Low-voltage electric leakage detection method, system and device based on power frequency phase anchor, medium

Abstract

The application discloses a low-voltage electric leakage detection method and system based on a power frequency phase anchor, equipment and a medium. The method performs integer cycle segmentation on a residual current signal sequence based on the power frequency phase anchor, ensures that each detection window corresponding to an event trigger includes multiple power frequency cycles of the residual current signal, makes the detection window consistent with the physical cycle of an alternating current system, and reduces feature aliasing caused by phase drift under the existing ordinary sliding window. Moreover, a reference cycle template is constructed to extract a deviation component of each cycle, so that the steady-state background leakage, load inherent leakage and repetitive disturbance synchronized with the working condition are inhibited, the recognition ability for weak electric leakage, intermittent electric leakage and composite interference electric leakage is enhanced, cross-cycle comparison can be performed at the same power frequency phase position, the stable background leakage and the real fault electric leakage can be more effectively distinguished, and therefore, the accuracy of the low-voltage electric leakage detection is greatly improved.

Description

Technical Field

[0001] This invention relates to the field of low-voltage leakage current detection technology, and in particular, to a low-voltage leakage current detection method and system, electronic equipment, and computer-readable storage medium based on power frequency phase anchoring. Background Technology

[0002] Low-voltage distribution networks, as the end point of the power system, have complex structures and variable environments, leading to frequent leakage faults. Accurate leakage detection is crucial for preventing electrical fires and electric shock accidents. Based on current research both domestically and internationally, existing low-voltage leakage detection methods can be mainly categorized into traditional residual current protection, supervised classification methods based on artificial features, and deep learning-based detection methods. Traditional residual current protection typically operates on a fixed threshold, detecting leakage by measuring the current vectors of the phase and neutral lines. However, in practical applications, normal ground leakage, capacitive leakage introduced by electromagnetic compatibility filters, harmonics and high-frequency components from inverters and rectifiers, and transient disturbances caused by switching operations all contribute to the residual current path. This makes it difficult for detection methods relying solely on fixed amplitude thresholds to distinguish between background leakage and fault leakage, easily resulting in false trips, failure to trip, or changes in sensitivity. Supervised classification methods based on artificial features rely on statistical characteristics such as mean, peak value, kurtosis, and spectral entropy. When dealing with weak leakage, intermittent leakage, and complex load switching conditions, they are prone to losing crucial waveform details. Furthermore, while deep learning-based detection methods can automatically extract features compared to supervised classification methods based on manual features, they typically require a large number of labeled samples. However, low-voltage leakage fault samples are often scarce in engineering sites, leading to insufficient model generalization ability. To address the problem of insufficient labeled samples, existing research attempts to apply self-supervised learning frameworks to low-voltage leakage detection. For example, the BYOL mechanism employs online and target networks for collaborative learning, with the target network obtained by the moving average of the online network parameters, allowing for the learning of representations without negative samples. Ultimately, the aforementioned low-voltage leakage detection methods directly process the raw residual current signal, extracting and analyzing the features of the raw residual current signal from multiple sampling points to achieve low-voltage leakage detection. In particular, they consider the absolute amplitude of a single sampling point as the most discriminative information. However, the raw residual current signal features of a single sampling point not only contain fault leakage information but also superimposed background factors such as normal ground leakage, capacitive leakage, harmonic components, and switching transients, resulting in insufficient fault specificity and thus poor accuracy in low-voltage leakage detection. Summary of the Invention

[0003] This invention provides a low-voltage leakage current detection method and system based on power frequency phase anchoring, an electronic device, and a computer-readable storage medium, which can improve the accuracy of low-voltage leakage current detection.

[0004] According to one aspect of the present invention, a low-voltage leakage current detection method based on power frequency phase anchoring is provided, characterized in that it includes the following: The residual current signal sequence and power frequency phase reference signal of the low-voltage distribution network are collected. The residual current signal sequence is segmented into integer cycles by event-driven method according to the power frequency phase reference signal to obtain at least one detection window. The detection window includes residual current signals of multiple power frequency cycles. Phase-synchronous resampling is performed on the residual current signal of each cycle within each detection window to obtain a resampled signal sequence. A reference cycle template corresponding to each detection window is constructed, and the deviation component of the resampled signal of each cycle relative to the reference cycle template is extracted from the resampled signal sequence to obtain the deviation sequence of each detection window relative to the reference cycle template. Based on the residual current signal sequence and the deviation sequence, a multi-view sample is constructed for each detection window; wherein, the multi-view sample includes a one-dimensional time domain view, a differential time-frequency diagram, a cycle-referenced recursive coding diagram, and a half-cycle asymmetric diagram, and the differential time-frequency diagram, the cycle-referenced recursive coding diagram, and the half-cycle asymmetric diagram constitute a two-dimensional scene constraint view; A self-supervised learning model based on the BYOL contrastive learning mechanism is constructed, and a one-dimensional temporal view and a two-dimensional scene constraint view are used as positive sample pairs to pre-train the self-supervised learning model. After pre-training, the model is fine-tuned using a small number of labeled samples, and a classifier is built to output the low-voltage leakage detection results.

[0005] Furthermore, the construction process of the cycle-referenced recursive coding graph is as follows: Phase space reconstruction is performed on the resampled signal and the reference cycle template for each cycle to obtain their state vectors. The cycle reference recursive coding matrix is ​​constructed based on the distance between the two state vectors in the phase space.

[0006] Furthermore, a cycle-referenced recursive encoding matrix is ​​constructed based on the following formula: ; in, This represents the similarity between two state vectors in phase space. This represents the state vector of the current cycle resampled signal in phase space. This represents the state vector of the reference cycle template in phase space. This represents the distance between two state vectors. This represents the phase value of the current cycle resampled signal. This represents the phase value of the reference cycle template. This indicates the preset phase difference threshold. This represents the similarity decay parameter.

[0007] Furthermore, the construction process of the semi-circular asymmetric graph is as follows: The positive half-cycle of each cycle is differentially divided with the mirror-aligned negative half-cycle to construct a half-cycle asymmetric sequence. Then, the half-cycle asymmetric sequences of multiple consecutive cycles within each detection window are stacked in cycle order to form a half-cycle asymmetric map.

