An intermittent arc fault sensing method for a distribution box

Abstract

This invention discloses a method for detecting intermittent arc faults in distribution boxes, relating to the field of distribution box operation status monitoring technology. The method includes: collecting distribution box operation data; extracting intermittent disturbance characteristic data; establishing constraint relationships between normal load voltage drop, line impedance voltage drop, and suspected arc channel voltage drop based on the distribution box operation data and intermittent disturbance characteristic data, and calculating arc channel residual characteristic data; inputting the intermittent disturbance characteristic data and arc channel residual characteristic data into an arc fault state prediction model to obtain arc fault prediction data; and correcting the arc fault prediction data based on acquisition uncertainty, prediction uncertainty, load confusion uncertainty, and intermittent consistency evaluation results to obtain fault detection result data. This invention can reduce false alarms caused by ordinary load disturbances and acquisition anomalies, improving the accuracy and reliability of intermittent arc fault identification.

Description

Technical Field

[0001] This invention relates to the field of power distribution box operation status monitoring technology, specifically to a method for detecting intermittent arc faults in power distribution boxes. Background Technology

[0002] As a crucial piece of equipment in low-voltage power distribution systems, distribution boxes undertake functions such as power distribution, circuit control, and electrical protection. Their operational status directly affects the safety of end-user equipment and power lines. During long-term operation, intermittent arcing faults may occur in the wiring terminals, circuit breaker contacts, branch connection points, or wire connections within the distribution box due to factors such as loose contact, insulation aging, moisture contamination, mechanical vibration, or overload heating. These faults are typically characterized by short duration, unstable frequency, discontinuous waveform disturbances, and weak early-stage energy. If not identified in time, they may further lead to localized overheating, insulation breakdown, equipment burnout, or even electrical fires.

[0003] Existing fault detection methods for distribution boxes mostly rely on sudden current changes, high-frequency energy, residual current changes, temperature rise, or protection device operation records for judgment, which can detect obvious short circuits, continuous arcs, or overload anomalies to a certain extent.

[0004] However, for intermittent arc faults, the abnormal signals are often similar to those of ordinary load start-up and shutdown, capacitive load switching, motor starting, contactor operation, or noise acquisition. Relying on a single threshold or single feature can easily lead to false alarms or missed alarms. For example, when an air conditioner compressor starts, it may generate short-term current surges and voltage dips, which can easily be misjudged as arc disturbances. On the other hand, early intermittent arcs caused by poor terminal contact may only generate high-frequency disturbances and phase anomalies within a few half-cycles, which can easily be ignored by ordinary threshold detection methods.

[0005] Furthermore, existing methods typically lack a comprehensive understanding of the electrical and physical relationships between the normal load response of the target circuit, line impedance voltage drop, and suspected arc channel voltage drop, making it difficult to determine whether abnormal disturbances truly conform to the characteristics of arc channel formation. Simultaneously, some model-based identification methods directly use the model output as the final alarm result, lacking comprehensive correction for sampling gaps, time synchronization deviations, sensor anomalies, load confusion, and intermittent repetitive consistency issues. This results in insufficient reliability of alarm results in field applications, making it difficult for maintenance personnel to accurately locate suspected faulty circuits and prioritize their handling based on alarm information.

[0006] Therefore, there is an urgent need for a method for detecting intermittent arc faults in distribution boxes that can integrate multi-source operating data, intermittent disturbance characteristics, physical residuals of arc channels, and result reliability correction, so as to improve the accuracy, anti-interference ability, and engineering usability of early intermittent arc fault identification. Summary of the Invention

[0007] In view of the above situation and to overcome the shortcomings of the prior art, the technical solution adopted by the present invention is as follows: The present invention provides a method for detecting intermittent arc faults in a distribution box, the method comprising the following steps:

[0008] Step S1: Data acquisition of the power distribution box operation;

[0009] Step S2: Extraction of intermittent perturbation features;

[0010] Step S3: Calculation of arc channel residuals;

[0011] Step S4: Arc fault state prediction;

[0012] Step S5: Fault perception result correction.

[0013] Further, in step S1, the distribution box operation data acquisition is used to obtain the basic operation data required for the intermittent arc fault detection of the distribution box. Specifically, it involves collecting voltage data, current data, residual current data, switch status data, box temperature data, and circuit identification data of the target circuit in the distribution box, and performing time base unification, circuit attribution matching, and abnormal collection point sorting on the collected data to obtain the distribution box operation data.

[0014] The operating data of the distribution box specifically includes voltage waveform data, current waveform data, residual current change data, switch action status data, box internal temperature change data, acquisition time identifier data, and circuit number identifier data.

[0015] Further, in step S2, the intermittent disturbance feature extraction is used to extract electrical features that can characterize short-term disturbances, repetitive disturbances, and phase disturbances from the distribution box operation data. Specifically, based on the distribution box operation data, a transient analysis window is constructed, and current mutation features, voltage distortion features, high-frequency disturbance features, zero-crossing phase features, and disturbance repetition features are extracted within the transient analysis window to obtain intermittent disturbance feature data.

[0016] The intermittent disturbance characteristic data specifically includes current abrupt change characteristics, voltage distortion characteristics, high-frequency disturbance characteristics, zero-crossing phase characteristics, and disturbance repetition characteristics.

[0017] Further, in step S3, the arc channel residual calculation is used to determine whether the abnormal disturbances in the intermittent disturbance characteristic data conform to the electrical and physical relationships in the arc channel formation process. Specifically, based on the distribution box operation data and the intermittent disturbance characteristic data, a constraint relationship is established between the normal load voltage drop, line impedance voltage drop, and suspected arc channel voltage drop of the target circuit. The arc channel residual is calculated based on the degree of deviation between the measured voltage and current and the constraint relationship to obtain the arc channel residual characteristic data, including the following steps:

[0018] Step S31: Candidate disturbance window determination, used to determine the analysis object that needs to be calculated for arc channel residuals. Specifically, based on the current mutation characteristics, high-frequency disturbance characteristics, voltage distortion characteristics and zero-crossing phase characteristics in the intermittent disturbance characteristic data, transient analysis windows that meet the disturbance intensity conditions and phase occurrence conditions are screened to obtain candidate disturbance window data.

[0019] Step S32: Reference window matching, used to determine the normal operation reference state corresponding to the candidate disturbance window. Specifically, based on the circuit number, switch status, current RMS value, load change trend and acquisition time identifier, adjacent operating windows without obvious disturbances are selected from the distribution box operation data to obtain normal reference window data.

[0020] Step S33: Load impedance deviation calculation, used to determine whether the voltage and current relationship in the candidate disturbance window deviates from the normal load relationship. Specifically, it calculates the normal equivalent impedance of the target circuit based on the normal reference window data, and compares the equivalent impedance in the candidate disturbance window with the normal equivalent impedance to obtain the normal load impedance deviation.

[0021] Step S34: Line voltage drop residual calculation, used to isolate the normal voltage drop response caused by the equivalent resistance and equivalent inductance of the line. Specifically, the line impedance voltage drop is calculated based on the line equivalent parameters, current change and current change rate of the target circuit, and the measured voltage disturbance is compared with the line impedance voltage drop to obtain the line voltage drop residual.

[0022] Step S35: Calculate the equivalent residual of the arc channel, which is used to characterize the suspected arc channel voltage drop that cannot be explained by the normal load response and line voltage drop response in the candidate disturbance window. Specifically, based on the deviation relationship between the normal load response, line impedance voltage drop and measured voltage disturbance, calculate the equivalent residual of the arc channel, the residual of current change and the residual of voltage change to obtain the residual data of the arc channel.

[0023] Step S36: Residual persistence processing, used to characterize the persistence and repetition of arc channel residuals in multiple transient analysis windows. Specifically, based on the time sequence, zero-crossing phase and residual amplitude changes of candidate disturbance windows, the residual duration, residual repetition count and residual phase consistency are statistically analyzed to obtain arc channel residual characteristic data.

[0024] The residual characteristic data of the arc channel specifically includes the deviation of normal load impedance, line voltage drop residual, arc channel equivalent residual, current change residual, voltage change residual, and residual persistence characteristics.

[0025] Further, in step S4, the arc fault state prediction is used to identify the arc fault state of the target circuit based on intermittent disturbance characteristic data and arc channel residual characteristic data. Specifically, the intermittent disturbance characteristic data and arc channel residual characteristic data are input into the arc fault state prediction model, and the arc fault state prediction model is used to classify and predict the normal operation state, ordinary load disturbance state, suspected arc state and arc fault state of the target circuit to obtain arc fault prediction data.

[0026] The arc fault prediction data specifically includes the target circuit fault category, fault occurrence time window, fault prediction probability, suspected arc disturbance segment, and initial fault level.

[0027] Further, in step S5, the fault perception result correction is used to correct the credibility of the arc fault prediction data and confirm intermittent faults. Specifically, based on the distribution box operation data, intermittent disturbance characteristic data, arc channel residual characteristic data, and arc fault prediction data, the signal acquisition uncertainty, model prediction uncertainty, load confusion uncertainty, and intermittent consistency evaluation results are calculated. The arc fault prediction data is then corrected based on these results to obtain the fault perception result data, including the following steps:

[0028] Step S51: Calculate the uncertainty of data acquisition, which is used to evaluate the reliability of the power distribution box operation data within the current fault prediction window. Specifically, it calculates the signal acquisition uncertainty based on the sampling missing situation, sampling noise level, time synchronization deviation and sensor abnormality indicators to obtain acquisition uncertainty data.

