A ring main unit fault diagnosis method, device and storage medium

Abstract

The application discloses a kind of ring network cabinet fault diagnosis method, equipment and storage medium, method includes: through the synchronous acquisition operation data of multiple source sensor group deployed on ring network cabinet;After the pre-processing of the collected multi-source data, respectively extract improved mel frequency cepstrum coefficient, transient voltage feature, ultrasonic feature and electrical feature based on high-order cumulant;The extracted features are spliced to form high-dimensional fusion feature vector, and the compact features are obtained by dimension reduction using sparse autoencoder;The improved aurora optimization algorithm is used to optimize the support vector machine, and the IPLO-SVM model is constructed to identify faults.The application can fully integrate multi-source information, efficiently extract fault features, achieve high-precision classification, and is suitable for intelligent operation and maintenance of distribution network ring network cabinet.

Description

Technical Field

[0001] This invention relates to the field of power equipment condition monitoring technology, specifically to a method, equipment, and storage medium for fault diagnosis of ring main units based on multi-source feature layer fusion and improved aurora optimization algorithm, which is applicable to intelligent operation and maintenance and health management of ring main units in distribution networks. Background Technology

[0002] Ring main units (RNBs) are core power supply control devices in distribution networks, undertaking crucial responsibilities for power distribution, line protection, and fault isolation. Their operational reliability directly impacts power quality and grid security. However, with the large-scale grid connection of high-proportion distributed photovoltaic, wind power, and electric vehicle charging facilities, the operating conditions of distribution networks are becoming increasingly complex. RNBs face serious challenges such as bidirectional power flow, frequent harmonic disturbances, and broadband oscillations. Simultaneously, due to their compact internal structure, limited heat dissipation, and complex operating environment, RNBs are prone to various faults during long-term operation, including insulation degradation, contact overheating, and mechanism jamming. This places higher demands on the intelligent sensing and online diagnostic capabilities of RNBs.

[0003] Existing methods for diagnosing ring main unit faults have the following main shortcomings: (1) Single information source: Most methods rely on only a single type of sensor data, such as only collecting acoustic signals or only collecting electrical signals. A single feature is difficult to fully characterize the operating status of the equipment, especially in the early stage of failure, where weak fault features are easily drowned out by environmental noise, resulting in limited diagnostic accuracy.

[0004] (2) Insufficient feature extraction capability: The acoustic signature signals generated by weak faults such as early partial discharge and contact overheating in ring main units are mainly concentrated in the high-frequency band and exhibit non-stationary and transient characteristics. The linear frequency-Mel frequency mapping relationship used by conventional Mel frequency cepstral coefficients (MFCC) has insufficient frequency resolution in the high-frequency band, resulting in weak high-order harmonic fault features being buried by environmental noise, low signal-to-noise ratio, and difficulty in effective extraction. Traditional signal processing methods (such as FFT and wavelet transform) have limited ability to represent complex non-stationary signals.

[0005] (3) Difficulty in classifier parameter optimization: The performance of machine learning models such as support vector machines (SVM) is highly dependent on the selection of penalty factors and kernel function parameters. Traditional parameter optimization methods (such as grid search and particle swarm optimization) are prone to getting trapped in local optima and have slow convergence speed, making it difficult to meet the needs of efficient global optimization of SVM parameters under multiple fault types and operating conditions in ring main units.

[0006] To address the problems in existing ring main unit (RNB) fault diagnosis methods, such as difficulty in extracting early weak fault features, insufficient generalization ability of diagnostic models, and poor real-time performance due to single information sources, fixed frequency band division signal processing methods, and parameter optimization algorithms that are prone to getting trapped in local optima, this invention proposes a RNB fault diagnosis method that can fully integrate multi-source information, efficiently extract fault features, and achieve high-precision classification. This method boasts excellent diagnostic accuracy and robustness, and its real-time performance is well-suited for engineering applications. Summary of the Invention

[0007] The main objective of this invention is to provide a fault diagnosis method for ring main units. This method involves simultaneously acquiring acoustic signature signals, partial discharge signals, and electrical signals from the ring main unit using multiple sensors; extracting improved Mel-frequency cepstral coefficient features, transient ground voltage features, ultrasonic features, and transient electrical features based on higher-order cumulants; constructing a feature layer fusion model and using a sparse autoencoder for dimensionality reduction to obtain low-dimensional, highly discriminative fusion features; finally, employing an improved polar light optimization (IPLO) algorithm with a dynamic back-learning strategy and adaptive convergence factor to globally optimize the penalty factor and kernel function parameters of the support vector machine (SVM), constructing an IPLO-SVM classifier to identify faults using the fusion features, thereby addressing the technical problems mentioned in the background.

