A method for locating an abnormal discharge fault point of a power transmission line

Abstract

The application discloses a power transmission line abnormal discharge fault point positioning method, comprising the following steps: S1, installing a high-frequency current transformer at a key node to collect a transient current signal during abnormal discharge at a frequency not lower than 1MHz and to carry out denoising and standardization preprocessing; S2, parallelly applying wavelet transform and Fourier transform, the wavelet transform extracts local detail features of the signal in the time-frequency domain, the Fourier transform obtains global frequency spectrum features of the signal, and the two are fused into a high-dimensional composite feature vector; S3, inputting the obtained feature vector into a pre-trained BP neural network to realize intelligent mapping from the features to the fault mode; S4, adopting an electromagnetic time reversal method, intercepting a fault initial traveling wave signal, reversing on a time axis, replaying in an accurate electromagnetic model of the line, calculating and scanning a spatial distribution function of electromagnetic energy along the line, and obtaining an accurate fault point; the application has the advantages of intelligent identification, high positioning precision and strong anti-interference performance.

Description

Technical Field

[0001] This invention belongs to the field of power system safety operation and maintenance technology, and specifically relates to a method for locating abnormal discharge fault points in transmission lines. Background Technology

[0002] Abnormal discharges in transmission lines (such as corona, flashover, internal discharge, etc.) are early signs of insulation deterioration or serious fault manifestations. Unlike obvious line breakage faults, they are insidious, gradual, and instantaneous, and pose a huge threat. Therefore, simply knowing that "a discharge has occurred" is far from enough. It is necessary to accurately locate the fault point in order to take effective measures, prevent further damage, and stop losses in time.

[0003] However, traditional transmission line fault diagnosis and location methods suffer from several problems: First, they rely on manual experience to analyze waveforms, which is inefficient, highly subjective, and prone to misjudging complex and combined discharge modes, making fault type identification difficult. Second, the double-ended traveling wave method relies on wavefront capture, which is susceptible to noise interference, and the impedance method is greatly affected by line parameters and arc resistance, often resulting in errors of hundreds of meters to kilometers. Third, the reliance on manual monitoring, diagnosis, and location is carried out by different systems or personnel, leading to fragmented processes and low efficiency. To solve these problems, it is necessary to develop a method for locating abnormal discharge fault points in transmission lines. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method for locating abnormal discharge fault points in transmission lines that is intelligent in identification, highly accurate in positioning, and has strong anti-interference capabilities. It can deeply integrate modern signal processing, artificial intelligence, and electromagnetic field theory, providing a complete and accurate closed-loop solution for condition-based maintenance of transmission lines, from diagnosis to location.

[0005] The objective of this invention is achieved as follows: a method for locating abnormal discharge fault points in transmission lines, comprising the following steps:

[0006] S1, Fault current signal acquisition and preprocessing: Install high-frequency current transformers at key nodes to acquire transient current signals during abnormal discharge at a frequency of not less than 1MHz, and then perform noise reduction and standardization preprocessing.

[0007] S2, Feature extraction based on time-frequency domain analysis: Wavelet transform and Fourier transform are used in parallel. Wavelet transform extracts the local detail features of the signal in the time-frequency domain, Fourier transform obtains the global spectral features of the signal, and finally the results of the two are fused to construct a high-dimensional composite feature vector.

[0008] S3, Fault type identification based on BP neural network: The feature vector obtained in step S2 is input into the pre-trained BP neural network. This network has the ability to classify by learning historical fault samples and can output the fault type discrimination result, realizing intelligent mapping from features to fault mode.

[0009] S4. Precise fault location based on electromagnetic time inversion: Using the electromagnetic time inversion method, the initial traveling wave signal of the fault is captured, reversed on the time axis, and then played back in the precise electromagnetic model of the line. By calculating and scanning the spatial distribution function of electromagnetic energy along the line, the location corresponding to its unique minimum point is the precise fault point.

