Power distribution terminal defect detection method and device and computer program product

Abstract

The invention discloses a power distribution terminal defect detection method and device and a computer program product, and the method comprises the steps: building a power distribution terminal operation model based on a dynamic simulation platform, analyzing and determining defect factors affecting the operation state of a power distribution terminal, and building a power distribution terminal defect type classification system; optimizing a VAE model structure of the variational auto-encoder, replacing an original back propagation BP neural network in the VAE model with a GRU network, and constructing a GRU-VAE hybrid network model; and on the basis of the GRU-VAE hybrid network model, feature extraction and anomaly detection are carried out on operation state data of the power distribution terminal equipment, a system fault state is identified through a preset fault determination threshold, and detected defects are classified according to the defect type classification system. According to the method, multiple defect types can be identified, the operation and maintenance requirements of a large-scale power distribution terminal can be met, powerful technical support is provided for stable operation of the power distribution Internet of Things, and the method has high practicability and popularization value.

Description

Technical Field

[0001] The present invention relates to the technical field of distribution networks, and particularly relates to a method, device and computer program product for detecting defects in distribution terminals. Background Art

[0002] With the rapid development of the distribution Internet of Things, distribution terminal devices show a trend of large-scale access, which poses a severe challenge to traditional distribution operation and maintenance methods. The existing operation and maintenance technologies face the following prominent problems: (1) The technical levels of distribution terminal devices vary widely, and the causes of defects are complex and diverse, resulting in low operation and maintenance efficiency; (2) The existing defect detection methods have obvious deficiencies in equipment information extraction and detection accuracy, and it is difficult to meet the operation and maintenance requirements of large-scale distribution terminals.

[0003] The existing defect detection methods are mainly divided into two categories: traditional machine learning methods and deep learning methods. In terms of traditional methods, researchers have proposed substation defect detection schemes such as the method combining histogram of oriented gradients and support vector machine, ultra-high frequency partial discharge detection technology, and time-frequency pulse conversion algorithm. However, these methods have obvious limitations: (1) The ability to extract device features is insufficient, and it is difficult to comprehensively reflect the operating status of distribution terminals; (2) The detection accuracy cannot meet the actual requirements; (3) The adaptability is limited, and it is impossible to effectively detect various defect types in multi-device scenarios.

[0004] Deep learning methods have shown significant advantages in the field of defect detection. Existing research has proposed a transmission line defect detection method based on an improved single-stage algorithm network, using the k-means algorithm for data clustering and realizing detection by using single-scale feature mapping; another research uses an improved phase grouping algorithm for visual appearance defect detection of transmission lines. However, these methods still have the following deficiencies: (1) The feature extraction dimension is single, and it is difficult to comprehensively reflect the operating characteristics of distribution terminals; (2) The detection range is limited, and it is mainly applicable to the detection of specific types of defects; (3) Insufficient consideration is given to the dynamic operating characteristics of distribution terminal devices.

[0005] Based on the above analysis, the existing technology is difficult to meet the defect detection requirements under the background of large-scale access of distribution terminals. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to provide a method for detecting defects in distribution terminals to improve the accuracy and adaptability of defect detection and meet the operation and maintenance requirements of large-scale distribution terminals.

[0007] To solve the above technical problem, the present invention provides a method for detecting defects in distribution terminals, including:

[0008] Step S1: Establish a distribution terminal operation model based on a dynamic simulation platform, analyze and determine the defect factors affecting the operation state of the distribution terminal, and establish a classification system for distribution terminal defect types.

[0009] Step S2: Optimize the variational autoencoder (VAE) model structure, replace the original backpropagation (BP) neural network in the VAE model with a gated recurrent unit (GRU) network, and construct a GRU-VAE hybrid network model.

[0010] Step S3: Based on the GRU-VAE hybrid network model, perform feature extraction and anomaly detection on the operation state data of distribution terminal equipment, identify the system fault state through a preset fault determination threshold, and classify the detected defects according to the defect type classification system.