[0008] Furthermore, the self-supervised learning model includes an online branch and a target branch. The online branch includes a temporal encoder, an image encoder, a condition-aware gated fusion network, a projection head, and a prediction head. The target branch includes a temporal encoder, an image encoder, and a projection head. The temporal encoder is used to extract temporal features for each detection window, and the image encoder is used to extract multi-scale image features for each detection window. The condition-aware gated fusion network is used to generate fusion weights for temporal features and multi-scale image features based on multi-dimensional condition indicators within each detection window, and to perform weighted fusion of temporal features and multi-scale image features.

[0009] Furthermore, the multi-dimensional operating condition indicators include cycle consistency coefficient, half-cycle asymmetry coefficient, harmonic and high-frequency energy ratio, zero-crossing burst energy ratio, and recursive texture sparsity. The cycle consistency coefficient is used to characterize the correlation between adjacent cycles, the half-cycle asymmetry coefficient is used to characterize the difference between the positive and negative half-cycles, the harmonic and high-frequency energy ratio is used to characterize the strength of high-frequency disturbances and harmonic components, the zero-crossing burst energy ratio is used to characterize the degree of local anomaly near the zero-crossing of the power frequency, and the recursive texture sparsity is used to characterize the abnormal texture distribution in the cycle-referenced recursive coding graph.

[0010] Furthermore, the loss function of the self-supervised learning model is: ; ; in, Indicates the total loss. This represents the BYOL alignment loss term. This represents the loss term due to the consistency constraint of the cycle. This indicates the number of cycles within the current detection window. This represents the cosine similarity between the encoding vectors corresponding to adjacent cycles. This represents the weighting coefficient.

[0011] In addition, another embodiment of the present invention provides a low-voltage leakage current detection system based on power frequency phase anchoring, comprising: The power frequency phase anchoring module is used to acquire the residual current signal sequence and the power frequency phase reference signal of the low-voltage distribution network, and to perform event-driven integer cycle segmentation of the residual current signal sequence according to the power frequency phase reference signal to obtain at least one detection window; wherein, the detection window includes residual current signals of multiple power frequency cycles; The deviation sequence construction module is used to perform phase-synchronous resampling of the residual current signal of each cycle in each detection window to obtain a resampled signal sequence, construct a reference cycle template corresponding to each detection window, and extract the deviation component of the resampled signal of each cycle in the resampled signal sequence relative to the reference cycle template to obtain the deviation sequence of each detection window relative to the reference cycle template. The multi-view sample construction module is used to construct multi-view samples for each detection window based on the residual current signal sequence and the deviation sequence; wherein, the multi-view sample includes a one-dimensional time domain view, a differential time-frequency diagram, a cycle-referenced recursive coding diagram, and a half-cycle asymmetric diagram, and the differential time-frequency diagram, the cycle-referenced recursive coding diagram, and the half-cycle asymmetric diagram constitute a two-dimensional scene constraint view; The self-supervised learning module is used to build a self-supervised learning model based on the BYOL contrastive learning mechanism. It uses a one-dimensional temporal view and a two-dimensional scene constraint view as positive sample pairs to pre-train the self-supervised learning model. The model fine-tuning module is used to fine-tune the model using a small number of labeled samples after pre-training, and to build a classifier to output low-voltage leakage detection results.

[0012] In addition, another embodiment of the present invention provides an electronic device including a processor and a memory, wherein the memory stores a computer program, and the processor executes the steps of the method described above by calling the computer program stored in the memory.

[0013] In addition, another embodiment of the present invention provides a computer-readable storage medium for storing a computer program for low-voltage leakage current detection based on power frequency phase anchoring, wherein the computer program executes the steps of the method described above when running on a computer.

[0014] The present invention has the following beneficial effects: This invention presents a low-voltage leakage current detection method based on power frequency phase anchoring. Taking advantage of the significant power frequency periodicity and operational condition correlation of residual current signals in low-voltage AC systems, the method segments the residual current signal sequence into integer cycles based on power frequency phase anchoring. This ensures that each detection window corresponding to an event trigger includes residual current signals across multiple power frequency cycles, aligning the detection window with the physical cycle of the AC system and reducing feature aliasing caused by phase drift in existing conventional sliding windows. Furthermore, by constructing a reference cycle template to extract the deviation component of each cycle, steady-state background leakage, inherent load leakage, and repetitive disturbances synchronized with the operating condition are suppressed. This enhances the ability to identify weak leakage currents, intermittent leakage currents, and complex interference leakage currents. It allows for cross-cycle comparisons at the same power frequency phase position, more effectively distinguishing between stable background leakage and actual fault leakage, thereby significantly improving the accuracy of low-voltage leakage current detection.

[0015] In addition, the low-voltage leakage current detection system based on power frequency phase anchoring of the present invention also has the above-mentioned advantages.

[0016] In addition to the objectives, features, and advantages described above, the present invention has other objectives, features, and advantages. The invention will now be described in further detail with reference to the figures. Attached Figure Description

[0017] The accompanying drawings, which form part of this application, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is a flowchart illustrating a preferred embodiment of the low-voltage leakage current detection method based on power frequency phase anchoring in this application. Figure 2 This is a schematic diagram showing the accuracy comparison between the low-voltage leakage detection model and the CNN-LSTM model of the present invention when detecting the same untrained leakage data. Figure 3 This is a leakage detection confusion matrix diagram of the CNN-LSTM model during experimental verification in a preferred embodiment of the present invention; Figure 4 This is a leakage current detection confusion matrix diagram of the low-voltage leakage current detection model of the present invention during experimental verification in a preferred embodiment of the present invention; Figure 5 This is a schematic diagram of the module structure of a low-voltage leakage current detection system based on power frequency phase anchoring, according to another embodiment of this application. Detailed Implementation