[0010] Preferably, step S1 specifically includes:

[0011] S11, Signal Acquisition: Install broadband current transformers at the beginning of the transmission line and at possible multi-point monitoring locations to synchronously acquire high-frequency transient current signals during abnormal discharges. sampling frequency ;

[0012] S12, Preprocessing: Processing the raw signal Noise reduction and standardization processes are performed.

[0013] Preferably, in step S12, denoising refers to using a moving average filter or an adaptive filter to suppress power frequency and white noise interference; normalization refers to normalizing the denoised signal. To normalize the amplitude and eliminate the influence of dimensions, the following formula is used:

[0014] ;

[0015] Where u is the signal mean. The standard deviation is denoted as .

[0016] Preferably, step S2 specifically comprises:

[0017] S21, Wavelet transform feature extraction:

[0018] S211, Transformation process: Select a mother wavelet suitable for analyzing transient signals, and perform multi-scale discrete wavelet transform on the signal, as shown in the following formula:

[0019] CWT ;

[0020] Where a is the scale factor (corresponding to frequency) and b is the translation factor (corresponding to time). For the mother wavelet function, Indicates complex conjugation;

[0021] S212, Feature Calculation: Selecting the detail coefficients of each layer after decomposition. and approximation coefficients Let j be the number of decomposition levels. Calculate the following statistical characteristics:

[0022] Energy entropy : ;in, , The energy of the wavelet coefficients at the j-th level;

[0023] variance : Calculate the variance of the coefficients at each level to characterize the intensity of signal fluctuations;

[0024] Wavelet packet energy spectrum: Further wavelet packet decomposition is performed to construct the energy distribution vector. ;

[0025] S22, Fourier feature extraction:

[0026] S221, Transformation process: Perform a Fast Fourier Transform on the signal to obtain its frequency domain identifier, as shown in the following formula:

[0027] ;

[0028] S222, Calculation of characteristic quantities:

[0029] Spectral centroid FC: ;

[0030] Spectrum Standard Deviation : Describes the degree of spectral dispersion;

[0031] Harmonic component ratio (HR): Calculates the ratio of the amplitude of a specific harmonic to the amplitude of the fundamental frequency.

[0032] Total harmonic distortion (THD);

[0033] S23, Feature Vector Construction: The time-frequency domain features described above are fused to form the final feature vector F for classification. .

[0034] Preferably, in step S211, the mother wavelet adopts the Daubechies series.

[0035] Preferably, step S3 specifically comprises:

[0036] S31, Network Construction: Construct a three-layer BP neural network as follows:

[0037] Input layer: The number of neurons is equal to the dimension n of the feature vector F;

[0038] Hidden layer: The number of neurons is determined through trial and error, with an initial value of 2n+1, and the activation function is either Sigmoid or ReLU.

[0039] Output layer: The number of neurons is equal to the number of discharge fault types m to be classified, and the Softmax function is used to output the probability of each category;

[0040] S32, Network Training: Supervised training is performed using a historical database, which contains feature vectors F and their labels T for various confirmed fault types, as follows:

[0041] S321, Forward Propagation: Calculates the network output Y;

[0042] S322, Error Backpropagation: Using Cross-Entropy Loss Function The network weights W and bias b are updated using the gradient descent algorithm until convergence.

[0043] Preferably, in step S31, the discharge fault type is normal, corona discharge, internal discharge, or surface discharge.

[0044] Preferably, step S4 specifically comprises:

[0045] S41, Establishment of the basic principle of time reversal:

[0046] Assuming the fault point is located at At the monitoring point The recorded fault current signal is ;

[0047] The core of electromagnetic time reversal is to reverse the recorded signal in time. And in a numerical model containing real lines, it is used as an excitation source for "replay";

[0048] S42, Constructing energy mirroring and positioning criteria:

[0049] In the numerical model, the electromagnetic energy density at each point x in the circuit space during the "playback" process is calculated. ;

[0050] Define the fault current energy mirror function The energy density is the integral or peak value in the time domain: or ;

