Machine learning-based power line live line alignment device and method

Abstract

The invention relates to an electric power circuit live line alignment device and method based on machine learning, and the device employs an anti-electromagnetic shielding high-frequency coil sensor to achieve complete non-contact detection, fundamentally avoids the electric shock and arc risks caused by the short-distance contact of an operator with a high-voltage line, and greatly improves the operation safety. Through fusion of optical fiber and wireless communication technologies and combination of wavelet transform and a support vector machine (SVM) machine learning algorithm, the system can accurately extract deep time-frequency domain features of current signals, effectively suppress noise interference in a strong electromagnetic environment, and improve the line identification accuracy. The device is integrated with an FPGA for real-time processing, the response speed is high, and a super capacitor is used for supplying power to adapt to intermittent operation. And finally, the three-dimensional visual result is displayed through the touch screen, so that the intelligent degree is high, the manual dependence and the misjudgment rate are remarkably reduced, and reliable technical support is provided for rapid, accurate and safe alignment of the high-voltage line.

Description

Technical Field

[0001] This disclosure relates to the field of power line detection technology, and more specifically, to a live-line pairing device and method for power lines based on machine learning. Background Technology

[0002] Traditional manual inspection methods heavily rely on the visual judgment and experience of maintenance personnel, and their accuracy is significantly affected by individual differences. Laser ranging technology is susceptible to interference from strong sunlight, rain, snow, fog, and other adverse weather conditions, and faces difficulties in identifying reflective surfaces. RFID-based identification schemes suffer from inherent drawbacks such as tag aging and detachment, limited information storage capacity, and potential signal interference in complex electromagnetic environments. Of particular note is that current inspection methods often require personnel to be in close contact with high-voltage lines, posing risks of direct electric shock, arc burns, and, in ultra-high-voltage environments, near-discharge phenomena. These technical shortcomings result in low line inspection efficiency and a high annual rate of misjudgments due to inspection errors, severely hindering the progress of smart grid construction. Furthermore, traditional Discrete Fourier Transform (DFT) analysis methods in power line inspection rely on physical contact measurements and are susceptible to power frequency noise and electromagnetic interference, leading to inaccurate harmonic feature extraction and affecting identification accuracy. Therefore, there is an urgent need for a device and method that can efficiently and accurately complete the alignment operation under energized conditions, is not easily affected by electromagnetic interference, and solves the problems of high dependence on manual labor, high safety risks, and insufficient adaptability of DFT analysis in the existing technology. Summary of the Invention

[0003] To overcome the existing problems, the present invention provides a live-line pairing device and method for power lines based on machine learning.

[0004] In a first aspect, this disclosure provides a live-line pairing method for power lines based on machine learning, including:

[0005] Step S1: Acquire the conductor current signal through a high-frequency current sampling coil with anti-electromagnetic shielding;

[0006] Step S2: The current signal is converted into a digital signal by a high-precision ADC, and the information is sent to the processing terminal through the fiber optic wireless module.

[0007] Step S3: The FPGA at the processing end performs wavelet packet decomposition to extract the energy features of 5 frequency bands;

[0008] Step S4: Input the 5-layer frequency band energy features into the SVM model for feature comparison and output the matching probability;

[0009] Step S5: Display the matching probability results on the touchscreen. A green light illuminates when a match is successful.

[0010] Optionally, step S1 includes:

[0011] Step S11: Clamp the high-frequency current sampling coil, which is resistant to electromagnetic interference, onto the conductor under test. For the measurement scenario of multi-phase lines, the system adopts a split coil design.

[0012] In step S12, the sensor works based on the principle of electromagnetic induction. When the changing current in the conductor generates a synchronously changing magnetic field, the magnetic field is efficiently coupled by the magnetic core and generates a changing magnetic flux. Finally, a voltage signal that precisely corresponds to the original rate of change of current is induced at the output end of the coil, thus completing the accurate capture of the current characteristics.

