Single-ended partial discharge accurate positioning method and system based on artificial intelligence and cable attenuation characteristics

Abstract

The application relates to a single-end partial discharge accurate positioning method and system based on artificial intelligence and cable attenuation characteristics, and belongs to the technical field of power equipment detection. The method solves the problems of traditional time domain reflectometers and frequency domain reflectometers, such as difficulty in identifying incident and reflected pulses, insufficient positioning accuracy and the like under complex environments such as pulse noise and multipath propagation. The technical scheme comprises the following steps: collecting signals in a single-end mode through a high-frequency current sensor, segmenting pulse fragments by using a sliding window correlation calculation, screening effective pulse pairs in combination with a cable attenuation characteristic curve, verifying multiple pulse pairs by using a full connection network, accurately identifying pulse arrival times by applying a U-Net deep learning model, and finally calculating the position of a partial discharge source based on a time difference and a propagation speed. The application significantly improves positioning accuracy and robustness, adapts to complex environments, and reduces implementation cost.

Claims

1. A method for precise localization of single-ended partial discharge based on artificial intelligence and cable attenuation characteristics, characterized in that: Includes the following steps: S1, a data acquisition system is used in conjunction with a high-frequency current sensor to collect and store the partial discharge signal of the target cable; S2, based on correlation calculation, uses a sliding window to calculate the correlation between the acquired partial discharge signal and the standard partial discharge sample, and segments the pulses near the peak in the correlation sequence to form pulse segments that may contain partial discharge. S3. The attenuation characteristic curve of the target cable is established by simulation or experimental methods. The peak ratio and time difference of each pair of pulse segments in the set of possible partial discharge pulse segments are calculated. Pulse pairs that fall into the attenuation characteristic region of the cable are screened out. If multiple pulse pairs fall into the attenuation characteristic region, the statistical characteristics of the pulse pairs are extracted and input into the Full Connected Network (FCN) model for further screening. S4, Input the incident pulse and the reflected pulse into the trained U-Net deep learning network model (U-Net) to calculate the arrival time of the incident pulse and the reflected pulse respectively; S5. Based on the arrival times of the incident and reflected pulses, the length of the cable, and the propagation speed of the partial discharge signal in the cable, determine the location of the partial discharge source in the cable.

2. The method for precise localization of single-ended partial discharge based on artificial intelligence and cable attenuation characteristics according to claim 1, characterized in that: In S1, the bandwidth of the high-frequency current sensor is 80kHz to 200MHz, and the sampling rate of the data acquisition system is not less than 200MHz.

3. The method for precise localization of single-ended partial discharge based on artificial intelligence and cable attenuation characteristics according to claim 1, characterized in that: In S2, the correlation calculation is determined by the following formula: in, , This indicates the correlation output. r The standard double-exponential partial discharge waveform, a and b It is a constant. s This represents the acquired time-domain digital signal sequence. n Indicates the window length. k Indicates the window step size. m Indicates the total number of segments in the signal. i Represents the index of sampling points within the window. j Represents the window position index.

4. The method for precise localization of single-ended partial discharge based on artificial intelligence and cable attenuation characteristics according to claim 1, characterized in that: In S3, the attenuation characteristic curve is established through experimental or simulation methods, wherein the amplitude ratio of the incident pulse to the reflected pulse is determined by the following formula: in, Indicates the incident pulse. Indicates a reflected pulse. Represents the attenuation constant. Indicates the cable length. d This indicates the distance between the local discharge source and the measuring end. It represents angular frequency.

5. The method for precise localization of single-ended partial discharge based on artificial intelligence and cable attenuation characteristics according to claim 1, characterized in that: In S3, the FCN model includes a flattening layer, a fully connected layer, a random deactivation layer, a first fully connected layer, a first random deactivation layer, a second fully connected layer, and an output layer. After the seven statistical features enter the network, they are first vectorized by the flattening layer. The statistical features are then processed sequentially through multiple fully connected operation units. In each fully connected layer, high-level discriminative features are extracted step by step through linear weighting and nonlinear mapping. The network introduces a random deactivation mechanism in the intermediate layer to limit overfitting. Finally, the model maps the high-dimensional discriminative features into a single scalar output through the output layer.

