A non-contact network fault identification and classification method and device

Abstract

The application provides a non-contact distribution network fault identification and classification method and device, and belongs to the technical field of distribution network fault detection. The method comprises the following steps: acquiring distance information between a detection device and a tower; adaptively selecting a detection mode based on the distance information, collecting electromagnetic field information around the tower through a non-contact electromagnetic sensor, and converting the electromagnetic field information into characteristic current information; performing hierarchical judgment according to the characteristic current amplitude to distinguish normal, hidden danger and fault states; extracting waveform time-frequency domain features in the fault state, identifying the fault type through a classification model, and outputting a diagnosis result in combination with positioning information. The application supports live and non-contact operation, is suitable for complex environments and overhead lines, has the advantages of safety in detection, strong adaptability, accurate diagnosis and the like, and can realize early warning and intelligent classification of distribution network faults.

Description

Technical Field

[0001] This invention relates to the field of distribution network fault inspection, specifically a non-contact method and device for distribution network fault identification and classification. Background Technology

[0002] As the "last mile" of the power system, the distribution network is a crucial link in delivering electricity to users, and its operational reliability directly affects the safety and quality of electricity consumption for users. However, due to the complex structure, diverse branches, and variable environment of distribution network lines, single-phase grounding faults occur frequently, accounting for 70%-80% of all fault types, posing a serious threat to power supply reliability.

[0003] Existing methods for detecting single-phase grounding faults in distribution networks are mostly intrusive or contact-based. For example, a smart circuit breaker connected in series in the line is used to detect the phase / line voltage, phase / line current, neutral voltage, and neutral current to determine whether a fault has occurred; or a signal is injected into the line during a power outage, and the location of the fault is determined by the feedback of the signal.

[0004] To detect faults using switches, a smart circuit breaker must first be installed in the line to detect, locate, and identify the fault. However, for many older lines, the hardware configuration or intelligence level of the switches is insufficient, resulting in a very low accuracy rate in identifying grounding faults.

[0005] Signal injection methods require power outages for troubleshooting; however, current applications are limited due to high requirements for power supply reliability and the need for shorter outage times. Therefore, a non-intrusive fault detection method is urgently needed. Summary of the Invention

[0006] To address the aforementioned technical issues, this solution provides a non-intrusive power distribution network fault detection method and device, which achieves fault diagnosis and location under energized conditions through non-contact leakage current detection and multi-dimensional fault characteristic analysis.

[0007] A non-contact method for identifying and classifying distribution network faults includes the following steps:

[0008] The mobile platform equipped with the detection device keeps the detection device within a preset range from the target tower; it checks whether the current detection distance meets the preset optimal detection distance; if it does, it enters the normal detection mode; if it does not, it enters the distance correction mode.

[0009] In the conventional detection mode, characteristic current information is directly acquired; in the distance correction mode, characteristic current information is calculated based on the electromagnetic field attenuation model by acquiring two magnetic field signals around the tower at different distance points.

[0010] The state is classified and judged based on the amplitude of the characteristic current information. If it is judged to be a fault state, the waveform feature analysis is performed on the characteristic current information to extract the time-frequency domain features of the characteristic current information. The time-frequency domain features include steady-state features and transient features. The steady-state features include harmonic features.

[0011] The extracted time-frequency domain features are input into a pre-trained classifier, which outputs fault information including fault type and confidence level.

[0012] By combining the location information of the detection device, the faulty tower can be located, and the location information of the faulty tower can be obtained.

[0013] The output includes diagnostic results, including fault information and location information.

[0014] Furthermore, the distance correction mode specifically includes:

[0015] The first magnetic field strength B1 was collected at the first position at a distance from the tower.

[0016] The mobile platform is controlled to move radially along the tower to the second position, the moving distance Δd of the mobile platform is recorded, and the second magnetic field strength B2 is collected; wherein, the radial direction represents the horizontal straight line direction in which the mobile platform moves away from or towards the tower;

[0017] Based on the electromagnetic field attenuation model, a set of equations is established. Substituting the first magnetic field strength, the second magnetic field strength, and the moving distance, the characteristic current value I is obtained as the characteristic current information.

