Power grid fault feature determining and positioning method, system, equipment and storage medium

Abstract

The invention discloses a power grid fault characteristic determination and positioning method, system, device and storage medium, and relates to the field of power grid fault processing, and the method comprises the steps: synchronously collecting the voltage, current, frequency and other electrical quantities and temperature, humidity and other environmental quantities of a power grid node through a multi-mode sensor, and forming an original time sequence data set; separating electrical quantity frequency domain fault features by using improved variational mode decomposition, and analyzing the correlation between the environmental quantity and the electrical quantity through a grey correlation degree to suppress noise; mapping the dynamic deviation of each order of derivative of the voltage / current into a spatial-temporal characteristic matrix, and constructing a distribution map in combination with power grid topology; based on the dynamic graph neural network, inputting node features and edge features, and establishing a fault-environment association model through feature propagation and a graph attention mechanism; and carrying out fine trimming on a positioning result by combining double-end traveling wave coarse positioning with an environment correlation degree correction coefficient, and verifying connectivity of a fault point and a power grid topology.

Description

Technical Field

[0001] This invention relates to the field of power grid fault handling, specifically to a method, system, device, and storage medium for determining and locating power grid fault characteristics. Background Technology

[0002] As a critical infrastructure for modern energy transmission, the safe and stable operation of the power grid is essential for social production and people's livelihood. However, during operation, the power grid is often affected by factors such as equipment aging, environmental fluctuations, and electromagnetic interference, which can easily lead to faults such as short circuits and grounding, resulting in power outages or equipment damage. Quickly and accurately identifying fault characteristics and locating fault points is a core requirement for improving the efficiency and reliability of power grid operation and maintenance.

[0003] Existing technologies face significant challenges in power grid fault analysis and location: First, data acquisition is often limited to a single dimension, focusing only on electrical quantities such as voltage, current, and frequency, without simultaneously incorporating environmental quantities such as temperature and humidity. This makes it difficult to reveal the potential impact of environmental factors on fault characteristics, resulting in feature extraction being hampered by environmental noise and yielding incomplete information. Second, frequency domain feature separation techniques have limitations. Traditional methods such as Fourier transform and fixed-parameter mode decomposition cannot adaptively separate fault characteristic frequencies from non-power frequency harmonic interference, making it difficult to accurately extract effective features in complex harmonic environments. Third, the correlation analysis between environmental and electrical quantities lacks systematic methods and fails to assess the degree of environmental factors' impact on faults through quantitative methods, leading to an inability to... Fourth, the existing methods lack spatiotemporal correlation modeling of fault features. They fail to map the dynamic deviations of voltage / current derivatives to grid node locations, making it impossible to construct a spatiotemporal feature matrix and topology map reflecting the temporal evolution and spatial distribution of faults. Furthermore, traditional traveling wave localization only calculates distance based on time difference, without verifying the actual connectivity of the fault point in conjunction with the grid topology, potentially leading to a mismatch between the localization results and the physical lines. Fifth, the existing algorithms lack dynamic correction for the impact of temperature and humidity on grid parameters such as line impedance and insulation resistance. The traveling wave localization process lacks a dynamic calibration mechanism based on environmental correlation, resulting in significant fluctuations in localization accuracy with changing environmental conditions. These issues render the existing technology insufficient in terms of fault detection accuracy and reliability in complex grid environments, necessitating innovative methods to overcome technical bottlenecks in data dimensionality, feature extraction, modeling analysis, and environmental adaptability. Summary of the Invention

[0004] The purpose of this invention is to provide a method, system, device, and storage medium for determining and locating power grid fault characteristics. The method involves collecting electrical and environmental quantities of power grid nodes through multimodal sensors, extracting fault characteristics through variational mode decomposition and grey relational analysis, modeling the correlation between faults and the environment using dynamic graph neural networks, and combining dual-end traveling wave localization and environmental correction to achieve accurate determination and location of power grid fault characteristics.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A method for determining and locating power grid fault characteristics, characterized by comprising the following steps:

[0007] S1: Multi-source heterogeneous data acquisition. Multimodal sensors are deployed at power grid nodes to simultaneously collect electrical quantities such as voltage, current, and frequency, as well as environmental quantities such as temperature and humidity, forming a raw time-series dataset stored according to timestamps.

[0008] S2: Decoupling of data spatiotemporal features: Frequency domain decomposition of electrical quantities is performed by improving variational mode decomposition to separate fault characteristic frequencies and suppress harmonic interference. The correlation between environmental quantities and electrical quantities is calculated using grey relational analysis to extract effective fault features and suppress noise.

[0009] S3: Fault feature matrix construction, calculate the dynamic deviation of voltage / current derivatives of each order with normal operation data, map the deviation values ​​to the grid node location according to the time series index to form a spatiotemporal feature matrix, quantify fault features and combine them with the grid topology to form a distribution map;

[0010] S4: Establish fault-environment correlation modeling based on dynamic graph neural network. The input includes node features containing electrical timing features and environmental correlation degree, as well as equilateral features such as line impedance and insulation resistance corrected for temperature and humidity. The correlation model between fault and environment is established through feature propagation and graph attention mechanism.

[0011] S5: Traveling wave positioning dynamic calibration. First, coarse positioning of the two ends of the traveling wave is performed based on the time difference of the transient traveling wave arriving at both ends, and the coordinates of the fault point are verified to ensure the positioning accuracy.

[0012] Step S1, multi-source heterogeneous data acquisition, specifically involves deploying multi-modal sensors at power grid nodes to simultaneously collect electrical and environmental quantities. The electrical quantities include voltage V(t), current I(t), and frequency f(t); the environmental quantities include temperature T(t) and humidity H(t). u (t); forming the original time series data {V(t),I(t),f(t),T(t),H u (t)}, stored according to timestamps.