[0051] Minimum criterion: Theoretically proven, in lossless or low-power circuit models, At the actual location of the fault It presents a single, sharp minimum point in space;

[0052] S43, Implementation steps:

[0053] S431, Signal Truncation and Inversion: Truncates the initial current traveling wave identified by the BP neural network as an abnormal discharge type. And by reversing time, we can obtain ;

[0054] S432, Numerical Simulation Playback: [This appears to be a fragment of a larger text, possibly related to simulation playback As an excitation, it is injected into the precise numerical model of the line to perform full-wave electromagnetic simulation;

[0055] S433, Energy Field Calculation and Scanning: Calculating the energy mirror function along a one-dimensional path within the simulation domain. ;

[0056] S434, Location Judgment: Search The coordinates of the position corresponding to the global minimum value This is the precise location result of the fault point.

[0057] Preferably, in step S41, the digital model of the actual circuit is constructed based on the finite element method or the method of moments.

[0058] Due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0059] (1) The present invention adopts wavelet + Fourier feature extraction + BP neural network identification, which can realize the automatic, intelligent and high-precision classification of fault features, and solves the problems of low efficiency, strong subjectivity, easy misjudgment of responsible and compound discharge modes, and difficulty in fault type identification that rely on manual experience to analyze waveforms.

[0060] (2) The present invention adopts electromagnetic time inversion based on the minimum value criterion of energy image, and directly uses the full waveform information of the fault signal for physical focusing, which can achieve meter-level high-precision positioning. It solves the problems of existing double-end traveling wave method relying on wavehead capture, being easily affected by noise interference, and impedance method being greatly affected by line parameters and arc resistance, with errors often reaching hundreds of meters to kilometers.

[0061] (3) The present invention adopts a seamless intelligent closed loop from signal → feature → diagnosis → positioning, which can realize an integrated solution of data input, automatic output of diagnostic results and precise location, and solves the problem that the existing reliance on manual monitoring, diagnosis and positioning is completed by different systems or personnel, resulting in fragmented processes and low efficiency.

[0062] In summary, this invention has the advantages of intelligent identification, high-precision positioning, and strong anti-interference capabilities. It can deeply integrate modern signal processing, artificial intelligence, and electromagnetic field theory to provide a complete and accurate closed-loop solution for transmission line condition-based maintenance, from diagnosis to positioning. Attached Figure Description

[0063] Figure 1This is a flowchart of the method steps of the present invention. Detailed Implementation

[0064] The technical solution of the present invention will be further described in detail below through embodiments and in conjunction with the accompanying drawings.

[0065] like Figure 1 As shown, this invention provides a method for locating abnormal discharge fault points in transmission lines, aiming to build a transmission line abnormal discharge fault handling system that integrates "feature extraction, intelligent identification, and precise location", which is implemented through the following four steps.

[0066] Step S1, Fault Current Signal Acquisition and Preprocessing: High-frequency current transformers are installed at key nodes such as substations to acquire transient current signals during abnormal discharge at a frequency of not less than 1MHz. Then, noise reduction and standardization preprocessing are performed to eliminate interference and provide clean input data for subsequent analysis. The purpose is to obtain high-quality, analyzable original fault current signals.

[0067] S11, Signal Acquisition: Install broadband current transformers at the beginning of the transmission line (outgoing line of the substation) and at possible multi-point monitoring locations to synchronously acquire high-frequency transient current signals during abnormal discharges. sampling frequency The Nyquist theorem must be satisfied, typically at a frequency of no less than 1 MHz, to capture discharge details.

[0068] S12, Preprocessing: Processing the raw signal Perform noise reduction and standardization processing;

[0069] Noise reduction: Use a moving average filter or adaptive filter to suppress power frequency and white noise interference;

[0070] Standardization: the standardization of the denoised signal To normalize the amplitude and eliminate the influence of dimensions, the following formula is used:

[0071] ;

[0072] Where u is the signal mean. The standard deviation is denoted as .

