A distribution network fault classification method, system, device and medium

Abstract

The present invention relates to the technical field of distribution network fault management and discloses a distribution network fault classification method, system, device, and medium. The method obtains virtual measurement data based on synchronized vector data within the fault area under various fault types of the target distribution network. Wavelet transform is used to extract wavelet energy datasets of the synchronized vector data and the virtual measurement data, respectively. A fault classification model is constructed based on an attention mechanism and a probabilistic neural network model, and the fault classification model is trained using the wavelet energy dataset to obtain a target fault classification model. During the training process, a squeeze excitation algorithm is used to enhance the adaptability of the wavelet energy dataset. The parameters of the fault classification model are optimized based on the similarity score obtained using the attention mechanism. The obtained real-time wavelet energy dataset is input into the target fault classification model to obtain the fault classification results of the target distribution network. The method disclosed in the present application significantly improves the accuracy of distribution network fault classification.

Description

Technical Field

[0001] The present invention relates to the technical field of distribution network fault management, and in particular to a distribution network fault classification method, system, equipment and medium. Background Art

[0002] During normal distribution network operation, power fluctuations from distributed power sources (DGs) cause frequent changes in electrical quantities such as voltage and current, disrupting the relatively stable operation of traditional distribution networks. Once a fault occurs, these DGs will have a complex impact on the distribution of fault current and voltage, resulting in highly complex fault signals. Traditional fault classification methods, whether based on fixed rules developed through expert experience or relying on single feature extraction, struggle to accurately capture the complex characteristics of fault signals in complex multi-branch topologies and diverse fault scenarios. They are unable to comprehensively and flexibly cover all possible scenarios, significantly reducing classification accuracy and severely impacting the rapid recovery and power supply reliability of distribution network failures.

[0003] It can be seen that how to improve the accuracy of distribution network fault classification has become a technical problem that needs to be solved urgently by those skilled in the art. Summary of the Invention

[0004] The present invention provides a distribution network fault classification method, system, device and medium to solve the technical problem of how to improve the accuracy and adaptability of distribution network fault classification, thereby achieving the effect of improving the accuracy of distribution network fault classification.

[0005] In a first aspect, the present invention provides a method for classifying distribution network faults, the method comprising:

[0006] Acquire synchronous vector data of original signals in the fault area under various fault types of the target distribution network, and obtain virtual measurement data based on the synchronous vector data;

[0007] Using wavelet transform to extract wavelet energy data sets of the synchronous vector data and the virtual measurement data respectively, the extraction process is configured to perform multi-scale decomposition on the synchronous vector data and the virtual measurement data respectively to obtain corresponding high-frequency components, and calculate the corresponding wavelet energy data set based on the high-frequency components;

[0008] A fault classification model for the target distribution network is constructed based on an attention mechanism and a probabilistic neural network model, and the fault classification model is trained using the wavelet energy dataset to obtain a target fault classification model. During the training process, a squeeze excitation algorithm is used to enhance the adaptability of the fault classification model to the wavelet energy dataset, and parameters of the fault classification model are optimized based on a similarity score between the wavelet energy dataset and a Gaussian kernel function obtained using the attention mechanism; the Gaussian kernel function is applied to each node of the dynamic pattern layer of the probabilistic neural network model;

[0009] In the actual distribution network fault classification process, the obtained wavelet energy real-time data set is input into the target fault classification model to obtain the fault classification result of the target distribution network.

[0010] Preferably, the fault classification model of the target distribution network is constructed based on the attention mechanism and the probabilistic neural network model, and the fault classification model is trained using the wavelet energy dataset to obtain the target fault classification model, including:

[0011] Constructing an initial fault classification model based on a probabilistic neural network model, wherein the initial fault classification model comprises at least: a feature channel selection layer and the dynamic mode layer, wherein the number of nodes in the dynamic mode layer is equal to the number of fault types;

[0012] Introducing a squeeze excitation module in the feature channel selection layer, the squeeze excitation module generates a channel weight vector according to the wavelet energy data set, and adjusts the channel weight of the wavelet energy data set according to the channel weight vector;

[0013] Assigning an adaptive weight smoothing factor to each of the dynamic pattern layers, and inserting an attention scoring module after each of the dynamic pattern layers to obtain an improved fault classification model;

[0014] The wavelet energy data set is input into the improved fault classification model for training, and a heuristic algorithm is used to optimize the parameters of the improved fault classification model to obtain a target fault classification model.

[0015] Preferably, the calculation formula of the similarity score is:

[0016]

[0017] in, is the Gaussian kernel function, Score for attention, For the The attention weights of the dynamic pattern layer nodes, is the wavelet energy dataset to be classified, is the total number of attention weights.

[0018] Preferably, the method further comprises:

[0019] According to the acquired newly added distribution network fault type, the number of nodes in the dynamic mode layer of the target fault classification model is expanded.

