Method for Detecting Short-Circuit Fault of Distribution Network Cable in Combination with Power Supply Network Topology Relationship

Abstract

This application relates to the technical field of cable short-circuit fault detection, specifically to a method for detecting cable short-circuit faults in a distribution network by combining the topology of the power supply network. The method includes: collecting the three-phase currents, three-phase voltages, and impedances of each node in the distribution network at each moment; determining the current mutation rate of each node at each moment; obtaining the primary fault characteristics of each node at each moment; obtaining the attenuation coefficient of each node at each moment; determining each fault isolation area in the distribution network through the connection status of the circuit breakers in the distribution network; determining the propagation attenuation weight of each node at each moment; weighting the primary fault characteristics by the reciprocal of the propagation attenuation weight, and using the weighted primary fault characteristics of all nodes at each moment, combining the node adjacency matrix of the distribution network and the three-phase currents and three-phase voltages of all nodes, and using a hybrid classifier to detect cable short-circuit faults in the distribution network. This application improves the accuracy of cable short-circuit fault detection.

Description

Technical Field

[0001] This application relates to the technical field of cable short - circuit fault detection, and specifically to a method for detecting cable short - circuit faults in a distribution network by combining the topological relationship of the power supply network. Background Art

[0002] The detection of cable short - circuit faults in a distribution network is a core link to ensure the safe operation of the power system. The arc high temperature generated by short - circuit faults can not only cause safety accidents such as fires and equipment explosions, but the instantaneous over - current it causes can also lead to insulation deterioration of the line and even permanent damage to equipment. With the development of a new - energy - high - penetration distribution network, the characteristics of short - circuit current are becoming more complex, and efficient detection technology has become a key defense line to ensure the safe and stable operation of the new power system. The topological relationship of the power supply network refers to the physical connection logic between electrical equipment in the power system and the structured description of the electric - energy transmission path.

[0003] Traditional methods for detecting cable short - circuit faults in a distribution network generally use static network models, which cannot respond to topological changes in the distribution network in real time and lack the combination of the topological relationship of the power supply network and electrical parameters. They often only detect short - circuit faults based on a single electrical parameter, resulting in low accuracy of short - circuit current feature extraction and interfering with the detection of cable short - circuit faults in the distribution network. Summary of the Invention

[0004] To solve the above - mentioned technical problems, this application provides a method for detecting cable short - circuit faults in a distribution network by combining the topological relationship of the power supply network to solve the existing problems.

[0005] The method for detecting cable short - circuit faults in a distribution network by combining the topological relationship of the power supply network in this application adopts the following technical solutions:

[0006] An embodiment of this application provides a method for detecting cable short - circuit faults in a distribution network by combining the topological relationship of the power supply network. The method includes the following steps:

[0007] Collect the three - phase current, three - phase voltage of each node at each moment in the distribution network, and the impedance of each node.

[0008] Construct the node adjacency matrix at each moment by traversing the closed circuit breaker paths in the distribution network; analyze the current difference between each node at each moment and its adjacent moment to determine the current mutation rate of each node at each moment.

[0009] Obtain the primary fault characteristics of each node at each moment through the zero - sequence transient energy and total harmonic distortion rate within the local time period of each node at each moment, combined with the current mutation rate.

[0010] Determine the topological connection degree of each node at each moment based on the connection distribution between nodes in the node adjacency matrix; use the line lengths and line impedances connected to each node at each moment to obtain the attenuation coefficient of each node at each moment; determine each fault isolation area in the distribution network through the connection status of the circuit breakers in the distribution network;

[0011] Use the number of hops from each node to the nearest fault isolation area at each moment, combine the topological connection degree, the attenuation coefficient, and the impedance deviation of each node to determine the propagation attenuation weight of each node at each moment;

[0012] Perform weighted processing on the primary fault features through the reciprocal of the propagation attenuation weight, use the weighted primary fault features of all nodes at each moment, combine the node adjacency matrix of the distribution network and the three-phase currents and three-phase voltages of all nodes, and use a hybrid classifier to detect cable short-circuit faults in the distribution network.