[0029] Step S52: Calculate the prediction uncertainty, which is used to evaluate the stability of the output results of the arc fault state prediction model. Specifically, it calculates the prediction uncertainty of the model based on the prediction probability distribution of different fault categories, the change of prediction results in adjacent windows, and the probability difference between the suspected arc state and the arc fault state, and obtains the prediction uncertainty data.

[0030] Step S53: Load confusion uncertainty calculation, used to evaluate whether the current suspected disturbance may be caused by ordinary load operation. Specifically, the load confusion uncertainty is calculated based on the switch operation status, load current change trend, high-frequency disturbance duration and arc channel residual characteristic data to obtain load confusion uncertainty data.

[0031] Step S54: Intermittent consistency evaluation, used to confirm whether the suspected arc disturbance has intermittent repetition and physical residual consistency characteristics. Specifically, the intermittent consistency evaluation results are calculated based on the number of residual repetitions, residual persistence characteristics, zero-crossing phase characteristics and adjacent disturbance intervals in the suspected arc disturbance segment, and intermittent consistency data is obtained.

[0032] Step S55: Fault credibility correction, used to correct the credibility of arc fault prediction data. Specifically, the fault credibility is calculated based on the fault prediction probability, signal acquisition uncertainty, model prediction uncertainty, load confusion uncertainty and intermittent consistency evaluation results to obtain corrected credibility data.

[0033] Step S56: Fault perception result output, used to generate the final intermittent arc fault perception result of the distribution box. Specifically, based on the correction confidence data, the initial fault level and the suspected arc disturbance segment, the corrected fault category, the intermittent arc confirmation result, the suspected fault circuit, the fault time period, the fault risk level and the operation and maintenance handling prompt information are determined to obtain the fault perception result data.

[0034] The fault perception result data specifically includes the corrected fault category, fault confidence, intermittent arc confirmation result, suspected fault circuit, fault time period, fault risk level, and operation and maintenance handling prompts.

[0035] The beneficial effects achieved by the present invention using the above solution are as follows:

[0036] (1) In view of the technical problems in the existing methods for detecting arc faults in distribution boxes, which mainly rely on single current mutations, high-frequency energy or threshold alarms for judgment, it is difficult to distinguish between intermittent arcs, ordinary load disturbances, switch action disturbances and acquisition anomalies, resulting in a high risk of missed and false alarms, this solution creatively adopts a comprehensive perception process that combines distribution box operation data acquisition, intermittent disturbance feature extraction, arc channel residual calculation, arc fault state prediction and fault perception result correction; by jointly analyzing multi-source operation data, short-term disturbance features, repetitive phase features, physical residual features and credibility correction results, the fault judgment is transformed from single-point threshold judgment to multi-dimensional evidence fusion judgment, thereby improving the accuracy and stability of intermittent arc fault identification;

[0037] (2) In view of the technical problem that existing arc fault identification methods directly use high-frequency disturbances or current changes as arc criteria, and fail to determine whether abnormal disturbances conform to the physical relationship of voltage, current and line voltage drop in the arc channel formation process, which makes it easy for ordinary load actions to be misidentified as arc faults, this solution creatively adopts the arc channel residual calculation method; by matching the normal reference window, the normal load impedance deviation, line voltage drop residual and arc channel equivalent residual are calculated, and the normal load response and line voltage drop response are separated from the measured voltage disturbance, thereby highlighting the suspected arc channel components that cannot be explained by conventional electrical relationships, and improving the ability to identify real arc disturbances;

[0038] (3) In view of the technical problems in the existing distribution box fault prediction methods, the model output results are directly used as the final alarm results without fully considering sampling missing, time synchronization deviation, sensor abnormality, unstable category probability, common load confusion and intermittent repetition consistency, resulting in insufficient credibility of on-site alarms and difficulty in suppressing false alarms. This solution creatively adopts the fault perception result correction method. By calculating the uncertainty of signal acquisition, the uncertainty of model prediction, the uncertainty of load confusion and the intermittent consistency evaluation results, the credibility of arc fault prediction data and the intermittent fault confirmation are performed. Thus, the final output results not only include the fault category, but also the fault credibility, suspected fault circuit, fault time period, risk level and operation and maintenance handling prompts, thereby improving the interpretability and handleability of alarm results. Attached Figure Description

[0039] Figure 1 A flowchart illustrating a method for detecting intermittent arc faults in a distribution box provided by the present invention;

[0040] Figure 2 This is a flowchart illustrating the calculation of the arc channel residual in step S3.

[0041] Figure 3 This is a flowchart illustrating the process of correcting the fault perception results in step S5.

[0042] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. Detailed Implementation

[0043] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0044] In the description of this invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0045] Example 1, see Figure 1 The present invention provides a method for detecting intermittent arc faults in a distribution box, the method comprising the following steps:

[0046] Step S1: Data acquisition of the power distribution box operation;

[0047] Step S2: Extraction of intermittent perturbation features;

[0048] Step S3: Calculation of arc channel residuals;

[0049] Step S4: Arc fault state prediction;

[0050] Step S5: Fault perception result correction.

[0051] By performing the above operations, this solution addresses the technical problem in existing distribution box arc fault detection methods, which rely primarily on single current surges, high-frequency energy, or threshold alarms for judgment, making it difficult to distinguish between intermittent arcs, ordinary load disturbances, switch action disturbances, and data acquisition anomalies, leading to a high risk of missed and false alarms. This solution creatively adopts a comprehensive sensing process that combines distribution box operation data acquisition, intermittent disturbance feature extraction, arc channel residual calculation, arc fault state prediction, and fault perception result correction. By jointly analyzing multi-source operation data, short-term disturbance features, repetitive phase features, physical residual features, and reliability correction results, fault judgment is transformed from single-point threshold judgment to multi-dimensional evidence fusion judgment, thereby improving the accuracy and stability of intermittent arc fault identification.

[0052] For example, when a branch repeatedly experiences sudden current changes and high-frequency energy increases within multiple half-cycles, but is accompanied by clear switching action records, this solution can reduce its fault reliability; when another branch experiences phase concentration, persistent residuals, and repetitive interval disturbances without switching action, it can improve its intermittent arc confirmation reliability.

[0053] Example 2, see Figure 1This embodiment is based on the above embodiment. In step S1, the distribution box operation data acquisition is used to obtain the basic operation data required for the intermittent arc fault detection of the distribution box. Specifically, it is to collect the voltage data, current data, residual current data, switch status data, box temperature data and circuit identification data of the target circuit in the distribution box, and to unify the time base, match the circuit affiliation and organize the abnormal collection points of the collected data to obtain the distribution box operation data.

[0054] In one optional implementation, a voltage acquisition unit, a current acquisition unit, a residual current acquisition unit, and a temperature acquisition unit are installed on the incoming line side, outgoing line side, or target branch of the distribution box.

[0055] The voltage acquisition unit is used to acquire the voltage waveform of the target circuit, the current acquisition unit is used to acquire the current waveform of the target circuit, the residual current acquisition unit is used to acquire the residual current change of the target circuit or the main circuit of the distribution box, the temperature acquisition unit is used to acquire the temperature inside the distribution box or the temperature near the terminal block, and the switch status data is obtained through auxiliary contact signals, communication status signals or action records of circuit breakers, contactors or protection devices.

[0056] In practical implementation, the voltage and current data can be collected at a sampling frequency higher than that required for power frequency cycle analysis, such as using a sampling frequency of 10kHz to 100kHz to collect voltage and current waveform data, in order to retain short-term sudden changes and high-frequency disturbances that may occur in intermittent arc faults; the box temperature data can be collected with a sampling period of 1s to 10s to reflect changes in the thermal state inside the distribution box; the switch status data can be recorded when the switch status changes, and the corresponding action time, action type, and circuit to which it belongs can be recorded. The above sampling frequency and sampling period can be adjusted according to the distribution box level, sensor performance, and local processing capabilities, and do not constitute a limitation on the scope of protection of this invention;

[0057] During the data acquisition process, in order to ensure the correspondence between data from different sources, the time base of various types of acquired data is unified. Specifically, the local clock of the acquisition device, the output time of the time synchronization module, or the synchronization time of the host system can be used as a unified time base to add acquisition time identifiers to voltage data, current data, residual current data, switch status data, and box temperature data. For data with different sampling frequencies, they can be unified to the same analysis time axis through timestamp alignment, interpolation, or window merging.

[0058] For example, voltage waveform data and current waveform data can be kept in their original high-frequency sampling sequences, while temperature data and switch status data can be mapped to the corresponding waveform analysis window according to timestamps, thereby forming multi-source operating data under the same time window;

[0059] Furthermore, the collected data is matched to the circuit. Specifically, the target circuit to which each type of collected data belongs is determined based on the correspondence between the circuit number, switch number, transformer installation location, sensor number, and acquisition channel number of the distribution box. When multiple branches are set in a distribution box, the voltage data, current data, residual current data, switch status data, and temperature data of different branches are respectively bound to the corresponding circuit number identifier.

