Multi-loop DC cable arcing fault line selection method based on discharge signal

Abstract

The invention relates to the technical field of direct-current power distribution networks, and particularly discloses a multi-loop direct-current cable arcing fault line selection method based on discharge signals, which comprises the following steps: acquiring an original signal data set through a high-frequency transient signal acquisition method; performing time-frequency domain feature extraction on the original signal data set to obtain a multi-dimensional feature vector; inputting the multi-dimensional feature vector into a dual-channel deep residual network model to obtain a fault loop candidate set; performing physical field verification on the fault loop candidate set, and matching real-time magnetic field data by adopting a magnetic field fingerprint database generated based on finite element simulation to obtain a verification result; determining a final fault loop according to the verification result, and triggering a fault isolation and alarm signal; discharge signals and temperature gradient data of the multi-loop direct-current cable are synchronously acquired through the high-frequency transient signal acquisition system, and data integrity and synchronism are ensured; and time-frequency domain feature extraction and a dual-channel deep residual network model are combined, so that the accuracy of fault line selection is remarkably improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of direct current (DC) distribution networks, and in particular relates to a method for selecting a multi-circuit DC cable arc fault line based on a discharge signal. Background Art

[0002] In DC distribution networks, multi-circuit DC cables are essential components of power transmission, and their safe and reliable operation is crucial to the stability of the power system. However, due to factors such as long-term operation, environmental corrosion, and mechanical stress, DC cables may experience arcing failures. If not detected and addressed promptly, these failures can lead to serious consequences such as fire and equipment damage.

[0003] Currently, line selection methods for arcing faults in multi-circuit DC cables primarily rely on traditional current protection principles and signal processing technologies. While these traditional current protection principles can detect fault currents to a certain extent, the accuracy of fault line selection is low due to the lack of clear fault current characteristics in DC power grids and their susceptibility to interference from factors such as load fluctuations and the addition of distributed power sources. Furthermore, while traditional signal processing techniques such as Fourier transforms and wavelet transforms can extract the time-frequency characteristics of signals, they often suffer from insufficient feature extraction and low recognition accuracy when processing discharge signals in complex noisy environments.

[0004] Furthermore, existing fault line selection methods mostly rely on a single data source or feature extraction method, lacking comprehensive consideration of the time-frequency characteristics and physical properties of the discharge signal, as well as the combined influence of external environmental factors (such as temperature gradients). Consequently, in the face of complex and changing fault scenarios, existing methods often struggle to accurately identify the fault circuit, resulting in delayed fault isolation and a high rate of false trips.

[0005] In summary, the existing technology has many shortcomings in multi-circuit DC cable arc fault line selection. A more accurate, fast and reliable fault line selection method is needed to meet the needs of DC distribution network for rapid response and precise fault positioning.

[0006] To address this issue, the inventors propose a multi-circuit DC cable arc fault line selection method based on discharge signals to solve the above problems. Summary of the Invention

[0007] The object of the present invention is to provide a method for selecting a multi-circuit DC cable arc fault line based on a discharge signal, so as to solve the problems raised in the above background technology.

[0008] To achieve the above object, the present invention provides the following technical solutions:

[0009] A multi-circuit DC cable arc fault line selection method based on discharge signals includes:

[0010] The discharge signal and temperature gradient data of the multi-circuit DC cable are acquired synchronously by using a high-frequency transient signal acquisition method to obtain the original signal data set.

[0011] Performing time-frequency domain feature extraction on the original signal data set to obtain a multi-dimensional feature vector;

[0012] Inputting the multi-dimensional feature vector into a dual-channel deep residual network model, and obtaining a candidate set of fault circuits by fusing time-frequency image features with physical features;

[0013] Performing physical field verification on the fault circuit candidate set, using a magnetic field fingerprint library generated based on finite element simulation to match real-time magnetic field data to obtain verification results;

[0014] The final fault circuit is determined based on the verification result, and fault isolation and alarm signals are triggered.

[0015] Preferably, the high-frequency transient signal acquisition method comprises the following steps:

[0016] The current waveform data of each loop is obtained by collecting the current signal of each loop through the wide-band Rogowski coil;

[0017] The cable temperature gradient data is collected through the distributed optical fiber temperature measurement unit to obtain the temperature distribution data;

[0018] The current waveform data and temperature distribution data are sampled synchronously by a GPS / IRIG-B time synchronization module to obtain the original signal data set synchronously.