[0018] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0019] Reference Figure 1 A preferred embodiment of this application provides a low-voltage leakage current detection method based on power frequency phase anchoring, including the following: Step S1: Collect the residual current signal sequence and power frequency phase reference signal of the low-voltage distribution network, and perform event-driven integer cycle segmentation on the residual current signal sequence according to the power frequency phase reference signal to obtain at least one detection window; wherein, the detection window includes residual current signals of multiple power frequency cycles; Step S2: Perform phase-synchronous resampling of the residual current signal of each cycle in each detection window to obtain a resampled signal sequence, construct a reference cycle template corresponding to each detection window, and extract the deviation component of the resampled signal of each cycle in the resampled signal sequence relative to the reference cycle template to obtain the deviation sequence of each detection window relative to the reference cycle template. Step S3: Based on the residual current signal sequence and the deviation sequence, construct a multi-view sample for each detection window; wherein, the multi-view sample includes a one-dimensional time-domain view, a differential time-frequency diagram, a cycle-referenced recursive coding diagram, and a half-cycle asymmetric diagram, and the differential time-frequency diagram, the cycle-referenced recursive coding diagram, and the half-cycle asymmetric diagram constitute a two-dimensional scene constraint view; Step S4: Construct a self-supervised learning model based on the BYOL contrastive learning mechanism, using the one-dimensional temporal view and the two-dimensional scene constraint view as positive sample pairs to pre-train the self-supervised learning model; Step S5: After pre-training, fine-tune the model using a small number of labeled samples, build a classifier, and output the low-voltage leakage detection results.

[0020] It is understood that the low-voltage leakage current detection method based on power frequency phase anchoring in this embodiment is based on the significant power frequency periodicity and operating condition correlation of the residual current signal in the low-voltage AC system. It segments the residual current signal sequence into integer cycles based on power frequency phase anchoring, ensuring that each detection window corresponding to the event trigger includes residual current signals of multiple power frequency cycles. This makes the detection window consistent with the physical cycle of the AC system, reducing feature aliasing caused by phase drift in existing ordinary sliding windows. Furthermore, by constructing a reference cycle template to extract the deviation component of each cycle, steady-state background leakage, inherent load leakage, and repetitive disturbances synchronized with the operating condition are suppressed. This enhances the ability to identify weak leakage current, intermittent leakage current, and composite interference leakage current. It allows for cross-cycle comparisons at the same power frequency phase position, more effectively distinguishing between stable background leakage and actual fault leakage current, thereby greatly improving the accuracy of low-voltage leakage current detection.

[0021] In step S1, zero-sequence current transformers are installed at the feeder terminals, floor distribution boxes, or key branch circuits of the low-voltage distribution network to collect the residual current signal sequence of the low-voltage distribution network. The power frequency phase reference signal can be obtained through phase voltage zero-crossing detection. If it is inconvenient to directly collect the voltage, it can also be obtained by estimating the fundamental component of the phase current or residual current using a phase-locked loop. The sampling frequency can be set to 5kHz to 100kHz, preferably 10kHz to 40kHz, for example, 20kHz. Furthermore, the residual current signal sequence can be preprocessed, including removing DC bias, limiting, eliminating abnormal spikes, normalizing, and anti-power frequency drift filtering. The specific preprocessing process is existing technology and will not be elaborated here. Then, instead of segmenting the residual current signal sequence data according to the existing sliding window mechanism, the residual current signal sequence is segmented into event-driven integer cycles based on the power frequency phase reference signal. The driving events include any one of the following: the root mean square increment of the residual current exceeds a threshold, the local energy increment near the zero crossover exceeds a threshold, the harmonic energy ratio increment exceeds a threshold, and the correlation coefficient between adjacent cycles decreases by more than a threshold. This ensures that each detection window corresponding to the event trigger includes residual current signals of multiple power frequency cycles. Each detection window may contain 2 to 20 power frequency cycles, preferably 4 to 8 power frequency cycles, such as 6 power frequency cycles.

[0022] It is understandable that, compared with the commonly used sliding window method, this invention performs integer cycle segmentation of the residual current signal sequence based on power frequency phase anchoring, ensuring that each detection window corresponding to the event trigger includes multiple power frequency cycles of residual current signal. This makes the feature extraction process consistent with the physical cycle of the AC system, reduces feature aliasing caused by phase drift under the existing ordinary sliding window, and helps to improve the accuracy of low-voltage leakage current detection.

[0023] In addition, in step S2, the residual current signal of each cycle within each detection window is resampled in phase synchronization so that each cycle is represented as a signal under a unified phase coordinate. , k Indicates the cycle number. m The location of the power frequency phase sampling point is indicated to obtain the resampled signal sequence. Then, a reference cycle template is constructed using the phase-by-phase median method, consisting of 3 to 20 stable cycles preceding the current detection window, preferably 5 to 10. Alternatively, a reference cycle template can be obtained by weighted averaging of stable cycles under the same historical load, time period, or operating mode. If historical stable data is unavailable, several cycles with the highest correlation can be selected from the current detection window and robustly averaged to obtain the reference cycle template. Next, the resampled signal for each cycle is calculated. Compared to referencing the Zhou Bo template The deviation component can be expressed as: This yields the deviation sequence of each detection window compared to the reference cycle template. d .

[0024] It is understood that this invention extracts deviation components by constructing a reference cycle template, retaining abnormal offsets, half-cycle imbalances, and local burst components caused by faults, while weakening background leakage and repetitive operating condition disturbances. This suppresses steady-state background leakage, inherent load leakage, and repetitive disturbances synchronized with the operating conditions, thereby enhancing the ability to identify weak leakage current, intermittent leakage current, and composite interference leakage current. It facilitates cross-cycle comparison at the same power frequency phase position and can more effectively distinguish between stable background leakage and real fault leakage current.

[0025] Additionally, in step S3, the residual current signal sequence within each detection window is... and deviation sequence d As a dual-channel one-dimensional time-domain input, a constructed one-dimensional time-domain view can be obtained, which can simultaneously display the original waveform information and the abnormal deviation information of each cycle relative to the reference cycle. The specific construction principle of the one-dimensional time-domain view is existing technology and will not be elaborated here. In addition, performing a short-time Fourier transform on the deviation sequence can yield a differential time-frequency plot. The frame length of the short-time Fourier transform can be set to 32–512 points, preferably 64–256 points, for example, 128 points. The frame shift can be set to 25%–75% of the frame length, preferably 50%. The differential time-frequency plot is used to characterize the energy transfer of the deviation components on the time and frequency axes.

