Power equipment real-time fault positioning method based on edge calculation

Abstract

The invention relates to the technical field of power system relay protection and power distribution network automation, and discloses a power equipment real-time fault positioning method based on edge calculation, which comprises the following steps: constructing an edge calculation logic topological graph, presetting a positive sequence impedance module value, and monitoring a zero sequence current signal in real time. And when the sudden change of energy is detected, the edge node is switched to the dominant node, and the spectrum fingerprint seed frequency is extracted by using the VMD unconstrained variational model and is sent to the adjacent node. And the adjacent nodes forcibly decompose local signals in the same frequency band based on the frequency by using a CC-VMD constraint optimization model to obtain frequency domain aligned characteristic mode components. And each node calculates a normalized permutation entropy and exchanges data, and calculates a unit electrical distance entropy dissipation rate in combination with a positive sequence impedance module value, thereby determining a fault section. According to the method, the problem of feature alignment under asynchronous sampling is solved by using a frequency domain locking mechanism, and the recognition sensitivity and the positioning accuracy of the high-resistance grounding fault are improved through the entropy dissipation rate.

Description

Technical Field

[0001] This invention relates to the field of power system relay protection and distribution network automation technology, specifically to a real-time fault location method for power equipment based on edge computing. Background Technology

[0002] With the high penetration rate of distributed generation in distribution networks, traditional distribution networks have gradually transformed into active distribution networks with multiple power sources and bidirectional power flow. To ensure power supply reliability, quickly and accurately locating faulty sections has become a core function of distribution automation systems. However, existing distribution network fault location technologies still face many technical challenges in practical applications.

[0003] While current dual-end traveling wave positioning technology has high theoretical accuracy, it has extremely high requirements for time synchronization, and usually relies on GPS or BeiDou satellite timing devices with nanosecond-level accuracy to ensure strict alignment of data at both ends.

[0004] In distribution networks with complex branches and numerous nodes, configuring high-precision timing hardware for each monitoring point not only significantly increases construction and maintenance costs, but also makes it difficult to keep the sampling clocks of distributed nodes synchronized when satellite signals are blocked or edge computing resources are limited.

[0005] This asynchronous sampling causes time-domain deviations in phase and amplitude in multi-point monitoring data, making it difficult to directly apply waveform comparison-based collaborative analysis methods and affecting the effectiveness of fault location.

[0006] Traditional impedance-based fault location techniques rely primarily on changes in the amplitude of power frequency voltage and current to calculate fault distance. However, in actual operation of distribution networks, faults that are grounded through high transition resistance (i.e., high-resistance grounding) frequently occur, at which point the fault point is accompanied by intermittent arcing or dielectric breakdown.

[0007] Because the zero-sequence current amplitude generated by high-resistance grounding faults is small, and due to the influence of the nonlinear characteristics of the transition resistance, the fault characteristics are not obvious at the amplitude level. This makes it difficult for the traditional impedance method, which relies on amplitude criteria, to sensitively identify such faults, and it is prone to failure to operate or misjudgment, failing to meet the detection requirements for weak fault characteristics.

[0008] In addition, the integration of distributed power sources makes the power flow of the distribution network more complex and variable. The presence of boosting current changes the distribution pattern of fault current, which in turn disturbs the linear relationship between measured impedance and fault distance.

[0009] Meanwhile, the significant differences in line length and electrical parameters among different feeder sections in the distribution network lead to inconsistent signal attenuation characteristics during transmission across different sections. Existing technologies for network-wide fault assessment often lack a standardized metric that can eliminate line parameter differences and is insensitive to bidirectional power flow fluctuations, making it difficult to achieve robust fault section location in active distribution network environments with severe source-load fluctuations. Summary of the Invention

[0010] To address the shortcomings of existing technologies, this invention provides a real-time fault location method for power equipment based on edge computing. This method solves the problems in existing technologies, such as the difficulty in data alignment under asynchronous sampling due to reliance on high-cost clock synchronization devices, the low sensitivity of amplitude-based criteria for identifying high-resistance grounding faults, and the lack of standardized metrics that can eliminate line parameter differences and resist source-load fluctuation interference.

[0011] This invention provides a real-time fault location method for power equipment based on edge computing, comprising the following steps: S1. Construct an edge computing logic topology graph, and pre-set the positive sequence impedance magnitude in the edge nodes of the edge computing logic topology graph. The edge nodes monitor the zero sequence current signal in real time. S2. When any edge node detects an energy change in the zero-sequence current signal, the corresponding edge node is switched to the dominant node mode. The zero-sequence current signal is decomposed using the VMD unconstrained variational model, the spectral fingerprint seed frequency is extracted and sent to the adjacent nodes in the edge computing logic topology graph. S3. In response to the received spectral fingerprint seed frequency, the neighboring nodes use the CC-VMD constraint optimization model to force the zero-sequence current signal monitored by the neighboring nodes to be decomposed in the same frequency band, and obtain the characteristic mode components aligned with the frequency domain of the edge nodes under the dominant node mode. S4. Edge nodes and adjacent nodes calculate the obtained feature mode components to obtain the normalized permutation entropy. S5. The normalized permutation entropy is exchanged between the edge node and the adjacent node. The entropy dissipation rate per unit electrical distance is calculated by combining the positive sequence impedance magnitude and the entropy dissipation rate per unit electrical distance. The fault section is determined based on the entropy dissipation rate per unit electrical distance.