[0013] 3. The method for determining and locating power grid fault characteristics according to claim 1, characterized in that step S2, data spatiotemporal feature decoupling, specifically includes separating effective fault characteristics from the collected raw time-series data, suppressing noise and extracting physical laws, including frequency domain decomposition of electrical quantities and correlation analysis of environmental quantities, comprising the following steps:

[0014] S21: Frequency domain decomposition of electrical quantities. An improved variational mode decomposition is performed on V(t), I(t), and f(t) to solve the variational constraint model as follows:

[0015]

[0016] In the formula, u k (t) represents the k-th frequency sub-signal, ω k Let be the center angular frequency of the k-th sub-signal, K be the total number of sub-signals obtained from the decomposition, δ(t) be the unit impulse function used for signal time-domain processing, j be the imaginary unit, π be the constant of pi, t be the time variable, and * be the convolution operation. The first-order partial derivative with respect to time t, The exponential factor for frequency domain transformation, used to convert a signal from the time domain to the frequency domain, L 2 Given the square of the norm and V(t) as the original voltage signal, K sub-signals of different frequencies, u, are adaptively separated. k (t), filtering out non-power frequency harmonic interference and retaining fault characteristic frequencies;

[0017] S22: Correlation analysis of environmental quantities, for T(t), H u (t) Perform grey relational analysis, using voltage / current as the reference sequence x0(j), and calculate the temperature / humidity sequence x. T (j) correlation ρ T , The formula is as follows:

[0018]

[0019] In the formula: ρ is the grey relational degree, x0(j) is the reference sequence, x i (j) represents the comparison sequence, where i is the sequence index and x is the index. 环境 (j) represents a specific environmental quantity sequence, including temperature T(t) and humidity H. u (t), j are the time points or sampling point indices in the sequence, and ξ is the resolution coefficient, with a value range of [0,1]; the correlation between output temperature and electrical quantity ρ T Correlation between humidity and electrical quantity The larger the value, the more significant the impact of environmental factors on the fault.

[0020] Step S3, fault feature matrix construction, specifically involves quantifying time-domain / frequency-domain anomalies into computable matrix features and combining them with the power grid topology to form a spatiotemporal distribution map. This includes the following steps:

[0021] S31: Dynamic deviation calculation. Using the i-th derivative of voltage / current (i = 1, 2, 3), corresponding to the rate of change and acceleration transient characteristics, the deviation from normal operating data is calculated. The formula is as follows:

[0022]

[0023] In the formula: F ij Let i be the i-th derivative, j be the dynamic deviation eigenvalue at time point j, i be the order of the derivative (i = 1, 2, 3), and j be the index of the time point and the sampling point. Let V(t) be the i-th time derivative of the voltage signal V(t), V norm (t) represents the voltage signal under normal operating conditions. Let I be the i-th time derivative of the current signal I(t), I norm (t) represents the current signal under normal operating conditions, S ij F is a normalization factor to eliminate the influence of dimensions. ij The higher the value, the higher the probability of failure;

[0024] S32: Matrix spatiotemporal mapping, using time series index j and power grid node location, to map F... ij The mapping is a spatiotemporal feature matrix, where rows represent the order of derivatives and columns represent time / nodes.

[0025] Step S4, based on the results of the above steps, establishes a fault-environment correlation model, using feature propagation from a dynamic graph neural network (GNN). The input node features include F... ij Mapped to spatiotemporal feature matrix and environmental correlation ρ T , The side characteristics include the temperature-corrected line impedance Z(T) = Z0(1+α) T (T-T0) and humidity-corrected insulation resistance Feature propagation is performed, and the feature propagation formula is:

[0026]

[0027] In the formula: The feature vector of node v in the l-th layer of the neural network includes the electrical timing features of node v. Temperature and humidity correlation at point v l represents the number of layers in the neural network, v represents a node in the graph structure, corresponding to a physical node in the power grid, u represents the neighboring nodes of node v, and W represents the number of nodes in the graph structure. (l) The node feature weight matrix of the l-th layer, Let h be the edge feature weight matrix of the l-th layer. e (v,u) represents the edge characteristics between nodes v and u, including the temperature-corrected line impedance Z(T) = Z0(1+α) T (T-T0) and humidity-corrected insulation resistance Where Z0 is the reference impedance, α T, where is the temperature coefficient, T is the current temperature, T0 is the reference temperature, R0 is the reference resistance, β is the humidity sensitivity coefficient, and H is the temperature coefficient. u Humidity; σ is the activation function;

[0028] α vu The attention weights for nodes v and u are calculated using the Graph Attention (GAT) mechanism, and the formula is as follows:

[0029]

[0030] Where: h e (v,u) represents the edge characteristics between nodes v and u, including the temperature-corrected line impedance Z(T) = Z0(1+α) T (T-T0) and humidity-corrected insulation resistance Where Z0 is the reference impedance, α T , where is the temperature coefficient, T is the current temperature, T0 is the reference temperature, R0 is the reference resistance, β is the humidity sensitivity coefficient, and H is the temperature coefficient. u For humidity; α vu Let be the attention weights between node v and its neighbor u, exp be the exponential function, LeakyReLU(·) be the linear rectified activation function with leakage, and a be the attention weights between node v and its neighbor u. T This represents the weight vector for the attention mechanism; This is a vector concatenation operation that concatenates the feature vectors of nodes v and u into a long vector. W is a shared feature transformation matrix used to unify the feature dimensions, and k is the index of the adjacent node.

