Switch cabinet fault detection method, device and equipment and storage medium

Abstract

The invention relates to the technical field of switch cabinet detection, in particular to a switch cabinet fault detection method, device and equipment and a storage medium. In data processing and model construction, historical operation data are firstly acquired, local features are automatically extracted and dimensionality is reduced by means of a convolutional neural network and cooperative work of a convolutional layer and a pooling layer, data features are deeply mined, key information is accurately extracted, and a foundation is built for fault diagnosis; in the model optimization stage, a sparrow search algorithm is applied to efficiently search and iteratively update hyper-parameters in a hyper-parameter space, so that a fault diagnosis model structure and parameters can more reasonably and accurately fit a data and fault relationship, and the model performance is improved; and finally, acquiring real-time operation data, predicting by using the updated model, quickly outputting a result through convolution, pooling and full-connection layer operation, assisting operation and maintenance personnel to check hidden dangers in time, and ensuring stable, reliable, safe and efficient operation of a power system.

Description

Technical Field

[0001] The present invention relates to the technical field of switch cabinet detection, and in particular to a switch cabinet fault detection method, device, equipment and storage medium. Background Art

[0002] With the rapid development of science and technology, switchgear detection technology has become more and more mature. Fault detection for switchgear usually relies on manual experience or simple trial and error. Simple trial and error is one of the most commonly used methods. When building a fault diagnosis model, technicians try different parameter combinations to find a relatively suitable model setting. However, this method lacks systematicity and scientificity. It is difficult to find the optimal hyperparameter combination to optimize the fault diagnosis model, which limits the model performance and cannot fully realize its potential. The operating data of the switchgear is highly complex and diverse, involving information in multiple dimensions such as current, voltage, temperature, partial discharge, etc., and the parameters are interrelated and Influence each other. Faced with the ready-made switchgear operation data, the processing capacity of the existing technology is obviously insufficient. Due to the lack of effective hyperparameter optimization methods, the performance of the constructed model is limited, and it is impossible to fully mine the potential information and fault characteristics in the switchgear operation data. This makes the constructed fault diagnosis model have low fault recognition accuracy in actual applications, unable to timely and accurately predict possible faults of the switchgear, and unable to meet the power system's strict requirements for safe and stable operation of the switchgear; with the growing demand for electricity and the continuous expansion of the power grid, these problems are becoming more and more prominent, and there is an urgent need for a more scientific and efficient switchgear fault detection method and model optimization technology. Summary of the Invention

[0003] In order to solve the above-mentioned shortcomings in the prior art, the present invention proposes a switch cabinet fault detection method, system, device and storage medium.

[0004] In order to solve the above technical problems, the technical solutions adopted by the present invention are as follows:

[0005] A switchgear fault detection method comprises: obtaining historical switchgear operating data; constructing a primary convolutional neural network model based on a preset convolutional neural network algorithm and the historical switchgear operating data; calculating hyperparameters of the primary convolutional neural network model based on a preset sparrow search algorithm to obtain optimized hyperparameters; updating the primary neural network model according to the optimized hyperparameters to obtain a fault diagnosis model; obtaining real-time switchgear operating data; and predicting the real-time switchgear operating data according to the fault diagnosis model to obtain a prediction result. In the data processing and model building stage, by obtaining historical switchgear operation data and utilizing the powerful feature extraction capability of convolutional neural networks, potential features related to faults can be mined from complex multi-dimensional data. The collaborative work of its convolutional layer and pooling layer not only automatically extracts local features, but also effectively reduces dimensionality, reduces the amount of calculation while retaining key information, and provides a solid data foundation and feature support for subsequent fault diagnosis, greatly improving the depth and accuracy of mining the switchgear operation data features. In terms of model optimization, the sparrow search algorithm is used to calculate the hyperparameters of the primary convolutional neural network model. As a swarm intelligence optimization algorithm, it can efficiently search in the hyperparameter space, and obtain optimized hyperparameters by continuously iteratively updating the hyperparameters, thereby making the fault diagnosis model The structure and parameter settings of the model are more reasonable; the optimized fault diagnosis model can better fit the relationship between switchgear operation data and faults, effectively improving the performance and accuracy of the model; in actual applications, after obtaining real-time switchgear operation data, the updated fault diagnosis model is used for prediction. The model quickly and accurately outputs the prediction results through convolution, pooling and full connection layer operations on real-time data, and judges the current operating status of the switchgear and possible fault types. This enables operation and maintenance personnel to detect potential faults in a timely manner and take corresponding measures, greatly reducing the probability of faults, ensuring the stable operation of the power system, reducing power outages caused by faults, improving the reliability and stability of power supply, and providing a strong guarantee for the safe and efficient operation of the power system.

[0006] Furthermore, the method of constructing a primary convolutional neural network model based on a preset convolutional neural network algorithm and historical switchgear operating data includes: calculating the historical switchgear operating data based on the interquartile range method to obtain the first and second quartiles; identifying and processing the historical switchgear operating data based on preset abnormal conditions, the first and second quartiles to obtain optimized operating data; and constructing a primary convolutional neural network model based on the convolutional neural network algorithm and the optimized operating data. Calculating the first and second quartiles using the interquartile range method provides a key reference for anomaly identification, helps determine whether the data distribution is symmetrical and the fluctuation range, and helps grasp the overall characteristics of the data; then, identifying and processing the historical switchgear operating data based on the abnormal conditions and quartiles, greatly improving the data quality; and finally, constructing a primary convolutional neural network model based on the optimized operating data. Using high-quality data as high-quality training material, the model can more accurately learn the characteristics and patterns of the switchgear operating data, enhance the model's accuracy and generalization ability, effectively improve the prediction and fault diagnosis capabilities of the switchgear operating status, and provide strong support for the stable operation of the power system.