[0073] Step S2, feature extraction based on time-frequency domain analysis: Wavelet transform and Fourier transform are used in parallel. Wavelet transform extracts local detailed features of the signal in the time-frequency domain (such as energy entropy and variance), and Fourier transform obtains global spectral features of the signal (such as spectral centroid and harmonic distortion rate). Finally, the results of the two are fused to construct a high-dimensional composite feature vector. The purpose is to extract multidimensional features that can characterize different discharge types from the preprocessed signal.

[0074] S21, Wavelet transform feature extraction:

[0075] S211, Transformation process: Select a suitable mother wavelet (such as the Daubechies series) for analyzing transient signals, and perform multi-scale discrete wavelet transform on the signal, as shown in the following formula:

[0076] CWT ;

[0077] Where a is the scale factor (corresponding to frequency) and b is the translation factor (corresponding to time). For the mother wavelet function, Indicates complex conjugation.

[0078] S212, Feature Calculation: Selecting the detail coefficients of each layer after decomposition. and approximation coefficients Let j be the number of decomposition levels. Calculate the following statistical characteristics:

[0079] Energy entropy : ;in, , The energy of the wavelet coefficients at the j-th level;

[0080] variance : Calculate the variance of the coefficients at each level to characterize the intensity of signal fluctuations;

[0081] Wavelet packet energy spectrum: Further wavelet packet decomposition is performed to construct the energy distribution vector. .

[0082] S22, Fourier feature extraction:

[0083] S221, Transformation process: Perform a Fast Fourier Transform on the signal to obtain its frequency domain identifier, as shown in the following formula:

[0084] .

[0085] S222, Calculation of characteristic quantities:

[0086] Spectral centroid FC: ;

[0087] Spectrum Standard Deviation : Describes the degree of spectral dispersion;

[0088] Harmonic component ratio (HR): Calculates the ratio of the amplitude of a specific harmonic (e.g., 3rd, 5th, 7th) to the amplitude of the fundamental frequency.

[0089] Total harmonic distortion (THD).

[0090] S23, Feature Vector Construction: The time-frequency domain features described above are fused to form the final feature vector F for classification. .

[0091] Step S3, Fault type identification based on BP neural network: The feature vector obtained in step S2 is input into the pre-trained BP neural network. This network has the ability to classify by learning historical fault samples and can output the fault type discrimination result, realizing intelligent mapping from features to fault mode; the purpose is to automatically identify the specific type of abnormal discharge (such as corona discharge, internal discharge, etc.).

[0092] S31, Network Construction: Construct a three-layer (input layer, hidden layer, output layer) BP neural network, as follows:

[0093] Input layer: The number of neurons is equal to the dimension n of the feature vector F;

[0094] Hidden layer: The number of neurons is determined through trial and error, with an initial value of 2n+1, and the activation function is either Sigmoid or ReLU.

[0095] Output layer: The number of neurons is equal to the number of discharge fault types m to be classified (such as normal, corona discharge, internal discharge, and surface discharge). The Softmax function is used to output the probability of each category.

[0096] S32, Network Training: Supervised training is performed using a historical database (containing feature vectors F and their labels T for various confirmed fault types), as follows:

[0097] S321, Forward Propagation: Calculates the network output Y;

[0098] S322, Error Backpropagation: Using Cross-Entropy Loss Function The network weights W and bias b are updated using a gradient descent algorithm (such as Adam) until convergence.

[0099] Step S4, precise fault location based on electromagnetic time inversion: Using the electromagnetic time inversion method, the initial traveling wave signal of the fault is intercepted, reversed on the time axis, and then played back in the precise electromagnetic model of the line. By calculating and scanning the spatial distribution function of electromagnetic energy along the line, the location corresponding to its unique minimum point is the precise fault point. The purpose is to accurately locate the fault point at the meter level after determining the discharge type.

[0100] S41, Establishment of the basic principle of time reversal:

[0101] Assuming the fault point is located at At the monitoring point The recorded fault current signal is ;

[0102] The core of electromagnetic time reversal is to reverse the recorded signal in time. And in a numerical model containing the real circuit (such as one built based on the finite element method or the method of moments), it is used as the excitation source for "replay".