[0031] Step S5, traveling wave positioning and dynamic calibration, combines fault characteristics and physical effects to locate the fault point, including the following steps:

[0032] S51: Coarse positioning of two-terminal traveling wave. Based on the time difference Δt = t1 - t2 of the transient traveling wave arriving at both ends, the initial distance x is calculated using the following formula:

[0033]

[0034] In the formula: v is the initial velocity of the traveling wave, x is the estimated distance from the fault point to the measuring end, v is the initial propagation velocity of the traveling wave in the transmission line, Δt is the time difference between the transient traveling wave and the two ends of the line, t1 and t2 are the times when the traveling wave arrives at the nodes at both ends of the line, taking the absolute value, and L is the total length of the transmission line.

[0035] S52: Verify the location result. Input the corrected fault point coordinates into the power grid topology map to verify whether it is located on the actual connected line. If it does not meet the requirements, trigger a re-verification and return to step 2 to supplement the data.

[0036] The coarse localization of the dual-end traveling wave is based on feature-assisted fine-tuning using dynamic calculation of correction coefficients based on fault-environment correlation, and the formula is as follows:

[0037]

[0038] In the formula: γ is the dynamic correction coefficient, γ0 is the basic correction coefficient, and η T ,η H For environmental correlation weighting factors, The correlation between temperature, humidity and electrical quantities obtained in step S2.

[0039] A system for determining and locating power grid fault characteristics is characterized by comprising a multi-source data acquisition module, a feature decoupling and analysis module, a fault feature modeling module, a fault-environment correlation modeling module, and a traveling wave location and calibration module. The multi-source data acquisition module includes deploying multimodal sensors at power grid nodes to simultaneously acquire electrical and environmental quantities to form an original time-series dataset. The feature decoupling and analysis module performs frequency domain decomposition on the electrical quantities and correlation analysis on the environmental quantities to separate effective fault features. The fault feature modeling module quantizes time-domain / frequency-domain anomalies into a spatiotemporal feature matrix and combines it with the power grid topology to form a distribution map. The fault-environment correlation modeling module uses a dynamic graph neural network to propagate features from node and edge features to establish a fault-environment correlation model. The traveling wave location and calibration module uses the time difference of arrival at both ends of the traveling wave for coarse location, and then refines and verifies the location result by combining a dynamic correction coefficient based on environmental correlation.

[0040] A terminal device, characterized in that it includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, it implements the method of any one of claims 1-7.

[0041] A computer-readable storage medium, characterized in that the storage medium stores a computer program that, when executed by a processor, implements the method of any one of claims 1-7.

[0042] The working mechanism of this power grid fault feature determination and location method is as follows: Electrical and environmental quantities of power grid nodes are synchronously collected using multimodal sensors to form a time-stamped raw time-series dataset; frequency domain decomposition of electrical quantities is performed using improved variational mode decomposition to separate fault characteristic frequencies and suppress harmonic interference; simultaneously, the correlation between environmental and electrical quantities is calculated using grey relational analysis to extract effective fault features and suppress noise; the dynamic deviation values ​​of voltage / current derivatives from normal operation data are mapped to a spatiotemporal feature matrix by time-series index and power grid node location, and combined with the power grid topology to form a distribution map, quantifying fault features; based on a dynamic graph neural network, node features including electrical time-series features and environmental correlation, as well as equilateral features of line impedance and insulation resistance corrected for temperature and humidity, are input, and a fault-environment correlation model is established through feature propagation and graph attention mechanisms; coarse localization of the traveling wave at both ends is first performed based on the time difference of the transient traveling wave arriving at both ends, and then feature-assisted refinement is performed by dynamically calculating the correction coefficient of the fault-environment correlation, and verifying whether the fault point coordinates are located on the actual connected line, thus achieving accurate fault point location.

[0043] Compared with the prior art, the beneficial effects of the present invention are:

[0044] This invention simultaneously collects electrical quantities such as voltage, current, and frequency, as well as environmental quantities such as temperature and humidity, to form a multi-source heterogeneous time-series dataset. It captures fault characteristics from a dual dimension of "electrical characteristics + environmental impact," avoiding the loss of information from a single data dimension.

[0045] By improving variational mode decomposition (VMD) to decompose electrical quantities in the frequency domain, multiple sub-signals of different frequencies are adaptively separated. Compared with traditional Fourier transform or fixed parameter decomposition methods, it can extract fault characteristic frequencies more accurately, effectively suppress harmonic interference, and improve the robustness of feature extraction.

[0046] By using grey relational analysis to quantify the correlation between environmental and electrical quantities, environmental factors that significantly affect faults are screened out, irrelevant noise is suppressed, and the problem of feature confusion caused by environmental interference in existing technologies is solved, thereby improving feature purity.

[0047] By mapping the dynamic deviation values ​​of voltage / current derivatives of each order to a spatiotemporal characteristic matrix and combining them with the power grid topology to form a distribution map, abstract fault characteristics are transformed into a calculable and visualized matrix structure, which facilitates intuitive analysis of the temporal evolution and spatial distribution patterns of faults, and is superior to traditional qualitative analysis methods.

[0048] Based on dynamic graph neural networks, this paper combines the electrical timing features of nodes with environmental correlation and the line impedance and insulation resistance corrected by temperature and humidity of edge features. Through feature propagation and graph attention mechanism, a dynamic correlation model between faults and the environment is established, which breaks through the limitation of ignoring the impact of environmental factors on faults in the existing technology and improves the adaptability of the model to complex operating conditions.

[0049] The dual-end traveling wave coarse positioning is combined with the environmental correlation correction coefficient for dynamic fine positioning, and the fault point is verified to be located on the actual connected line. This solves the positioning deviation problem caused by environmental changes in traditional traveling wave positioning, ensures that the positioning results are consistent with the physical reality, and improves the accuracy of fault positioning and engineering practicality. Attached Figure Description

[0050] Figure 1 This is a flowchart of a method for determining and locating power grid fault characteristics according to the present invention. Detailed Implementation

[0051] The technical solutions of the present invention will now be described in detail with reference to the accompanying drawings.