[0020] Preferably, the step of obtaining synchronous vector data of original signals in the fault area under various fault types of the target distribution network and obtaining virtual measurement data based on the synchronous vector data includes:

[0021] Acquire synchronous vector data of original signals in the fault area of the target distribution network under various fault types collected by the synchronous vector measurement unit, wherein the synchronous vector data includes: three-way voltage signals, unidirectional current signals and zero-sequence current signals;

[0022] Based on the synchronous vector data and the topological structure model of the target power distribution network, virtual measurement data are generated, where the virtual measurement data include: a neutral point voltage signal and a positive sequence current signal.

[0023] Preferably, the step of extracting the wavelet energy data sets of the synchronous vector data and the virtual measurement data respectively by using wavelet transform includes:

[0024] Constructing a wavelet base library, and screening the best wavelet bases dynamically matching the various fault types from the wavelet base library to construct a mother wavelet library;

[0025] Sequentially selecting different mother wavelets from the mother wavelet library, and performing multi-scale decomposition on the synchronous vector data and the virtual measurement data according to the function of the mother wavelet, to obtain first high-frequency components of different frequency bands corresponding to the synchronous vector data and second high-frequency components of different frequency bands corresponding to the virtual measurement data;

[0026] The wavelet energies of the first high-frequency components in different frequency bands and the second high-frequency components in different frequency bands are calculated, and a wavelet energy data set is constructed according to the calculation results.

[0027] Preferably, the calculation formula of the wavelet energy is:

[0028]

[0029] in, is the number of decomposition layers, is the number of wavelet decomposition coefficients, is the wavelet energy, is the wavelet decomposition coefficient.

[0030] In a second aspect, the present invention further provides a distribution network fault classification system to implement the distribution network fault classification method described above, the system comprising: the system comprising: a data acquisition unit, a wavelet transform extraction unit, a model construction and training unit, and a fault classification unit;

[0031] The data acquisition unit is used to acquire synchronous vector data of the original signal in the fault area under various fault types of the target distribution network, and obtain virtual measurement data based on the synchronous vector data;

[0032] The wavelet transform extraction unit is configured to respectively extract wavelet energy datasets of the synchronous vector data and the virtual measurement data using wavelet transform, wherein the extraction process is configured to perform multi-scale decomposition on the synchronous vector data and the virtual measurement data to obtain corresponding high-frequency components, and calculate the corresponding wavelet energy dataset based on the high-frequency components;

[0033] The model construction and training unit is used to construct a fault classification model of the target distribution network based on the attention mechanism and the probabilistic neural network model, and use the wavelet energy data set to train the fault classification model to obtain a target fault classification model. During the training process, a squeeze excitation algorithm is used to enhance the adaptability of the fault classification model to the wavelet energy data set, and similarity scores between the wavelet energy data set obtained by the attention mechanism and the Gaussian kernel function of each node of the dynamic pattern layer of the probabilistic neural network model are used to optimize the parameters of the fault classification model.

[0034] The fault classification unit is used to input the obtained wavelet energy real-time data set during the actual distribution network fault classification process.

[0035] In a third aspect, the present invention also provides a computer device, which includes a memory, a processor and a transceiver, which are connected via a bus; the memory is used to store a set of computer program instructions and data, and transmit the stored data to the processor, and the processor executes the program instructions stored in the memory to execute the above-mentioned sensitive equipment failure probability assessment method.

[0036] In a fourth aspect, the present invention further provides a computer-readable storage medium, wherein a computer program is stored in the computer-readable storage medium. When the computer program is executed, the above-mentioned method for evaluating the failure probability of sensitive equipment is implemented.

[0037] This application provides a method, system, device, and medium for evaluating the probability of failure of sensitive equipment. Compared with the prior art, the embodiments of this application have the following beneficial effects:

[0038] The distribution network fault classification method disclosed in the present application embeds an attention mechanism and a squeeze excitation algorithm in a probabilistic neural network model, highlights the wavelet energy data set obtained based on different mother wavelet transforms, adapts to different types of fault characteristics, can accurately extract fault signal characteristics, and significantly improves the distribution network fault classification accuracy. It uses a heuristic algorithm to optimize the kernel width parameter of the model to achieve dynamic adjustment of the fault classification model, designs the dynamic mode layer as an expandable node network, and can learn new fault types online to adapt to complex working conditions such as distributed power supply access and grid topology changes. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 This is a schematic diagram of the steps of a distribution network fault classification method provided by a preferred embodiment of the present invention;

[0040] Figure 2 This is a schematic diagram of a model training process provided by a preferred embodiment of the present invention;

[0041] Figure 3 Schematic diagram of a confusion matrix of training samples provided by a preferred embodiment of the present invention;