6. The method for precise localization of single-ended partial discharge based on artificial intelligence and cable attenuation characteristics according to claim 5, characterized in that: The statistical features include peak ratio, time difference, rise time ratio, fall time ratio, area ratio, kurtosis ratio, and skewness ratio. When the FCN model output is 1, it indicates that the input pulse pair constitutes the incident pulse and reflected pulse pair of the same partial discharge event. When the output is 0, it indicates that the input pulse is an invalid pulse or noise pulse.

7. The method for precise localization of single-ended partial discharge based on artificial intelligence and cable attenuation characteristics according to claim 1, characterized in that: In S4, the U-Net model processes the input signal by: resampling the time-domain signal segments containing the determined incident pulse and reflected pulse to form a fixed-length one-dimensional time-domain signal; feeding this sequence signal as input into a pre-trained U-Net network; the U-Net network performs end-to-end processing on the input signal segments; the model outputs a probability distribution sequence with the same length as the input sequence to characterize the probability distribution of pulse arrival time in the time series; post-processing the output results; determining the position indices of the incident pulse and reflected pulse on the time axis by threshold determination; and converting them into actual time values ​​by combining the sampling rate.

8. The method for precise localization of single-ended partial discharge based on artificial intelligence and cable attenuation characteristics according to claim 7, characterized in that: The network architecture of the U-Net model includes an encoder and a decoder. The encoder's network structure includes a first convolutional layer, a second convolutional layer, a first pooling layer, a third convolutional layer, a fourth convolutional layer, a second pooling layer, a fifth convolutional layer, and a sixth convolutional layer connected in sequence. The decoder's network structure includes a first upsampling layer, a first skip connection layer, a seventh convolutional layer, an eighth convolutional layer, a second upsampling layer, a second skip connection layer, a ninth convolutional layer, a tenth convolutional layer, and an eleventh convolutional layer connected in sequence.

9. The method for precise localization of single-ended partial discharge based on artificial intelligence and cable attenuation characteristics according to claim 1, characterized in that: In S5, the location of the partial discharge power source is determined by the following formula: in, d This indicates the distance between the partial discharge source and the measuring end. Indicates the length of the cable. v This indicates the propagation speed of a partial discharge signal in a cable. Indicates the arrival time of the incident pulse. This indicates the arrival time of the reflected pulse.

10. An artificial intelligence-driven cable partial discharge location system based on time-domain reflectometry and transfer function analysis, characterized in that... include: The sensing unit is used to sense partial discharge signals; A data acquisition unit, connected to the sensing unit, is used to acquire and digitize the signal; The first signal filtering module is used to identify pulse segments that may contain partial discharge based on correlation calculations. The second signal filtering module is used to filter incident and reflected pulse pairs; The arrival time identification module is used to calculate the arrival time of the incident pulse and the reflected pulse using the U-Net model; Partial discharge location module, used to determine the location of partial discharge source based on arrival time, cable length and signal propagation speed; The second signal filtering module includes an FCN model, and the arrival time recognition module includes a U-Net model. The network architecture includes an encoder and a decoder. The encoder's network structure includes a first convolutional layer, a second convolutional layer, a first pooling layer, a third convolutional layer, a fourth convolutional layer, a second pooling layer, a fifth convolutional layer, and a sixth convolutional layer connected in sequence. The decoder's network structure includes a first upsampling layer, a first skip connection layer, a seventh convolutional layer, an eighth convolutional layer, a second upsampling layer, a second skip connection layer, a ninth convolutional layer, a tenth convolutional layer, and an eleventh convolutional layer connected in sequence.

Images