[0052] like Figure 1 As shown, a method for determining and locating power grid fault characteristics is characterized by comprising the following steps:

[0053] S1: Multi-source heterogeneous data acquisition. Multimodal sensors are deployed at power grid nodes to simultaneously collect electrical quantities such as voltage, current, and frequency, as well as environmental quantities such as temperature and humidity, forming a raw time-series dataset stored according to timestamps.

[0054] S2: Decoupling of data spatiotemporal features: Frequency domain decomposition of electrical quantities is performed by improving variational mode decomposition to separate fault characteristic frequencies and suppress harmonic interference. The correlation between environmental quantities and electrical quantities is calculated using grey relational analysis to extract effective fault features and suppress noise.

[0055] S3: Fault feature matrix construction, calculate the dynamic deviation of voltage / current derivatives of each order with normal operation data, map the deviation values ​​to the grid node location according to the time series index to form a spatiotemporal feature matrix, quantify fault features and combine them with the grid topology to form a distribution map;

[0056] S4: Establish fault-environment correlation modeling based on dynamic graph neural network. The input includes node features containing electrical timing features and environmental correlation degree, as well as equilateral features such as line impedance and insulation resistance corrected for temperature and humidity. The correlation model between fault and environment is established through feature propagation and graph attention mechanism.

[0057] S5: Traveling wave positioning dynamic calibration. First, coarse positioning of the two ends of the traveling wave is performed based on the time difference of the transient traveling wave arriving at both ends, and the coordinates of the fault point are verified to ensure the positioning accuracy.

[0058] Step S1, multi-source heterogeneous data acquisition, specifically involves deploying multi-modal sensors at power grid nodes to simultaneously collect electrical and environmental quantities. The electrical quantities include voltage V(t), current I(t), and frequency f(t); the environmental quantities include temperature T(t) and humidity H(t). u (t); forming the original time series data {V(t),I(t),f(t),T(t),H u (t)}, stored according to timestamps.

[0059] 3. The method for determining and locating power grid fault characteristics according to claim 1, characterized in that step S2, data spatiotemporal feature decoupling, specifically includes separating effective fault characteristics from the collected raw time-series data, suppressing noise and extracting physical laws, including frequency domain decomposition of electrical quantities and correlation analysis of environmental quantities, comprising the following steps:

[0060] S21: Frequency domain decomposition of electrical quantities. An improved variational mode decomposition is performed on V(t), I(t), and f(t) to solve the variational constraint model as follows:

[0061]

[0062] In the formula, u k (t) represents the k-th frequency sub-signal, ω k Let be the center angular frequency of the k-th sub-signal, K be the total number of sub-signals obtained from the decomposition, δ(t) be the unit impulse function used for signal time-domain processing, j be the imaginary unit, π be the constant of pi, t be the time variable, and * be the convolution operation. The first-order partial derivative with respect to time t, The exponential factor for frequency domain transformation, used to convert a signal from the time domain to the frequency domain, L 2 Given the square of the norm and V(t) as the original voltage signal, K sub-signals of different frequencies, u, are adaptively separated. k (t), filtering out non-power frequency harmonic interference and retaining fault characteristic frequencies;

[0063] S22: Correlation analysis of environmental quantities, for T(t), H u (t) Perform grey relational analysis, using voltage / current as the reference sequence x0(j), and calculate the temperature / humidity sequence. correlation The formula is as follows:

[0064]

[0065] In the formula: ρ is the grey relational degree, x0(j) is the reference sequence, x i (j) represents the comparison sequence, where i is the sequence index and x is the index. 环境 (j) represents a specific environmental quantity sequence, including temperature T(t) and humidity H. u (t), j are the time points or sampling point indices in the sequence, and ξ is the resolution coefficient, with a value range of [0,1]; the correlation between output temperature and electrical quantity ρ T Correlation between humidity and electrical quantity The larger the value, the more significant the impact of environmental factors on the fault.

[0066] Step S3, fault feature matrix construction, specifically involves quantifying time-domain / frequency-domain anomalies into computable matrix features and combining them with the power grid topology to form a spatiotemporal distribution map. This includes the following steps:

[0067] S31: Dynamic deviation calculation. Using the i-th derivative of voltage / current (i = 1, 2, 3), corresponding to the rate of change and acceleration transient characteristics, the deviation from normal operating data is calculated. The formula is as follows:

[0068]

[0069] In the formula: F ij Let i be the i-th derivative, j be the dynamic deviation eigenvalue at time point j, i be the order of the derivative (i = 1, 2, 3), and j be the index of the time point and the sampling point. Let V(t) be the i-th time derivative of the voltage signal V(t), V norm (t) represents the voltage signal under normal operating conditions. Let I be the i-th time derivative of the current signal I(t), I norm (t) represents the current signal under normal operating conditions, S ij F is a normalization factor to eliminate the influence of dimensions. ij The higher the value, the higher the probability of failure;

[0070] S32: Matrix spatiotemporal mapping, using time series index j and power grid node location, to map F... ij The mapping is a spatiotemporal feature matrix, where rows represent the order of derivatives and columns represent time / nodes.