[0042] Figure 4 is a schematic diagram of the classification results of a test set provided by a preferred embodiment of the present invention;

[0043] Figure 5 This is a structural diagram of a distribution network fault classification system provided by a preferred embodiment of the present invention;

[0044] Figure 6 1 is a diagram showing the internal structure of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0045] The following is a detailed explanation of the embodiments of the present invention in conjunction with the accompanying drawings. The embodiments are provided for illustrative purposes only and cannot be understood as limitations on the present invention. The accompanying drawings are for reference and illustration purposes only and do not constitute a limitation on the scope of patent protection of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention. In the description of the present invention, the terms "first", "second", "third", etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first", "second", "third", etc. may explicitly or implicitly include one or more of the features. In the description of the present invention, unless otherwise specified, the meaning of "multiple" is two or more.

[0046] In the description of the present invention, it should be noted that, unless otherwise expressly specified and limited, the terms "installed", "connected" and "connected" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or an indirect connection through an intermediate medium, or it can be a communication between the two components. The terms "vertical", "horizontal", "left", "right", "up", "down" and similar expressions used herein are for illustrative purposes only, and do not indicate or imply that the device or component referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention. The term "and / or" used herein includes any and all combinations of one or more related listed items. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0047] In describing the present invention, it should be noted that, unless otherwise defined, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art. The terms used in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. Those skilled in the art will understand the specific meanings of the above terms in the present invention in specific circumstances.

[0048] Wavelet transforms can effectively capture both high-frequency and low-frequency components of fault signals. However, the quality of feature extraction by wavelet transforms is highly dependent on the selection of the mother wavelet and the optimization of the decomposition parameters. When used independently, they struggle to effectively classify complex fault types. Probabilistic neural networks (PNNs), with their superior classification capabilities and fault tolerance, have shown great potential for fault classification. However, relying solely on PNNs for fault classification remains limited by their reliance on feature extraction and their ability to generalize across multiple fault scenarios.

[0049] In view of this, in an embodiment of the present invention, a method for classifying distribution network faults is provided. Figure 1 , the method comprising:

[0050] S1. Acquire synchronous vector data of the original signals within the fault area under various fault types of the target distribution network, and obtain virtual measurement data based on the synchronous vector data; the synchronous vector data includes: three-phase voltage signals, unidirectional current signals, and zero-sequence current signals; the virtual measurement data includes: neutral point voltage signals and positive-sequence current signals. In one embodiment of the present application, a synchronous vector measurement unit (PMU) device is used to collect three-phase voltage signals, three-phase current signals, and zero-sequence current signals within the fault area under various fault types of the target distribution network. The sampling frequency must meet the dynamic changes in the fault signal characteristics and ensure data synchronization. For the virtual measurement data, the topological structure model of the target distribution network is used to generate the neutral point voltage signal and positive-sequence current signal in real time based on the synchronous vector data, thereby providing more complete analysis data for fault classification.

[0051] With its high sampling rate and precise time synchronization, the synchronous vector measurement unit (PMU) device can collect voltage and current phasor data at each node in the power system in real time and with high precision, providing rich and accurate electrical measurement data support for fault classification.

[0052] Furthermore, the synchronous vector data and virtual measurement data are normalized to remove the DC component and noise interference in the signal to ensure the integrity and accuracy of the data.

[0053] S2. Wavelet transform is used to extract the wavelet energy datasets of the synchronous vector data and the virtual measurement data, respectively. The extraction process is configured to perform multi-scale decomposition on the synchronous vector data and the virtual measurement data to obtain corresponding high-frequency components, and calculate the corresponding wavelet energy dataset based on the high-frequency components. The key to wavelet transform lies in the selection of mother wavelets. In one embodiment of the present application, a wavelet basis library matching the distribution network fault is constructed. Based on the fault type, the optimal wavelet basis that dynamically matches various fault types is selected from the wavelet basis library. The optimal wavelet basis that can reflect various types of faults is summarized into a mother wavelet library. During the wavelet transform decomposition process, each mother wavelet is selected from the mother wavelet library in turn. Based on the function of each selected mother wavelet, multi-scale decomposition is performed on the synchronous vector data and the virtual measurement data, respectively, to obtain first high-frequency components of different frequency bands corresponding to the synchronous vector data and second high-frequency components of different frequency bands corresponding to the virtual measurement data. Based on the first high-frequency components and the second high-frequency components, the wavelet energy of the different frequency bands corresponding to the synchronous vector data and the virtual measurement data is calculated. The wavelet energy calculation formula is:

[0054]

[0055] in, is the number of decomposition layers, is the number of wavelet decomposition coefficients, is the wavelet energy, is the wavelet decomposition coefficient.

[0056] The wavelet energies of different frequency bands calculated under different mother wavelets are vectorized to obtain the wavelet energy data set.