[0013] Optionally, step S2 includes:

[0014] Step S21: The original analog signal output by the high-frequency current sensor is processed by a preamplifier with high input impedance and low noise to improve the signal-to-noise ratio and driving capability.

[0015] In step S22, after processing by the preamplifier, the signal enters a dynamic gain adjustment circuit controlled by a microcontroller. This circuit can automatically adapt to different current ranges from milliamps to kiloamps, ensuring that the signal is optimized to the best range suitable for processing by the analog-to-digital converter (ADC).

[0016] Step S23: Before entering the ADC, the signal passes through a high-order active low-pass filter. The cutoff frequency of the filter is precisely set to ensure that the final digital signal can reproduce the original analog characteristics with high fidelity.

[0017] Step S24: Data frames are encapsulated according to a preset standard communication protocol. Each data frame contains specific structural information, laying the foundation for subsequent transmission, identification and verification.

[0018] Step S25: An identification code ID and a high-precision timestamp are embedded in the encapsulated data frame to ensure the traceability of the data source and the accuracy of time synchronization. At the end of the data frame, a cyclic redundancy check code is added to verify the integrity of the data during transmission and to determine whether any errors have occurred.

[0019] Optionally, step S2 further includes:

[0020] Step S25: The encapsulated data is transmitted through gigabit single-mode fiber, taking advantage of the high bandwidth, low latency and natural electromagnetic interference resistance of fiber.

[0021] In step S26, the system simultaneously constructs a LoRa wireless link as an auxiliary channel. This channel performs real-time analysis on the raw data, extracts key characteristic values ​​such as effective value, peak value, and harmonic content, and then compresses and transmits them.

[0022] Optionally, step S3 includes:

[0023] Step S31: Perform 5-level wavelet packet decomposition using the DB4 wavelet basis to extract energy features across 32 frequency bands.

[0024] The DB4 wavelet is chosen because its tight support and regularity are suitable for capturing local abrupt changes in current signals.

[0025] ;

[0026] Where t represents the time variable; gk represents the coefficients of the 8th-order high-pass filter of DB4; The mother wavelet function is represented by φ, also known as the analytical wavelet; φ represents the scaling function corresponding to the mother wavelet, representing the smoothed or approximate part of the signal. Its function is to generate mother wavelets at different scales and to capture the low-frequency profile of the signal; k represents the discrete translation parameter or sequence index, which is an integer indicating that the wavelet or scaling function is translated on the time axis to cover the entire signal for analysis.

[0027] The amplitude of the current signal x(t) acquired by the ADC is normalized:

[0028] ;

[0029] Where μ represents the arithmetic mean or DC bias of signal x(t) within the analysis time window; σ represents the standard deviation of signal x(t) within the analysis time window; and xnom(t) represents the standardized characteristic signal.

[0030] After decomposing into 5 layers, 32 sub-bands {W} are obtained. 5.1 W 5.2 , … , W 5.32}

[0031] The Mallat algorithm is used for 5-level decomposition to extract detail coefficients d. j and approximation coefficient a j :

[0032] ;

[0033] Where j represents the wavelet decomposition level or scale; n represents the discrete-time index; a j [n] represents the approximation coefficients of the decomposition at the j-th level; d j[n] represents the detail coefficients at the j-th level; h[k] represents the DB4 low-pass filter coefficients; g[k] represents the DB4 high-pass filter coefficients; the number of decomposition levels j=1, 2, … , 5 corresponds to the frequency range; d1 ranges from 2.5 to 5 kHz; d3 ranges from 312.5 to 625 Hz; d5 ranges from 78.125 to 156.25 Hz;

[0034] Step S32, calculate the energy proportion E3 of the wavelet coefficients in layer d3:

[0035] ;

[0036] Where k represents the discrete translation parameter or sequence index; j represents the number of layers or scale of the wavelet decomposition;

[0037] The normalized entropy value H3 is calculated as follows:

[0038] ;

[0039] Detecting the local extremum point M5 of the d5 coefficient:

[0040] ;

[0041] The Canny edge detection algorithm is used to locate extreme points;

[0042] Step S33: Calculate the phase difference between the two conductor signals using wavelet coherence:

[0043]

[0044] in, W represents the instantaneous phase difference between the current signals of the two conductors in the 5th harmonic frequency band. xy The cross wavelet spectrum represents the d5 layer of the two signals.