[0060] For example, the first outgoing circuit can be denoted as circuit L1, the second outgoing circuit can be denoted as circuit L2, and the current waveform corresponding to acquisition channel CH1, the voltage waveform corresponding to acquisition channel CH2, and the state data of switch K1 can be collectively attributed to circuit L1;

[0061] Furthermore, abnormal sampling points are identified in the collected data. Specifically, missing sampling points, over-range sampling points, communication interruption segments, and obvious isolated pulse points in voltage and current waveform data are marked; short-term jump values ​​and long-term unchanged values ​​in the chamber temperature data are quality-labeled; and repetitive action records, invalid state records, and missing timestamp records in the switch status data are organized. For short-term missing data that does not affect subsequent analysis, neighboring sample values, window averaging, or linear interpolation can be used to fill in the missing data. For data segments that are continuously missing for more than a preset duration, the missing data marker is retained and used as a source of uncertainty in subsequent fault perception result correction.

[0062] The operating data of the distribution box specifically includes voltage waveform data, current waveform data, residual current change data, switch action status data, box internal temperature change data, acquisition time identifier data, and circuit number identifier data;

[0063] The voltage waveform data is used to characterize the voltage amplitude change, zero-crossing position, and voltage distortion of the target circuit.

[0064] The current waveform data is used to characterize the load current changes, short-time sudden changes, and high-frequency disturbances of the target circuit.

[0065] The residual current change data is used to characterize the insulation status or leakage current change of the target circuit, and serves as auxiliary verification information for the fault result in step S5; when the residual current increases synchronously within the suspected arc disturbance segment, the attention to insulation abnormality accompanied by arc fault is increased, but it is not used as the sole criterion for arc fault judgment.

[0066] The switch operation status data is used to distinguish electrical disturbances caused by normal switch operation and abnormal disturbances that occur under non-switch operation conditions.

[0067] The temperature change data inside the box is used to characterize the thermal state inside the distribution box and assists in determining operation and maintenance handling prompts in step S5; when the temperature inside the box or the temperature near the wiring terminals continuously exceeds the corresponding normal operation benchmark, the priority of fault handling is increased, but it is not involved in the direct identification of short-term arc disturbance windows.

[0068] The collected time identifier data is used to ensure the correspondence between multi-source data in the time dimension;

[0069] The loop number identifier data is used to determine the target loop to which the abnormal data belongs.

[0070] Example 3, see Figure 1 This embodiment is based on the above embodiment. In step S2, the intermittent disturbance feature extraction is used to extract electrical features that can characterize short-term disturbances, repetitive disturbances and phase disturbances from the distribution box operation data. Specifically, based on the distribution box operation data, a transient analysis window is constructed, and current change features, voltage distortion features, high-frequency disturbance features, zero-crossing phase features and disturbance repetition features are extracted in the transient analysis window to obtain intermittent disturbance feature data.

[0071] In one alternative implementation, the power frequency period and voltage zero-crossing position are determined based on the voltage waveform data, and a transient analysis window is constructed based on the voltage zero-crossing position or a fixed sampling length.

[0072] The transient analysis window can be a data window corresponding to a power frequency half-cycle, a power frequency cycle, or multiple consecutive power frequency cycles.

[0073] For example, in a power distribution box circuit with a power frequency of 50Hz, one power frequency cycle is 20ms and one half-cycle is 10ms. One half-cycle can be used as the basic window, and several adjacent half-cycles can be combined into a continuous analysis window to observe whether short-term disturbances have the characteristic of recurrence.

[0074] Furthermore, current abrupt change features are extracted within the transient analysis window; these current abrupt change features are used to characterize the short-term deviation of the target circuit current from the normal power frequency variation trend.

[0075] Specifically, a reference current waveform is constructed based on a normal window adjacent to the current transient analysis window that does not exhibit significant disturbances. And calculate the current deviation sequence within the k-th transient analysis window. ;

[0076] according to Calculate the current surge amplitude in the k-th transient analysis window; according to Calculate the maximum rate of change of current; where, This represents the amplitude of the current surge in the k-th transient analysis window. This represents the maximum rate of change of current. This is the k-th transient analysis window; The time interval between adjacent sampling points;

[0077] The preset abrupt change threshold is determined based on the 99th percentile of the current deviation amplitude during normal operation. ;Will First time exceeding The time is determined as the mutation initiation time, and will continue for more than [time period]. The duration of the mutation is determined by the time length of the mutation.

[0078] The current mutation characteristics specifically include the current mutation amplitude, the maximum current change rate, the mutation start time, and the mutation duration;

[0079] Furthermore, voltage distortion features are extracted within the transient analysis window; specifically, the voltage waveform within the current transient analysis window is extracted. Phase-aligned reference voltage waveform Compare;

[0080] according to Calculate the degree of local voltage distortion;

[0081] according to Calculate the voltage amplitude deviation;

[0082] In the formula, The degree of local voltage distortion in the k-th transient analysis window; This represents the number of valid sampling points within the window. The root mean square value of the reference voltage waveform; and These are the effective voltage values ​​for the current window and the reference window, respectively.

[0083] The absolute time difference between the current window voltage zero-crossing time and the corresponding zero-crossing time of the reference window is determined as the voltage zero-crossing offset;

[0084] The ratio of the sum of the energy of the non-fundamental frequency components in the current window to the total frequency energy is determined as the proportion of the voltage non-fundamental frequency components.

[0085] The voltage distortion characteristics specifically include voltage amplitude deviation, degree of local voltage distortion, voltage zero-crossing offset, and the proportion of non-fundamental voltage components.

[0086] Furthermore, high-frequency disturbance features are extracted within the transient analysis window; these high-frequency disturbance features are used to characterize the short-term high-frequency energy enhancement phenomenon that may be caused by intermittent electric arcs.

[0087] In one alternative implementation, a bandpass filter is used to extract the high-frequency components from the current or voltage waveform. and in accordance with Calculate the high-frequency energy within the k-th transient analysis window;

[0088] The upper limit frequency of the passband is lower than half of the sampling frequency, preferably not higher than 40% of the sampling frequency; when the sampling frequency is between 10kHz and 50kHz, a frequency band between 2kHz and 40% of the sampling frequency is selected; when the sampling frequency is higher than 50kHz, a frequency band between 5kHz and 20kHz can be selected.

[0089] The difference between the high-frequency energy in the current transient analysis window and the median high-frequency energy in the normal reference window is determined as the high-frequency energy change; the duration for which the high-frequency energy is continuously higher than the 99th percentile of the high-frequency energy in the normal operation window is determined as the high-frequency energy duration; and the time position corresponding to the sampling point with the maximum absolute value of the high-frequency component is determined as the high-frequency energy peak position.

[0090] In the formula, The high-frequency energy within the k-th transient analysis window; The high-frequency component corresponding to the nth sampling point; This represents the number of valid sampling points within the k-th transient analysis window;

[0091] The high-frequency disturbance characteristics specifically include high-frequency energy value, high-frequency energy change, high-frequency energy duration, and high-frequency energy peak position;

[0092] Further, zero-crossing phase features are extracted within the transient analysis window. These zero-crossing phase features characterize the phase relationship between the disturbance occurrence time and the voltage zero-crossing position. Specifically, two adjacent voltage zero-crossing times are determined based on the voltage waveform data, and the disturbance occurrence time is mapped to the corresponding power frequency cycle or half-cycle to obtain the disturbance occurrence phase after the zero-crossing. For example, the disturbance occurrence phase after the zero-crossing can be calculated using the following formula:

[0093] ;

[0094] In the formula, For the first The perturbation that occurs after the zero-crossing corresponds to the next perturbation has a phase. The time when the disturbance occurs. For the voltage zero-crossing time corresponding to the half-cycle, The power frequency period is used to characterize whether disturbances are concentrated in a specific phase range after the voltage crosses zero. The zero-crossing phase characteristics specifically include the disturbance occurrence phase, the degree of disturbance phase dispersion, and the phase offset of adjacent disturbances.

[0095] The time when the disturbance occurs This refers to the earliest moment when the current deviation amplitude, the absolute value of the high-frequency component, or the voltage waveform deviation within the current transient analysis window first exceeds the corresponding normal threshold.

[0096] For multiple disturbance events within the same continuous analysis period, the standard deviation of the phase of the disturbance after each disturbance event crosses zero is determined as the disturbance phase dispersion; the absolute value of the difference between the phases of the disturbances after the zero crossing of two adjacent disturbance events is determined as the adjacent disturbance phase offset.

[0097] The difference between the start times of adjacent disturbance events is determined as the adjacent disturbance interval; the number of disturbance events identified within the preset statistical period is determined as the disturbance repetition count; the number of transient analysis windows that are adjacent and both identified as disturbance events is determined as the number of consecutive disturbance windows; the ratio of the standard deviation of the adjacent disturbance interval to its mean is determined as the disturbance interval fluctuation degree.