[0007] Furthermore, the updating of the primary neural network model according to the optimized hyperparameters to obtain the fault diagnosis model includes: obtaining a historical fault diagnosis sample set; performing a proximity analysis on the historical fault diagnosis sample set using a preset K-nearest neighbor algorithm to obtain a proximity value data set; performing interpolation calculation on the historical fault diagnosis sample set according to a linear interpolation method and a proximity value data set to obtain a balanced sample set; mapping the balanced sample set to a preset orthogonal coordinate axis based on a linear transformation method to obtain a mapped sample set; generating a covariance matrix according to the mapped sample set, and performing a feature analysis on the covariance matrix to obtain eigenvalues and a eigenvector set; sorting the eigenvector set according to the eigenvalues to obtain a sorted eigenvector set; generating a projection matrix according to the sorted eigenvector set and the optimized operation data; and updating the primary neural network model based on the projection matrix and the optimized hyperparameters to obtain the fault diagnosis model. After obtaining the historical fault diagnosis sample set, the K-nearest neighbor algorithm is used to find the similarity and correlation between samples, laying the foundation for subsequent processing; through linear interpolation calculation, the sample imbalance problem is successfully solved, making the model's ability to identify various fault types more balanced and avoiding diagnostic bias caused by differences in sample quantity; linear transformation is used to map the balanced sample set to a new space for in-depth analysis and processing, generate a covariance matrix and analyze the features, clarify the variance and main feature directions of the data in the direction of each eigenvector, and highlight the key feature directions after sorting; based on this, a projection matrix is generated to achieve data dimensionality reduction and feature extraction, remove redundant information, and focus on key features. Finally, the primary neural network model is updated in combination with optimized hyperparameters to make the model more focused on key features, significantly improve the performance of the fault diagnosis model, improve the accuracy and reliability of fault diagnosis, help to timely discover potential faults in switchgear, and ensure the stable operation of the power system.

[0008] Furthermore, the interpolation calculation of the historical fault diagnosis sample set is performed according to the linear interpolation method and the neighboring value data set to obtain a balanced sample set, including: performing type statistics on the historical fault diagnosis samples to obtain an interpolation range; selecting a neighboring value from the neighboring value data set according to a preset retrieval order, and selecting a historical fault diagnosis sample from the historical fault diagnosis sample set according to the neighboring value; performing interpolation calculation on the historical fault diagnosis samples according to the linear interpolation method, the interpolation range and the neighboring value to obtain a balanced sample; returning to select a neighboring value from the neighboring value data set according to the retrieval order until a preset iteration stop condition is met, and then obtaining all balanced samples to obtain a balanced sample set. The interpolation range is determined by performing type statistics on historical fault diagnosis samples, accurately locating fault types with scarce samples and indicating the direction for sample balance processing; adjacent values and historical fault diagnosis samples are selected according to the preset retrieval order, ensuring the orderliness and pertinence of sample selection, so that the interpolation calculation is based on meaningful sample pairs; the linear interpolation method is combined with the interpolation range to supplement fault types with few samples, optimize the sample distribution, make the proportion of various fault samples more balanced, avoid model deviation caused by uneven sample distribution, and provide a better data foundation for subsequent model training.

[0009] Furthermore, the method calculates the hyperparameters of the primary convolutional neural network model based on a preset sparrow search algorithm to obtain optimized hyperparameters, including: obtaining the hyperparameters in the primary convolutional neural network model, and obtaining the individual parameters of the sparrows according to the hyperparameters; performing moderate evaluation on the individual parameters of the sparrows to obtain the individual fitness set; sorting the individual fitness set to obtain the individual position with the best fitness and the individual position with the worst fitness; generating a fitness function according to the individual fitness set, the individual position with the best fitness and the individual position with the worst fitness; analyzing the optimization operation data based on the fitness function to obtain the population size and the number of iterations; iteratively optimizing the hyperparameters based on the sparrow search algorithm, the population size, the individual parameters of the sparrows and the number of iterations to obtain the optimized hyperparameters. Converting model hyperparameters into individual sparrow parameters paves the way for subsequent hyperparameter optimization using the sparrow search algorithm. Evaluating individual sparrow parameters yields a fitness set, using model performance metrics to measure individual performance, allowing the algorithm to focus on hyperparameter combinations that improve model performance. The sorted positions of individuals with the best and worst fitness provide key guidance for algorithm iteration, making the search process more directional. The generated fitness function accurately describes the relationship between fitness and hyperparameters, becoming the core basis for algorithm optimization. Optimizing operational data based on the fitness function determines the population size and number of iterations, fully accounting for data characteristics and problem complexity, avoiding blind searches, and improving algorithm efficiency and optimization effectiveness. During the iterative optimization process, individual sparrows continuously evolve based on the fitness function, ultimately finding the optimal hyperparameters to update the model. This effectively improves model performance, enhancing its accuracy and generalization capabilities.

[0010] Furthermore, the optimization operation data is analyzed based on the fitness function to obtain the population size and the number of iterations, including: dividing the optimization operation data into populations according to the fitness function to obtain a discoverer population; obtaining the position and self-state of the discoverer population; generating a safety value according to the position and self-state, and judging whether the safety value is greater than a preset warning value; when the safety value is greater than the warning value, performing a normal distribution calculation according to the position and self-state to obtain a first population area; dividing the optimization operation data into populations according to the fitness function to obtain a follower population; obtaining the hunger value of the follower population, and generating a second population area according to the hunger value, position and self-state; and calculating the population size and the number of iterations according to the first population area and the second population area. The fitness function is used to divide the discoverer and follower populations, laying the foundation for efficient search. For the discoverer population, the search strategy can be flexibly adjusted by calculating the safety value and comparing it with the warning value. When the safety value is greater than the warning value, the normal distribution is used to calculate the first population area, allowing the discoverer to randomly explore in a safe environment, increasing the probability of finding a better solution; considering the hunger value of the follower population, which reflects the degree of desire for a better solution, the second population area is generated according to the hunger value, position and its own status, so that the follower can move to a better area more specifically; combining the information of the two population areas to determine the population size and number of iterations provides strong support for optimizing hyperparameters and improving model performance.

