Medium-voltage distribution network line fault positioning method and intelligent switch

Abstract

The application provides a medium-voltage distribution network line fault positioning method and a smart switch, and belongs to the technical field of distribution networks. The method comprises the following steps: performing Hilbert-Huang transformation on a distribution network fault current signal connected with a distributed energy source to obtain the instantaneous frequency and instantaneous amplitude of a key component signal; performing multidimensional signal phase space reconstruction based on the instantaneous frequency and instantaneous amplitude of the key component signal to obtain a phase space reconstruction signal; using a convolutional neural network to extract the characteristic quantity of the time sequence of the distribution network fault current signal and the characteristic quantity of the phase space reconstruction signal, mixing them, and determining the fault occurrence position according to the mixed characteristic quantity. The application firstly uses Hilbert-Huang transformation to mine the fault current signal and its time-frequency characteristics in the medium-voltage distribution network containing distributed resources, then uses a phase space reconstruction method to capture the internal law of the fault signal, and finally realizes fault line selection and accurate positioning of the fault in the medium-voltage distribution network containing distributed resources through a convolutional neural network.

Description

Technical Field

[0001] This invention relates to the field of power distribution network technology, specifically to a method for locating faults in medium-voltage power distribution lines and a smart switch. Background Technology

[0002] The integration of large-scale distributed generation (DG) resources significantly increases the uncontrollability, volatility, and randomness of medium-voltage distribution networks, transforming them from traditional radial networks into complex grids with multiple power sources, and changing power flow from unidirectional to bidirectional. The dynamic switching and random power output of DG resources make the current flow in traditional power grids more complex and variable. Traditional traveling wave ranging and fault location methods become ineffective, leading to damage to electrical equipment, disruption of normal power supply and distribution, and even production safety accidents. Therefore, it is necessary to research rapid and accurate fault location methods suitable for large-scale DG-integrated distribution networks.

[0003] Fault location in distribution networks is mainly divided into three categories: fault line selection, fault distance measurement, and section location. Fault line selection locates the faulty line based on reported fault information, quickly isolates the fault point, restores power supply to non-faulty areas, and prevents further escalation of the accident. Fault distance measurement mainly determines the location of the fault based on the relationship between various electrical quantities at the time of the fault. Existing fault distance measurement methods mainly include impedance methods, traveling wave methods, and signal injection methods. One traveling wave method combines frequency domain reflection and inverse fast Fourier transform to estimate the precise location of the fault point in the cable. Another traveling wave method identifies the fault interval based on the arrival time series of voltage traveling waves, and then uses the double-ended method to achieve precise fault location within the determined interval. Yet another method is based on traveling wave characteristics and reference measurement points, calculating the distance from each fault point to the reference point, and determining the fault point location by selecting the maximum value. However, none of the above-mentioned distribution network fault location methods consider the situation where the distribution network contains highly pervasive distributed resources, making it impossible to achieve accurate location of faults in distribution network lines with large-scale distributed resource access.

[0004] As core edge-side equipment in distribution networks, smart switches integrate control, metering, and communication functions, enabling them to better perform line interruption tasks, improve interruption reliability, and achieve intelligent operation of circuit breakers. Smart switches can replace traditional circuit breakers, contactors, thermal relays, fuses, etc., and can achieve fault isolation of distribution network sections and grid connection after fault recovery. Therefore, it is necessary to research smart switches suitable for rapid and accurate fault location, fault isolation, and fault recovery in large-scale distributed resource access distribution networks. Summary of the Invention

[0005] To address the aforementioned technical deficiencies, this invention provides a method for locating faults in medium-voltage distribution network lines and a smart switch.

[0006] This invention provides a method for locating faults in medium-voltage distribution network lines, comprising:

[0007] The fault current signal of the distribution network connected to the distributed energy source is subjected to Hilbert-Huang transform to obtain the instantaneous frequency and instantaneous amplitude of multiple key component signals of the fault current signal.

[0008] Multidimensional signal phase space reconstruction is performed based on the instantaneous frequency and instantaneous amplitude of multiple key component signals to obtain a phase space reconstructed signal; wherein, the delay time of the signal phase space reconstruction is determined according to the cross-correlation function of the instantaneous frequency and instantaneous amplitude of multiple key component signals;

[0009] The feature quantities of the time series of the distribution network fault current signal and the feature quantities of the phase space reconstruction signal are extracted by using a convolutional neural network. The feature quantities of the time series of the distribution network fault current signal and the feature quantities of the phase space reconstruction signal are mixed, and the fault location is determined based on the mixed feature quantities.

[0010] In this embodiment of the invention, the step of performing a Hilbert-Huang transform on the distribution network fault current signal connected to distributed energy sources to obtain the instantaneous frequency and instantaneous amplitude of multiple key component signals of the distribution network fault current signal includes: decomposing the distribution network fault current signal connected to distributed energy sources into multiple different frequency components, and extracting multiple key component signals from the multiple different frequency components; performing a Hilbert transform on the extracted multiple key component signals respectively to obtain the instantaneous frequency and instantaneous amplitude of the multiple key component signals.

[0011] In this embodiment of the invention, performing Hilbert transform on multiple key component signals to obtain the instantaneous frequency and instantaneous amplitude of the multiple key component signals includes:

[0012] A complex signal is obtained by performing a Hilbert transform on each key component signal. This complex signal is composed of the key component signal and the Hilbert transform signal of the key component signal.

[0013] Based on the key component signal and its Hilbert transform, the instantaneous amplitude and instantaneous phase of the key component signal are calculated, and the instantaneous frequency of the key component signal is calculated based on its instantaneous phase.