[0019] Preferably, the time-frequency domain feature extraction includes pulse front steepness, wavelet packet energy entropy and recursion graph fractal dimension, and the time-frequency domain feature extraction includes the following steps:

[0020] Perform continuous wavelet transform on the original signal data set to generate time-frequency images;

[0021] Performing singular value decomposition on the time-frequency image to extract principal component features;

[0022] Calculate the fractal dimension and Lyapunov index of the recurrence graph of the original signal data set to obtain the nonlinear eigenvector;

[0023] The time-frequency image, the principal component features, and the nonlinear feature vector are fused into a multi-dimensional feature vector.

[0024] Preferably, the calculation formula of the wavelet packet energy entropy is:

[0025]

[0026] Where pi: energy proportion of the i-th wavelet packet node;

[0027] Ei: energy of the i-th wavelet packet node;

[0028] N: The total number of nodes of wavelet packet decomposition.

[0029] Preferably, the formula of the continuous wavelet transform is:

[0030]

[0031] Where x(t): original discharge signal;

[0032] ψ(t): mother wavelet function;

[0033] a: scale parameter, which controls the expansion and contraction of the wavelet;

[0034] b: translation parameter, controlling the position of the wavelet;

[0035] ψ*: complex conjugate of the mother wavelet.

[0036] Preferably, the step of obtaining a fault loop candidate set through the dual-channel deep residual network model includes:

[0037] Input the time-frequency image into the first channel and extract the time-frequency features through the pre-trained ResNet-50 network;

[0038] Input the multi-dimensional feature vector into the second channel and extract the physical features through the fully connected layer;

[0039] The time-frequency features and the physical features are fused through a cross-channel attention mechanism to obtain a candidate set of faulty circuits.

[0040] Preferably, the calculation formula of the cross-channel attention mechanism is:

[0041]

[0042] Where fi: the i-th eigenvector;

[0043] Wq: learnable weight matrix;

[0044] αi: Attention weight of the ith feature.

[0045] Preferably, the physical field verification includes:

[0046] Build an electromagnetic-thermal coupling model of multi-conductor cables based on finite element simulation and generate a magnetic field fingerprint library;

[0047] The real-time magnetic field data is collected by a magnetometer array, and a dynamic time-warping matching is performed with the magnetic field fingerprint library to obtain a matching degree;

[0048] The verification result is determined based on the matching degree. If the matching degree is greater than 90%, the faulty circuit is confirmed; otherwise, the manual review process is initiated.

[0049] Preferably, the calculation formula of the dynamic time warping is:

[0050]

[0051] Where Q = [q1,q2,…,qm]: query sequence;

[0052] C = [c1, c2, …, cn]: reference sequence;

[0053] π: Optimal path, satisfying boundary conditions, monotonicity and continuity.

[0054] Preferably, the fault isolation and triggering of the alarm signal include:

[0055] Issue fault isolation instructions in real time through the edge computing platform to isolate the fault circuit;

[0056] Generate alarm signals through the SCADA system and record holographic data before and after the fault.

[0057] Compared with the prior art, the present invention has the following beneficial effects:

[0058] (1) The present invention synchronously acquires the discharge signals and temperature gradient data of multi-circuit DC cables through a high-frequency transient signal acquisition system to ensure data integrity and synchronization; combined with time-frequency domain feature extraction and a dual-channel deep residual network (Dual-ResNet) model, the accuracy of fault line selection is significantly improved.

[0059] (2) The present invention uses multi-dimensional feature extraction (such as wavelet packet energy entropy, recursive graph fractal dimension, etc.) to enhance the adaptability to complex noise environments. Real-time fault line selection is achieved through the edge computing platform, and the total time from fault occurrence to isolation completion is low. To meet the DC distribution network's demand for rapid fault isolation, the magnetic field fingerprint library matching verification is introduced to ensure the reliability of fault line selection, avoid misjudgment and missed judgment, and have a low false alarm rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 This is a flow chart of the method for selecting a multi-circuit DC cable arc fault line based on discharge signals of the present invention. DETAILED DESCRIPTION

[0061] The following will provide a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0062] Example 1:

[0063] See also Figure 1 As shown in FIG, a multi-circuit DC cable arc fault line selection method based on discharge signals includes:

[0064] The discharge signal and temperature gradient data of the multi-circuit DC cable are acquired synchronously by using a high-frequency transient signal acquisition method to obtain the original signal data set.