[0018] Furthermore, the step of establishing a system of equations based on the electromagnetic field attenuation model, substituting the first magnetic field strength, the second magnetic field strength, and the moving distance to obtain the characteristic current value I as the characteristic current information includes:

[0019] The electromagnetic field attenuation model is as follows:

[0020] ;

[0021] Substituting the first magnetic field strength into the electromagnetic field attenuation model, we obtain...

[0022] ;

[0023] Substituting the second magnetic field strength into the electromagnetic field attenuation model, we obtain...

[0024] ;

[0025] in, To preset system coefficients, The absolute distance is given; by solving the equations simultaneously, the characteristic current value I is calculated.

[0026] Furthermore, the state classification judgment includes:

[0027] The effective value of the characteristic current information is compared with a first threshold and a second threshold.

[0028] Among them, values ​​below the first threshold are considered normal, values ​​between the first and second thresholds are considered insulation defects, and values ​​above the second threshold are considered fault conditions.

[0029] Furthermore, the classifier is one of a decision tree, support vector machine, or shallow neural network, used to match fault types based on steady-state and transient features.

[0030] A non-contact power distribution network fault identification and classification device, comprising:

[0031] Non-contact signal acquisition module, used to acquire electromagnetic field signals around the tower;

[0032] A distance sensing module is used to obtain the detection distance between the device and the tower;

[0033] The processing module, connected to both the non-contact signal acquisition module and the distance sensing module, is used for:

[0034] Check if the current detection distance meets the preset optimal detection distance; if it does, enter the normal detection mode; if not, enter the distance correction mode.

[0035] In the conventional detection mode, characteristic current information is directly acquired; in the distance correction mode, characteristic current information is calculated based on the electromagnetic field attenuation model by acquiring two magnetic field signals around the tower at different distance points.

[0036] The state is classified and judged based on the amplitude of the characteristic current information. If it is judged to be a fault state, the waveform feature analysis is performed on the characteristic current information to extract the time-frequency domain features of the characteristic current information. The time-frequency domain features include steady-state features and transient features. The steady-state features include harmonic features.

[0037] The extracted time-frequency domain features are input into a pre-trained classifier, which outputs fault information including fault type and confidence level.

[0038] The positioning module is used to acquire the position information of the device and obtain the positioning information of the faulty tower.

[0039] The communication module is used to output diagnostic results, including fault information and location information.

[0040] Furthermore, the distance sensing module includes at least one of laser ranging, ultrasonic ranging, visual ranging, or radar ranging.

[0041] Furthermore, the non-contact signal acquisition module includes at least one of a Rogowski coil, a magnetoresistive sensor, or an electric field sensor.

[0042] Furthermore, the distance correction mode specifically includes:

[0043] The first magnetic field strength B1 was collected at the first position at a distance from the tower.

[0044] The mobile platform is controlled to move radially along the tower to the second position, the moving distance Δd of the mobile platform is recorded, and the second magnetic field strength B2 is collected; wherein, the radial direction represents the horizontal straight line direction in which the mobile platform moves away from or towards the tower;

[0045] Based on the electromagnetic field attenuation model, the following model was established. Substituting the first magnetic field strength, the second magnetic field strength, and the moving distance, the characteristic current value I is obtained as the characteristic current information.

[0046] Furthermore, it also includes a mobile support platform for carrying the non-contact signal acquisition module, distance sensing module, processing module, positioning module, and communication module.

[0047] Furthermore, the mobile carrier platform can be a drone, an inspection robot, or a handheld inspection device.

[0048] The present invention has the following beneficial effects:

[0049] 1. This invention proposes a non-invasive fault detection method. First, leakage current is inverted by analyzing the changes in the electromagnetic field around the tower. No electrical contact is required, and live-line work is supported, which improves the safety and convenience of detection. Second, the detection device of this invention can be mounted on handheld devices, drones or inspection robots, and is suitable for complex environments and high-altitude lines, enabling multi-scenario adaptation.

[0050] 2. This invention proposes a distance correction algorithm that upgrades single ranging to differential gradient detection, which greatly improves the anti-interference capability of the device and enables it to obtain accurate data even in complex environments.