[0103] S42, Constructing energy mirroring and positioning criteria:

[0104] In the numerical model, the electromagnetic energy density at each point x in the circuit space during the "playback" process is calculated. ;

[0105] Define the fault current energy mirror function The energy density is the integral or peak value in the time domain: or ;

[0106] Minimum criterion: Theoretically proven, in lossless or low-power circuit models, At the actual location of the fault It exhibits a single, sharp minimum point in space; this is because the time reversal process enables the wave field to achieve spatiotemporal synchronous focusing at the source point.

[0107] S43, Implementation steps:

[0108] S431, Signal Truncation and Inversion: Truncates the initial current traveling wave identified by the BP neural network as an abnormal discharge type. And by reversing time, we can obtain ;

[0109] S432, Numerical Simulation Playback: [This appears to be a fragment of a larger text, possibly related to simulation playback As an excitation, it is injected into the precise numerical model of the line to perform full-wave electromagnetic simulation;

[0110] S433, Energy Field Calculation and Scanning: Calculating the energy mirror function along a one-dimensional path within the simulation domain. ;

[0111] S434, Location Judgment: Search The coordinates of the position corresponding to the global minimum value This is the precise location result of the fault point.

[0112] In practice, a high-fidelity numerical model needs to be built based on the actual parameters of the line, and sufficient samples need to be accumulated to train and optimize the neural network.

[0113] In summary, the four steps of this invention are logically closely linked, forming a complete technical closed loop of "signal acquisition → feature extraction → intelligent diagnosis → physical positioning", realizing fully automated processing from anomaly detection to precise fault location.

[0114] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of the claims of the present invention.

Claims

1. A method for locating abnormal discharge fault points in transmission lines, characterized in that: Includes the following steps: S1, Fault current signal acquisition and preprocessing: Install high-frequency current transformers at key nodes to acquire transient current signals during abnormal discharge at a frequency of not less than 1MHz, and then perform noise reduction and standardization preprocessing. S2, Feature extraction based on time-frequency domain analysis: Wavelet transform and Fourier transform are used in parallel. Wavelet transform extracts the local detail features of the signal in the time-frequency domain, Fourier transform obtains the global spectral features of the signal, and finally the results of the two are fused to construct a high-dimensional composite feature vector. S3, Fault type identification based on BP neural network: The feature vector obtained in step S2 is input into the pre-trained BP neural network. This network has the ability to classify by learning historical fault samples and can output the fault type discrimination result, realizing intelligent mapping from features to fault mode. S4. Precise fault location based on electromagnetic time inversion: Using the electromagnetic time inversion method, the initial traveling wave signal of the fault is captured, reversed on the time axis, and then played back in the precise electromagnetic model of the line. By calculating and scanning the spatial distribution function of electromagnetic energy along the line, the location corresponding to its unique minimum point is the precise fault point.

2. The method for locating abnormal discharge fault points in transmission lines according to claim 1, characterized in that, Step S1 specifically involves: S11, Signal Acquisition: Install broadband current transformers at the beginning of the transmission line and at possible multi-point monitoring locations to synchronously acquire high-frequency transient current signals during abnormal discharges. sampling frequency ; S12, Preprocessing: Processing the raw signal Noise reduction and standardization processes are performed.

3. The method for locating abnormal discharge fault points in transmission lines according to claim 2, characterized in that: In step S12, denoising refers to using a moving average filter or an adaptive filter to suppress power frequency and white noise interference; normalization refers to standardizing the denoised signal. To normalize the amplitude and eliminate the influence of dimensions, the following formula is used: ； Where u is the signal mean. The standard deviation is denoted as .