[0011] Furthermore, the predicting of the real-time switchgear operating data according to the fault diagnosis model to obtain a prediction result includes: performing feature analysis on the real-time switchgear operating data based on a preset time series algorithm to obtain time domain features; performing spectrum analysis on the time domain features to obtain frequency domain features; performing variance analysis on the frequency domain features to obtain signal fluctuation features; performing thermal balance analysis on the real-time switchgear operating data to obtain thermal balance analysis data; obtaining partial discharge pulse current, integration time interval, and detection capacitance from the real-time switchgear operating data; and calculating the partial discharge amount based on a preset integral type formula, partial discharge pulse current, integration time interval, and detection capacitance. The expression of the partial discharge amount is as follows:

[0012] Where i(t) is the partial discharge pulse current, [t1, t2] is the integration time interval, t1 is the start time, t2 is the end time, and C is the detection capacitor. The time domain characteristics, frequency domain characteristics, signal fluctuation characteristics, thermal balance analysis data, and partial discharge amount are predicted based on the fault diagnosis model to obtain the prediction results. In terms of feature extraction, time domain features intuitively present the changes in data over time, laying the foundation for analyzing the operating status at different time points; frequency domain features can reveal hidden periodicity and frequency components, helping to capture specific fault signals; signal fluctuation characteristics reflect frequency domain stability, and large fluctuations may indicate equipment abnormalities; thermal balance analysis data can assist in judging the thermal status of the switchgear and discover fault signs in advance. The calculation of partial discharge is also very critical. It is an important indicator for evaluating the insulation performance of the switchgear. Accurate calculation helps to grasp the insulation status in a timely manner; finally, the fault diagnosis model integrates multi-dimensional feature information for prediction, and can accurately judge the operating status of the switchgear and possible fault types; this multi-dimensional analysis and precise prediction greatly improves the accuracy and reliability of fault diagnosis, can discover potential fault hazards in advance, and provide a scientific basis for operation and maintenance personnel to formulate maintenance plans, effectively reducing the risk of equipment damage and power outages, and ensuring the stable operation of the power system.

[0013] Furthermore, a switch cabinet fault detection device includes: a first data acquisition module for acquiring historical switch cabinet operation data; a model construction module for constructing a primary convolutional neural network model based on a preset convolutional neural network algorithm and historical switch cabinet operation data; a parameter calculation module for calculating the hyperparameters of the primary convolutional neural network model based on a preset sparrow search algorithm to obtain optimized hyperparameters; a model update module for updating the primary neural network model according to the optimized hyperparameters to obtain a fault diagnosis model; a second data acquisition module for acquiring real-time switch cabinet operation data; and a result prediction module for predicting the real-time switch cabinet operation data according to the fault diagnosis model to obtain a prediction result. During data processing and model building, historical operating data is obtained. Leveraging the powerful feature extraction capabilities of convolutional neural networks, its convolutional and pooling layers work together to automatically extract local features and reduce dimensionality, retaining key information, providing solid support for fault diagnosis, and improving the depth and accuracy of data feature mining. For model optimization, the sparrow search algorithm is used to efficiently search in the hyperparameter space, continuously iteratively updating hyperparameters to make the structure and parameters of the fault diagnosis model more reasonable, better fit the relationship between data and faults, and improve model performance and accuracy. Real-time operating data is obtained and predicted using the updated model. After convolution, pooling, and fully connected layer operations, the results are quickly and accurately output. Operations and maintenance personnel can promptly detect potential faults and take measures to reduce the probability of faults, minimize power outage losses, and ensure stable, reliable, safe, and efficient operation of the power system.

[0014] Furthermore, a switch cabinet fault detection device includes: a memory and at least one processor, wherein the memory stores instructions; at least one of the processors calls the instructions in the memory so that the switch cabinet fault detection device performs each step of a switch cabinet fault detection method as described in any one of the above.

[0015] Furthermore, a computer-readable storage medium is provided, wherein instructions are stored on the computer-readable storage medium, and when the instructions are executed by a processor, the steps of the switch cabinet fault detection method as described in any one of the above are implemented.

[0016] The beneficial effects of the switch cabinet fault detection method of the present invention are:

[0017] In the data processing and model building stage, by obtaining historical switchgear operation data and utilizing the powerful feature extraction capability of convolutional neural networks, potential features related to faults can be mined from complex multi-dimensional data. The collaborative work of its convolutional layer and pooling layer not only automatically extracts local features, but also effectively reduces dimensionality, reduces the amount of calculation while retaining key information, and provides a solid data foundation and feature support for subsequent fault diagnosis, greatly improving the depth and accuracy of mining the switchgear operation data features. In terms of model optimization, the sparrow search algorithm is used to calculate the hyperparameters of the primary convolutional neural network model. As a swarm intelligence optimization algorithm, it can efficiently search in the hyperparameter space, and obtain optimized hyperparameters by continuously iteratively updating the hyperparameters, thereby making the fault diagnosis model The structure and parameter settings of the model are more reasonable; the optimized fault diagnosis model can better fit the relationship between switchgear operation data and faults, effectively improving the performance and accuracy of the model; in actual applications, after obtaining real-time switchgear operation data, the updated fault diagnosis model is used for prediction. The model quickly and accurately outputs the prediction results through convolution, pooling and full connection layer operations on real-time data, and judges the current operating status of the switchgear and possible fault types. This enables operation and maintenance personnel to detect potential faults in a timely manner and take corresponding measures, greatly reducing the probability of faults, ensuring the stable operation of the power system, reducing power outages caused by faults, improving the reliability and stability of power supply, and providing a strong guarantee for the safe and efficient operation of the power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments with reference to the accompanying drawings, in which:

[0019] Figure 1 A first flow chart of a switch cabinet fault detection method provided by an embodiment of the present invention;

[0020] Figure 2A second flow chart of a switch cabinet fault detection method provided by an embodiment of the present invention;

[0021] Figure 3 A third flow chart of a switch cabinet fault detection method provided by an embodiment of the present invention;

[0022] Figure 4 A fourth flow chart of a switch cabinet fault detection method provided by an embodiment of the present invention;

[0023] Figure 5 A fifth flow chart of a switch cabinet fault detection method provided by an embodiment of the present invention;

[0024] Figure 6 A sixth flow chart of a switch cabinet fault detection method provided by an embodiment of the present invention;

[0025] Figure 7 A seventh flow chart of a switch cabinet fault detection method provided by an embodiment of the present invention;

[0026] Figure 8 A schematic structural diagram of a switch cabinet fault detection device provided by an embodiment of the present invention;

[0027] Figure 9 A structural schematic diagram of a switch cabinet fault detection device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0028] The technical solution of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0029] The terms "first," "second," "third," "fourth," and the like (if any) in the description and claims of the present invention and in the accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a particular order or precedence. It should be understood that the terms used in this manner are interchangeable where appropriate so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product, or apparatus that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to these processes, methods, products, or apparatus.

