Low-voltage area electric leakage and backflow monitoring and early warning method based on edge computing

Abstract

The present application relates to the technical field of power system, disclose a kind of based on edge computing's low-voltage area leakage return flow monitoring early warning method, comprising: in edge subnode extraction zero sequence active current phasor and positive sequence voltage phase angle and upload;The reference positive sequence voltage phase angle of acquisition edge main node is calculated phase angle drift deviation, and phase alignment compensation is executed to zero sequence active current phasor;According to physical topology, the topological phasor residual of adjacent level node is calculated, and space covariance matrix is constructed and fundamental wave maximum eigenvalue and main eigenvector are extracted;Orthogonal projection operation is carried out to harmonic covariance matrix to main eigenvector to calculate projection coincidence degree coefficient;Based on the out-of-limit situation of maximum eigenvalue and projection coincidence degree coefficient, distinguish nonlinear leakage state and linear error state and trigger early warning.The present application eliminates the phase alignment deviation caused by communication delay, distinguishes topology file error and real leakage fault using spatial collinearity characteristics, and avoids leakage protection misoperation.

Description

Technical Field

[0001] This invention relates to the field of power system technology, specifically to a method for monitoring and early warning of leakage current backflow in low-voltage distribution areas based on edge computing. Background Technology

[0002] Leakage monitoring in low-voltage distribution network areas plays a crucial role in ensuring power safety and stable operation of distribution lines. With the development of distribution Internet of Things (IoT) technology, distributed collaborative monitoring methods based on edge computing architecture are gradually being applied to leakage protection in distribution areas. This mainly involves deploying edge computing devices at the main node and various sub-nodes in the distribution area to collect electrical phasor data such as current and voltage and perform calculations based on the physical topology to locate and warn of leakage anomalies.

[0003] In actual operating environments, the aforementioned distributed collaborative computing method faces multiple technical bottlenecks. Due to the nondeterministic latency in the underlying communication network of the distribution area during data transmission, it is difficult for the phasor data uploaded by child nodes distributed in different physical locations to achieve accurate phase alignment on a unified historical time section. This latency jitter will directly lead to distortion in the calculation of topological phasor residuals between nodes. At the same time, the inherent distributed capacitance of low-voltage distribution lines will continuously generate capacitive leakage current, which, combined with high-frequency random interference noise in the field environment, further reduces the numerical stability of the underlying current vector operation.

[0004] In long-term operation and maintenance of distribution transformers, it is common for discrepancies to exist between the physical connection records of users and the actual topology on site. When the system performs Kirchhoff's Current Law calculations based on the discrepancies in the distribution transformer topology records, pure topology recording errors can transform the normal load current of ordinary users into false topology phasor residuals at the calculation level. Existing leakage current determination mechanisms rely solely on the single case of residual amplitude exceeding the limit for judgment, and cannot distinguish from the physical characteristics of linear mathematical errors caused by incorrect connection records from nonlinear leakage current faults caused by actual insulation degradation of the lines. This leads to the system triggering incorrect leakage current alarms or issuing incorrect tripping commands, resulting in unplanned power outages and reducing the power supply reliability of the distribution network. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a low-voltage distribution area leakage current monitoring and early warning method based on edge computing. This method solves the problems of residual distortion caused by communication delays and the inability to distinguish between topology file errors and actual leakage faults in existing low-voltage distribution area leakage current monitoring, which can lead to false alarms.

[0006] To achieve the above objectives, the present invention provides the following technical solution: This invention provides a method for monitoring and early warning of leakage current backflow in low-voltage distribution areas based on edge computing. The method includes the following steps: Extract the zero-sequence active current phasor of the fundamental and lower harmonics and the phase angle of the fundamental positive-sequence voltage at the edge sub-nodes, and upload the set of characteristic parameters to the edge master node. Obtain the reference positive sequence voltage phase angle of the edge master node, calculate the positive sequence phase angle drift deviation, and construct a rotation operator to perform phase alignment compensation on the zero sequence active current phasor. Based on the physical topology architecture, the topological phasor residuals between adjacent level nodes are calculated, and the covariance matrix is ​​constructed within the time window. The fundamental covariance matrix is ​​then decomposed into eigenvalues ​​to extract the maximum eigenvalue and the principal eigenvector. The harmonic covariance matrix is ​​orthogonally projected onto the space spanned by the principal eigenvectors to calculate the projection overlap coefficient. Based on the over-limit situation of the maximum eigenvalue and the projection overlap coefficient, the nonlinear leakage state and the linear error state are distinguished and an early warning is triggered.

[0007] Using the above technical solution, this invention calculates the phase angle drift deviation between nodes using the positive sequence voltage of the edge master node as the time reference benchmark to compensate for the time delay jitter introduced by the underlying communication, and realizes the unified alignment of the distributed phasor data of the distribution area in the time section. Furthermore, it solves the topological phasor residuals and constructs the spatiotemporal covariance matrix according to Kirchhoff's current law to suppress the interference of independent random noise and extract the spatial synchronization characteristics of the residual fluctuations in each physical interval. Finally, based on the collinearity characteristics of the fundamental wave and the associated harmonics generated by high-resistance leakage due to insulation deterioration in the spatial distribution, it performs orthogonal projection operation on the harmonic covariance matrix to quantify the energy distribution ratio, thereby distinguishing between non-fault residuals caused by errors in the physical topology file and real nonlinear leakage faults, and avoiding false operation of the distribution area leakage protection.

[0008] In a preferred embodiment of the present invention, the step of extracting the zero-sequence active current phasors of the fundamental and lower harmonics at the edge sub-nodes includes: High-frequency sampling and discrete Fourier transform are performed on the three-phase voltage and three-phase current locally at the edge sub-nodes to obtain the original zero-sequence current phasor and the original zero-sequence voltage phasor at a specific frequency. Preset leakage current calculation and lockout threshold; When the amplitude of the original zero-sequence voltage is less than the leakage current calculation latch-up threshold, the unified time reference reference phase of the entire distribution area is used as the reference direction for orthogonal projection, and orthogonal projection calculation is performed on the original zero-sequence current phasor to obtain the zero-sequence active current phasor. When the amplitude of the original zero-sequence voltage is greater than or equal to the leakage current calculation latch-up threshold, the original zero-sequence current phasor is orthogonally projected onto the direction of the original zero-sequence voltage phasor of the corresponding frequency, and an anti-overflow minimum positive threshold is introduced into the denominator of the projection calculation to prevent overflow after division by zero, and the zero-sequence active current phasor after calculating the complex real part is obtained.

[0009] In the above technical solution, the capacitive leakage component in the original zero-sequence current is eliminated by orthogonal projection operation, and the zero-sequence active current phasor reflecting the resistive leakage characteristics is extracted. At the same time, leakage calculation lockout threshold and overflow prevention minimum positive threshold are introduced to prevent division-to-zero overflow and phase change caused by small random noise when the three-phase load is balanced and the zero-sequence voltage is close to zero, thus ensuring the numerical stability of the distribution network current vector calculation process.