[0065] Performing time-frequency domain feature extraction on the original signal data set to obtain a multi-dimensional feature vector;

[0066] The high-frequency transient signal acquisition method comprises the following steps:

[0067] The current waveform data of each loop is obtained by collecting the current signal of each loop through the wide-band Rogowski coil;

[0068] The cable temperature gradient data is collected through the distributed optical fiber temperature measurement unit (DTS) to obtain temperature distribution data;

[0069] The current waveform data and temperature distribution data are sampled synchronously by a GPS / IRIG-B time synchronization module to ensure that the time error is less than 10ns, and the original signal data set is obtained synchronously;

[0070] The time-frequency domain feature extraction includes pulse front steepness, wavelet packet energy entropy and recurrence graph fractal dimension, and the time-frequency domain feature extraction includes the following steps:

[0071] Perform continuous wavelet transform (CWT) on the original signal data set to generate time-frequency images;

[0072] Performing singular value decomposition (SVD) on the time-frequency image to extract principal component features;

[0073] Calculate the fractal dimension and Lyapunov index of the recurrence graph of the original signal data set to obtain the nonlinear eigenvector;

[0074] fusing the time-frequency image, the principal component features, and the nonlinear feature vector into a multi-dimensional feature vector;

[0075] Inputting the multi-dimensional feature vector into a dual-channel deep residual network (Dual-ResNet) model, and fusing the time-frequency image features with the physical features to obtain a candidate set of fault circuits;

[0076] The time-frequency image is input into the first channel, and the time-frequency features are extracted through the pre-trained ResNet-50 network.

[0077] The multi-dimensional feature vector is input into the second channel and physical features are obtained by extraction through a fully connected layer.

[0078] The time-frequency features and the physical features are fused through a cross-channel attention mechanism (CCA) to obtain a candidate set of fault circuits, including:

[0079] The time-frequency image is input into the first channel, and the time-frequency features are extracted through the pre-trained ResNet-50 network.

[0080] The multi-dimensional feature vector is input into the second channel and physical features are obtained by extraction through a fully connected layer.

[0081] fusing the time-frequency features and the physical features through a cross-channel attention mechanism (CCA) to obtain a candidate set of faulty circuits;

[0082] Performing physical field verification on the fault circuit candidate set, using a magnetic field fingerprint library generated based on finite element simulation to match real-time magnetic field data to obtain verification results;

[0083] The physical field verification includes:

[0084] Build an electromagnetic-thermal coupling model of multi-conductor cables based on finite element simulation and generate a magnetic field fingerprint library;

[0085] Real-time magnetic field data is collected through a magnetometer array, and dynamic time warping (DTW) matching is performed with the magnetic field fingerprint library to obtain a matching degree;

[0086] Determine the verification result based on the matching degree. If the matching degree is greater than 90%, the faulty circuit is confirmed; otherwise, a manual review process is initiated.

[0087] Determine the final fault circuit based on the verification result, and trigger fault isolation and alarm signals;

[0088] The fault isolation and alarm signal triggering include:

[0089] Issue fault isolation instructions in real time through the edge computing platform to isolate the fault circuit;

[0090] Generate alarm signals through the SCADA system and record holographic data before and after the fault.

[0091] Specifically, the time-frequency domain feature extraction includes pulse front steepness, wavelet packet energy entropy and recursive graph fractal dimension, and the time-frequency domain feature extraction includes the following steps:

[0092] Perform continuous wavelet transform (CWT) on the original signal data set to generate time-frequency images;

[0093] Performing singular value decomposition (SVD) on the time-frequency image to extract principal component features;

[0094] Calculate the fractal dimension and Lyapunov index of the recurrence graph of the original signal data set to obtain the nonlinear eigenvector;

[0095] The time-frequency image, the principal component features, and the nonlinear feature vector are fused into a multi-dimensional feature vector.

[0096] Specifically, the calculation formula of the wavelet packet energy entropy (WPE) is:

[0097]

[0098] Where pi: energy proportion of the i-th wavelet packet node;

[0099] Ei: energy of the i-th wavelet packet node;

[0100] N: the total number of nodes of wavelet packet decomposition;

[0101] The above formula is used to quantify the complexity of the signal's energy distribution in the frequency domain, characterizing the randomness and irregularity of the discharge signal. It serves as a frequency domain feature to distinguish normal signals from arcing fault signals.

