An intelligent automatic reclosing and protection system for power distribution networks

Abstract

The present invention relates to the field of power grid technology and discloses an intelligent automatic reclosing and protection system for a distribution network, wherein the system comprises: a power grid topology structure construction module, which is used to construct a power grid topology model; a signal acquisition and processing module, which is used to obtain high-quality signals; a feature fusion and fault identification module, which extracts multidimensional features based on the high-quality signals and obtains preliminary classification results; a fault property judgment module, which is used to determine the fault property; a reclosing strategy formulation module, which is used to combine reclosing strategies; a fault isolation module, which determines automatic reclosing nodes and executes reclosing strategies; the present invention improves the accuracy of fault type identification through multidimensional feature fusion and deep learning models, can accurately distinguish temporary faults from permanent faults and dynamically formulate optimal reclosing strategies, including intelligent adjustment of reclosing time and reclosing times, thereby improving the power supply reliability and safety of the distribution network.

Description

Technical Field

[0001] The present invention relates to the technical field of power grids, and more particularly to an intelligent automatic reclosing and protection system for a power distribution network. Background Art

[0002] The distribution network is a crucial component of the power system, responsible for delivering electricity from substations to end users. During its operation, various faults often occur due to external environmental factors such as lightning strikes, windstorms, and tree branches, as well as internal equipment failures. Most of these faults are temporary and can be resolved by automatically reclosing the circuit breaker after a brief power outage, often restoring power.

[0003] The existing Chinese patent, CN112462193B, discloses a method for determining automatic reclosing in a distribution network based on real-time fault filtering data. The method establishes a fault extraction model based on wavelet transforms, a comprehensive fault analysis and judgment model based on a support vector machine classification model, and simulation modeling and analysis. The method uses wavelet transforms to extract three-phase voltage, three-phase current, and zero-sequence current as feature quantities. The wavelet coefficients from the wavelet analysis are reconstructed and used as the support vector machine data set. The algorithm model training data is derived from actual recorded data, enhancing the model's reliability and practicality.

[0004] However, in complex power grid topologies under extreme weather conditions, especially in densely populated urban power grids with multiple branches and feeders, fault waveforms may be superimposed on each other due to multiple factors such as reflected waves and refracted waves, and the nature of the fault may change rapidly, resulting in unclear wavelet feature extraction and making it difficult for the classification model to accurately distinguish between permanent and transient faults. Forced binary classification can easily lead to misjudgment, such as mistaking permanent faults for temporary ones, resulting in repeated reclosing and damage to equipment. Summary of the Invention

[0005] The purpose of the present invention is to provide an intelligent automatic reclosing and protection system for a power distribution network in order to solve the above problems.

[0006] The present invention provides an intelligent automatic reclosing protection system for a power distribution network, comprising:

[0007] The grid topology structure construction module is used to construct a grid topology model. The grid topology model includes nodes and edges. Nodes include bus nodes, transformer nodes, automatic reclosing nodes, and load nodes, and edges represent lines.

[0008] The signal acquisition and processing module is used to collect the original signals of multiple nodes in the power grid topology model and pre-process the original signals to obtain high-quality signals;

[0009] The feature fusion and fault identification module extracts multi-dimensional features during the fault period based on high-quality signals, fuses the multi-dimensional features using a neural network to obtain fused features, and inputs the fused features into the recognition model to obtain preliminary classification results;

[0010] The fault nature judgment module determines the fault nature based on the fusion features and preliminary classification results;

[0011] The reclosing strategy formulation module determines the fault location and optimal reclosing time by integrating features and the grid topology model, sets the number of reclosing times based on the nature of the fault, and combines the optimal reclosing time and reclosing times into a reclosing strategy;

[0012] The fault isolation module determines the automatic reclosing node that executes the reclosing strategy based on the fault location and the power grid topology model and executes the reclosing strategy.

[0013] Furthermore, the specific method for obtaining the preliminary classification results includes:

[0014] Step 301: extract multidimensional features during the fault period, where the multidimensional features include a basic feature set, a time domain feature set, a frequency domain feature set, an energy feature set, a topology feature set, and a meteorological data feature set;

[0015] Step 302: Build a recognition model based on deep learning, input the fusion features into the recognition model, output the probability of occurrence of various fault types, and combine the probabilities of occurrence of all fault types into a preliminary classification result.

[0016] Furthermore, the method for determining the nature of the fault includes:

[0017] Step 401: Establish a mapping relationship between fault type and permanent fault probability to obtain the permanent fault condition probability corresponding to each fault type;

[0018] Step 402: Using the probability of occurrence of each fault type and the probability of a permanent fault condition corresponding to each fault type, a Bayesian network is used to calculate the permanent probability of the fault nature. The fault nature can be a temporary fault or a permanent fault. The sum of the temporary probability and the permanent probability is 1, and the temporary probability is calculated using the permanent probability.

[0019] Step 403: Use a decision tree to determine the nature of the fault. When the temporary probability of the fault nature exceeds the temporary threshold, it is determined to be a temporary fault; when the permanent probability of the fault nature exceeds the permanent threshold, it is determined to be a permanent fault; when both the temporary probability and the permanent probability of the fault nature do not exceed the corresponding threshold, it is determined to be an uncertain fault.

[0020] Furthermore, the specific methods for determining the fault location and optimal reclosing time by integrating features and the grid topology model include:

[0021] Calculate the comprehensive importance index of the affected loads in the power grid topology model;

[0022] The fused features and the grid topology model are input into the fault location model, which outputs the probability of each line failing and selects the line with the highest probability as the fault location.

[0023] Calculate the topology impact factor based on the power grid topology model, fault location, and comprehensive importance index;

[0024] Calculate the optimal overlap time. The optimal overlap time is obtained by adding the basic overlap time and the time adjustment amount. The time adjustment amount is the product of the time adjustment step and the time adjustment function. The time adjustment function is the weighted sum of the fault property score, weather condition score and topology structure influencing factor score. The weather condition score is calculated based on the meteorological data characteristics of the fault location, and includes the weighted sum of the scores of three meteorological factors: wind speed, rainfall and temperature.

[0025] Furthermore, the determination method during the fault period includes:

[0026] When the high-quality signal change rate exceeds the preset normal range threshold, it is marked as the fault starting moment;

[0027] When the high-quality signal change rate returns to the normal range threshold and remains stable for a preset time, it is marked as the fault end time. For persistent faults, if the high-quality signal change rate continues to exceed the normal range threshold and exceeds the preset maximum allowable time, it will be forcibly marked as the fault end time.

[0028] The time period from the fault start time to the fault end time is defined as the fault duration.

[0029] Furthermore, specific methods for setting the number of coincidence times based on the nature of the fault include:

[0030] When the fault nature of the fault location is a permanent fault, the reclosing times is 0 and no reclosing operation is performed; when the fault nature of the fault location is an uncertain fault, the reclosing times is 1 and one reclosing attempt is performed; when the fault nature of the fault location is a temporary fault, the reclosing times is 2 and a maximum of two reclosing attempts are performed.