[0011] Preferably, step S1 specifically includes:

[0012] Based on the similarity principle and four additional similarity conditions, establish dynamic simulation models for each module of the distribution terminal. Each module of the distribution terminal includes a central processing unit, a communication module, a collection module, an operation control loop, and a power supply module.

[0013] Simulate the operation characteristics of the distribution terminal through the dynamic simulation platform, and analyze multi-dimensional defect influencing factors. The influencing factors include external factors and internal factors.

[0014] Label the defect types according to the phenomena, including no-fault defect (A), telemetry signal defect (B), remote control failure (C), telemetry defect (D), terminal offline (E), and frequent switching on and off (F).

[0015] Preferably, in step S2, the GRU network is introduced to optimize the VAE model, which specifically includes:

[0016] Replace the BP neural network in the VAE model with a GRU network. The GRU network includes a reset gate, a hidden gate, and an update gate.

[0017] Optimize the model training process by variational inference and KL divergence to constrain the latent variable distribution and maximize the evidence lower bound.

[0018] Combine the time series processing ability of the GRU network and the generation ability of the VAE to improve the data reconstruction quality and feature learning ability.

[0019] Preferably, the calculation formula for the reconstruction error is:

[0020] E q(z|x) [logp(x′|z)]

[0021] where p(x′|z) is the distribution function of the decoder; x′ is the reconstructed data, and E q(z|x)It represents taking the expectation of the distribution q(z|x) of the latent variable z under the condition of the given input data x; logp(x′|z) represents the logarithmic likelihood probability of the reconstructed data x′ generated by the decoder under the condition of the given latent variable z.

[0022] Preferably, in the step S2, VAE uses the KL divergence to measure the difference between the distribution learned by the encoder and the prior distribution, and the final loss function is expressed as:

[0023]

[0024] where D KL is the KL divergence.

[0025] Preferably, the step S3 specifically includes:

[0026] Step S31, collect the data of the distribution terminal system according to the transmission of each sensor data, and use the support vector machine to fill in a small amount of missing data;

[0027] Step S32, perform data normalization processing, and divide the data set into a training set and a test set;

[0028] Step S33, input the processed training set data into the improved GRU-VAE model for training, input the output of the residual GRU network into the GMM fault classifier for fault classification, calculate the reconstruction error of the data, update the network parameters through backpropagation, and end the training when the model converges;

[0029] Step S34, input the test set data into the trained model, calculate the reconstruction error of each system data value, compare it with the threshold set by the model, determine whether the distribution terminal system has a fault, and judge the defect type according to the fault classification;

[0030] Step S35, output the defect detection result of the data distribution terminal according to the determination result.

[0031] Preferably, the GMM fault classifier classifies the defect types according to the latent variable features output by the GRU-VAE model.

[0032] The present invention also provides a distribution terminal defect detection device, including:

[0033] A dynamic simulation module, which is used to establish a distribution terminal operation model based on a dynamic simulation platform, analyze and determine the defect factors affecting the operation state of the distribution terminal, and establish a distribution terminal defect type classification system;

[0034] A model optimization module, which is used to optimize the variational autoencoder (VAE) model structure. It replaces the original backpropagation (BP) neural network in the VAE model with a gated recurrent unit (GRU) network to construct a GRU-VAE hybrid network model.

[0035] A defect detection module, which is used to perform feature extraction and anomaly detection on the operation status data of distribution terminal equipment based on the GRU-VAE hybrid network model. It identifies the system fault status through a preset fault determination threshold and classifies the detected defects according to the defect type classification system.

[0036] The present invention also provides a distribution terminal defect detection device, including:

[0037] One or more processors;

[0038] A memory;

[0039] One or more applications, where the one or more applications are stored in the memory and configured to be executed by the one or more processors. The one or more applications are configured to execute the above-mentioned distribution terminal defect detection method.