[0008] The present invention solves the above-mentioned technical problems by adopting the following technical solutions: A method for diagnosing ring main unit faults involves performing the following steps using computer equipment: Step S1. Multi-source data acquisition: Simultaneously acquire operational data through multiple sensors deployed on the ring network cabinet; Step S2. Preprocessing and feature extraction: After filtering, denoising and normalizing the collected multi-source data, the improved Mel frequency cepstral coefficients of the acoustic signature signal, the statistical features of the transient ground voltage signal, the features of the ultrasonic signal and the transient features of the electrical signal based on higher-order cumulants are extracted respectively. Step S3. Feature layer fusion and dimensionality reduction: The extracted features of various types are concatenated to form a high-dimensional fused feature vector. The high-dimensional fused feature vector is then dimensionality reduced by a sparse autoencoder to obtain compact features. Step S4. Classifier Construction and Optimization: The improved Aurora Optimization (IPLO) algorithm, which introduces a dynamic back learning initialization strategy and an adaptive convergence factor, is used to optimize the support vector machine (SVM) classifier, and an IPLO-SVM model is constructed to identify fault types based on the dimensionality-reduced fused features. Step S5. Fault Identification and Output: Input the compact features as the fusion features to be diagnosed into the trained IPLO-SVM model, and output the fault category. The fault categories include: normal operation, partial discharge fault, contact overheating fault, and mechanism jamming fault.

[0009] Preferably, the plurality of sensors in step S1 includes: Acoustic emission sensors are installed on the outer wall of the ring main unit to collect acoustic signals during equipment operation and fault conditions. Transient ground voltage (TEV) sensors and ultrasonic sensors are installed in each compartment to collect partial discharge signals; Current transformers and voltage transformers are used to collect electrical signals of three-phase current and voltage, respectively. Preferably, in step S2: Extracting the improved Mel-frequency cepstral coefficients (IMFCC), the frequency band division function is as follows:

[0010] in, Frequency (unit: Hz) Sampling frequency, and These are adjustable adjustment coefficients, with values ​​ranging from α∈[10,50] to β∈[0.5,2.0]. This formula introduces a sinusoidal perturbation term into the traditional Mel frequency mapping, causing the filter bank to generate periodic fluctuations in the high-frequency band, thereby enhancing the sensitivity to high-frequency harmonic components in weak fault acoustic features and improving the signal-to-noise ratio of feature extraction. Peak value, pulse count, and average amplitude are extracted from transient ground voltage signals as statistical features. The center frequency, bandwidth, and energy attenuation coefficient of the ultrasonic signal are extracted as features. For electrical signals (three-phase current and voltage), calculate transient characteristic quantities based on fourth-order cumulants. It is used to capture sudden changes in current and voltage during a fault, and the formula is:

[0011] in, For signal sampling points, The number of sampling points. The mean of the signal. Standard deviation; This characteristic quantity has a natural ability to suppress Gaussian white noise. When a fault occurs, the current or voltage waveform will be distorted. The value will change significantly, clearly indicating the time and type of fault.

[0012] Preferably, step S3 concatenates the features to form a high-dimensional fused feature vector. The specific operating procedure is as follows:

[0013] in, To improve the Mel frequency cepstral coefficients, This is a transient ground voltage characteristic. These are characteristics of ultrasound. This is a high-order cumulative characteristic.

[0014] Preferably, step S3 employs a deep feature learning network based on a sparse autoencoder. This deep feature learning network consists of an encoder and a decoder, and by setting sparsity constraints, it automatically learns low-dimensional, highly discriminative compact features from high-dimensional fused features. Its encoding process is represented as follows:

[0015] in, It is a non-linear activation function. and These are the weight matrix and bias vector of the encoding layer, respectively. The sparse autoencoder can adopt a three-layer network structure including one hidden layer. The number of nodes in the hidden layer can be determined according to the input dimension, generally set to 1 / 3 to 1 / 2 of the input dimension, and KL divergence sparsity constraints and mean squared error loss function are set.