[0031] Furthermore, the method of determining an automatic reclosing node for executing a reclosing strategy based on a fault location and a power grid topology model and executing the reclosing strategy includes:

[0032] Step 601, the method for determining the automatic reclosing node includes:

[0033] Starting from the fault location, trace back to the power source in the grid topology model;

[0034] Identify the set of all auto-reclosing nodes that cut off the line at the fault location;

[0035] Calculate the electrical distance from each auto-reclosing node to the fault location;

[0036] Select the automatic reclosing node with the smallest electrical distance as the execution node;

[0037] If there are multiple auto-reclosing nodes with the same distance, the auto-reclosing node with the smallest impact range is selected as the execution node;

[0038] Step 602: Perform a reclosing operation on the automatic reclosing node determined to execute the reclosing strategy according to the optimal reclosing time and reclosing times.

[0039] Furthermore, the specific method of constructing the power grid topology model includes:

[0040] Step 101: construct a power grid topology model based on a node-edge representation method. The power grid topology model includes a node set and an edge set. The node set includes multiple nodes, and the edge set includes multiple edges. In a power distribution network, when there is an electrical connection between two nodes, an edge is formed. Each edge includes an electrical distance and impedance of the line. The electrical distance is the edge length, and the impedance is the sum of the line resistance and reactance.

[0041] Step 102, calculating the connectivity matrix, node admittance matrix and state estimation of the power grid topology model;

[0042] Step 103: Update the grid topology model according to the automatic reclosing state.

[0043] The present invention provides an intelligent automatic reclosing device for a power distribution network, which is used for communication connection with a controller, and the controller is used for executing the aforementioned intelligent automatic reclosing protection system for a power distribution network.

[0044] The beneficial effects of the present invention are as follows: by combining multi-dimensional feature fusion and power grid topology model, the present invention effectively overcomes the difficulty of waveform feature extraction caused by the superposition of reflected waves and refracted waves in multi-branch, multi-feeder urban dense power grids, improves the feature extraction capability in complex power grid environments, and enhances the robustness and recognition of fault features; based on multi-dimensional feature fusion and deep learning models, it integrates multi-dimensional information such as time domain, frequency domain, energy, topology and meteorology, compensates for the limitations of single wavelet features in complex interference scenarios, can more accurately distinguish temporary faults from permanent faults, and improves the accuracy of fault type identification.

[0045] Taking into account multiple factors such as the nature of the fault, load importance, meteorological conditions, and network topology, the overlap time and number of overlaps are dynamically adjusted to significantly improve the overlap success rate and reduce the risk of equipment damage. Through the graph convolutional neural network (GCN) and power grid topology model, the fault location is accurately located, the fault isolation and power supply restoration time are shortened, and the interference of reflected and refracted waves on the judgment of the fault propagation path is reduced.

[0046] A conservative reclosing strategy is adopted for uncertain faults, and the nature of the fault is re-evaluated through real-time data, balancing the speed of power supply restoration and equipment safety, and adapting to the rapidly changing characteristics of complex power grids; the overall fault recovery efficiency of the distribution network is optimized, reducing misjudgments and delays caused by waveform interference, and effectively reducing the risk of power outages for important loads. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 This is a module example diagram of an intelligent automatic reclosing protection system for a power distribution network according to the present invention;

[0048] Figure 2 This is an example diagram of a power grid topology model of an intelligent automatic reclosing protection system for a power distribution network according to the present invention;

[0049] Figure 3 This is an example diagram of a process for determining the nature of a fault in an intelligent automatic reclosing protection system for a power distribution network according to the present invention;

[0050] Figure 4 This is an example diagram of a reclosing strategy formulation process for an intelligent automatic reclosing protection system for a power distribution network according to the present invention. DETAILED DESCRIPTION

[0051] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed solely to enable those skilled in the art to better understand and implement the subject matter described herein, and that the functions and arrangements of the elements discussed may be varied without departing from the scope of this specification. Various examples may omit, substitute, or add various processes or components as needed. Furthermore, features described for some examples may be combined in other examples.

[0052] An intelligent automatic reclosing protection system for distribution networks, such as Figure 1 As shown, including:

[0053] The power grid topology structure construction module 100: constructs a power grid topology model, which includes a node set, an edge set, a connectivity matrix, a node admittance matrix and a state estimate. The node set includes multiple nodes, and the nodes include bus nodes, transformer nodes, automatic reclosing nodes and load nodes.

[0054] The specific implementation is as follows:

[0055] Step 101: construct a power grid topology model based on the node-edge representation method. The power grid topology model example is shown in FIG. Figure 2 As shown in the figure, using the grid topology model, the system can analyze fault propagation paths from a network structure perspective, reducing the interference of reflected and refracted waves on fault location. In densely populated urban power grids with multiple branches and feeders, auxiliary topology judgment can effectively circumvent the difficulty in feature extraction caused by waveform superposition.

[0056] The power grid topology model includes a node set and an edge set. The node set and edge set include multiple nodes and edges. In the distribution network, the nodes include bus nodes, transformer nodes, automatic reclosing nodes, load nodes and other key electrical equipment; among them, the bus node is the collection point of electric energy in the distribution network, which may be connected to multiple load nodes; the transformer node is the transformer equipment in the distribution network; the automatic reclosing node is an intelligent switching device in the distribution network, which is used to automatically open the circuit breaker after detecting a fault and attempt to reclose the circuit breaker after a short delay to restore power supply; the load node is the physical node connected to the load in the distribution network. The load includes hospitals, data centers, residential areas, industrial production lines, farmland irrigation, etc., representing the location of actual electricity consumption.

[0057] Edges represent lines, which are electrical pathways connecting different nodes and responsible for the transmission and distribution of electrical energy. When two nodes are electrically connected, an edge is formed between them. This representation method can intuitively reflect the physical connections of the power grid, facilitating subsequent topological analysis and fault location. The power grid topology model can simplify complex power grids into abstract models with clear mathematical structures, making them easier to process by algorithms. The specific representation is as follows:

[0058]

[0059] in, represents the power grid topology model, Represents a collection of nodes, Represents an edge set; a node set Each node in Represents an electrical device in the distribution network. Each node contains attribute data such as the type and operating parameters of the electrical device. For example, the bus node contains the rated voltage level and maximum load capacity; the transformer node contains the transformation ratio and rated capacity; the autorecloser node contains the reclosing status and operation time; the load node contains the load type and load quantity; each edge contains the electrical distance and impedance of the line. The electrical distance is the edge length, and the impedance is the sum of the line resistance and reactance. The specific expression of impedance is as follows:

[0060]

[0061] in, Indicates line The impedance, Represents a connected node and nodes The line, Indicates line The resistance, Indicates line The reactance, Represents an imaginary unit; by assigning actual electrical parameters to each edge in the power grid topology model, the power grid topology model not only includes connection relationships but also electrical characteristics, which can more accurately reflect the operating status and characteristics of the actual distribution network.