[0013] In one embodiment, the determination of the current mutation rate includes:

[0014] Calculate the difference between the in-phase currents at each moment and its adjacent moment, and the current mutation rate is positively correlated with the difference and negatively correlated with the in-phase current at the adjacent moment.

[0015] In one embodiment, calculate the ratio of the difference to the in-phase current at the adjacent moment, and the current mutation rate is the average value of the ratios of the three-phase currents at each moment.

[0016] In one embodiment, the topological connection degree is the number of connections between each node and other nodes in the node adjacency matrix at each moment.

[0017] In one embodiment, the attenuation coefficient is the product of the sum of the actual lengths of all lines connected to each node at each moment and the impedance per unit length of the line.

[0018] In one embodiment, the determination of the fault isolation area includes:

[0019] If the circuit breaker at node Y is disconnected, then mark node Y and all nodes directly connected to node Y as the fault isolation area; if the circuit breaker between node Y and node X is disconnected, then mark node Y and node X as the fault isolation area.

[0020] In one embodiment, the determination of the propagation attenuation weight includes:

[0021] Obtain the calculation result of the exponential function with the attenuation coefficient as the base and the number of hops as the exponent, calculate the multiplication result of the calculation result and the topological connection degree, and the propagation attenuation weight is positively correlated with the multiplication result and the impedance deviation.

[0022] In one embodiment, the impedance deviation is the absolute value of the difference between the impedance measurement value and the nominal value of each node at each moment.

[0023] In one embodiment, the hybrid classifier consists of a graph neural network, a deep belief network, and a D-S evidence theory fusion layer.

[0024] This application has at least the following beneficial effects:

[0025] This application first collects the three-phase current, three-phase voltage, and impedance of each node in the distribution network at each moment; constructs the node adjacency matrix at each moment by traversing the closed circuit breaker paths in the distribution network; analyzes the current difference between each node at each moment and its adjacent moments to determine the current mutation rate of each node at each moment; the current mutation rate reflects the possibility of a short circuit fault occurring at each node at each moment, improving the reliability of short circuit fault detection at the node; obtains the primary fault characteristics of each node at each moment through the zero-sequence transient energy and total harmonic distortion rate within the local time period of each node at each moment, combined with the current mutation rate; the primary fault characteristics fuse the characteristic signals of multiple aspects of electrical parameters, and can more sensitively capture the transient characteristics in the initial stage of the cable short circuit fault, improving the accuracy of cable short circuit fault detection; determines the topological connectivity of each node at each moment based on the connection distribution between nodes in the node adjacency matrix; the topological connectivity reflects the number of connections between each node and other nodes at each moment, reflecting the degree of closeness of the connection between each node and other nodes, and helping to measure the fault propagation effect between nodes; obtains the attenuation coefficient of each node at each moment using the line length and line impedance connected to each node at each moment; determines each fault isolation area in the distribution network through the connection state of the circuit breakers in the distribution network, and the determination of the fault isolation area improves the efficiency of cable short circuit fault detection; determines the propagation attenuation weight of each node at each moment using the number of hops from each node at each moment to the nearest fault isolation area, combined with the topological connectivity and the attenuation coefficient, and the impedance deviation of each node; the propagation attenuation weight reflects the degree of influence of each node at each moment by the short circuit fault, comprehensively considering the real-time topological relationship of the power supply network, and improving the accuracy and reliability of cable short circuit fault detection; performs weighted processing on the primary fault characteristics by taking the reciprocal of the propagation attenuation weight, uses the weighted primary fault characteristics of all nodes at each moment, combines the node adjacency matrix of the distribution network and the three-phase current and three-phase voltage of all nodes, and uses a hybrid classifier to detect cable short circuit faults in the distribution network; through the synergistic effect of multi-dimensional feature fusion and topological perception mechanism, this application achieves a triple breakthrough in detection accuracy, response speed, and complex scenario adaptability in distribution network short circuit detection; expands the traditional single electrical quantity analysis to a multi-physical field joint perception system: effectively and accurately detects cable short circuit faults in the distribution network by dynamically mapping the power supply network topology with the node adjacency matrix. Brief Description of the Drawings