[0012] The phase of constructing the edge computing logical topology and presetting parameters specifically includes: In the local database of each edge node, an edge computing logical topology diagram describing the connection relationship between the set of edge nodes and the set of transmission lines is established; In the local database, the electrical parameters of the transmission lines between the edge node and its directly adjacent edge nodes are pre-written; the positive sequence impedance magnitude of the transmission line is selected as the key physical parameter. This positive sequence impedance magnitude is calculated based on the unit length impedance parameter of the transmission line and the actual physical length of the line, and is used as the physical benchmark for subsequent calculation of the entropy dissipation rate per unit electrical distance.

[0013] Regarding the triggering and feature extraction of the dominant node, the edge node continuously monitors the energy of the zero-sequence current signal within the buffer. When the detected energy mutation value of the zero-sequence current signal exceeds the quiescent threshold, an interrupt is triggered, switching the operating state of the edge node from monitoring mode to dominant node mode; this quiescent threshold is preset based on the background noise level during normal operation of the distribution network.

[0014] In the dominant node mode, edge nodes use the VMD unconstrained variational model to establish a variational constraint problem, that is, to find the minimum sum of bandwidth of each modal component decomposed from the zero-sequence current signal and the sum of each modal component equals the zero-sequence current signal. In this process, a quadratic penalty factor and Lagrange multipliers are introduced to construct an augmented Lagrange variational objective function; the alternating direction multiplier method is used to iteratively solve the augmented Lagrange variational objective function to obtain a set of modal components and their corresponding center frequencies.

[0015] Subsequently, the energy amplitude of each modal component in the set of modal components obtained by iterative solution is calculated, the modal component with the largest energy amplitude is selected, and the center frequency corresponding to the modal component is determined as the spectral fingerprint seed frequency; the spectral fingerprint seed frequency is encapsulated into an induced request data packet and sent to the directly connected adjacent nodes in the edge computing logical topology graph through the communication network.

[0016] Regarding the collaborative processing of adjacent nodes, adjacent nodes enter the master-slave collaborative induced decomposition process, and initiate the center frequency constrained variational mode decomposition to decompose the zero-sequence current signal. Specifically, the CC-VMD constrained optimization model is used to add a frequency penalty term to the variational model to construct a frequency domain locking constraint objective function.

[0017] The frequency domain locking constraint objective function is used to constrain the frequency domain position of a specific mode component by introducing a quadratic penalty term, based on the premise of finding the minimum sum of bandwidths of all mode components and the sum of all mode components equal to the zero-sequence current signal.

[0018] The frequency-domain locking constraint objective function includes frequency locking weight coefficients, spectral fingerprint seed frequencies sent by edge nodes under the dominant node mode, and indicator functions for selecting target modal components. In solving the frequency-domain locking constraint objective function, the center frequency of the modal components is iteratively calculated using a center frequency iterative update formula. In the center frequency iterative update formula, the center frequency of the modal component is updated based on the power spectrum centroid of the modal component calculated in the current iteration step, and the spectral fingerprint seed frequency is introduced by the frequency locking weight coefficient and the indicator function for weighted balancing. Through iterative updates, the center frequencies of the target modal components output by adjacent nodes converge to the spectral fingerprint seed frequency, thereby obtaining frequency-domain aligned feature modal components.

[0019] When calculating the normalized permutation entropy, edge nodes and adjacent nodes use a signal complexity quantification model and employ the time delay embedding theorem to reconstruct the phase space of the feature mode components, resulting in multiple reconstructed vectors. The frequency of occurrence of symbol permutation patterns in all reconstructed vectors is counted, and the probability of occurrence of each symbol permutation pattern is calculated. Based on the occurrence probability, the normalized permutation entropy is calculated using the normalized permutation entropy calculation formula; this formula uses the factorial of the embedding dimension and the occurrence probability of the symbol permutation pattern for logarithmic operation.

[0020] When determining the faulty section, depending on the strategy adopted, the adjacent nodes encapsulate the calculated normalized permutation entropy into the response data packet and send it back to the edge node in the dominant node mode, or the edge node in the dominant node mode exchanges the normalized permutation entropy calculated by each of the adjacent nodes; then, the preset positive sequence impedance magnitude value is retrieved, and the entropy dissipation rate per unit electrical distance is calculated using the formula for the entropy dissipation rate per unit electrical distance.

[0021] The formula for the unit electrical distance entropy dissipation rate is calculated based on the ratio of the absolute value of the difference between the normalized permutation entropy between two edge nodes to the positive sequence impedance magnitude.

[0022] Finally, the fault section determination criteria are applied, and the calculated unit electrical distance entropy dissipation rate is compared with the natural attenuation benchmark threshold. If the unit electrical distance entropy dissipation rate is greater than the natural attenuation benchmark threshold, it is determined that there is nonlinear energy leakage in the line between the two edge nodes, and the line is identified as a fault section. The natural attenuation benchmark threshold is pre-set based on statistical analysis of historical data of the distribution network under normal operating conditions and external disturbances.

[0023] This invention provides a real-time fault location method for power equipment based on edge computing. It has the following advantages: 1. This invention extracts the seed frequency of the spectral fingerprint by the dominant node and uses a CC-VMD constraint optimization model to force adjacent nodes to perform signal decomposition in the same frequency band. This frequency domain locking mechanism enables feature alignment of distributed nodes under asynchronous sampling conditions, avoiding the dependence of traditional traveling wave positioning technology on nanosecond-level high-precision clock synchronization devices. While reducing the hardware configuration cost of edge computing nodes, it solves the waveform phase deviation problem caused by asynchronous sampling clocks and improves the effectiveness of multi-point collaborative analysis.

