurs alone or has a linkage effect, and the fault correlation is used to assist maintenance personnel in making decisions.

[0102] Based on fault type, location, and correlation, it provides decision support and maintenance strategies for operations and maintenance personnel.

[0103] In one specific embodiment, after analyzing the current signal characteristics, the model determines the fault type to be a high-impedance grounding fault. For high-impedance grounding faults, maintenance personnel will prioritize using a high-impedance grounding detector for verification. The model calculates the fault distance to substation A to be 3.42 kilometers, with a location accuracy of ±50 meters. This is achieved by analyzing the propagation time difference T of the traveling wave signal. d =2.4ms, zero-sequence component amplitude of 0.3pu in the standing wave, and wavefront arrival time help maintenance personnel locate fault points in power lines. When multiple fault points exist, the correlation between faults assists maintenance personnel in making decisions, determining whether the occurrence of F2 may be affected by the expansion of insulation damage caused by F1. Based on the fault type of high-resistance grounding, location (3.42km, 0), (7.89km, 0), and correlation that F2 may be caused by F1, decision support and maintenance strategies are provided for maintenance personnel. It is recommended that maintenance personnel prioritize handling F1 and check whether there are potential hazards in the vicinity of F1 that may lead to F2, so as to formulate emergency repair plans more efficiently and accurately.

[0104] like Figure 2 The system architecture diagram of the artificial intelligence analysis system for traveling wave fault location is shown in the present invention. The present invention provides an artificial intelligence analysis system for traveling wave fault location, comprising:

[0105] Data acquisition and preprocessing module: includes a data acquisition unit and a data preprocessing unit; the data acquisition unit acquires traveling wave data and standing wave data through current sensors and voltage sensors; the data preprocessing unit performs alignment and synchronization, noise reduction filtering, feature extraction, and correction of missing and outlier values ​​on the data;

[0106] The deep feature extraction and fusion module includes a hybrid model structure unit, a feature extraction unit, and a feature fusion and attention mechanism unit. The hybrid model structure unit constructs a hybrid network structure comprising a Convolutional Neural Network (CNN) and a Long Short-Term Memory (LSTM) network, used to process spatial features and temporal series features, respectively. The feature extraction unit uses CNN branches to extract spatial structural features from the input data and LSTM branches to extract temporal evolution features. The feature fusion and attention mechanism unit fuses spatial and temporal features into spatiotemporal features and applies an attention mechanism to dynamically emphasize important features and suppress irrelevant information.

[0107] Fault type identification module: includes a typical fault identification unit and a complex fault identification unit; the typical fault identification unit, based on the features fused by the spatiotemporal attention mechanism, focuses on analyzing the transient characteristics of traveling wave grounding and the zero-sequence component information of standing wave, and identifies typical fault types including single-phase grounding, two-phase short-circuit grounding, two-phase short circuit, and three-phase short circuit; the complex fault identification unit, based on the features fused by the spatiotemporal attention mechanism, integrates and analyzes the characteristic combination of traveling wave and standing wave behavior to identify high-resistance grounding faults and intermittent faults.

[0108] The fault location module includes a typical fault location unit and a complex fault location unit. The typical fault location unit calculates the fault distance based on the traveling wave time difference, verifies it with standing wave data, and locates the location coordinates of typical faults by combining line information. The complex fault location unit integrates and analyzes the traveling wave multi-wavehead and standing wave steady-state characteristics, inputs a trained hybrid model, and outputs the location coordinates of complex faults.

[0109] The multi-fault correlation analysis module includes a time relationship analysis unit, a spatial relationship analysis unit, and a correlation comprehensive judgment unit. The time relationship analysis unit analyzes the arrival time of each wavefront to identify the temporal sequence between different fault points. The spatial relationship analysis unit utilizes spatiotemporal attention mechanisms and sensor location information to evaluate the spatial propagation characteristics of fault signals and determine whether there are spatial propagation or influence relationships between various fault points. The correlation comprehensive judgment unit integrates the analysis results from both time and spatial dimensions to determine the mutual influence between multiple fault points and draw conclusions about fault correlation.