[0026] It is understandable that, since recursion graphs can characterize the phase space similarity of time series, existing technologies typically construct recursion graphs using the original residual current signal sequence. However, recursion graphs constructed for a single sequence are naturally symmetrical around the main diagonal, easily introducing mirror redundancy, requiring further diagonal cutting, trimming, or reconstruction of the recursion graph. Optionally, in this invention, phase space reconstruction is first performed on the resampled signal and the reference cycle template for each cycle, assuming the embedding dimension is... p The delay step size is τ Then the state vectors of the two can be obtained as follows: , ,in, This represents the state vector reconstructed in phase space by the current cycle resampled signal. This represents the state vector reconstructed in phase space using a reference cycle template, with the embedding dimension being... p The delay step size can be 2 to 8, preferably 3 to 5, for example, 3. τOne to ten sampling points can be selected, for example, two sampling points. Then, a cycle-reference recursive coding matrix is ​​constructed based on the distance between the two state vectors in phase space, thus obtaining the cycle-reference recursive coding map. The cycle-reference recursive coding matrix is ​​constructed based on the following formula: ; in, This represents the similarity between two state vectors in phase space, i.e., the (th)th state vector in the cycle-referenced recursive encoding matrix. i , j ) elements, This represents the state vector of the current cycle resampled signal in phase space. This represents the state vector of the reference cycle template in phase space. This represents the distance between two state vectors. This represents the phase value of the current cycle resampled signal. This represents the phase value of the reference cycle template. This represents the preset phase difference threshold, which can be 5 to 30, preferably 8 to 15, for example, 10. This represents the similarity attenuation parameter, which can be taken as 0.5 to 2 times the median distance.

[0027] It is understood that the horizontal and vertical axes of the recursive coding diagram of the cycle reference in this invention are derived from the current cycle and the reference cycle, respectively, rather than from the same sequence. Its structure is naturally different from the traditional single-sequence recursive diagram. It does not rely on the symmetry feature of the main diagonal, and can more directly characterize the difference evolution of the current detection window relative to the reference cycle. Furthermore, it limits the participation of the same phase or near-phase positions in the similarity calculation, and can characterize the cross-cycle change at the same power frequency phase position, thereby highlighting the difference evolution direction between background leakage and fault leakage.

[0028] In addition, the present invention also performs differential analysis on the positive half-cycle and the mirror-aligned negative half-cycle of each cycle to construct a half-cycle asymmetric sequence. Then, the half-cycle asymmetric sequences of multiple consecutive cycles within each detection window are stacked in cycle order to form a half-cycle asymmetric diagram, which is used to characterize the differences between the positive and negative half-cycles in amplitude, local energy and burst behavior.

[0029] The differential time-frequency diagram, the cycle reference recursive coding diagram, and the half-cycle asymmetric diagram form a two-dimensional scene constraint view, which allows the model to simultaneously utilize frequency domain offset, in-phase cross-cycle difference, and half-cycle imbalance information, thereby enhancing the model's ability to distinguish between weak leakage current, intermittent leakage current, and composite interference leakage current.

[0030] It is understandable that existing technologies typically use the absolute amplitude of a single sampling point as the most discriminative feature. However, the absolute amplitude mainly reflects the instantaneous operating point and does not directly reflect the fault attributes. The residual current signal in a low-voltage power distribution system is inherently superimposed with normal ground leakage, capacitive leakage, harmonic / high-frequency components, and switching transient disturbances. Therefore, the absolute amplitude of a certain sampling point is often simultaneously affected by the load state, power frequency phase, filter branches, electromagnetic compatibility devices, rectifier / frequency converters, and transient disturbances. The absolute amplitude contains a large amount of operating condition information and background information, and is not equivalent to fault information. This is the fundamental reason why the traditional threshold method is prone to false tripping, failure to tripping, and sensitivity variations. In addition, from the perspective of signal processing, the absolute amplitude is a raw observation, with large intra-class fluctuations and large inter-class overlap. Normal background leakage may be large at some times, and weak leakage may not be prominent at some times. Therefore, using the absolute amplitude of a single sampling point as the most discriminative feature in existing technologies leads to insufficient separability between fault and non-fault classes.