[0014] In this embodiment of the invention, the multi-dimensional signal phase space reconstruction based on the instantaneous frequency and instantaneous amplitude of multiple key component signals includes: performing three-dimensional signal phase space reconstruction based on the instantaneous frequency of each key component signal to obtain a three-dimensional spatial signal vector; and performing three-dimensional signal phase space reconstruction based on the instantaneous amplitude of each key component signal to obtain a four-dimensional spatial signal vector.

[0015] In this embodiment of the invention, the delay time for phase space reconstruction of the signal is determined based on the cross-correlation function of the instantaneous frequency and instantaneous amplitude of multiple key component signals, including: calculating the mutual information between the original key component signal and the delayed key component signal based on the cross-correlation function of the instantaneous frequency and instantaneous amplitude of each key component signal, and selecting the time point at which the first local minimum value is reached as the delay time of the phase space reconstruction signal.

[0016] In this embodiment of the invention, the formula for calculating the mutual information between the original key component signal and the delayed key component signal is as follows:

[0017]

[0018] Where I(τ) is mutual information, x(t) is the original critical component signal, x(t+τ) is the critical component signal after a delay of τ, P(x(t)) is the probability distribution of the original critical component signal x(t), P(x(t+τ)) is the probability distribution of the critical component signal x(t+τ) after a delay of τ, and P(x(t),x(t+τ)) is the joint probability distribution of x(t) and x(t+τ).

[0019] In this embodiment of the invention, the step of extracting the time series features of the distribution network fault current signal and the phase space reconstructed signal using a convolutional neural network includes: mining the time series features of the distribution network fault current signal using a one-dimensional convolutional neural network; and extracting the features of the three-dimensional spatial signal vector and the four-dimensional spatial signal vector using a point voxel convolutional neural network.

[0020] In this embodiment of the invention, the feature quantities of the time series of the distribution network fault current signal are mixed with the feature quantities of the phase space reconstruction signal, and the fault location is determined based on the mixed feature quantities. This includes: mixing the feature quantities of the time series of the distribution network fault current signal with the feature quantities of the three-dimensional spatial signal vector, and performing fault line selection based on the mixed feature quantities to determine the line where the fault occurred; and mixing the feature quantities of the time series of the distribution network fault current signal with the feature quantities of the four-dimensional spatial signal vector, and performing fault distance measurement based on the mixed feature quantities to determine the fault location.

[0021] Another aspect of the present invention provides a smart switch, comprising: a data acquisition module and an edge computing module;

[0022] The edge computing module includes: a fault segment location module and a fault ranging module;

[0023] The fault section location module is used to perform preliminary location of the fault section based on the fault current signal of the distribution network connected to distributed energy collected by the acquisition module.

[0024] The fault location module is used to locate the fault location within the initially identified fault section, specifically including:

[0025] The fault current signal of the distribution network connected to the distributed energy source is subjected to Hilbert-Huang transform to obtain the instantaneous frequency and instantaneous amplitude of multiple key component signals of the fault current signal.

[0026] Multidimensional signal phase space reconstruction is performed based on the instantaneous frequency and instantaneous amplitude of multiple key component signals to obtain a phase space reconstructed signal; wherein, the delay time of the signal phase space reconstruction is determined according to the cross-correlation function of the instantaneous frequency and instantaneous amplitude of multiple key component signals;

[0027] The feature quantities of the time series of the distribution network fault current signal and the feature quantities of the phase space reconstruction signal are extracted by using a convolutional neural network. The feature quantities of the time series of the distribution network fault current signal and the feature quantities of the phase space reconstruction signal are mixed, and the fault location is determined based on the mixed feature quantities.

[0028] In this embodiment of the invention, signal phase space reconstruction based on the instantaneous frequency and instantaneous amplitude of multiple key component signals includes: three-dimensional signal phase space reconstruction based on the instantaneous frequency of each key component signal to obtain a three-dimensional spatial signal vector; and three-dimensional signal phase space reconstruction based on the instantaneous amplitude of each key component signal to obtain a four-dimensional spatial signal vector.

[0029] In this embodiment of the invention, the fault location module is specifically used for:

[0030] One-dimensional convolutional neural networks are used to mine the time series features of the distribution network fault current signal.

[0031] The feature quantities of the three-dimensional spatial signal vector and the feature quantities of the four-dimensional spatial signal vector are extracted using a point voxel convolutional neural network.

[0032] The feature quantities of the time series of the distribution network fault current signal are mixed with the feature quantities of the three-dimensional spatial signal vector, and the fault line selection is performed based on the mixed feature quantities to determine the line where the fault occurred.

[0033] The feature quantities of the time series of the distribution network fault current signal are mixed with the feature quantities of the four-dimensional spatial signal vector, and the fault location is determined by fault ranging based on the mixed feature quantities.

[0034] In this embodiment of the invention, the smart switch further includes: a strategy library module and a control module;

[0035] The strategy library module is used to receive the fault location and fault information sent by the edge computing module, and search for the recovery strategy corresponding to the current fault in the control strategy library of the strategy library module.

[0036] The control module is used to receive fault segment location information sent by the fault segment location module and recovery strategy and fault location sent by the strategy library module, perform isolation operation according to the fault segment location information, and restore the line at the fault location according to the recovery strategy.

[0037] In this embodiment of the invention, the smart switch further includes: a remote communication module;

[0038] The remote communication module is used to communicate with the remote master station. When no recovery strategy corresponding to the current fault is found in the strategy library module, the fault location and fault information are uploaded to the remote master station, the recovery strategy issued by the remote master station is received, and the recovery strategy issued by the remote master station is sent to the strategy library module.

[0039] The policy library module is also used to send the recovery policy issued by the remote master station to the control module;

[0040] The control module is also used to isolate the faulty section corresponding to the location of the fault according to the recovery strategy issued by the remote master station.