[0010] In a preferred embodiment of the present invention, the steps of obtaining the reference positive-sequence voltage phase angle of the edge master node, calculating the positive-sequence phase angle drift deviation, and constructing a rotating operator to perform phase alignment compensation on the zero-sequence active current phasor include: Based on the beacon frame mechanism of the underlying communication network, the fundamental positive sequence voltage phase angle of the edge master node at the same historical sampling time section as the edge child node is extracted as a physical reference quantity. The difference between the physical reference value and the fundamental positive sequence voltage phase angle of the corresponding edge sub-node is calculated to obtain the positive sequence phase angle drift deviation of the edge sub-node relative to the edge master node. In the complex plane, a corresponding phase rotation compensation operator is constructed using the positive sequence phase angle drift deviation and the proportional amplification factor of the corresponding harmonic frequency. The phase rotation compensation operator is applied to the original zero-sequence active current phasor of the corresponding frequency to obtain the zero-sequence active current phasor after phase alignment compensation.

[0011] In the above technical solution, based on the high consistency of positive sequence voltage distribution in the physical space of the power grid, it serves as a reference benchmark for phase alignment of the entire distribution area. By calculating the positive sequence phase angle drift deviation of each edge sub-node relative to the edge master node, the nondeterministic equivalent time delay of the underlying communication system is obtained. The rotation operator formed by the deviation and the harmonic frequency ratio is used to perform phase rotation compensation, correct the phasor angle deviation caused by time delay jitter, and complete the spatial phase alignment of multi-node phasors.

[0012] In a preferred embodiment of the present invention, the step of calculating the topological phasor residuals between adjacent level nodes based on the physical topology includes: The station area is divided into multiple monitoring intervals, and data integrity verification is performed before calculation. If the data of the edge sub-nodes corresponding to any monitoring interval in the current calculation period is not fully received, the current calculation period is determined to be an invalid section with missing data, and the residual data of all monitoring intervals in the entire station area in the current calculation period are discarded. For a monitoring interval with complete data, the zero-sequence active current phasor of the parent node at the beginning of the monitoring interval after phase alignment compensation is extracted, and the sum of the zero-sequence active current phasors of all direct child nodes at the end of the monitoring interval after phase alignment compensation is extracted. Perform a vector subtraction operation on the sum of the phasor corresponding to the first parent node and the phasor corresponding to the direct child node to obtain the topological phasor residual of the corresponding monitoring interval at the corresponding frequency.

[0013] In the above technical solution, the phasor difference between adjacent level nodes is calculated according to Kirchhoff's current law. The topological phasor residual objectively reflects the ground leakage return component in the corresponding physical interval. The pre-calculation data integrity verification mechanism removes invalid data sections caused by communication packet loss, prevents missing data from causing a surge in artificial calculation residuals, and ensures the structural integrity of the subsequent residual snapshot matrix in the spatiotemporal dimension.

[0014] In a preferred embodiment of the present invention, the step of constructing the covariance matrix within a time window includes: A sliding time window containing multiple consecutive calculation cycles is set. The topological phasor residuals of each monitoring interval collected within the sliding time window are combined according to the spatial and temporal dimensions to construct a residual snapshot matrix. Row mean centering is then performed to obtain the complex snapshot matrix of topological phasor residuals after centering and mean removal. The complex snapshot matrix of the topological phasor residual is multiplied by its complex conjugate transpose, and then averaged in combination with the total number of discrete calculation cycles within the sliding time window to generate a multidimensional spatiotemporal covariance matrix of the corresponding frequency.

[0015] In the above technical solution, the residual data of a single-point discrete time section is converted into a sliding time window snapshot matrix across the calculation cycle, and further solved into a multidimensional spatiotemporal covariance matrix. This process suppresses non-stationary random high-frequency noise interference through correlation calculation and extracts the spatiotemporal correlation characteristics of leakage current fluctuations from the same source in the interval with electrical connection relationship.

[0016] In a preferred embodiment of the present invention, the step of performing an orthogonal projection operation on the space spanned by the principal eigenvectors to calculate the projection coincidence coefficient includes: The complex conjugate transpose of the principal eigenvector, the harmonic covariance matrix, and the principal eigenvector are multiplied together and used as the numerator of the projected energy. Calculate the trace of the harmonic covariance matrix and add a minimal positive tolerance to prevent the denominator from being zero, as the total energy denominator term; Divide the numerator of the projected energy by the denominator of the total energy to obtain the projection overlap coefficient at the corresponding frequency.

[0017] In the above technical solution, orthogonal projection operation is used to quantify the degree of spatial collinearity between the energies of different frequencies. The numerator of the projected energy reflects the energy mapping value of the harmonic wave in the direction of the fundamental wave principal eigenvector, and the denominator of the total energy reflects the total energy scale of the spatial distribution of the harmonics at that frequency. The ratio of the two is used to calculate the projection overlap coefficient, which is used to quantitatively evaluate the concentration ratio of harmonic wave energy in the direction of the fundamental wave dominant wave.

[0018] In a preferred embodiment of the present invention, the step of distinguishing between nonlinear leakage current state and linear error state based on the limit-crossing condition of the maximum eigenvalue and the projection overlap coefficient includes: Determine whether the largest eigenvalue of the fundamental covariance matrix is ​​greater than the largest eigenvalue threshold, and simultaneously determine whether any of the calculated projection overlap coefficients of each specified lower harmonic are greater than the projection overlap threshold. When the maximum eigenvalue exceeds the limit and at least one low-order harmonic projection overlap coefficient exceeds the limit, the current transformer area is determined to be in a nonlinear leakage state. When the maximum eigenvalue exceeds the limit and the projection coincidence coefficients of all lower harmonics do not exceed the limit, the current transformer area is determined to be in a linear error state.

[0019] In the above technical solution, the residual caused by the physical connection error is essentially the user's normal load current, which is characterized by the fundamental frequency and the generated high-order harmonic components are weak. It is difficult to form a synchronous and spatially concentrated energy distribution with the fundamental residual, and the corresponding projection overlap coefficient is maintained at the level that does not exceed the limit. However, the actual insulation deterioration leakage current has nonlinear impedance characteristics, and the associated low-order harmonics it generates show a high degree of spatial collinearity with the fundamental frequency. Based on this, the dual-condition judgment logic can physically separate the normal linear load error from the nonlinear leakage fault.

[0020] In a preferred embodiment of the present invention, the step of triggering the warning includes: When a linear error state is determined, the leakage alarm signal is intercepted, and a transformer topology correction work order is generated instead. The transformer topology correction work order is then reported to the upper-level master station system through the communication network. When a non-linear leakage condition is detected, a leakage alarm signal is triggered to issue a low-voltage switch trip command or prompt for on-site troubleshooting.

[0021] In the above technical solution, a hierarchical and classified response mechanism is executed based on the status determination result. When the status is determined to be linear error, the trip signal is intercepted and converted into an error correction work order to eliminate malfunctions caused by inaccurate topology files. When the status is determined to be a real leakage current state, an action command is triggered to ensure the reliable operation of the leakage current protection function of the distribution area.

[0022] In a preferred embodiment of the present invention, the maximum eigenvalue threshold is set based on the background leakage current noise level obtained during the adaptive learning period of the system in the early stage of commissioning, and is configured to be 1.5 to 3 times the background leakage current noise level; the projection overlap threshold is configured to be 0.75 to 0.9.