[0031] In this invention, considering that AC low-voltage systems are strongly periodic systems, for the same circuit, the same power frequency phase position corresponds to approximately the same power excitation conditions. Therefore, comparing points at the same phase position in different cycles is equivalent to observing the system response differences under the most similar external excitation possible. This eliminates the natural waveform differences caused by phase differences, retaining only whether abnormal changes occur between cycles. Because background leakage usually has strong periodic repeatability, if the system is in a stable operating condition, the response of adjacent cycles at the same phase position should be basically consistent or only fluctuate slightly. However, fault leakage often disrupts this cycle consistency. Even if the absolute amplitude change is not large, the difference between cycles at the same phase position will increase significantly. Therefore, cross-cycle variation actually measures the degree of deviation from the periodic steady state, which is more reflective of whether the anomaly truly originates from a fault than single-point amplitude. Furthermore, under ideal or relatively stable AC operating conditions, many normal leakage components have a certain correspondence with the responses of the positive and negative half-cycles. In particular, the half-cycle morphology of linear, stable, and repetitive background leakage is often relatively balanced. Conversely, the actual leakage fault path often has nonlinearity, time-varying characteristics, contact instability, or unidirectional conduction tendency. This leads to significant differences in the degree of conduction, peak position, local energy, and sudden behavior between the positive and negative half-cycles. From a mechanistic perspective, the impedance of the fault branch may not be nonlinearly constant. Faults such as insulation deterioration, moisture, carbonization, and intermittent contact will cause the conductivity state of the leakage path to be not completely the same in the two half-cycles. Partial discharge, intermittent breakdown, or anomalies near zero crossover are more likely to show half-cycle imbalance. Some faults do not exist uniformly throughout the entire cycle but are more active near specific polarities or phases. Therefore, background repetitive disturbances are more likely to be canceled or identified at the half-cycle level. When the positive half-cycle is subtracted from the mirror-aligned negative half-cycle, repetitive and symmetrical background components are weakened, while asymmetrical anomalies are amplified. Therefore, the half-cycle difference is essentially a measurement of whether the fault has changed the symmetrical structure of the AC response. Furthermore, since the original residual current signal = stable background component + repetitive operating condition component + transient disturbance component + fault abnormal component, if we directly look at the original amplitude, the fault component is only a part of the sum and is often masked by the background. However, this invention uses the current cycle-reference cycle template to extract the deviation component, so the stable background and repetitive operating condition components will be canceled out. What remains is mainly the relatively new, relatively abnormal, and relatively inconsistent part. The deviation component is essentially a differential mode information or abnormal residual information. It is closer to the fault itself than the original value, so it has a higher signal-to-noise ratio and stronger class discrimination ability. Especially for weak leakage current, intermittent leakage current, and compound interference leakage current, the fault itself may not bring a large increase in absolute amplitude, but it often brings a continuous offset or local change relative to the normal template.Therefore, this invention performs cross-cycle comparison at the same power frequency phase position, which can better characterize the disruption of cycle consistency caused by faults. The difference analysis between the positive and negative half-cycles can better reflect the half-cycle imbalance or asymmetrical conduction characteristics caused by the fault path. By extracting the deviation component with reference to the cycle template, it can suppress steady-state background leakage and repetitive operating condition disturbances, highlight the new abnormal components caused by the fault, and has a higher discriminative value than the absolute amplitude, which can greatly improve the accuracy of low-voltage leakage detection.

[0032] Therefore, this invention uses differential time-frequency diagrams, cycle-referenced recursive coding diagrams, and half-cycle asymmetry diagrams to characterize the energy migration, in-phase cross-cycle differences, and half-cycle imbalance information of the deviation components in each detection window on the time and frequency axes, respectively. These can respectively reflect the deviation of the fault from the steady-state background, the disruption of period consistency, and the disruption of half-cycle symmetry. This can more directly reflect the disturbance of leakage faults on the system response structure and has higher discriminative value than absolute amplitude, which can greatly improve the accuracy of low-voltage leakage detection.

[0033] In addition, in step S4, a self-supervised learning model based on the BYOL contrastive learning mechanism is constructed. This self-supervised learning model includes an online branch and a target branch. The online branch includes a temporal encoder, an image encoder, a condition-aware gating fusion network, a projection head, and a prediction head. The target branch includes a temporal encoder, an image encoder, and a projection head. The temporal encoder is used to extract temporal features for each detection window, preferably using BiLSTM, 1D-CNN, or a combination of both. The image encoder is used to extract multi-scale image features for each detection window, specifically including three image scale branches, preferably using 2D-CNN. The condition-aware gating fusion network uses... The system generates fusion weights for temporal features and multi-scale image features based on multi-dimensional working condition indicators within each detection window, and performs weighted fusion of the temporal features and multi-scale image features. The projection head is used to project the features output by the encoder into the embedding space. For example, the projection head in the online branch projects the fused features output by the working condition-aware gated fusion network into the embedding space, and the projection head in the target branch projects the features output by the temporal encoder and the image encoder into the embedding space. The prediction head is used to predict the projected representation of the target branch. The target branch does not have a working condition-aware gated fusion network or prediction head, and does not participate in gradient backpropagation; instead, it is updated through the exponential moving average of the online branch parameters. The self-supervised learning model based on the BYOL contrastive learning mechanism is an existing self-supervised model, and its specific principles and network structure for contrastive learning are existing technologies and will not be elaborated here. Optionally, the working condition perception gating fusion network in this invention does not use the frequency-space fusion module in general image tasks, but designs a special fusion module based on the low-voltage leakage scenario. It generates fusion weights of time-domain features and multi-scale image features based on multi-dimensional working condition indicators in each detection window, and performs weighted fusion of time-domain features and multi-scale image features.

[0034] The multi-dimensional operating condition indicators include cycle consistency coefficient, half-cycle asymmetry coefficient, harmonic and high-frequency energy ratio, zero-crossing burst energy ratio, and recursive texture sparsity. The cycle consistency coefficient is used to characterize the correlation between adjacent cycles, the half-cycle asymmetry coefficient is used to characterize the difference between the positive and negative half-cycles, the harmonic and high-frequency energy ratio is used to characterize the strength of high-frequency disturbances and harmonic components, the zero-crossing burst energy ratio is used to characterize the degree of local anomaly near the zero-crossing of the power frequency, and the recursive texture sparsity is used to characterize the abnormal texture distribution in the cycle-referenced recursive coding graph. After inputting the above multi-dimensional operating condition indicators into the gating network, the fusion weights of temporal features and multi-scale image features can be generated.

[0035] In this invention, the cycle consistency coefficient is defined as the average normalized correlation coefficient between adjacent cycles, which can be expressed as: ,in, Represents the cycle consistency coefficient. K This indicates the number of cycles within the current detection window. Indicates the first k The average of each cycle, This is used to prevent extremely small positive numbers with a denominator of zero.

[0036] In addition, the formula for calculating the semi-circular asymmetry coefficient is: ,in, This represents the semi-circular asymmetry coefficient. Indicates the first k One positive half-cycle of a cycle, , This represents the negative half-cycle after mirror alignment. .

[0037] In addition, the formula for calculating the energy proportion of harmonics and high frequencies is: ,in, This indicates the proportion of harmonic and high-frequency energy. This represents the spectrum of the bias sequence, with the fundamental frequency being... , Indicates harmonic frequency band, B h 2 can be selected f 0 to H f The harmonic frequency band corresponding to 0 Indicates the high-frequency disturbance band. The proportion of fundamental wave energy, harmonics, and high-frequency energy can highlight the degree of anomaly relative to the fundamental wave.