[0051] 2. This invention proposes a fault diagnosis strategy. First, it sets a multi-level threshold of "hidden danger - fault" to separate early warning of hidden dangers from the identification of faults that have already occurred, thereby improving the targeted nature of operation and maintenance. Second, it combines the time-domain characteristics (peak value, effective value) and frequency-domain characteristics (harmonic distribution, transient components) of leakage current to achieve fine classification of fault types (such as distinguishing between transient faults and permanent faults).

[0052] 3. This invention proposes a cloud-based collaborative diagnosis method that combines local preliminary diagnosis with comparison of historical data in the cloud to improve diagnostic accuracy and support fault trend prediction. Attached Figure Description

[0053] Figure 1 This is a schematic diagram showing the position of the drone in the radial direction, either away from or close to the tower.

[0054] Figure 2 This is a flowchart of the non-contact power distribution network fault identification and classification method of the present invention. Detailed Implementation

[0055] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0056] like Figure 2 As shown, this embodiment of the invention provides a non-contact distribution network fault identification and classification method, including the following steps:

[0057] Step 110: Using a mobile platform equipped with the detection device, maintain the detection device within a preset range from the target tower; check whether the current detection distance meets the preset optimal detection distance; if it does, enter the normal detection mode; if it does not, enter the distance correction mode.

[0058] Step 121: Under the normal detection mode, directly collect characteristic current information;

[0059] Step 122: In distance correction mode, the characteristic current information is calculated based on the electromagnetic field attenuation model by collecting two magnetic field signals around the tower at different distance points.

[0060] Step 130: Based on the amplitude of the characteristic current information, a state classification judgment is made. If the judgment is a fault state, waveform feature analysis is performed on the characteristic current information to extract the time-frequency domain features of the characteristic current information. The time-frequency domain features include steady-state features and transient features. The steady-state features include harmonic features.

[0061] Step 140: Input the extracted time-frequency domain features into the pre-trained classifier and output fault information including fault type and confidence level.

[0062] Step 150: Combine the location information of the detection device to locate the faulty tower and obtain the location information of the faulty tower;

[0063] Step 160: Output diagnostic results including fault information and location information.

[0064] To facilitate understanding, the following steps are explained in detail:

[0065] Step 110: Distance Calibration

[0066] The detection device is mounted on platforms such as handheld devices, drones, and inspection robots to keep it at a certain distance from the tower.

[0067] Use laser or ultrasonic ranging to ensure the detection device maintains the optimal detection distance (e.g., 1 meter) between itself and the pole (or line) to avoid electromagnetic interference affecting data accuracy. If measurement can be performed at the optimal detection distance, proceed to step 121 for routine inspection.

[0068] If the optimal detection distance cannot be maintained due to environmental factors (for example, there are many tree branches in mountainous areas, making it difficult for drones to fly in), then switch to step 122 distance correction mode.

[0069] Step 120: Characteristic Current Detection

[0070] Characteristic current information of the spatial electromagnetic field near the tower is collected using a high-sensitivity electromagnetic sensor (such as a Rogowski coil or magnetoresistive sensor) for a continuous period of time. Specifically, the characteristic current detection includes the following two modes:

[0071] Step 121: Routine testing

[0072] When the distance of the detection device meets the detection requirements, the characteristic current of the tower sampled at the preset sampling rate within a period of time (such as 5-10 power frequency cycles) is recorded. (The characteristic current is current data that is strongly correlated with the tower leakage current. Since it is a non-contact measurement, the specific value of the tower leakage current cannot be directly obtained; however, the presence and type of fault can be determined by observing the magnitude and trend of the characteristic current.)

[0073] Step 122: Distance Correction Mode

[0074] Utilizing the physical property that electromagnetic fields decay with distance (e.g.) or By using measurements taken by a drone at two different distance points, a system of equations is established to eliminate distance errors and directly calculate the characteristic current value I.

[0075] For ease of understanding, a specific example is given below:

[0076] 1) The drone hovers at a distance from the tower. At this location (this distance does not need to be the exact optimal detection distance, only the relative position needs to be recorded), measure the magnetic field strength. .