[0030] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1 , an embodiment of a switch cabinet fault detection method according to an embodiment of the present invention includes:

[0031] 101. Obtain historical switchgear operation data;

[0032] 102. A primary convolutional neural network model is constructed based on a preset convolutional neural network algorithm and historical switchgear operation data;

[0033] In this embodiment, the convolutional neural network has powerful feature extraction capabilities. Its convolution layer can automatically extract local features from the data by sliding the convolution kernel on the data. The pooling layer performs dimensionality reduction on the extracted features, reducing the amount of computation while retaining key information. When processing switchgear operation data, it can mine potential features related to faults from complex data.

[0034] 103. Calculate the hyperparameters of the primary convolutional neural network model based on the preset sparrow search algorithm to obtain the optimized hyperparameters;

[0035] In this embodiment, the sparrow search algorithm is a swarm intelligence optimization algorithm that can be used to efficiently search in the hyperparameter space of the primary convolutional neural network model and continuously iteratively update the hyperparameters to obtain optimized hyperparameters;

[0036] 104. Update the primary neural network model according to the optimized hyperparameters to obtain a fault diagnosis model;

[0037] In this embodiment, the primary neural network model is updated according to the optimized hyperparameters to obtain a fault diagnosis model. The optimized model is more reasonable in structure and parameter settings and can better fit the relationship between switchgear operation data and faults.

[0038] 105. Obtain real-time switchgear operation data;

[0039] In this embodiment, the real-time switchgear operation data can be measured by current sensors, voltage sensors, temperature sensors, partial discharge detection devices, etc.

[0040] 106. Predict the real-time switchgear operation data based on the fault diagnosis model to obtain the prediction results.

[0041] In this embodiment, after acquiring real-time switchgear operating data, the updated fault diagnosis model is used for prediction. The model takes the real-time data as input and, after a series of convolution, pooling, and fully connected layer operations, outputs the prediction result, i.e., the prediction result, to determine whether there is a fault in the current operating state of the switchgear and the possible fault type;

[0042] In this embodiment, during the data processing and model building stage, by acquiring historical switchgear operation data and utilizing the powerful feature extraction capability of the convolutional neural network, potential features related to faults can be mined from complex multi-dimensional data. The collaborative work of the convolutional layer and the pooling layer not only automatically extracts local features, but also effectively reduces the dimensionality, reduces the amount of calculation while retaining key information, and provides a solid data foundation and feature support for subsequent fault diagnosis, greatly improving the depth and accuracy of mining the switchgear operation data features. In terms of model optimization, the sparrow search algorithm is used to calculate the hyperparameters of the primary convolutional neural network model. As a swarm intelligence optimization algorithm, it can efficiently search in the hyperparameter space, and obtain optimized hyperparameters by continuously iteratively updating the hyperparameters, thereby optimizing the fault diagnosis. The structure and parameter settings of the fault diagnosis model are more reasonable; the optimized fault diagnosis model can better fit the relationship between switchgear operation data and faults, effectively improving the performance and accuracy of the model; in actual applications, after obtaining real-time switchgear operation data, the updated fault diagnosis model is used for prediction. The model quickly and accurately outputs the prediction results through convolution, pooling and full connection layer operations on real-time data, and judges the current operating status of the switchgear and possible fault types. This enables operation and maintenance personnel to detect potential faults in a timely manner and take corresponding measures, greatly reducing the probability of faults, ensuring the stable operation of the power system, reducing power outages caused by faults, improving the reliability and stability of power supply, and providing strong guarantees for the safe and efficient operation of the power system.

[0043] See also Figure 2 A second embodiment of a switch cabinet fault detection method according to an embodiment of the present invention includes:

[0044] 201. Calculate the historical switchgear operation data based on the interquartile range method to obtain the first quartile and the second quartile;

[0045] In this embodiment, these two quantiles can help us understand the central tendency and dispersion of the data, providing a key reference benchmark for subsequent anomaly identification. For example, these two quartiles can be used to preliminarily determine whether the distribution of historical switchgear operation data is symmetrical and the fluctuation range of the data.

[0046] 202. Identify and process abnormalities in historical switchgear operation data based on preset abnormal conditions, the first quartile, and the second quartile to obtain optimized operation data;

[0047] In this embodiment, let the first quartile be Q1, the second quartile be Q2, the IQR be the quartile coefficient, IQR = Q2 - Q1, and the abnormal condition be: a data point less than Q1 - 1.5 IQR is considered an outlier; the abnormality processing operation includes replacing the outlier with the interpolation of adjacent data points in the historical switchgear operation data to obtain optimized operation data; the advantage of replacing the outlier with the interpolation of adjacent data points in the historical switchgear operation data is that it fully utilizes the information of the original data and preserves the continuity and trend of the data as much as possible. Compared with simply deleting outliers, interpolation replacement does not lead to a reduction in the amount of data, avoids information loss due to missing data, and thus ensures the integrity and reliability of the data;

[0048] 203. A primary convolutional neural network model is constructed based on the convolutional neural network algorithm and optimized operating data.

[0049] In this embodiment, the optimized operating data obtained after abnormality identification and processing has significantly improved data quality. This data can better reflect the actual operating status of the switchgear and provide high-quality training materials for the convolutional neural network.

[0050] In this embodiment, the first and second quartiles are calculated using the interquartile range method, which provides a key reference for anomaly identification, helps determine whether the data distribution is symmetrical and the fluctuation range, and helps grasp the overall characteristics of the data; then, the historical switchgear operation data is processed for anomaly identification based on abnormal conditions and quartiles, greatly improving the data quality; finally, a primary convolutional neural network model is constructed based on the optimized operation data. The high-quality data provides high-quality training material for the model, enabling the model to more accurately learn the characteristics and laws of the switchgear operation data, enhance the model's accuracy and generalization ability, effectively improve the prediction of the switchgear operation status and fault diagnosis capabilities, and provide strong support for the stable operation of the power system.