[0024] 2. This invention changes the traditional impedance method's reliance on voltage and current amplitude changes for judgment, instead employing normalized permutation entropy to quantify the nonlinear complexity of fault signals. By calculating the entropy dissipation rate per unit electrical distance during signal transmission, this method can effectively identify weak arc discharge and dielectric breakdown characteristics caused by high-impedance grounding faults. Compared to directly comparing amplitudes, the criterion based on signal complexity is more sensitive to waveform distortion, reducing missed detections in high-impedance fault scenarios.

[0025] 3. This invention introduces the positive-sequence impedance magnitude as a physical benchmark and constructs the entropy dissipation rate per unit electrical distance as a fault criterion. This calculation process eliminates the decision bias caused by differences in line lengths and parameters in the distribution network, achieving a standardized measurement of fault characteristics. Furthermore, since the entropy dissipation rate is based on the nonlinear rate of change of the signal waveform structure, rather than simply the power flow direction or amplitude superposition, this method is less affected by power flow fluctuations caused by distributed generation, making it suitable for active distribution network environments with severe source-load fluctuations and exhibiting strong adaptability. Attached Figure Description

[0026] Figure 1 This is a flowchart of a real-time fault location method for power equipment based on edge computing according to the present invention. Figure 2 This is a flowchart of the adjacent node frequency domain locking cooperative decomposition process of the present invention; Figure 3 This is a flowchart of the signal complexity quantification of the present invention. Detailed Implementation

[0027] 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.

[0028] Please see the appendix Figure 1 This invention provides a real-time fault location method for power equipment based on edge computing, comprising the following steps: Step S1: Edge computing logical topology construction and parameter initialization. The system constructs a logical topology diagram in the distribution network edge computing system. .in, Represents a set of edge computing nodes. This represents a set of transmission lines. The system pre-sets the positive-sequence impedance magnitudes between each edge node and its adjacent nodes in the local database of each edge node. This parameter serves as the physical benchmark for subsequent dissipation rate calculations. Simultaneously, each edge node monitors the zero-sequence current signal of the line in real time via current transformers. .

[0029] Next, proceed to step S2: Dominant node triggering and VMD unconstrained variational mode extraction. When any edge node... When a sudden change in signal energy exceeding the quiescent threshold is detected, the node immediately switches to dominant node mode. Edge node. The processor executes the VMD unconstrained variational model to decompose the locally acquired signal and extract the intrinsic mode functions. During this process, the augmented Lagrange variational objective function is constructed as follows: ; In the formula: This represents the value of the augmented Lagrange function; The decomposition yields the first... The time-domain components of each intrinsic mode function (IMF); Indicates the first The central angular frequency of each modal component; This represents the original zero-sequence current signal input. Represent the Lagrange multiplier function; Represents the imaginary unit; Represents pi; Represents the Dirac distribution function; This represents the convolution operator; Represents the L2 norm; Represents the time partial derivative; This indicates the inner product operation.

[0030] After solving the above function, the nodes The center frequency of the mode with the highest energy was selected as the seed frequency of the spectral fingerprint. The frequency is then encapsulated into an inducement request data packet and sent to neighboring nodes in the topology via the communication network.

[0031] Step S3: Constraint decomposition of neighboring nodes based on frequency-domain locked cooperative mode. Neighboring nodes Received Then, the master-slave collaborative induced decomposition process begins, initiating center frequency constrained variational mode decomposition (CC-VMD). Adjacent nodes utilize the CC-VMD constraint optimization model to force local signals to be decomposed within the same frequency band. During this process, the frequency domain locking constraint objective function is constructed as follows: ; In the formula: This represents the objective function value after applying frequency domain locking constraints; This represents the frequency-locked weighting coefficient; This indicates the frequency of the spectral fingerprint seed sent by the dominant node; Indicates the indicator function, when the modal index Equal to the target modal level index The function value is 1 when the condition is met, and 0 otherwise.

[0032] In solving this constraint model, an iterative update formula for the center frequency is used to ensure that the frequency converges to the locked value. ; In the formula: Indicates the first The result of the iteration calculation is the first The center frequency of each mode; Indicates the first In the nth iteration Fourier transform results of each modal component; It represents the frequency differential increment.

[0033] Through this step, the node Get and Node The frequency-domain strictly aligned characteristic mode components eliminate frequency deviations caused by asynchronous sampling.

[0034] Step S4: Signal Complexity Measurement. Each node calculates the signal complexity of its characteristic modes using a signal complexity measurement model. First, phase space reconstruction is performed, and the statistical symbol arrangement pattern is determined. The probability distribution is used to calculate the entropy value using the normalized permutation entropy formula: ; In the formula: This represents the normalized permutation entropy value, and its range is... ; The embedding dimension determines the dimension of the phase space; This represents the factorial operation; Indicates the first The probability of occurrence of a certain symbol arrangement pattern; Indicates the symbol arrangement pattern; Represents the natural logarithm operation.