[0045] Optionally, step S4 includes:

[0046] Step S41: Combine the above features into a three-dimensional vector:

[0047] ;

[0048] Where F is the original feature vector, E3 represents the dimensionless ratio, M5 represents the amplitude change rate, and Δφ represents the phase angle;

[0049] Normalization process:

[0050] ;

[0051] Among them, F norm This represents the normalized standard feature vector;

[0052] Step S42: The support vector machine (SVM) model pre-trained on the embedded Linux system is compared with the current feature and the line feature library to output the matching probability P (0~1). The system performs a three-level response based on the probability value: output matching probability P (0~1); if P ≥ 95%, it is determined to be "same wire"; if 80% ≤ P < 95%, a yellow warning is triggered and manual review is recommended; if 80% P < 80%, it is determined to be "different wire" and a red alarm is issued.

[0053] Step S43: The model constructs a feature library based on a large-scale experimental sample, collecting more than 1,000 sets of sample data to provide a reliable data foundation for the training and comparison of the SVM model.

[0054] Optionally, step S5 includes:

[0055] Step S51 displays a comparison of the time-domain waveforms of the original electric field signal and the filtered curve, visually presenting the characteristic changes and noise suppression effects before and after signal processing;

[0056] Step S52: The energy proportion of each layer of wavelet decomposition is displayed by frequency domain energy distribution bar chart, and the spatial distribution of electric field intensity around the conductor is reflected by three-dimensional electric field cloud map, with color gradient clearly representing intensity change.

[0057] In step S53, the system automatically generates a matching conclusion of "match successful" or "no match" and marks the precise confidence level to provide a quantitative basis for the diagnostic results.

[0058] Secondly, this disclosure provides a live-line pairing device for power lines based on machine learning, comprising:

[0059] The acquisition module is used to acquire conductor current signals through an electromagnetically shielded high-frequency current sampling coil;

[0060] The conversion module is used to convert the current signal into a digital signal via a high-precision ADC, and then transmit the information to the processing end via a fiber optic wireless module.

[0061] The decomposition module is used by the FPGA at the processing end to perform wavelet packet decomposition and extract the energy features of five frequency bands.

[0062] The feature comparison module is used to input the energy features of the 5-layer frequency bands into the SVM model for feature comparison and output the matching probability.

[0063] The display module is used to display the matching probability results via a touchscreen, with a green light illuminating when a match is successful.

[0064] The technical solutions provided by the embodiments of this disclosure may include the following beneficial effects:

[0065] This machine learning-based live-line matching method for power lines involves several key components and their functions. Electromagnetically shielded high-frequency current sampling coils accurately acquire conductor current signals in environments with strong electromagnetic interference, while a high-precision ADC converts these current signals into digital signals for subsequent precise processing. A fiber optic wireless module further ensures stable signal transmission, sending data securely and efficiently to the processing end. At the processing end, an FPGA (Field-Programmable Gate Array) extracts energy features from five different frequency bands using wavelet packet decomposition technology; these features are crucial for identification and matching. These features are then input into a Support Vector Machine (SVM) model, which compares the features and outputs the matching probability. Finally, the touchscreen displays the matching result, with a green light indicating a successful match. The entire system works by first acquiring and digitizing conductor current signals, then extracting key features through signal processing, performing matching analysis using a machine learning model, and finally visually displaying the matching results through a user interface.