[0041] In this embodiment of the invention, the smart switch further includes: a fault recording module;

[0042] The fault recording module is used to record fault information, fault location, and fault handling method, and update it to the strategy library module;

[0043] The remote communication module is also used to upload the fault information, fault location, and fault handling method updated by the policy library module to the remote master station, so as to update the fault recovery policy stored in the remote master station.

[0044] In this embodiment of the invention, the smart switch further includes a self-testing and self-recovery module, which is used to perform fault self-testing and self-recovery on the smart switch.

[0045] This invention first utilizes the Hilbert-Huang transform technique to fully mine the fault current signal and its time-frequency characteristics in medium-voltage distribution networks containing distributed resources. Then, it employs a phase space reconstruction method to capture the inherent patterns of the fault signal. Furthermore, it uses a convolutional neural network to achieve fault line selection and precise fault location in medium-voltage distribution networks with distributed resources. This invention also proposes a smart switch with an embedded fault location module. By installing the smart switch in the line, it achieves automated restoration and grid connection of the faulty line, eliminating the need for additional equipment and saving costs.

[0046] Other features and advantages of the technical solution of the present invention will be described in detail in the following detailed embodiments section. Attached Figure Description

[0047] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this invention, illustrate exemplary embodiments of the invention and are used to explain the invention, but do not constitute an undue limitation of the invention. In the drawings:

[0048] Figure 1 This is a flowchart of the medium-voltage distribution network line fault location method provided in the embodiments of the present invention;

[0049] Figure 2 This is an architecture diagram of fault line selection and fault location based on convolutional neural networks provided in an embodiment of the present invention;

[0050] Figure 3 This is a block diagram of the smart switch provided in an embodiment of the present invention;

[0051] Figure 4 This is a flowchart of the power distribution line fault recovery process provided in an embodiment of the present invention. Detailed Implementation

[0052] To make the technical solutions and advantages of the embodiments of the present invention clearer, the exemplary embodiments of the present invention will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not an exhaustive list of all embodiments. It should be noted that, unless otherwise specified, the embodiments and features in the embodiments of the present invention can be combined with each other.

[0053] As described in the background section, distribution network fault location mainly falls into three categories: fault line selection, fault distance measurement, and section location. Fault line selection locates the faulty line based on reported fault information, quickly isolates the fault point, and restores power supply to non-faulty areas. Fault distance measurement primarily determines the fault location based on the relationship between various electrical quantities at the time of the fault. Existing distribution network fault location methods do not consider situations where the distribution network contains highly permeable distributed resources, making it impossible to accurately locate faults in distribution network lines with large-scale distributed resource access. Early on, distribution network fault location relied on fault indicators installed on the distribution network and branch lines; a change in the color of the fault indicator indicated a fault in a downstream area. With the increasing maturity of distribution network automation, distribution switch monitoring terminals (FTUs) are widely installed at feeder circuit breakers and automatic switches, requiring the cooperation of FTUs and fault indicators for fault location, resulting in high installation costs.

[0054] To address the aforementioned problems, this invention proposes a fault location method for medium-voltage distribution networks containing distributed energy resources. First, it utilizes Hilbert-Huang transform technology to fully mine the fault current signal and its time-frequency characteristics in medium-voltage distribution networks with distributed resources. Second, it employs phase space reconstruction to capture the inherent patterns of the fault signal. Finally, it uses a convolutional neural network to achieve fault line selection and accurate fault location in medium-voltage distribution networks with distributed resources. This invention also proposes a smart switch with an embedded fault location module. This smart switch, installed in the line, enables automated restoration and grid connection of the faulty line, achieving fault recovery without the need for additional equipment, thus saving costs.

[0055] Figure 1 This is a flowchart of the medium-voltage distribution network line fault location method provided in an embodiment of the present invention. Figure 1 As shown, the medium-voltage distribution network line fault location method provided in this embodiment includes the following steps:

[0056] S100 performs Hilbert-Huang transform on the distribution network fault current signal connected to distributed energy sources to obtain the instantaneous frequency and instantaneous amplitude of multiple key component signals of the distribution network fault current signal.

[0057] S200 performs multi-dimensional signal phase space reconstruction based on the instantaneous frequency and instantaneous amplitude of multiple key component signals to obtain the phase space reconstructed signal;

[0058] S300 uses a convolutional neural network to extract the feature quantities of the time series of the distribution network fault current signal and the feature quantities of the phase space reconstruction signal. It then mixes the feature quantities of the time series of the distribution network fault current signal and the feature quantities of the phase space reconstruction signal to determine the location of the fault based on the mixed feature quantities.

[0059] Before step S100 above, for the distribution network connected to distributed energy, the distribution network fault current signal under different topologies, different fault types, different lines, and different noise interference conditions is collected. The sampling frequency is 10kHz. The distribution network fault current signal of the two cycles after the fault is used as the original fault sample signal for fault line selection.

[0060] The Hilbert-Huang Transhorm (HHT) is a time-frequency analysis method suitable for non-stationary and nonlinear signals. It mainly consists of two steps: Empirical Mode Decomposition (EMD) and Hilbert Spectral Analysis (HSA). Compared to traditional Fourier Transform or wavelet Transform, the HHT offers better adaptability and higher accuracy in handling complex nonlinear signals.

[0061] When faults occur in traditional distribution networks, such as single-phase grounding faults, phase-to-phase short-circuit faults, and two-phase grounding faults, different fault types will generate different fault currents in the distribution lines. However, with the high proportion of new energy sources added to the distribution network, distributed power sources of different capacities, states, and grid connection locations have an unknown impact on the waveform characteristics of traditional fault currents. Therefore, it is necessary to first extract the key component signals from the original fault current signals of distribution networks containing distributed energy sources through empirical mode decomposition (EMD), and then perform Hilbert spectrum analysis on each key component signal.