[0040] The present invention also provides a computer program product, including computer instructions, and the computer instructions direct a computer device to perform the operations corresponding to the above method.

[0041] Implementing the present invention has the following beneficial effects: By constructing a dynamic simulation system for distribution terminals based on a dynamic simulation platform, the present invention can accurately simulate the operation characteristics of distribution terminals, comprehensively analyze multi-dimensional factors affecting defects, and propose a clear defect type classification system, providing a reliable data basis for defect detection. Using a GRU network to optimize the VAE model significantly improves the model's feature extraction ability for multi-dimensional time series data and overcomes the limitations of traditional BP neural networks in processing complex non-linear data. The improved GRU-VAE model can efficiently analyze the operation status data of distribution terminal equipment, accurately determine system faults and classify defects, effectively improving the accuracy and adaptability of defect detection. The present invention can not only identify multiple defect types but also meet the operation and maintenance requirements of large-scale distribution terminals, providing strong technical support for the stable operation of the distribution Internet of Things, and having high practicality and promotion value. Description of the Drawings

[0042] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0043] Figure 1 It is a schematic flow chart of a method for detecting defects in a distribution terminal according to Embodiment 1 of the present invention.

[0044] Figure 2 It is a specific schematic flow chart of a method for detecting defects in a distribution terminal according to Embodiment 1 of the present invention.

[0045] Figure 3 It is a schematic structural diagram of a distribution terminal.

[0046] Figure 4 It is a schematic structural diagram of an improved VAE model in the embodiments of the present invention.

[0047] Figure 5 It is a schematic diagram of the detection results of terminal defects in an example.

[0048] Figure 6 It is a schematic diagram of the detection accuracy of each defect type in an example. Detailed implementation manners

[0049] The following descriptions of the embodiments are for reference to the accompanying drawings to illustrate specific embodiments in which the present invention can be implemented.

[0050] Please refer to Figure 1 As shown, Embodiment 1 of the present invention provides a method for detecting defects in a distribution terminal, including:

[0051] Step S1, establish an operation model of the distribution terminal based on a dynamic simulation platform, analyze and determine the defect factors affecting the operation state of the distribution terminal, and establish a classification system for the defect types of the distribution terminal;

[0052] Step S2, optimize the structure of the variational autoencoder VAE model, replace the original backpropagation BP neural network in the VAE model with a gated recurrent unit GRU network, and construct a GRU-VAE hybrid network model;

[0053] Step S3, based on the GRU-VAE hybrid network model, perform feature extraction and anomaly detection on the operation state data of the distribution terminal equipment, identify the system fault state through a preset fault determination threshold, and classify the detected defects according to the defect type classification system.

[0054] As can be seen from the above steps, in the embodiment of the present invention, by constructing a dynamic simulation system for distribution terminals, the operating characteristics of distribution terminals can be accurately reflected, providing a reliable data basis for defect detection; the GRU network is used to optimize the VAE model, significantly improving the feature extraction ability and the ability to process time-series data, and overcoming the limitations of traditional BP neural networks in processing complex non-linear data; based on the improved GRU-VAE hybrid network model, the efficient analysis of the operating state of distribution terminal equipment and the accurate classification of defects are realized, effectively improving the accuracy and adaptability of defect detection, and providing reliable technical support for the operation and maintenance of large-scale distribution terminals.

[0055] Specifically, please combine Figure 2 As shown in the figure, for experimental research, it is often necessary to model the experimental object. There are generally two types of model building, namely mathematical models and dynamic simulation models. Among them, dynamic simulation is to use the similarity principle to scale up or down the actual physical system in size to establish a dynamic simulation model that does not change the physical characteristics. Then, by analyzing the dynamic simulation model, some physical phenomena that cannot be directly observed using mathematical models can be intuitively observed. Therefore, the dynamic simulation method has an irreplaceable role compared with the mathematical model simulation method.