[0066] This device employs an electromagnetically shielded high-frequency coil sensor to achieve completely non-contact detection, fundamentally avoiding the risks of electric shock and arcing from close contact with high-voltage lines, thus greatly improving operational safety. By integrating fiber optic and wireless communication technologies, and combining wavelet transform and support vector machine (SVM) machine learning algorithms, the system can accurately extract deep time-frequency domain features of current signals, effectively suppressing noise interference in strong electromagnetic environments and improving line identification accuracy. The device integrates an FPGA for real-time processing, ensuring fast response speed, and utilizes supercapacitor power supply to adapt to intermittent operation. Finally, the results are displayed in three dimensions via a touchscreen, demonstrating a high degree of intelligence and significantly reducing reliance on manual labor and the rate of misjudgment, providing reliable technical support for rapid, accurate, and safe alignment of high-voltage lines. Attached Figure Description

[0067] Figure 1 This is a flowchart illustrating a live-line pairing method for power lines based on machine learning, according to an exemplary embodiment of this disclosure. Detailed Implementation

[0068] The specific embodiments of this disclosure will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit this disclosure.

[0069] In this disclosure, unless otherwise stated, directional terms such as "upper," "lower," "front," "rear," "left," and "right" are used for ease of description based on the drawing orientations of the corresponding figures, while "inner" and "outer" are defined based on the contours of the corresponding components themselves. Terms such as "first" and "second" used in this disclosure are used to distinguish one element from another and do not have sequential or importance implications. Furthermore, when the following description refers to the figures, unless otherwise indicated, the same numbers in different figures represent the same or similar elements.

[0070] Please see Figure 1 In a first aspect, this disclosure provides a live-line pairing method for power lines based on machine learning, comprising:

[0071] Step S1: Acquire the conductor current signal through a high-frequency current sampling coil with anti-electromagnetic shielding;

[0072] Step S2: The current signal is converted into a digital signal by a high-precision ADC (high-precision analog-to-digital converter), and the information is sent to the processing terminal through the fiber optic wireless module.

[0073] Step S3: The FPGA at the processing end performs wavelet packet decomposition to extract the energy features of 5 frequency bands;

[0074] Step S4: Input the 5-layer frequency band energy features into the SVM model (support vector machines) for feature comparison and output the matching probability;

[0075] Step S5: Display the matching probability results on the touchscreen. A green light illuminates when a match is successful.

[0076] Understandably, this machine learning-based live-line matching method for power lines involves multiple key components and their functions. Electromagnetically shielded high-frequency current sampling coils are used to accurately acquire conductor current signals in environments with strong electromagnetic interference, while a high-precision ADC is responsible for converting these current signals into digital signals for subsequent precise processing. A fiber optic wireless module further ensures stable signal transmission, sending data securely and efficiently to the processing end. At the processing end, an FPGA (Field-Programmable Gate Array) extracts energy features from five different frequency bands using wavelet packet decomposition technology; these features are crucial for identification and matching. These features are then input into a Support Vector Machine (SVM) model, which compares the features and outputs the matching probability. Finally, the touchscreen displays the matching result, with a green light indicating a successful match. The entire system works by first acquiring and digitizing the conductor current signal, then extracting key features through signal processing, performing matching analysis using a machine learning model, and finally visually displaying the matching results through a user interface.

[0077] This device employs an electromagnetically shielded high-frequency coil sensor to achieve completely non-contact detection, fundamentally avoiding the risks of electric shock and arcing from close contact with high-voltage lines, thus greatly improving operational safety. By integrating fiber optic and wireless communication technologies, and combining wavelet transform and support vector machine (SVM) machine learning algorithms, the system can accurately extract deep time-frequency domain features of current signals, effectively suppressing noise interference in strong electromagnetic environments and improving line identification accuracy. The device integrates an FPGA for real-time processing, ensuring fast response speed, and utilizes supercapacitor power supply to adapt to intermittent operation. Finally, the results are displayed in three dimensions via a touchscreen, demonstrating a high degree of intelligence and significantly reducing reliance on manual labor and the rate of misjudgment, providing reliable technical support for rapid, accurate, and safe alignment of high-voltage lines.