[0051] See also Figure 3 A third embodiment of a switch cabinet fault detection method according to an embodiment of the present invention includes:

[0052] 301. Obtain a historical fault diagnosis sample set;

[0053] 302. Using a preset K-nearest neighbor algorithm to perform proximity analysis on the historical fault diagnosis sample set to obtain a proximity value data set;

[0054] In this embodiment, the K-nearest neighbor algorithm is used to perform proximity analysis, which can find the similarity and correlation between samples and obtain a proximity value data set;

[0055] 303. Perform interpolation calculation on the historical fault diagnosis sample set according to the linear interpolation method and the neighboring value data set to obtain a balanced sample set;

[0056] In this embodiment, this process can effectively solve the problem of sample imbalance, making the number of samples of different fault types more balanced, and avoiding the model's insufficient recognition ability of certain fault types due to too few samples of certain fault types;

[0057] 304. Map the equalized sample set to a preset orthogonal coordinate axis based on a linear transformation method to obtain a mapped sample set;

[0058] In this embodiment, spatial transformation of data is achieved, making the data easier to analyze and process in the new space;

[0059] 305. Generate a covariance matrix based on the mapped sample set, and perform feature analysis on the covariance matrix to obtain an eigenvalue and an eigenvector set;

[0060] In this embodiment, the eigenvalue represents the variance of the data in the direction of each eigenvector, and the eigenvector represents the main characteristic direction of the data;

[0061] 306. Sort the eigenvector set according to the eigenvalues to obtain a sorted eigenvector set;

[0062] In this embodiment, the feature vector set is sorted according to the eigenvalue to obtain a sorted feature vector set, which highlights the most important feature directions in the data;

[0063] 307. Generate a projection matrix based on the sorted feature vector set and the optimized operation data;

[0064] In this embodiment, the projection matrix can project the original data onto important feature directions to achieve data dimensionality reduction and feature extraction;

[0065] 308. Update the primary neural network model based on the projection matrix and optimized hyperparameters to obtain a fault diagnosis model.

[0066] In this embodiment, the primary neural network model is updated based on the projection matrix and optimized hyperparameters, so that the model can pay more attention to the key features of the data, thereby obtaining a fault diagnosis model with better performance.

[0067] In this embodiment, after obtaining the historical fault diagnosis sample set, the K-nearest neighbor algorithm is used to find the similarity and correlation between samples, laying the foundation for subsequent processing; through linear interpolation calculation, the sample imbalance problem is successfully solved, making the model's ability to identify various fault types more balanced, avoiding diagnostic bias caused by differences in sample quantity; linear transformation is used to map the balanced sample set to a new space, which is convenient for in-depth analysis and processing, generating a covariance matrix and analyzing the features, clarifying the variance and main feature directions of the data in the direction of each eigenvector, and highlighting the key feature directions after sorting; based on this, a projection matrix is generated to achieve data dimensionality reduction and feature extraction, remove redundant information, and focus on key features. Finally, the primary neural network model is updated in combination with optimized hyperparameters, so that the model is more focused on key features, significantly improving the performance of the fault diagnosis model, improving the accuracy and reliability of fault diagnosis, helping to timely discover potential faults in switchgear, and ensuring stable operation of the power system.

[0068] See also Figure 4 A fourth embodiment of a switch cabinet fault detection method according to an embodiment of the present invention includes:

[0069] 401. Perform type statistics on historical fault diagnosis samples to obtain an interpolation range;

[0070] In this embodiment, historical fault diagnosis samples are statistically analyzed to understand the distribution of samples of different fault types, thereby determining the interpolation range. This method can identify which fault types have fewer samples and need to be supplemented through interpolation, providing guidance for subsequent sample balancing. For example, if the number of samples of a certain type of fault only accounts for 10% of the total samples, far lower than that of other fault types, then this type of fault sample and its related neighboring values will become the focus of interpolation.

[0071] 402. Select a neighboring value from the neighboring value data set according to a preset search order, and select a historical fault diagnosis sample from the historical fault diagnosis sample set based on the neighboring value;

[0072] In this embodiment, this selection method ensures the orderliness and pertinence of sample selection, so that interpolation calculations can be performed based on meaningful sample pairs;

[0073] 403. Perform interpolation calculation on historical fault diagnosis samples according to a linear interpolation method, an interpolation range, and a neighboring value to obtain a balanced sample;

[0074] In this embodiment, combined with the previously determined interpolation range, fault types with a small number of samples can be supplemented in a targeted manner, so that the newly generated balanced samples are more reasonably distributed;

[0075] 404. Return and select a neighboring value from the neighboring value data set according to the search order until the preset iteration stop condition is met, thereby obtaining all balanced samples to obtain a balanced sample set;

[0076] In this embodiment, the iteration stopping condition is that the number of samples of a specific fault type reaches a certain limit. Assume that there are three fault types A, B, and C in the historical fault diagnosis sample set, and their ratio is 3:2:1. If you want to adjust the ratio to 1:1:1 in the balanced sample set, monitor the ratio of each type of fault sample in real time during the iteration process. When it reaches 1:1:1 or close to this ratio (for example, a certain error range is allowed, such as 1±0.1:1±0.1:1±0.1), the iteration is stopped.

[0077] In this embodiment, the interpolation range is determined by performing type statistics on historical fault diagnosis samples, accurately locating fault types with scarce samples, and indicating the direction for sample balancing processing; neighboring values and historical fault diagnosis samples are selected according to a preset retrieval order, ensuring the orderliness and pertinence of sample selection, so that the interpolation calculation is based on meaningful sample pairs; the fault types with few samples are supplemented by the linear interpolation method combined with the interpolation range, and the sample distribution is optimized, so that the proportion of various fault samples is more balanced, avoiding model deviation due to uneven sample distribution, and providing a better data foundation for subsequent model training.

[0078] See also Figure 5 A fifth embodiment of a switch cabinet fault detection method according to an embodiment of the present invention includes:

[0079] 501. Obtain hyperparameters in the primary convolutional neural network model, and calculate individual parameters of the sparrow based on the hyperparameters;

[0080] 502. Perform appropriate evaluation on the individual parameters of the sparrows to obtain the individual fitness set;

[0081] In this embodiment, fitness is an indicator to measure the quality of each individual sparrow. Through a specific evaluation method, the individual parameters of the sparrow are converted into a specific fitness value. This evaluation method is usually related to the performance indicators of the model, such as accuracy, loss function value, etc.