[0026] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0027] Figure 1 It is a flowchart of the steps of the method for detecting short - circuit faults of distribution network cables combined with the power supply network topology relationship provided by the present application;

[0028] Figure 2 It is a flowchart of the relationship of the detection indexes for short - circuit faults of distribution network cables. Detailed Embodiments

[0029] In order to further elaborate on the technical means and effects adopted by the present application to achieve the intended invention purpose, the following, in combination with the drawings and preferred embodiments, details the specific embodiments, structures, features and effects of the method for detecting short - circuit faults of distribution network cables combined with the power supply network topology relationship proposed according to the present application. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. In addition, the specific features, structures or characteristics in one or more embodiments can be combined in any suitable form.

[0030] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present application belongs.

[0031] The following specifically describes the specific solution of the method for detecting short - circuit faults of distribution network cables combined with the power supply network topology relationship provided by the present application with reference to the drawings.

[0032] A method for detecting short - circuit faults of distribution network cables combined with the power supply network topology relationship provided by an embodiment of the present application. Specifically, the following method for detecting short - circuit faults of distribution network cables combined with the power supply network topology relationship is provided. Please refer to Figure 1 , and the method includes the following steps:

[0033] Step S001: Collect the three - phase currents, three - phase voltages of each node at each moment in the distribution network, and the impedance of each node.

[0034] In this embodiment, distributed sensors are deployed at each node in the distribution network to continuously collect the three-phase current, three-phase voltage, impedance of each node, and impedance of each line. Among them, the three-phase current, three-phase voltage, and impedance of the node are collected synchronously. In this embodiment, the sampling frequency is 20 kHz, which can be set by the implementer according to the actual situation and is not limited in this embodiment.

[0035] Step S002: Construct the node adjacency matrix at each moment by traversing the closed breaker paths in the distribution network; analyze the current differences between each node at each moment and its adjacent moments to determine the current mutation rate of each node at each moment.

[0036] During the operation of the distribution network, its topological structure is not static. The opening and closing operations of breakers, the access and withdrawal of distributed power sources, etc. will all cause the real-time change of the distribution network topological structure. Therefore, in order to accurately analyze the propagation path and scope of fault current and thus provide a reliable basis for fault diagnosis, it is necessary to construct an accurate and real-time updated power supply network topological structure.

[0037] This embodiment uses the breadth-first search algorithm to construct the real-time topological structure of the distribution network. During the construction process, the adjacency matrix will be updated in real time and the line impedance parameters will be marked for subsequent short-circuit fault feature analysis and fault location.

[0038] The outgoing switch of the substation is the starting point of the power output of the distribution network. Taking it as the root node and searching along the closed breaker path can comprehensively cover the energized lines and nodes in the distribution network. In this way, the physical connection relationship of the distribution network can be transformed into a mathematical matrix form, that is, the adjacency matrix. The elements in the adjacency matrix are used to represent the connection state between nodes. If there is a direct connection between two nodes, the corresponding matrix element is 1, otherwise it is 0.

[0039] Cable short-circuit faults usually cause a sharp change in current instantaneously. During normal operation, the current in the distribution network is relatively stable with a small change range. When a short-circuit fault occurs, the current near the fault point will increase rapidly. Therefore, in order to be able to keenly capture this rapid change and thus provide an important basis for the early detection of short-circuit faults. In this embodiment, the current mutation rate of each node at each moment is calculated, and the specific calculation method is as follows:

[0040] ; where is the current mutation rate of node Y at the j-th moment, is the value of the A-phase current of node Y at the j-th moment, is the value of the A-phase current of node Y at the (j - 1)-th moment, is the value of the B-phase current of node Y at the j-th moment, is the value of the B-phase current of node Y at the (j - 1)-th moment, is the C-phase current value of node Y at the j-th moment, is the C-phase current value of node Y at the (j - 1)-th moment.