[0062] Step 102 is to calculate the connectivity matrix, node admittance matrix, and state estimation of the power grid topology model. These are important components of the power grid topology model and are used to describe and analyze the structural characteristics and operating status of the power grid. Specifically, the following steps are involved:

[0063] Calculate the connectivity matrix of the power grid topology model. The connectivity matrix is ​​a The matrix, the matrix elements Represents the space between two nodes, i.e., nodes and nodes Is it connected? If it is connected, the element value is 1 ( ), otherwise 0 ( ); The connectivity matrix provides a mathematical representation of the topology of the distribution network, which is convenient for computer analysis.

[0064] Calculate the node admittance matrix of the power grid topology model. The node admittance matrix is ​​the core mathematical expression of the power grid topology model. The matrix, the matrix elements Representation node and nodes The admittance between them, that is, the inverse of the impedance, reflects the electrical connection relationship between the nodes in the power grid.

[0065] Generate the grid state vector (GSV). The GSV is a dynamic component of the grid topology model. It includes the voltage phase angles and voltage amplitudes at multiple nodes. The GSV is a key vector for obtaining the real-time operating status of the grid.

[0066] Step 103: Update the grid topology model based on the auto-reclosing status. Changes in the auto-reclosing status directly affect the grid topology, so the grid topology model must be updated in real time to reflect the current actual operating status. The status of all auto-reclosing nodes, including closing / opening, is monitored in real time. When a status change is detected, steps 101 and 102 are re-executed.

[0067] Signal acquisition and processing module 200: collects current and voltage waveform data of multiple nodes in the power grid topology model to obtain original signals, removes noise and eliminates interference from the original signals to obtain high-quality signals.

[0068] The specific implementation is as follows:

[0069] In step 201, oscilloscopes installed at key nodes in the power grid topology model collect current and voltage waveform data to obtain raw signals. Key nodes are important nodes in the power grid topology model, including main busbars, important transformers, and load nodes in concentrated load areas. These nodes have high connectivity or carry significant loads within the power grid topology, making them key locations for fault monitoring. The collected raw waveform data serves as input for subsequent signal processing.

[0070] Step 202: Apply wavelet transform to the collected original signal to remove noise, decompose the time domain signal into wavelet coefficients at different scales and positions, and obtain the noise-reduced signal. In scale and location The wavelet coefficients at Equal to the original signal The integral of the conjugate function of the wavelet basis function after scale transformation and translation, where there is a scale before the integral A normalization factor that is inversely proportional to the square root. The specific formula is:

[0071]

[0072] in, Represents the original signal In scale and location The wavelet coefficients at is the wavelet basis function, is its conjugate function, represents the time variable, Indicates the signal at time Relative position By choosing an appropriate scale parameter and positional parameters , the noise components in the fault signal can be effectively separated, and the noise reduction processing of the original current and voltage signals can be achieved.

[0073] Step 203: Input the noise-reduced signal into the adaptive filter to eliminate interference and obtain a high-quality signal; the adaptive filter uses an adaptive filtering algorithm to continuously adjust the filter coefficients so that the filtered high-quality signal retains useful information to the greatest extent while suppressing interference; the specific process is: the high-quality signal is equal to the filter coefficient and the corresponding The cumulative sum of the products of the denoised signals at time , where The value range is 0 to , is the order of the filter. The specific expression is:

[0074]

[0075] in, For the current moment The filter coefficients of For the current moment High-quality signal, express The noise-reduced signal at time Indicates the order of the filter. The filter coefficients The update adopts the minimum mean square error (LMS) algorithm; the filter coefficient at the current moment Equal to the filter coefficient of the previous moment Add the step size parameter and the error signal and The product of the noise-reduced signals at time . The specific expression is:

[0076]

[0077] in, Indicates The filter coefficients at time , The step size parameter is used to control the update speed of the filter coefficients. Its selection directly affects the convergence speed and stability of the filter and usually needs to be adjusted according to the specific signal characteristics. The error signal refers to the difference between the filter output signal and the expected signal, which reflects the current performance of the filter. The method to obtain the error signal is to compare the filter output signal with the known reference signal, and then calculate the difference between the two. This method can adaptively eliminate the interference of specific frequencies according to the signal characteristics and retain fault information.

[0078] Feature fusion and fault identification module 300: Based on high-quality signals, it extracts multi-dimensional features during the fault period, and uses a neural network to fuse the multi-dimensional features to obtain fused features. The fused features are input into the recognition model to obtain preliminary classification results.

[0079] The specific implementation is as follows:

[0080] Step 301 extracts multidimensional features from the fault period based on high-quality signals and fuses them using a neural network to generate fused features. These multidimensional features include a basic feature set, a time-domain feature set, a frequency-domain feature set, an energy feature set, a topology feature set, and a meteorological data feature set. Multidimensional feature fusion effectively overcomes the difficulty in waveform feature extraction in densely populated urban power grids with multiple branches and feeders, caused by the superposition of reflected and refracted waves. This improves feature extraction capabilities in complex power grid environments and enhances the robustness and discernibility of fault features.

[0081] The specific methods for extracting multidimensional features during a fault period and fusing them using a neural network to obtain fused features include:

[0082] Extract basic feature sets, based on high-quality signals, to extract basic feature sets during faults The basic feature set includes multiple basic features, including fault current amplitude, phase angle, and frequency characteristics. The basic features during the fault period are extracted mainly through the following methods:

[0083] Fault onset detection is performed by setting a normal range threshold for current or voltage. When the signal change rate exceeds the normal range threshold, it is marked as the fault onset moment. For example, the normal range threshold for current is set to 200% of the normal current per cycle, or the normal range threshold for voltage drop is set to 30% of the rated voltage. When the measured current increases by more than twice the normal value within one cycle, or the voltage suddenly drops by more than 30% of the rated value, the system determines it as the fault onset moment. The threshold setting can also be adaptively adjusted based on historical fault data in different seasons and regions to improve detection accuracy.

[0084] When the current or voltage returns to the normal range threshold and remains stable for a preset time, usually 3-5 cycles, the fault is marked as over. For persistent faults (such as three-phase short circuits), if the current continues to exceed the normal range threshold for more than the preset maximum allowable time, the system will forcibly mark it as the fault end time. This is to prevent damage to the equipment caused by prolonged excessive current.

[0085] Fault period determination: The time period from the fault start time to the fault end time is defined as the fault period, and features are extracted within this interval;

[0086] Feature calculation: During the specified fault period, various feature parameters are calculated, such as maximum current value, current change rate, and spectrum characteristics.