[0071] Step S4, based on the results of the above steps, establishes a fault-environment correlation model, using feature propagation from a dynamic graph neural network (GNN). The input node features include F... ij Mapped to spatiotemporal feature matrix and environmental correlation The side characteristics include the temperature-corrected line impedance Z(T) = Z0(1+α) T (T-T0) and humidity-corrected insulation resistance Feature propagation is performed, and the feature propagation formula is:

[0072]

[0073] In the formula: The feature vector of node v in the l-th layer of the neural network includes the electrical timing features of node v. The temperature and humidity correlation ρ corresponding to point v T , l represents the number of layers in the neural network, v represents a node in the graph structure, corresponding to a physical node in the power grid, u represents the neighboring nodes of node v, and W represents the number of nodes in the graph structure. (l) The node feature weight matrix of the l-th layer, Let h be the edge feature weight matrix of the l-th layer. e (v,u) represents the edge characteristics between nodes v and u, including the temperature-corrected line impedance Z(T) = Z0(1+α) T (T-T0) and humidity-corrected insulation resistance Where Z0 is the reference impedance, α T , where is the temperature coefficient, T is the current temperature, T0 is the reference temperature, R0 is the reference resistance, β is the humidity sensitivity coefficient, and H is the temperature coefficient. u Humidity; σ is the activation function;

[0074] α vu The attention weights for nodes v and u are calculated using the Graph Attention (GAT) mechanism, and the formula is as follows:

[0075]

[0076] Where: h e (v,u) represents the edge characteristics between nodes v and u, including the temperature-corrected line impedance Z(T) = Z0(1+α) T (T-T0) and humidity-corrected insulation resistance Where Z0 is the reference impedance, α T , where is the temperature coefficient, T is the current temperature, T0 is the reference temperature, R0 is the reference resistance, β is the humidity sensitivity coefficient, and H is the temperature coefficient. u For humidity; α vu Let be the attention weights between node v and its neighbor u, exp be the exponential function, LeakyReLU(·) be the linear rectified activation function with leakage, and a be the attention weights between node v and its neighbor u. T This represents the weight vector for the attention mechanism; This is a vector concatenation operation that concatenates the feature vectors of nodes v and u into a long vector. W is a shared feature transformation matrix used to unify the feature dimensions, and k is the index of the adjacent node.

[0077] Step S5, traveling wave positioning and dynamic calibration, combines fault characteristics and physical effects to locate the fault point, including the following steps:

[0078] S51: Coarse positioning of two-terminal traveling wave. Based on the time difference Δt = t1 - t2 of the transient traveling wave arriving at both ends, the initial distance x is calculated using the following formula:

[0079]

[0080] In the formula: v is the initial velocity of the traveling wave, x is the estimated distance from the fault point to the measuring end, v is the initial propagation velocity of the traveling wave in the transmission line, Δt is the time difference between the transient traveling wave and the two ends of the line, t1 and t2 are the times when the traveling wave arrives at the nodes at both ends of the line, taking the absolute value, and L is the total length of the transmission line.

[0081] S52: Verify the location result. Input the corrected fault point coordinates into the power grid topology map to verify whether it is located on the actual connected line. If it does not meet the requirements, trigger a re-verification and return to step 2 to supplement the data.

[0082] The coarse localization of the dual-end traveling wave is based on feature-assisted fine-tuning using dynamic calculation of correction coefficients based on fault-environment correlation, and the formula is as follows:

[0083]

[0084] In the formula: γ is the dynamic correction coefficient, γ0 is the basic correction coefficient, and η T ,η H For environmental correlation weighting factors, The correlation between temperature, humidity and electrical quantities obtained in step S2.

[0085] A system for determining and locating power grid fault characteristics is characterized by comprising a multi-source data acquisition module, a feature decoupling and analysis module, a fault feature modeling module, a fault-environment correlation modeling module, and a traveling wave location and calibration module. The multi-source data acquisition module includes deploying multimodal sensors at power grid nodes to simultaneously acquire electrical and environmental quantities to form an original time-series dataset. The feature decoupling and analysis module performs frequency domain decomposition on the electrical quantities and correlation analysis on the environmental quantities to separate effective fault features. The fault feature modeling module quantizes time-domain / frequency-domain anomalies into a spatiotemporal feature matrix and combines it with the power grid topology to form a distribution map. The fault-environment correlation modeling module uses a dynamic graph neural network to propagate features from node and edge features to establish a fault-environment correlation model. The traveling wave location and calibration module uses the time difference of arrival at both ends of the traveling wave for coarse location, and then refines and verifies the location result by combining a dynamic correction coefficient based on environmental correlation.

[0086] A terminal device, characterized in that it includes a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, it implements the method of any one of claims 1-7.

[0087] A computer-readable storage medium, characterized in that the storage medium stores a computer program that, when executed by a processor, implements the method of any one of claims 1-7.

[0088] Its specific implementation method is as follows:

[0089] S1: Multi-source heterogeneous data acquisition involves deploying multimodal sensors at power grid nodes to simultaneously collect electrical quantities such as voltage, current, and frequency, as well as environmental quantities such as temperature and humidity, forming a raw time-series dataset stored by timestamps. Specifically, this involves deploying multimodal sensors at power grid nodes to simultaneously collect electrical and environmental quantities. The electrical quantities include voltage V(t), current I(t), and frequency f(t); the environmental quantities include temperature T(t), humidity H(t), and so on. u (t); forming the original time series data {V(t),I(t),f(t),T(t),H u (t)}, stored according to timestamps.