[0057] S3. Construct a fault classification model for the target distribution network based on the attention mechanism and the probabilistic neural network model, and train the fault classification model using the wavelet energy dataset to obtain a target fault classification model. During the training process, a squeeze excitation algorithm is used to enhance the adaptability of the fault classification model to the wavelet energy dataset, and the parameters of the fault classification model are optimized based on the similarity score between the wavelet energy dataset and the Gaussian kernel function obtained using the attention mechanism. The Gaussian kernel function is applied to each node of the dynamic pattern layer of the probabilistic neural network model. The probabilistic neural network model is a neural network based on probabilistic statistics. It classifies input data by calculating the probability that the input data belongs to different categories. Its basic structure includes an input layer, a pattern layer, a summation layer, and an output layer. The input layer receives data features, the pattern layer matches the input data with patterns in the training data, the summation layer sums the matching results, and the output layer calculates the category probability based on the summation result. A fault classification model for the target distribution network is constructed based on an attention mechanism and a probabilistic neural network model. The model consists of five parts: an input layer, a feature channel selection layer, a dynamic pattern layer, a summation layer, and an output layer. The input layer receives a wavelet energy dataset extracted by wavelet transform. The feature channel selection layer dynamically selects channels within the input wavelet energy dataset to suppress noise or redundant information. In a preferred embodiment of the present application, a squeeze-excitation module (SENet) is added to the feature channel selection layer, and an attention scoring module is inserted after each node in the dynamic pattern layer.

[0058] SENet uses the Squeeze-Excitation algorithm to enhance the model's feature adaptability. The Squeeze operation performs global average pooling on the input wavelet energy dataset, compressing the features of each channel into a single value. This results in a one-dimensional vector with the same number of channels as the input wavelet energy dataset. This one-dimensional vector can be viewed as a summary of the global statistical information of the entire wavelet energy dataset across each channel. The Excitation operation feeds the one-dimensional vector obtained by the Squeeze operation into a fully connected neural network, which typically consists of two fully connected layers and a Reluctant Unit (ReLU) activation function. The first fully connected layer reduces the dimensionality of the one-dimensional vector, while the second fully connected layer restores the dimensionality to the same number of channels as the input wavelet energy dataset. Finally, a Sigmoid activation function is used to output a channel weight vector with the same number of channels as the input wavelet energy dataset. This channel weight vector represents the importance of each channel. Feature rescaling multiplies the channel weight vector obtained by the Excitation operation by each channel of the original input wavelet energy dataset to adjust the channel weights of the wavelet energy dataset, thereby enhancing the features of important channels and suppressing those of unimportant channels.

[0059] The introduction of SENet can enhance the fault classification model's ability to represent features, better mine important features in the data, and suppress irrelevant or redundant features. This improves the distribution network fault classification model's ability to represent complex data features and enhances the accuracy of distribution network fault classification. SENet's relatively simple structure and minimal increase in computational effort and parameters effectively improves model performance without significantly increasing model complexity, offering a good price-performance ratio.

[0060] In a preferred embodiment of the present application, the dynamic model layer uses a Gaussian kernel function to calculate the matching relationship between the input wavelet energy data set and various fault types in the distribution network fault sample data. The probability density function expression is as follows:

[0061]

[0062] in, is the wavelet energy dataset to be classified, is the wavelet energy dataset used for training, is the kernel width parameter, is the sample dimension of the wavelet energy dataset, is the training set number.

[0063] Furthermore, an adaptive weight smoothing factor is assigned to each dynamic pattern layer, replacing the fixed value of the traditional algorithm, so as to optimize the initial fault classification model through back-propagation fine-tuning, highlight the training sample pattern most relevant to the current input, and improve the efficiency of distribution network fault classification.

[0064] Furthermore, an attention scoring module is inserted after each dynamic pattern layer to obtain an improved fault classification model. The attention scoring module calculates the similarity score between the wavelet energy dataset and the Gaussian kernel function of each node in the dynamic pattern layer. The calculation formula is as follows:

[0065]

[0066] in, is the Gaussian kernel function, Score for attention, For the The attention weights of the dynamic pattern layer nodes, is the total number of attention weights.

[0067] The summation layer calculates the posterior probability based on the probability density function of each distribution network fault type:

[0068]

[0069] in, Indicates the Class failure.

[0070] The output layer is used to output the distribution network fault type according to the posterior probability. The distribution network fault types of this application include at least single-phase grounding, two-phase grounding, two-phase-to-phase and three-phase grounding.

[0071] After the model is built, the wavelet energy data set is divided into a training set and a validation set. The training set is input into the improved fault classification model for training, and then the validation set is input into the trained improved fault classification model. The heuristic algorithm is used to optimize the parameters of the improved fault classification model to obtain the target fault classification model. In the preferred embodiment of this application, the kernel width parameter is mainly Optimization is performed to improve the classification accuracy and generalization ability of the target fault classification model.