[0023] In the above technical solution, the background noise level of adaptive learning in the early stage of system commissioning is used as the benchmark for setting the maximum feature value threshold, so that the judgment boundary is adjusted according to the inherent insulation leakage state of different distribution network areas, thereby improving the objective adaptability of state identification.

[0024] In a preferred embodiment of the present invention, the method further includes the step of locking the fault region based on the main feature vector, the step comprising: Calculate the modulus of each element in each dimension of the main feature vector; By comparing the magnitudes of the elements in each dimension, the index of the dimension with the largest magnitude is extracted, and the index of the dimension with the largest magnitude is mapped back to the preset physical topology of the transformer area, thereby locking the interval corresponding to the largest magnitude as the actual physical location where leakage occurs.

[0025] In the above technical solution, each dimension element inside the main feature vector directly corresponds to different physical monitoring intervals within the transformer area. The magnitude of the value represents the relative distribution ratio of leakage current fluctuation energy in each interval. By extracting the dimension index corresponding to the maximum magnitude and combining it with the topology mapping, the electrical topology positioning of the leakage current fault interval can be achieved.

[0026] This invention provides a method for monitoring and early warning of leakage current backflow in low-voltage distribution areas based on edge computing. It has the following beneficial effects: 1. This invention extracts the reference positive-sequence voltage phase angle of the edge master node, calculates the positive-sequence phase angle drift deviation between each node, and constructs a phase rotation compensation operator to perform phase alignment compensation on the zero-sequence active current phasor. This feature utilizes the consistency of the positive-sequence voltage distribution in the physical space of the power grid to offset the phase deviation introduced by the nondeterministic delay of the underlying communication network, realizes the alignment of distributed phasor data across the entire distribution area on a unified time section, and eliminates the distortion in the calculation of topological phasor residuals caused by communication delay jitter.

[0027] 2. This invention constructs a multidimensional spatiotemporal covariance matrix of topological phasor residuals, extracts the fundamental wave principal eigenvector, and performs orthogonal projection operation on the harmonic covariance matrix onto the principal eigenvector to calculate the projection overlap coefficient. This feature utilizes the physical law that the associated low-order harmonics excited by insulation deterioration leakage current are collinear with the fundamental wave in spatial distribution, and uses the projection overlap as a distinguishing criterion. At the algorithm level, it separates the linear mathematical error caused by the error in the physical connection file of the transformer area from the actual nonlinear leakage fault, and avoids the leakage protection malfunction caused by the inaccuracy of the topological file.

[0028] 3. This invention extracts the zero-sequence active current phasor by using a reference phase orthogonal projection method on the edge sub-node side. It introduces a leakage current calculation lockout threshold and an anti-overflow minimum positive threshold into the calculation model, and combines it with mean-neutralization processing within the sliding time window. The above feature combination eliminates the capacitive leakage component generated by the line distributed capacitance and suppresses high-frequency random noise. At the same time, it prevents zero-divide-off overflow and phase abrupt changes caused by small noise under the three-phase load balance state of the transformer area, ensuring the numerical stability of the leakage current state determination process. Attached Figure Description

[0029] Figure 1 This is a schematic diagram of the system architecture of the present invention; Figure 2 This is a schematic diagram of the method flow of the present invention; Figure 3 This is a schematic diagram of the feature extraction, phase compensation, and matrix construction process of the present invention; Figure 4 This is a logic diagram of the true / false reflux identification and response based on orthogonal projection of the present invention. Figure 5 This is a spatial clustering distribution diagram of true and false faults in this invention.

[0030] Among them, 10 is the feature extraction module; 20 is the phase compensation module; 30 is the matrix construction module; and 40 is the state determination module. Detailed Implementation

[0031] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0032] See attached document Figure 1 The hardware system that operates on the method of the present invention may include: an edge master node, an edge child node, and a communication network connecting the edge master node and the edge child node.

[0033] Edge master nodes are deployed on the low-voltage side of transformers in low-voltage distribution network areas. They adopt intelligent distribution transformer terminals with independent computing units. Edge master nodes serve as computing entities for data aggregation and matrix analysis within the distribution area, and internally integrate the physical topology archive records of the distribution area.

[0034] Edge sub-nodes are distributed and deployed at various levels of branch boxes and user terminals of low-voltage lines, including intelligent branch monitoring units and smart meters. Edge sub-nodes have local high-frequency electrical parameter sampling and discrete signal processing capabilities.

[0035] The communication network is a high-speed power line carrier communication network. The edge master node and each edge sub-node exchange messages through the communication network and rely on the beacon frame mechanism at the bottom layer of the communication network to achieve time synchronization. The beacon frame mechanism provides millisecond-level system-level coarse time synchronization to ensure the basic alignment of the sampling time windows of each node.

[0036] Based on the aforementioned hardware environment limitations, in order to realize the perception and analysis of leakage current return status, the system includes a feature extraction module 10, a phase compensation module 20, a matrix construction module 30, and a status determination module 40 at the logic processing level.

[0037] The feature extraction module 10 operates within the edge master node and each edge sub-node. The feature extraction module 10 performs high-frequency sampling and fast Fourier transform on the three-phase voltage and three-phase current at the edge master node and each edge sub-node level. The feature extraction module 10 extracts the zero-sequence active current phasor and the fundamental positive-sequence voltage phase angle at a specified frequency locally. For the feature parameter set extracted by each edge sub-node, the feature extraction module 10 sends it to the edge master node through the communication network; for the feature parameter set extracted by the edge master node itself, the feature extraction module 10 directly transmits it locally to the phase compensation module 20.

[0038] The phase compensation module 20 operates at the edge master node. After the edge master node receives the characteristic parameters of the edge sub-nodes, the phase compensation module 20 extracts the reference positive sequence voltage phase angle of the edge master node itself. The phase compensation module 20 calculates the phase angle deviation between different spatial nodes and constructs a rotation transformation operator. The phase compensation module 20 performs dynamic phase rotation compensation on the zero-sequence active current phasors of each frequency reported by the edge sub-nodes to align the time cross-section reference between multiple nodes.

[0039] The matrix construction module 30 operates based on the compensation phasor output by the phase compensation module 20. The matrix construction module 30 performs vector subtraction operation between nodes according to the physical topology tree structure preset by the edge master node, and outputs the topology phasor residual sequence of the fundamental wave and the lower harmonics. The matrix construction module 30 constructs the covariance matrix of the residual sequence within a preset time window and completes the eigenvalue decomposition operation.

[0040] The state determination module 40 receives the eigenvalues ​​and eigenvector parameters output by the matrix construction module 30. The state determination module 40 performs spatial mapping operations between covariance matrices of different frequencies. The state determination module 40 uses the calculated projection overlap coefficients combined with the eigenvalue over-limit conditions to distinguish between nonlinear leakage current state and linear error state. Based on the identification results, the state determination module 40 executes alarm logic instructions.