[0035] Step S5: Fault section determination based on entropy dissipation rate per unit electrical distance. The entropy values ​​calculated between nodes are exchanged, and the entropy dissipation rate is calculated according to the formula for entropy dissipation rate per unit electrical distance. : ; In the formula: Represents edge nodes With edge nodes Entropy dissipation rate per unit electrical distance between them; Represents a node The normalized permutation entropy value; Represents a node The normalized permutation entropy value; This represents the positive sequence impedance magnitude of the transmission line connecting the two nodes.

[0036] System application fault section determination criteria: if the calculated If the value exceeds the system's preset natural attenuation threshold, then the node is considered... With nodes The line between them exhibits nonlinear energy leakage, identifying the faulty section as such.

[0037] Step S1 involves constructing the edge computing logical topology and initializing parameters. This step mainly provides the physical model foundation and data source for subsequent collaborative computing, and can be implemented through the following sub-steps S101 to S103: Step S101: Establish the logical topology diagram of the distribution network edge computing system.

[0038] Edge computing units are deployed at various physical nodes of the distribution network. These physical nodes include, but are not limited to, pole-mounted switch controllers (FTUs), distribution room monitoring terminals (DTUs), and smart meters with edge computing capabilities. Each edge computing unit is equipped with a processor and memory, enabling it to perform distributed computing tasks. The system maps these edge computing units into a logical topology diagram. The vertices in the middle.

[0039] in, This represents a set of edge computing nodes, covering all monitoring points on the power distribution network feeders; This represents a set of power transmission lines, corresponding to the conductors or cables connecting each physical node.

[0040] The logical topology graph is constructed by storing it in the local database of each edge computing node in the form of an adjacency list or adjacency matrix. Unlike the traditional centralized master station that maintains the entire network topology, in this embodiment, each edge node only needs to maintain a list of neighboring nodes within its one-hop range.

[0041] This distributed storage method reduces the dependence on the central server, enabling edge nodes to still perform collaborative operations based on local topology information even when the communication network is partially damaged.

[0042] Step S102: Preset line parameters and positive sequence impedance magnitude.

[0043] In the local database of each edge node, the electrical parameters of the transmission line between that node and its direct neighboring nodes are pre-written. In this embodiment, the key physical parameter selected is the positive sequence impedance magnitude. , Represents a node With nodes The positive sequence impedance magnitude of the transmission line between them, in ohms (Ω). ).

[0044] This parameter This is calculated based on the impedance parameter per unit length of the transmission line (determined by the conductor type) and the actual physical length of the line. Positive-sequence impedance is chosen over zero-sequence impedance as the benchmark because positive-sequence parameters are less affected by environmental factors, grounding methods, and operating conditions, and can more stably reflect the electrical distance between nodes. In subsequent calculations... It will be used as the physical denominator to measure signal transmission attenuation and computational entropy dissipation rate, and will be used to eliminate the influence of different line lengths on the rate of change of signal complexity, thereby achieving standardized distance normalization processing.

[0045] Step S103: Monitor the zero-sequence current signal of the line in real time.

[0046] Each edge computing node calculates and synthesizes the real-time zero-sequence current signal of the line by using zero-sequence current transformers installed on the line or by using three-phase current data collected by three-phase current transformers. Current signal For time A changing analog quantity or a digital sequence after analog-to-digital conversion.

[0047] To ensure the capture of high-frequency transient fault characteristics, the analog front-end (AFE) of the edge computing node is equipped with an anti-aliasing filter, and the sampling frequency must be set to satisfy the Nyquist sampling theorem, typically set to 128 times or higher than the power frequency. The acquired signal... It is stored in the circular buffer of the edge node.

[0048] This circular buffer employs a first-in, first-out (FIFO) overlay strategy, always retaining waveform data within the most recent time window. When the triggering conditions in subsequent steps are met, the processor directly locks and extracts time-domain data segments before and after the fault from this buffer for subsequent variational mode decomposition operations. This mechanism ensures that the system can acquire the complete transient waveform without delay at the moment a fault occurs, without waiting for data to be uploaded to the cloud.

[0049] Step S2, which involves the leading node triggering and VMD unconstrained variational mode extraction, is a crucial starting point for achieving asynchronous coordination. By extracting characteristic frequency references from the fault-causing node, a unified reference is provided for subsequent frequency domain locking across the entire network. This can be achieved through the following sub-steps S201 to S204: Step S201: Fault Trigger Detection and Operating Mode Switching.

[0050] Edge computing nodes continuously monitor the zero-sequence current signal within the buffer. Energy monitoring is performed. The processor uses a sliding window algorithm to calculate the instantaneous energy or root mean square value of the signal and to calculate the energy difference between the current window and historical windows. The system has a preset silence threshold, which is set according to the background noise level of the distribution network during normal operation, aiming to shield against normal load fluctuation interference.

[0051] When any edge node When the detected energy surge of the signal exceeds the quiescent threshold, it indicates that a disturbance event has occurred near the node. At this time, the node... The processor immediately triggers an interrupt, switching its operating state from monitor mode to master node mode. Master node mode means that the node will act as the initiator of this collaborative computation, responsible for defining the reference frequency of the fault characteristics.

[0052] Step S202: Construct an unconstrained variational model of VMD.

[0053] After entering the dominant node mode, the node The internal digital signal processing unit is invoked to perform variational mode decomposition (VMD) on the transient zero-sequence current signal locked in the buffer.

[0054] In order to decompose the original signal into several intrinsic mode functions (IMFs) with specific center frequencies and finite bandwidths, the system first establishes a variational constraint problem, that is, to find the minimum sum of bandwidths of each mode component, while the sum of each mode component is equal to the original signal.