[0087] Extract time domain feature set, time domain feature set Includes multiple time domain features, including maximum current , current change rate and the effective value of current Time-domain features are extracted directly from the current waveform of high-quality signals. The maximum current value is obtained by performing peak detection on the processed current waveform; the current rate of change is calculated by dividing the current difference between adjacent sampling points by the sampling interval; and the effective current value is obtained by calculating the square mean root of the instantaneous current values ​​during the fault period. This involves averaging the squared current values ​​at each moment during the fault period and then taking the square root. Time-domain features can effectively distinguish different types of faults, providing the system with a preliminary basis for determining the nature and severity of the fault.

[0088] Extract frequency domain feature set, frequency domain feature set Includes multiple frequency domain features, including harmonic components , time-frequency spectrum and power spectral density The frequency domain characteristics are obtained by frequency domain transformation of the voltage and current waveforms of high-quality signals. Obtained by performing fast Fourier transform (FFT) on the fault signal; time spectrum The short-time Fourier transform (STFT) is used to perform time-frequency analysis on high-quality signals; power spectral density Obtained by calculating the Fourier transform of the autocorrelation function of high-quality signals. Frequency domain features can reveal fault characteristics that are difficult to detect through time domain analysis, and are particularly suitable for identifying instantaneous faults and harmonic interference faults containing high-frequency components. and power spectral density It has obvious advantages in identifying complex waveforms composed of superposition of reflected waves and refracted waves.

[0089] Extract energy feature set, energy feature set Includes multiple energy signatures, including the accumulated energy during the fault The energy signature data is derived from high-quality current waveform data. The cumulative energy is calculated by numerically integrating the squared current value during the fault period. The energy signature during a fault period directly reflects the impact of the fault on the system and is an important indicator for distinguishing temporary from permanent faults. The cumulative energy of a permanent fault is typically significantly higher than that of a temporary fault.

[0090] Combined with topological feature set, topological feature set Includes multiple topological features, including the node centrality of multiple nodes , node betweenness and node degree Topological features are derived from the power grid topology model and connectivity matrix. Node centrality is calculated by calculating the shortest paths between a single node and all other nodes; node betweenness is calculated by counting the number of shortest paths from a single node; and node degree is calculated by counting the number of edges connected to a single node in the connectivity matrix. Combined with the physical structure of the power grid, topological features help analyze the network importance of the fault location and the scope of its impact on the system, providing a network structural basis for formulating reclosing strategies.

[0091] Integrate meteorological data feature set, meteorological data feature set Includes multiple meteorological data features, including wind speed features , rainfall characteristics and temperature characteristics Meteorological data features obtain meteorological data in real time from the meteorological monitoring system through the meteorological data interface. Meteorological data includes meteorological station data and meteorological forecast data.

[0092] For wind speed characteristics To calculate the wind speed, we first spatially interpolate the wind speed data within the area covered by the grid topology model to obtain wind speed estimates for each node. We then calculate the average and maximum wind speeds within the window before the fault occurs and use them as wind speed features. Next, we normalize the wind speed features to a value between 0 and 1.

[0093] For rainfall characteristics To perform the same calculation, spatial interpolation is performed on rainfall data within the area covered by the power grid topology model to obtain an estimated rainfall intensity at each node. The cumulative rainfall, as well as the average and maximum rainfall intensities within the time window before the fault, are then calculated as rainfall features. These rainfall features are then normalized to a value between 0 and 1.

[0094] For temperature characteristics The system then performs spatial interpolation of temperature data within the area covered by the power grid topology model to obtain temperature estimates for each node. The temperature value at the time of the fault, as well as the average temperature and temperature change rate within the preceding window, are then calculated as temperature features. Finally, the temperature features are normalized to a value between 0 and 1.

[0095] The wind speed characteristics, rainfall characteristics and temperature characteristics obtained by the above method constitute a complete meteorological data feature set After normalization, these meteorological features are converted into feature vectors that can be used for fault analysis. Meteorological data is valuable for determining the cause and nature of faults, especially under extreme weather conditions. For example, high wind speeds often cause temporary short circuits due to line swaying, heavy rainfall can cause ground faults due to insulation degradation, and temperature fluctuations can affect equipment status.

[0096] Apply neural network to fuse multi-dimensional features, and use neural network to transform time domain feature set into , frequency domain feature set , energy feature set , topological feature set , meteorological data feature set and the basic feature set Perform weighted fusion to obtain fusion features , the specific expressions are as follows:

[0097]

[0098] in, To fusion features, is the time domain feature set, is the frequency domain feature set, is the energy feature set, is the topological feature set, is the meteorological data feature set, is the basic feature set, 、 、 、 、 、 is the weight coefficient of each feature, is the activation function (such as ReLU, tanh, etc.). The feature fusion process realizes the comprehensive utilization of information of different dimensions and gives full play to the advantages of various features.

[0099] Step 302: Build a recognition model based on deep learning. The recognition model calculates the probability distribution of fault type (FaultType) under given features (Features), ranging from 0 to 1 and the sum of the probabilities of all fault types is 1. Features refer to the features input into the recognition model, which are fusion features. Fault types include a variety of common faults in distribution networks, such as single-phase grounding fault, two-phase short circuit fault, three-phase short circuit fault, interphase short circuit fault, high impedance fault, and intermittent fault. The recognition model outputs the probability of occurrence of various fault types under the condition of a known specific feature set, and combines the probabilities of occurrence of all fault types into a preliminary classification result.

[0100] The training steps of the recognition model are as follows:

[0101] Data preparation: Collect a large amount of historical fault data, including current and voltage waveforms at the time of the fault, grid topology, and meteorological data. Clean and preprocess the historical fault data to obtain fused features. These fused features are then used as a dataset. The dataset is divided into training, validation, and test sets.

[0102] Label preparation: Label the training data, using the fault type corresponding to each sample as the label. One-hot encoding can be used to encode the labels.

[0103] Model construction: Build a convolutional neural network model and define hyperparameters such as the network structure, number of layers, convolution kernel size, and activation function. Add a fully connected layer and a softmax layer at the end of the network to output the probability distribution of fault types.

[0104] Model training: The model is trained using the training set data, using the backpropagation algorithm and optimizer to update the model parameters. During the training process, model performance is optimized by adjusting hyperparameters such as the learning rate, batch size, and regularization coefficient.

[0105] Model Validation: Use validation data to evaluate the performance of the trained model, calculating metrics such as accuracy, precision, recall, and F1 score. Based on the validation results, adjust the model structure and hyperparameters, performing multiple rounds of iterative optimization.

[0106] Model testing: Use test set data to perform performance testing on the optimized model to evaluate the model's generalization ability and actual application effect.

[0107] The purpose of using fused features as input to the recognition model is:

[0108] The fusion features integrate basic features, time domain features, frequency domain features, energy features, topological features and meteorological features, providing multi-dimensional information of the fault, enabling the model to understand the fault characteristics from different angles.