[0072] In the preferred embodiment of the present application, the heuristic algorithm is a sparrow search algorithm, such as Figure 2 As shown, the specific process is as follows:

[0073] The kernel width parameter It is represented as the individual position of the sparrow search algorithm. The position of each individual is randomly distributed within the domain of definition. The domain of definition is .

[0074] Initialize the parameters of the sparrow search algorithm, including population size, maximum number of iterations, finder ratio, joiner ratio, warning value and safety threshold.

[0075] The fitness function is defined according to the classification performance of the target fault classification model, which is expressed as follows:

[0076]

[0077] Among them, the accuracy rate represents the classification accuracy of the target fault classification model on the validation set, and the goal is to minimize , that is, maximize the classification accuracy.

[0078] The forager position is updated, and the update rule is:

[0079]

[0080] in: Indicates forager In the The position in the iteration, Indicates forager In the The position in the iteration, is the total number of iterations, is the parameter that controls the step size, is a random number that follows a standard normal distribution.

[0081] The monitor position is updated and the monitor adjusts its position according to the fitness of the current solution:

[0082]

[0083] in, Indicates forager In the The position in the iteration, Indicates forager In the The position in the iteration, represents the position of the current optimal individual, is the proportionality coefficient, represents a random uniformly distributed disturbance.

[0084] The algorithm determines whether it has reached its termination condition. When the maximum number of iterations is reached or the fitness function changes by less than a set threshold, the sparrow search algorithm stops and outputs the optimal kernel width parameter, which is then applied to the target fault classification model. A heuristic algorithm is used to optimize the kernel width parameter, enabling dynamic adjustment of the fault classification model and improving the accuracy of distribution network fault classification.

[0085] S4. In the actual distribution network fault classification process, the obtained wavelet energy real-time data set is input into the target fault classification model to obtain the fault classification result of the target distribution network; when a fault occurs in the target distribution network, the real-time synchronous vector data of the original signal in the fault area is collected, and the real-time virtual measurement data is calculated based on the real-time synchronous vector data and the topological structure model of the target distribution network. The wavelet energy data set of the real-time synchronous vector data and the real-time virtual measurement data is extracted by wavelet transform, and the obtained wavelet energy real-time data set is input into the target fault classification model to obtain the fault type of the target distribution network fault.

[0086] In a preferred embodiment of the present application, when a new fault type occurs in the target distribution network, the number of nodes in the dynamic mode layer is expanded according to the fault type, and the model does not need to be retrained, thereby realizing online learning of the target fault classification model.

[0087] In a specific embodiment of the present application, based on a 10kV distribution network radial line simulation model, a fault classification study is conducted for different distribution network fault types, fault initial angles, and transition resistances, a wavelet energy data set sample is constructed, and the classification performance of the target fault classification model is verified.

[0088] First, a radial line simulation model of a 10kV distribution network was established to simulate the actual operating environment of the distribution network, including loads, distributed generation (DGs), and line parameter configurations. Four fault types were set on different lines: single-phase grounding, two-phase short circuit, two-phase grounding, and three-phase grounding. The initial fault angles were set from 0 to 120°, and the transient resistance was set from 10 to 500Ω.

[0089] Furthermore, under different fault types, fault initiation angles, and transition resistance conditions, three-phase fault voltage, three-phase fault current, and zero-sequence fault current signals were collected from each busbar measurement point. Based on the distribution network topology model, the neutral point voltage and positive-sequence current signals were calculated. Wavelet decomposition was performed on the three-phase fault voltage, three-phase fault current, zero-sequence fault current, neutral point voltage, and positive-sequence current signals. The wavelet energy of the high-frequency components in different frequency bands of each signal was extracted. This generated a wavelet energy dataset containing 1000 sets of different conditions, divided into a 70% training set and a 30% validation set.

[0090] Then, the fault classification model is trained using the training set, and the output is the fault type, where single-phase grounding is 1, two-phase interphase is 2, two-phase grounding is 3, and three-phase grounding is 4. After training, the confusion matrix of the training samples is intercepted (such as Figure 3 ) and the classification results of the test set (as shown in Figure 4The classification accuracy of the trained fault classification model on the validation set is 94.4% for single-phase grounding, 93.3% for two-phase grounding, 100% for two-phase-to-phase, and 98.9% for three-phase grounding, with an overall accuracy of 96.67%.