[0082] 503. Sort the individual fitness set to obtain the individual position with the best fitness and the individual position with the worst fitness;

[0083] In this embodiment, these two positions represent the best and worst performing individuals in the current population, respectively, and they play an important guiding role in subsequent algorithm iterations;

[0084] 504. Generate a fitness function according to the individual fitness set, the individual position with the best fitness, and the individual position with the worst fitness;

[0085] In this embodiment, the fitness function is used to describe the relationship between the fitness of individual sparrows and hyperparameters, which is the basis for the optimization of the sparrow search algorithm;

[0086] 505. Analyze the optimization operation data based on the fitness function to obtain the population size and the number of iterations;

[0087] In this embodiment, the number of iterations indicates the number of search rounds that the algorithm needs to perform. By analyzing the optimization operation data, the appropriate population size and number of iterations can be determined according to the characteristics of the data and the complexity of the problem, thereby improving the search efficiency and optimization effect of the algorithm.

[0088] 506. Iteratively optimize the hyperparameters based on the sparrow search algorithm, population size, individual sparrow parameters, and iteration number to obtain optimized hyperparameters;

[0089] In this embodiment, during the iteration process, individual sparrows continuously adjust their positions according to the fitness function, evolving towards a better hyperparameter combination, and ultimately finding an optimal set of hyperparameters to update the primary convolutional neural network model;

[0090] In this embodiment, model hyperparameters are converted into individual sparrow parameters, paving the way for subsequent hyperparameter optimization using the sparrow search algorithm. Individual sparrow parameters are evaluated to obtain a fitness set, and the individual performance is measured using model performance indicators, allowing the algorithm to focus on hyperparameter combinations that improve model performance. The sorted positions of the individuals with the best and worst fitness provide key guidance for algorithm iteration, making the search process more directional. The generated fitness function accurately describes the relationship between fitness and hyperparameters, becoming the core basis for algorithm optimization. Based on the fitness function, the optimized running data is analyzed and optimized to determine the population size and number of iterations, fully considering the data characteristics and problem complexity, avoiding blind searches, and improving algorithm efficiency and optimization results. During the iterative optimization process, the individual sparrows continuously evolve according to the fitness function, ultimately finding the optimal hyperparameters to update the model. This effectively improves model performance and enhances the accuracy and generalization ability of the model.

[0091] See also Figure 6 A sixth embodiment of a switch cabinet fault detection method according to an embodiment of the present invention includes:

[0092] 601. Divide the optimization operation data into populations according to the fitness function to obtain a discoverer population;

[0093] In this embodiment, the fitness function reflects the degree of excellence of individual sparrows in the search space. The discoverer population divided by it is a set of individuals in the population that are relatively more capable of exploring new areas, laying the foundation for subsequent searches for better solutions.

[0094] 602. Obtain the location and status of the discoverer population;

[0095] In this embodiment, the position represents the current coordinates in the search space, and the state itself includes various factors that may affect the search behavior, such as energy and fitness;

[0096] 603. Generate a safety value based on the location and its own state, and determine whether the safety value is greater than a preset warning value;

[0097] In this embodiment, the safety value can be understood as the safety level of the current environment of the discoverer, and the warning value is a pre-set threshold used to determine whether the search strategy needs to be changed;

[0098] 604. When the safety value is greater than the warning value, a normal distribution calculation is performed based on the position and the self-state to obtain the first population area;

[0099] In this embodiment, when the safety value is greater than the warning value, it indicates that the current environment is relatively safe. At this time, a normal distribution calculation is performed based on the location and its own state to obtain the first population area. The normal distribution calculation allows the discoverer to explore new areas more randomly within a certain range, increasing the possibility of finding a better solution.

[0100] 605. Divide the optimized operation data into populations according to the fitness function to obtain a follower population;

[0101] In this embodiment, the follower population will act according to the finder's search results to improve the search efficiency of the entire population;

[0102] 606. Obtain the hunger value of the follower population and generate a second population area based on the hunger value, location, and own status;

[0103] In this embodiment, the hunger value reflects the follower's desire for a better solution. The higher the hunger value, the more likely the follower is to actively approach the discoverer or move to a better area.

[0104] 607. Calculate the population size and the number of iterations based on the first population area and the second population area.

[0105] In this embodiment, the population size determines the number of individuals participating in the search, and the number of iterations determines the number of search rounds. By integrating information from the two population regions to determine these two key parameters, the algorithm can better adapt to the complexity of the problem and the characteristics of the data.

[0106] In this embodiment, the discoverer and follower populations are divided by the fitness function, laying the foundation for efficient search. For the discoverer population, the search strategy can be flexibly adjusted by calculating the safety value and comparing it with the warning value. When the safety value is greater than the warning value, the first population area is calculated using the normal distribution, allowing the discoverer to randomly explore in a safe environment, increasing the probability of finding a better solution; considering the hunger value of the follower population, which reflects the degree of desire for a better solution, the second population area is generated according to the hunger value, position and its own status, so that the follower can move to a better area in a more targeted manner; combining the information of the two population areas to determine the population size and the number of iterations, provides strong support for optimizing hyperparameters and improving model performance.

[0107] See also Figure 7 A seventh embodiment of a switch cabinet fault detection method according to an embodiment of the present invention includes:

[0108] 701. Perform feature analysis on the real-time switchgear operation data based on a preset time series algorithm to obtain time domain features;

[0109] In this embodiment, the time domain features can intuitively reflect the changes in data over time, such as the specific values and change trends of parameters such as current and voltage at different times, providing a basis for subsequent analysis of the operating status of the switchgear at different time points;

[0110] 702. Perform spectrum analysis on the time domain features to obtain frequency domain features;

[0111] In this embodiment, frequency domain features help reveal the periodic information and frequency components hidden in the data. Different fault types may correspond to specific frequency features, and frequency domain analysis can more sensitively capture these fault signals.

[0112] 703. Perform variance analysis on the frequency domain features to obtain signal fluctuation characteristics;

[0113] In this embodiment, the signal fluctuation characteristics reflect the stability of the switchgear operating data in the frequency domain. Large fluctuations may indicate an abnormality in the equipment.