[0062] In step S100 above, the fault current signal of the distribution network connected to the distributed energy source (i.e., the original fault sample signal) is first decomposed into multiple different frequency components (IMFs) and a residual component using the Empirical Mode Decomposition (EMD) algorithm. Each IMF component represents a different frequency component of the original fault sample signal. Then, multiple key component signals are extracted from the multiple different frequency components, and Hilbert transforms are performed on the extracted key component signals to obtain the instantaneous frequency and instantaneous amplitude of the multiple key component signals. Specifically, performing a Hilbert transform on each key component signal yields a complex signal, which is composed of the key component signal and its Hilbert transform signal. Based on the key component signal and its Hilbert transform signal, the instantaneous amplitude and instantaneous phase of the key component signal are calculated, and the instantaneous frequency of the key component signal is calculated based on its instantaneous phase.

[0063] In a specific example, the EMD algorithm flow is as follows:

[0064] Step 1: Assuming the original signal of the fault current received by the smart switch is x(t), use spline interpolation to fit the maximum and minimum points of the original signal to form the upper and lower envelopes, and calculate the average envelope x. o (t), then subtract the mean from the original signal to obtain a new signal h1(t), which is expressed as: h1(t) = x(t) - x o (t);

[0065] Step 2: Determine whether signal h1(t) satisfies the characteristic conditions of the Intrinsic Mode Function (IMF). If it does, signal h1(t) is the first IMF component IMF1. If it does not, continue to repeat Step 1 until h1(t) satisfies the characteristic conditions of the Intrinsic Mode Function (IMF).

[0066] Step 3: Separate the IMF1 component from the original signal x(t) to obtain a new signal r1(t). The expression for signal r1(t) is: r1(t) = x(t) - IMF1. Repeat this process until the second IMF component IMF2 is obtained.

[0067] Step four, repeat the above steps until the Kth IMF component is obtained. Finally, the result of the empirical mode decomposition can be expressed as:

[0068]

[0069] Where, r k The remainder represents the central trend of the signal.

[0070] Ultimately, the original signal is decomposed into a series of IMF components and a residual component, with each IMF component representing a different frequency component of the signal. For example, after EMD decomposition, the original distribution network fault current signal yields the intrinsic mode functions of the original distribution network fault current signal, which is the time-frequency data matrix of the original signal, including its time-frequency characteristics.

[0071] In a specific example, during Hilbert spectral analysis (HSA), a complex signal z(t) is obtained by performing a Hilbert transform on the IMF component signal x(t):

[0072]

[0073] in, It is the Hilbert transform of x(t);

[0074] The expression for calculating the instantaneous amplitude A(t) of the component signal x(t) is:

[0075]

[0076] Instantaneous phase of component signal x(t) The calculation expression is:

[0077]

[0078] The expression for calculating the instantaneous frequency f(t) of the component signal x(t) is:

[0079]

[0080] In step S200 above, multi-dimensional signal phase space reconstruction is performed based on the instantaneous frequency and instantaneous amplitude of each key component (IMF) signal. For signal phase space reconstruction, the reconstruction dimension and delay time need to be determined. The reconstruction dimension determines the dimension of the phase space; an automatic information dimension algorithm can be used to select a suitable dimension to capture the dynamic characteristics of the signal. The delay time determines how to extract time series values ​​from the signal. The choice of delay time is crucial; if the delay is too small, it may lead to redundancy in the phase space reconstruction trajectory; if the delay is too large, the dynamic characteristics of the signal may not be accurately represented. In this embodiment, the delay time for signal phase space reconstruction is determined based on the cross-correlation function of the instantaneous frequency and instantaneous amplitude of multiple key component signals. Based on the dimension of the signal phase space reconstruction and the delay time of the phase space reconstruction signal, a delay embedding method is used to transform multiple key component signals into trajectories in a high-dimensional phase space.

[0081] Since instantaneous frequency can reflect the frequency changes caused by fault events (such as short circuits and ground faults), this invention extracts the instantaneous frequencies of intrinsic mode components (IMFs) using the Hilbert-Huang transform, constructs a three-dimensional spatial signal vector of the instantaneous frequency of each IMF component, and constructs a four-dimensional spatial signal vector of the energy of the IMF component signal. For fault line selection, the cross-correlation function (CCF) of the instantaneous frequency and instantaneous amplitude of the IMF component is calculated; for fault location, the auto-correlation function (ACF) of the energy of the IMF component is calculated to determine the delay time.

[0082] In one embodiment, three-dimensional images of the instantaneous frequency and instantaneous amplitude of the IMF component signals are constructed for the fault location process, and a four-dimensional image of the energy of the IMF component signals is also constructed. Specifically, a three-dimensional spatial signal vector is constructed based on the instantaneous frequency of each key component signal, and a three-dimensional spatial signal vector is constructed based on the instantaneous amplitude of each key component signal; simultaneously, a four-dimensional spatial signal vector is constructed based on the instantaneous amplitude (energy) of each key component signal (signal amplitude is used as the fourth-dimensional signal parameter). Based on the cross-correlation function of the instantaneous frequency and instantaneous amplitude of each key component signal, the mutual information between the original key component signal and the delayed key component signal is calculated, and the time point at which the first local minimum value is reached is selected as the delay time of the phase space reconstruction signal.

[0083] Let f(t) be the instantaneous frequency signal of the fault current signal after Hilbert-Huang transform (HHT). The process of signal phase space reconstruction is as follows:

[0084] F(t)=[f(t), f(t+τ), f(t+2τ),..., f(t+(m-1)τ)];

[0085] Where F(t) is the reconstructed vector containing the signal's delay coordinates; m is the embedding dimension, which is 2 when the reconstructed signal is three-dimensional; and τ is the delay time.