[0056] The dynamic simulation method mainly calculates the per-unit values of the system, analyzes and obtains the similarity criteria of the actual system, and then converts the actual system into a dynamic simulation system to establish a physical simulation system with consistent characteristics to analyze the characteristics of the actual physical system. In addition to meeting the three theorems of similarity theory, it also needs to meet four additional similarity conditions.

[0057] (1) For two composite systems, if their subsystems are respectively similar and the boundary conditions are also similar, then these two composite systems are similar.

[0058] (2) If the relative characteristics of the parameters of the linear system corresponding to a non-linear system coincide, then the similarity conditions of the linear system can be used for this non-linear system.

[0059] (3) If two systems have the same anisotropic and inhomogeneous characteristics, then the similarity conditions of the anisotropic and inhomogeneous systems can be derived from the similarity conditions of the isotropic and homogeneous systems.

[0060] (4) In systems with dissimilar geometric structures, they may exhibit similarity in physical processes.

[0061] The actual distribution terminal system is a strongly non-linear system. In the actual process of constructing the dynamic simulation, only the characteristic parameters of each component of the distribution terminal need to be considered, and only the characteristics of each component of the distribution terminal need to be accurately simulated. Then, the simulation of the distribution terminal system is accurate and effective.

[0062] Therefore, step S1 specifically constructs dynamic simulation models for each module of the distribution terminal based on the dynamic simulation method, the similarity principle, and additional conditions, analyzes the information acquisition process of each module of the distribution terminal in the dynamic simulation platform, simulates the operating characteristics of the distribution terminal, and provides data for the defect analysis of the distribution terminal.

[0063] As Figure 3 shown, the distribution terminal is decomposed into a central processing unit, a communication module, a collection module, an operation control loop, and a power supply module, and the functions and characteristics of each module are accurately modeled:

[0064] (1) The central processing unit is the core of the distribution terminal. Its main tasks are to provide action instructions to each module and to undertake the self-diagnosis task of the distribution terminal for faults.

[0065] (2) The communication module is the node for information interaction between the distribution terminal and other devices as well as the master station.

[0066] (3) The main task of the collection module is to collect telemetry and telecommand data of voltage transformers, current transformers, and switches.

[0067] (4) The operation control loop is used to implement the remote control function of the line switch and complete fault isolation.

[0068] (5) The power supply module includes a storage battery and a charging module connected to the grid side.

[0069] The failures and defects of the distribution terminal are caused by multiple factors. By collecting terminal defect data, summarizing and analyzing the key factors, the multi-dimensional influencing factors can be divided into external factors and internal factors, specifically including environmental temperature and humidity changes, wireless signal quality, on-site power outage faults, improper human operations, impacts of primary equipment, hardware operating status, software operating status, equipment upgrade and debugging, equipment family defects, and too long operating years. There are many types of defects in the distribution terminal. For the convenience of defect identification, the embodiments of the present invention select 5 types of defect types, label the defect types according to the defect phenomena, where the defect of the distribution terminal without faults is A, the telecommand defect is B, the remote control failure is C, the telemetry defect is D, the terminal offline is E, and the frequent switching on and off is F.

[0070] Through the dynamic simulation platform, the operating characteristics of the distribution terminal system are simulated, and simulation data highly consistent with the actual operation are generated, providing a high-quality data basis for defect analysis. This simulation method can comprehensively reflect the state changes of the distribution terminal under different operating conditions and provide reliable support for the training and verification of defect detection algorithms.

[0071] In summary, in step S1, a comprehensive and accurate defect analysis framework for distribution terminals is constructed through modular modeling, defect factor analysis, and dynamic simulation, laying a solid foundation for the implementation of subsequent defect detection algorithms.

[0072] In step S2, the GRU network is introduced to optimize the VAE model, replacing the BP neural network of the VEA model itself to improve the feature extraction ability of the VAE model.