[0109] Different types of features have complementary effects. For example, time domain features may be sensitive to short circuit faults, while frequency domain features are more effective for high impedance faults. After fusion, they can make up for the limitations of a single feature.

[0110] Multi-source feature fusion reduces the noise and error impact that may be brought by a single data source, and improves the robustness and anti-interference ability of the model.

[0111] By weightedly fusing different features through neural networks, the system can learn the importance of various features, optimize feature combinations, and improve the accuracy of fault type identification.

[0112] The fault conditions of power distribution networks are complex and changeable. A single feature cannot cope with all scenarios. However, fused features can adapt to a variety of complex fault environments and conditions.

[0113] Fault nature judgment module 400: Based on the fusion features and the preliminary classification results, the fault nature is determined. The fault nature includes temporary fault, permanent fault, and uncertain fault.

[0114] The specific implementation method is as follows Figure 3 As shown, including:

[0115] Step 401: Establish a mapping relationship between fault type and permanent fault probability to obtain the permanent fault condition probability corresponding to each fault type. Different types of faults have different permanent probabilities, which mainly depends on the cause and nature of the fault, for example:

[0116] A three-phase short circuit has a 95% probability of a permanent fault condition. This is because a three-phase short circuit is typically caused by severe physical damage, such as complete insulation failure or wire breakage, and is difficult to recover from on its own. For example, a three-phase short circuit caused by a burned-out transformer winding requires repairing or replacing the transformer to eliminate the fault.

[0117] High-impedance faults correspond to a permanent fault condition with a probability of 85%. High-impedance faults typically occur when a conductor comes into contact with a non-conductive object, such as a tree or the ground. This can cause persistent damage to the conductor or insulation, requiring manual cleaning of the fault point to restore normal operation. For example, long-term friction between a conductor and a tree branch can cause wear and tear.

[0118] The probability of a permanent fault condition corresponding to a single-phase ground fault is relatively low, at 30%. Single-phase ground faults are often caused by brief contact between a conductor and the ground. Unless the insulation is completely damaged, power can generally be restored by reclosing the circuit breaker, such as if the conductor is blown to the ground by strong wind.

[0119] The probability of a permanent fault condition corresponding to a two-phase short circuit fault is 40%. Two-phase short circuits are often caused by external forces damaging the insulation of the conductors. If the damage is not severe, the fault can be temporarily eliminated. For example, a two-phase short circuit can occur when construction machinery accidentally hits an overhead line.

[0120] The probability of a permanent fault condition corresponding to a phase-to-phase short circuit is 70%. Phase-to-phase short circuits often occur when the conductors themselves have quality defects or have aged due to long-term operation. Once a fault occurs, the conductors are usually already damaged and difficult to recover on their own. For example, aging and damage to the conductor insulation can reduce the distance between phases and cause discharge.

[0121] The probability of a permanent fault condition corresponding to an intermittent fault is only 25%. Intermittent faults are mostly caused by factors such as random contact between conductors and environmental factors. They can be effectively eliminated by reclosing, but they are prone to recurrence. For example, strong winds can cause conductors to sway, resulting in intermittent discharges from tree branches.

[0122] The above probability of permanent fault conditions is derived from a statistical analysis of a large number of historical fault cases, providing prior knowledge for subsequent Bayesian network fault probability reasoning. By considering the permanence tendency of different fault types, it is possible to more accurately determine the nature of the fault and develop a reasonable reclosing strategy.

[0123] In step 402, the preliminary classification results provide the probability of each fault type occurring. Using the probability of each fault type occurring, combined with the corresponding conditional probability of a permanent fault for each fault type, a Bayesian network is used to calculate the probability of the fault being permanent. The specific calculation method is as follows: the total probability of a permanent fault is equal to the weighted sum of the conditional probabilities of permanent faults for each fault type, where the weight is the probability of each fault type occurring. The calculation begins by first obtaining the probability of each fault type occurring, then multiplying it by the corresponding conditional probability of permanent faults for that fault type. Finally, all the results are summed to obtain the total probability of the fault being permanent under the current characteristic conditions. By linking the probability of a fault type occurring with the conditional probability of permanent faults, probabilistic reasoning from fault type to fault nature is achieved, addressing the issue of rapidly changing fault nature.

[0124] The specific expression formula is as follows:

[0125]

[0126] in, is the probability that the fault type is permanent under specific features (fusion features), Fault type The probability of occurrence, Is the fault type By associating the probability of the fault type with the probability of the permanent fault condition, probabilistic reasoning from the fault type to the fault nature is achieved.

[0127] Similarly, since a fault can only be temporary or permanent, and the sum of the two probabilities is 1, the temporary probability can be directly calculated from the permanent probability. The temporary probability of the fault type is 1 minus the probability of the fault type being permanent, which is calculated as follows:

[0128]

[0129] in, is the probability that the fault type is transient.

[0130] Step 403 uses a decision tree to determine the nature of the fault. Fault nature includes temporary fault, permanent fault, and uncertain fault. The specific determination method is as follows: when the temporary probability of the fault nature exceeds the temporary threshold, it is determined to be a temporary fault; when the permanent probability of the fault nature exceeds the permanent threshold, it is determined to be a permanent fault; when both probabilities do not exceed the corresponding thresholds, it is determined to be an uncertain fault. The preset logic of the temporary threshold and the permanent threshold is as follows:

[0131] The temporary threshold is set to 0.75, which is determined based on historical data analysis and expert experience. It takes into account the unnecessary power outage costs that may result from misjudging a temporary fault as a permanent fault. Under special weather conditions such as thunderstorms, the threshold can be dynamically adjusted to 0.80 to reduce misjudgments.

[0132] The permanent threshold is usually set to 0.85, which is higher than the temporary threshold. This is because misjudging a permanent fault as a temporary fault can lead to repeated reclosing failures, increasing equipment loss and the risk of system instability.

[0133] The difference between the two thresholds constitutes an uncertainty interval, providing the system with a buffer for decision-making, preventing oscillation when probabilities approach the boundary. This hierarchical judgment mechanism enhances the system's decision-making reliability, especially for challenging edge cases. By setting appropriate thresholds, a balance can be achieved between accuracy and sensitivity.

[0134] Classifying faults into temporary, permanent, and uncertain faults can solve the problem of misjudgment caused by forced binary classification. This classification not only meets actual application needs but also optimizes fault handling strategies. The following is a detailed analysis and example:

[0135] Different fault types require different handling strategies. Temporary faults (such as instantaneous flashovers caused by lightning) can usually be automatically corrected with rapid reclosing. Permanent faults (such as aging and damaged cable insulation) require manual repair. For uncertain faults, a progressive strategy can be employed, such as first attempting delayed reclosing and then dispatching a human for inspection if this fails. For example, in the case of a metallic ground fault, where the system cannot determine whether the fault was caused by a metal object temporarily contacting the conductor or a broken conductor, it can be classified as an uncertain fault and a delayed reclosing strategy can be employed.