[0098] The preset statistical time period can be set from 1 second to 10 seconds, preferably 5 seconds;

[0099] Furthermore, disturbance repetition features are extracted within the transient analysis window. These disturbance repetition features characterize whether short-term disturbances repeatedly occur within multiple consecutive or interval windows. Specifically, disturbance events are determined based on current mutation features, high-frequency disturbance features, and voltage distortion features, and the time interval, number of repetitions, and repetition stability between adjacent disturbance events are statistically analyzed. The disturbance repetition features specifically include the adjacent disturbance interval, the number of disturbance repetitions, the number of consecutive disturbance windows, and the degree of disturbance interval fluctuation.

[0100] In practical implementation, candidate disturbance events can be determined based on current change characteristics, voltage distortion characteristics, and high-frequency disturbance characteristics, and preliminary screening of candidate disturbance events can be performed based on zero-crossing phase characteristics and disturbance repetition characteristics.

[0101] For example, when a transient analysis window simultaneously exhibits an increase in current amplitude, an increase in high-frequency energy, and a local voltage distortion, the transient analysis window is marked as a candidate disturbance window; when multiple candidate disturbance windows repeatedly appear within multiple consecutive half-cycles, and the phases of the disturbances are in a similar range, they are considered as suspected intermittent disturbance segments.

[0102] The intermittent disturbance characteristic data specifically includes current abrupt change characteristics, voltage distortion characteristics, high-frequency disturbance characteristics, zero-crossing phase characteristics, and disturbance repetition characteristics;

[0103] The current mutation characteristics include the current mutation amplitude, the maximum current change rate, the mutation start time, and the mutation duration;

[0104] The voltage distortion characteristics include voltage amplitude deviation, degree of local voltage distortion, voltage zero-crossing offset, and the proportion of non-fundamental voltage components.

[0105] The high-frequency disturbance characteristics include high-frequency energy value, high-frequency energy change, high-frequency energy duration, and high-frequency energy peak position;

[0106] The zero-crossing phase characteristics include the phase in which the disturbance occurs, the degree of dispersion of the disturbance phase, and the phase offset of adjacent disturbances;

[0107] The perturbation repetition characteristics include the adjacent perturbation interval, the number of perturbation repetitions, the number of consecutive perturbation windows, and the degree of fluctuation in the perturbation interval.

[0108] Example 4, see Figure 1 , Figure 2 This embodiment is based on the above embodiment. In step S3, the arc channel residual calculation is used to determine whether the abnormal disturbances in the intermittent disturbance characteristic data conform to the electrical and physical relationships in the arc channel formation process. Specifically, based on the distribution box operation data and the intermittent disturbance characteristic data, a constraint relationship is established between the normal load voltage drop, line impedance voltage drop, and suspected arc channel voltage drop of the target circuit. The arc channel residual is calculated based on the degree of deviation between the measured voltage and current and the constraint relationship to obtain the arc channel residual characteristic data. This includes the following steps:

[0109] Step S31: Candidate disturbance window determination, used to determine the analysis object that needs to be calculated for arc channel residuals. Specifically, based on the current mutation characteristics, high-frequency disturbance characteristics, voltage distortion characteristics and zero-crossing phase characteristics in the intermittent disturbance characteristic data, transient analysis windows that meet the disturbance intensity conditions and phase occurrence conditions are screened to obtain candidate disturbance window data.

[0110] The normalized value of the current mutation is calculated according to the following formula:

[0111] ;

[0112] The normalized value of the high-frequency disturbance is calculated according to the following formula:

[0113] ;

[0114] The voltage distortion normalization value is calculated according to the following formula:

[0115] ;

[0116] In the formula, , and These are the normalized values ​​of current surge, high-frequency disturbance, and voltage distortion, respectively, for the k-th transient analysis window. This represents the amplitude of the sudden change in current. This refers to high-frequency energy values. This represents the degree of local voltage distortion within the k-th transient analysis window. , and These are the medians of the corresponding features within the historical normal operating window; , and These represent the median absolute deviation of the corresponding feature; 1.4826 is the correction factor for converting the median absolute deviation to the standard deviation scale. To prevent tiny positive numbers with a denominator of zero, it is preferable to take... ;

[0117] The disturbance intensity evaluation value is calculated based on each standardized characteristic:

[0118] ;

[0119] when Not less than 2.5, and , and When at least two of the values ​​are not less than 2, the corresponding transient analysis window is determined as the initial disturbance window; the above weights are used to enhance the role of high-frequency disturbance characteristics in the initial screening, and can be adjusted according to the load type of the target circuit. Adjust within the range, while keeping the sum of all weights at 1;

[0120] Furthermore, the initial disturbance window is validated for phase occurrence conditions; for the initial disturbance window within the same continuous analysis period, the median value of its disturbance occurrence phase is calculated based on the zero-crossing phase characteristics. And calculate the phase deviation of the k-th initial perturbation window according to the following formula:

[0121] ;

[0122] In the formula, The phase of the perturbation appears after the zero-crossing of the k-th initial perturbation window; when If the initial disturbance window is not greater than the preset phase tolerance, it is determined that the phase occurrence condition is met. The preset phase tolerance can be set to 10° to 30°, preferably 20°. This range is used to take into account sampling error, load change and arc ignition moment fluctuation. When the sampling frequency is high and the voltage zero crossing positioning error is small, a smaller phase tolerance can be selected. When the target circuit load fluctuation is large, a larger phase tolerance can be selected.

[0123] When there is only one initial disturbance window in the continuous analysis period, the initial disturbance window is retained and its phase state is marked as to be reconfirmed; the initial disturbance window that meets the disturbance intensity condition and the phase occurrence condition is determined as the candidate disturbance window, and its loop, window start and end time, disturbance intensity evaluation value and disturbance occurrence phase are recorded to obtain the candidate disturbance window data;

[0124] Step S32: Reference window matching, used to determine the normal operation reference state corresponding to the candidate disturbance window. Specifically, based on the circuit number, switch status, current RMS value, load change trend and acquisition time identifier, adjacent operating windows without obvious disturbances are selected from the distribution box operation data to obtain normal reference window data.

[0125] Furthermore, taking the start time of the candidate disturbance window as the center, a transient analysis window with the same loop number and the same switching state as the candidate disturbance window is selected within a preset search period before and after it; the preset search period can be set to 1 second to 5 seconds before and after, preferably 2 seconds before and after.

[0126] Exclude the following transient analysis windows: windows containing circuit breaker or contactor operation records within 100 ms before and after the window's start and end times; disturbance intensity evaluation values. A window with a value not less than 1.0; a window with continuous sampling missing, sensor malfunction, or time synchronization malfunction is used to obtain a reference candidate window;

[0127] The preset action isolation period can be set from 50ms to 200ms, preferably 100ms, to cover the short transient process caused by the operation of a general low-voltage circuit breaker or contactor. A value less than 1.0 indicates that the overall perturbation characteristics within this window are within the normal statistical fluctuation range;

[0128] Calculate the RMS current and load variation trend for each candidate disturbance window and each reference candidate window, respectively. The load variation trend for the j-th transient analysis window is calculated according to the following formula:

[0129] ;

[0130] In the formula, The load change trend for the j-th transient analysis window; The effective value of the current within the j-th transient analysis window; It is the effective value of the current in the previous transient analysis window;

[0131] When the relative deviation of the effective current value between the reference candidate window and the candidate disturbance window is no greater than 15%, and the difference in load change trend is no greater than 0.2, the reference candidate window is determined as a valid reference window. The 15% deviation of the effective current value is used to limit the reference window and the candidate disturbance window to be at similar load levels, and the difference in load change trend of 0.2 is used to exclude windows where the load is rapidly increasing or decreasing. The above limits can be set according to the statistical results of the normal load fluctuation of the target circuit, preferably the 95th percentile value of the corresponding difference under normal operating conditions.

[0132] According to the order of acquisition time and candidate disturbance windows from near to far, select 3 to 8 windows from the effective reference windows; when there are less than 3 effective reference windows, gradually expand the preset search time period, and the expanded single-sided search time period shall not exceed 10 seconds;

[0133] The selected effective reference window is phase-aligned according to the voltage zero-crossing position, and the median of the voltage and current sample values ​​at the same sampling position is taken to obtain the reference voltage waveform. and reference current waveform The normal reference window data is composed of the reference voltage waveform, the reference current waveform, and the corresponding window identifier.

[0134] Step S33: Load impedance deviation calculation, used to determine whether the voltage and current relationship in the candidate disturbance window deviates from the normal load relationship. Specifically, it calculates the normal equivalent impedance of the target circuit based on the normal reference window data, and compares the equivalent impedance in the candidate disturbance window with the normal equivalent impedance to obtain the normal load impedance deviation.

[0135] Furthermore, the fundamental frequency waveforms of the voltage and current waveforms within the normal reference window and the candidate disturbance window are extracted to obtain the fundamental frequency phasor of the reference voltage. Reference current fundamental phasor Candidate voltage fundamental phasor and candidate current fundamental phasor The fundamental phasor can be calculated using a single-cycle discrete Fourier transform, and the fundamental frequency used is consistent with the power frequency of the target circuit.

[0136] Calculate the normal equivalent impedance and the candidate window equivalent impedance respectively according to the following formula:

[0137] ;

[0138] ;

[0139] In the formula, The normal equivalent impedance corresponding to the k-th candidate disturbance window; The measured equivalent impedance within the k-th candidate perturbation window;

[0140] When the fundamental amplitude of the candidate current is less than the preset low load threshold, impedance deviation calculation is not performed, and the corresponding window is marked as an invalid window for low load impedance. The preset low load threshold can be set to 3% to 10% of the rated current of the target circuit, preferably 5%, to avoid excessive impedance calculation error caused by phasor division when the current is close to zero.