[0016] Preferably, the specific operation process of step S4 includes: The improved aurora optimization algorithm is used to simulate the motion of particles in the Earth's magnetic field, including three stages: particle engine motion, global walk, and particle collision. For the i-th particle in the current iteration population, its position on the j-th dimension decision variable is updated according to the following formula:

[0017] in, For particle velocity, For the random variable components of the global walk, , For adaptive weights, A random number uniformly distributed in the range [0, 1]. The initial positions of the particles before the update. The updated particle positions; Updated particle positions Search for key parameters of Support Vector Machine (SVM) (penalty factor c and kernel function parameter g) and construct a validation set for SVM; Finally, the classification accuracy of the Support Vector Machine (SVM) on the validation set was used as the fitness function, and the improved Aurora Optimization algorithm was applied to the SVM as a penalty factor. and radial basis kernel function parameters To find the best option, we have:

[0018] Constraints: , ; Objective function: The optimization objective is to maximize the classification margin and minimize the classification error; in, The fitness function value is used to measure the current penalty factor. and radial basis kernel function parameters The advantages and disadvantages of parameter combinations For the validation set, the parameters are and The classification accuracy of the trained Support Vector Machine (SVM) model. , Punishment factors The lower and upper bounds of the search, , These are the lower and upper bounds for searching the radial basis kernel function parameters, respectively.

[0019] Preferably, the improved aurora optimization algorithm employs a dynamic back-learning initialization operation during particle update, specifically including: During the population initialization phase, after generating the initial population, the dynamic reverse population of the initial population is calculated:

[0020] in, The first in the original population The position of each particle. For the first The dynamic reverse solution of the position of each particle and A random number in the range [0,1]. and To solve the spatial boundaries (lower and upper bounds). This represents the average position of the current population (population center). The original population and the reverse population are merged, and the N individuals with the best fitness are selected as the updated initial population.

[0021] Preferably, the improved aurora optimization algorithm also introduces an adaptive convergence factor during the particle update process, specifically including: Introducing improved adaptive weights during the global walk phase , Its formula is:

[0022]

[0023] in, This represents the current iteration number. This represents the maximum number of iterations. In the early stages of the iteration ( ), Larger When the size is small, the algorithm performs a global exploration with large step sizes; as iterations proceed, Gradually decrease, As the step size gradually increases, the algorithm automatically switches to fine-grained local search with small steps, effectively balancing global search and local exploration capabilities. The original Probability of Loop (PLO) algorithm is prone to getting trapped in local optima when optimizing complex parameters. This invention improves the quality of the initial population through a dynamic back-learning initialization strategy to avoid premature convergence. At the same time, it designs an adaptive convergence factor, emphasizing global exploration in the early stage of iteration and fine-grained local search in the later stage, thus effectively overcoming the shortcomings of the original algorithm, such as slow convergence speed and susceptibility to local optima.

[0024] In another aspect, the present invention also discloses a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the steps of the method described above.

[0025] In another aspect, the present invention also discloses a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of the method described above.

[0026] As can be seen from the above technical solution, the present invention provides a method for diagnosing ring main unit faults. Compared with the prior art, the present invention has the following advantages: 1. This invention integrates acoustic signature, partial discharge, and electrical multi-source signals at the feature layer, comprehensively utilizing fault characterization information from multiple physical quantities. This enables a complete characterization of the ring main unit's operating status, effectively overcoming the limitations of a single information source and significantly improving fault diagnosis accuracy.

[0027] 2. This invention proposes an improved Mel-frequency cepstral coefficient (MFCC) method and higher-order cumulant features. The improved MFCC method can optimize the frequency band division through sinusoidal perturbation terms, thereby enhancing the ability to capture weak fault acoustic signatures. The higher-order cumulant features can enhance the ability to capture weak fault features and suppress Gaussian noise, thus significantly improving the noise robustness of the diagnostic method.