[0136] The introduction of uncertain fault categories helps optimize the allocation of maintenance resources. For faults identified as permanent, maintenance personnel can be dispatched immediately; for temporary faults, automated measures can be used to address them; and for uncertain faults, resources can be flexibly allocated based on their importance and risk level. For example, after a widespread thunderstorm, multiple faults may occur. The system can prioritize permanent faults and implement a monitoring and waiting strategy for uncertain faults.

[0137] The operating environment of power distribution networks is complex and ever-changing, and fault characteristics are often ambiguous due to multiple factors. A three-class classification approach is more adaptable to this complexity than a two-class classification approach. For example, during rainy seasons, a high-impedance fault could be caused by a wet tree branch temporarily touching a conductor or by contaminated or damaged insulators. In such cases, it would be more appropriate to classify it as an uncertain fault.

[0138] In distribution networks, fault diagnosis is inherently uncertain. Introducing an uncertain fault category can avoid potentially erroneous binary judgments when insufficient evidence is available, reducing decision-making risk. For example, if the system detects a single-phase ground fault, but its characteristics neither fully match those of a typical temporary fault, such as a lightning-induced flashover, nor a permanent fault, such as equipment damage, it is more reasonable to classify it as an uncertain fault.

[0139] By introducing the uncertain fault category, the system can take a more cautious approach to edge cases, avoiding power outages or equipment damage caused by misjudgment and improving overall reliability. For example, even if a fault on a critical load line has a 70% probability of being temporary, the system will classify it as an uncertain fault because it does not meet the 75% temporary threshold. A more conservative handling strategy will be adopted to ensure the power supply security of critical loads.

[0140] Reclosing strategy formulation module 500: determines the fault location and optimal reclosing time by integrating features and grid topology model, sets the number of reclosing times based on the nature of the fault, and combines the optimal reclosing time and the number of reclosing times into a reclosing strategy.

[0141] The specific implementation method is as follows Figure 4 As shown, including:

[0142] Step 501: Calculate the comprehensive importance index of the affected loads in the power grid topology model. The comprehensive importance index is calculated by the importance factors of different load types. The specific calculation method is as follows:

[0143]

[0144] in, is a comprehensive importance indicator. is the weight coefficient, For the Importance factor of each load type, is the number of load types; examples of load types and their corresponding importance factors are as follows:

[0145] The importance factor of super-critical loads (such as hospitals, data centers, emergency command centers, etc.) is 1.0;

[0146] The importance factor of first-level important loads (such as industrial production lines, large commercial centers, etc.) is 0.8;

[0147] The importance factor of secondary important loads (such as schools, medium-sized enterprises, etc.) is 0.6;

[0148] The importance factor of the third-level general load (such as residential areas, small commercial facilities, etc.) is 0.4;

[0149] The importance factor of level 4 non-critical loads (such as farmland irrigation, temporary power facilities, etc.) is 0.2;

[0150] The weight coefficient is the product of seasonal factors, time factors, electricity consumption factors and the basic weights of various types of loads. The seasonal factor is 1.2 during the peak seasons of summer and winter, and 1.0 during the normal seasons of spring and autumn. The time factor is 1.2 during the peak seasons of weekdays, 1.0 during the normal seasons of weekdays, and 0.9 during holidays. The electricity consumption factor is 1.3 when it is greater than 80% of the rated capacity, 1.0 when it is 50%-80% of the rated capacity, and 0.8 when it is less than 50% of the rated capacity. The basic weights of various types of loads are 5.0 for super-important loads, 4.0 for first-level important loads, 3.0 for second-level important loads, 2.0 for third-level important loads, and 1.0 for fourth-level non-critical loads.

[0151] Through this multi-dimensional weighting approach, the system can quantitatively assess the comprehensive importance of loads in the fault-affected area. Load importance assessment provides a key basis for formulating reclosing strategies, ensuring that critical loads are restored to power first.

[0152] In step 502, the fused features and the power grid topology model are input into the fault location model. The fault location model outputs the probability of each line fault occurring and selects the line with the highest probability as the fault location. The fused features contain multi-dimensional characteristic information, including time, frequency, and energy, of the fault, providing the electrical characteristics of the fault. The power grid topology model provides the connection relationships and electrical parameters between lines, reflecting the fault propagation path and electrical distance. The combination of these two types of information enables the fault location model to simultaneously consider both the electrical characteristics of the fault and the network structure, reducing the interference of reflected and refracted waves on the fault propagation path determination, thereby achieving more accurate fault location.

[0153] In one embodiment of the present invention, the fault localization model uses a graph convolutional neural network (GCN), and the specific training steps are:

[0154] Data preparation: Collect a location dataset, including fusion features, the grid topology at the time of the fault, and the actual location label of the fault. Divide the location dataset into training, validation, and test sets.

[0155] Model architecture: The input layer receives the fused features and the grid topology. Three graph convolution layers are set in the middle. Each layer is followed by batch normalization and ReLU activation function. Finally, the fully connected layer and Softmax function are used to output the fault probability of each line.

[0156] Loss function: cross entropy loss function is used.

[0157] Training strategy: The Adam optimizer was used, with an initial learning rate of 0.001 and a 10% decay every 50 epochs. An early stopping strategy was used to prevent overfitting, with training terminated when the validation set loss did not decrease for 10 consecutive epochs.

[0158] Model evaluation: Use validation data to evaluate the performance of the trained model, calculating metrics such as accuracy, precision, recall, and F1 score. Based on the validation results, adjust the model structure and hyperparameters, performing multiple rounds of iterative optimization.

[0159] Model testing: Use test set data to perform performance testing on the optimized model to evaluate the model's generalization ability and actual application effect.

[0160] Step 503 , calculating a topology impact factor based on the power grid topology model, the fault location, and the comprehensive importance index. The topology impact factor has a value range of 0-1 and is obtained by calculating a topology function, which is a composite evaluation function.

[0161] The topology function first analyzes the centrality and connectivity of the fault location within the power grid topology model to determine the extent of its impact on overall network connectivity. Centrality is quantified by calculating the betweenness centrality of the fault location within the network. This is calculated by counting the proportion of the shortest paths between all node pairs that pass through the fault location. A higher ratio indicates a higher centrality for the fault location and a greater impact on network connectivity. Connectivity is measured by counting the number of edges connected to the fault location. A higher connectivity indicates that the location is associated with more lines and equipment, and the fault's impact may be wider.

[0162] The topology function then evaluates the electrical distance from the fault location to each node and the number of power paths (edges) to determine the range of affected loads. The number of power paths is calculated using algorithms such as depth-first search, starting from the fault location and counting the number of paths to each load node. A greater number of paths indicates greater redundant power supply capability and a lower risk of power loss to that load due to a fault.