[0141] For the valid candidate disturbance window, the deviation of the normal load impedance is calculated according to the following formula:

[0142] ;

[0143] In the formula, This represents the deviation of the normal load impedance for the k-th candidate disturbance window; The larger the value, the greater the deviation of the voltage-current relationship within the candidate disturbance window from the normal load condition;

[0144] Step S34: Line voltage drop residual calculation, used to isolate the normal voltage drop response caused by the equivalent resistance and equivalent inductance of the line. Specifically, the line impedance voltage drop is calculated based on the line equivalent parameters, current change and current change rate of the target circuit, and the measured voltage disturbance is compared with the line impedance voltage drop to obtain the line voltage drop residual.

[0145] Furthermore, the equivalent resistance of the circuit is pre-set for each target circuit. Equivalent inductance of the line The equivalent resistance of the line is preferably determined using the line as-built data or on-site resistance measurement results, or it can be determined according to... Calculate, where, The resistivity of the conductor material at the current temperature. Let A be the round-trip length of the target loop conductor, and A be the cross-sectional area of ​​the conductor. The equivalent inductance of the line is preferably determined by the results of on-site impedance measurement, or it can be determined by the product of the inductance per unit length of the cable and the length of the line.

[0146] When the circuit parameters are derived from design data, the equivalent resistance of the circuit can be corrected for temperature based on the internal temperature of the enclosure or the estimated temperature of the conductors. When the equivalent inductance of the circuit cannot be obtained, for circuits with short circuit lengths and low load change rates, the inductance voltage drop can be ignored and... Set to zero, while retaining the simplified identifier for line parameters;

[0147] Calculate the measured voltage disturbance and current change based on the candidate disturbance window and the normal reference window:

[0148] ;

[0149] ;

[0150] In the formula, The measured voltage disturbance of the k-th candidate disturbance window relative to the normal reference window; This corresponds to the change in current. and These are the measured voltage waveform and measured current waveform within the k-th candidate disturbance window, respectively;

[0151] The voltage drop across the line impedance caused by the equivalent resistance and equivalent inductance of the line is calculated using the following formula:

[0152] ;

[0153] For discrete sampled data, the rate of change of current is calculated using the central difference:

[0154] ;

[0155] In the formula, The voltage drop across the line impedance caused by the line's equivalent resistance and equivalent inductance; The sampling time interval for the current waveform;

[0156] By comparing the measured voltage disturbance with the line impedance voltage drop, the line voltage drop residual sequence is obtained:

[0157] ;

[0158] The line voltage drop residual for the k-th candidate disturbance window is calculated using the following formula:

[0159] ;

[0160] In the formula, The line voltage drop residual for the k-th candidate disturbance window; This represents the number of valid sampling points within the k-th candidate perturbation window; The line voltage drop residual corresponding to the nth valid sampling point;

[0161] Step S35: Calculate the equivalent residual of the arc channel, which is used to characterize the suspected arc channel voltage drop that cannot be explained by the normal load response and line voltage drop response in the candidate disturbance window. Specifically, based on the deviation relationship between the normal load response, line impedance voltage drop and measured voltage disturbance, calculate the equivalent residual of the arc channel, the residual of current change and the residual of voltage change to obtain the residual data of the arc channel.

[0162] Furthermore, the equivalent parameters of the normal load are determined based on the normal equivalent impedance and the line equivalent parameters; firstly, the power frequency equivalent impedance of the line is calculated according to the following formula:

[0163] ;

[0164] In the formula, is the power frequency equivalent impedance of the target circuit; j is the imaginary unit; It is the power frequency angular frequency; f is the power frequency of the target circuit, which is 50Hz in a 50Hz power distribution system and 60Hz in a 60Hz power distribution system.

[0165] Calculate the equivalent impedance of normal load using the following formula:

[0166] ;

[0167] In the formula, The equivalent impedance of the normal load corresponding to the k-th candidate disturbance window;

[0168] The real part of the equivalent impedance of a normal load is defined as the equivalent resistance of a normal load:

[0169] ;

[0170] Convert the imaginary part of the equivalent impedance under normal load to the equivalent inductance under normal load:

[0171] ;

[0172] In the formula, Re() represents taking the real part of the complex number; Im() represents taking the imaginary part of the complex number; when When it is less than zero, set it to zero; when When the value is less than zero, it indicates that the target circuit may have capacitive load characteristics, and in this case, it is retained. The complex impedance form is used for subsequent fundamental residual calculation, and the inductance term in the normal load response in the time domain is set to zero, while the simplified identifier of the load parameters is retained.

[0173] Calculate the normal load response based on the current change:

[0174] ;

[0175] In the formula, This refers to the change in load voltage drop caused by changes in current under normal load equivalent parameters;

[0176] The equivalent residual sequence of the arc channel is calculated according to the following formula:

[0177] ;

[0178] And calculate the equivalent residual of the arc channel:

[0179] ;

[0180] In the formula, The voltage residual sequence within the k-th candidate disturbance window that cannot be explained by line impedance voltage drop and normal load response; The root mean square value of the corresponding voltage residual sequence is used to characterize the overall intensity of the voltage drop in the suspected arc channel;

[0181] Furthermore, the fundamental voltage and fundamental current variations are calculated based on the fundamental phasors of the normal reference window and the candidate disturbance window:

[0182] ;

[0183] ;

[0184] The predicted current variation that can be explained by normal electrical relationships is calculated based on the equivalent impedance of the normal load and the equivalent impedance of the line at power frequency.

[0185] ;

[0186] The residual current change is calculated using the following formula:

[0187] ;

[0188] Calculate the voltage change residual using the following formula:

[0189] ;

[0190] In the formula, The fundamental voltage change of the k-th candidate disturbance window relative to the normal reference window; This corresponds to the change in the fundamental current. The residual of the current change is used to represent the relative deviation between the measured current change and the predicted result of the normal impedance relationship; The voltage change residual is used to represent the relative deviation between the measured voltage change and the predicted result of the normal impedance relationship;

[0191] Step S36: Residual persistence processing, used to characterize the persistence and repetition of arc channel residuals in multiple transient analysis windows. Specifically, based on the time sequence, zero-crossing phase and residual amplitude changes of candidate disturbance windows, the residual duration, residual repetition count and residual phase consistency are statistically analyzed to obtain arc channel residual characteristic data.

[0192] Furthermore, the upper limit of normal fluctuation of the equivalent residual of the arc channel is calculated using a normal reference window; the residuals of each normal reference window are calculated according to steps S34 and S35 to obtain the set of normal residuals, and the effective threshold of the residuals is determined according to the following formula:

[0193] ;

[0194] In the formula, The effective threshold for the residual; The median of the normal residual set; This represents the median absolute deviation of the normal residual set; in practice, it can also be determined based on the 99th percentile of the normal operating data. ;

[0195] When the arc channel equivalent residual of the candidate disturbance window Greater than the effective threshold of residual When this happens, the candidate disturbance window is determined as the effective residual window;

[0196] The effective residual windows are arranged in the order of acquisition time. When the interval between two adjacent effective residual windows is not greater than the preset merging interval, the effective residual windows are merged into the same residual segment. The preset merging interval can be set to 1 to 3 transient analysis windows, preferably 2 transient analysis windows, to allow intermittent arcs to be briefly interrupted in adjacent half-cycles or power frequency cycles.

[0197] The residual duration of the residual segment is calculated using the following formula:

[0198] ;

[0199] In the formula, The duration of the residual segment; This is the start time of the first valid residual window within the residual segment; This is the start time of the last valid residual window within the residual segment; The duration of a single transient analysis window;

[0200] The number of mutually independent residual segments is counted within a preset continuous statistical period and determined as the residual repetition count. The preset continuous statistical period can be set from 1s to 10s, preferably 5s. When the target circuit load operates frequently, a shorter statistical period can be used, and when the interval between intermittent arcs is long, a longer statistical period can be used.

[0201] For N valid residual windows within the same preset duration statistical period, the residual phase consistency is calculated based on the phase of the disturbance after the zero crossing in each window:

[0202] ;

[0203] In the formula, For residual phase consistency, the value ranges from 0 to 1; The phase of the perturbation appears after the zero crossing of the k-th effective residual window; The closer it is to 1, the more concentrated the disturbances in multiple effective residual windows appear at similar zero-crossing phase positions;

[0204] The residual duration, residual repetition count, and residual phase consistency are combined into residual persistence features to obtain residual feature data of the arc channel;

[0205] The residual characteristic data of the arc channel specifically includes the deviation of normal load impedance, line voltage drop residual, arc channel equivalent residual, current change residual, voltage change residual and residual persistence characteristics.