[0041] See attached document Figure 2 This invention provides a method for monitoring and early warning of leakage current backflow in low-voltage distribution areas based on edge computing. The method includes the following steps: S1, the feature extraction module 10 performs high-frequency sampling and discrete signal processing on the three-phase voltage and three-phase current locally on the edge sub-node, extracts the zero-sequence active current phasor of the fundamental wave and the specified low-order harmonics and the phase angle of the fundamental wave positive-sequence voltage, and uploads the extracted feature parameter set to the edge master node. S2, the phase compensation module 20 obtains the reference positive sequence voltage phase angle of the edge master node itself, calculates the positive sequence phase angle drift deviation between the master node and each edge sub-node, uses the deviation to construct a rotation operator and performs phase alignment compensation on the reported zero sequence active current phasor. S3, the matrix construction module 30 calculates the frequency topological phasor residuals between adjacent level nodes based on the preset physical topology architecture of the transformer area, constructs the covariance matrix of the residual time series within the preset time window, and performs eigenvalue decomposition on the fundamental covariance matrix to extract the maximum eigenvalue and the principal eigenvector. S4, the state determination module 40 performs orthogonal projection operation on the space spanned by the harmonic covariance matrix and the fundamental main eigenvector to calculate the projection overlap coefficient. Based on the over-limit situation of the maximum eigenvalue and the projection overlap coefficient, it distinguishes between nonlinear leakage state and linear error state, and triggers the corresponding early warning response action according to the determination result.

[0042] See attached document Figure 3 In specific implementations, step S1 provided by the present invention may include the following steps: S101, the feature extraction module 10 performs high-frequency sampling and discrete Fourier transform on the three-phase voltage and three-phase current locally acquired by the edge sub-nodes.

[0043] In a specific embodiment of the present invention, multiple edge sub-nodes are configured in the low-voltage distribution network area. The feature extraction module 10 obtains the instantaneous values ​​of the three-phase voltage and the three-phase current of the distribution line where the edge sub-node is located by relying on the analog-to-digital conversion circuit built into each edge sub-node. For the acquisition of instantaneous values, existing AC sampling methods are used. In order to prevent the spectral leakage of the discrete Fourier transform from interfering with the phasor calculation accuracy, the truncation window length of the high-frequency sampling is configured to be an integer multiple of 1 to 5 times the power frequency period of the power grid.

[0044] After acquiring the discrete sampling sequence, the feature extraction module 10 converts the time-domain sequence to the frequency domain through discrete Fourier transform, and extracts the zero-sequence voltage phasor and zero-sequence current phasor at a specific frequency. It can be understood that, in order to cover the typical electrical frequency band generated by high-resistance leakage, the specific frequency is set as the fundamental and low-order harmonic frequency band, specifically the fundamental, third harmonic and fifth harmonic.

[0045] S102, the feature extraction module 10 orthogonally projects the zero-sequence current phasor of each frequency to the zero-sequence voltage phasor of the same frequency to calculate the zero-sequence active current phasor.

[0046] In actual power distribution lines, due to the influence of the distributed capacitance of the insulation layer of long-distance cables, capacitive leakage current will be generated during normal operation of the line. The capacitive leakage current is superimposed on the resistive leakage current caused by three-phase imbalance or insulation aging, which leads to false alarms when the leakage current is directly judged by the current amplitude. According to basic electrical engineering theory, the resistive leakage current is in phase with the zero-sequence voltage, while the capacitive leakage current is orthogonal to the zero-sequence voltage.

[0047] To eliminate capacitive components, the feature extraction module 10 uses projection transformation to obtain the component in the zero-sequence current that is in phase with the zero-sequence voltage. Considering the problem of calculation overflow caused by the zero-sequence voltage amplitude approaching zero when the three phases of the power grid are highly balanced, this step introduces a small positive number of 0.001V to 0.05V as an overflow prevention threshold. In addition, to prevent the phase angle from fluctuating drastically due to random noise interference when the zero-sequence voltage is extremely small, thus causing false topology residuals, this step also sets a leakage current calculation latch-up threshold, such as 0.5V. When the locally acquired original zero-sequence voltage amplitude is less than the latching threshold, to prevent false topological residuals caused by drastic fluctuations in the local zero-sequence voltage phase angle due to random noise, the feature extraction module 10 abandons the use of the local zero-sequence voltage phase angle and instead adopts the unified time reference phase of the entire distribution area as the reference direction for orthogonal projection. Orthogonal projection calculations are then performed on the original zero-sequence current phasor to filter out random noise interference in the phase angle while maintaining the Kirchhoff current vector balance of the topology network. When the original zero-sequence voltage amplitude is greater than or equal to the latching threshold, the orthogonal projection calculation model performed by the feature extraction module 10 is as follows: ; In the formula, edge child nodes In frequency The zero-sequence active current phasor; This is an operation function for taking the real part of a complex number; edge child nodes In frequency The original zero-sequence current phasor; edge child nodes In frequency The original zero-sequence voltage phasor; To prevent overflow of a minimum positive number threshold when the denominator is zero; The imaginary unit; This is the spatial index label for the edge child node. The value of the spatial index label ranges from 1 to... Positive integers between; The total number of edge child nodes configured within the transformer area; Specific electrical frequencies for feature extraction.

[0048] The zero-sequence active current phasor here can characterize the resistive leakage current characteristics, and the zero-sequence active current phasor eliminates capacitive current interference in the mathematical space.

[0049] S103, the feature extraction module 10 extracts the local fundamental positive sequence voltage phase angle and uploads the reduced feature parameter set to the edge master node.

[0050] When the feature extraction module 10 acquires the fundamental frequency band phasor, it simultaneously extracts the fundamental positive sequence voltage phase angle at the edge sub-node. The symmetrical component method for solving the positive sequence voltage phasor is performed based on the existing power system asymmetric analysis methods. The fundamental positive sequence voltage phase angle serves as the basic reference anchor point for subsequent spatiotemporal synchronization calculations of the system.

[0051] The feature extraction module 10 packages the zero-sequence active current phasors and the fundamental positive-sequence voltage phase angle of each frequency according to a preset calculation period of 1s to 300s, forming a set of feature parameters for the time section where the node is located. The feature extraction module 10 transmits the set of feature parameters to the edge master node through the communication network. The duration of the preset calculation period can be configured in combination with the node polling interval of the high-speed power line carrier communication network, and is generally set in the range of 1s to 300s. By only reporting the extracted phasor and phase angle data parameters, the feature extraction module 10 compresses the amount of data transmitted in a single transmission to adapt to the bandwidth limitations of the communication network.

[0052] See attached document Figure 3 In specific implementations, step S2 provided by the present invention may include the following steps: S201, the phase compensation module 20 obtains the reference positive sequence voltage phase angle of the edge master node itself, and calculates the positive sequence phase angle drift deviation between the edge master node and each edge sub-node.

[0053] In low-voltage distribution network environments, the underlying communication network adopts high-speed power line carrier communication technology. High-speed power line carrier communication technology is subject to channel contention and routing relay mechanisms, resulting in non-deterministic communication delays. If differential operations are performed on the phasors of each edge sub-node based directly on the time when the edge master node receives the data packet, the synchronization error will produce residuals in the phasor subtraction operation.