[0073] In the variational autoencoder VAE, the main role of the encoder is to map complex input data to a latent variable space, and this process can be represented by the formula:

[0074]

[0075] In the formula: q(z|x) is the distribution function of the encoder, z is the latent feature, x is the input data, and μ(x), σ 2 (x) are the mean and variance of the latent variable respectively.

[0076] The decoding process can be represented as:

[0077]

[0078] In the formula: p(x′|z) is the distribution function of the decoder; x′ is the reconstructed data, are the mean and variance of the reconstructed data generated according to the latent variable z respectively.

[0079] Since the distribution of the latent feature is not directly observable, variational inference is used during the training process to approximate the true posterior distribution by maximizing the evidence lower bound (ELBO). To make the distribution learned by the encoder as close as possible to the prior distribution (usually the standard normal distribution), VAE uses the KL divergence to measure the difference between the two. The formula for the KL divergence is as follows:

[0080]

[0081] The final loss function can be represented as:

[0082]

[0083] In the formula: E q(z|x) [logp(x′|z)] is the reconstruction error. The reconstruction error E q(z|x) [logp(x′|z)] measures the similarity between the reconstructed data x′ generated by the decoder and the input data x under the condition of the given input data x. Specifically:

[0084] logp(x′|z) represents the log-likelihood probability of the reconstructed data x′, and the larger the value, the closer the reconstructed data is to the original data;

[0085] E q(z|x) denotes taking the expectation of the distribution of the latent variable z to ensure minimizing the average reconstruction error under different latent variable samplings.

[0086] By minimizing the reconstruction error, the VAE model can learn an effective feature representation of the input data and generate high-quality reconstructed data, thus providing a reliable basis for defect detection.

[0087] The reconstruction error and the KL divergence work together to ensure that the VAE can constrain the distribution of the latent variable while maintaining the data generation ability, making it as close as possible to the standard normal distribution.

[0088] The core mechanism of GRU consists of three parts: the reset gate, the update gate, and the hidden gate. The calculation process of the entire network is as follows:

[0089] 1) Reset gate: The gate for forgetting information, whose size is determined by the current input data and the hidden gate at the previous moment. The expression is as follows:

[0090] r t = σ(W r x t + U r h t-1 + b r )

[0091] In the formula, r t is the reset gate at time t; σ is the Sigmoid activation function; W r is the parameter matrix of the input data of the reset gate; x t is the input at time t; U r is the parameter matrix of the hidden gate at time t - 1; h t-1 is the hidden gate at time t - 1; b r is the overall bias size of the reset gate.

[0092] 2) Hidden gate: Its size is determined by the hidden gate at the previous moment and the reset gate at the current moment. The expression is as follows:

[0093]

[0094] In the formula, h t is the hidden gate at time t; is the dot product operation; U h is the parameter matrix of the hidden gate at time t - 1; W h is the parameter matrix of the input data of the hidden gate; b h is the overall bias size of the hidden gate.

[0095] 3) Update gate: The gate for retaining information, whose size is determined by the hidden gate at the previous moment and the input at the current moment. The expression is as follows:

[0096] u t = σ(W u x t + U u h t-1 + b u )

[0097] Wherein, u t is the update gate; W u is the input state parameter matrix at the current moment; U u is the input data parameter matrix of the update gate; h t-1 is the hidden gate at time t-1; b u is the overall bias size of the update gate.

[0098] Please refer to Figure 4 As shown, the GRU network combines the generative ability of VAE. The GRU network can not only effectively capture the temporal characteristics of complex data, but also improve the quality of data generation and reconstruction.

[0099] Replace the BP neural network in the VAE model itself with a GRU network to enhance the feature learning ability of the VAE model, improve the quality of data generation and reconstruction, and strengthen the object detection ability of the VAE model.