[0086] At the moment of a fault, the amplitude change is usually significant. Capturing the change in signal strength over time helps identify peaks in the current. The instantaneous amplitude of each IMF is calculated using Hilbert transform, and the three-dimensional spatial signal of the instantaneous amplitude of each IMF is reconstructed. Let the instantaneous amplitude of the fault current signal after HHT transform be A(t). The signal phase space reconstruction process is as follows:

[0087] A'(t)=[A(t), A(t+τ), A(t+2τ),...,A(t+(m-1)τ)];

[0088] Where A'(t) is the reconstructed vector, m is the embedding dimension, and τ is the delay time.

[0089] The method for selecting the delay time τ is as follows:

[0090] Calculate the mutual information between the critical component signal and the delayed critical component signal:

[0091]

[0092] Where I(τ) is mutual information, x(t) is the original critical component signal, x(t+τ) is the critical component signal after a delay of τ, P(x(t)) is the probability distribution of the original critical component signal x(t), P(x(t+τ)) is the probability distribution of the critical component signal x(t+τ) after a delay of τ, and P(x(t),x(t+τ)) is the joint probability distribution of x(t) and x(t+τ).

[0093] Calculate the values ​​of I(τ1), I(τ2), I(τ3), etc. corresponding to P(x(t)) and P(x(t+2τ)), P(x(t+τ)) and P(x(t+2τ)), P(x(t+2τ)) and P(x(t+3τ)) respectively. The delay time τ corresponding to the minimum value I(τ) is the time point when the local minimum is first reached.

[0094] In step S300 above, a convolutional neural network is used to extract the time-series features of the distribution network fault current signal and the features of the phase space reconstructed signal. (Refer to...) Figure 2 This study utilizes a one-dimensional convolutional neural network (1D-CNN) to mine the time-series features of distribution network fault current signals, and a point-voxel convolutional neural network (PVCNN) to extract the features of three-dimensional and four-dimensional spatial signal vectors. The features of the time-series fault current signals and the features of the three-dimensional spatial signal vectors are then combined to perform fault line selection based on the combined features, thus identifying the faulty line. Conversely, the features of the time-series fault current signals and the features of the four-dimensional spatial signal vectors are combined to perform fault location measurement based on the combined features, thus determining the fault location.

[0095] In a specific embodiment, a one-dimensional convolutional neural network (1D-CNN) includes: a one-dimensional convolutional layer, a normalization layer, an activation function layer, a presentation layer, and a fully connected layer. The process of extracting three-dimensional features using a point-voxel convolutional neural network (PVCNN) includes: inputting a three-dimensional spatial signal vector into the PVCNN network, with a data size of n×3, where n represents n points, each represented by three-dimensional coordinates (x, y, z); feature extraction layers (PV Conv Layers) extract local features from the data; each layer extracts local features of points through Pointwise Convolution (PV Conv) in PointNet; and feature extraction layers (PV Conv Layers) form a high-dimensional feature space by increasing the input dimension. The process involves four layers: the first layer (n×3→n×64) extracts initial features using PointNet's Pointwise Convolution; the second layer (n×64→n×128) extracts higher-dimensional features; the third layer (n×128→n×128) further enriches the local features of the points; and the fourth layer (n×128→n×512) extracts higher-order point features. Then, a fully connected layer (FC) maps the point features (n×512→n×2048), and max pooling aggregates all point features into a single global feature (1×2048). This global feature represents the semantic information of the entire point cloud.

[0096] For fault line selection, the three-dimensional features extracted by the one-dimensional convolutional neural network (1D-CNN) and the point voxel convolutional neural network (PVCNN) are mixed. The mixed features are used as the input of the classifier Sigmoid, and the probability of the fault line is output in the form of probability, so as to realize the selection of the fault line.

[0097] For fault location, the energy distribution across different frequency bands of a signal can be understood by calculating the energy of the IMF components (e.g., through squared amplitude calculation). During a fault, energy may be concentrated in certain frequency bands, and energy analysis can identify the characteristics of the fault mode. By analyzing changes in instantaneous amplitude or energy, the timing and duration of the fault can be determined, which can be used for fault location on faulty lines.

[0098] Let the energy of the fault current after HHT transformation be E(t). The signal reconstruction process is: E'(t)=[E(t), E(t+τ), E(t+2τ), ...,E(t+(m-1)τ)], constructing a four-dimensional spatial signal of the signal energy, and selecting the amplitude of the energy signal as the fourth-dimensional signal parameter.

[0099] Four-dimensional feature extraction: Four-dimensional signal features can be represented as n×(x,y,z,e), where n represents the number of signal sampling points; x,y,z represent the first three dimensions of spatial coordinates; and e is the fourth dimension, representing the signal energy. The energy e in n×4 is directly used as one of the initial input dimensions of convolutional or fully connected layers. Feature fusion: Energy information and geometric information (x,y,z) are jointly processed to calculate the joint features of the energy distribution. The input data has an n×4 shape, with each point containing four-dimensional information. Local feature extraction (PV Conv Layers) is performed on the data. Each layer extracts local features of points through Pointwise Convolution (PVConv) in PointNet. The feature extraction layers (PV Conv Layers) form a high-dimensional feature space by increasing the input dimension. The process consists of four layers: the first layer (n×4→n×64) extracts initial features using PointNet's Pointwise Convolution; the second layer (n×64→n×128) extracts higher-dimensional features; the third layer (n×128→n×128) further enriches the local features of the points; and the fourth layer (n×128→n×512) extracts higher-order point features. Fully connected layers (FC) are used to map the point features: n×512→n×2048. Max pooling is then used to aggregate the features of all points into a single global feature: 1×2048.

[0100] The four-dimensional features extracted by the one-dimensional convolutional neural network (1D-CNN) and the point voxel convolutional neural network (PVCNN) are mixed. The mixed features are then processed by the CNN to map the relationship between the features and the distance to the fault point. The mapping result is used as the input of the classifier Sigmoid, and its accuracy is calculated. The distance from the fault location to the smart switch is mapped to the corresponding probability. The distance with the highest accuracy is output separately. In order to ensure that no possible fault location information is missed, the fault point location with an accuracy greater than a certain threshold is also output as a reference to realize fault ranging.