[0028] 3. This invention improves the Aurora Optimization Algorithm by using dynamic back learning and adaptive convergence factor. It can improve the quality of the initial population through dynamic back learning strategy and balance global and local search capabilities through adaptive convergence factor, thereby achieving efficient optimization of Support Vector Machine (SVM) parameters, accelerating convergence speed and improving classification accuracy. As a result, the IPLO-SVM model outperforms the standard comparison model in both convergence speed and optimization accuracy.

[0029] 4. This invention can achieve lightweight feature dimensionality reduction and end-side deployable model structure, thereby meeting the real-time requirements of online monitoring of ring main units, thus having good engineering applicability and providing reliable support for intelligent operation and maintenance of distribution networks.

[0030] It should be understood that the descriptions in this section are not intended to identify key or essential features of embodiments of the invention, nor are they intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Of course, implementing any product of the invention does not necessarily require achieving all of the advantages described above simultaneously. Attached Figure Description

[0031] The accompanying drawings, which form part of this application, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is a schematic diagram of the overall process of the present invention.

[0032] Figure 2 This is a comparison of the convergence curves of the IPLO algorithm, PLO, and PSO optimization algorithms in this invention for SVM parameter optimization.

[0033] Figure 3 This is a confusion matrix diagram for fault diagnosis in a traditional standard SVM model.

[0034] Figure 4 This is a fault diagnosis confusion matrix diagram of the IPLO-SVM model of the present invention. Detailed Implementation

[0035] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Unless otherwise specified, the embodiments and features in the embodiments of this application can be combined with each other. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0036] For details in the embodiments, please refer to Figures 1 to 4 .

[0037] like Figure 1 As shown in the embodiment of the present invention, a method for diagnosing ring main unit faults includes the following steps: Step S1. Multi-source data acquisition: Simultaneously acquire operational data through multiple sensors deployed on the ring network cabinet.

[0038] Several of these sensors include: Acoustic emission sensors are installed on the outer wall of the ring main unit to collect acoustic signals during equipment operation and fault conditions. Transient ground voltage (TEV) sensors and ultrasonic sensors are installed in each compartment to collect partial discharge signals; Current transformers and voltage transformers are used to collect electrical signals of three-phase current and voltage.

[0039] Step S2. Preprocessing and Feature Extraction: After filtering, denoising and normalizing the collected multi-source data, fault features of the specified signals are extracted, including improved Mel frequency cepstral coefficients (IMFCC), transient ground voltage features, ultrasonic features and electrical features based on higher-order cumulants.

[0040] The frequency band division function for the improved Mel frequency cepstral coefficients is as follows:

[0041] in, Frequency (unit: Hz) Sampling frequency, and These are adjustable coefficients, with values ​​ranging from α∈[10,50] to β∈[0.5,2.0].

[0042] This formula introduces a sinusoidal perturbation term into the traditional Mel frequency mapping, causing the filter bank to generate periodic fluctuations in the high-frequency band, thereby enhancing the sensitivity to high-frequency harmonic components in weak fault acoustic features and improving the signal-to-noise ratio of feature extraction. Peak value, pulse count, and average amplitude are extracted from transient ground voltage signals as statistical features. The center frequency, bandwidth, and energy attenuation coefficient of the ultrasonic signal are extracted as features. For electrical signals (three-phase current and voltage), calculate transient characteristic quantities based on fourth-order cumulants. It is used to capture sudden changes in current and voltage during a fault, and has the following formula:

[0043] in, For signal sampling points, The number of sampling points. The mean of the signal. The standard deviation is denoted as .

[0044] This characteristic quantity has a natural ability to suppress Gaussian white noise. When a fault occurs, the current or voltage waveform will be distorted. The value will change significantly, clearly indicating the time and type of fault.

[0045] This invention proposes an improved Mel-frequency cepstral coefficient (MFCC) method and higher-order cumulant features. The improved MFCC method can optimize the frequency band division through sinusoidal perturbation terms, thereby enhancing the ability to capture weak fault acoustic signatures. The higher-order cumulant features can enhance the ability to capture weak fault features and suppress Gaussian noise, thus significantly improving the noise robustness of the diagnostic method.

[0046] Step S3. Feature layer fusion and dimensionality reduction: The extracted features of various types are concatenated to form a high-dimensional fused feature vector to construct a feature layer fusion model, and compact features are obtained by encoding with a sparse autoencoder.