[0100] Step S3 specifically includes:

[0101] Step S31, collect the data of the distribution terminal system according to the data transmission of each sensor, and use the support vector machine to fill in a small amount of missing data;

[0102] Step S32, perform data normalization processing, and divide the data set into a training set and a test set;

[0103] Step S33, input the processed training set data into the improved GRU-VAE model for training, input the output of the residual GRU network into the GMM fault classifier for fault classification, calculate the reconstruction error of the data, update the network parameters through backpropagation, and end the training when the model converges;

[0104] Step S34, input the test set data into the trained model, calculate the reconstruction error of each system data value, compare it with the threshold set by the model, determine whether the distribution terminal system has a fault, and judge the defect type according to the fault classification;

[0105] Step S35, output the defect detection result of the data distribution terminal according to the determination result.

[0106] To verify the effectiveness of the distribution terminal defect detection method according to the embodiments of the present invention, a dynamic simulation test model of the distribution terminal is built by collecting sensor data of each module of the distribution terminal, and the simulated operation data of the distribution terminal is collected. The defect types of the distribution terminal defect detection are respectively selected as the remote signal defect, the remote control failure, the remote measurement defect, the terminal offline, and the frequent switching on and off. The defect detection is carried out on the simulated data of the distribution terminal, and the state of the distribution terminal and the defect type analysis are judged.

[0107] In the distribution terminal defect detection, the accuracy rate P A , the recall rate R, and the F1 value are usually used as three indicators to detect the effect of the method.

[0108]

[0109] In the formula: TP is the number of correctly detected defect samples; FP is the number of incorrectly detected normal samples; TN is the number of correctly detected normal samples; FN is the number of incorrectly detected defect samples.

[0110] To ensure the optimal threshold, the threshold is calculated by taking values in the range of [0.1 - 0.3], increasing by 0.01 each time. When the threshold is 0.24, the detection effect of the abnormal data is the best. The threshold is set to 0.24, and the number of iterations is 1000.

[0111] By collecting 500 sample data of the distribution terminal with 5 different defects in the dynamic simulation platform of the distribution terminal, the first 250 sample data are used as the training set of the improved variational autoencoder, and the last 250 sample data are used as the test set of the model. The final results are as Figure 5 , Figure 6 shown. The detection results of the method of the present invention in the distribution network terminal defect detection are basically consistent with the actual results, and only 2 misjudgment results appear, indicating that the present invention can dig out the required rules among a large number of terminal defect data by analyzing the operation characteristics and data defects of the distribution terminal, and effectively identify the defect types of the distribution terminal; at the same time, the F1 value of the detection results of the present invention all exceeds 0.95, and the average value is 0.979. Facing the long-term and multi-sample system data feature learning detection, it reflects excellent feature learning ability and rule mining ability, can effectively identify the distribution terminal defects and classify the types, and reflects the feasibility and practicability of the proposed detection method.

[0112] To verify the superiority of the defect detection method according to the embodiments of the present invention, the GRU-VAE model is compared and analyzed with the VAE model, the GA-LSSVM model, and the DR-VAE-AM model. The detection results of different algorithm models are shown in Table 1:

[0113] Table 1 Comparison of results of different algorithms

[0114]

[0115] As can be seen from Table 1, compared with the traditional VAE model, GA-LSSVM model, and DR-VAE-AM model, the proposed detection method shows higher accuracy, recall rate, and F1 value. The accuracy is increased by 7.8%, 0.4%, and 2.1% compared with the VAE model, GA-LSSVM model, and DR-VAE-AM model respectively; the recall rate is increased by 5.4%, 0.9%, and 3.5% respectively; the F1 value is increased by 7.1%, 2.3%, and 3.8% respectively. These results indicate that the present invention has obvious advantages in the defect detection of distribution terminals.