[0055] To solve the above variational problem, a quadratic penalty factor and Lagrange multipliers are introduced to transform the constrained variational problem into an unconstrained problem, thereby constructing the augmented Lagrange variational objective function: ; In the formula: This represents the value of the augmented Lagrange function, used to evaluate the quality of the decomposition effect; The decomposition yields the first... The time-domain components of each intrinsic mode function (IMF); Indicates the first The central angular frequency of each modal component; This represents the original zero-sequence current signal input. This represents the Lagrange multiplier function, used to strictly enforce the reconstruction constraints; Represents the imaginary unit; Represents pi; Represents the Dirac distribution function; This represents the convolution operator; Represents the L2 norm; Represents the time partial derivative; This indicates the inner product operation.

[0056] Step S203: Solve for the modes and extract the seed frequency of the spectral fingerprint.

[0057] The processor uses the Alternating Direction Multiplier Method (ADMM) to iteratively solve the above augmented Lagrange variational objective function, continuously updating... and This continues until the convergence condition is met. After convergence, a set of eigenmode functions is obtained. and its corresponding center frequency .

[0058] Since distribution network fault signals typically contain power frequency components, harmonic components, and transient high-frequency components, in order to pinpoint the frequency components that best reflect the fault characteristics, nodes... Calculate the energy amplitude of each decomposed mode. The system selects the mode component with the largest energy amplitude, and the center frequency corresponding to this mode component is determined as the seed frequency of the spectral fingerprint. This frequency It represents the main energy concentration point of the current fault signal in the frequency domain, has extremely strong fingerprint characteristics, and can serve as a physical anchor point for cross-node collaboration.

[0059] Step S204: Encapsulate and send the inducement request data packet.

[0060] node Extracted spectral fingerprint seed frequency The data payload written into the communication protocol generates an inducement request packet. The header of this packet contains the dominant node's information. A unique identifier (ID) and collaboration request instructions.

[0061] Subsequently, the node Query the logical topology of local storage It identifies all directly connected neighboring nodes. Nodes can connect via fiber optic Ethernet or wireless communication modules. The induced request packet is sent unicast or multicast to these neighboring nodes. This process does not transmit large waveform data, but only a floating-point frequency value, reducing communication bandwidth usage and ensuring millisecond-level response of the coordination mechanism.

[0062] Please see the appendix Figure 2 Step S3, the constraint decomposition of adjacent nodes based on the frequency domain locking cooperative mode, solves the problem of feature frequency alignment of distributed nodes under asynchronous sampling conditions by introducing an external frequency fingerprint as a strong constraint. Specifically, it can be implemented through the following sub-steps S301 to S303: Step S301: Receive the induction signal and start the collaborative process.

[0063] In a logical topology network, with the dominant node Directly adjacent edge nodes Receive inducement request packets through the communication interface. Node The communication processing unit unpacks and parses the data packets, extracting the spectral fingerprint seed frequency determined by the dominant node. .

[0064] At this time, node Upon detecting a high-priority active localization task in the network, the current routine inspection or low-power mode is immediately aborted, and a master-slave collaborative induced decomposition process is initiated. This process aims to enable nodes to... The local signal decomposition process is no longer an unconstrained adaptive decomposition, but rather... Controlled decomposition under the guidance of [the relevant authority / organization].

[0065] Step S302: Construct a CC-VMD constrained optimization model.

[0066] node The zero-sequence current signal from the local cache is read, and the Center Frequency Constrained Variational Mode Decomposition (CC-VMD) algorithm is started. Unlike the traditional VMD algorithm, which only aims to minimize the signal reconstruction error, the CC-VMD algorithm adds a frequency penalty term to the variational model to force the center frequency of the specified mode to converge to the spectral fingerprint seed frequency.

[0067] node The processor constructs a frequency-domain locking constraint objective function, which, while ensuring signal sparsity and fidelity, constrains the frequency domain position of specific modes by introducing a quadratic penalty term: ; In the formula: This represents the objective function value after applying frequency domain locking constraints. The smaller the value, the better the decomposition result conforms to the physical constraints. This represents the frequency locking weighting coefficient. The larger the coefficient value, the greater the penalty for frequency deviation and the more obvious the effect of forced locking. This represents the frequency of the spectral fingerprint seed sent by the dominant node, i.e., the collaborative benchmark; This indicates an indicator function used to select the modal level to which constraints need to be applied, when the modal index... Equal to the target modal level index The function value is 1 when the condition is met, and 0 otherwise.

[0068] Step S303: Perform iterative update of the center frequency under constraints.

[0069] In solving the above-mentioned constrained objective function, the Alternating Direction Multiplier Method (ADMM) is used to decompose the multivariate optimization problem into a series of subproblems for iterative solution. In each iteration, for the modal center frequency... The update no longer relies solely on the centroid of the current modal power spectrum, but comprehensively considers both the centroid of the power spectrum and the seed frequency of the externally input spectral fingerprint. The weighted balance.

[0070] node Calculate the first... based on the constrained center frequency iterative update formula. The center frequency of the next iteration: ; In the formula: Indicates the first The result of the iteration calculation is the first The center frequency of each mode; Indicates the first In the nth iteration The Fourier transform result of each modal component, i.e., the frequency domain representation; It represents the frequency differential increment.