[0102] Specifically, the formula of the continuous wavelet transform is:

[0103]

[0104] Where x(t): original discharge signal;

[0105] ψ(t): mother wavelet function (such as Morlet wavelet);

[0106] a: scale parameter, which controls the expansion and contraction of the wavelet;

[0107] b: translation parameter, controlling the position of the wavelet;

[0108] ψ*: complex conjugate of mother wavelet;

[0109] Convert the time domain signal into a time-frequency domain image to capture the transient characteristics of the discharge signal; provide time-frequency localization information of the high-frequency discharge pulse, laying the foundation for subsequent feature extraction.

[0110] Specifically, the calculation formula of the cross-channel attention mechanism is:

[0111]

[0112] Where fi: the i-th eigenvector (time-frequency feature or physical feature);

[0113] Wq: learnable weight matrix;

[0114] αi: attention weight of the i-th feature;

[0115] Dynamically weighted fusion of time-frequency features and physical features improves feature expression capabilities, enhances the model's ability to process multi-source heterogeneous data, and improves the accuracy of fault line selection.

[0116] Specifically, the calculation formula of the dynamic time warping is:

[0117]

[0118] Where Q = [q1,q2,…,qm]: query sequence (real-time magnetic field data);

[0119] C = [c1, c2, ..., cn]: reference sequence (magnetic field fingerprint library data);

[0120] π: optimal path, satisfying boundary conditions, monotonicity and continuity;

[0121] Calculate the similarity between real-time magnetic field data and the magnetic field fingerprint library; provide a basis for physical field verification to ensure the reliability of fault line selection.

[0122] As can be seen from the above, the discharge signals and temperature gradient data of multi-circuit DC cables are synchronously acquired through the high-frequency transient signal acquisition system, ensuring data integrity and synchronization. Combined with time-frequency domain feature extraction and the dual-channel deep residual network (Dual-ResNet) model, the accuracy of fault line selection is significantly improved.

[0123] Multi-dimensional feature extraction (such as wavelet packet energy entropy and recursive graph fractal dimension) enhances adaptability to complex noisy environments. Real-time fault line selection is achieved through an edge computing platform, minimizing the total time from fault occurrence to isolation. To meet the DC distribution network's need for rapid fault isolation, the system incorporates magnetic field fingerprint matching verification to ensure reliable fault line selection, avoid misjudgments and missed detections, and minimize false alarm rates.

[0124] Example 2:

[0125] DC microgrid ring main unit arc fault line selection:

[0126] Application scenario system configuration:

[0127] 4 DC cables (±750V), total length 500 meters.

[0128] Loose mechanical contacts in the ring network cabinet cause arcing faults.

[0129] Sensor deployment:

[0130] Wideband Rogowski coil (bandwidth 0.1Hz-50MHz).

[0131] Distributed fiber optic temperature measurement unit (DTS), with a spatial resolution of 1 meter.

[0132] Magnetometer array (16 sensors, covering the perimeter of the ring main unit). Implementation steps Signal acquisition and preprocessing:

[0133] Sampling frequency: 10MHz, synchronization error <10ns.

[0134] The collected fault current waveform amplitude is: ±50A, and the discharge pulse width is: 5μs.

[0135] Temperature gradient data: The temperature near the fault point rises by 10°C.

[0136] Feature extraction:

[0137] Time-frequency domain features:

[0138] Wavelet packet energy entropy (WPE): WPE for the faulty circuit = 1.8, and WPE for the normal circuit = 0.6.

[0139] Fractal dimension of the recurrence graph: fault loop = 2.3, normal loop = 1.5.

[0140] Nonlinear characteristics:

[0141] Lyapunov index: faulty loop = 0.12, normal loop = 0.02.

[0142] Intelligent recognition:

[0143] Dual-ResNet input:

[0144] Time-frequency image (256×256RGB).

[0145] Multidimensional feature vector (dimension 32).

[0146] Output: Faulty circuit candidate set (confidence 98.5%).

[0147] Physics Verification:

[0148] Magnetic field fingerprint library matching:

[0149] The DTW similarity between real-time magnetic field data and the simulated fingerprint library is 92%.

[0150] Confirm the faulty circuit: Circuit 2.

[0151] Fault isolation and alarm:

[0152] The edge computing platform issues isolation instructions with a delay of <50ms.