[0041] The current mutation rate of node Y at the j-th moment reflects the possibility of a short-circuit fault occurring at node Y at the j-th moment. The greater the current mutation rate, the greater the possibility of a short-circuit fault occurring at node Y at the j-th moment.

[0042] Step S003: Obtain the primary fault characteristics of each node at each moment by using the zero-sequence transient energy and total harmonic distortion rate within the local time period of each node at each moment, in combination with the current mutation rate.

[0043] When the power system is operating normally, the waveforms of current and voltage are basically sinusoidal. However, a short-circuit fault will disrupt the normal operating state of the power system, resulting in a large number of harmonic components in the current and voltage. The harmonic distortion rate is used to measure the proportion of these harmonic components in the total power. By monitoring the change of the harmonic distortion rate, it is possible to determine whether a fault has occurred and the severity of the fault.

[0044] Therefore, in this embodiment, for node Y at the j-th moment, if node Y is not in a power-off state, all moments within 2 s before and 2 s after the j-th moment of node Y are used as the local time period of node Y at the j-th moment. If node Y is in a power-off state at the j-th moment, all moments within 4 s before the j-th moment of node Y are used as the local time period of node Y at the j-th moment. It should be noted that the local time period includes the j-th moment.

[0045] Calculate the total harmonic distortion rate of the three-phase voltage data of all moments within the local time period of node Y at the j-th moment as the total harmonic distortion rate of node Y at the j-th moment. Among them, the calculation of the total harmonic distortion rate is a well-known prior art, and this embodiment will not elaborate in detail here.

[0046] In a three-phase AC power system, the zero-sequence component mainly reflects the asymmetry of the power system. A short-circuit fault often causes asymmetry in the power system, resulting in zero-sequence current and zero-sequence voltage. The zero-sequence transient energy is the energy accumulation of the zero-sequence current and zero-sequence voltage during the transient process. It is very sensitive to asymmetric faults and can effectively detect and distinguish different types of asymmetric faults. Therefore, in this embodiment, based on the three-phase current and three-phase voltage of all moments within the local time period of node Y at the j-th moment, calculate the zero-sequence transient energy as the zero-sequence transient energy of node Y at the j-th moment. The calculation of the zero-sequence transient energy is a well-known prior art, and the specific process will not be elaborated.

[0047] Arrange the current mutation rate, total harmonic distortion rate, and zero-sequence transient energy of node Y at the j-th moment in sequence to form the primary fault feature of node Y at the j-th moment. The primary fault feature can comprehensively describe the fault state from multiple dimensions. Different fault types and fault locations may cause these features to exhibit different change patterns. Considering these features comprehensively can greatly improve the accuracy and reliability of fault diagnosis and reduce the possibility of misjudgment and missed judgment.

[0048] Adopt the same acquisition method as the primary fault feature of node Y at the j-th moment to obtain the primary fault features of each node at each moment.

[0049] Step S004: Based on the connection distribution between nodes in the node adjacency matrix, determine the topological connection degree of each node at each moment; use the line length and line impedance connected to each node at each moment to obtain the attenuation coefficient of each node at each moment; determine each fault isolation area in the distribution network through the connection state of the circuit breakers in the distribution network.

[0050] The fault features extracted above only simply describe the characteristics of the fault itself and do not consider the influence of the complex topological structure of the distribution network on the fault propagation. In the distribution network, after a fault occurs, the fault signal will propagate along the network topological structure, and nodes at different positions are affected by the fault to different degrees. Therefore, it is necessary to incorporate the topological information of the distribution network into the fault features so that the fused features can more comprehensively and accurately reflect the propagation characteristics of the fault in the entire distribution network, thereby improving the accuracy of fault diagnosis.

[0051] Based on the dynamic topological relationship of the distribution network, that is, the adjacency matrix at each moment, for each node, calculate the number of its connections with other nodes as the topological connection degree of each node. The topological connection degree of a node reflects the importance and position characteristics of the node in the distribution network topological structure. The higher the topological connection degree, the closer the connection between the node and other nodes, and the greater its role may be in the fault propagation process.