[0101] This invention also provides a smart switch, which has on-grid and off-grid functions and edge computing functions, as well as fault detection functions. It can perform preliminary section location of medium-voltage line faults connected to the smart switch, and based on this, realize fault line selection and accurate fault distance measurement.

[0102] like Figure 3 As shown, the smart switch includes: a data acquisition module, an edge computing module, a policy library module, a control module, and a remote communication module.

[0103] The edge computing module includes a fault section location module and a fault ranging module. The fault section location module is used to perform preliminary fault section location based on the fault current signal of the distribution network connected to distributed energy sources collected by the acquisition module; the fault ranging module is used to perform fault ranging within the initially located fault section to locate the fault location.

[0104] The strategy library module receives the fault location and fault information sent by the edge computing module, searches for the recovery strategy corresponding to the current fault in the control strategy library of the strategy library module; it stores historical control strategies issued by the remote master station in the control strategy library, and first searches for the recovery strategy that matches the current fault in the control strategy library when a fault occurs.

[0105] The control module receives fault location information from the fault location module and recovery strategies and fault location from the strategy library module. It performs isolation operations (opening and closing operations of the fault section) based on the fault location information and restores the line at the fault location according to the recovery strategy.

[0106] The remote communication module is used to communicate with a remote master station (e.g., a remote intelligent fusion terminal). When no recovery strategy corresponding to the current fault is found in the strategy library module, the module uploads the fault location and fault information to the remote master station, receives the recovery strategy issued by the remote master station, and sends the recovery strategy issued by the remote master station to the strategy library module.

[0107] The policy library module is also used to send recovery policies issued by the remote master station to the control module. The control module then isolates the faulty section corresponding to the location of the fault according to the recovery policies issued by the remote master station.

[0108] The edge computing module of the smart switch in this embodiment of the invention uses the medium-voltage distribution network line fault location method provided in the above embodiment for fault selection and fault location. Specifically, the fault location module performs Hilbert-Huang transform on the distribution network fault current signal connected to distributed energy sources to obtain the instantaneous frequency and instantaneous amplitude of multiple key component signals of the distribution network fault current signal; based on the instantaneous frequency and instantaneous amplitude of multiple key component signals, multi-dimensional signal phase space reconstruction is performed to obtain the phase space reconstructed signal; the feature quantities of the time series of the distribution network fault current signal and the feature quantities of the phase space reconstructed signal are extracted using a convolutional neural network, and the feature quantities of the time series of the distribution network fault current signal and the feature quantities of the phase space reconstructed signal are mixed, and the fault location is determined according to the mixed feature quantities.

[0109] In a specific embodiment, the fault location module is pre-configured with a Hilbert-Huang transform unit, a signal reconstruction unit, and a convolutional neural network model. First, the Hilbert-Huang transform unit decomposes the incoming distribution network fault current signal into multiple different frequency components, and extracts multiple key component signals from these components. Hilbert transforms are then performed on each of these key component signals to obtain their instantaneous frequency and amplitude. Next, the signal reconstruction unit performs three-dimensional signal phase space reconstruction based on the instantaneous frequency of each key component signal to obtain a three-dimensional signal vector. Similarly, it performs three-dimensional signal phase space reconstruction based on the instantaneous amplitude of each key component signal to obtain a four-dimensional signal vector. Finally, based on the dimension of the reconstructed signal phase space and the delay time of the reconstructed signal, a delay embedding method is used to transform the multiple key component signals into trajectories in a high-dimensional phase space.

[0110] The delay time for signal phase space reconstruction can be determined based on the cross-correlation function of the instantaneous frequencies and instantaneous amplitudes of multiple key component signals. Specifically, the mutual information between the key component signals and the delayed key component signals is calculated:

[0111]

[0112] Where I(τ) is mutual information, x(t) is the original critical component signal, x(t+τ) is the critical component signal after a delay of τ, P(x(t)) is the probability distribution of the original critical component signal x(t), P(x(t+τ)) is the probability distribution of the critical component signal x(t+τ) after a delay of τ, and P(x(t),x(t+τ)) is the joint probability distribution of x(t) and x(t+τ).

[0113] Calculate the values ​​of I(τ1), I(τ2), I(τ3), etc. corresponding to P(x(t)) and P(x(t+2τ)), P(x(t+τ)) and P(x(t+2τ)), P(x(t+2τ)) and P(x(t+3τ)) respectively. The delay time τ corresponding to the minimum value I(τ) is the time point when the local minimum is first reached.

[0114] In a specific embodiment, the fault location module utilizes a convolutional neural network model to extract the time-series features of the distribution network fault current signal and the features of the phase space reconstructed signal. Specifically, a one-dimensional convolutional neural network is used to mine the time-series features of the distribution network fault current signal, and a point-voxel convolutional neural network is used to extract the features of the three-dimensional spatial signal vector and the four-dimensional spatial signal vector. Then, the features of the time-series of the distribution network fault current signal and the features of the three-dimensional spatial signal vector are mixed, and fault line selection is performed based on the mixed features to determine the line where the fault occurred; the features of the time-series of the distribution network fault current signal and the features of the four-dimensional spatial signal vector are mixed, and fault location is performed based on the mixed features to determine the location of the fault.

[0115] In one embodiment, the smart switch further includes a fault recording module. The fault recording module records fault information, fault location, and fault handling methods, and packages these information together with the fault information to be updated in the policy library module. The remote communication module then uploads the updated fault information, fault location, and fault handling methods from the policy library module to the remote master station to update the fault recovery policy stored on the remote master station.