[0163] Finally, the topology function combines the comprehensive importance index with a weighted calculation of the influence degree and load range to derive the topology impact factor. The influence degree weight is proportional to the centrality and connectivity, while the load range weight is inversely proportional to the electrical distance and directly proportional to the number of power supply paths. The comprehensive importance index serves as the weight coefficient for each load node, and the weighted sum of the influence degree and load range is taken to obtain the final topology impact factor.

[0164] For mainline faults, the topology function returns a higher topology impact factor due to their high connectivity and wide impact range. However, for branch line faults that only affect a small number of non-critical loads, a lower topology impact factor is returned. This analysis allows the system to identify the impact of the fault and provide a topology-level reference for developing reclosing strategies.

[0165] Step 504: Calculate the optimal overlap time. The optimal overlap time is obtained by adding the basic overlap time and the time adjustment amount. The time adjustment amount is the product of the time adjustment step and the time adjustment function. The specific calculation formula is as follows:

[0166]

[0167] in, is the optimal overlap time, As the basic overlap time, is the time adjustment, where Adjust the step size for time, is the time adjustment function, Score the nature of the fault, Rate the weather conditions, To score the topology impact factor, the specific calculation method of the time adjustment function is as follows:

[0168] Time adjustment function It is the weighted sum of the fault nature score, weather condition score, and topology impact factor score. The default weight coefficients for the fault nature, weather condition, and topology impact factors are set to 0.4, 0.35, and 0.25, respectively.

[0169] The fault nature score ranges from -1 to 1. A temporary fault is scored as -1 (shortening the reclosing time), a permanent fault is scored as 1 (extending the reclosing time), and an uncertain fault is scored as 0 (not adjusting the reclosing time).

[0170] The weather condition score is calculated based on the meteorological data characteristics of the fault location, and the value range is [-1,1]. The meteorological data features of the two nodes connected to the fault location are extracted, and the initial weather condition scores of the two nodes are calculated respectively. The initial weather condition scores of the two nodes are averaged to obtain the weather condition score. The initial weather condition score is the weighted sum of the scores of the three meteorological factors: wind speed, rainfall and temperature. The default values ​​of the weight coefficients are set to 0.4, 0.4 and 0.2 respectively, which can be adjusted according to regional characteristics. The wind speed feature score is calculated by adding the wind speed feature Mapped to the [-1,1] interval. When the wind speed is high ( Close to 1), the score is close to 1, indicating that the overlap time needs to be extended; when the wind speed is low ( The score is close to 0), and the score is close to -1, which means that the overlap time can be shortened. Mapped to the interval [-1, 1] to obtain. When the rainfall intensity is large ( Close to 1), a score close to 1 indicates that the overlap time needs to be extended; when there is no rainfall ( The temperature characteristic score takes into account that too high or too low temperature may require longer reclosing time, while moderate temperature can shorten reclosing time. When the temperature is close to the extreme value ( close to 0 or 1), the score is close to 1; when the temperature is moderate ( close to 0.5), and the score is close to -1.

[0171] The topology impact factor ranges from [0, 1]. To maintain consistency with other factors, it is mapped to the interval [-1, 1] to obtain the topology impact factor score. When the topology impact is significant (the topology impact factor is close to 1), a score close to 1 indicates that the reclosing time needs to be extended to ensure system stability. When the topology impact is minor (the topology impact factor is close to 0), a score close to -1 indicates that the reclosing time can be shortened to quickly restore power.

[0172] The default value of the basic coincidence time is set to 2 seconds, which is the benchmark coincidence time value commonly used in power distribution networks. The default value of the time adjustment step is set to 1 second, indicating a maximum adjustable range of ±1 second.

[0173] Using this calculation method, the optimal reclosing time is dynamically adjusted between 1 and 3 seconds, depending on a comprehensive assessment of the fault nature, weather conditions, and topological impacts. This multi-factor dynamic adjustment mechanism calculates the optimal reclosing time for different fault scenarios and environmental conditions, improving reclosing success rates and reducing the risk of equipment damage, while maintaining a balanced balance between power supply reliability and system security.

[0174] Step 505, setting the number of reclosing times based on the nature of the fault. When the fault nature of the fault location is a permanent fault, the number of reclosing times is 0, and no reclosing operation is performed; when the fault nature of the fault location is an uncertain fault, the number of reclosing times is 1, and one reclosing attempt is performed; when the fault nature of the fault location is a temporary fault, the number of reclosing times is 2, and a maximum of two reclosing attempts are performed.

[0175] Fault isolation module 600: determines the automatic reclosing node for executing the reclosing strategy according to the fault location and the power grid topology model and executes the reclosing strategy.

[0176] The specific implementation is as follows:

[0177] Step 601: Determine the automatic reclosing node that executes the reclosing strategy based on the fault location and the grid topology model. The automatic reclosing node is the nearest upstream automatic reclosing node in the grid topology model that can cut off the line at the fault location. The determination method is as follows:

[0178] Starting from the fault location, trace back to the power source in the grid topology model;

[0179] Identify the set of all automatic reclosing nodes that can disconnect the line at the fault location;

[0180] Calculate the electrical distance from each auto-reclosing node to the fault location;

[0181] Select the automatic reclosing node with the smallest electrical distance as the execution node;

[0182] If there are multiple auto-reclosing nodes with the same distance, the auto-reclosing node with the smallest impact range (i.e., the node with the least number of power outage loads after disconnection) is selected as the execution node.

[0183] Step 602: Execute the reclosing strategy. For a temporary fault:

[0184] Performing reclosing operation on the automatic reclosing node determined to execute the reclosing strategy according to the determined reclosing time and reclosing number;

[0185] The grid topology construction module is used to update the grid topology; the signal acquisition and processing module is used to monitor the system status after reclosing in real time; the feature fusion and fault identification module is used to detect the preliminary classification results after reclosing; the fault nature judgment module is used to determine the nature of the fault after reclosing; and the reclosing strategy formulation module is used to determine the fault location after reclosing.

[0186] If there is another fault at the same fault location, it becomes a permanent fault.

[0187] In case of permanent failure:

[0188] The automatic reclosing node that is determined to execute the reclosing strategy performs the reclosing operation according to the determined reclosing time and reclosing number. If the reclosing number is 0, the reclosing operation is not performed.

[0189] Use the analog signal acquisition and processing module to monitor the system status after reclosing in real time, including current, voltage and other parameters;

[0190] Updating the grid topology using the grid topology building module;

[0191] Notify maintenance personnel to handle the problem on site.

[0192] For uncertain failure situations:

[0193] Performing reclosing operation on the automatic reclosing node determined to execute the reclosing strategy according to the determined reclosing time and reclosing number;

[0194] The grid topology construction module is used to update the grid topology; the signal acquisition and processing module is used to monitor the system status after reclosing in real time; the feature fusion and fault identification module is used to detect the preliminary classification results after reclosing; the fault nature judgment module is used to determine the nature of the fault after reclosing; and the reclosing strategy formulation module is used to determine the fault location after reclosing.