[0077] 2) The drone moves backward or forward a certain distance in a radial direction away from or towards the tower. (Precisely controlled by airborne vision or radar) to arrive Measuring magnetic field strength Wherein, the radial direction represents the straight line connecting the drone and the tower, see [reference needed]. Figure 1 As shown.

[0078] 3) Construct a decay model (for ease of understanding, this is used here) It should be noted that the formula given here is only an example and is not actually limited to the formula form.

[0079] in, As the preset system coefficients at the factory, they are calibrated in advance in the laboratory:

[0080] Before leaving the factory, the testing device is placed under a known current. and known distance Record sensor readings in a standard laboratory environment. According to the formula

[0081] Calculate and store system coefficients :

[0082] this Once the value is calibrated, it is fixed as a constant in the MCU of the detection device and will not be changed during subsequent field operations.

[0083] 4) Solve the simultaneous equations: and .

[0084] At this point, there is no need to know the absolute distance of the initial hovering position. Only the relative displacement needs to be known. (Not only is it easy to measure, but it is also highly accurate), so the current can be calculated. .

[0085] Step 130: Fault Diagnosis

[0086] To perform more detailed fault diagnosis and classification, this step will be divided into two stages:

[0087] Step 131: Amplitude threshold grading;

[0088] The purpose of this sub-step is to quickly distinguish between "insulation hazards" and "actual faults" that have already occurred. Since the follow-up action after identifying the hazard status is to issue a timely warning and notify maintenance personnel to check or replace it, there is no need to make specific classifications.

[0089] Therefore, the amplitude threshold classification can be initially divided into three states:

[0090] Normal state: If the effective value of leakage current is... If the current is consistently below the threshold of 1 (e.g., 100mA), it is considered to be in a normal state.

[0091] ;

[0092] Hazard status: If the effective value of leakage current is consistently higher than (greater than or equal to) threshold 1 (e.g., 100mA) but lower than threshold 2 (e.g., 500mA), it is judged as "insulation hazard"; indicating that the insulation has deteriorated but has not yet triggered the protection action, and planned inspection is required.

[0093] ;

[0094] Fault status: The effective value of leakage current is continuously higher than the threshold 2; this indicates that a ground fault may have occurred, and we should immediately proceed to the next step of waveform characteristic analysis.

[0095]

[0096] Note that the thresholds given here are reference thresholds. In actual applications, they may be similar to but not exactly the same as the thresholds. They need to be set according to the specific situation, but the principle of threshold 1 < threshold 2 must be followed.

[0097] Step 132: Waveform feature analysis;

[0098] When a single-phase ground fault occurs in a distribution network, the nonlinear characteristics of the fault point (such as high-frequency harmonics and intermittent pulses of the arc) will be simultaneously reflected in the zero-sequence current of the line and the ground current of the faulty tower. Although the amplitudes differ, their time-frequency characteristics and transient polarities are of the same origin. Therefore, the fault type (such as arc grounding, high-resistance grounding, and open-circuit grounding) can be distinguished by analyzing the waveform characteristics of the leakage current.

[0099] Starting from the time-domain waveform, harmonic components, and transient characteristics of tower leakage current detected by non-contact methods, a feature library is established for matching based on experimental data and expert research results.

[0100] First, the detected characteristic currents are filtered and their features are extracted:

[0101] Steady-state characteristics: Calculate the effective value and phase (phase difference with reference voltage) within the power frequency cycle after the fault. This can be obtained synchronously through an electric field sensor.

[0102] The steady-state characteristics also include harmonic characteristics, which are extracted through FFT analysis to determine the content and proportion of the 3rd, 5th, and 7th harmonics. For example, arc grounding faults generate abundant high-order harmonics.

[0103] Transient characteristics: Using tools such as wavelet transform, the amplitude, polarity, and spectral distribution of transient currents in the first half-wave or the initial stage of a fault are analyzed. The transient components of high-resistivity grounding faults are usually more significant than the steady-state components.

[0104] Secondly, the rules for classifying faults are determined. Characteristics of some fault types can be found in Table 1. It should be noted that due to the complexity of on-site conditions, actual faults are not limited to the following situations; Table 1 shows some typical faults.