[0116] In one embodiment, the smart switch further includes a self-testing and self-recovery module and a power supply module. The self-testing and self-recovery module is used to detect and handle faults in the smart switch, achieving fault self-testing and self-recovery. The power supply module is used to provide overall power to the smart switch.

[0117] This invention, through direct application to smart switches installed in power lines, eliminates the need for knowledge of the distribution network topology, focusing only on line information. First, it fully exploits the fault current signal and its time-frequency characteristics in medium-voltage distribution networks containing distributed resources using Hilbert-Huang transform technology. Second, it employs phase space reconstruction to capture the inherent patterns of the fault signal. Finally, it utilizes convolutional neural networks to achieve fault line selection and accurate fault location in medium-voltage distribution networks containing distributed resources.

[0118] The smart switch of this invention features both grid-connected and off-grid functionality, edge computing capabilities, and fault detection, reducing the cost of pairing it with feeder equipment. This smart switch enables precise fault location in medium-voltage distribution networks with high penetration rates of distributed energy. Through intelligent algorithms (specifically referring to the medium-voltage distribution network line fault location method described above), it ensures both accuracy and efficiency in fault location. Furthermore, it can achieve automated fault recovery and grid connection of faulty lines through edge-end collaboration, eliminating the need for additional equipment, thus saving costs and reducing grid losses caused by line faults.

[0119] Reference Figure 4 The fault recovery process for medium-voltage lines based on the aforementioned smart switches and achieving edge-end coordination is as follows:

[0120] Initial stage: The acquisition module sends the acquired distribution network fault current signal to the edge computing module, which then extracts the line information features;

[0121] Fault section location: The fault section location module performs preliminary location of the fault section based on the distribution network fault current signal, and the control module isolates the fault section based on the fault section location result;

[0122] The edge computing module's processing: The fault location module performs fault location within the initially located fault section, locates the fault location (i.e., the distance to the faulty line), and sends the fault location and fault information (fault type, etc.) to the policy library module.

[0123] Execution of fault recovery strategy: The fault location module uses reinforcement learning to locate the faulty line based on the faulty section; it searches the control strategy library of the strategy library module for a recovery strategy corresponding to the current fault (checking if there is a reported fault-corresponding strategy): If a recovery strategy corresponding to the current fault exists in the control strategy library, the recovery strategy is sent to the control module, which then restores the line at the fault location according to the recovery strategy; if no recovery strategy corresponding to the current fault exists in the control strategy library, it is uploaded to the remote master station to trigger the remote master station to calculate and issue the recovery strategy, and the control module automatically restores the faulty line according to the strategy issued by the strategy library.

[0124] Fault recovery check: If the fault recovery is successful, record the fault information and recovery strategy, and update the recovery strategy in the strategy library; if the fault recovery is unsuccessful, check if the request time exceeds the set threshold X. S Seconds (the threshold time is determined according to actual needs); if yes, request manual intervention for recovery; if no, request a new recovery strategy from the terminal.

[0125] Triggering recovery strategy: The remote master station sends the recovery strategy to the strategy library module, and the control module performs automatic recovery according to the strategy sent by the strategy library.

[0126] The present invention also provides a computer device, including: a memory and a processor, wherein the memory stores a computer program, and the processor is used to execute the computer program to implement the above-described method for locating faults in medium-voltage distribution network lines.

[0127] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of the present invention can be implemented using various computer languages, such as the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0128] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0129] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0130] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0131] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention. Clearly, those skilled in the art can make various alterations and modifications to the invention without departing from its spirit and scope. Thus, if these modifications and modifications of the invention fall within the scope of the claims and their equivalents, the invention is also intended to include these modifications and modifications.

Claims

1. A method for locating faults in medium-voltage distribution network lines, characterized in that, include: The fault current signal of the distribution network connected to the distributed energy source is subjected to Hilbert-Huang transform to obtain the instantaneous frequency and instantaneous amplitude of multiple key component signals of the fault current signal. Multidimensional signal phase space reconstruction is performed based on the instantaneous frequency and instantaneous amplitude of multiple key component signals to obtain a phase space reconstructed signal. This includes: three-dimensional signal phase space reconstruction based on the instantaneous frequency of each key component signal to obtain a three-dimensional spatial signal vector; and three-dimensional signal phase space reconstruction based on the instantaneous amplitude of each key component signal to obtain a four-dimensional spatial signal vector. The delay time of the signal phase space reconstruction is determined by: calculating the mutual information between the original key component signal and the delayed key component signal based on the cross-correlation function of the instantaneous frequency and instantaneous amplitude of each key component signal, and selecting the time point at which the first local minimum value is reached as the delay time of the phase space reconstructed signal. The method involves using a convolutional neural network to extract the time-series features of the distribution network fault current signal and the features of the phase space reconstructed signal. The features of the time-series features of the distribution network fault current signal and the features of the phase space reconstructed signal are then mixed. The fault location is determined based on the mixed features. This includes: mixing the features of the time-series features of the distribution network fault current signal with the features of a three-dimensional spatial signal vector, and using the mixed features for fault line selection to determine the faulty line; and mixing the features of the time-series features of the distribution network fault current signal with the features of a four-dimensional spatial signal vector, and using the mixed features for fault distance measurement to determine the fault location.

2. The method for locating faults in medium-voltage distribution network lines according to claim 1, characterized in that, The process of performing a Hilbert-Huang transform on the distribution network fault current signal connected to distributed energy resources to obtain the instantaneous frequency and instantaneous amplitude of multiple key component signals of the distribution network fault current signal includes: The fault current signal of the distribution network connected to distributed energy is decomposed into multiple different frequency components, and multiple key component signals are extracted from multiple different frequency components. Hilbert transforms were performed on the extracted key component signals to obtain the instantaneous frequency and instantaneous amplitude of the key component signals.