[0078] Optionally, step S1 includes:

[0079] Step S11: Clamp the high-frequency current sampling coil, which is resistant to electromagnetic interference, onto the conductor under test. For the measurement scenario of multi-phase lines, the system adopts a split coil design.

[0080] In step S12, the sensor works based on the principle of electromagnetic induction. When the changing current in the conductor generates a synchronously changing magnetic field, the magnetic field is efficiently coupled by the magnetic core and generates a changing magnetic flux. Finally, a voltage signal that precisely corresponds to the original rate of change of current is induced at the output end of the coil, thus completing the accurate capture of the current characteristics.

[0081] Optionally, step S2 includes:

[0082] Step S21: The original analog signal output by the high-frequency current sensor is processed by a preamplifier with high input impedance and low noise to improve the signal-to-noise ratio and driving capability.

[0083] In step S22, after processing by the preamplifier, the signal enters a dynamic gain adjustment circuit controlled by a microcontroller. This circuit can automatically adapt to different current ranges from milliamps to kiloamps, ensuring that the signal is optimized to the best range suitable for processing by the analog-to-digital converter (ADC).

[0084] Step S23: Before entering the ADC, the signal passes through a high-order active low-pass filter. The cutoff frequency of the filter is precisely set to ensure that the final digital signal can reproduce the original analog characteristics with high fidelity.

[0085] Step S24: Data frames are encapsulated according to a preset standard communication protocol. Each data frame contains specific structural information, laying the foundation for subsequent transmission, identification and verification.

[0086] Step S25: An identification code ID and a high-precision timestamp are embedded in the encapsulated data frame to ensure the traceability of the data source and the accuracy of time synchronization. At the end of the data frame, a cyclic redundancy check code is added to verify the integrity of the data during transmission and to determine whether any errors have occurred.

[0087] Optionally, step S2 further includes:

[0088] Step S25: The encapsulated data is transmitted through gigabit single-mode optical fiber. With the high bandwidth, low latency and natural electromagnetic interference resistance of optical fiber, the integrity and real-time performance of massive high-speed sampled waveform data are ensured.

[0089] In step S26, the system simultaneously constructs a LoRa wireless link as an auxiliary channel. This channel performs real-time analysis on the raw data, extracts key characteristic values ​​such as effective value, peak value, and harmonic content, and then compresses and transmits them.

[0090] Optionally, step S3 includes:

[0091] Step S31: Perform 5-level wavelet packet decomposition using the DB4 wavelet basis to extract energy features across 32 frequency bands.

[0092] The DB4 wavelet is chosen because its tight support and regularity are suitable for capturing local abrupt changes in current signals.

[0093] ;

[0094] Where t represents the time variable; gk represents the coefficients of the 8th-order high-pass filter of DB4; The mother wavelet function is represented by φ, also known as the analytical wavelet; φ represents the scaling function corresponding to the mother wavelet, representing the smoothed or approximate part of the signal. Its function is to generate mother wavelets at different scales and to capture the low-frequency profile of the signal; k represents the discrete translation parameter or sequence index, which is an integer indicating that the wavelet or scaling function is translated on the time axis to cover the entire signal for analysis.

[0095] The amplitude of the current signal x(t) acquired by the ADC is normalized:

[0096] ;

[0097] Where μ represents the arithmetic mean or DC bias of signal x(t) within the analysis time window; σ represents the standard deviation of signal x(t) within the analysis time window; and xnom(t) represents the standardized characteristic signal.