[0054] To quantify and eliminate the impact of communication delay, this embodiment introduces the positive sequence voltage of the transformer substation as a spatial reference benchmark. The physical spatial range of the same low-voltage transformer substation is within 50m to 500m, and the electrical impedance of the line is mainly resistance. When leakage faults or single-phase unbalanced load fluctuations occur, the phase of the positive sequence voltage of the power grid is less affected. The distribution of the positive sequence voltage phase in a specific substation space maintains good consistency. The difference in the phase angle of the positive sequence voltage between different spatial nodes can map the phase angle misalignment caused by the equivalent delay of the underlying communication system. The phase compensation module 20 uses the drift deviation calculated by the positive sequence voltage phase angle to eliminate microsecond-level residual communication delay jitter and sampling phase deviation, thereby achieving microsecond-level high-precision spatiotemporal alignment and thus forming a two-level hierarchical synchronization architecture with the underlying beacon frame mechanism.

[0055] In practical implementation, based on the same-source sampling timestamps recorded by the beacon frame mechanism at the bottom layer of the communication network, the phase compensation module 20 extracts the fundamental positive-sequence voltage phase angle of the edge master node at the same historical sampling time section as the edge child node as the physical reference quantity. The phase compensation module 20 calculates the positive-sequence phase angle drift deviation between the edge master node and the corresponding edge child node. The calculation model for the positive-sequence phase angle drift deviation is as follows: ; In the formula, edge child nodes The positive sequence phase angle drift deviation relative to the edge master node; The phase angle of the fundamental positive sequence voltage of the edge master node; edge child nodes The phase angle of the fundamental positive sequence voltage; This is the spatial index label for the edge child node. The value of the spatial index label ranges from 1 to... Positive integers between; The total number of edge child nodes configured within the transformer area; This is the spatial index code for the edge master node.

[0056] S202, the phase compensation module 20 constructs a rotating operator using the positive sequence phase angle drift deviation and performs phase alignment compensation on the reported zero sequence active current phasor.

[0057] After obtaining the positive-sequence phase angle drift deviation of each edge sub-node, the phase compensation module 20 constructs the corresponding phase rotation compensation operator in the complex plane. Considering that the phase offset caused by the same time delay at different frequencies is proportional to the harmonic order, the phase compensation module 20 introduces the harmonic frequency as a proportional amplification factor when constructing the rotation operator for higher harmonics.

[0058] Phase compensation module 20 applies the constructed rotation operator to the zero-sequence active current phasors of each frequency reported by the edge sub-nodes to compensate for the phase offset caused by communication delay. The calculation model for phase alignment compensation is as follows: ; In the formula, edge child nodes In frequency The zero-sequence active current phasor after phase alignment compensation; edge child nodes In frequency The original zero-sequence active current phasor; The imaginary unit; Specific electrical frequencies for feature extraction; edge child nodes The positive sequence phase angle drift deviation relative to the edge master node; This is the spatial index label for the edge child node. The value of the spatial index label ranges from 1 to... Positive integers between; This represents the total number of edge child nodes configured within the transformer area.

[0059] Through dynamic phase rotation compensation, the phase compensation module 20 reduces the impact of time delay jitter caused by nondeterministic communication networks, and adjusts the phasors of nodes scattered in different physical locations with different communication arrival times to the same time section reference with the edge master node as the reference, providing a data basis for spatial topology differential calculation.

[0060] See attached document Figure 3 In specific implementations, step S3 provided by the present invention may include the following steps: S301, the matrix construction module 30 calculates the frequency topological phasor residuals between adjacent level nodes based on the preset physical topology architecture of the transformer area.

[0061] In one embodiment of the present invention, the current distribution of the low-voltage distribution network follows Kirchhoff's current law. Under an ideal tree-like distribution topology with no leakage and good insulation, the current vector flowing into a certain line section theoretically approaches the sum of the current vectors at the ends of all branches flowing out of the line section. When there is insulation degradation or high-resistance leakage in the line section, the zero-sequence active current vector sum flowing into and out of the line section will have a significant deviation. The significant deviation can be used to characterize the return current component ionized to ground in the corresponding physical section.

[0062] The matrix construction module 30 retrieves the internally integrated physical topology archive records of the transformer area. Based on the tree structure, the transformer area is divided into 5 to 30 monitoring intervals. To avoid the calculation of false abrupt residuals due to packet loss in the underlying communication, the matrix construction module 30 performs data integrity verification before calculation. If the data of the edge child nodes corresponding to any monitoring interval in the current calculation period is not successfully received, the current calculation period is determined to be an invalid section with missing data. The residual data of all monitoring intervals in the entire transformer area in the current calculation period are discarded. This calculation period is not used as a valid snapshot column in the sliding time window to participate in the construction of the subsequent covariance matrix, thereby ensuring the structural integrity of the spatiotemporal dimension of the residual snapshot matrix and the executability of mathematical operations.

[0063] For monitoring intervals with complete data, the matrix construction module 30 extracts the zero-sequence active current phasor after phase compensation from the parent node at the beginning of the monitoring interval, and extracts the zero-sequence active current phasors after compensation from all direct child nodes at the end of the monitoring interval. The matrix construction module 30 performs vector subtraction on the above phasors to obtain the topological phasor residual. The calculation model for the topological phasor residual is as follows: ; In the formula, For the Taiwan region Each monitoring interval in frequency Topological phasor residuals; For monitoring interval First parent node In frequency The zero-sequence active current phasor after phase alignment compensation; For direct connection in the monitoring range The spatial index number of the direct child node at the end; To monitor the interval The set of spatial index labels of the next level direct child nodes directly connected to the first parent node; For set Inner The frequency of direct child nodes The zero-sequence active current phasor after phase alignment compensation; Spatial index labels for the physical monitoring intervals within the transformer area; For monitoring interval The spatial index number of the parent node at the beginning; Specific electrical frequencies for feature extraction.

[0064] S302, the matrix construction module 30 constructs the covariance matrix of the residual time series within a preset time window.

[0065] Leakage current has certain dynamic fluctuation characteristics. Considering that the sampling of the underlying hardware is affected by high-frequency white noise, the residual data on a single time section is difficult to stably reflect the physical characteristics of leakage current. By transforming the discrete node phasors on the time series into a covariance matrix, it is helpful to filter out independent and unrelated random noise and extract the synchronous correlation of residual fluctuations between different spatial nodes. Leakage current fluctuations that are of the same origin often exhibit high covariance characteristics between nodes at different levels with electrical connections.

[0066] The matrix construction module 30 sets a sliding time window containing multiple consecutive calculation cycles. As time progresses, the matrix construction module 30 combines the topological phasor residuals of each monitoring interval collected within the sliding time window according to the spatial and temporal dimensions to construct a residual snapshot matrix. The rows of the matrix correspond to different monitoring intervals, and the columns correspond to different time sections within the sliding window. After performing row mean centering on the residual snapshot matrix, the spatiotemporal covariance matrix of the corresponding frequency is calculated using the complex conjugate transpose rule. The residual snapshot matrix calculation model is as follows: ; In the formula, Frequency The multidimensional spatiotemporal covariance matrix; This represents the total number of discrete computation cycles contained within the sliding time window. Frequency The complex snapshot matrix of the topological phasor residuals after centering and mean-removing processing; This is the matrix conjugate transpose operator; Specific electrical frequencies for feature extraction.