[0038] In addition, the formula for calculating the zero-crossing burst energy ratio is: ,in, Indicates the zero-crossing burst energy ratio. This represents the phase neighborhood corresponding to a zero-crossing at the power frequency, for example, taking ±Δ sampling points centered at each zero-crossing point.

[0039] In addition, the formula for calculating the sparsity of recursive textures is: ,in, Indicates the sparsity of recursive textures. This indicates the side length of the reference recursive encoding matrix for the Zhou wave. This represents the threshold for determining abnormal textures. Represents the characteristic function, Represents an abnormal texture map. , Indicates the first k Each cycle corresponds to a cycle reference recursive encoding matrix. The larger the value, the more significant the difference between the current cycle and the reference cycle. The more scattered and localized the abnormal texture regions are in the entire recursive coding graph, the higher their sparsity.

[0040] Then, the extracted multi-dimensional working condition indicators are used to construct a working condition vector, which is then normalized and input into a gating network. The gating network outputs the weight parameters corresponding to the temporal branch and the shallow, medium, and deep image branches through fully connected layers and nonlinear activation functions. These weights are then normalized using Softmax to obtain the fusion weights. Finally, the temporal features and the multi-scale image features after pooling and projection are weighted and summed according to the fusion weights to obtain the fusion feature representation. For example, the working condition vector can be represented as: To eliminate the dimensional differences between indicators under different operating conditions, the operating condition vector is first normalized to obtain the normalized operating condition vector: The normalization method can be any of min-max normalization, Z-score normalization, or layer normalization. Then, the normalized condition vector is input into the gating network, which uses a multilayer perceptron structure. Its computation process can be represented as follows: , , , This represents a trainable weight matrix. , Indicates the bias term. To represent a non-linear activation function, any one of ReLU, LeakyReLU, or GELU can be used. Indicates the output of the hidden layer. Let represent the output vector of the gating network. The three image scale branches of the image encoder correspond to shallow, medium, and deep image features, respectively. Therefore, the output vector of the gating network can be expressed as: , The original weight parameters represent the time-domain features. , , These represent the original weight parameters for shallow, mid-level, and deep image features, respectively. Then, the Softmax function is used to normalize the original weight parameters to obtain the final fused weights, i.e.: , ,in, This indicates the proportion of temporal features in the final fused features. , , These represent the proportions of shallow, mid-level, and deep image features in the final fused features, respectively. Assume the temporal features output by the temporal encoder are... f t The shallow, medium, and deep image features output by the image encoder are respectively F1. F 2 and F 3. Since temporal features and image features may differ in dimensionality, it is necessary to perform dimensional alignment processing on each branch feature separately. For example, a unified dimensional branch feature vector can be obtained through global average pooling, flattening, and linear mapping. , , , , Indicates global average pooling. This represents a linear mapping layer. In addition to outputting branch-level fusion weights, the gated network can also output channel recalibration weights for image features at various scales, enabling adaptive enhancement or suppression of shallow, mid-, and deep image features based on the current conditions. Let the th... l The channel recalibration weight vector for each scale image feature is: Then it can be expressed as: , This represents the Sigmoid activation function. , Let represent the trainable parameters at the corresponding scale. Then, the image features after channel recalibration are: , This indicates that the elements are weighted by channel. Indicates the first channel after recalibration l Image features at each scale are then mapped to a unified dimension to obtain: Finally, the temporal features and multi-scale image features are weighted and fused to obtain the fused feature representation: .

[0041] It is understood that the working condition perception gating fusion network of the present invention does not use a general image feature fusion module. Instead, it performs weighted fusion based on electrical working condition indicators such as cycle consistency, half-cycle asymmetry, harmonic / high-frequency energy ratio, zero-crossing burst energy, and recursive texture sparsity. This makes the fusion result more consistent with the low-voltage leakage detection task itself and can adaptively adjust the contribution ratio of the time domain branch and the image branches of different scales. This allows the online branch to select more discriminative feature representations under different leakage conditions, different background leakage conditions, and different disturbance intensities.

[0042] In addition, during the pre-training phase, the one-dimensional temporal view and the two-dimensional scene constraint view of the same detection window are used as positive sample pairs to input the online branch and the target branch. Cross-view representation learning is performed using a BYOL-style alignment loss. The target branch parameters are updated using an exponential moving average, and the momentum coefficient can be set to 0.99–0.9999, preferably 0.995–0.999. Optionally, based on the BYOL loss, this invention also introduces a cycle consistency constraint loss according to the characteristics of the low-voltage leakage current detection scenario. The loss function of the self-supervised learning model can be expressed as: ; ; in, Indicates the total loss. This represents the BYOL alignment loss term. This represents the loss term due to the consistency constraint of the cycle. This indicates the number of cycles within the current detection window. This represents the cosine similarity between the encoding vectors corresponding to adjacent cycles. This represents the weighting coefficient, which can be 0.1 to 2.0, preferably 0.2 to 1.0, for example 0.5.

[0043] It is understood that this invention introduces cycle consistency constraints on the basis of the BYOL self-supervised learning framework, which not only enables pre-training using a large amount of unlabeled running data under conditions with few labels, but also improves the model's generalization ability to different loads, different transformer areas and different interference conditions.

[0044] In addition, in step S5, after completing the self-supervised pre-training, a small number of labeled samples are used for downstream fine-tuning, and a classifier is constructed to obtain a low-voltage leakage detection model, which outputs the leakage detection results. The leakage detection results can include normal operating conditions, background leakage or non-fault leakage, slight leakage, severe leakage, intermittent leakage, and suspected fault leakage under combined disturbance conditions. The classifier can be a multilayer perceptron. The specific classifier construction process is existing technology and will not be elaborated here. Furthermore, in the actual detection process, after acquiring the residual current signal sequence, it sequentially undergoes power frequency phase anchoring, integer cycle segmentation, deviation sequence construction, and multi-view sample construction. The obtained multi-view samples are then input into the trained low-voltage leakage detection model to output the low-voltage leakage detection results.