[0090] S2: Decoupling of spatiotemporal features of data. This involves improving variational mode decomposition to perform frequency domain decomposition of electrical quantities to separate fault characteristic frequencies and suppress harmonic interference. Grey relational analysis is used to calculate the correlation between environmental and electrical quantities, extracting effective fault features and suppressing noise. Specifically, this includes separating effective fault features from the collected raw time-series data, suppressing noise, and extracting physical laws, including frequency domain decomposition of electrical quantities and correlation analysis of environmental quantities. The steps include:

[0091] S21: Frequency domain decomposition of electrical quantities. An improved variational mode decomposition is performed on V(t), I(t), and f(t) to solve the variational constraint model as follows:

[0092]

[0093] In the formula, u k (t) represents the k-th frequency sub-signal, ω k Let be the center angular frequency of the k-th sub-signal, K be the total number of sub-signals obtained from the decomposition, δ(t) be the unit impulse function used for signal time-domain processing, j be the imaginary unit, π be the constant of pi, t be the time variable, and * be the convolution operation. The first-order partial derivative with respect to time t, The exponential factor for frequency domain transformation, used to convert a signal from the time domain to the frequency domain, L 2 Given the square of the norm and V(t) as the original voltage signal, K sub-signals of different frequencies, u, are adaptively separated. k (t), filtering out non-power frequency harmonic interference and retaining fault characteristic frequencies;

[0094] S22: Correlation analysis of environmental quantities, for T(t), H u(t) Perform grey relational analysis, using voltage / current as the reference sequence x0(j), and calculate the temperature / humidity sequence. correlation The formula is as follows:

[0095]

[0096] In the formula: ρ is the grey relational degree, x0(j) is the reference sequence, x i (j) represents the comparison sequence, where i is the sequence index and x is the index. 环境 (j) represents a specific environmental quantity sequence, including temperature T(t) and humidity H. u (t), j are the time points or sampling point indices in the sequence, and ξ is the resolution coefficient, with a value range of [0,1]; the correlation between output temperature and electrical quantity ρ T Correlation between humidity and electrical quantity The larger the value, the more significant the impact of environmental factors on the fault.

[0097] S3: Fault feature matrix construction involves calculating the dynamic deviations between the voltage / current derivatives and normal operation data, mapping the deviation values ​​to the grid node locations using time-series indices, quantifying fault characteristics, and combining this with the grid topology to form a distribution map. Specifically, this involves quantifying time-domain / frequency-domain anomalies into computable matrix features and combining them with the grid topology to form a spatiotemporal distribution map, including the following steps:

[0098] S31: Dynamic deviation calculation. Using the i-th derivative of voltage / current (i = 1, 2, 3), corresponding to the rate of change and acceleration transient characteristics, the deviation from normal operating data is calculated. The formula is as follows:

[0099]

[0100] In the formula: F ij Let i be the i-th derivative, j be the dynamic deviation eigenvalue at time point j, i be the order of the derivative (i = 1, 2, 3), and j be the index of the time point and the sampling point. Let V(t) be the i-th time derivative of the voltage signal V(t), V norm (t) represents the voltage signal under normal operating conditions. Let I be the i-th time derivative of the current signal I(t), I norm (t) represents the current signal under normal operating conditions, S ij F is a normalization factor to eliminate the influence of dimensions. ij The higher the value, the higher the probability of failure;

[0101] S32: Matrix spatiotemporal mapping, using time series index j and power grid node location, to map F... ij The mapping is a spatiotemporal feature matrix, where rows represent the order of derivatives and columns represent time / nodes.

[0102] S4: Establish fault-environment correlation modeling based on dynamic graph neural networks. Inputs include node features containing electrical timing characteristics and environmental correlation, as well as equilateral features such as line impedance and insulation resistance corrected for temperature and humidity. A fault-environment correlation model is established through feature propagation and graph attention mechanisms. Based on the results of the above steps, fault-environment correlation modeling is established using feature propagation in a dynamic graph neural network (GNN). Input node features include F... ij Mapped to spatiotemporal feature matrix and environmental correlation The side characteristics include the temperature-corrected line impedance Z(T) = Z0(1+α) T (T-T0) and humidity-corrected insulation resistance Feature propagation is performed, and the feature propagation formula is:

[0103]

[0104] In the formula: The feature vector of node v in the l-th layer of the neural network includes the electrical timing features of node v. Temperature and humidity correlation at point v l represents the number of layers in the neural network, v represents a node in the graph structure, corresponding to a physical node in the power grid, u represents the neighboring nodes of node v, and W represents the number of nodes in the graph structure. (l) The node feature weight matrix of the l-th layer, Let h be the edge feature weight matrix of the l-th layer. e (v,u) represents the edge characteristics between nodes v and u, including the temperature-corrected line impedance Z(T) = Z0(1+α) T (T-T0) and humidity-corrected insulation resistance Where Z0 is the reference impedance, α T , where is the temperature coefficient, T is the current temperature, T0 is the reference temperature, R0 is the reference resistance, β is the humidity sensitivity coefficient, and H is the temperature coefficient. u Humidity; σ is the activation function;

[0105] α vu The attention weights for nodes v and u are calculated using the Graph Attention (GAT) mechanism, and the formula is as follows:

[0106]

[0107] Where: h e (v,u) represents the edge characteristics between nodes v and u, including the temperature-corrected line impedance Z(T) = Z0(1+α) T (T-T0) and humidity-corrected insulation resistance Where Z0 is the reference impedance, α T, where is the temperature coefficient, T is the current temperature, T0 is the reference temperature, R0 is the reference resistance, β is the humidity sensitivity coefficient, and H is the temperature coefficient. u For humidity; α vu Let be the attention weights between node v and its neighbor u, exp be the exponential function, LeakyReLU(·) be the linear rectified activation function with leakage, and a be the attention weights between node v and its neighbor u. T This represents the weight vector for the attention mechanism; This is a vector concatenation operation that concatenates the feature vectors of nodes v and u into a long vector. W is a shared feature transformation matrix used to unify the feature dimensions, and k is the index of the adjacent node.