[0071] This formula shows that when When, it degenerates into the standard VMD update formula; when At that time, the new center frequency was approached Through multiple iterations, nodes The output of the first Each modal component will be strictly locked to Nearby. This process ensures the node and nodes Even when the sampling times are not synchronized, the extracted modal components are responses to the same physical frequency components, thus providing a physically comparable prerequisite for subsequent comparisons based on signal waveform complexity.

[0072] Please see the appendix Figure 3 Step S4, signal complexity quantification, aims to convert the time-domain waveform obtained from the co-decomposition into a numerical indicator that can quantify the degree of signal distortion. This is particularly important in high-resistance grounding fault scenarios where the current amplitude change is not significant, but the nonlinear complexity of the waveform increases. Specifically, this can be achieved through the following sub-steps S401 to S403: Step S401: Phase space reconstruction of characteristic modes.

[0073] Each edge computing node (including the master node) and adjacent nodes The characteristic mode components extracted in step S3 are processed using a signal complexity quantification model. Since one-dimensional time series cannot fully describe the characteristics of nonlinear dynamic systems, the processor first uses Takens' embedding theorem to reconstruct the phase space of the discretized modal signal sequence.

[0074] System setting embedding dimension and time delay For a length of modal signal sequence Reconstruct Each reconstructed vector. Include Each component is represented as Embedding dimension The value of is usually between 3 and 7, in order to balance computational efficiency with the ability to capture dynamic features.

[0075] Step S402: Statistical analysis of symbol arrangement patterns.

[0076] In the reconstructed phase space, the processor does not directly process the magnitude values, but instead focuses on the relative magnitudes between them to enhance robustness to magnitude drift. For each reconstructed vector... In The processor sorts the elements in ascending order of their numerical values ​​and obtains the original index order of the elements.

[0077] The order of these subscripts constitutes a unique symbol arrangement pattern. For embedding dimension is Theoretically, such a system exists. ( There are (factorial) possible symbol permutations. The system counts the frequency of each symbol permutation in all reconstructed vectors and calculates its probability of occurrence. .

[0078] Here, Defined as the first The ratio of the number of occurrences of a symbol arrangement pattern to the total number of reconstructed vectors. When the signal waveform is relatively regular (such as a standard sine wave), the arrangement patterns are relatively simple and the probability distribution is concentrated; when the signal waveform is distorted or contains random noise, the distribution of arrangement patterns tends to be uniform.

[0079] Step S403: Calculate the normalized permutation entropy.

[0080] Based on the probability distribution obtained through statistics, the processor applies information entropy theory to calculate the complexity of the signal. To eliminate the dimensional differences caused by different embedding dimensions and to make the results calculated by different nodes comparable, normalization is employed.

[0081] The nodes calculate the final entropy value according to the normalized permutation entropy calculation formula: ; In the formula: This represents the normalized permutation entropy value, and its range is... ; The embedding dimension determines the dimension of the phase space; This represents the factorial operation; Indicates the first The probability of occurrence of a certain symbol arrangement pattern; Indicates the symbol arrangement pattern; Represents the natural logarithm operation.

[0082] Calculated The value directly reflects the complexity of the characteristic mode components. When When the value approaches 0, it indicates that the signal exhibits a regular and definite trend, corresponding to normal operation or low-frequency oscillation; when... When the value approaches 1, it indicates that the time series of the signal exhibits highly random or complex chaotic characteristics, corresponding to the nonlinear distortion state at the time of the fault. This entropy value will serve as the sole physical quantization input for subsequent judgment of the fault segment.

[0083] Step S5 is based on the fault section determination of the entropy dissipation rate per unit electrical distance. This step S5 is the decision-making link for fault location. By comparing the rate of change of signal complexity differences between spatially adjacent nodes, it identifies whether there is a nonlinear fault source. Specifically, it can be implemented through the following sub-steps S501 to S503: Step S501: Exchange entropy data between nodes.

[0084] After completing the calculation of the normalized permutation entropy, the system performs data interaction according to the preset decision strategy.

[0085] When a single-end decision strategy is adopted, adjacent nodes The local entropy value calculated from it The response data is encapsulated in a response packet and sent back to the master node via the communication link of the edge computing network. .

[0086] When a distributed decision-making strategy is adopted, the dominant node mode then dominates the edge nodes. With neighboring nodes They will exchange the normalized permutation entropy calculated by each other. and .

[0087] Single-end decision strategy and distributed decision strategy are operating modes that the system pre-configures based on communication bandwidth resources or reliability requirements. Single-end decision strategy is generally suitable for scenarios with limited communication resources, while distributed decision strategy is generally suitable for scenarios with high requirements for decision reliability.

[0088] To ensure data timeliness, the data exchange process is equipped with a strict timeout mechanism. If entropy data from a neighboring node is not received within a preset time window, the system will mark the link communication as abnormal and trigger a backup route or perform a limited-precision evaluation based solely on single-end data. Typically, because the transmitted data is a single floating-point value, the data volume is extremely small, and the exchange process can be completed within milliseconds.

[0089] Step S502: Calculate the entropy dissipation rate per unit electrical distance.

[0090] After receiving entropy data from neighboring nodes, the processor of the edge computing node retrieves the preset line parameters from its local database, i.e., the connection nodes. With nodes The positive sequence impedance magnitude of the transmission line .