[0047] The features are concatenated to form a high-dimensional fused feature vector. The specific operating procedure is as follows:

[0048] in, To improve the Mel frequency cepstral coefficients, This is a transient ground voltage characteristic. These are characteristics of ultrasound. This is a high-order cumulative characteristic.

[0049] Furthermore, in one embodiment, a deep feature learning network based on a sparse autoencoder is employed during the feature concatenation process. This deep feature learning network consists of an encoder and a decoder, and by setting sparsity constraints, it automatically learns low-dimensional, highly discriminative compact features from high-dimensional fused features. Its encoding process is represented as follows:

[0050] in, It is a non-linear activation function. and These are the weight matrix and bias vector of the encoding layer, respectively.

[0051] This invention integrates acoustic signatures, partial discharge signals, and multiple electrical sources in the feature layer. By comprehensively utilizing fault characterization information from various physical quantities, it can fully characterize the operating status of the ring main unit, thereby effectively overcoming the limitations of a single information source and significantly improving the accuracy of fault diagnosis.

[0052] Step S4. Classifier Construction and Optimization: The improved Aurora Optimization Algorithm (IPLO) with the introduction of a dynamic back learning initialization strategy and an adaptive convergence factor is used to optimize the Support Vector Machine (SVM) classifier and construct the IPLO-SVM model.

[0053] The specific operational procedures include: The improved aurora optimization algorithm is used to simulate the motion of particles in the Earth's magnetic field, including three stages: particle engine motion, global walk, and particle collision. For the i-th particle in the current iteration population, its position on the j-th dimension decision variable is updated by the following formula:

[0054] in, For particle velocity, For the random variable components of the global walk, A random number uniformly distributed in the range [0, 1]. , For adaptive weights, The initial positions of the particles before the update. The updated particle positions; Updated particle positions Initially search for key parameters of Support Vector Machine (SVM) (penalty factor c and kernel function parameter g), and construct a validation set for SVM; Finally, the classification accuracy of the Support Vector Machine (SVM) on the validation set was used as the fitness function, and the improved Aurora Optimization algorithm was applied to the SVM as a penalty factor. and radial basis kernel function parameters To find the best option, we have:

[0055] Constraints: , ; Objective function: The optimization objective is to maximize the classification margin and minimize the classification error; in, The fitness function value is used to measure the current penalty factor. and radial basis kernel function parameters The advantages and disadvantages of parameter combinations For the validation set, the parameters are and The classification accuracy of the trained Support Vector Machine (SVM) model. , Punishment factors The lower and upper bounds of the search, , These are the lower and upper bounds for searching the radial basis kernel function parameters, respectively.

[0056] In a further embodiment, the improved aurora optimization algorithm employs a dynamic reverse learning initialization operation during particle update, specifically including: During the population initialization phase, after generating the initial population, the dynamic reverse population of the initial population is calculated as follows:

[0057] in, The first in the original population The position of each particle. For the first The dynamic reverse solution of the position of each particle and A random number in the range [0,1]. and To solve the spatial boundaries (lower and upper bounds). This represents the average position of the current population (population center). By merging the original population with the reverse population, and selecting the N individuals with the best fitness as the updated initial population, the diversity and quality of the initial solutions are effectively enhanced.

[0058] Furthermore, in another embodiment, the improved aurora optimization algorithm introduces an adaptive convergence factor during the particle update process, specifically including: Improved adaptive weights introduced during the global walk phase , Its formula is:

[0059]

[0060] in, This represents the current iteration number. This represents the maximum number of iterations. In the early stages of the iteration ( ), Larger The size is relatively small, so the algorithm performs global exploration with large step sizes; As the iterations proceed... Gradually decrease, As the size of the step increases, the algorithm automatically switches to fine-grained local search with smaller step sizes, thus effectively balancing global search and local development capabilities.

[0061] The original PLO algorithm is prone to getting stuck in local optima when searching for complex parameters. This invention improves the quality of the initial population by using a dynamic back learning initialization strategy to avoid premature convergence. At the same time, it designs an adaptive convergence factor that focuses on global exploration in the early stage of iteration and local fine search in the later stage, thereby effectively overcoming the shortcomings of the original algorithm, such as slow convergence speed and easy getting stuck in local optima.