[0110] The fault integration and decision support module includes a fault information integration unit and a decision support unit. The fault information integration unit integrates fault type, fault location, and correlation information obtained from the fault type identification module, fault location module, and multi-fault correlation analysis module. Based on the integrated fault information, the decision support unit provides maintenance personnel with complete fault information and proposes preliminary maintenance strategy suggestions.

[0111] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. An artificial intelligence analysis method applied to traveling wave fault location, characterized in that, include: Traveling wave data and standing wave data are collected using current sensors and voltage sensors, and then preprocessed. By using a hybrid model combining convolutional neural networks and long short-term memory networks, spatiotemporal features are extracted from preprocessed traveling wave and standing wave data. A spatiotemporal attention mechanism is incorporated to explore the correlation and complementarity between traveling wave signals and standing wave signals, thereby achieving deep information fusion. Based on the fused features, a hybrid model is used for fault type identification and fault location; by combining the characteristics of traveling wave grounding and the identification of zero-sequence components in standing waves, the fault types of typical faults are obtained; by fusing and analyzing the information dimensions provided by the combination of features of traveling wave and standing wave performance, the fault types of complex faults are obtained. In terms of fault location, for typical faults, we focus on the time difference between the arrival of the initial traveling wave front at both measurement points, and calculate the fault distance by combining the line parameters, with standing wave information used as an auxiliary verification. For complex faults, the time intervals and energy distribution of multiple wavefronts are analyzed, and the fault location is determined by comprehensively considering the reflection characteristics of the fault point reflected in the standing wave information. When multiple fault points are detected near the fault location, a hybrid model is used to perform correlation analysis to evaluate whether there is a temporal sequence and spatial propagation relationship between the fault signals, and to obtain the fault correlation. Based on the fault type, fault location, and fault correlation, it assists maintenance personnel in making decisions and diagnosing problems.

2. The artificial intelligence analysis method for traveling wave fault location according to claim 1, characterized in that, The process of acquiring traveling wave and standing wave data through current and voltage sensors, and performing preprocessing, includes: Deploy current and voltage sensors at key nodes of transmission lines to measure traveling wave transient current and voltage signals at key nodes, and collect voltage and current signals under steady-state and quasi-steady-state conditions after a fault. The collected raw data is preprocessed, and data alignment and synchronization verification are performed. GPS timestamps are used to adjust the time alignment of data collected by different sensor nodes to synchronize the data in the time dimension. Digital filtering technology is used to denoise and filter the data to remove high-frequency noise and power frequency interference. Wavelet transform is used to analyze the traveling wave signal, extract the instantaneous amplitude, and highlight the key features and instantaneous changes in the signal that are related to the fault. Data loss and outliers are handled and corrected using the mean filling method.

3. The artificial intelligence analysis method for traveling wave fault location according to claim 1, characterized in that, The method utilizes a hybrid model combining convolutional neural networks and long short-term memory networks to extract spatiotemporal features from preprocessed traveling and standing wave data. It incorporates a spatiotemporal attention mechanism to explore the correlation and complementarity between traveling and standing wave signals, achieving deep information fusion. This includes: The hybrid model comprises a Convolutional Neural Network (CNN) branch and a Long Short-Term Memory (LSTM) branch. Preprocessed data is used as input to the hybrid model. The CNN branch captures the structured information of the signal within local regions. Convolutional layers operate on the time series of the signal using sliding filters, automatically learning and extracting local spatial features. Pooling layers reduce feature dimensionality and enhance the translation invariance of the hybrid model. The LSTM branch captures the evolutionary features of the signal over time, processing temporal dynamics in sequence data. It learns and memorizes dependencies and dynamic patterns in long-term sequences through gating mechanisms, including input gates, forget gates, and output gates. After extracting spatial and temporal features from the two branches respectively, the outputs of the two branches are fused and input into the spatiotemporal attention mechanism module. The weights are dynamically calculated and assigned according to the importance of features to fault type identification and localization. The importance of features on the time axis and the spatial dimension are evaluated, so that the hybrid model can automatically focus on the most relevant spatiotemporal region, suppress noise and redundant information, and effectively explore the correlation and complementarity between traveling wave signals and standing wave signals.