[0206] Wherein, the normal load impedance deviation is used to represent the degree of deviation of the equivalent impedance of the candidate disturbance window from the equivalent impedance of the normal reference window; the line voltage drop residual is used to represent the remaining part of the measured voltage disturbance that cannot be explained by the voltage drop caused by the equivalent resistance and equivalent inductance of the line; the arc channel equivalent residual is used to represent the remaining voltage amplitude of the measured voltage disturbance that cannot be explained by the normal load response and the line impedance voltage drop; the current change residual and voltage change residual are used to represent the degree of deviation of the voltage and current changes within the candidate disturbance window from the normal impedance constraint relationship; the residual persistence characteristic is used to represent the persistence, repetition and phase concentration of the arc channel equivalent residual in multiple transient analysis windows.

[0207] By performing the above operations, this solution addresses the technical problem in existing arc fault identification methods that directly use high-frequency disturbances or current surges as arc criteria, failing to determine whether abnormal disturbances conform to the physical relationships of voltage, current, and line voltage drop during the arc channel formation process. This leads to ordinary load actions being easily misidentified as arc faults. The solution creatively employs an arc channel residual calculation method. By matching a normal reference window, it calculates the normal load impedance deviation, line voltage drop residual, and arc channel equivalent residual. This isolates the normal load response and line voltage drop response from the measured voltage disturbance, thereby highlighting suspected arc channel components that cannot be explained by conventional electrical relationships and improving the ability to identify genuine arc disturbances.

[0208] For example, when an air conditioner compressor starts up, it may generate sudden current changes and short-term voltage dips, but its voltage changes can usually be explained by load changes and line voltage drops, so the equivalent residual of the arc path is low; while when poor contact of the wiring terminals generates intermittent arcs, residuals that cannot be explained by line voltage drops will repeatedly appear in the same phase window, and thus be more accurately identified.

[0209] Example 5, see Figure 1This embodiment is based on the above embodiment. In step S4, the arc fault state prediction is used to identify the arc fault state of the target circuit based on intermittent disturbance feature data and arc channel residual feature data. Specifically, the intermittent disturbance feature data and arc channel residual feature data are input into the arc fault state prediction model. The arc fault state prediction model is used to classify and predict the normal operation state, ordinary load disturbance state, suspected arc state and arc fault state of the target circuit to obtain arc fault prediction data.

[0210] In one optional implementation, the intermittent disturbance feature data and arc channel residual feature data are first processed to obtain the model input feature sequence. Specifically, the current mutation features, voltage distortion features, high-frequency disturbance features, zero-crossing phase features, disturbance repetition features, normal load impedance deviation, line voltage drop residual, arc channel equivalent residual, current change residual, voltage change residual and residual persistence features within the same target circuit, the same candidate disturbance window or the same continuous window group are time aligned and normalized to form feature vectors or feature matrices corresponding to the candidate disturbance window.

[0211] For continuous features with different dimensions, such as current abrupt amplitude, high-frequency energy change, and arc channel equivalent residual, standardization is performed using the mean and standard deviation from the training data. The standardization results are then processed according to... Calculate; where, Let r be the original eigenvalue of the r-th term. To correspond to the standardized eigenvalues, and Let be the mean and standard deviation of the r-th feature in the training data, respectively. To prevent tiny positive numbers with a denominator of zero, it can be set to ;

[0212] For discrete features related to the source of disturbance, such as switch status, one-hot encoding can be used; loop number is only used to associate data of the same target loop and is not used as a model input feature; fault category is a training label and is not used as a model input feature to avoid the model directly using category information to form prediction results;

[0213] For data formed by multiple consecutive transient analysis windows, a serialized input is constructed according to the acquisition time sequence. The serialized input is composed of consecutive windows before and after the current candidate disturbance window as the center. The number of windows M can be set to 5 to 21, preferably 11. When the transient analysis window is a half-cycle, 11 windows can cover the waveform change process of about 110ms before and after the current disturbance. The specific number of windows can be determined according to the typical duration of the arc disturbance and the computing power of the model deployment equipment.

[0214] In one optional implementation, the arc fault state prediction model adopts a combined model consisting of a one-dimensional convolutional neural network and a gated recurrent unit network.

[0215] Specifically, the model input feature sequence is sequentially fed into a one-dimensional convolutional layer, a pooling layer, a gated recurrent unit layer, a fully connected layer, and a classification output layer. In one optional implementation, the one-dimensional convolutional layer is configured with 32 convolutional kernels of length 3, the pooling length is set to 2, the gated recurrent unit layer is configured with 64 hidden units, the fully connected layer is configured with 32 output nodes, and the classification output layer is configured with 4 output nodes. The Softmax function is used to output the predicted probabilities of normal operation, normal load disturbance, suspected arcing, and arcing fault states. The above network parameters can be adjusted according to the model input scale and the computing power of the deployed equipment.

[0216] The model training uses the cross-entropy loss function and the adaptive moment estimation optimization algorithm. The initial learning rate can be set to... to Training is stopped when the validation set loss does not decrease for 10 consecutive rounds, and the model parameters at which the validation set loss is lowest are retained.

[0217] For example, when the input data is continuous Each transient analysis window contains [number] windows. When there are 1 feature term, the model input can be represented as:

[0218] ;

[0219] In the formula, The input feature sequence is used for the arc fault state prediction model; For the first The feature vectors corresponding to each transient analysis window; Enter the number of windows; The number of feature terms within each window; this expression is only used to describe the organization of the model input data and does not limit the specific neural network structure.

[0220] Each transient analysis window contains at least the following features: current surge amplitude, maximum current change rate, local voltage distortion, high-frequency energy change, high-frequency energy duration, disturbance occurrence phase, disturbance phase offset, disturbance repetition count, normal load impedance deviation, line voltage drop residual, arc channel equivalent residual, current change residual, voltage change residual, residual duration, and residual phase consistency. P is determined based on the actual number of selected features, and the same feature order is used for all training samples and samples to be predicted.

[0221] Furthermore, the operating states of the training samples are labeled. Normal operating state refers to an operating state where no switching action, load abrupt change, or arc fault occurs, and all disturbance characteristics are within the normal statistical range. Ordinary load disturbance state refers to a disturbance state caused by motor starting, capacitive load switching, switching action, or other known load changes, and which can be explained by switching status, normal load response, or line impedance voltage drop. Suspected arc state refers to an operating state exhibiting arc-related characteristics such as current abrupt change, high-frequency energy enhancement, or phase concentration, but where the arc channel equivalent residual, repetition count, or duration has not yet reached the conditions for arc fault confirmation. Arc fault state refers to an operating state where the existence of an arc channel is confirmed through artificial arc initiation tests, fault circuit verification, maintenance records, or protection device records, and the corresponding disturbance has obvious arc channel equivalent residual and intermittent repetition characteristics.

[0222] Samples whose status cannot be determined through test records, maintenance records, or multi-source operation data will not be used as model training samples to avoid uncertain labels reducing the reliability of model training.

[0223] Furthermore, the input feature sequence of the model is input into the arc fault state prediction model, and the predicted probability of the target circuit belonging to different operating states is output.

[0224] The different operating states include at least the normal operating state, the ordinary load disturbance state, the suspected arcing state, and the arcing fault state; for example, the model output can be expressed as:

[0225] ;

[0226] In the formula, The state prediction probability vector output by the model; The predicted probability of normal operation; The predicted probability for a normal load disturbance state; The predicted probability of a suspected electric arc state; Let be the predicted probability of an arc fault state, and ;

[0227] Furthermore, the initial fault category of the target loop is determined based on the state prediction probability;

[0228] when It is the maximum value among the four types of prediction probabilities, and Not less than the preset fault judgment threshold At that time, the target circuit is initially determined to be in an arc fault state; the preset fault judgment threshold It can be set to 0.6 to 0.9, preferably 0.75;

[0229] When the conditions for determining the state of an arc fault are not met, and The maximum value among the four predicted probabilities, or Not less than the preset suspected arc threshold At that time, the target circuit is initially determined to be in a suspected arcing state; the preset suspected arcing threshold It can be set to 0.5 to 0.8, preferably 0.65;

[0230] When the conditions for determining arc fault state and suspected arc state are not met, and When the probability is the maximum among the remaining predicted probabilities, the target loop is initially determined to be in a normal load disturbance state; otherwise, the target loop is initially determined to be in a normal operating state.

[0231] The preset fault judgment threshold and preset suspected arc threshold The threshold is determined based on the false alarm rate and false negative rate of the validation set. When the field has a low tolerance for false alarms, the corresponding threshold is increased; when the field has a low tolerance for false faults, the corresponding threshold is decreased. The preferred threshold is one that ensures that the false alarm rate of arc faults in the validation set is no higher than 5%.

[0232] Furthermore, the fault occurrence time window and suspected arc disturbance segment are determined based on the predicted category, predicted probability, and temporal location of the candidate disturbance window; specifically, candidate disturbance windows with an initial fault category of suspected arc state or arc fault state are determined as suspected arc windows and arranged in order of acquisition time.

[0233] When the interval between two adjacent suspected arc windows is not greater than the preset segment merging interval, and the phase difference between their post-zero-crossing disturbances is not greater than the preset phase tolerance, the adjacent suspected arc windows are merged into the same suspected arc disturbance segment; the preset segment merging interval can be set to 1 to 3 transient analysis windows, preferably 2 transient analysis windows; the preset phase tolerance is consistent with the preset phase tolerance in step S31; the start time of the first window to the end time of the last window in the same suspected arc disturbance segment is determined as the fault occurrence time window;

[0234] Furthermore, the probability is predicted based on the arc fault state. , Probability of Predicted Suspected Arc State The initial fault level is determined using residual characteristic data of the arc channel;

[0235] When the initial fault category is an arc fault state, and the arc channel equivalent residual Not less than the effective residual threshold When the risk level is 1.5 times that of the suspected arc disturbance segment and the number of effective residual windows within the same suspected arc disturbance segment is not less than 3, the initial fault level is determined to be a high-risk level.