[0062] Step S5. Fault Identification and Output: Input the compact features as the fusion features to be diagnosed into the trained IPLO-SVM model, and output the fault categories, which include: normal operation, partial discharge fault, contact overheating fault, and mechanism jamming fault.

[0063] In summary, this invention can achieve lightweight feature dimensionality reduction and end-side deployable model structure, thereby meeting the real-time requirements of online monitoring of ring main units, thus possessing good engineering applicability and providing reliable support for intelligent operation and maintenance of distribution networks.

[0064] Based on the above operational procedures, in the specific data acquisition process, a 10kV distribution network ring main unit was used as the test object. This ring main unit includes a circuit breaker compartment, a cable compartment, and a busbar compartment, with a rated voltage of 12kV and a rated current of 630A. The specific implementation process is as follows: Step L1. Install sensors at key locations in the ring main unit. Wherein: One acoustic emission sensor (frequency range 50kHz-150kHz) is installed on the outer wall of both the circuit breaker compartment and the cable compartment to collect acoustic signals. One transient ground voltage sensor (bandwidth: 3MHz-100MHz) and one ultrasonic sensor (center frequency 40kHz) are installed on the inner wall of the cable room to collect partial discharge signals. Install three-phase current transformers and voltage transformers on the incoming line side, and set the sampling frequency to 1MHz; All sensors here are connected to the edge computing unit via a synchronous acquisition card.

[0065] Step L2. Through simulation experiments and field data collection, a dataset containing four states is constructed and randomly divided into training set, validation set, and test set according to a preset ratio, as shown in Table 1 below: Table 1: Examples of Training Set, Validation Set, and Test Set Partitions

[0066] Step L3. Following step S2, extract features for each sample: IMFCC extraction: The power spectrum is obtained by pre-emphasizing the voiceprint signal, framing it (frame length 25ms, frame shift 10ms), adding a Hamming window, and then performing a Fast Fourier Transform (FFT). The power spectrum is then passed through an improved Mel filter bank defined by the frequency band division function formula, with the number of filters set to 24. After taking the logarithm, a discrete cosine transform is performed to obtain 24-dimensional IMFCC features (the first 13 dimensions are taken after removing the 0th dimension). Parameter settings: =30, =1.2, =1MHz; TEV and ultrasonic features: Extracting 3D features of peak value (maximum value), pulse count (number of pulses exceeding the 50mV threshold), and average amplitude from TEV signals; Extracting 3D features of center frequency, -3dB bandwidth, and energy attenuation coefficient from ultrasonic signals; Higher-order cumulant characteristics: Calculate the fourth-order cumulant for the phase A current signal with a window length N=200 to obtain a one-dimensional feature. ; By concatenating the above features, we obtain a 13+3+3+1=20-dimensional fused feature vector. .

[0067] Step L4. Construct a three-layer sparse autoencoder: 20 nodes in the input layer, 8 nodes in the hidden layer, and 20 nodes in the output layer; Set sparsity parameters =0.05, sparse penalty weight =3; After pre-training for 200 rounds on the training set, extract the 8-dimensional compact features from the encoder output. .

[0068] Step L5. Set the parameters for the Improved Aurora Optimization Algorithm (IPLO): Population size =30, maximum number of iterations =100, the penalty factor for Support Vector Machine (SVM) Search range [0.1, 100], kernel function parameters Search range [0.01, 10]; The average accuracy of 5-fold cross-validation on the training set is used as the fitness function. like Figure 2 As shown, the IPLO algorithm converges to the optimal fitness in the 42nd generation, while the standard PLO converges in the 58th generation and PSO converges in the 67th generation. The convergence speed of IPLO is improved by 27.6% and 37.3%, respectively. The final optimized parameters are: =48.62, =0.53; The SVM model is trained on the entire training set using the optimal parameters, and the IPLO-SVM model for this data instance is finally obtained.

[0069] Step L6. On the test samples, the classification results of the IPLO-SVM model and the comparison model show that the diagnostic accuracy of the method of the present invention is significantly higher, and the overall recognition effect is far better than that of various comparison models, as shown in Table 2 below; Table 2: Comparison of Classification Results

[0070] Figure 3 This is a confusion matrix diagram for fault diagnosis in a traditional standard SVM model. Figure 4 The diagram shows the fault diagnosis confusion matrix of the IPLO-SVM model of this invention. As can be seen from the comparison, the model of this invention significantly reduces the identification error rate in each fault category.