[0098] After decomposing into 5 layers, 32 sub-bands {W} are obtained. 5.1 W 5.2 , … , W5.32}

[0099] The Mallat algorithm is used for 5-level decomposition to extract detail coefficients d. j and approximation coefficient a j :

[0100] ;

[0101] Where j represents the wavelet decomposition level or scale; n represents the discrete-time index; a j [n] represents the approximation coefficients of the decomposition at the j-th level; d j [n] represents the detail coefficients at the j-th level; h[k] represents the DB4 low-pass filter coefficients; g[k] represents the DB4 high-pass filter coefficients; the number of decomposition levels j=1, 2, … , 5 corresponds to the frequency range; d1 ranges from 2.5 to 5 kHz; d3 ranges from 312.5 to 625 Hz; d5 ranges from 78.125 to 156.25 Hz;

[0102] Step S32, calculate the energy proportion E3 of the wavelet coefficients in layer d3:

[0103] ;

[0104] Where k represents the discrete translation parameter or sequence index; j represents the number of layers or scale of the wavelet decomposition;

[0105] The normalized entropy value H3 is calculated as follows:

[0106] ;

[0107] Detecting the local extremum point M5 of the d5 coefficient:

[0108] ;

[0109] The Canny edge detection algorithm is used to locate extreme points;

[0110] Step S33: Calculate the phase difference between the two conductor signals using wavelet coherence:

[0111] ;

[0112] in, W represents the instantaneous phase difference between the current signals of the two conductors in the 5th harmonic frequency band. xy The cross wavelet spectrum represents the d5 layer of the two signals.

[0113] Optionally, step S4 includes:

[0114] Step S41: Combine the above features into a three-dimensional vector:

[0115] ;

[0116] Where F is the original feature vector, E3 represents the dimensionless ratio, M5 represents the amplitude change rate, and Δφ represents the phase angle;

[0117] Normalization process:

[0118] ;

[0119] Among them, F norm This represents the normalized standard feature vector;

[0120] Step S42: The support vector machine (SVM) model pre-trained on the embedded Linux system is compared with the current feature and the line feature library to output the matching probability P (0~1). The system performs a three-level response based on the probability value: output matching probability P (0~1); if P ≥ 95%, it is determined to be "same wire"; if 80% ≤ P < 95%, a yellow warning is triggered and manual review is recommended; if 80% P < 80%, it is determined to be "different wire" and a red alarm is issued.

[0121] Step S43: The model constructs a feature library based on a large-scale experimental sample, collecting more than 1,000 sets of sample data to provide a reliable data foundation for the training and comparison of the SVM model.

[0122] Optionally, step S5 includes:

[0123] Step S51 displays a comparison of the time-domain waveforms of the original electric field signal and the filtered curve, visually presenting the characteristic changes and noise suppression effects before and after signal processing;

[0124] Step S52: The energy proportion of each layer of wavelet decomposition is displayed by frequency domain energy distribution bar chart, and the spatial distribution of electric field intensity around the conductor is reflected by three-dimensional electric field cloud map, with color gradient clearly representing intensity change.

[0125] In step S53, the system automatically generates a matching conclusion of "match successful" or "no match" and marks the precise confidence level to provide a quantitative basis for the diagnostic results.

[0126] Secondly, this disclosure provides a live-line pairing device for power lines based on machine learning, comprising:

[0127] The acquisition module is used to acquire conductor current signals through an electromagnetically shielded high-frequency current sampling coil;

[0128] The conversion module is used to convert the current signal into a digital signal via a high-precision ADC, and then transmit the information to the processing end via a fiber optic wireless module.

[0129] The decomposition module is used by the FPGA at the processing end to perform wavelet packet decomposition and extract the energy features of five frequency bands.

[0130] The feature comparison module is used to input the energy features of the 5-layer frequency bands into the SVM model for feature comparison and output the matching probability.

[0131] The display module is used to display the matching probability results via a touchscreen, with a green light illuminating when a match is successful.

[0132] This invention has been described through embodiments. Those skilled in the art will understand that various changes or equivalent substitutions can be made to these features and embodiments without departing from the spirit and scope of the invention. Furthermore, under the teachings of this invention, these features and embodiments can be modified to adapt to specific situations and materials without departing from the spirit and scope of the invention. Therefore, this invention is not limited to the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of this application are within the protection scope of this invention.