[0108] S5: Traveling wave positioning dynamic calibration. First, coarse positioning of the two ends of the traveling wave is performed based on the time difference of the transient traveling wave arrival at both ends. Then, the coordinates of the fault point are verified to ensure that they are located on the actual connected line, thus ensuring positioning accuracy. This includes the following steps:

[0109] S51: Coarse positioning of two-terminal traveling wave. Based on the time difference Δt = t1 - t2 of the transient traveling wave arriving at both ends, the initial distance x is calculated using the following formula:

[0110]

[0111] In the formula: v is the initial velocity of the traveling wave, x is the estimated distance from the fault point to the measuring end, v is the initial propagation velocity of the traveling wave in the transmission line, Δt is the time difference between the transient traveling wave and the two ends of the line, t1 and t2 are the times when the traveling wave arrives at the nodes at both ends of the line, taking the absolute value, and L is the total length of the transmission line.

[0112] S52: Verify the location result. Input the corrected fault point coordinates into the power grid topology map to verify whether it is located on the actual connected line. If it does not meet the requirements, trigger a re-verification and return to step 2 to supplement the data.

[0113] The coarse localization of the dual-end traveling wave is based on feature-assisted fine-tuning using dynamic calculation of correction coefficients based on fault-environment correlation, and the formula is as follows:

[0114]

[0115] In the formula: γ is the dynamic correction coefficient, γ0 is the basic correction coefficient, and η T ,η H For environmental correlation weighting factors, The correlation between temperature, humidity and electrical quantities obtained in step S2.

Claims

1. A method for determining and locating power grid fault characteristics, characterized in that, Includes the following steps: S1: Multi-source heterogeneous data acquisition. Multimodal sensors are deployed at power grid nodes to simultaneously collect electrical quantities such as voltage, current, and frequency, as well as environmental quantities such as temperature and humidity, forming a raw time-series dataset stored according to timestamps. S2: Decoupling of data spatiotemporal features: Frequency domain decomposition of electrical quantities is performed by improving variational mode decomposition to separate fault characteristic frequencies and suppress harmonic interference. The correlation between environmental quantities and electrical quantities is calculated using grey relational analysis to extract effective fault features and suppress noise. S3: Fault feature matrix construction, calculate the dynamic deviation of voltage / current derivatives of each order with normal operation data, map the deviation values ​​to the grid node location according to the time series index to form a spatiotemporal feature matrix, quantify fault features and combine them with the grid topology to form a distribution map; S4: Establish fault-environment correlation modeling based on dynamic graph neural network. The input includes node features containing electrical timing features and environmental correlation degree, as well as equilateral features such as line impedance and insulation resistance corrected for temperature and humidity. The correlation model between fault and environment is established through feature propagation and graph attention mechanism. S5: Traveling wave positioning dynamic calibration. First, coarse positioning of the two ends of the traveling wave is performed based on the time difference of the transient traveling wave arriving at both ends, and the coordinates of the fault point are verified to ensure the positioning accuracy.

2. The method for determining and locating power grid fault characteristics according to claim 1, characterized in that, Step S1, multi-source heterogeneous data acquisition, specifically involves deploying multi-modal sensors at power grid nodes to simultaneously collect electrical and environmental quantities. The electrical quantities include voltage V(t), current I(t), and frequency f(t); the environmental quantities include temperature T(t) and humidity H(t). u (t); forming the original time series data {V(t),I(t),f(t),T(t),H u (t)}, stored according to timestamps.

3. The method for determining and locating power grid fault characteristics according to claim 1, characterized in that, Step S2, data spatiotemporal feature decoupling, specifically includes separating effective fault features from the collected raw time-series data, suppressing noise, and extracting physical laws, including frequency domain decomposition of electrical quantities and correlation analysis of environmental quantities, comprising the following steps: S21: Frequency domain decomposition of electrical quantities. An improved variational mode decomposition is performed on V(t), I(t), and f(t) to solve the variational constraint model as follows: In the formula, u k (t) represents the k-th frequency sub-signal, ω k Let be the center angular frequency of the k-th sub-signal, K be the total number of sub-signals obtained from the decomposition, δ(t) be the unit impulse function used for signal time-domain processing, j be the imaginary unit, π be the constant of pi, t be the time variable, and * be the convolution operation. The first-order partial derivative with respect to time t, The exponential factor for frequency domain transformation, used to convert a signal from the time domain to the frequency domain, L 2 Given the square of the norm and V(t) as the original voltage signal, K sub-signals of different frequencies, u, are adaptively separated. k (t), filtering out non-power frequency harmonic interference and retaining fault characteristic frequencies; S22: Correlation analysis of environmental quantities, for T(t), H u (t) Perform grey relational analysis, using voltage / current as the reference sequence x0(j), and calculate the temperature / humidity sequence x. T (j) correlation ρ T , The formula is as follows: In the formula: ρ is the grey relational degree, x0(j) is the reference sequence, x i (j) represents the comparison sequence, where i is the sequence index and x is the index. 环境 (j) represents a specific environmental quantity sequence, including temperature T(t) and humidity H. u (t), j are the time points or sampling point indices in the sequence, and ξ is the resolution coefficient, with a value range of [0,1]; the correlation between output temperature and electrical quantity ρ T Correlation between humidity and electrical quantity The larger the value, the more significant the impact of environmental factors on the fault.