3. The method for locating faults in medium-voltage distribution network lines according to claim 2, characterized in that, The step of performing Hilbert transforms on multiple key component signals to obtain the instantaneous frequency and instantaneous amplitude of the multiple key component signals includes: A complex signal is obtained by performing a Hilbert transform on each key component signal. This complex signal is composed of the key component signal and the Hilbert transform signal of the key component signal. Based on the key component signal and its Hilbert transform, the instantaneous amplitude and instantaneous phase of the key component signal are calculated, and the instantaneous frequency of the key component signal is calculated based on its instantaneous phase.

4. The method for locating faults in medium-voltage distribution network lines according to claim 1, characterized in that, The formula for calculating the mutual information between the original key component signal and the delayed key component signal is as follows: ； in, I(τ) For mutual information, x(t) The original key component signal, x(t+τ) The key component signal after a delay time τ, P(x(t)) It is the original key component signal x(t) The probability distribution, P(x(t+τ)) It is the key component signal after the delay time τ. x(t+τ) The probability distribution, P(x(t),x(t+τ)) yes x(t) and x(t+τ) The joint probability distribution.

5. The method for locating faults in medium-voltage distribution network lines according to claim 1, characterized in that, The extraction of time-series features of the distribution network fault current signal and features of the phase space reconstructed signal using a convolutional neural network includes: One-dimensional convolutional neural networks are used to mine the time series features of the distribution network fault current signal. The feature quantities of the three-dimensional spatial signal vector and the feature quantities of the four-dimensional spatial signal vector are extracted using a point voxel convolutional neural network.

6. A smart switch, characterized in that, include: Acquisition module and edge computing module; The edge computing module includes: a fault segment location module and a fault ranging module; The fault section location module is used to perform preliminary location of the fault section based on the fault current signal of the distribution network connected to distributed energy collected by the acquisition module. The fault location module is used to locate the fault location within the initially identified fault section, specifically including: The fault current signal of the distribution network connected to the distributed energy source is subjected to Hilbert-Huang transform to obtain the instantaneous frequency and instantaneous amplitude of multiple key component signals of the fault current signal. Multidimensional signal phase space reconstruction is performed based on the instantaneous frequency and instantaneous amplitude of multiple key component signals to obtain a phase space reconstructed signal. This includes: three-dimensional signal phase space reconstruction based on the instantaneous frequency of each key component signal to obtain a three-dimensional spatial signal vector; and three-dimensional signal phase space reconstruction based on the instantaneous amplitude of each key component signal to obtain a four-dimensional spatial signal vector. The delay time of the signal phase space reconstruction is determined by: calculating the mutual information between the original key component signal and the delayed key component signal based on the cross-correlation function of the instantaneous frequency and instantaneous amplitude of each key component signal, and selecting the time point at which the first local minimum value is reached as the delay time of the phase space reconstructed signal. The method involves using a convolutional neural network to extract the time-series features of the distribution network fault current signal and the features of the phase space reconstructed signal. The features of the time-series features of the distribution network fault current signal and the features of the phase space reconstructed signal are then mixed. The fault location is determined based on the mixed features. This includes: mixing the features of the time-series features of the distribution network fault current signal with the features of a three-dimensional spatial signal vector, and using the mixed features for fault line selection to determine the faulty line; and mixing the features of the time-series features of the distribution network fault current signal with the features of a four-dimensional spatial signal vector, and using the mixed features for fault distance measurement to determine the fault location.

7. The smart switch according to claim 6, characterized in that, The fault location module is specifically used for: One-dimensional convolutional neural networks are used to mine the time series features of the distribution network fault current signal. The feature quantities of the three-dimensional spatial signal vector and the feature quantities of the four-dimensional spatial signal vector are extracted using a point voxel convolutional neural network. The feature quantities of the time series of the distribution network fault current signal are mixed with the feature quantities of the three-dimensional spatial signal vector, and the fault line selection is performed based on the mixed feature quantities to determine the line where the fault occurred. The feature quantities of the time series of the distribution network fault current signal are mixed with the feature quantities of the four-dimensional spatial signal vector, and the fault location is determined by fault ranging based on the mixed feature quantities.

8. The smart switch according to claim 6, characterized in that, Also includes: The strategy library module and the control module; The strategy library module is used to receive the fault location and fault information sent by the edge computing module, and search for the recovery strategy corresponding to the current fault in the control strategy library of the strategy library module. The control module is used to receive fault segment location information sent by the fault segment location module and recovery strategy and fault location sent by the strategy library module, perform isolation operation according to the fault segment location information, and restore the line at the fault location according to the recovery strategy.

9. The smart switch according to claim 8, characterized in that, Also includes: Remote communication module; The remote communication module is used to communicate with the remote master station. When no recovery strategy corresponding to the current fault is found in the strategy library module, the fault location and fault information are uploaded to the remote master station, the recovery strategy issued by the remote master station is received, and the recovery strategy issued by the remote master station is sent to the strategy library module. The policy library module is also used to send the recovery policy issued by the remote master station to the control module; The control module is also used to isolate the faulty section corresponding to the location of the fault according to the recovery strategy issued by the remote master station.

10. The smart switch according to claim 9, characterized in that, Also includes: Fault logging module; The fault recording module is used to record fault information, fault location, and fault handling method, and update it to the strategy library module; The remote communication module is also used to upload the fault information, fault location, and fault handling method updated by the policy library module to the remote master station, so as to update the fault recovery policy stored in the remote master station.

11. The smart switch according to claim 10, characterized in that, Also includes: The self-test and self-recovery module is used to perform fault self-testing and self-recovery on the smart switch.

12. A computer device, characterized in that, include: Memory, which stores computer programs; A processor is used to execute the computer program to implement the medium-voltage distribution network line fault location method according to any one of claims 1-5.

Images