[0091] In a preferred embodiment of the present invention, synchronous vector data of original signals in a fault area under various fault types of a target distribution network are obtained, and virtual measurement data are obtained based on the synchronous vector data; wavelet energy data sets of the synchronous vector data and the virtual measurement data are respectively extracted using wavelet transform, and the extraction process is configured to perform multi-scale decomposition on the synchronous vector data and the virtual measurement data to obtain corresponding high-frequency components, and corresponding wavelet energy data sets are calculated based on the high-frequency components; a fault classification model of the target distribution network is constructed based on an attention mechanism and a probabilistic neural network model, and the fault classification model is trained using the wavelet energy data set to obtain a target fault classification model. During the training process, a squeezing excitation algorithm is used to enhance the adaptability of the fault classification model to the wavelet energy data set, and the parameters of the fault classification model are optimized using the similarity score between the wavelet energy data set obtained by the attention mechanism and the Gaussian kernel function of each node in the dynamic pattern layer of the probabilistic neural network model; in the actual distribution network fault classification process, the obtained wavelet energy real-time data set is input into the target fault classification model to obtain a fault classification result of the target distribution network. The distribution network fault classification method disclosed in this application embeds an attention mechanism and a squeeze excitation algorithm in a probabilistic neural network model, highlights the wavelet energy data set obtained based on different mother wavelet transforms, adapts to different types of fault characteristics, can accurately extract fault signal characteristics, and significantly improves the distribution network fault classification accuracy.

[0092] Accordingly, if Figure 5 As shown, based on a distribution network fault classification method, an embodiment of the present invention further provides a distribution network fault classification system, which implements the distribution network fault classification method disclosed in an embodiment of the present invention, including: a data acquisition unit 1, a wavelet transform extraction unit 2, a model construction and training unit 3 and a fault classification unit 4;

[0093] The data acquisition unit 1 is used to acquire synchronous vector data of the original signal in the fault area under various fault types of the target distribution network, and obtain virtual measurement data based on the synchronous vector data;

[0094] The wavelet transform extraction unit 2 is configured to respectively extract wavelet energy datasets of the synchronous vector data and the virtual measurement data using wavelet transform, wherein the extraction process is configured to perform multi-scale decomposition on the synchronous vector data and the virtual measurement data to obtain corresponding high-frequency components, and calculate the corresponding wavelet energy dataset based on the high-frequency components;

[0095] The model construction and training unit 3 is used to construct a fault classification model of the target distribution network based on the attention mechanism and the probabilistic neural network model, and use the wavelet energy data set to train the fault classification model to obtain a target fault classification model. During the training process, a squeeze excitation algorithm is used to enhance the adaptability of the fault classification model to the wavelet energy data set, and based on the similarity score between the wavelet energy data set and the Gaussian kernel function obtained by using the attention mechanism, the parameters of the fault classification model are optimized; the Gaussian kernel function is applied to each node of the dynamic mode layer of the probabilistic neural network model;

[0096] The fault classification unit 4 is used to input the obtained wavelet energy real-time data set into the target fault classification model during the actual distribution network fault classification process to obtain the fault classification result of the target distribution network.

[0097] For the specific definition of a distribution network fault classification system, please refer to the above-mentioned definition of a distribution network fault classification method, which will not be repeated here. Those skilled in the art will appreciate that the various modules and steps described in conjunction with the embodiments disclosed in the present invention can be implemented in hardware, software, or a combination of both. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0098] like Figure 6 As shown, an embodiment of the present invention provides a computer device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, the steps in the above-mentioned distribution network fault classification embodiment are implemented, for example Figure 1 Steps S1 to S4 described in .

[0099] Those skilled in the art will understand that the schematic Figure 6 These are merely examples of computer devices and do not constitute limitations on the computer device. The computer device may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, the computer device may also include input and output devices, network access devices, buses, etc.

[0100] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the computer device, connecting various parts of the entire computer device using various interfaces and lines.

[0101] The memory can be used to store the computer programs and / or modules. The processor implements the various functions of the computer device by running or executing the computer programs and / or modules stored in the memory and accessing the data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store an operating system and at least one application required for a function (such as a sound playback function, an image playback function, etc.); the data storage area may store data generated based on the use of the mobile phone (such as audio data, a phone book, etc.). In addition, the memory may include high-speed random access memory and non-volatile memory, such as a hard disk, internal memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one disk storage device, a flash memory device, or other volatile solid-state storage device.

[0102] If the module integrated into the computer device is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention can also implement all or part of the process steps in the above-mentioned method embodiments by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When executed by a processor, the computer program can implement the steps of each of the above-mentioned method embodiments. The computer program includes computer program code, which can be in source code form, object code form, executable file, or some intermediate form. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium.

[0103] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing the relevant hardware through a computer program. The program can be stored in a computer-readable storage medium, and when executed, the program can include the processes in the above-described method embodiments. The storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0104] Accordingly, an embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium includes a stored computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located is controlled to perform the steps in the distribution network fault classification of the above embodiment, for example Figure 1 Steps S1 to S4 described in .