[0067] S303, the matrix construction module 30 performs eigenvalue decomposition on the fundamental covariance matrix and extracts the largest eigenvalue and the principal eigenvector.

[0068] After constructing the multi-frequency covariance matrix, the matrix construction module 30 performs eigenvalue decomposition operations on the covariance matrix corresponding to the fundamental frequency band. For the eigenvalue decomposition of the multidimensional matrix, existing matrix algebra calculation methods are used.

[0069] After eigenvalue decomposition, the matrix construction module 30 obtains multiple sets of eigenvalues ​​and their corresponding eigenvectors. At the analytical level, the magnitude of the eigenvalues ​​of the covariance matrix characterizes the intensity of system state fluctuations to a certain extent. The eigenvalue with the largest value corresponds to the dominant fluctuation energy source. Therefore, the matrix construction module 30 extracts the eigenvalue with the largest value as the data basis for subsequent state discrimination.

[0070] Meanwhile, the matrix construction module 30 extracts the principal feature vector corresponding to the feature value with the largest value. The principal feature vector is a multi-dimensional column vector, and its number of dimensions is consistent with the total number of monitoring intervals divided in the transformer area. The value of each element in each dimension of the principal feature vector reflects the distribution of the dominant leakage energy in each physical interval of the entire transformer area, providing a reference mapping direction for subsequent spatial projection calculations.

[0071] See attached document Figure 4 In specific implementations, step S4 provided by the present invention may include the following steps: S401, the state determination module 40 performs orthogonal projection operation on the harmonic covariance matrix to the space spanned by the fundamental wave principal eigenvector, in order to calculate the projection overlap coefficient.

[0072] In the actual operation of low-voltage distribution network areas, if there are errors in the physical connection records of user meters, the system will often generate false topological phasor residuals during calculation. The false topological phasor residuals exhibit a purely mathematical linear proportional relationship at various electrical frequencies. This is because when a user's meter is incorrectly assigned, causing a linear error state, the source of the topological error is essentially the user's normal load current itself. Since the load current of ordinary users is mainly concentrated in the fundamental frequency band, its inherent low-order harmonic components are extremely small and will be submerged by various randomly distributed and spatially diffuse background harmonic noises throughout the low-voltage distribution network area.

[0073] This means that topology errors in the low harmonic frequency band do not generate wave sources that are synchronized with the fundamental residual and are physically concentrated. As a result, the spatial energy distribution of each low harmonic covariance matrix is ​​uncorrelated with the fundamental principal eigenvector, i.e., the spatial collinearity is extremely weak. Mathematically, this is manifested in the calculated projection overlap coefficient being at an extremely low level without exceeding the limit. Conversely, when real insulation degradation or high-resistance leakage occurs within the transformer area, the nonlinear breakdown of the dielectric at the fault point will excite specific low harmonics. The leakage current fluctuations caused by real insulation degradation or high-resistance leakage have high collinearity characteristics in the spatial distribution of the fundamental and the associated low harmonics.

[0074] To quantitatively assess the spatial collinearity of residual fluctuations at different frequencies, the state determination module 40 performs an orthogonal projection operation on the extracted low-order harmonic covariance matrix into the one-dimensional space spanned by the fundamental wave principal eigenvector. This projection operation can assess the concentration ratio of harmonic wave energy in the main direction of the fundamental wave. The calculation model for the projection overlap coefficient performed by the state determination module 40 is as follows: ; In the formula, Frequency The projection overlap coefficient; The principal eigenvectors extracted after performing eigenvalue decomposition on the fundamental spatiotemporal covariance matrix; It is the complex conjugate transpose of the fundamental principal eigenvector; Frequency The multidimensional spatiotemporal covariance matrix; This is the trace operation function for a matrix; Specific electrical frequencies for feature extraction; To prevent tolerance for extremely small positive numbers with a denominator of zero, the value of the extremely small positive number tolerance can be set according to the underlying floating-point precision of the computing system, for example, set to 10. -6 .

[0075] S402, the state determination module 40 distinguishes between nonlinear leakage state and linear error state based on the over-limit situation of the maximum eigenvalue and the projection overlap coefficient.

[0076] After calculating the projection overlap coefficient, the state determination module 40 constructs a dual-condition fake detection logic based on the nonlinear physical law of high-resistance leakage current-associated harmonics. The state determination module 40 internally presets the maximum eigenvalue threshold and the projection overlap threshold. The specific value of the maximum eigenvalue threshold is set according to the background leakage current noise level under normal fault-free conditions of the transformer area, and is configured to be 1.5 to 3 times the background noise level to intercept random noise fluctuations. The background leakage current noise level is obtained by statistically averaging or calculating the maximum value of the natural capacitive and resistive leakage current generated under normal conditions in each monitoring interval during the adaptive learning period of the system in the early stage of commissioning, such as the first 7 days of normal operation without faults. The specific value of the projection overlap threshold is set in combination with the data distribution characteristics under historical real leakage conditions. The value range of the projection overlap threshold is generally configured between 0.75 and 0.9 to define the strong and weak boundaries of spatial collinearity.

[0077] The state determination module 40 determines whether the largest eigenvalue of the fundamental covariance matrix is ​​greater than the largest eigenvalue threshold, and simultaneously determines whether there is a case where the calculated projection overlap coefficients of each specified low harmonic are greater than the projection overlap threshold. The state determination module 40 performs a logical AND operation between the judgment result of the largest eigenvalue and the judgment result of the projection overlap coefficient.

[0078] When the maximum eigenvalue exceeds the limit and the projection coincidence coefficient of at least one low-order harmonic exceeds the limit, the state determination module 40 determines that the current transformer area is in a nonlinear leakage state, indicating that a real physical leakage return current has been detected. If the maximum eigenvalue exceeds the limit but the projection coincidence coefficient of all low-order harmonics does not exceed the limit, the state determination module 40 determines that the current transformer area is in a linear error state, indicating that the residual detected by the system is mainly caused by the error in the physical topology file of the transformer area, rather than an actual electrical fault.

[0079] S403, the status determination module 40 triggers the corresponding early warning response action based on the determination result, and uses the magnitude of the main feature vector to compare and lock the fault range.

[0080] The status determination module 40 executes a classification response strategy based on the final identified status type. When the status is determined to be linear error, the status determination module 40 intercepts the leakage alarm signal, generates a transformer topology correction work order, and reports the transformer topology correction work order to the upper-level master station system through the communication network, prompting maintenance personnel to check the physical connection files of the corresponding branch box and user meter. When the status is determined to be nonlinear leakage, the status determination module 40 triggers a leakage alarm signal to issue a low-voltage switch trip command or prompt on-site investigation.

[0081] While executing the response, the state determination module 40 needs to determine the specific topological interval where the fault occurred. Since the covariance matrix originates from complex phasors, the eigenvector elements corresponding to the multidimensional spatiotemporal covariance matrix are in complex form. In order to accurately measure the energy weight of each interval and avoid the logical dead zone of size comparison, the state determination module 40 calculates the magnitude of each dimension element inside the fundamental main eigenvector. Since each dimension of the main eigenvector corresponds one-to-one with each pre-divided physical monitoring interval, the state determination module 40 compares the magnitude of each dimension element, extracts the dimension index with the largest magnitude, and maps the dimension index with the largest magnitude back to the preset physical topology of the transformer area, thereby locking the interval corresponding to the maximum magnitude as the actual physical location where leakage occurs.