4. The method for determining and locating power grid fault characteristics according to claim 1, characterized in that, Step S3, fault feature matrix construction, specifically involves quantifying time-domain / frequency-domain anomalies into computable matrix features and combining them with the power grid topology to form a spatiotemporal distribution map. This includes the following steps: S31: Dynamic deviation calculation. Using the i-th derivative of voltage / current (i = 1, 2, 3), corresponding to the rate of change and acceleration transient characteristics, the deviation from normal operating data is calculated. The formula is as follows: In the formula: F ij Let i be the i-th derivative, j be the dynamic deviation eigenvalue at time point j, i be the order of the derivative (i = 1, 2, 3), and j be the index of the time point and the sampling point. Let V(t) be the i-th time derivative of the voltage signal V(t), V norm (t) represents the voltage signal under normal operating conditions. Let I be the i-th time derivative of the current signal I(t), I norm (t) represents the current signal under normal operating conditions, S ij F is a normalization factor to eliminate the influence of dimensions. ij The higher the value, the higher the probability of failure; S32: Matrix spatiotemporal mapping, using time series index j and power grid node location, to map F... ij The mapping is a spatiotemporal feature matrix, where rows represent the order of derivatives and columns represent time / nodes.

5. The method for determining and locating power grid fault characteristics according to claim 1, characterized in that, Step S4, based on the results of the above steps, establishes a fault-environment correlation model, using feature propagation from a dynamic graph neural network (GNN). The input node features include F... ij Mapped to spatiotemporal feature matrix and environmental correlation ρ T , The side characteristics include the temperature-corrected line impedance Z(T) = Z0(1+α) T (T-T0) and humidity-corrected insulation resistance Perform feature propagation. Its feature propagation formula is: In the formula: The feature vector of node v in the l-th layer of the neural network includes the electrical timing features of node v. The temperature and humidity correlation ρ corresponding to point v T , l represents the number of layers in the neural network, v represents a node in the graph structure, corresponding to a physical node in the power grid, u represents the neighboring nodes of node v, and W represents the number of nodes in the graph structure. (l) The node feature weight matrix of the l-th layer, Let h be the edge feature weight matrix of the l-th layer. e (v,u) represents the edge characteristics between nodes v and u, including the temperature-corrected line impedance Z(T) = Z0(1+α) T (T-T0) and humidity-corrected insulation resistance Where Z0 is the reference impedance, α T , where is the temperature coefficient, T is the current temperature, T0 is the reference temperature, R0 is the reference resistance, β is the humidity sensitivity coefficient, and H is the temperature coefficient. u Humidity; σ is the activation function; α vu The attention weights for nodes v and u are calculated using the Graph Attention (GAT) mechanism, and the formula is as follows: Where: h e (v,u) represents the edge characteristics between nodes v and u, including the temperature-corrected line impedance Z(T) = Z0(1+α) T (T-T0) and humidity-corrected insulation resistance Where Z0 is the reference impedance, α T , where is the temperature coefficient, T is the current temperature, T0 is the reference temperature, R0 is the reference resistance, β is the humidity sensitivity coefficient, and H is the temperature coefficient. u For humidity; α vu Let be the attention weights between node v and its neighbor u, exp be the exponential function, LeakyReLU(·) be the linear rectified activation function with leakage, and a be the attention weights between node v and its neighbor u. T This represents the weight vector for the attention mechanism; This is a vector concatenation operation that concatenates the feature vectors of nodes v and u into a long vector. W is a shared feature transformation matrix used to unify the feature dimensions, and k is the index of the adjacent node.

6. The method for determining and locating power grid fault characteristics according to claim 1, characterized in that, Step S5, traveling wave positioning and dynamic calibration, combines fault characteristics and physical effects to locate the fault point, including the following steps: S51: Coarse positioning of two-terminal traveling wave. Based on the time difference Δt = t1 - t2 of the transient traveling wave arriving at both ends, the initial distance x is calculated using the following formula: In the formula: v is the initial velocity of the traveling wave, x is the estimated distance from the fault point to the measuring end, v is the initial propagation velocity of the traveling wave in the transmission line, Δt is the time difference between the transient traveling wave and the two ends of the line, t1 and t2 are the times when the traveling wave arrives at the nodes at both ends of the line, taking the absolute value, and L is the total length of the transmission line. S52: Verify the location result. Input the corrected fault point coordinates into the power grid topology map to verify whether it is located on the actual connected line. If it does not meet the requirements, trigger a re-verification and return to step 2 to supplement the data.

7. The method for determining and locating power grid fault characteristics according to claim 6, characterized in that, The coarse localization of the dual-end traveling wave is based on feature-assisted fine-tuning using dynamic calculation of correction coefficients based on fault-environment correlation, and the formula is as follows: In the formula: γ is the dynamic correction coefficient, γ0 is the basic correction coefficient, and η T ,η H For environmental correlation weighting factor, ρ T , The correlation between temperature, humidity and electrical quantities obtained in step S2.

8. A system for determining and locating power grid fault characteristics, characterized in that, The system includes a multi-source data acquisition module, a feature decoupling and analysis module, a fault feature modeling module, a fault-environment correlation modeling module, and a traveling wave localization and calibration module. The multi-source data acquisition module involves deploying multi-modal sensors at power grid nodes to simultaneously acquire electrical and environmental quantities, forming a raw time-series dataset. The feature decoupling and analysis module performs frequency domain decomposition on electrical quantities and correlation analysis on environmental quantities to separate effective fault features. The fault feature modeling module quantifies time-domain / frequency-domain anomalies into a spatiotemporal feature matrix and combines it with the power grid topology to form a distribution map. The fault-environment correlation modeling module uses a dynamic graph neural network to propagate features from node and edge features to establish a fault-environment correlation model. The traveling wave localization and calibration module uses the time difference between the arrival ends of the traveling wave for coarse localization, and then refines and verifies the localization results using a dynamic correction coefficient based on environmental correlation.

9. A terminal device, characterized in that, It includes a memory and a processor, wherein the memory stores a computer program that, when executed by the processor, implements the method of any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The storage medium stores a computer program that, when executed by a processor, implements the method of any one of claims 1-7.

Images