[0105] Table 1:

[0106]

[0107] Step 140: Intelligent Diagnostic Fusion

[0108] The extracted features (transient, steady-state, and harmonic) are input into a lightweight classifier (such as a decision tree, support vector machine (SVM), or shallow neural network; these have low performance requirements, fast decision-making capabilities, and can be integrated into the MCU or host computer of the detection device). The classifier, trained on a large amount of laboratory fault data, can automatically match and output the most probable fault type and its confidence level.

[0109] Step 150: Location and Early Warning

[0110] By combining the location information (GPS / BeiDou) of the detection device, the faulty tower can be accurately located, and the diagnostic results (fault diagnosis and location) can be uploaded to the monitoring center in real time via wireless network.

[0111] Step 160: Output the diagnostic results.

[0112] To facilitate understanding, the following is a specific implementation example of an actual operation process, specifically including a case study of non-contact fault inspection of a mountain power distribution network based on a drone platform:

[0113] 1. System hardware environment setup

[0114] This embodiment uses a drone as a platform to inspect overhead power distribution lines in mountainous areas with a voltage level of 10kV.

[0115] The main hardware components of the system include:

[0116] The main body of the detection device is equipped with a high-sensitivity magnetoresistive sensor (or Rogowski coil) for collecting space magnetic field signals.

[0117] Onboard MCU (Microcontroller Unit): Employs STM32 series chips, with pre-stored system coefficients calibrated in the laboratory. .

[0118] Displacement measurement module: Employs airborne millimeter-wave radar or binocular vision module to accurately measure the radial displacement of the UAV relative to the tower. .

[0119] Communication and positioning module: Integrates Beidou / GPS positioning module and 4G / 5G wireless transmission module.

[0120] 2. Preliminary steps: System coefficients It has already been calibrated before leaving the factory (refer to step 122), so it will not be repeated here.

[0121] 3. On-site operation phase:

[0122] Step 210: The inspection drone flies along the power distribution line. When it reaches the vicinity of the tower to be tested (e.g., tower #15), it hovers beside the tower's grounding wire. The BeiDou module is used to record the current geographical coordinates of the tower.

[0123] Step 220: When a situation is detected where the optimal detection distance cannot be maintained due to obstruction by trees in the mountainous area, the "distance correction mode" is activated;

[0124] First measurement: The drone hovers at its current position (denoted as position A, the actual distance from the tower at this moment). (Unknown), the sensor collects the first magnetic induction intensity Subsequently, the drone was controlled to retreat horizontally in a radial direction away from the tower. The retreat distance was precisely recorded using onboard vision or radar. .

[0125] Second measurement: After the drone reaches location B and hovers, the sensors collect the second magnetic induction intensity. The MCU automatically calculates the effective value of the characteristic current at this moment:

[0126] A system of equations was established based on the attenuation model:

[0127] By solving the system of equations, we get:

[0128]

[0129]

[0130] at this time, , , , All are known quantities; no variables need to be known. The effective value of the characteristic current of the tower grounding wire can be directly calculated. .

[0131] Step 230: Multi-level threshold initial screening:

[0132] MCU will display the RMS value of the characteristic current. Compare with a preset threshold:

[0133] Scenario 1: If ( Once the condition is determined to be "normal", the drone flies to the next pole.

[0134] Scenario 2: If ( The signal was identified as an "insulation hazard" and a yellow warning signal was generated, indicating that insulation deterioration was present.

[0135] Scenario 3: If ( If the condition is determined to be "fault state", high-frequency sampling is immediately triggered, and the waveform analysis process begins.

[0136] Step 240: Waveform Analysis

[0137] When the condition is determined to be a "fault state", the system performs time-frequency domain analysis on the acquired current waveform and performs matching diagnosis based on the classifier:

[0138] For example, if the MCU performs FFT analysis on the waveform and finds that the waveform is severely distorted and contains a large number of 3rd, 5th, and 7th high-frequency harmonics, and at the same time the time-domain waveform shows that the current amplitude is unstable and there are intermittent pulses and beat frequency phenomena, then the background will judge it as an intermittent arc grounding fault.