Claims

1. A live-line pairing method for power lines based on machine learning, characterized in that, include: Step S1: Acquire the conductor current signal through a high-frequency current sampling coil with anti-electromagnetic shielding; Step S2: The current signal is converted into a digital signal by a high-precision ADC, and the information is sent to the processing terminal through the fiber optic wireless module. Step S3: The FPGA at the processing end performs wavelet packet decomposition to extract the energy features of 5 frequency bands; Step S4: Input the 5-layer frequency band energy features into the SVM model for feature comparison and output the matching probability; Step S5: Display the matching probability results on the touchscreen. A green light illuminates when a match is successful.

2. The method for live-line connection of power lines according to claim 1, characterized in that, Step S1 includes: Step S11: Clamp the high-frequency current sampling coil, which is resistant to electromagnetic interference, onto the conductor under test. For the measurement scenario of multi-phase lines, the system adopts a split coil design. In step S12, the sensor works based on the principle of electromagnetic induction. When the changing current in the conductor generates a synchronously changing magnetic field, the magnetic field is efficiently coupled by the magnetic core and generates a changing magnetic flux. Finally, a voltage signal that precisely corresponds to the original rate of change of current is induced at the output end of the coil, thus completing the accurate capture of the current characteristics.

3. The method for live-line connection of power lines according to claim 2, characterized in that, Step S2 includes: Step S21: The original analog signal output by the high-frequency current sensor is processed by a preamplifier with high input impedance and low noise to improve the signal-to-noise ratio and driving capability. In step S22, after processing by the preamplifier, the signal enters a dynamic gain adjustment circuit controlled by a microcontroller. This circuit can automatically adapt to different current ranges from milliamps to kiloamps, ensuring that the signal is optimized to the best range suitable for processing by the analog-to-digital converter (ADC). Step S23: Before entering the ADC, the signal passes through a high-order active low-pass filter. The cutoff frequency of the filter is precisely set to ensure that the final digital signal can reproduce the original analog characteristics with high fidelity. Step S24: Data frames are encapsulated according to a preset standard communication protocol. Each data frame contains specific structural information, laying the foundation for subsequent transmission, identification and verification. Step S25: An identification code ID and a high-precision timestamp are embedded in the encapsulated data frame to ensure the traceability of the data source and the accuracy of time synchronization. At the end of the data frame, a cyclic redundancy check code is added to verify the integrity of the data during transmission and to determine whether any errors have occurred.

4. The method for live-line connection of power lines according to claim 3, characterized in that, Step S2 also includes: Step S25: The encapsulated data is transmitted through gigabit single-mode fiber, taking advantage of the high bandwidth, low latency and natural electromagnetic interference resistance of fiber. In step S26, the system simultaneously constructs a LoRa wireless link as an auxiliary channel. This channel performs real-time analysis on the raw data, extracts key characteristic values ​​such as effective value, peak value, and harmonic content, and then compresses and transmits them.