4. The artificial intelligence analysis method for traveling wave fault location according to claim 1, characterized in that, The combination of traveling wave grounding characteristics and zero-sequence component identification in standing waves yields typical fault types, including: The features fused through a spatiotemporal attention mechanism are combined with a hybrid model to identify and locate fault types. In the fault type identification stage, the hybrid model focuses on analyzing the transient characteristics of traveling wave grounding and the zero-sequence component information in standing waves reflected in the fused features to distinguish typical fault types. Typical fault types include single-phase grounding, two-phase short-circuit grounding, two-phase short circuit, and three-phase short circuit. When identifying a single-phase ground fault, the initial transient signal of the traveling wave generated by the single-phase ground fault is captured. This is manifested as the voltage of the faulty phase being lower than the preset single-phase ground fault phase voltage threshold, while the voltage of the non-faulty phase is higher than the preset single-phase ground fault non-faulty phase voltage threshold. At the same time, the faulty phase generates an initial traveling wave current with an amplitude exceeding the preset single-phase threshold. The hybrid model can identify this single-phase asymmetric initial disturbance characteristic, including the difference in amplitude and polarity of the traveling waves between the faulty and non-faulty phases. During the steady-state or quasi-steady-state period after the fault, the single-phase ground fault generates a zero-sequence current and zero-sequence voltage higher than the preset single-phase ground zero-sequence threshold. The hybrid model analyzes and fuses the spectral or time-domain information of the standing wave component in the characteristics to detect the existence and characteristics of the zero-sequence component. Since the zero-sequence component has a specific amplitude and mode during a single-phase ground fault, when the single-phase asymmetric disturbance of the initial traveling wave is detected, and the presence of a zero-sequence component conforming to the single-phase ground fault mode in the standing wave is confirmed, it can be determined as a single-phase ground fault. When identifying a two-phase short-circuit-to-ground fault, the system captures the initial transient signal of a traveling wave where both phases are simultaneously short-circuited and at least one phase is grounded. The traveling wave signal is characterized by the voltages of the two faulted phases being lower than their respective preset two-phase short-circuit-to-ground fault phase voltage thresholds, while the voltages of the non-faulted phases are higher than or equal to the preset two-phase short-circuit-to-ground non-faulted phase voltage thresholds. The hybrid model analyzes the initial transient signal of the traveling wave, learns and identifies the specific characteristics of the phase relationship between the currents of the two faulted phases under this fault, which is characterized by the initial phase difference being within the preset two-phase phase difference threshold range. The two phases generate initial traveling wave currents with amplitudes exceeding the preset two-phase thresholds, and there is a specific phase relationship. The two-phase short-circuit-to-ground fault generates zero-sequence current and zero-sequence voltage, but their magnitude and characteristics are different from those of a single-phase ground fault. Here, the zero-sequence component is caused by the simultaneous grounding of both phases or by one phase being grounded and short-circuited with the other phase, and has identifiable amplitude and phase characteristics. When a two-phase asymmetrical disturbance of the traveling wave signal is detected, and it is confirmed that there is a zero-sequence component in the standing wave that conforms to the two-phase ground fault mode, it is determined to be a two-phase short-circuit-to-ground fault. When identifying a two-phase short-circuit fault, the hybrid model analyzes the initial transient characteristics of the traveling wave signal, which are manifested as the voltage and current of the two involved phases being lower than or deviating from the preset normal range of two-phase short-circuit fault voltage and current, while the voltage and current of the non-faulty phases remain within the normal range. Since there is no grounding, when the hybrid model analyzes the fusion characteristics of the standing wave component, it will detect a zero-sequence component with an amplitude lower than or equal to the preset two-phase short-circuit fault zero-sequence threshold, which is also a key feature for differentiation. When the hybrid model detects the initial transient characteristics of two phases being disturbed simultaneously in the traveling wave signal and confirms that the amplitude of the standing wave zero-sequence component is lower than the preset two-phase short-circuit fault zero-sequence threshold, it can be determined as a two-phase short-circuit fault. When identifying three-phase short-circuit faults, the hybrid model analyzes the initial transient characteristics of simultaneous short circuits in the three phases reflected in the fusion features. The traveling wave signal is characterized by three-phase voltages simultaneously being lower than their respective preset three-phase short-circuit fault voltage thresholds, and three-phase currents also deviating from their respective preset normal ranges for three-phase short-circuit fault currents. Furthermore, there is a specific phase relationship between the three phases, manifested as the initial time difference of the initial changes in the three-phase voltage and current being less than the preset three-phase time threshold, and the similarity of the changing waveforms being higher than the preset three-phase similarity threshold. The hybrid model analyzes the fusion features of the standing wave component and detects zero-sequence components with amplitudes lower than or equal to the preset three-phase short-circuit fault zero-sequence threshold. When the hybrid model detects the initial transient characteristics of simultaneous disturbance in the three phases in the traveling wave signal and confirms that the amplitude of the zero-sequence component in the standing wave is lower than the preset three-phase short-circuit fault zero-sequence threshold, it can be determined as a three-phase short-circuit fault.