[0236] When the initial fault category is an arc fault state but does not meet the high-risk level conditions, or when the initial fault category is a suspected arc state and the number of effective residual windows in the same suspected arc disturbance segment is not less than 2, the initial fault level is determined to be a medium-risk level.

[0237] When the initial fault category is suspected arcing, but there is only one valid residual window or the residual phase state is to be reconfirmed, the initial fault level is determined to be the level to be confirmed.

[0238] When the initial fault category is a normal load disturbance state, the initial fault level is determined to be a low-risk level; when the initial fault category is a normal operation state, the initial fault level is determined to be a fault-free level.

[0239] in, The effective threshold for the residual determined in step S36; Not less than 1.5 This is used to indicate that the equivalent residual of the arc channel has significantly exceeded the upper limit of normal residual fluctuation; the number of effective residual windows is 2 to 3 to distinguish between single occasional disturbances, repeated suspected disturbances and continuous arc disturbances. The specific number can be adjusted according to the length of the transient analysis window and the arc test data.

[0240] The arc fault prediction data specifically includes the target circuit fault category, fault occurrence time window, fault prediction probability, suspected arc disturbance segment, and initial fault level. The target circuit fault category indicates whether the target circuit is in normal operation, normal load disturbance, suspected arc state, or arc fault state. The fault occurrence time window marks the acquisition time range corresponding to the suspected arc or arc fault. The fault prediction probability represents the category probability output by the arc fault state prediction model. The suspected arc disturbance segment represents a continuous or intermittent disturbance window identified by the model as a suspected arc or arc fault. The initial fault level represents the fault risk level obtained before confidence correction.

[0241] Example 6, see Figure 1 and Figure 3 This embodiment is based on the above embodiment. In step S5, the fault perception result correction is used to correct the credibility of the arc fault prediction data and confirm intermittent faults. Specifically, based on the distribution box operation data, intermittent disturbance characteristic data, arc channel residual characteristic data, and arc fault prediction data, the signal acquisition uncertainty, model prediction uncertainty, load confusion uncertainty, and intermittent consistency evaluation results are calculated. Then, the arc fault prediction data is corrected based on the signal acquisition uncertainty, model prediction uncertainty, load confusion uncertainty, and intermittent consistency evaluation results to obtain fault perception result data, including the following steps:

[0242] Step S51: Calculate the uncertainty of data acquisition, which is used to evaluate the reliability of the power distribution box operation data within the current fault prediction window. Specifically, it calculates the signal acquisition uncertainty based on the sampling missing situation, sampling noise level, time synchronization deviation and sensor abnormality indicators to obtain acquisition uncertainty data.

[0243] Furthermore, the continuous transient analysis window used to form the primary model input in Example 5 is determined as the current fault prediction window; the proportion of missing sampling points, the ratio of the root mean square noise value to the upper limit of normal noise, the ratio of time synchronization deviation to the allowable synchronization deviation, and the sensor abnormal state within the current fault prediction window are normalized to 0 to 1 respectively to obtain the degree of missing sampling. Noise level Synchronization deviation degree and sensor anomaly degree Among them, when the sensor is normal Set to 0 if there is an over-range error, communication interruption, or self-test anomaly. Take 1;

[0244] according to Calculate signal acquisition uncertainty When any normalized index exceeds 1, it is counted as 1; the allowable synchronization deviation is preferably set to 5% of the transient analysis window duration, and the upper limit of normal noise is determined based on the 95th percentile value of the normal operation data of the target loop;

[0245] Step S52: Calculate the prediction uncertainty, which is used to evaluate the stability of the output results of the arc fault state prediction model. Specifically, it calculates the prediction uncertainty of the model based on the prediction probability distribution of different fault categories, the change of prediction results in adjacent windows, and the probability difference between the suspected arc state and the arc fault state, and obtains the prediction uncertainty data.

[0246] Furthermore, the maximum value among the four predicted probabilities is denoted as... Let the second largest prediction probability be denoted as And calculate the difference in prediction probability vectors between two adjacent fault prediction windows. ;according to Computational model prediction uncertainty ;

[0247] in, It is the sum of the absolute differences of each component of two adjacent prediction probability vectors divided by 2, with a value ranging from 0 to 1; and The closer the windows are, or the greater the difference in prediction results between adjacent windows, the higher the uncertainty of the model prediction.

[0248] Step S53: Load confusion uncertainty calculation, used to evaluate whether the current suspected disturbance may be caused by ordinary load operation. Specifically, the load confusion uncertainty is calculated based on the switch operation status, load current change trend, high-frequency disturbance duration and arc channel residual characteristic data to obtain load confusion uncertainty data.

[0249] Furthermore, when there are corresponding circuit switching action records within the preset action correlation period before and after a suspected arc disturbance segment, the switching correlation index will be... Set to 1 otherwise set to 0; the preset action association time period can be set from 50ms to 200ms, preferably 100ms;

[0250] The relative difference between the average value of the load current variation trend within the current suspected arc disturbance segment and the median value of the historical normal load operation trend is: and in accordance with Calculate load trend similarity; where, The upper limit of the allowable difference in the trend of normal load operation can be determined based on the 95th percentile value of historical normal load operation samples.

[0251] The duration of the current high-frequency disturbance is denoted as . The median value of the transient duration of historical ordinary switches is denoted as and in accordance with Calculate the degree of similarity in duration;

[0252] according to Calculate the degree of low residual; when Not less than hour, Set to 0;

[0253] Based on switch correlation indicators Load trend similarity and low residual degree Determine the uncertainty of load obfuscation:

[0254] ;

[0255] In the formula, To obfuscate uncertainty in the load; , , and The values ​​range from 0 to 1;

[0256] Step S54: Intermittent consistency evaluation, used to confirm whether the suspected arc disturbance has intermittent repetition and physical residual consistency characteristics. Specifically, the intermittent consistency evaluation results are calculated based on the number of residual repetitions, residual persistence characteristics, zero-crossing phase characteristics and adjacent disturbance intervals in the suspected arc disturbance segment, and intermittent consistency data is obtained.

[0257] Furthermore, the number of times the residual is repeated is denoted as... and in accordance with Calculate the degree of residual repeatability, where, It refers to the degree of residual repeatability;

[0258] The effective residual accumulation time is denoted as and in accordance with Calculate the persistence of the residual; where, The duration of a single transient analysis window. It refers to the persistence of residuals;

[0259] The residual phase consistency obtained in step S36 is directly used as the phase consistency degree. ,Right now ;

[0260] Based on the coefficient of variation of adjacent perturbation intervals Calculate interval stability ;when When it is not greater than 0.2, Take 1, when When it is not less than 1, Take 0, when At that time, according to Calculation; when there are fewer than 3 disturbance events, Set to 0;

[0261] according to Calculate the intermittent consistency evaluation results, where, It is the result of intermittent consistency evaluation;

[0262] Step S55: Fault credibility correction, used to correct the credibility of arc fault prediction data. Specifically, the fault credibility is calculated based on the fault prediction probability, signal acquisition uncertainty, model prediction uncertainty, load confusion uncertainty and intermittent consistency evaluation results to obtain corrected credibility data.

[0263] The sum of the predicted probabilities of suspected arc state and arc fault state is taken as the arc-related prediction probability. ,Right now ;

[0264] uncertainty in signal acquisition Model prediction uncertainty And load confusion uncertainty The average value is determined as the comprehensive uncertainty. ,Right now:

[0265] ;

[0266] according to Calculate the reliability of the fault The calculation results are limited to the range of 0 to 1.

[0267] In the formula, For arc-related prediction probabilities, ; The higher the value, the more significantly the prediction results are affected by data quality, model stability, or common load confusion. The higher the value, the stronger the repeatability and phase consistency of the suspected perturbation; Due to the uncertainty in signal acquisition, To predict uncertainty in the model, To mitigate the uncertainty of the load, To account for uncertainties, This is the result of the intermittent consistency evaluation;

[0268] Step S56: Fault perception result output, used to generate the final intermittent arc fault perception result of the distribution box. Specifically, based on the correction confidence data, the initial fault level and the suspected arc disturbance segment, the corrected fault category, the intermittent arc confirmation result, the suspected fault circuit, the fault time period, the fault risk level and the operation and maintenance handling prompt information are determined to obtain the fault perception result data.

[0269] when If the value is not less than 0.80, but the initial fault category is suspected arcing, it is determined to be a high-confidence suspected arcing state or a suspected arcing state; when A value not less than 0.80 and an initial fault category of arc fault state are considered confirmed arc faults; when... When the value is not less than 0.60 and less than 0.80, it is determined to be a suspected arcing condition; when When the value is not less than 0.40 and less than 0.60, it is determined to be in a state pending review; when When the value is less than 0.40, compare the predicted probability p_1 of the normal load disturbance state with the predicted probability of the normal operation state. ;when Greater than If there are corresponding circuit switching action records before and after a suspected arc disturbance segment, the system is corrected to a normal load disturbance state; otherwise, it is corrected to a normal operating state.