5. The method for live-line connection of power lines according to claim 4, characterized in that, Step S3 includes: Step S31: Perform 5-level wavelet packet decomposition using the DB4 wavelet basis to extract energy features across 32 frequency bands. The DB4 wavelet is chosen because its tight support and regularity are suitable for capturing local abrupt changes in current signals. ； Where t represents the time variable; gk represents the coefficients of the 8th-order high-pass filter of DB4; The mother wavelet function is represented by φ, also known as the analytical wavelet; φ represents the scaling function corresponding to the mother wavelet, representing the smoothed or approximate part of the signal. Its function is to generate mother wavelets at different scales and to capture the low-frequency profile of the signal; k represents the discrete translation parameter or sequence index, which is an integer indicating that the wavelet or scaling function is translated on the time axis to cover the entire signal for analysis. The amplitude of the current signal x(t) acquired by the ADC is normalized: ； Where μ represents the arithmetic mean or DC bias of signal x(t) within the analysis time window; σ represents the standard deviation of signal x(t) within the analysis time window; and xnom(t) represents the standardized characteristic signal. After decomposing into 5 layers, 32 sub-bands {W} are obtained. 5.1 W 5.2 , … , W 5.32 }; The Mallat algorithm is used for 5-level decomposition to extract detail coefficients d. j and approximation coefficient a j : ； Where j represents the wavelet decomposition level or scale; n represents the discrete-time index; a j [n] represents the approximation coefficients of the decomposition at the j-th level; d j [n] represents the detail coefficients at the j-th level; h[k] represents the DB4 low-pass filter coefficients; g[k] represents the DB4 high-pass filter coefficients; the number of decomposition levels j=1, 2, … , 5 corresponds to the frequency range; d1 ranges from 2.5 to 5 kHz; d3 ranges from 312.5 to 625 Hz; d5 ranges from 78.125 to 156.25 Hz; Step S32, calculate the energy proportion E3 of the wavelet coefficients in layer d3: ； Where k represents the discrete translation parameter or sequence index; j represents the number of layers or scale of the wavelet decomposition; The normalized entropy value H3 is calculated as follows: ； Detecting the local extremum point M5 of the d5 coefficient: ； The Canny edge detection algorithm is used to locate extreme points; Step S33: Calculate the phase difference between the two conductor signals using wavelet coherence: ； in, W represents the instantaneous phase difference between the current signals of the two conductors in the 5th harmonic frequency band. xy The cross wavelet spectrum represents the d5 layer of the two signals.

6. The method for live-line connection of power lines according to claim 5, characterized in that, Step S4 includes: Step S41: Combine the above features into a three-dimensional vector: ； Where F is the original feature vector, E3 represents the dimensionless ratio, M5 represents the amplitude change rate, and Δφ represents the phase angle; Normalization process: ； Among them, F norm This represents the normalized standard feature vector; Step S42: The support vector machine (SVM) model pre-trained on the embedded Linux system is compared with the current feature and the line feature library to output the matching probability P (0~1). The system performs a three-level response based on the probability value: output matching probability P (0~1); if P ≥ 95%, it is determined to be "same wire"; if 80% ≤ P < 95%, a yellow warning is triggered and manual review is recommended; if 80% P < 80%, it is determined to be "different wire" and a red alarm is issued. Step S43: The model constructs a feature library based on a large-scale experimental sample, collecting more than 1,000 sets of sample data to provide a reliable data foundation for the training and comparison of the SVM model.

7. The method for live-line connection of power lines according to claim 6, characterized in that, Step S5 includes: Step S51 displays a comparison of the time-domain waveforms of the original electric field signal and the filtered curve, visually presenting the characteristic changes and noise suppression effects before and after signal processing; Step S52: The energy proportion of each layer of wavelet decomposition is displayed by frequency domain energy distribution bar chart, and the spatial distribution of electric field intensity around the conductor is reflected by three-dimensional electric field cloud map, with color gradient clearly representing intensity change. In step S53, the system automatically generates a matching conclusion of "match successful" or "no match" and marks the precise confidence level to provide a quantitative basis for the diagnostic results.

8. A live-line pairing device for power lines based on machine learning, characterized in that, include: The acquisition module is used to acquire conductor current signals through an electromagnetically shielded high-frequency current sampling coil; The conversion module is used to convert the current signal into a digital signal via a high-precision ADC, and then transmit the information to the processing end via a fiber optic wireless module. The decomposition module is used by the FPGA at the processing end to perform wavelet packet decomposition and extract the energy features of five frequency bands. The feature comparison module is used to input the energy features of the 5-layer frequency bands into the SVM model for feature comparison and output the matching probability. The display module is used to display the matching probability results via a touchscreen, with a green light illuminating when a match is successful.

Images