[0153] As can be seen from the above, the SCADA system generates alarm signals and records holographic data before and after the fault. The accuracy rate is 98.5%.

[0154] Noise resistance: Under 30dB Gaussian white noise, the accuracy remains above 95%.

[0155] Response time: From fault occurrence to isolation completion, the total time is less than 100ms.

[0156] As can be seen from the above, the combination of time-frequency domain characteristics, nonlinear characteristics, and physical field characteristics provides comprehensive fault characterization capabilities, breaking through the limitations of traditional single criteria and adapting to complex working conditions. Intelligent identification is achieved through a dual-channel deep residual network (Dual-ResNet). Combined with magnetic field fingerprint library matching and verification, a three-level closed-loop mechanism of "signal characteristics-intelligent identification-physical verification" is formed.

[0157] This method has been successfully applied in DC microgrids and flexible DC distribution networks, verifying its feasibility and superiority in practical projects. It can be extended to DC systems with higher voltage levels and more complex topologies.

[0158] The degree of contact degradation is assessed based on the accumulated discharge energy, triggering preventive maintenance strategies to achieve full life cycle coverage from fault line selection to equipment health management; fast and accurate fault line selection reduces power outage time and equipment damage, lowering operation and maintenance costs.

[0159] Intelligent operation and maintenance reduces the frequency of manual inspections and improves operational efficiency; it integrates cutting-edge technologies such as edge computing, deep learning, and digital twins to provide technical support for the intelligent development of DC distribution networks.

[0160] Example 3:

[0161] Flexible DC distribution network cable arc fault line selection:

[0162] Application scenario system configuration:

[0163] 6 DC cables (±10kV), total length 2km.

[0164] Carbonization of cable insulation causes arcing failure.

[0165] Sensor deployment:

[0166] Wideband Rogowski coil (bandwidth 0.1Hz-50MHz).

[0167] Distributed fiber optic temperature measurement unit (DTS), with a spatial resolution of 0.5 meters.

[0168] Magnetometer array (24 sensors, covering the entire cable line). Implementation steps Signal acquisition and preprocessing:

[0169] Sampling frequency: 20MHz, synchronization error <10ns.

[0170] The collected fault current waveform amplitude is: ±200A, and the discharge pulse width is: 10μs.

[0171] Temperature gradient data: The temperature near the fault point rises by 15°C.

[0172] Feature extraction:

[0173] Time-frequency domain features:

[0174] Wavelet packet energy entropy (WPE): WPE for the faulty circuit = 2.1, and WPE for the normal circuit = 0.7.

[0175] Fractal dimension of the recurrence graph: fault loop = 2.5, normal loop = 1.6.

[0176] Nonlinear characteristics:

[0177] Lyapunov index: faulty loop = 0.15, normal loop = 0.03.

[0178] Intelligent recognition:

[0179] Dual-ResNet input:

[0180] Time-frequency image (256×256RGB).

[0181] Multidimensional feature vector (dimension 32).

[0182] Output: Faulty circuit candidate set (confidence 99.2%).

[0183] Physics Verification:

[0184] Magnetic field fingerprint library matching:

[0185] The DTW similarity between real-time magnetic field data and the simulated fingerprint library is 94%.

[0186] Confirm the faulty circuit: circuit 4.

[0187] Fault isolation and alarm:

[0188] The edge computing platform issues isolation instructions with a delay of <50ms.

[0189] As can be seen from the above, the SCADA system generates alarm signals and records holographic data before and after the fault. The accuracy rate is 99.2%.

[0190] Noise resistance: Under 30dB Gaussian white noise, the accuracy remains above 96%.

[0191] Response time: From fault occurrence to isolation completion, the total time is less than 100ms.

[0192] In the description of this specification, the reference terms "one embodiment", "some embodiments", "examples", "specific examples" or "some examples" mean that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and features of different embodiments or examples without contradiction.

[0193] In the drawings of the embodiments disclosed in the present invention, only the structures related to the embodiments disclosed in the present invention are involved. Other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of the present invention can be combined with each other.

