Power equipment anomaly detection method and system based on deep learning

Abstract

The invention provides a power equipment anomaly detection method and system based on deep learning, and the method comprises the steps: obtaining a time sequence current and voltage signal, an infrared thermal imaging image and an equipment vibration frequency spectrum of power equipment every first preset time, performing standard processing on the time sequence current and voltage signals, the infrared thermal imaging image and the equipment vibration frequency spectrum; respectively extracting a first feature set, a second feature set and a third feature set from the time sequence current and voltage signal, the infrared thermal imaging image and the equipment vibration spectrum which are subjected to standard processing, and fusing the first feature set, the second feature set and the third feature set to obtain fused features; inputting the fusion features into a pre-trained anomaly detection model to obtain the anomaly probability of each anomaly type; and traversing all the abnormal probabilities according to the abnormal types, and outputting a final detection result according to a traversing result. The method can achieve the precise recognition of a composite fault and the self-adaptive diagnosis of a working condition.

Description

Technical Field

[0001] This invention relates to the field of power equipment anomaly detection technology, and in particular to a power equipment anomaly detection method and system based on deep learning. Background Technology

[0002] Current power equipment anomaly detection technologies mainly rely on single-type data (such as electrical signals or vibration spectra), making it difficult to comprehensively reflect the multi-physics coupling characteristics of equipment faults. For example, localized overheating is often accompanied by current fluctuations and mechanical vibration changes, but traditional methods lack multi-source data collaborative analysis mechanisms, easily leading to missed diagnoses of complex faults. Furthermore, the preprocessing workflow is not optimized for modal characteristics: time-series signals are susceptible to noise interference, infrared image contrast and edge fidelity are insufficient, and an excessively wide dynamic range of the vibration spectrum can mask weak features, directly affecting the effectiveness of subsequent feature extraction. Summary of the Invention

[0003] The purpose of this invention is to provide a method and system for detecting power equipment anomalies based on deep learning, which aims to solve the problem that the accuracy of power equipment anomaly detection is low due to the reliance on a single type of data in traditional technologies.

[0004] In a first aspect, the present invention provides a method for detecting anomalies in power equipment based on deep learning, the method comprising:

[0005] The system acquires the time-series current and voltage signals, infrared thermal imaging images, and equipment vibration spectrum of the power equipment at first preset time intervals, and performs standard processing on the time-series current and voltage signals, infrared thermal imaging images, and equipment vibration spectrum.

[0006] The first feature set, the second feature set, and the third feature set are extracted from the time-series current and voltage signals, infrared thermal imaging images, and equipment vibration spectra after standard processing, respectively. The first feature set, the second feature set, and the third feature set are then fused to obtain the fused features.

[0007] The fused features are input into a pre-trained anomaly detection model to obtain the anomaly probability for each anomaly type;

[0008] Iterate through all the anomaly probabilities according to the anomaly type, and output the final detection result based on the iteration results.

[0009] In some embodiments, the step of standard processing of the time-series current-voltage signal, infrared thermal imaging image, and device vibration spectrum includes:

[0010] Segmented overlapping sampling is used for time-series current and voltage signals, and Z-Score normalization is applied to eliminate dimensional differences;

[0011] Histogram equalization is performed on infrared thermal imaging images to enhance contrast, while bilateral filtering is used to preserve edge thermal radiation features;

[0012] The vibration spectrum of the equipment is generated into a time-frequency spectrum using short-time Fourier transform, and the dynamic range is compressed to the [-5, 5] dB interval using logarithmic transform.

[0013] In some embodiments, the first feature set is extracted according to the following formula:

[0014] The first feature set includes mean, standard deviation, skewness, kurtosis, RMS value, and peak factor.

[0015] Calculate the skewness using the following formula:

[0016]

[0017] Calculate kurtosis using the following formula:

[0018]

[0019] Calculate the effective value using the following formula:

[0020]

[0021] The peak factor is calculated using the following formula:

[0022]

[0023] Among them, S i Let X(t, i) be the skewness of the i-th electrical signal, and let X(t, i) be the sampled value of the i-th electrical signal at time t. i Let σ be the mean of the i-th electrical signal. i K is the standard deviation of the i-th electrical signal. i R is the kurtosis of the i-th electrical signal. i Let C be the effective value of the i-th electrical signal, T be the total number of electrical signals, and C be the effective value of the i-th electrical signal. i Let be the peak factor of the i-th electrical signal.