[0045] In addition, to verify the superiority of the low-voltage leakage current detection method based on power frequency phase anchoring of the present invention, comparative experiments were also conducted. The detection accuracy of the CNN-LSTM model and the low-voltage leakage current detection model of the present invention on untrained leakage current data was compared. The comparison results are as follows: Figure 2As shown, the CNN-LSTM model maintains an accuracy of approximately 0.88 for untrained leakage current detection data, while the low-voltage leakage current detection model of this invention maintains an accuracy of approximately 0.98 for untrained leakage current detection data, which is nearly 10% higher than the CNN-LSTM model.

[0046] In addition, to verify the comprehensive performance of the low-voltage leakage current detection method of this invention, this invention also verified the classification ability of this method and existing CNN-LSTM models between normal operating conditions and different leakage faults (such as minor leakage, severe leakage, and intermittent leakage), and analyzed its accuracy, false alarm rate, and false negative rate. The experiment used actual current data from a low-voltage distribution network, which included current signals under different operating conditions, including: normal operation: three-phase current signals collected under normal operating conditions; minor leakage: small-scale leakage caused by slight insulation damage to the equipment casing or grounding problems; severe leakage: significant leakage caused by severe insulation damage to the equipment or grounding faults; intermittent leakage: intermittent current leakage, which may be caused by intermittent equipment failure or sensor problems. The dataset contains 500 samples, each consisting of a current signal over a sampling period, labeled with the corresponding leakage type or normal operating condition. The overall performance of the model is evaluated using accuracy, precision, recall, and F1 score, represented by the characters A, P, R, and F1 respectively. F 1 indicates that the calculation formula is: , , , , TP This indicates a true class (i.e., a leakage fault is correctly detected as a leakage fault). TN This indicates a true negative class (i.e., normal operating conditions are correctly detected as normal). FP This indicates a false positive (i.e., a leakage fault is mistakenly detected under normal operating conditions). FN The CNN-LSTM model's confusion matrix for detecting leakage faults (i.e., false negatives, where leakage faults are mistakenly detected as normal) is shown in the diagram below. Figure 3 As shown, the leakage current detection confusion matrix diagram of the present invention is as follows. Figure 4 As shown in Table 1, the comparison results between the two are as follows: Table 1. Performance Comparison Results ; It can be seen that the low-voltage leakage detection method of the present invention is superior to the CNN-LSTM model in terms of accuracy, precision, recall and F1 score. This proves that the low-voltage leakage detection method of the present invention has a strong detection capability for different types of leakage faults, especially in the multi-domain feature fusion of leakage signals.

[0047] In addition, such as Figure 5 As shown, another embodiment of the present invention also provides a low-voltage leakage current detection system based on power frequency phase anchoring, preferably employing the low-voltage leakage current detection method based on power frequency phase anchoring as described above, including: The power frequency phase anchoring module is used to acquire the residual current signal sequence and the power frequency phase reference signal of the low-voltage distribution network, and to perform event-driven integer cycle segmentation of the residual current signal sequence according to the power frequency phase reference signal to obtain at least one detection window; wherein, the detection window includes residual current signals of multiple power frequency cycles; The deviation sequence construction module is used to perform phase-synchronous resampling of the residual current signal of each cycle in each detection window to obtain a resampled signal sequence, construct a reference cycle template corresponding to each detection window, and extract the deviation component of the resampled signal of each cycle in the resampled signal sequence relative to the reference cycle template to obtain the deviation sequence of each detection window relative to the reference cycle template. The multi-view sample construction module is used to construct multi-view samples for each detection window based on the residual current signal sequence and the deviation sequence; wherein, the multi-view sample includes a one-dimensional time domain view, a differential time-frequency diagram, a cycle-referenced recursive coding diagram, and a half-cycle asymmetric diagram, and the differential time-frequency diagram, the cycle-referenced recursive coding diagram, and the half-cycle asymmetric diagram constitute a two-dimensional scene constraint view; The self-supervised learning module is used to build a self-supervised learning model based on the BYOL contrastive learning mechanism. It uses a one-dimensional temporal view and a two-dimensional scene constraint view as positive sample pairs to pre-train the self-supervised learning model. The model fine-tuning module is used to fine-tune the model using a small number of labeled samples after pre-training, and to build a classifier to output low-voltage leakage detection results.

[0048] In addition, another embodiment of the present invention provides an electronic device including a processor and a memory, wherein the memory stores a computer program, and the processor executes the steps of the method described above by calling the computer program stored in the memory.

[0049] In addition, another embodiment of the present invention provides a computer-readable storage medium for storing a computer program for low-voltage leakage current detection based on power frequency phase anchoring, wherein the computer program executes the steps of the method described above when running on a computer.

[0050] Common computer-readable storage media include: floppy disks, flexible disks, hard disks, magnetic tapes, any other magnetic media, CD-ROMs, any other optical media, punch cards, paper tape, any other physical media with perforated patterns, random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), flash erasable programmable read-only memory (FLASH-EPROM), any other memory chips or cartridges, or any other media readable by a computer. Instructions may further be transmitted or received by a transmission medium. The term transmission medium can include any tangible or intangible medium used to store, encode, or carry instructions for execution by a machine, and includes digital or analog carrier communication signals or intangible media that facilitate communication of such instructions. Transmission media include coaxial cables, copper wires, and optical fibers, which contain conductors for transmitting a bus of computer data signals.