[0105] In summary, the embodiments of the present application provide a distribution network fault classification method, system, device, and medium that solve the technical problem of how to improve the accuracy of distribution network fault classification. The method includes: obtaining synchronous vector data of original signals in a fault area under various fault types of a target distribution network, and obtaining virtual measurement data based on the synchronous vector data; using wavelet transform to extract wavelet energy data sets of the synchronous vector data and the virtual measurement data, respectively. The extraction process is configured to perform multi-scale decomposition on the synchronous vector data and the virtual measurement data to obtain corresponding high-frequency components, and calculate corresponding wavelet energy data sets based on the high-frequency components; constructing a fault classification model for the target distribution network based on an attention mechanism and a probabilistic neural network model, and training the fault classification model using the wavelet energy data set to obtain a target fault classification model. During the training process, a squeezing excitation algorithm is used to enhance the adaptability of the fault classification model to the wavelet energy data set, and the parameters of the fault classification model are optimized using the similarity score between the wavelet energy data set obtained by the attention mechanism and the Gaussian kernel function of each node in the dynamic pattern layer of the probabilistic neural network model; in the actual distribution network fault classification process, the obtained wavelet energy real-time data set is input into the target fault classification model to obtain a fault classification result for the target distribution network. The distribution network fault classification method disclosed in this application embeds an attention mechanism and a squeeze excitation algorithm in a probabilistic neural network model, highlights the wavelet energy data set obtained based on different mother wavelet transforms, adapts to different types of fault characteristics, can accurately extract fault signal characteristics, and significantly improves the distribution network fault classification accuracy.

[0106] Each embodiment in this specification is described in a progressive manner, and the same or similar parts of each embodiment can be directly referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. It should be noted that the various technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the various technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0107] The above-described embodiments merely represent several preferred implementations of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that a person skilled in the art could make several improvements and substitutions without departing from the technical principles of the present application, and such improvements and substitutions should also be considered within the scope of protection of the present application. Therefore, the scope of protection of the present patent application shall be based on the scope of protection of the claims.

Claims

1. A distribution network fault classification method, characterized in that: The method comprises: Acquire synchronous vector data of original signals in the fault area under various fault types of the target distribution network, and obtain virtual measurement data based on the synchronous vector data; Using wavelet transform to extract wavelet energy data sets of the synchronous vector data and the virtual measurement data respectively, the extraction process is configured to perform multi-scale decomposition on the synchronous vector data and the virtual measurement data respectively to obtain corresponding high-frequency components, and calculate the corresponding wavelet energy data set based on the high-frequency components; A fault classification model for the target distribution network is constructed based on an attention mechanism and a probabilistic neural network model, and the fault classification model is trained using the wavelet energy dataset to obtain a target fault classification model. During the training process, a squeeze excitation algorithm is used to enhance the adaptability of the fault classification model to the wavelet energy dataset, and parameters of the fault classification model are optimized based on a similarity score between the wavelet energy dataset and a Gaussian kernel function obtained using the attention mechanism; the Gaussian kernel function is applied to each node of the dynamic pattern layer of the probabilistic neural network model; In the actual distribution network fault classification process, the obtained wavelet energy real-time data set is input into the target fault classification model to obtain the fault classification result of the target distribution network; The fault classification model of the target distribution network is constructed based on the attention mechanism and the probabilistic neural network model, and the fault classification model is trained using the wavelet energy dataset to obtain a target fault classification model, including: Constructing an initial fault classification model based on a probabilistic neural network model, wherein the initial fault classification model comprises at least: a feature channel selection layer and the dynamic mode layer, wherein the number of nodes in the dynamic mode layer is equal to the number of fault types; Introducing a squeeze excitation module in the feature channel selection layer, the squeeze excitation module generates a channel weight vector according to the wavelet energy data set, and adjusts the channel weight of the wavelet energy data set according to the channel weight vector; Assigning an adaptive weight smoothing factor to each of the dynamic pattern layers and inserting an attention scoring module after each of the dynamic pattern layers to obtain an improved fault classification model, wherein the attention scoring module calculates a similarity score between the wavelet energy dataset and the Gaussian kernel function of each node in the dynamic pattern layer; Inputting the wavelet energy data set into the improved fault classification model for training, and optimizing the parameters of the improved fault classification model using a heuristic algorithm to obtain a target fault classification model; The calculation formula of the similarity score is: in, is the Gaussian kernel function, is the attention score function, For the The weight of the dynamic mode layer node, is the wavelet energy dataset to be classified, is the total number of nodes in the dynamic mode layer.

2. The distribution network fault classification method according to claim 1, characterized in that: The method further comprises: According to the acquired newly added distribution network fault type, the number of nodes in the dynamic mode layer of the target fault classification model is expanded.