5. The artificial intelligence analysis method for traveling wave fault location according to claim 1, characterized in that, The information dimensions provided by the fusion analysis of the characteristic combination of traveling wave and standing wave behavior yield the fault types of complex faults, including: Complex fault types include high-resistance grounding faults and intermittent faults; When identifying high-resistance grounding faults, the hybrid model analyzes the following characteristics: the amplitude of the initial transient signal of the traveling wave is lower than the preset high-resistance grounding fault traveling wave amplitude threshold, the waveform change exceeds the preset normal range of the high-resistance grounding traveling wave waveform, and the transient signal is accompanied by noise or oscillation with an amplitude lower than the preset noise threshold and a frequency higher than the preset high-frequency threshold. At the same time, in the standing wave analysis, the amplitude of the zero-sequence component is lower than that of a typical grounding fault, the rate of change with time is lower than the preset zero-sequence component change threshold, and the fluctuation exceeds the preset normal range of high-resistance grounding zero-sequence component fluctuation. The hybrid model determines a high-resistance grounding fault by analyzing this combination of characteristics: the traveling wave transient is lower than the preset high-resistance grounding fault traveling wave amplitude threshold, and the standing wave zero-sequence component amplitude is low and unstable. When identifying intermittent faults, the hybrid model analyzes the time-varying nature of the traveling wave signal in the fusion features. This is manifested in the following: the interval between the appearance of the traveling wave transient signals of the fault phase voltage and current exceeds the preset minimum interval time, the interval between their disappearance exceeds the preset maximum disappearance time, the amplitude exceeds the preset intermittent fault amplitude range, the polarity change exceeds the preset polarity regular change range, and the arrival time interval stability is lower than the preset intermittent fault interval stability threshold. At the same time, in the standing wave analysis, the existence time of the zero-sequence component is lower than the preset intermittent fault zero-sequence component persistence time threshold. The hybrid model determines the fault as intermittent by capturing this combination of features where both the traveling wave transient and the standing wave zero-sequence component exhibit obvious intermittent characteristics and do not meet the preset continuity or stability.