[0071] From Table 2, Figure 3 and Figure 4 It can be seen that all four types of fault states can be effectively distinguished, the model file size is about 150KB, and even easily confused faults can be accurately identified, which fully verifies the superiority of the method of the present invention.

[0072] Step L7. The single diagnostic time on the edge computing unit is tested to be less than 15 milliseconds, which fully meets the real-time requirements of the ring main unit's online monitoring. The edge computing unit uses an embedded processor based on the ARM Cortex-A53 architecture, with a main frequency of 1.5GHz and 2GB of memory.

[0073] In summary, this embodiment fully demonstrates the effectiveness, accuracy, and engineering applicability of the method of the present invention in the fault diagnosis of ring main units.

[0074] In another aspect, the present invention also discloses a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the steps of the method described above.

[0075] In another aspect, the present invention also discloses a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor performs the steps of the method described above.

[0076] In another embodiment provided in this application, a computer program product containing instructions is also provided, which, when run on a computer, causes the computer to execute any of the ring main unit fault diagnosis methods in the above embodiments.

[0077] It is understood that the system provided in the embodiments of the present invention corresponds to the method provided in the embodiments of the present invention, and the explanation, examples and beneficial effects of the relevant content can be referred to the corresponding parts of the above methods.

[0078] This application also provides an electronic device, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, communication interface, and memory communicate with each other via the communication bus. Memory, used to store computer programs; The processor, when executing the program stored in the memory, implements the above-mentioned ring main unit fault diagnosis method.

[0079] The communication bus mentioned in the above-mentioned electronic devices can be a standard bus for interconnecting peripheral components or an extended industrial standard structure bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc.

[0080] The communication interface is used for communication between the aforementioned electronic devices and other devices.

[0081] The memory may include random access memory or non-volatile memory, such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.

[0082] The processors mentioned above can be general-purpose processors, including central processing units, network processors, etc.; they can also be digital signal processors, application-specific integrated circuits, field-programmable gate arrays or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0083] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless means. The computer-readable storage medium can be any available medium accessible to a computer or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium, an optical medium, or a semiconductor medium, etc.

[0084] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0085] Furthermore, it should be noted that if any directional indication (such as up, down, left, right, front, back, etc.) is involved in the embodiments of the present invention, the directional indication is only used to explain the relative positional relationship and movement of each component in a specific posture. If the specific posture changes, the directional indication will also change accordingly.

[0086] Furthermore, those skilled in the art should understand that in the actual use of the embodiments of this application, there may be preset thresholds used as the basis for judging the corresponding technical solutions. These thresholds are conventional technical means commonly used in the field to implement functions such as state judgment, condition recognition, and control logic switching. The specific values, setting basis, value selection methods, determination methods, and adjustment rules of the thresholds involved in this technical solution are all conventional technical choices that can be reasonably determined by those skilled in the art based on conventional technical factors such as actual application scenarios, system working states, characteristics of the detection object, hardware performance parameters, and functional requirements, through conventional experiments, calibrations, and debugging. The specific setting and adjustment of the aforementioned thresholds will not cause this technical solution to be unimplementable as a whole, nor will it affect the realization of the core concept and the achievement of the technical effects of this technical solution.

[0087] Furthermore, if the embodiments of this invention involve descriptions such as "first" or "second," these descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, the meaning of "and / or" throughout the text includes three parallel solutions; for example, "A and / or B" includes solution A, solution B, or a solution where both A and B are satisfied simultaneously. Furthermore, in the embodiments of this invention, "multiple" refers to two or more. Moreover, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by this invention.