[0052] Determine the attenuation coefficient when the short-circuit fault propagates between nodes according to factors such as the distance between nodes and the line impedance. Generally speaking, the farther the distance and the greater the line impedance, the greater the attenuation degree of the fault propagation. In this embodiment, for each moment, the product of the sum of the actual lengths of all lines connected to each node and the impedance per unit length of the line is used as the attenuation coefficient of each node at each moment, so that in the process of short-circuit fault propagation, the fault features of nodes farther from the fault point are attenuated more significantly.

[0053] Secondly, the tripping of the circuit breaker in the distribution network is an important sign of the change in the distribution network topology. After detecting the circuit breaker tripping signal, it is necessary to quickly update the adjacency matrix to reflect the new topology. Specifically, if the circuit breaker at node Y trips, in this embodiment, node Y and all nodes directly connected to node Y are marked as the fault isolation area; if the circuit breaker between node Y and node X trips, then node Y and node X are marked as the fault isolation area. Marking the fault isolation area is to clarify the scope affected by the fault and improve the efficiency of fault detection.

[0054] Step S005: Using the hop count from each node to the nearest fault isolation area at each moment, combining the topological connectivity and the attenuation coefficient, and the impedance deviation of each node, determine the propagation attenuation weight of each node at each moment.

[0055] Comprehensively considering the topological connectivity and attenuation coefficient of each node at each moment, calculate the propagation attenuation weight for the fault characteristics of each node. For nodes with high topological connectivity and close to the fault point, the propagation attenuation weight of their fault characteristics should be relatively large, indicating that the fault characteristics of this node have a greater impact on the overall fault detection result; while for nodes with low topological connectivity or far from the fault point, the propagation attenuation weight of their fault characteristics should be relatively small.

[0056] Based on this, in this embodiment, calculate the propagation attenuation weight of each node at each moment, and the specific calculation method is as follows:

[0057] ; where is the propagation attenuation weight of node W at the j-th moment, is the topological connectivity of node W at the j-th moment, is the attenuation coefficient of node W at the j-th moment, is the hop count from node W to the nearest fault isolation area at the j-th moment, is the absolute value of the difference between the impedance measurement value and the nominal value of node W at the j-th moment.

[0058] It should be understood that the hop count reflects the topological distance between the node and the fault isolation area. The more hop counts, the farther the node is from the fault point, and the more lines and nodes the fault signal passes through during propagation, and the greater the attenuation degree. Impedance is one of the important factors affecting the propagation of fault current. In actual operation, due to reasons such as equipment aging and environmental factors, the impedance measurement value may deviate from the nominal value. This deviation will affect the propagation and attenuation of the fault signal. When is larger, it indicates that the actual change in impedance is larger, and the impact on the propagation of the fault signal is also larger. Therefore, the propagation attenuation weight of the node should be increased. The propagation attenuation weight represents the attenuation degree of the fault characteristics of the node during propagation. The larger the propagation attenuation weight, the smaller the contribution of the fault characteristics of this node to fault detection.

[0059] Step S006: Weight the primary fault features by the reciprocal of the propagation attenuation weight. Using the weighted primary fault features of all nodes at each moment, combined with the node adjacency matrix of the distribution network and the three-phase currents and three-phase voltages of all nodes, use a hybrid classifier to detect cable short-circuit faults in the distribution network.

[0060] In this embodiment, by analyzing the real-time power supply network topology relationship, the propagation attenuation weights of each node at each moment are obtained. Combining the primary fault features of each node at each moment, the cable short-circuit fault detection of the distribution network deeply integrates the power supply network topology relationship and real-time electrical parameters, breaking through the dependence on local information of traditional detection methods.

[0061] Based on this, in this embodiment, the product of the reciprocal of the propagation attenuation weight of each node at each moment and each element in the primary fault features is calculated as the weighted primary fault feature of each node at each moment. The weighted primary fault features of all nodes at each moment, the adjacency matrix at each moment, the three-phase currents and three-phase voltages of all nodes at each moment are used as the input of the hybrid classifier, and the output is the node where the short-circuit fault occurs, completing the detection and location of the cable short-circuit fault in the distribution network. The flow chart of the relationship between the cable short-circuit fault detection indexes in the distribution network is as Figure 2 shown.