[0024] In some embodiments, the second feature set includes the highest temperature, the lowest temperature, the maximum temperature difference, the proportion of high-temperature areas, and the maximum temperature gradient;

[0025] The percentage of high-temperature areas is calculated using the following formula:

[0026]

[0027] Ω={(x,y)X(x,y)>T max –5%ΔT};

[0028] The maximum temperature gradient is calculated using the following formula:

[0029]

[0030] Where H represents the proportion of high-temperature regions, Ω represents the set of high-temperature regions, h and w are the length and width of the infrared thermal image, respectively, X(x, y) is the temperature value at coordinate (x, y) in the infrared thermal image, and T... max The highest temperature is ΔT, the maximum temperature difference is G. max This represents the maximum temperature gradient.

[0031] In some embodiments, the third feature set includes total energy, first frequency energy, second frequency energy, third frequency energy, and frequency center;

[0032] The frequencies in the vibration spectrum of the equipment are sorted, and the total energy of the highest third is taken as the first frequency energy, the total energy of the next third is taken as the second frequency energy, and the total energy of the lowest third is taken as the third frequency energy.

[0033] In some embodiments, the step of fusing the first feature set, the second feature set, and the third feature set to obtain fused features includes:

[0034] The features in the first feature set, the second feature set, and the third feature set are concatenated to obtain a fused feature matrix, which is a one-dimensional feature matrix.

[0035] In some embodiments, the step of traversing all anomaly probabilities according to the anomaly type and outputting the final detection result based on the traversal result includes:

[0036] Based on the anomaly type, obtain the probability threshold for each anomaly type from a preset association table;

[0037] And determine whether the anomaly probability is greater than the corresponding probability threshold;

[0038] If the abnormal probability is greater than the corresponding probability threshold, an abnormality is determined and the abnormality type is output.

[0039] If the abnormal probability is less than or equal to the corresponding probability threshold, it is judged as normal;

[0040] All traversal results are summarized to obtain the final detection result.

[0041] Secondly, the present invention provides a deep learning-based power equipment anomaly detection system, the system comprising:

[0042] The data acquisition module is used to acquire the time-series current and voltage signals, infrared thermal imaging images, and equipment vibration spectrum of the power equipment at first preset time intervals, and to perform standard processing on the time-series current and voltage signals, infrared thermal imaging images, and equipment vibration spectrum.

[0043] The feature extraction module is used to extract a first feature set, a second feature set, and a third feature set from the time-series current and voltage signals, infrared thermal imaging images, and equipment vibration spectra that have undergone standard processing, and to fuse the first feature set, the second feature set, and the third feature set to obtain fused features;

[0044] The probability prediction module is used to input the fused features into a pre-trained anomaly detection model to obtain the anomaly probability for each anomaly type.

[0045] The abnormal result output module is used to traverse all abnormal probabilities according to the abnormal type and output the final detection result based on the traversal result.

[0046] Thirdly, the present invention provides a storage medium that stores one or more programs, which, when executed by a processor, implement the above-described deep learning-based power equipment anomaly detection method.

[0047] Fourthly, the present invention provides an electronic device, the electronic device comprising a memory and a processor, wherein:

[0048] The memory is used to store computer programs;

[0049] When the processor executes the computer program stored in the memory, it implements the above-described deep learning-based power equipment anomaly detection method.

[0050] Compared with the prior art, the present invention has the following advantages:

[0051] This invention integrates multimodal data from time-series current and voltage signals, infrared thermal imaging images, and equipment vibration spectra. It employs segmented overlapping sampling and Z-score standardization to eliminate time-series signal noise and dimensional differences. Histogram equalization and bilateral filtering enhance infrared image contrast while preserving thermal radiation edge features. Short-time Fourier transform and logarithmic transform standardize the dynamic range of the vibration spectrum, ensuring high fidelity in data preprocessing. Furthermore, it extracts a composite feature set encompassing statistical moments (mean, skewness, kurtosis), energy indicators (RMS, frequency band energy), and spatial distribution characteristics (proportion of high-temperature areas, temperature gradient). This set is then spliced ​​and fused to form a multidimensional correlation feature matrix, comprehensively characterizing the multi-physical field coupling state of the equipment. Finally, by combining a dynamic threshold adjustment mechanism with a pre-trained model to output anomaly probabilities, it achieves accurate identification of composite faults and adaptive diagnosis of operating conditions, significantly improving detection reliability and reducing false negatives and false negatives, providing a highly robust solution for intelligent operation and maintenance of power equipment. Attached Figure Description