6. The artificial intelligence analysis method for traveling wave fault location according to claim 1, characterized in that, Regarding fault location, for typical faults, the focus is on the time difference between the initial traveling wavefront arriving at both measurement points. The fault distance is calculated by combining this with line parameters, and standing wave information is used as auxiliary verification, including: After identifying the fault type as a typical fault, the time difference T between the initial traveling wave front and the measurement points at both ends of the line, extracted from the traveling wave data, is used. d The formula for calculating fault distance is: ； Where D is the fault distance, representing the distance from one end of the line to the fault point, c is the wave velocity, and T is the wave velocity. d The time difference between the arrival of the initial traveling wavefront at the measurement points at both ends of the line; In this process, the steady-state features extracted from the standing wave data are used as auxiliary verification information. By comparing them with the expected standing wave reflection characteristics corresponding to the fault distance D calculated based on the traveling wave, the traveling wave location results are verified, the confidence of the location results is improved, the fault distance D is output, and the mapping relationship between the fault distance D and the fault location coordinates is established by combining the geographical or topological information of the line. The specific coordinates of the fault location are obtained by linear interpolation calculation.

7. The artificial intelligence analysis method for traveling wave fault location according to claim 1, characterized in that, For complex faults, the analysis of time intervals and energy distribution of multiple wavefronts, relying on the reflection characteristics of the fault point reflected in the standing wave information, comprehensively determines the fault location, including: When the fault type is identified as a complex fault, the hybrid model analyzes the time interval sequence of multiple traveling wave fronts arriving at the measurement point and the corresponding energy distribution characteristics extracted from the traveling wave data; it also analyzes the steady-state characteristics extracted from the standing wave data, including the evolution of the standing wave ratio and reflection coefficient with frequency and time. Based on a comprehensive analysis of multiple wavefront time intervals, energy distributions, and standing wave reflection characteristics, a unified multidimensional feature vector is formed. The hybrid model establishes a mapping relationship between the fault location and the multidimensional feature vector. Using historical fault data, it learns the feature vectors generated by high-resistance grounding and intermittent faults at different locations. Through training, the hybrid model automatically learns and fits the nonlinear relationship between the feature vector and the fault location. The currently extracted feature vector is input into the trained hybrid model, and through calculation, the specific fault location coordinates are output. These coordinates are the final determined fault location coordinates, realizing the location of high-resistance grounding and intermittent faults.

8. The artificial intelligence analysis method for traveling wave fault location according to claim 1, characterized in that, When multiple fault points are detected near the fault location, a hybrid model is used for correlation analysis to evaluate whether there is a temporal sequence and spatial propagation relationship between the fault signals, thus obtaining fault correlation, including: When multiple fault points exist, the hybrid model analyzes the arrival time of each wavefront to identify the temporal relationship between different fault points and determine the propagation sequence of the fault signal. The hybrid model utilizes a spatiotemporal attention mechanism to evaluate the spatial propagation characteristics of fault signals and determine whether there are spatial propagation relationships between various fault points, including signal reflection, refraction, and scattering phenomena. In terms of time analysis, the hybrid model analyzes the time difference of the fault wavefront, infers the order of occurrence of fault points, and helps to locate the sequential relationship between multiple fault points; in terms of spatial analysis, the hybrid model combines the location of sensor nodes to evaluate the signal propagation path, fault distance and fault location, and further confirms the spatial correlation of different fault points. Based on the correlation analysis of time and space dimensions, the hybrid model comprehensively judges the mutual influence between multiple fault points and obtains the fault correlation.