[0062] It should be noted that the hybrid classifier in this embodiment specifically includes:

[0063] (1) Graph Neural Network (GNN). GNN is a neural network model specifically designed to process data with graph structures. In the distribution network fault detection scenario, the topology of the distribution network can be abstracted as a graph, where nodes represent various electrical devices in the distribution network, and edges represent electrical connections between nodes. GNN allows each node to interact and aggregate information with its adjacent nodes through a message passing mechanism.

[0064] Through the aggregation of the distribution network topology neighborhood features by GNN, the potential associations and mutual influences between nodes in the distribution network can be fully explored. In a complex distribution network topology, short-circuit fault signals often spread between different nodes through electrical connections. GNN can capture this propagation law, thereby more accurately judging the short-circuit fault possibility of each node.

[0065] (2) Deep Belief Network (DBN). DBN is a generative model based on deep learning, which is stacked by multiple Restricted Boltzmann Machines (RBM). In fault detection, DBN is mainly used to analyze the characteristics of time series waveforms. The waveforms of three-phase current and three-phase voltage changing with time in the distribution network contain rich fault information. DBN learns the characteristic patterns of these waveforms through an unsupervised pre-training and a supervised fine-tuning process. In the pre-training stage, each RBM learns the local features of the input data. By continuously adjusting its own weights and biases, the RBM can reconstruct the input data as accurately as possible. After stacking multiple RBMs, the entire DBN can learn the multi-level and abstract feature representations of the input data. In the fine-tuning stage, the DBN is supervised and learned using labeled training data so that it can accurately classify the fault types.

[0066] DBN can automatically extract the deep features in the time series waveform without manually designing complex feature extraction methods. It has high accuracy and robustness in classifying fault types, can process fault data under different working conditions, and can better identify the fault types even in the presence of noise and interference.

[0067] (3) D-S evidence theory fusion layer. D-S evidence theory is a method for processing the fusion of uncertain information. In fault detection, there is a certain degree of uncertainty in both the node-level fault probability output by the graph neural network layer and the fault type classification result output by the deep belief network. D-S evidence theory represents the support degree of different evidences for cable short-circuit faults by defining the basic probability assignment function (BPA). Then, the Dempster combination rule is used to fuse the BPAs of multiple evidences to calculate the combined BPA, so as to obtain a more accurate and reliable fault detection result.

[0068] The D-S evidence theory fusion layer can make full use of the respective advantages of GNN and DBN to organically fuse different types of information. It can effectively reduce the risk of misjudgment and missed judgment that may be generated by a single model and improve the overall accuracy and reliability of fault detection.

[0069] The input of the D-S evidence theory fusion layer is the fault probability and fault type of each node. These are the output results of the graph neural network layer and the deep belief network layer and serve as the input evidences for the D-S evidence theory fusion layer. The final short-circuit fault nodes will be output. By fusing two different types of evidences and comprehensively considering the topological neighborhood features and time series waveform features, a more accurate node short-circuit fault detection result can be obtained. Among them, the graph neural network, the deep belief network, and the D-S evidence theory are all existing well-known technologies, and the specific processes will not be elaborated.

[0070] In this embodiment, the three levels of the hybrid classifier work together, which can make full use of the topological structure information of the distribution network and the time-series waveform information of electrical parameters to achieve fast and accurate detection of cable short-circuit faults in the distribution network, providing a strong guarantee for the safe and stable operation of the distribution network.

[0071] It should be noted that the above sequence of embodiments of the present application is only for description and does not represent the superiority or inferiority of the embodiments. And the above specific embodiments of this specification have been described. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0072] The various embodiments in this specification are described in a progressive manner. For the same or similar parts between the various embodiments, reference can be made to each other. Each embodiment focuses on the differences from other embodiments.