[0051] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of this application can be implemented in various computer languages, such as the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0052] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0053] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0054] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0055] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0056] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

[0057] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A low-voltage leakage current detection method based on power frequency phase anchoring, characterized in that, Includes the following: The residual current signal sequence and power frequency phase reference signal of the low-voltage distribution network are collected. The residual current signal sequence is segmented into integer cycles by event-driven method according to the power frequency phase reference signal to obtain at least one detection window. The detection window includes residual current signals of multiple power frequency cycles. Phase-synchronous resampling is performed on the residual current signal of each cycle within each detection window to obtain a resampled signal sequence. A reference cycle template corresponding to each detection window is constructed, and the deviation component of the resampled signal of each cycle relative to the reference cycle template is extracted from the resampled signal sequence to obtain the deviation sequence of each detection window relative to the reference cycle template. Based on the residual current signal sequence and the deviation sequence, a multi-view sample is constructed for each detection window; wherein, the multi-view sample includes a one-dimensional time domain view, a differential time-frequency diagram, a cycle-referenced recursive coding diagram, and a half-cycle asymmetric diagram, and the differential time-frequency diagram, the cycle-referenced recursive coding diagram, and the half-cycle asymmetric diagram constitute a two-dimensional scene constraint view; A self-supervised learning model based on the BYOL contrastive learning mechanism is constructed, and a one-dimensional temporal view and a two-dimensional scene constraint view are used as positive sample pairs to pre-train the self-supervised learning model. After pre-training, the model is fine-tuned using a small number of labeled samples, and a classifier is built to output the low-voltage leakage detection results.

2. The low-voltage leakage current detection method based on power frequency phase anchoring as described in claim 1, characterized in that, The process of constructing the cycle-referenced recursive coding graph is as follows: Phase space reconstruction is performed on the resampled signal and the reference cycle template for each cycle to obtain their state vectors. A cycle reference recursive coding matrix is ​​constructed based on the distance between the two state vectors in phase space.

3. The low-voltage leakage current detection method based on power frequency phase anchoring as described in claim 2, characterized in that, The cycle-referenced recursive encoding matrix is ​​constructed based on the following formula: ； in, This represents the similarity between two state vectors in phase space. This represents the state vector of the current cycle resampled signal in phase space. This represents the state vector of the reference cycle template in phase space. This represents the distance between two state vectors. This represents the phase value of the current cycle resampled signal. This represents the phase value of the reference cycle template. This indicates the preset phase difference threshold. This represents the similarity decay parameter.

4. The low-voltage leakage current detection method based on power frequency phase anchoring as described in claim 1, characterized in that, The process of constructing the semi-circular asymmetric graph is as follows: The positive half-cycle of each cycle is differentially divided with the mirror-aligned negative half-cycle to construct a half-cycle asymmetric sequence. Then, the half-cycle asymmetric sequences of multiple consecutive cycles within each detection window are stacked in cycle order to form a half-cycle asymmetric map.

5. The low-voltage leakage current detection method based on power frequency phase anchoring as described in claim 1, characterized in that, The self-supervised learning model includes an online branch and a target branch. The online branch includes a temporal encoder, an image encoder, a condition-aware gated fusion network, a projection head, and a prediction head. The target branch includes a temporal encoder, an image encoder, and a projection head. The temporal encoder is used to extract temporal features for each detection window, and the image encoder is used to extract multi-scale image features for each detection window. The condition-aware gated fusion network is used to generate fusion weights for temporal features and multi-scale image features based on multi-dimensional condition indicators within each detection window, and to perform weighted fusion of temporal features and multi-scale image features.

6. The low-voltage leakage current detection method based on power frequency phase anchoring as described in claim 5, characterized in that, The multi-dimensional operating condition indicators include cycle consistency coefficient, half-cycle asymmetry coefficient, harmonic and high-frequency energy ratio, zero-crossing burst energy ratio, and recursive texture sparsity. The cycle consistency coefficient is used to characterize the correlation between adjacent cycles, the half-cycle asymmetry coefficient is used to characterize the difference between the positive and negative half-cycles, the harmonic and high-frequency energy ratio is used to characterize the strength of high-frequency disturbances and harmonic components, the zero-crossing burst energy ratio is used to characterize the degree of local anomaly near the zero-crossing of the power frequency, and the recursive texture sparsity is used to characterize the abnormal texture distribution in the cycle-referenced recursive coding graph.

7. The low-voltage leakage current detection method based on power frequency phase anchoring as described in claim 1, characterized in that, The loss function of the self-supervised learning model is: ； ； in, Indicates the total loss. This represents the BYOL alignment loss term. This represents the loss term due to the consistency constraint of the cycle. This indicates the number of cycles within the current detection window. This represents the cosine similarity between the encoding vectors corresponding to adjacent cycles. This represents the weighting coefficient.

8. A low-voltage leakage current detection system based on power frequency phase anchoring, characterized in that, include: The power frequency phase anchoring module is used to acquire the residual current signal sequence and the power frequency phase reference signal of the low-voltage distribution network, and to perform event-driven integer cycle segmentation of the residual current signal sequence according to the power frequency phase reference signal to obtain at least one detection window; wherein, the detection window includes residual current signals of multiple power frequency cycles; The deviation sequence construction module is used to perform phase-synchronous resampling of the residual current signal of each cycle in each detection window to obtain a resampled signal sequence, construct a reference cycle template corresponding to each detection window, and extract the deviation component of the resampled signal of each cycle in the resampled signal sequence relative to the reference cycle template to obtain the deviation sequence of each detection window relative to the reference cycle template. The multi-view sample construction module is used to construct multi-view samples for each detection window based on the residual current signal sequence and the deviation sequence; wherein, the multi-view sample includes a one-dimensional time domain view, a differential time-frequency diagram, a cycle-referenced recursive coding diagram, and a half-cycle asymmetric diagram, and the differential time-frequency diagram, the cycle-referenced recursive coding diagram, and the half-cycle asymmetric diagram constitute a two-dimensional scene constraint view; The self-supervised learning module is used to build a self-supervised learning model based on the BYOL contrastive learning mechanism. It uses a one-dimensional temporal view and a two-dimensional scene constraint view as positive sample pairs to pre-train the self-supervised learning model. The model fine-tuning module is used to fine-tune the model using a small number of labeled samples after pre-training, and to build a classifier to output low-voltage leakage detection results.

9. An electronic device, characterized in that, The method includes a processor and a memory, wherein the memory stores a computer program, and the processor executes the steps of the method as described in any one of claims 1 to 7 by calling the computer program stored in the memory.

10. A computer-readable storage medium for storing a computer program for low-voltage leakage current detection based on power frequency phase anchoring, characterized in that, The computer program, when run on a computer, performs the steps of the method as described in any one of claims 1 to 7.

Images