3. The distribution network fault classification method according to claim 1, characterized in that: The step of obtaining synchronous vector data of original signals within a fault area under various fault types of the target distribution network and obtaining virtual measurement data based on the synchronous vector data includes: Acquire synchronous vector data of original signals in the fault area under various fault types of the target distribution network collected by the synchronous vector measurement unit, wherein the synchronous vector data includes: three-phase voltage signals, single-phase current signals and zero-sequence current signals; Based on the synchronous vector data and the topological structure model of the target power distribution network, virtual measurement data are generated, where the virtual measurement data include: a neutral point voltage signal and a positive sequence current signal.

4. The distribution network fault classification method according to claim 1, characterized in that: The step of extracting the wavelet energy data sets of the synchronous vector data and the virtual measurement data respectively by using wavelet transform includes: Constructing a wavelet base library, and screening the best wavelet bases dynamically matching the various fault types from the wavelet base library to construct a mother wavelet library; Sequentially selecting different mother wavelets from the mother wavelet library, and performing multi-scale decomposition on the synchronous vector data and the virtual measurement data according to the function of the mother wavelet, to obtain first high-frequency components of different frequency bands corresponding to the synchronous vector data and second high-frequency components of different frequency bands corresponding to the virtual measurement data; The wavelet energies of the first high-frequency components in different frequency bands and the second high-frequency components in different frequency bands are calculated, and a wavelet energy data set is constructed according to the calculation results.

5. The distribution network fault classification method according to claim 1, characterized in that: The calculation formula of the wavelet energy is: in, is the number of decomposition layers, is the number of wavelet decomposition coefficients, is the wavelet energy, is the wavelet decomposition coefficient, is the decomposition level index, is the index of wavelet decomposition coefficients.

6. A distribution network fault classification system, implementing the distribution network fault classification method according to any one of claims 1 to 5, characterized in that: The system includes: a data acquisition unit, a wavelet transform extraction unit, a model building and training unit, and a fault classification unit; The data acquisition unit is used to acquire synchronous vector data of the original signal in the fault area under various fault types of the target distribution network, and obtain virtual measurement data based on the synchronous vector data; The wavelet transform extraction unit is configured to respectively extract wavelet energy datasets of the synchronous vector data and the virtual measurement data using wavelet transform, wherein the extraction process is configured to perform multi-scale decomposition on the synchronous vector data and the virtual measurement data to obtain corresponding high-frequency components, and calculate the corresponding wavelet energy dataset based on the high-frequency components; The model construction and training unit is used to construct a fault classification model of the target distribution network based on the attention mechanism and the probabilistic neural network model, and use the wavelet energy data set to train the fault classification model to obtain a target fault classification model. During the training process, a squeeze excitation algorithm is used to enhance the adaptability of the fault classification model to the wavelet energy data set, and similarity scores between the wavelet energy data set obtained by the attention mechanism and the Gaussian kernel function of each node of the dynamic pattern layer of the probabilistic neural network model are used to optimize the parameters of the fault classification model. The Gaussian kernel function is applied to each node of the dynamic pattern layer of the probabilistic neural network model. The fault classification unit is used to input the obtained wavelet energy real-time data set into the target fault classification model during the actual distribution network fault classification process to obtain the fault classification result of the target distribution network; The fault classification model of the target distribution network is constructed based on the attention mechanism and the probabilistic neural network model, and the fault classification model is trained using the wavelet energy dataset to obtain a target fault classification model, including: Constructing an initial fault classification model based on a probabilistic neural network model, wherein the initial fault classification model comprises at least: a feature channel selection layer and the dynamic mode layer, wherein the number of nodes in the dynamic mode layer is equal to the number of fault types; Introducing a squeeze excitation module in the feature channel selection layer, the squeeze excitation module generates a channel weight vector according to the wavelet energy data set, and adjusts the channel weight of the wavelet energy data set according to the channel weight vector; Assigning an adaptive weight smoothing factor to each of the dynamic pattern layers and inserting an attention scoring module after each of the dynamic pattern layers to obtain an improved fault classification model, wherein the attention scoring module calculates a similarity score between the wavelet energy dataset and the Gaussian kernel function of each node in the dynamic pattern layer; Inputting the wavelet energy data set into the improved fault classification model for training, and optimizing the parameters of the improved fault classification model using a heuristic algorithm to obtain a target fault classification model; The calculation formula of the similarity score is: in, is the Gaussian kernel function, is the attention score function, For the The weight of the dynamic mode layer node, is the wavelet energy dataset to be classified, is the total number of nodes in the dynamic mode layer.

7. A computer device, characterized in that: The computer device includes a memory, a processor and a transceiver, which are connected via a bus; the memory is used to store a set of computer program instructions and data, and transmit the stored data to the processor, and the processor executes the program instructions stored in the memory to perform the distribution network fault classification method according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed, the distribution network fault classification method according to any one of claims 1 to 5 is implemented.

Images