[0052] Figure 1 This is a flowchart of a deep learning-based power equipment anomaly detection method proposed in an embodiment of the present invention;

[0053] Figure 2 This is a schematic diagram of the structure of a deep learning-based power equipment anomaly detection system proposed in an embodiment of the present invention.

[0054] The following detailed description, in conjunction with the accompanying drawings, will further illustrate the present invention. Detailed Implementation

[0055] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Unless otherwise defined, the technical or scientific terms used herein should have the ordinary meaning understood by those skilled in the art. The terms "comprising" and similar expressions used herein mean that the element or object preceding the word covers the element or object listed after the word and its equivalents, but does not exclude other elements or objects.

[0056] like Figure 1 As shown, an embodiment of the present invention proposes a method for detecting anomalies in power equipment based on deep learning. The method includes steps S101 to S104, wherein:

[0057] Step S101: Acquire the time-series current and voltage signals, infrared thermal imaging images, and equipment vibration spectrum of the power equipment at first preset time intervals, and perform standard processing on the time-series current and voltage signals, infrared thermal imaging images, and equipment vibration spectrum.

[0058] It should be noted that during the monitoring of power equipment operation, in order to comprehensively and accurately grasp its operating status, multi-dimensional data collection is required at pre-set time intervals. Specifically, high-precision current and voltage sensors are used to acquire time-series current and voltage signals of the power equipment over a period of time at pre-set time intervals. These signals can intuitively reflect the changes in power parameters during equipment operation, such as current fluctuations and voltage stability, providing basic data for subsequent analysis of whether the equipment has any abnormalities. Simultaneously, infrared thermal imagers are used to photograph the power equipment at the same time intervals to acquire infrared thermal images. Infrared thermal imaging technology can detect infrared radiation emitted from the surface of the equipment and convert it into a visualized thermal image, thus clearly showing the temperature distribution of different parts of the equipment and helping to detect abnormal heating phenomena such as localized overheating. Furthermore, vibration sensors are used to collect the vibration spectrum of the power equipment at pre-set time intervals. The vibration spectrum reflects the vibration characteristics of the equipment during operation. Different types of equipment faults often lead to specific change patterns in the vibration spectrum; for example, bearing damage may cause an increase in vibration at specific frequencies. After obtaining this raw data, in order to facilitate subsequent feature extraction and model analysis, it is necessary to perform standard processing on the time-series current and voltage signals, infrared thermal imaging images, and equipment vibration spectra.

[0059] In addition, in some embodiments, the standard processing is as follows: segmented overlapping sampling is used for the time-series current and voltage signals, and Z-Score normalization is applied to eliminate dimensional differences; histogram equalization is performed on the infrared thermal imaging image to enhance contrast, while bilateral filtering is used to preserve edge thermal radiation characteristics; the vibration spectrum of the equipment is generated into a time-spectrum map by short-time Fourier transform, and logarithmic transform is used to compress the dynamic range to the [-5,5]dB range.

[0060] In summary, segmented overlapping sampling of time-series current and voltage signals increases the number of samples, allowing the model to learn richer signal features, avoiding information omissions due to insufficient sampling, and improving the accuracy of anomaly detection. Applying Z-Score normalization to eliminate dimensional differences ensures that data of different dimensions are on the same order of magnitude, facilitating subsequent feature extraction and model processing, and enhancing the model's stability and generalization ability. Histogram equalization of infrared thermal imaging images enhances contrast, making the differences between different temperature regions in the image more obvious, highlighting the temperature anomalies of the equipment, and facilitating the discovery of potential problems. Bilateral filtering preserves edge thermal radiation features, removing noise without destroying the temperature information at the edges of the equipment in the image, resulting in higher image quality and providing a reliable basis for accurate analysis of the equipment's thermal state. Generating a time-frequency spectrum from the equipment vibration spectrum using short-time Fourier transform simultaneously acquires both time and frequency domain information of the signal, more comprehensively reflecting the equipment's vibration characteristics. Using logarithmic transform to compress the dynamic range to the [-5, 5] dB interval makes the data distribution more concentrated, reducing the impact of extreme values ​​on the model and improving the model's learning effect on vibration features.