[0114] 704. Perform thermal balance analysis on the real-time switchgear operation data to obtain thermal balance analysis data;

[0115] In this embodiment, the balance condition is closely related to the operating state of the switchgear. Abnormal temperature changes may be an early sign of equipment failure. The thermal balance analysis data can help determine whether the switchgear is in a normal thermal state.

[0116] 705. Obtaining partial discharge pulse current, integration time interval, and detection capacitance from real-time switchgear operation data;

[0117] 706. The partial discharge amount is calculated based on the preset integral type formula, the partial discharge pulse current, the integral time interval, and the detection capacitance. The expression of the partial discharge amount is as follows:

[0118] Where i(t) is the partial discharge pulse current, [t1, t2] is the integration time interval, t1 is the start time, t2 is the end time, and C is the detection capacitance;

[0119] In this embodiment, partial discharge is an important manifestation of the degradation of the insulation performance of the switchgear. Accurate calculation of the partial discharge amount is crucial for evaluating the insulation status of the switchgear.

[0120] 707. Predict the time domain characteristics, frequency domain characteristics, signal fluctuation characteristics, thermal balance analysis data, and partial discharge amount according to the fault diagnosis model to obtain a prediction result;

[0121] In this embodiment, the fault diagnosis model is trained to comprehensively analyze these multi-dimensional feature information and accurately determine whether the current operating status of the switch cabinet is normal and the possible fault type;

[0122] In this embodiment, in terms of feature extraction, the time domain features intuitively present the changes in data over time, laying the foundation for analyzing the operating status at different time points; the frequency domain features can reveal hidden periodicity and frequency components, helping to capture specific fault signals; the signal fluctuation characteristics reflect the stability of the frequency domain, and large fluctuations may indicate equipment abnormalities; the thermal balance analysis data can assist in judging the thermal state of the switchgear and discover fault signs in advance. The calculation of partial discharge is also very critical. It is an important indicator for evaluating the insulation performance of the switchgear. Accurate calculation helps to grasp the insulation status in a timely manner; finally, the fault diagnosis model integrates multi-dimensional feature information for prediction, and can accurately judge the operating status of the switchgear and possible fault types; this multi-dimensional analysis and precise prediction greatly improves the accuracy and reliability of fault diagnosis, can discover potential fault hazards in advance, and provide a scientific basis for operation and maintenance personnel to formulate maintenance plans, effectively reducing the risk of equipment damage and power outages, and ensuring the stable operation of the power system.

[0123] The above describes a switch cabinet fault detection method according to an embodiment of the present invention. The following describes a switch cabinet fault detection device according to an embodiment of the present invention. Figure 8 In one embodiment of the present invention, a switch cabinet fault detection device includes:

[0124] The first data acquisition module 1 is used to acquire historical switchgear operation data;

[0125] Model building module 2, used to build a primary convolutional neural network model based on a preset convolutional neural network algorithm and historical switchgear operation data;

[0126] Parameter calculation module 3, used to calculate the hyperparameters of the primary convolutional neural network model based on a preset sparrow search algorithm to obtain optimized hyperparameters;

[0127] The model updating module 4 is used to update the primary neural network model according to the optimized hyperparameters to obtain a fault diagnosis model;

[0128] The second data acquisition module 5 is used to obtain real-time switchgear operation data;

[0129] The result prediction module 6 is used to predict the real-time switchgear operation data according to the fault diagnosis model to obtain the prediction result.

[0130] In this embodiment, during data processing and model construction, historical operating data is obtained. With the powerful feature extraction capability of the convolutional neural network, its convolution layer and pooling layer work together to automatically extract local features and reduce dimensionality, retain key information, provide solid support for fault diagnosis, and improve the depth and accuracy of data feature mining; in model optimization, the sparrow search algorithm is used to efficiently search in the hyperparameter space, and the hyperparameters are continuously updated iteratively to make the structure and parameters of the fault diagnosis model more reasonable, better fit the relationship between data and faults, and improve model performance and accuracy; real-time operating data is obtained, and the updated model is used for prediction. After convolution, pooling, and fully connected layer operations, the results are output quickly and accurately. Operation and maintenance personnel can promptly discover potential faults and take measures to reduce the probability of faults, reduce power outage losses, and ensure stable, reliable, safe, and efficient operation of the power system.

[0131] Figure 9 is a schematic structural diagram of a switch cabinet fault detection device provided by an embodiment of the present invention. The switch cabinet fault detection device 900 may have relatively large differences due to different configurations or performances, and may include one or more processors (central processing units, CPU) 913 (for example, one or more processors) and a memory 920, and one or more storage media 930 (for example, one or more mass storage devices) storing application programs 933 or data 932. The memory 920 and the storage medium 930 may be temporary storage or permanent storage. The program stored in the storage medium 930 may include one or more modules (not shown in the figure), each of which may include a series of instruction operations in the switch cabinet fault detection device 900. Furthermore, the processor 913 may be configured to communicate with the storage medium 930, and execute a series of instruction operations in the storage medium 930 on the switch cabinet fault detection device 900 to implement the steps of the switch cabinet fault detection method provided by the above-mentioned method embodiments.

[0132] A switch cabinet fault detection device 900 may also include one or more power supplies 940, one or more wired or wireless network interfaces 950, one or more input and output interfaces 960, and / or one or more operating systems 931, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. It will be understood by those skilled in the art that Figure 9 The structure of a switch cabinet fault detection device shown does not constitute a limitation on the switch cabinet fault detection device 900 , and may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently.

[0133] A computer-readable storage medium stores instructions, which, when executed by a processor, implement the steps of the switch cabinet fault detection method as described above.

[0134] The present invention and its embodiments are described above. This description is not restrictive. What is shown in the accompanying drawings is only one of the embodiments of the present invention, and the actual content is not limited to this. In short, if ordinary technicians in this field are inspired by it and do not depart from the purpose of the invention, they can creatively design structural methods and embodiments similar to the technical solution, which should all fall within the scope of protection of the present invention.