[0091] The processor calculates the entropy dissipation rate of the current line segment based on the formula for entropy dissipation rate per unit electrical distance. This physical quantity is used to quantify the degree of change in waveform complexity (characterized by permutation entropy) of a signal after it has traveled a unit electrical distance. The calculation formula is as follows: ; In the formula: Represents edge nodes With edge nodes Entropy dissipation rate per unit electrical distance, expressed in dimensionless entropy per ohm ( ); Represents a node The normalized permutation entropy value; Represents a node The normalized permutation entropy value; This represents the positive sequence impedance magnitude of the transmission line connecting the two nodes, serving as a normalization factor for the physical distance.

[0092] This calculation process eliminates the decision bias caused by varying line lengths. For long lines, even if the entropy values ​​at both ends differ significantly, the dissipation rate may be low if the impedance modulus is also large; conversely, if the entropy values ​​at both ends of a short line change abruptly, the dissipation rate will increase sharply.

[0093] Step S503: Application of fault section determination criteria.

[0094] The system has a preset natural attenuation reference threshold. This threshold is derived through statistical analysis of historical data from the distribution network under normal operating conditions and external disturbances (such as load switching and capacitor bank operation). Under non-fault conditions, when the signal is transmitted along the line, the waveform undergoes linear attenuation or slight distortion due to the influence of line impedance and distributed capacitance, and its entropy dissipation rate is usually maintained at a low level.

[0095] The processor will calculate Comparison with the natural decay benchmark threshold: like If the value is less than or equal to the natural decay reference threshold, the node is determined. With nodes The lines between them are in normal transmission condition or there is only external disturbance, and no fault has occurred.

[0096] like If the value is greater than the natural decay reference threshold, the node is determined. With nodes There is nonlinear energy leakage in the line between them. This nonlinear energy leakage is due to the additional signal complexity introduced by nonlinear physical processes such as arc discharge and dielectric breakdown at the fault point (especially the high-resistance grounding fault point), causing the entropy value of the signal to increase abnormally sharply when passing through this section. At this time, the system identifies this line section as the fault section, generates alarm information and reports it to the distribution network management master station to complete the fault location.

[0097] In summary, the real-time fault location method for power equipment based on edge computing provided by this invention changes the conventional path of fault location in traditional power distribution networks that relies on high-precision time synchronization.

[0098] By constructing a logical topology and establishing a master-slave collaboration mechanism at the edge, and using the seed frequency of the spectral fingerprint extracted by the dominant node as a physical anchor point, adjacent nodes are induced to perform variational mode decomposition under constraints, thereby achieving forced locking and alignment of distributed nodes in the frequency domain.

[0099] This frequency domain locking mechanism effectively avoids phase and amplitude errors caused by asynchronous sampling clocks, enabling comparable fault characteristic data to be obtained without the need for high-cost satellite timing devices.

[0100] Meanwhile, this invention abandons the impedance method approach that relies solely on current amplitude for judgment, and instead uses normalized permutation entropy to quantify the nonlinear complexity of the signal waveform, and uses the entropy dissipation rate per unit electrical distance as the final fault criterion.

[0101] This physical quantity is highly sensitive to the weak arc distortion caused by high-resistance grounding faults. Furthermore, by introducing line impedance as a normalization factor, it eliminates the judgment interference caused by the varying lengths of the lines, thereby improving the accuracy and robustness of the location.

[0102] Furthermore, the edge node method of the present invention can be implemented in the form of a software product and stored in a computer-readable storage medium, or embedded in the hardware processor of an electronic device.

[0103] The computer-readable storage medium may be a non-volatile memory, such as a read-only memory (ROM), flash memory, hard disk, etc., which stores computer program instructions. When these instructions are executed by the arithmetic unit of a computer or processor, they can implement the method steps of the edge nodes in the above embodiments.

[0104] The corresponding electronic device may include a processor, a communication interface, a memory, and a communication bus. The processor reads and executes the program in the memory through the communication bus, thereby implementing the fault location logic of the present invention at the hardware level.

Claims

1. An edge computing-based real-time fault location method for power equipment, characterized in that, The method comprises the following steps: S1, constructing an edge computing logical topology graph, presetting a positive sequence impedance modulus in an edge node, and monitoring a zero sequence current signal; S2, switching to a leading node mode when any edge node detects an energy mutation of the zero sequence current signal, decomposing the zero sequence current signal by using a VMD unconstrained variational model to extract a spectral fingerprint seed frequency and sending the spectral fingerprint seed frequency to a neighboring node; S3, after the neighboring node receives the spectral fingerprint seed frequency, decomposing the zero sequence current signal in the same frequency band by using a CC-VMD constrained optimization model to obtain a characteristic modal component aligned with the edge node in the frequency domain in the leading node mode; S4, calculating a normalized permutation entropy of the characteristic modal component; S5, exchanging the normalized permutation entropy, calculating a unit electrical distance entropy dissipation rate in combination with the positive sequence impedance modulus, and determining a fault section. 2.The power equipment real-time fault locating method based on edge computing according to claim 1, characterized in that, The S1 step specifically comprises: In a local database of the edge node, establishing the edge computing logical topology graph describing the connection relationship between the edge node set and the power transmission line set; In the local database, prewriting electrical parameters of the power transmission line between the edge node and the directly adjacent edge node; Selecting the positive sequence impedance modulus of the power transmission line as a key physical parameter, which is calculated based on the unit length impedance parameter of the power transmission line and the actual physical length of the line, and is used as a physical reference for subsequent calculation of the unit electrical distance entropy dissipation rate; The edge node continuously monitors the energy of the zero sequence current signal in the buffer area. 3.The power equipment real-time fault locating method based on edge computing according to claim 1, characterized in that, In the S2 step, when any edge node detects an energy mutation of the zero sequence current signal, switching to a leading node mode specifically comprises: When detecting that the energy mutation value of the zero sequence current signal exceeds a silence threshold, triggering an interruption, and switching the running state of the edge node from a monitoring mode to a leading node mode; The silence threshold is pre-set according to the background noise level during normal operation of the distribution network. 4.The method of claim 1, wherein, In the S2 step, decomposing the zero sequence current signal by using a VMD unconstrained variational model specifically comprises: Establishing a variational constraint problem by using the VMD unconstrained variational model, that is, seeking the sum of the bandwidths of each modal component decomposed from the zero sequence current signal to be minimum and the sum of each modal component to be equal to the zero sequence current signal; Introducing a quadratic penalty factor and a Lagrange multiplier to construct an augmented Lagrange variational objective function; Iteratively solving the augmented Lagrange variational objective function by using an alternating direction multiplier method to obtain a group of modal components and corresponding center frequencies.