Claims

1. A method for diagnosing faults in a ring main unit, characterized in that, include: Step S1. Collect operational data synchronously using multiple sensors deployed on the ring main unit; Step S2. After filtering, denoising and normalizing the collected multi-source data, the improved Mel frequency cepstral coefficients, transient ground voltage characteristics, ultrasonic signal characteristics and transient characteristic quantities of electrical signals based on higher-order cumulants are extracted respectively. Step S3. The extracted features are concatenated to form a high-dimensional fusion feature vector. The high-dimensional fusion feature vector is then reduced in dimensionality using a sparse autoencoder to obtain compact features. Step S4. Combine the improved Aurora Optimization algorithm to optimize the Support Vector Machine classifier, which is used to construct the IPLO-SVM model for fault type identification of the dimensionality-reduced fused features; Step S5. Input the compact features as the fusion features to be diagnosed into the trained IPLO-SVM model, and output the fault category; The fault characteristics of the specified signal in step S2 include: Improved Mel-frequency cepstral coefficients are extracted from the voiceprint signal, and the frequency band division function is as follows: in, For frequency, Sampling frequency, and This is an adjustable adjustment coefficient; Peak value, pulse count, and average amplitude are extracted from transient ground voltage signals as statistical features. The center frequency, bandwidth, and energy attenuation coefficient of the ultrasonic signal are extracted as features. For electrical signals, calculate transient characteristic quantities based on fourth-order cumulants. It is used to capture sudden changes in current and voltage during a fault, and its formula is: in, For signal sampling points, The number of sampling points. The mean of the signal. The standard deviation is denoted as .

2. The ring main unit fault diagnosis method as described in claim 1, characterized in that, The multi-source sensor group in step S1 includes: Acoustic emission sensors are installed on the outer wall of the ring main unit to collect acoustic signals during equipment operation and fault conditions. Transient ground voltage sensors and ultrasonic sensors are installed in each compartment to collect partial discharge signals; Current transformers and voltage transformers are used to collect electrical signals of three-phase current and voltage.

3. The ring main unit fault diagnosis method as described in claim 1, characterized in that, Step S3 concatenates the features to form a high-dimensional fused feature vector. The specific operating procedures are as follows: in, To improve the Mel frequency cepstral coefficients, This is a transient ground voltage characteristic. These are characteristics of ultrasound. This is a high-order cumulative characteristic.

4. The ring main unit fault diagnosis method as described in claim 3, characterized in that, In step S3, a deep feature learning network based on a sparse autoencoder is used during the feature concatenation process. This deep feature learning network consists of an encoder and a decoder. By setting sparsity constraints, it learns low-dimensional, highly discriminative compact features from high-dimensional fused features. Its encoding process is represented as follows: in, It is a non-linear activation function. and These are the weight matrix and bias vector of the encoding layer, respectively.

5. The ring main unit fault diagnosis method as described in claim 1, characterized in that, The specific operation process of step S4 includes: The improved aurora optimization algorithm is used to simulate the motion of particles in the Earth's magnetic field, including three stages: particle engine motion, global walk, and particle collision. For the i-th particle in the current iteration population, its position on the j-th dimension decision variable is updated by the following formula: in, For particle velocity, For the random variable components of the global walk, , For adaptive weights, A random number uniformly distributed in [0,1]. The initial positions of the particles before the update. The updated particle positions; Updated particle positions Initially search for key parameters of the support vector machine and construct a validation set for the support vector machine; Finally, the classification accuracy of the support vector machine on the validation set is used as the fitness function, and the improved aurora optimization algorithm is used to apply a penalty factor to the support vector machine. and radial basis kernel function parameters To find the best option.

6. The ring main unit fault diagnosis method as described in claim 5, characterized in that, The improved aurora optimization algorithm performs a dynamic reverse learning initialization operation during the particle update process, specifically including: During the population initialization phase, after generating the initial population, the dynamic reverse population of the initial population is calculated as follows: in, The first in the original population The position of each particle. For the first The dynamic reverse solution of the position of each particle and A random number in the range [0,1]. and To solve the spatial boundary, As the population center; The original population and the reverse population are merged, and the N individuals with the best fitness are selected as the updated initial population.

7. The ring main unit fault diagnosis method as described in claim 5, characterized in that, The improved aurora optimization algorithm introduces an adaptive convergence factor during the particle update process, specifically including: Improved adaptive weights introduced during the global walk phase , Its formula is: in, This represents the current iteration number. This represents the maximum number of iterations.

8. A computer-readable storage medium, characterized in that, The device stores a computer program that, when executed by a processor, causes the processor to perform the steps of the method as described in any one of claims 1 to 7.

9. A computer device, characterized in that, It includes a memory and a processor, the memory storing a computer program that, when executed by the processor, causes the processor to perform the steps of the method as described in any one of claims 1 to 7.

Images