[0116] Corresponding to the distribution terminal defect detection method described in Embodiment 1 of the present invention, the present invention also provides a distribution terminal defect detection device, including:

[0117] A dynamic simulation module, configured to establish an operation model of a distribution terminal based on a dynamic simulation platform, analyze and determine defect factors affecting the operation state of the distribution terminal, and establish a classification system for distribution terminal defect types;

[0118] A model optimization module, configured to optimize the structure of the variational autoencoder (VAE) model, replace the original backpropagation (BP) neural network in the VAE model with a gated recurrent unit (GRU) network, and construct a GRU-VAE hybrid network model;

[0119] A defect detection module, configured to perform feature extraction and anomaly detection on the operation state data of distribution terminal devices based on the GRU-VAE hybrid network model, identify the system fault state through a preset fault determination threshold, and classify the detected defects according to the defect type classification system.

[0120] Corresponding to the distribution terminal defect detection method described in Embodiment 1 of the present invention, Embodiment 3 of the present invention also provides a distribution terminal defect detection device, including:

[0121] One or more processors;

[0122] A memory;

[0123] One or more applications, wherein the one or more applications are stored in the memory and are configured to be executed by the one or more processors, and the one or more applications are configured to execute the distribution terminal defect detection method described above.

[0124] Corresponding to the distribution terminal defect detection method described in Embodiment 1 of the present invention above, Embodiment 4 of the present invention also provides a computer program product, including computer instructions, and the computer instructions instruct a computer device to perform the operations corresponding to the distribution terminal defect detection method described in Embodiment 1 of the present invention above.

[0125] Preferably, the processor may be a Central Processing Unit (CPU), or may also be other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field-Programmable Gate Arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor, or the processor may also be any conventional processor. The processor is the control center of the device, and connects various parts of the device through various interfaces and circuits.

[0126] The memory mainly includes a program storage area and a data storage area. Among them, the program storage area can store an operating system, application programs required for at least one function, etc., and the data storage area can store relevant data, etc. In addition, the memory may be a high-speed random access memory, or may also be a non-volatile memory, such as a plug-in hard disk, a SmartMedia Card (SMC), a Secure Digital (SD) card, a Flash Card, etc., or the memory may also be other volatile solid-state storage devices.

[0127] It should be noted that the above device may include but is not limited to a processor and a memory, which can be understood by those skilled in the art.

[0128] For the working principle and process of this embodiment, please refer to the description of Embodiment 1 of the present invention above, and details will not be repeated here.

[0129] Compared with the prior art, the beneficial effects brought by the embodiments of the present invention are as follows: By constructing a dynamic simulation system for distribution terminals based on a dynamic simulation platform, the present invention can accurately simulate the operating characteristics of distribution terminals, comprehensively analyze multi-dimensional factors affecting defects, and propose a clear defect type classification system, providing a reliable data basis for defect detection. Using the GRU network to optimize the VAE model significantly improves the model's feature extraction ability for multi-dimensional time series data, overcoming the limitations of traditional BP neural networks in processing complex non-linear data. The improved GRU-VAE model can efficiently analyze the data of distribution terminal devices and operating states, accurately determine system faults and classify defects, effectively improving the accuracy and adaptability of defect detection. The present invention can not only identify multiple defect types, but also meet the operation and maintenance requirements of large-scale distribution terminals, providing strong technical support for the stable operation of the distribution Internet of Things, and having high practicality and promotion value.

[0130] The above-disclosed are only the preferred embodiments of the present invention. Of course, the scope of the rights of the present invention cannot be limited thereby. Therefore, equivalent changes made according to the claims of the present invention still fall within the scope covered by the present invention.

Claims

1. A method for detecting defects in a power distribution terminal, characterized in that: include: Step S1, establishing a distribution terminal operation model based on a dynamic simulation platform, analyzing and determining defect factors that affect the operation status of the distribution terminal, and establishing a distribution terminal defect type classification system; Step S2, optimizing the variational autoencoder VAE model structure, using the gated recurrent unit GRU network to replace the original back propagation BP neural network in the VAE model, and constructing a GRU-VAE hybrid network model; Step S3, based on the GRU-VAE hybrid network model, feature extraction and anomaly detection are performed on the operating status data of the power distribution terminal equipment, the system fault state is identified through a preset fault judgment threshold, and the detected defects are classified according to the defect type classification system.