5. The edge computing based real-time fault locating method for power equipment according to claim 4, characterized in that, In the S2 step, extracting a spectral fingerprint seed frequency and sending it to a neighboring node specifically comprises: Calculating the energy amplitude of the group of modal components decomposed by the VMD unconstrained variational model; Screening out the modal component with the largest energy amplitude, and determining the center frequency corresponding to the modal component with the largest energy amplitude as the spectral fingerprint seed frequency; Packaging the spectral fingerprint seed frequency into an induction request data packet and sending it to the neighboring node directly connected in the edge computing logical topology graph through a communication network.

6. The edge computing based real-time fault locating method for power equipment according to claim 1, characterized in that, In step S3, the specific steps of using the CC-VMD constraint optimization model to force the zero-sequence current signal to decompose within the same frequency band include: The adjacent nodes enter the master-slave collaborative induced decomposition process and start the center frequency constrained variational mode decomposition to decompose the zero-sequence current signal. Specifically, the CC-VMD constrained optimization model is used to add a frequency penalty term to the variational model to construct a frequency domain locking constraint objective function. The frequency domain locking constraint objective function includes frequency locking weight coefficients, the spectral fingerprint seed frequency sent by the edge nodes under the dominant node mode, and an indication function for selecting the target modal component.

7. The edge computing based real-time fault locating method for power equipment according to claim 6, characterized in that, In step S3, obtaining the characteristic modal components aligned with the frequency domain of the edge nodes under the dominant node mode specifically includes: In solving the frequency domain locking constraint objective function, the center frequency of the modal component is iteratively calculated using the center frequency iterative update formula; In the center frequency iterative update formula, the center frequency of the modal component is updated by weighting the power spectrum centroid of the modal component calculated in the current iteration and by using the frequency locking weight coefficient and the indicator function to introduce the spectral fingerprint seed frequency for weighted balance. By updating the center frequency of the modal components, the center frequency of the target modal components output by adjacent nodes converges to the spectral fingerprint seed frequency, thereby obtaining frequency-domain aligned feature modal components. 8.The method of claim 1, wherein, In step S4, calculating the normalized permutation entropy of the characteristic mode components specifically includes: The edge nodes and adjacent nodes utilize a signal complexity quantification model and employ a time delay embedding theorem to reconstruct the phase space of the feature mode components, resulting in multiple reconstructed vectors. The frequency of occurrence of symbol arrangement patterns in the multiple reconstructed vectors is counted, and the probability of occurrence of each symbol arrangement pattern is calculated. Based on the occurrence probability, the normalized permutation entropy is calculated using the normalized permutation entropy calculation formula. 9.The power equipment real-time fault locating method based on edge computing according to claim 1, wherein, In step S5, the edge node exchanges the normalized permutation entropy with its neighboring nodes, and calculates the entropy dissipation rate per unit electrical distance by combining the positive sequence impedance magnitude. Specifically, this includes: When a single-end decision strategy is adopted, the neighboring node encapsulates the calculated normalized permutation entropy into the response data packet and sends it back to the edge node in the dominant node mode. When a distributed decision strategy is adopted, the edge nodes and their neighboring nodes in the dominant node mode exchange the normalized permutation entropy calculated by each other. Retrieve the preset positive sequence impedance magnitude; The entropy dissipation rate per unit electrical distance is calculated using the formula for entropy dissipation rate per unit electrical distance.

10. The edge computing based real-time fault locating method for power equipment according to claim 1, characterized in that, In step S5, determining the faulty section specifically includes: The fault section determination criteria are applied, and the calculated entropy dissipation rate per unit electrical distance is compared with the natural attenuation reference threshold. If the entropy dissipation rate per unit electrical distance is greater than the natural attenuation reference threshold, it is determined that there is nonlinear energy leakage in the line between the two edge nodes, and the line is identified as a fault section. The natural attenuation reference threshold is preset according to historical data of the power distribution network under normal operation conditions and external disturbances. The natural attenuation reference threshold is preset according to historical data of the power distribution network under normal operation conditions and external disturbances. The natural attenuation reference threshold is preset according to historical data of the power distribution network under normal operation conditions and external disturbances.

Images