Claims

1. A switch cabinet fault detection method, characterized in that: include: Obtain historical switchgear operation data; A primary convolutional neural network model is constructed based on the preset convolutional neural network algorithm and historical switchgear operation data; Calculate the hyperparameters of the primary convolutional neural network model based on the preset sparrow search algorithm to obtain the optimized hyperparameters; The primary neural network model is updated according to the optimized hyperparameters to obtain a fault diagnosis model; Obtain real-time switchgear operation data; The real-time switchgear operation data is predicted according to the fault diagnosis model to obtain the prediction results.

2. A switch cabinet fault detection method according to claim 1, characterized in that: The primary convolutional neural network model is constructed based on the preset convolutional neural network algorithm and historical switchgear operation data, including: The historical switchgear operation data is calculated based on the interquartile range method to obtain the first quartile and the second quartile; Identify and process abnormalities in historical switchgear operating data based on preset abnormal conditions, first quartile, and second quartile to obtain optimized operating data; A primary convolutional neural network model is constructed based on the convolutional neural network algorithm and optimized running data.

3. A switch cabinet fault detection method according to claim 2, characterized in that: The updating of the primary neural network model according to the optimized hyperparameters to obtain a fault diagnosis model includes: Obtain historical fault diagnosis sample sets; The preset K-nearest neighbor algorithm is used to perform proximity analysis on the historical fault diagnosis sample set to obtain a proximity value data set; Perform interpolation calculation on the historical fault diagnosis sample set according to the linear interpolation method and the neighboring value data set to obtain a balanced sample set; Mapping the equalized sample set to a preset orthogonal coordinate axis based on a linear transformation method to obtain a mapped sample set; Generate a covariance matrix according to the mapped sample set, and perform eigenanalysis on the covariance matrix to obtain eigenvalues and eigenvector sets; Sorting the eigenvector set according to the eigenvalue to obtain a sorted eigenvector set; generating a projection matrix based on the sorted feature vector set and the optimization run data; The primary neural network model is updated based on the projection matrix and optimized hyperparameters to obtain a fault diagnosis model.

4. A switch cabinet fault detection method according to claim 3, characterized in that: The interpolation calculation of the historical fault diagnosis sample set according to the linear interpolation method and the neighboring value data set to obtain the balanced sample set includes: Perform type statistics on historical fault diagnosis samples to obtain the interpolation range; Selecting a neighboring value from the neighboring value data set according to a preset retrieval order, and selecting a historical fault diagnosis sample from the historical fault diagnosis sample set according to the neighboring value; Perform interpolation calculation on historical fault diagnosis samples according to linear interpolation method, interpolation range and neighboring values to obtain balanced samples; Return and select a neighboring value from the neighboring value data set according to the retrieval order until the preset iteration stop condition is met, and then obtain all balanced samples to obtain a balanced sample set.

5. A switch cabinet fault detection method according to claim 2, characterized in that: The hyperparameters of the primary convolutional neural network model are calculated based on the preset sparrow search algorithm to obtain optimized hyperparameters, including: Obtain the hyperparameters in the primary convolutional neural network model and calculate the individual parameters of the sparrow based on the hyperparameters; The individual parameters of sparrows are moderately evaluated to obtain the individual fitness set; Sort the individual fitness set to obtain the individual position with the best fitness and the individual position with the worst fitness; Generate a fitness function based on the individual fitness set, the best fitness individual position and the worst fitness individual position; Analyze the optimization operation data based on the fitness function to obtain the population size and number of iterations; The hyperparameters are iteratively optimized based on the sparrow search algorithm, population size, sparrow individual parameters and number of iterations to obtain the optimized hyperparameters.

6. A switch cabinet fault detection method according to claim 5, characterized in that: The optimization operation data is analyzed based on the fitness function to obtain the population size and the number of iterations, including: Divide the optimization operation data into populations according to the fitness function to obtain the discoverer population; Get the location and status of the discoverer population; Generate a safety value based on the location and its own status, and determine whether the safety value is greater than the preset warning value; When the safety value is greater than the warning value, a normal distribution calculation is performed based on the location and its own state to obtain the first population area; Divide the optimized operation data into populations according to the fitness function to obtain the follower population; Get the hunger value of the follower population and generate a second population area based on the hunger value, position and own status; The population size and number of iterations are calculated based on the first population area and the second population area.

7. A switch cabinet fault detection method according to claim 1, characterized in that: The predicting of real-time switchgear operation data according to the fault diagnosis model to obtain a prediction result includes: Perform feature analysis on real-time switchgear operation data based on a preset time series algorithm to obtain time domain features; Perform spectrum analysis on the time domain features to obtain frequency domain features; Perform variance analysis on frequency domain features to obtain signal fluctuation characteristics; Conduct thermal balance analysis on real-time switchgear operation data to obtain thermal balance analysis data; Obtain partial discharge pulse current, integration time interval and detection capacitance from real-time switchgear operation data; The partial discharge amount is calculated based on the preset integral type formula, partial discharge pulse current, integral time interval, and detection capacitance. The expression of partial discharge amount is as follows: Where i(t) is the partial discharge pulse current, [t1, t2] is the integration time interval, t1 is the start time, t2 is the end time, and C is the detection capacitance; According to the fault diagnosis model, the time domain characteristics, frequency domain characteristics, signal fluctuation characteristics, thermal balance analysis data, and partial discharge amount are predicted to obtain the prediction results.

8. A switch cabinet fault detection device, characterized in that: include: A first data acquisition module is used to acquire historical switchgear operation data; A model building module is used to build a primary convolutional neural network model based on a preset convolutional neural network algorithm and historical switchgear operation data; A parameter calculation module is used to calculate the hyperparameters of the primary convolutional neural network model based on a preset sparrow search algorithm to obtain optimized hyperparameters; A model updating module is used to update the primary neural network model according to the optimized hyperparameters to obtain a fault diagnosis model; The second data acquisition module is used to obtain real-time switchgear operation data; The result prediction module is used to predict the real-time switchgear operation data according to the fault diagnosis model to obtain the prediction result.

9. A switch cabinet fault detection device, characterized in that: include: a memory and at least one processor, wherein instructions are stored in the memory; At least one of the processors calls the instructions in the memory to enable the switch cabinet fault detection device to execute each step of the switch cabinet fault detection method according to any one of claims 1 to 7.

10. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the steps of the switch cabinet fault detection method according to any one of claims 1 to 7 are implemented.

Images