[0139] Step 250: Result Reporting:

[0140] The detection device packages the data of "#15 tower, intermittent arc grounding, and intermittent arc grounding with a confidence level of 95%" along with the fault waveform data and sends it to the local area or the cloud in real time via the communication module. The data is then further confirmed by humans or by the cloud in combination with historical data for intelligent confirmation, and a maintenance work order is dispatched.

[0141] This invention also provides a non-contact power distribution network fault identification and classification device, comprising:

[0142] Non-contact signal acquisition module, used to acquire electromagnetic field signals around the tower;

[0143] A distance sensing module is used to obtain the detection distance between the device and the tower;

[0144] The processing module, connected to both the non-contact signal acquisition module and the distance sensing module, is used for:

[0145] Check if the current detection distance meets the preset optimal detection distance; if it does, enter the normal detection mode; if not, enter the distance correction mode.

[0146] In the conventional detection mode, characteristic current information is directly acquired; in the distance correction mode, characteristic current information is calculated based on the electromagnetic field attenuation model by acquiring two magnetic field signals around the tower at different distance points.

[0147] The state is classified and judged based on the amplitude of the characteristic current information. If it is judged to be a fault state, the waveform feature analysis is performed on the characteristic current information to extract the time-frequency domain features of the characteristic current information. The time-frequency domain features include steady-state features and transient features. The steady-state features include harmonic features.

[0148] The extracted time-frequency domain features are input into a pre-trained classifier, which outputs fault information including fault type and confidence level.

[0149] The positioning module is used to acquire the position information of the device and obtain the positioning information of the faulty tower.

[0150] The communication module is used to output diagnostic results, including fault information and location information.

[0151] This invention has the following features and effects:

[0152] 1. Enables non-contact live-line detection, improving operational safety and adaptability.

[0153] This invention collects electromagnetic field information around power poles using a non-contact electromagnetic sensing method, eliminating the need for electrical contact or power outage operations. It supports live-line inspections, significantly improving detection safety. This method can be mounted on mobile platforms such as drones and inspection robots, making it suitable for complex terrain and high-altitude lines, thus expanding the application scenarios for power distribution network fault detection.

[0154] 2. A distance-adaptive detection mechanism is proposed to enhance data reliability in complex environments.

[0155] By introducing detection distance sensing and distance correction algorithms, the characteristic current information can still be accurately retrieved even at non-ideal detection distances. This effectively overcomes the impact of environmental obstruction and positioning deviation on the measurement results, and improves the system's anti-interference capability and detection accuracy in complex environments such as the field and mountains.

[0156] 3. Establish a diagnostic strategy that integrates hidden danger-fault classification with waveform characteristics to achieve precise fault identification.

[0157] Multi-level threshold judgments are performed based on characteristic current amplitude to distinguish between normal, potential hazards, and fault states, supporting early warning. Furthermore, time-domain and frequency-domain feature analysis is combined with a classification model to achieve intelligent identification of fault types (such as arc grounding, high-resistance grounding, etc.), improving the accuracy of fault classification and the targeted nature of maintenance responses.

[0158] 4. Supports cloud-based collaboration and precise positioning, enhancing the level of intelligent operation and maintenance.

[0159] Diagnostic results can be linked to location information and uploaded to the monitoring system in real time, supporting rapid location of faulty towers and work order dispatch. Combined with historical cloud data and intelligent analysis, it enables fault trend prediction and diagnostic result verification, constructing an integrated hardware and software intelligent inspection system for power distribution networks.

[0160] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A non-contact method for identifying and classifying distribution network faults, characterized in that, Includes the following steps: The mobile platform carrying the detection device keeps the detection device within a preset range from the target tower; it then checks whether the current detection distance meets the preset optimal detection distance. If the conditions are met, proceed to the normal detection mode; otherwise, proceed to the distance correction mode. In the conventional detection mode, characteristic current information is directly acquired; In distance correction mode, characteristic current information is calculated based on the electromagnetic field attenuation model by collecting two magnetic field signals around the tower at different distance points. The state is classified and judged based on the amplitude of the characteristic current information. If it is judged to be a fault state, the waveform feature analysis is performed on the characteristic current information to extract the time-frequency domain features of the characteristic current information. The time-frequency domain features include steady-state features and transient features. The steady-state features include harmonic features. The extracted time-frequency domain features are input into a pre-trained classifier, which outputs fault information including fault type and confidence level. By combining the location information of the detection device, the faulty tower can be located, and the location information of the faulty tower can be obtained. The output includes diagnostic results with fault information and location information; The distance correction mode specifically includes: The first magnetic field strength B1 was collected at the first position at a distance from the tower. The mobile platform is controlled to move radially along the tower to the second position, the moving distance Δd of the mobile platform is recorded, and the second magnetic field strength B2 is collected; wherein, the radial direction represents the horizontal straight line direction in which the mobile platform moves away from or towards the tower; Based on the electromagnetic field attenuation model, a set of equations is established. Substituting the first magnetic field strength, the second magnetic field strength, and the moving distance, the characteristic current value I is obtained as the characteristic current information.