[0073] The above-described embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them; modifying the technical solutions recorded in the foregoing embodiments, or equivalently replacing some of the technical features, does not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the various embodiments of the present application, and should all be included within the protection scope of the present application.

Claims

1. A method for detecting short-circuit faults in a distribution network cable in combination with the topological relationship of a power supply network, characterized in that: The method comprises the following steps: Collect the three-phase current, three-phase voltage and impedance of each node in the distribution network at each time; By traversing the closed circuit breaker paths in the distribution network, the node adjacency matrix at each time is constructed; the current difference between each node at each time and its neighboring time is analyzed to determine the current mutation rate of each node at each time; The primary fault characteristics of each node at each moment are obtained by combining the zero-sequence transient energy and the total harmonic distortion rate in the local time period of each node at each moment with the current mutation rate; The number of connections between each node and other nodes in the node adjacency matrix at each moment is determined as the topological connectivity of each node at each moment; the attenuation coefficient of each node at each moment is obtained by using the line length and line impedance connected to each node at each moment; and each fault isolation zone in the distribution network is determined by the connection status of the circuit breaker in the distribution network; Determine the propagation attenuation weight of each node at each moment by using the number of hops from each node to the nearest fault isolation area at each moment, combining the topological connectivity and the attenuation coefficient, and the impedance deviation of each node; The primary fault feature is weighted by the inverse of the propagation attenuation weight, and the weighted primary fault features of all nodes at each moment are used, combined with the node adjacency matrix of the distribution network and the three-phase current and three-phase voltage of all nodes, and a hybrid classifier is used to detect the distribution network cable short-circuit fault.

2. The method for detecting short-circuit faults in a distribution network cable in combination with the topological relationship of the power supply network according to claim 1, characterized in that: The determination of the current mutation rate includes: The difference between the in-phase current at each moment and the adjacent moments is calculated, and the current mutation rate is positively correlated with the difference and negatively correlated with the in-phase current at the adjacent moments.

3. The method for detecting short-circuit faults in a distribution network cable in combination with the topological relationship of the power supply network according to claim 2, characterized in that: The ratio of the difference to the same-phase current at the adjacent moments is calculated, and the current mutation rate is the average of the ratios of the three-phase currents at each moment.

4. The method for detecting short-circuit faults in a distribution network cable in combination with the topological relationship of the power supply network according to claim 1, characterized in that: The primary fault characteristics are composed of zero-sequence transient energy and total harmonic distortion rate in a local time period at each node at each moment, and current mutation rate at each node at each moment.

5. The method for detecting short-circuit faults in a distribution network cable in combination with the topological relationship of a power supply network according to claim 1, characterized in that: The attenuation coefficient is the product of the sum of the actual lengths of all lines connected to each node at each moment and the impedance per unit length of the line.

6. The method for detecting short-circuit faults in a distribution network cable in combination with the topological relationship of the power supply network according to claim 1, characterized in that: The determination of the fault isolation area includes: If the circuit breaker at node Y is disconnected, node Y and all nodes directly connected to node Y are marked as a fault isolation zone; if the circuit breaker between node Y and node X is disconnected, node Y and node X are marked as a fault isolation zone.

7. The method for detecting short-circuit faults in a distribution network cable in combination with the topological relationship of the power supply network according to claim 1, characterized in that: The determination of the propagation attenuation weight includes: The calculation result of an exponential function with the attenuation coefficient as the base and the number of hops as the exponent is obtained, and the product of the calculation result and the topological connectivity is calculated. The propagation attenuation weight is positively correlated with the product and the impedance deviation.

8. The method for detecting short-circuit faults in a distribution network cable in combination with the topological relationship of a power supply network according to claim 1, characterized in that: The impedance deviation is the absolute value of the difference between the impedance measurement value of each node at each moment and the nominal value.

9. The method for detecting short-circuit faults in a distribution network cable in combination with the topological relationship of a power supply network according to claim 1, characterized in that: The hybrid classifier consists of a graph neural network, a deep belief network, and a DS evidence theory fusion layer.

Images