[0270] The confidence level threshold can be adjusted based on the validation set subject operating characteristic curve or the field false alarm tolerance, preferably ensuring that the false alarm rate of confirmed arc faults is no higher than 5%.

[0271] The fault risk level is determined based on the corrected fault category and the initial fault level. When the correction result lowers the status level by one level, the initial fault level is correspondingly lowered by one level. When the correction result is upgraded to a confirmed arc fault, the initial fault level is maintained or increased. The circuit to which the suspected arc disturbance segment belongs is identified as the suspected fault circuit, and the time from the start of its first window to the end of its last window is defined as the fault time period. Maintenance and operation prompts are generated based on the fault risk level. Among these, no fault handling prompts are output for no fault level, continuous monitoring prompts are output for low-risk levels, on-site verification prompts are output for pending confirmation levels and medium-risk levels, and prompts for priority power outage inspection of suspected fault circuits and connections are output for high-risk levels. When the residual current increases synchronously or the temperature inside the box continuously exceeds the normal operating benchmark, the priority of on-site handling can be increased.

[0272] The fault perception result data specifically includes the corrected fault category, fault confidence, intermittent arc confirmation result, suspected fault circuit, fault time period, fault risk level, and operation and maintenance handling prompts.

[0273] By performing the above operations, this solution addresses the technical problems in existing distribution box fault prediction methods, where model output results are directly used as the final alarm results without fully considering sampling gaps, time synchronization deviations, sensor anomalies, unstable category probabilities, common load confusion, and intermittent repetitive consistency, leading to insufficient reliability of on-site alarms and difficulty in suppressing false alarms. This solution creatively adopts a fault perception result correction method. By calculating signal acquisition uncertainty, model prediction uncertainty, load confusion uncertainty, and intermittent consistency evaluation results, the reliability of arc fault prediction data is corrected and intermittent faults are confirmed. This ensures that the final output not only includes the fault category but also the fault reliability, suspected fault circuit, fault time period, risk level, and maintenance handling prompts, improving the interpretability and manageability of alarm results.

[0274] For example, when the model outputs a high probability of arc fault, but there are continuous sampling gaps in the current window or drastic changes in the predicted categories of adjacent windows, this scheme can reduce the reliability of the fault. When the suspected disturbance simultaneously meets the requirements of residual repetition, zero-crossing phase concentration, stable interval between adjacent disturbances and low load confusion, it can improve the reliability of the intermittent arc confirmation results.

[0275] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0276] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention.

[0277] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. A method for detecting intermittent arc faults in a distribution box, characterized in that: The method includes the following steps: Step S1: Data acquisition of the distribution box operation, obtaining the distribution box operation data; Step S2: Intermittent disturbance feature extraction. Based on the operating data of the distribution box, a transient analysis window is constructed, and current mutation features, voltage distortion features, high-frequency disturbance features, zero-crossing phase features, and disturbance repetition features are extracted within the transient analysis window to obtain intermittent disturbance feature data. Step S3: Arc channel residual calculation. Based on the distribution box operation data and intermittent disturbance characteristic data, establish the constraint relationship between the normal load voltage drop, line impedance voltage drop and suspected arc channel voltage drop of the target circuit, and calculate the arc channel residual based on the degree of deviation between the measured voltage and current and the constraint relationship to obtain the arc channel residual characteristic data. Step S4: Arc fault state prediction. The intermittent disturbance characteristic data and arc channel residual characteristic data are input into the arc fault state prediction model. The arc fault state prediction model is used to classify and predict the normal operation state, ordinary load disturbance state, suspected arc state and arc fault state of the target circuit to obtain arc fault prediction data. Step S5: Fault perception result correction. Based on the distribution box operation data, intermittent disturbance characteristic data, arc channel residual characteristic data, and arc fault prediction data, calculate the signal acquisition uncertainty, model prediction uncertainty, load confusion uncertainty, and intermittent consistency evaluation results. Then, correct the arc fault prediction data based on the signal acquisition uncertainty, model prediction uncertainty, load confusion uncertainty, and intermittent consistency evaluation results to obtain fault perception result data.

2. The method for detecting intermittent arc faults in a distribution box according to claim 1, characterized in that: The data acquisition of the distribution box operation is used to obtain the basic operation data required for the intermittent arc fault detection of the distribution box. Specifically, it involves collecting voltage data, current data, residual current data, switch status data, box temperature data, and circuit identification data of the target circuit in the distribution box, and performing time base unification, circuit attribution matching, and abnormal collection point sorting on the collected data to obtain the distribution box operation data. The operating data of the distribution box includes voltage waveform data, current waveform data, residual current change data, switch action status data, box internal temperature change data, acquisition time identifier data, and circuit number identifier data.

3. The method for detecting intermittent arc faults in a distribution box according to claim 2, characterized in that: In step S3, the calculation of the arc channel residual includes the following steps: candidate disturbance window determination, reference window matching, load impedance deviation calculation, line voltage drop residual calculation, arc channel equivalent residual calculation, and residual persistence adjustment. The candidate disturbance window is determined by screening transient analysis windows that meet the disturbance intensity condition and phase occurrence condition based on the current sudden change characteristic, high frequency disturbance characteristic, voltage distortion characteristic and zero-crossing phase characteristic in the intermittent disturbance characteristic data, thus obtaining candidate disturbance window data. The reference window matching process involves selecting adjacent operating windows without significant disturbances from the distribution box operating data based on the circuit number, switch status, current RMS value, load change trend, and acquisition time identifier, thus obtaining normal reference window data.

4. The method for detecting intermittent arc faults in a distribution box according to claim 3, characterized in that: In step S3, the load impedance deviation calculation involves calculating the normal equivalent impedance of the target circuit based on the normal reference window data, and comparing the equivalent impedance within the candidate disturbance window with the normal equivalent impedance to obtain the normal load impedance deviation. The line voltage drop residual calculation is performed by calculating the line impedance voltage drop based on the line equivalent parameters, current change, and current change rate of the target circuit, and comparing the measured voltage disturbance with the line impedance voltage drop to obtain the line voltage drop residual.

5. The method for detecting intermittent arc faults in a distribution box according to claim 4, characterized in that: In step S3, the equivalent residual of the arc channel is calculated by calculating the equivalent residual of the arc channel, the residual of current change, and the residual of voltage change based on the deviation relationship between the normal load response, the line impedance voltage drop, and the measured voltage disturbance, so as to obtain the residual data of the arc channel. The residual persistence processing involves statistically analyzing the residual duration, residual repetition count, and residual phase consistency based on the temporal sequence of candidate disturbance windows, zero-crossing phase, and residual amplitude changes, to obtain residual characteristic data for the arc channel.

6. The method for detecting intermittent arc faults in a distribution box according to claim 5, characterized in that: In step S3, the residual characteristic data of the arc channel specifically includes the deviation of normal load impedance, line voltage drop residual, arc channel equivalent residual, current change residual, voltage change residual and residual persistence characteristics.

7. The method for detecting intermittent arc faults in a distribution box according to claim 6, characterized in that: In step S4, the arc fault prediction data specifically includes the target circuit fault category, fault occurrence time window, fault prediction probability, suspected arc disturbance segment, and initial fault level.

8. The method for detecting intermittent arc faults in a distribution box according to claim 7, characterized in that: In step S5, the fault perception result correction includes the following steps: acquisition uncertainty calculation, prediction uncertainty calculation, load confusion uncertainty calculation, intermittent consistency evaluation, fault credibility correction, and fault perception result output. The acquisition uncertainty calculation is performed based on sampling missing cases, sampling noise levels, time synchronization deviations, and sensor anomaly indicators to calculate signal acquisition uncertainty data. The prediction uncertainty calculation is performed by calculating the model prediction uncertainty based on the prediction probability distribution of different fault categories, the change of prediction results in adjacent windows, and the probability difference between suspected arc state and arc fault state, and obtaining prediction uncertainty data. The load confusion uncertainty calculation is performed based on the switching action state, load current change trend, high-frequency disturbance duration, and arc channel residual characteristic data to obtain load confusion uncertainty data.

9. The method for detecting intermittent arc faults in a distribution box according to claim 8, characterized in that: The intermittent consistency evaluation is calculated based on the number of residual repetitions, residual persistence characteristics, zero-crossing phase characteristics, and adjacent disturbance intervals in the suspected arc disturbance segment, to obtain intermittent consistency data. The fault credibility correction is performed by calculating the fault credibility based on the fault prediction probability, signal acquisition uncertainty, model prediction uncertainty, load confusion uncertainty, and intermittent consistency evaluation results, and obtaining the corrected credibility data. The fault perception result output determines the corrected fault category, intermittent arc confirmation result, suspected fault circuit, fault time period, fault risk level, and operation and maintenance handling prompts based on the correction confidence data, initial fault level, and suspected arc disturbance segment, thus obtaining fault perception result data. The fault perception result data specifically includes the corrected fault category, fault confidence, intermittent arc confirmation result, suspected fault circuit, fault time period, fault risk level, and operation and maintenance handling prompts.

Images