[0082] To aid in understanding the technical solution of this invention, the following provides an example of a low-voltage distribution area leakage current backflow monitoring and early warning application based on edge computing.

[0083] The low-voltage distribution network area is set to have a tree-like distribution network. The physical environment of this area is configured with one edge master node and three edge child nodes. The system divides the first physical monitoring section and the second physical monitoring section according to the line connection relationship. The first end access point of the first physical monitoring section is the aforementioned edge master node, and the last branch nodes are the first edge child node and the second edge child node. The first end access point of the second physical monitoring section is the first edge child node, and the last branch node is the third edge child node. The system's calculation cycle is set to 60 seconds, the preset maximum feature value threshold is set to 450, and the projection overlap threshold is set to 0.75.

[0084] During the current calculation cycle, a resistive leakage fault occurs in the first physical monitoring interval, and the underlying communication network experiences a time delay deviation. The system then performs the following calculation process.

[0085] During the feature extraction stage, the first edge sub-node locally acquires the original zero-sequence voltage, which has an amplitude of 1.15 volts, satisfying the precondition of a latch-up threshold greater than 0.5 volts. The corresponding acquisition phase angle is 12.4 degrees. The first edge sub-node simultaneously acquires the original fundamental zero-sequence current, which has an amplitude of 78.6 mA and a phase angle of 43.1 degrees. The feature extraction module 10 performs orthogonal projection calculation, and the calculated amplitude of the extracted fundamental zero-sequence active current phasor is approximately 67.5 mA, with the phase angle maintained at 12.4 degrees. The system simultaneously extracts the phase angle of the fundamental positive-sequence voltage of each node. The phase angle at the edge master node is 0 degrees, and the phase angle at the first edge sub-node is -3.2 degrees.

[0086] During the phase compensation stage, the phase compensation module 20 calculates the positive sequence phase angle drift deviation between nodes. The phase angle difference between the edge master node and the first edge child node is positive 3.2 degrees. Based on this value, the system constructs a phase rotation compensation operator to perform dynamic phase compensation on the fundamental zero-sequence active current phasor sent by the first edge child node. After compensation, the amplitude of the fundamental zero-sequence active current phasor remains unchanged, and the phase angle reference is aligned to 15.6 degrees. Other nodes perform corresponding compensation according to the same logic.

[0087] During the matrix construction phase, the matrix construction module 30 performs vector subtraction on the phasors of each node after unifying the time base, and calculates that the residual amplitude of the fundamental wave topological phasor in the first physical monitoring interval is about 42.3 mA. The system combines the data of 10 consecutive calculation cycles to construct the covariance matrix of the residual time series. After performing eigenvalue decomposition on the fundamental wave covariance matrix, the matrix construction module 30 extracts the eigenvalue with the largest value, which is 1845.6. The matrix construction module 30 simultaneously extracts the principal eigenvector corresponding to this eigenvalue.

[0088] During the state determination phase, the state determination module 40 retrieves the covariance matrix of the third harmonic and performs an orthogonal projection operation on it onto the space where the fundamental main eigenvector is located. The calculated projection overlap coefficient is 0.865. The state determination module 40 compares the preset numerical boundary. The eigenvalue 1845.6 is greater than 450, and the projection overlap coefficient 0.865 is greater than 0.75. According to the determination rules, the state determination module 40 identifies that the current leakage current state of the platform area is nonlinear. The state determination module 40 further calculates the magnitude of each dimension of the main eigenvector. The magnitude of the element corresponding to the first dimension is about 0.976, and the magnitude of the element corresponding to the second dimension is about 0.041. The comparison determines that the magnitude of the first dimension is the largest. The state determination module 40 locks the first physical monitoring interval as the actual leakage current location and triggers the corresponding action response.

[0089] Based on the aforementioned application examples, the test data for various electrical conditions output by the physical experiment platform are shown in the table below.

[0090] Table 1: Test Data of Leakage Current Characteristic Parameters in Transformer Areas According to Table 1 and Figure 5 It can be seen that the calculation parameters of the system under various electrical conditions exhibit distinct characteristics. In test sequence 1, the zero-sequence voltage amplitude of the first edge sub-node is lower than the set blocking threshold. When the amplitude is low, the system limits the amplification effect of the random flipping of the local voltage phase angle on the residual calculation and maintains the stability of the basic data calculation. In test sequence 2, the system is subjected to high-frequency capacitive interference, and the zero-sequence voltage amplitude increases, but the maximum eigenvalue of the fundamental covariance matrix remains within the preset threshold range. The system eliminates the interference of capacitive reactive current components through zero-sequence orthogonal projection calculation.

[0091] In tests 3 and 4, the transformer area access topology file was incorrect. In tests 5 and 6, the transformer area accessed a real high-resistivity leakage condition. Both test conditions caused the maximum eigenvalue of the fundamental covariance matrix to exceed the limit, with the specific eigenvalue generally rising above 1700. Analyzing the projection overlap coefficient parameter, the source of the residuals in tests 3 and 4 was the fundamental current of the normal load of ordinary users. This type of current did not produce a concentrated energy distribution in the 3rd harmonic frequency band, and the calculated 3rd harmonic projection overlap coefficient was between 0.12 and 0.15, failing to reach the anti-counterfeiting threshold. Based on this, the status determination module 40 identified it as a linear error state and intercepted the warning signal. Tests 5 and 6 reflected real line insulation degradation and nonlinear physical breakdown. The 3rd harmonic fluctuations excited by physical breakdown were consistent with the fundamental frequency in spatial distribution, and the corresponding projection overlap coefficient rose to the range of 0.84 to 0.89. Based on this, the status determination module 40 identified a nonlinear leakage state.

[0092] The residual fluctuation mean data in Table 1 reflects the operational effect of the phase alignment compensation step. Before the compensation operation was performed, the latency of the underlying communication network caused the time dimension misalignment of the phasors reported by multiple nodes, resulting in redundant increments in the spatial differential calculation. After the phase rotation operator compensation was performed, the residual fluctuation mean of each test interval decreased as a whole. The positive sequence phase angle drift deviation was aligned with the time reference of the multi-node data section, causing the spatial topology residual data to converge towards the real physical noise floor.