[0061] Step S102: Extract the first feature set, the second feature set, and the third feature set from the time-series current and voltage signal, the infrared thermal imaging image, and the device vibration spectrum after standard processing, respectively, and fuse the first feature set, the second feature set, and the third feature set to obtain the fused feature;

[0062] It should be noted that after standardizing the various types of raw data, the next step is to extract representative feature information from these processed data. For the standardized time-series current and voltage signals, specialized algorithms in the field of signal processing, such as Fourier transform and wavelet transform, are used to extract a first feature set that reflects the signal characteristics. This first feature set can describe the characteristics of the current and voltage signals from different perspectives, providing a basis for judging whether the equipment's power parameters are normal. For the standardized infrared thermal imaging images, image processing and computer vision techniques are used to extract a second feature set. The second feature set can effectively reflect the temperature distribution and thermal state characteristics of the equipment surface. For the standardized equipment vibration spectrum, spectral analysis methods are used to extract a third feature set. The third feature set is closely related to the mechanical vibration state of the equipment and can provide important clues for detecting whether the equipment has mechanical faults. After extracting the first, second, and third feature sets respectively, in order to fully utilize the information contained in these different types of data and improve the accuracy and comprehensiveness of anomaly detection, these three feature sets need to be fused. The fusion method can use a simple feature splicing method, combining the three feature sets in a certain order into a new feature vector, i.e., the fused feature.

[0063] Specifically, the first feature set includes mean, standard deviation, skewness, kurtosis, RMS value, and peak factor. The mean and standard deviation can be obtained by conventional methods and will not be described in detail in this embodiment.

[0064] Calculate the skewness using the following formula:

[0065]

[0066] Calculate kurtosis using the following formula:

[0067]

[0068] Calculate the effective value using the following formula:

[0069]

[0070] The peak factor is calculated using the following formula:

[0071]

[0072] Among them, S i Let X(t,i) be the skewness of the i-th electrical signal, and let X(t,i) be the sampled value of the i-th electrical signal at time t. i Let σ be the mean of the i-th electrical signal. i K is the standard deviation of the i-th electrical signal. i R is the kurtosis of the i-th electrical signal. i Let C be the effective value of the i-th electrical signal, T be the total number of electrical signals, and C be the effective value of the i-th electrical signal. i Let be the peak factor of the i-th electrical signal.

[0073] In summary, the mean reflects the average level of electrical signals, helping to understand the overall power parameters of the equipment; the standard deviation reflects the dispersion of the signal, which can determine the fluctuation of power parameters; skewness describes the asymmetry of the signal distribution, helping to detect abnormal asymmetric power phenomena; kurtosis reflects the sharpness of the signal distribution, which can identify extremely abnormal power signals; the RMS value measures the energy of the signal and is related to the operating status of the equipment, such as power; the kurtosis factor highlights the peak characteristics in the signal, enabling timely detection of sudden changes in power parameters. These characteristics characterize the properties of time-series current and voltage signals from different perspectives, providing rich information for accurately detecting anomalies in power equipment.

[0074] In addition, in some embodiments, the second feature set includes the highest temperature, the lowest temperature, the maximum temperature difference, the proportion of high temperature area, and the maximum temperature gradient. The highest temperature and the lowest temperature can be obtained directly, and the maximum temperature difference is the difference between the highest temperature and the lowest temperature.

[0075] The percentage of high-temperature areas is calculated using the following formula:

[0076]

[0077] Ω={(x,y)|X(x,y)>T max –5%ΔT};

[0078] The maximum temperature gradient is calculated using the following formula:

[0079]

[0080] Where H represents the proportion of high-temperature regions, Ω represents the set of high-temperature regions, h and w are the length and width of the infrared thermal image, respectively, X(x,y) is the temperature value at coordinates (x,y) in the infrared thermal image, and T... max The highest temperature is ΔT, the maximum temperature difference is G. max This represents the maximum temperature gradient.