4. The method for locating abnormal discharge fault points in transmission lines according to claim 1, characterized in that, Step S2 specifically involves: S21, Wavelet transform feature extraction: S211, Transformation process: Select a mother wavelet suitable for analyzing transient signals, and perform multi-scale discrete wavelet transform on the signal, as shown in the following formula: CWT ; Where a is the scale factor (corresponding to frequency) and b is the translation factor (corresponding to time). For the mother wavelet function, Indicates complex conjugation; S212, Feature Calculation: Selecting the detail coefficients of each layer after decomposition. and approximation coefficients Let j be the number of decomposition levels. Calculate the following statistical characteristics: Energy entropy : ;in, , The energy of the wavelet coefficients at the j-th level; variance : Calculate the variance of the coefficients at each level to characterize the intensity of signal fluctuations; Wavelet packet energy spectrum: Further wavelet packet decomposition is performed to construct the energy distribution vector. ; S22, Fourier feature extraction: S221, Transformation process: Perform a Fast Fourier Transform on the signal to obtain its frequency domain identifier, as shown in the following formula: ； S222, Calculation of characteristic quantities: Spectral centroid FC: ; Spectrum Standard Deviation : Describes the degree of spectral dispersion; Harmonic component ratio (HR): Calculates the ratio of the amplitude of a specific harmonic to the amplitude of the fundamental frequency. Total harmonic distortion (THD); S23, Feature Vector Construction: The time-frequency domain features described above are fused to form the final feature vector F for classification. .

5. The method for locating abnormal discharge fault points in transmission lines according to claim 4, characterized in that: In step S211, the mother wavelet adopts the Daubechies series.

6. The method for locating abnormal discharge fault points in transmission lines according to claim 1, characterized in that, Step S3 specifically involves: S31, Network Construction: Construct a three-layer BP neural network as follows: Input layer: The number of neurons is equal to the dimension n of the feature vector F; Hidden layer: The number of neurons is determined through trial and error, with an initial value of 2n+1, and the activation function is either Sigmoid or ReLU. Output layer: The number of neurons is equal to the number of discharge fault types m to be classified, and the Softmax function is used to output the probability of each category; S32, Network Training: Supervised training is performed using a historical database, which contains feature vectors F and their labels T for various confirmed fault types, as follows: S321, Forward Propagation: Calculates the network output Y; S322, Error Backpropagation: Using Cross-Entropy Loss Function The network weights W and bias b are updated using the gradient descent algorithm until convergence.

7. The method for locating abnormal discharge fault points in transmission lines according to claim 6, characterized in that: In step S31, the discharge fault types are normal, corona discharge, internal discharge, and surface discharge.

8. The method for locating abnormal discharge fault points in transmission lines according to claim 1, characterized in that: Step S4 specifically involves: S41, Establishment of the basic principle of time reversal: Assuming the fault point is located at At the monitoring point The recorded fault current signal is ; The core of electromagnetic time reversal is to reverse the recorded signal in time. And in a numerical model containing real lines, it is used as an excitation source for "replay"; S42, Constructing energy mirroring and positioning criteria: In the numerical model, the electromagnetic energy density at each point x in the circuit space during the "playback" process is calculated. ; Define the fault current energy mirror function The energy density is the integral or peak value in the time domain: or ; Minimum criterion: Theoretically proven, in lossless or low-power circuit models, At the actual location of the fault It presents a single, sharp minimum point in space; S43, Implementation steps: S431, Signal Truncation and Inversion: Truncates the initial current traveling wave identified by the BP neural network as an abnormal discharge type. And by reversing time, we can obtain ; S432, Numerical Simulation Playback: [This appears to be a fragment of a larger text, possibly related to simulation playback As an excitation, it is injected into the precise numerical model of the line to perform full-wave electromagnetic simulation; S433, Energy Field Calculation and Scanning: Calculating the energy mirror function along a one-dimensional path within the simulation domain. ; S434, Location Judgment: Search The coordinates of the position corresponding to the global minimum value This is the precise location result of the fault point.

9. The method for locating abnormal discharge fault points in transmission lines according to claim 8, characterized in that: In step S41, the digital model of the actual circuit is constructed based on the finite element method or the method of moments.

Images