[0093] 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, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for monitoring and early warning of leakage current return in low-voltage distribution areas based on edge computing, characterized in that, The method includes the following steps: Extract the zero-sequence active current phasor of the fundamental and lower harmonics and the phase angle of the fundamental positive-sequence voltage at the edge sub-nodes, and upload the set of characteristic parameters to the edge master node. Obtain the reference positive sequence voltage phase angle of the edge master node, calculate the positive sequence phase angle drift deviation, and construct a rotation operator to perform phase alignment compensation on the zero sequence active current phasor. Based on the physical topology architecture, the topological phasor residuals between adjacent level nodes are calculated, and the covariance matrix is ​​constructed within the time window. The fundamental covariance matrix is ​​then decomposed into eigenvalues ​​to extract the maximum eigenvalue and the principal eigenvector. The harmonic covariance matrix is ​​orthogonally projected onto the space spanned by the principal eigenvector to calculate the projection overlap coefficient. Based on the over-limit situation of the maximum eigenvalue and the projection overlap coefficient, the nonlinear leakage state and the linear error state are distinguished and an early warning is triggered. The steps for distinguishing between nonlinear leakage current state and linear error state based on the out-of-limit condition of the maximum eigenvalue and the projection coincidence coefficient include: Determine whether the largest eigenvalue of the fundamental covariance matrix is ​​greater than the largest eigenvalue threshold, and simultaneously determine whether any of the calculated projection overlap coefficients of each specified lower harmonic are greater than the projection overlap threshold. When the maximum eigenvalue exceeds the limit and at least one low-order harmonic projection overlap coefficient exceeds the limit, the current transformer area is determined to be in a nonlinear leakage state. When the maximum eigenvalue exceeds the limit and the projection coincidence coefficients of all lower harmonics do not exceed the limit, the current transformer area is determined to be in a linear error state.

2. The low-voltage distribution area leakage current monitoring and early warning method based on edge computing according to claim 1, characterized in that, The step of extracting the fundamental and lower harmonic zero-sequence active current phasors at the edge sub-nodes includes: High-frequency sampling and discrete Fourier transform are performed on the three-phase voltage and three-phase current locally at the edge sub-nodes to obtain the original zero-sequence current phasor and the original zero-sequence voltage phasor at a specific frequency. Preset leakage current calculation and lockout threshold; When the amplitude of the original zero-sequence voltage is less than the leakage current calculation latch-up threshold, the unified time reference reference phase of the entire distribution area is used as the reference direction for orthogonal projection, and orthogonal projection calculation is performed on the original zero-sequence current phasor to obtain the zero-sequence active current phasor. When the amplitude of the original zero-sequence voltage is greater than or equal to the leakage current calculation latch-up threshold, the original zero-sequence current phasor is orthogonally projected onto the direction of the original zero-sequence voltage phasor of the corresponding frequency, and an anti-overflow minimum positive threshold is introduced into the denominator of the projection calculation to prevent overflow after division by zero, and the zero-sequence active current phasor after calculating the complex real part is obtained. The specific frequency is set as the fundamental frequency and lower harmonic frequency band, and the specific frequency is the fundamental frequency, the third harmonic and the fifth harmonic.

3. The low-voltage distribution area leakage current backflow monitoring and early warning method based on edge computing according to claim 1, characterized in that, The steps of obtaining the reference positive-sequence voltage phase angle of the edge master node, calculating the positive-sequence phase angle drift deviation, and constructing a rotating operator to perform phase alignment compensation on the zero-sequence active current phasor include: Based on the beacon frame mechanism of the underlying communication network, the fundamental positive sequence voltage phase angle of the edge master node at the same historical sampling time section as the edge child node is extracted as a physical reference quantity. The difference between the physical reference value and the fundamental positive sequence voltage phase angle of the corresponding edge sub-node is calculated to obtain the positive sequence phase angle drift deviation of the edge sub-node relative to the edge master node. In the complex plane, a corresponding phase rotation compensation operator is constructed using the positive sequence phase angle drift deviation and the proportional amplification factor of the corresponding harmonic frequency. The phase rotation compensation operator is applied to the original zero-sequence active current phasor of the corresponding frequency to obtain the zero-sequence active current phasor after phase alignment compensation.

4. The low-voltage distribution area leakage current backflow monitoring and early warning method based on edge computing according to claim 1, characterized in that, The steps for calculating the topological phasor residuals between adjacent level nodes based on the physical topology architecture include: The station area is divided into multiple monitoring intervals, and data integrity verification is performed before calculation. If the data of the edge sub-nodes corresponding to any monitoring interval in the current calculation period is not fully received, the current calculation period is determined to be an invalid section with missing data, and the residual data of all monitoring intervals in the entire station area in the current calculation period are discarded. For a monitoring interval with complete data, the zero-sequence active current phasor of the parent node at the beginning of the monitoring interval after phase alignment compensation is extracted, and the sum of the zero-sequence active current phasors of all direct child nodes at the end of the monitoring interval after phase alignment compensation is extracted. Perform a vector subtraction operation on the sum of the phasor corresponding to the first parent node and the phasor corresponding to the direct child node to obtain the topological phasor residual of the corresponding monitoring interval at the corresponding frequency.

5. The low-voltage distribution area leakage current monitoring and early warning method based on edge computing according to claim 4, characterized in that, The steps for constructing the covariance matrix within the time window include: A sliding time window containing multiple consecutive calculation cycles is set. The topological phasor residuals of each monitoring interval collected within the sliding time window are combined according to the spatial and temporal dimensions to construct a residual snapshot matrix. Row mean centering is then performed to obtain the complex snapshot matrix of topological phasor residuals after centering and mean removal. The complex snapshot matrix of the topological phasor residual is multiplied with the complex conjugate transpose matrix, and then averaged in combination with the total number of discrete calculation cycles within the sliding time window to generate a multidimensional spatiotemporal covariance matrix of the corresponding frequency.

6. The low-voltage distribution area leakage current backflow monitoring and early warning method based on edge computing according to claim 1, characterized in that, The step of performing an orthogonal projection operation on the space spanned by the principal eigenvectors to calculate the projection coincidence coefficient includes: The complex conjugate transpose of the principal eigenvector, the harmonic covariance matrix, and the principal eigenvector are multiplied together and used as the numerator of the projected energy. Calculate the trace of the harmonic covariance matrix and add a minimal positive tolerance to prevent the denominator from being zero, as the total energy denominator term; Divide the numerator of the projected energy by the denominator of the total energy to obtain the projection overlap coefficient at the corresponding frequency.

7. The low-voltage distribution area leakage current backflow monitoring and early warning method based on edge computing according to claim 1, characterized in that, The steps for triggering the warning include: When a linear error state is determined, the leakage alarm signal is intercepted, and a transformer topology correction work order is generated instead. The transformer topology correction work order is then reported to the upper-level master station system through the communication network. When a non-linear leakage condition is detected, a leakage alarm signal is triggered to issue a low-voltage switch trip command or prompt for on-site troubleshooting.

8. The low-voltage distribution area leakage current backflow monitoring and early warning method based on edge computing according to claim 1, characterized in that, The maximum eigenvalue threshold is set based on the background leakage current noise level obtained during the adaptive learning period of the system in the early stage of commissioning, and is configured to be 1.5 to 3 times the background leakage current noise level. The projection overlap threshold is configured to range from 0.75 to 0.

9.

9. The low-voltage distribution area leakage current backflow monitoring and early warning method based on edge computing according to claim 1, characterized in that, The method further includes a step of locking the fault region based on the main feature vector, the step including: Calculate the modulus of each element in each dimension of the main feature vector; By comparing the magnitudes of the elements in each dimension, the index of the dimension with the largest magnitude is extracted, and the index of the dimension with the largest magnitude is mapped back to the preset physical topology of the transformer area, thereby locking the interval corresponding to the largest magnitude as the actual physical location where leakage occurs.

Images