[0194] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A multi-circuit DC cable arc fault line selection method based on discharge signals, characterized in that: The following steps are involved: The discharge signal and temperature gradient data of the multi-circuit DC cable are acquired synchronously by using a high-frequency transient signal acquisition method to obtain the original signal data set. Performing time-frequency domain feature extraction on the original signal data set to obtain a multi-dimensional feature vector; Inputting the multi-dimensional feature vector into a dual-channel deep residual network model, and fusing the time-frequency image features with the physical features to obtain a candidate set of fault circuits; Performing physical field verification on the fault circuit candidate set, using a magnetic field fingerprint library generated based on finite element simulation to match real-time magnetic field data to obtain verification results; The final fault circuit is determined based on the verification result, and fault isolation and alarm signals are triggered.

2. The method for selecting a multi-circuit DC cable arc fault line based on discharge signals according to claim 1, characterized in that: The high-frequency transient signal acquisition method comprises the following steps: The current signals of each loop are collected by a wide-band Rogowski coil to obtain current waveform data; The cable temperature gradient data is collected through the distributed optical fiber temperature measurement unit to obtain the temperature distribution data; The current waveform data and temperature distribution data are sampled synchronously by a GPS / IRIG-B time synchronization module to obtain the original signal data set synchronously.

3. The method for selecting a multi-circuit DC cable arc fault line based on discharge signals according to claim 1, characterized in that: The time-frequency domain feature extraction includes pulse front steepness, wavelet packet energy entropy and recurrence graph fractal dimension, and the time-frequency domain feature extraction includes the following steps: Perform continuous wavelet transform on the original signal data set to generate time-frequency images; Performing singular value decomposition on the time-frequency image to extract principal component features; Calculate the fractal dimension and Lyapunov index of the recurrence graph of the original signal data set to obtain the nonlinear eigenvector; The time-frequency image, the principal component features, and the nonlinear feature vector are fused into a multi-dimensional feature vector.

4. The method for selecting a multi-circuit DC cable arc fault line based on discharge signals according to claim 3, characterized in that: The calculation formula of the wavelet packet energy entropy is: Where pi: energy proportion of the i-th wavelet packet node; Ei: energy of the i-th wavelet packet node; N: The total number of nodes of wavelet packet decomposition.

5. The method for selecting a multi-circuit DC cable arc fault line based on discharge signals according to claim 3, characterized in that: The formula of the continuous wavelet transform is: Where x(t): original discharge signal; ψ(t): mother wavelet function; a: scale parameter, which controls the expansion and contraction of the wavelet; b: translation parameter, controlling the position of the wavelet; ψ*: complex conjugate of the mother wavelet.

6. The method for selecting a multi-circuit DC cable arc fault line based on discharge signals according to claim 3, characterized in that: The step of obtaining a fault loop candidate set by using the dual-channel deep residual network model includes: Input the time-frequency image into the first channel and extract the time-frequency features through the pre-trained ResNet-50 network; Input the multi-dimensional feature vector into the second channel and extract the physical features through the fully connected layer; The time-frequency features and the physical features are fused through a cross-channel attention mechanism to obtain a candidate set of faulty circuits.

7. The method for selecting a multi-circuit DC cable arc fault line based on discharge signals according to claim 6, characterized in that: The calculation formula of the cross-channel attention mechanism is: Where fi: the i-th eigenvector; Wq: learnable weight matrix; αi: Attention weight of the ith feature.

8. The method for selecting a multi-circuit DC cable arc fault line based on discharge signals according to claim 1, characterized in that: The physical field verification includes: Build an electromagnetic-thermal coupling model of multi-conductor cables based on finite element simulation and generate a magnetic field fingerprint library; The real-time magnetic field data is collected by a magnetometer array, and a dynamic time-warping matching is performed with the magnetic field fingerprint library to obtain a matching degree; The verification result is determined based on the matching degree. If the matching degree is greater than 90%, the faulty circuit is confirmed; otherwise, the manual review process is initiated.

9. The method for selecting a multi-circuit DC cable arc fault line based on discharge signals according to claim 8, characterized in that: The calculation formula of the dynamic time warping is: Where Q = [q1,q2,…,qm]: query sequence; C = [c1, c2, …, cn]: reference sequence; π: Optimal path, satisfying boundary conditions, monotonicity and continuity.

10. The method for selecting a multi-circuit DC cable arc fault line based on discharge signals according to claim 1, characterized in that: The fault isolation and alarm signal triggering include: Fault isolation instructions are issued in real time through the edge computing platform to isolate the fault circuit; alarm signals are generated through the SCADA system, and holographic data before and after the fault is recorded.

Images