2. The method according to claim 1, characterized in that: The step S1 specifically includes: Based on the similarity principle and four similarity additional conditions, a dynamic simulation model of each module of the distribution terminal is established. Each module of the distribution terminal includes a central processing unit, a communication module, a collection module, an operation control loop and a power supply module. The operating characteristics of the distribution terminal are simulated through a dynamic simulation platform to analyze the multi-dimensional defect influencing factors, including external factors and internal factors; The defect types are labeled according to the phenomenon, including no-fault defect (A), remote signal defect (B), remote control failure (C), telemetry defect (D), terminal offline (E) and frequent startup and shutdown (F).

3. The method according to claim 1, characterized in that The step S2 introduces the GRU network to optimize the VAE model, which specifically includes: A GRU network is used to replace the BP neural network in the VAE model, wherein the GRU network includes a reset gate, a hidden gate, and an update gate; Through variational inference and KL divergence constraints on latent variable distribution, the lower bound of evidence is maximized and the model training process is optimized; Combining the time series processing capability of the GRU network and the generation capability of VAE, the data reconstruction quality and feature learning capabilities are improved.

4. The method according to claim 3, characterized in that: The calculation formula of the reconstruction error is: E q(z|x) [logp(x ′ |z)] Among them, p(x ′ |z) is the distribution function of the decoder; x ′ is the reconstructed data, E q(z|x) It means that under the condition of given input data x, the distribution q(z|x) of the latent variable z is expected; logp(x ′ |z) represents the reconstructed data x generated by the decoder under the condition of given hidden variable z ′ The log-likelihood probability.

5. The method according to claim 4, characterized in that In step S2, VAE uses KL divergence to measure the difference between the distribution learned by the encoder and the prior distribution, and the final loss function is expressed as: Among them, D KL is the KL divergence.

6. The method according to claim 1, characterized in that The step S3 specifically includes: Step S31, collecting distribution terminal system data according to the data transmission of each sensor, and using support vector machine to fill in a small amount of missing data; Step S32, normalizing the data and dividing the data set into a training set and a test set; Step S33, input the processed training set data into the improved GRU-VAE model for training, input the residual GRU network output into the GMM fault classifier for fault classification, calculate the data reconstruction error, update the network parameters through back propagation, and end the training when the model converges; Step S34, input the test set data into the trained model, calculate the reconstruction error of each system data value, and compare it with the threshold set by the model to determine whether the power distribution terminal system has a fault, and determine the defect type according to the fault classification; Step S35, outputting the data distribution terminal defect detection result according to the determination result.

7. The method according to claim 6, characterized in that The GMM fault classifier classifies the defect type according to the latent variable features output by the GRU-VAE model.

8. A distribution terminal defect detection device, characterized in that: include: Dynamic simulation module, used to establish a distribution terminal operation model based on the dynamic simulation platform, analyze and determine the defect factors that affect the operation status of the distribution terminal, and establish a classification system for distribution terminal defect types; The model optimization module is used to optimize the variational autoencoder VAE model structure, using the gated recurrent unit GRU network to replace the original back-propagation BP neural network in the VAE model and construct a GRU-VAE hybrid network model; The defect detection module is used to perform feature extraction and anomaly detection on the operating status data of the power distribution terminal equipment based on the GRU-VAE hybrid network model, identify the system fault state through a preset fault judgment threshold, and classify the detected defects according to the defect type classification system.

9. A distribution terminal defect detection device, characterized in that: include: one or more processors; Memory; One or more applications, wherein the one or more applications are stored in the memory and are configured to be executed by the one or more processors, and the one or more applications are configured to execute the distribution terminal defect detection method as described in any one of claims 1 to 7.

10. A computer program product, characterized in that The method comprises computer instructions, wherein the computer instructions instruct a computer device to execute operations corresponding to the method according to any one of claims 1 to 7.

Images