[0195] If there is a temporary fault at the same fault location, it will be converted to a temporary fault. If there is a permanent fault or an uncertain fault at the same fault location, it will be converted to a permanent fault.

[0196] In at least one embodiment of the present invention, an intelligent automatic reclosing system for a power distribution network is provided, which is used to communicate with a controller, and the controller is used to execute the aforementioned intelligent automatic reclosing protection system for a power distribution network.

[0197] The above describes an embodiment of the present invention, but this embodiment is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Ordinary technicians in this field can also make more forms of equivalent embodiments based on the inspiration of this embodiment, all of which are protected by this embodiment.

Claims

1. An intelligent automatic reclosing protection system for a power distribution network, characterized in that: include: The grid topology structure construction module is used to construct a grid topology model. The grid topology model includes nodes and edges. Nodes include bus nodes, transformer nodes, automatic reclosing nodes, and load nodes, and edges represent lines. The signal acquisition and processing module is used to collect the original signals of multiple nodes in the power grid topology model and pre-process the original signals to obtain high-quality signals; The feature fusion and fault identification module extracts multi-dimensional features during the fault period based on high-quality signals, fuses the multi-dimensional features using a neural network to obtain fused features, and inputs the fused features into the recognition model to obtain preliminary classification results; The fault nature judgment module determines the fault nature based on the fusion features and preliminary classification results: Establish a mapping relationship between fault type and permanent fault probability to obtain the permanent fault condition probability corresponding to each fault type; fault types include single-phase grounding fault, two-phase short circuit fault, three-phase short circuit fault, interphase short circuit fault, high impedance fault, and intermittent fault; Obtain the probability of each fault type, multiply it by the probability of the permanent fault condition corresponding to that fault type, and add all the results to obtain the total probability that the fault is a permanent fault under the current characteristic conditions. Fault properties can be temporary or permanent. The sum of the temporary probability and the permanent probability is 1, and the temporary probability is calculated using the permanent probability. A decision tree is used to determine the nature of the fault. When the fault type is temporary and the probability exceeds the temporary threshold, it is determined to be a temporary fault; when the fault type is permanent and the probability exceeds the permanent threshold, it is determined to be a permanent fault; when the temporary and permanent probabilities of the fault type do not exceed the corresponding thresholds, it is determined to be an uncertain fault. The reclosing strategy formulation module determines the fault location and optimal reclosing time by integrating features and the grid topology model, sets the number of reclosing times based on the nature of the fault, and combines the optimal reclosing time and reclosing times into a reclosing strategy; The fault isolation module determines the automatic reclosing node that executes the reclosing strategy based on the fault location and the power grid topology model and executes the reclosing strategy.

2. The intelligent automatic reclosing protection system for a power distribution network according to claim 1, characterized in that: The specific methods for obtaining preliminary classification results include: Step 301: extract multidimensional features during the fault period, where the multidimensional features include a basic feature set, a time domain feature set, a frequency domain feature set, an energy feature set, a topology feature set, and a meteorological data feature set; Step 302: Build a recognition model based on deep learning, input the fusion features into the recognition model, output the probability of occurrence of various fault types, and combine the probabilities of occurrence of all fault types into a preliminary classification result.

3. The intelligent automatic reclosing protection system for a power distribution network according to claim 1, characterized in that: The specific methods for determining the fault location and optimal reclosing time by integrating features and power grid topology models include: Calculate the comprehensive importance index of the affected loads in the power grid topology model; The fused features and the grid topology model are input into the fault location model, which outputs the probability of each line failing and selects the line with the highest probability as the fault location. Calculate the topology impact factor based on the power grid topology model, fault location, and comprehensive importance index; Calculate the optimal overlap time. The optimal overlap time is obtained by adding the basic overlap time and the time adjustment amount. The time adjustment amount is the product of the time adjustment step and the time adjustment function. The time adjustment function is the weighted sum of the fault property score, weather condition score and topology structure influencing factor score. The weather condition score is calculated based on the meteorological data characteristics of the fault location, and includes the weighted sum of the scores of three meteorological factors: wind speed, rainfall and temperature.

4. The intelligent automatic reclosing protection system for a power distribution network according to claim 1, characterized in that: Methods for determining during a fault include: When the high-quality signal change rate exceeds the preset normal range threshold, it is marked as the fault starting moment; When the high-quality signal change rate returns to the normal range threshold and remains stable for a preset time, it is marked as the fault end time. For persistent faults, if the high-quality signal change rate continues to exceed the normal range threshold and exceeds the preset maximum allowable time, it will be forcibly marked as the fault end time. The time period from the fault start time to the fault end time is defined as the fault duration.

5. The intelligent automatic reclosing protection system for a power distribution network according to claim 1, characterized in that: The specific methods for setting the number of coincidence times based on the nature of the fault include: When the fault nature of the fault location is a permanent fault, the reclosing times is 0 and no reclosing operation is performed; when the fault nature of the fault location is an uncertain fault, the reclosing times is 1 and one reclosing attempt is performed; when the fault nature of the fault location is a temporary fault, the reclosing times is 2 and a maximum of two reclosing attempts are performed.

6. The intelligent automatic reclosing protection system for a power distribution network according to claim 1, characterized in that: The method for determining an automatic reclosing node for executing a reclosing strategy according to a fault location and a power grid topology model and executing the reclosing strategy includes: Step 601, the method for determining the automatic reclosing node includes: Starting from the fault location, trace back to the power source in the grid topology model; Identify the set of all auto-reclosing nodes that cut off the line at the fault location; Calculate the electrical distance from each auto-reclosing node to the fault location; Select the automatic reclosing node with the smallest electrical distance as the execution node; If there are multiple auto-reclosing nodes with the same distance, the auto-reclosing node with the smallest impact range is selected as the execution node; Step 602: Perform a reclosing operation on the automatic reclosing node determined to execute the reclosing strategy according to the optimal reclosing time and reclosing times.

7. The intelligent automatic reclosing protection system for a power distribution network according to claim 1, characterized in that: The specific methods for constructing the power grid topology model include: Step 101: construct a power grid topology model based on a node-edge representation method. The power grid topology model includes a node set and an edge set. The node set includes multiple nodes, and the edge set includes multiple edges. In a power distribution network, when there is an electrical connection between two nodes, an edge is formed. Each edge includes an electrical distance and impedance of the line. The electrical distance is the edge length, and the impedance is the sum of the line resistance and reactance. Step 102, calculating the connectivity matrix, node admittance matrix and state estimation of the power grid topology model; Step 103: Update the grid topology model according to the automatic reclosing state.

8. An intelligent automatic reclosing device for a power distribution network, characterized in that: It is used for communication connection with a controller, and the controller is used for executing an intelligent automatic reclosing protection system for a power distribution network as described in any one of claims 1 to 7.

Images