2. The method according to claim 1, characterized in that, The process of establishing a system of equations based on the electromagnetic field attenuation model, substituting the first magnetic field strength, the second magnetic field strength, and the moving distance to obtain the characteristic current value I as the characteristic current information includes: The electromagnetic field attenuation model is as follows: ； Substituting the first magnetic field strength into the electromagnetic field attenuation model, we obtain... ； Substituting the second magnetic field strength into the electromagnetic field attenuation model, we obtain... ； in, To preset system coefficients, The absolute distance is given; by solving the equations simultaneously, the characteristic current value I is calculated.

3. The method according to claim 1, characterized in that, The state classification judgment includes: The effective value of the characteristic current information is compared with a first threshold and a second threshold. Among them, values ​​below the first threshold are considered normal, values ​​between the first and second thresholds are considered insulation defects, and values ​​above the second threshold are considered fault conditions.

4. The method according to claim 1, characterized in that, The classifier is one of a decision tree, support vector machine, or shallow neural network, used to match fault types based on steady-state and transient features.

5. A non-contact power distribution network fault identification and classification device, characterized in that, include: Non-contact signal acquisition module, used to acquire electromagnetic field signals around the tower; A distance sensing module is used to obtain the detection distance between the device and the tower; The processing module, connected to both the non-contact signal acquisition module and the distance sensing module, is used for: Check if the current detection distance meets the preset optimal detection distance; if it does, enter the normal detection mode; if not, enter the distance correction mode. In the conventional detection mode, characteristic current information is directly acquired; In distance correction mode, characteristic current information is calculated based on the electromagnetic field attenuation model by collecting two magnetic field signals around the tower at different distance points. The state is classified and judged based on the amplitude of the characteristic current information. If it is judged to be a fault state, the waveform feature analysis is performed on the characteristic current information to extract the time-frequency domain features of the characteristic current information. The time-frequency domain features include steady-state features and transient features. The steady-state features include harmonic features. The extracted time-frequency domain features are input into a pre-trained classifier, which outputs fault information including fault type and confidence level. The positioning module is used to acquire the position information of the device and obtain the positioning information of the faulty tower. The communication module is used to output diagnostic results, including fault information and location information; The distance correction mode specifically includes: The first magnetic field strength B1 was collected at the first position at a distance from the tower. The mobile platform is controlled to move radially along the tower to the second position, the moving distance Δd of the mobile platform is recorded, and the second magnetic field strength B2 is collected; wherein, the radial direction represents the horizontal straight line direction in which the mobile platform moves away from or towards the tower; Based on the electromagnetic field attenuation model, a set of equations is established. Substituting the first magnetic field strength, the second magnetic field strength, and the moving distance, the characteristic current value I is obtained as the characteristic current information.

6. The apparatus according to claim 5, characterized in that, The distance sensing module includes at least one of laser ranging, ultrasonic ranging, visual ranging, or radar ranging.

7. The apparatus according to claim 5, characterized in that, The non-contact signal acquisition module includes at least one of a Rogowski coil, a magnetoresistive sensor, or an electric field sensor.

8. The apparatus according to claim 5, characterized in that, It also includes a mobile support platform for carrying the non-contact signal acquisition module, distance sensing module, processing module, positioning module and communication module.

9. The apparatus according to claim 8, characterized in that, The mobile platform can be a drone, an inspection robot, or a handheld inspection device.

Images