9. The artificial intelligence analysis method for traveling wave fault location according to claim 1, characterized in that, The method, based on fault type, fault location, and fault correlation, assists maintenance personnel in decision-making and diagnosis, including: Based on the in-depth analysis of traveling wave and standing wave signals using a hybrid model, fault types are accurately identified, including single-phase grounding, two-phase short-circuit grounding, two-phase short circuit, three-phase short circuit, high-resistance grounding, and intermittent faults. By feeding back the identified fault types to maintenance personnel, the nature of the fault is provided, and corresponding handling measures are determined. By combining fault type and signal propagation analysis, the hybrid model provides fault location information to determine the specific location of the fault; by analyzing the propagation time difference of the traveling wave signal, the zero-sequence component in the standing wave, and the wavefront arrival time, it helps maintenance personnel locate the fault point in the power line. When multiple fault points exist, spatiotemporal correlation analysis is used to identify the temporal and spatial relationships between the fault points; hybrid model analysis is used to analyze the mutual influence between the fault points to determine whether the fault occurs alone or has a linkage effect, and the fault correlation is used to assist maintenance personnel in making decisions. Based on fault type, location, and correlation, it provides decision support and maintenance strategies for operations and maintenance personnel.

10. An artificial intelligence analysis system for traveling wave fault location, using the artificial intelligence analysis method for traveling wave fault location as described in any one of claims 1-9, characterized in that, include: Data acquisition and preprocessing module: includes a data acquisition unit and a data preprocessing unit; the data acquisition unit acquires traveling wave data and standing wave data through current sensors and voltage sensors; the data preprocessing unit performs alignment and synchronization, noise reduction filtering, feature extraction, and correction of missing and outlier values ​​on the data; The deep feature extraction and fusion module includes a hybrid model structure unit, a feature extraction unit, and a feature fusion and attention mechanism unit. The hybrid model structure unit constructs a hybrid network structure comprising a Convolutional Neural Network (CNN) and a Long Short-Term Memory (LSTM) network, used to process spatial features and temporal series features, respectively. The feature extraction unit uses CNN branches to extract spatial structural features from the input data and LSTM branches to extract temporal evolution features. The feature fusion and attention mechanism unit fuses spatial and temporal features into spatiotemporal features and applies an attention mechanism to dynamically emphasize important features and suppress irrelevant information. Fault type identification module: includes a typical fault identification unit and a complex fault identification unit; the typical fault identification unit, based on the features fused by the spatiotemporal attention mechanism, focuses on analyzing the transient characteristics of traveling wave grounding and the zero-sequence component information of standing wave, and identifies typical fault types including single-phase grounding, two-phase short-circuit grounding, two-phase short circuit, and three-phase short circuit; the complex fault identification unit, based on the features fused by the spatiotemporal attention mechanism, integrates and analyzes the characteristic combination of traveling wave and standing wave behavior to identify high-resistance grounding faults and intermittent faults. The fault location module includes a typical fault location unit and a complex fault location unit. The typical fault location unit calculates the fault distance based on the traveling wave time difference, verifies it with standing wave data, and locates the location coordinates of typical faults by combining line information. The complex fault location unit integrates and analyzes the traveling wave multi-wavehead and standing wave steady-state characteristics, inputs a trained hybrid model, and outputs the location coordinates of complex faults. The multi-fault correlation analysis module includes a time relationship analysis unit, a spatial relationship analysis unit, and a correlation comprehensive judgment unit. The time relationship analysis unit analyzes the arrival time of each wavefront to identify the temporal sequence between different fault points. The spatial relationship analysis unit utilizes spatiotemporal attention mechanisms and sensor location information to evaluate the spatial propagation characteristics of fault signals and determine whether there are spatial propagation or influence relationships between various fault points. The correlation comprehensive judgment unit integrates the analysis results from both time and spatial dimensions to determine the mutual influence between multiple fault points and draw conclusions about fault correlation. The fault integration and decision support module includes a fault information integration unit and a decision support unit. The fault information integration unit integrates fault type, fault location, and correlation information obtained from the fault type identification module, fault location module, and multi-fault correlation analysis module. Based on the integrated fault information, the decision support unit provides maintenance personnel with complete fault information and proposes preliminary maintenance strategy suggestions.

Images