[0081] In summary, the highest temperature directly reflects the maximum heat generation on the equipment surface, allowing for timely detection of localized overheating issues; the lowest temperature helps determine the overall temperature range of the equipment; the maximum temperature difference reflects the uniformity of temperature distribution on the equipment surface, and an excessively large temperature difference may indicate a malfunction; the proportion of high-temperature areas quantifies the extent of high-temperature regions on the equipment surface, helping to assess the severity of abnormal heating; and the maximum temperature gradient reflects the drastic changes in surface temperature, with abnormal temperature gradients potentially suggesting mechanical or electrical faults. These characteristics enable comprehensive and accurate analysis of infrared thermal imaging images, effectively detecting thermal anomalies in power equipment.

[0082] Furthermore, in some embodiments, the third feature set includes total energy, first frequency energy, second frequency energy, third frequency energy, and frequency center. The formula for calculating the total energy is a conventional technique and will not be described in detail in this embodiment.

[0083] The frequencies in the vibration spectrum of the equipment are sorted, and the total energy of the highest third is taken as the first frequency energy, the total energy of the next third is taken as the second frequency energy, and the total energy of the lowest third is taken as the third frequency energy.

[0084] In summary, total energy reflects the overall energy level of the equipment's vibration spectrum and is related to the vibration intensity. Dividing the frequencies into one-third intervals and calculating the corresponding first, second, and third frequency energies reflects the distribution of vibration energy in different frequency bands. Energy anomalies in different frequency bands may correspond to different types of equipment faults. The frequency center can determine the location of the concentrated frequency of the equipment's vibration energy, helping to identify the equipment's main vibration frequency and providing crucial information for analyzing the equipment's vibration characteristics. These features describe the characteristics of the equipment's vibration spectrum from different perspectives, providing an important basis for detecting mechanical faults in equipment.

[0085] Finally, the features from the first, second, and third feature sets are concatenated to obtain a fused feature matrix, which is a one-dimensional feature matrix. For example, the mean, standard deviation, skewness, kurtosis, RMS value, peak factor, highest temperature, lowest temperature, maximum temperature difference, proportion of high-temperature region, maximum temperature gradient, total energy, first frequency energy, second frequency energy, third frequency energy, and frequency center are concatenated sequentially to obtain the fused feature matrix.

[0086] Step S103: Input the fused features into the pre-trained anomaly detection model to obtain the anomaly probability of each anomaly type;

[0087] It's important to note that in this step, after obtaining the fused features, they are input into a pre-trained anomaly detection model to accurately determine abnormal conditions in power equipment. This pre-trained anomaly detection model is typically trained on a large amount of historical data and known anomaly cases. It learns the pattern distribution of fused features under normal equipment operation and the feature patterns corresponding to various anomaly types. Common anomaly detection models can include machine learning-based models such as Support Vector Machines (SVM) and Random Forests, or deep learning-based models such as Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), and their variants (such as LSTM and GRU). When the fused features are input into the anomaly detection model, the model analyzes and calculates the input features based on its internally learned knowledge and algorithms, outputting the anomaly probability corresponding to each anomaly type. For example, assuming that the possible anomaly types of power equipment include electrical faults, mechanical faults, and overheating faults, the model will calculate the probability values ​​of the input fused features belonging to these types of anomalies. These probability values ​​reflect the degree of matching between the current state of the equipment and various anomaly types; the higher the probability value, the greater the likelihood that the equipment has experienced that type of anomaly. This approach provides a quantitative basis for subsequent anomaly assessment and decision-making.

[0088] Step S104: Traverse all the anomaly probabilities according to the anomaly type, and output the final detection result based on the traversal result.

[0089] Specifically, the probability threshold for each abnormality type is obtained from a preset association table based on the abnormality type; and it is determined whether the abnormality probability is greater than the corresponding probability threshold; if the abnormality probability is greater than the corresponding probability threshold, an abnormality is determined and the abnormality type is output; if the abnormality probability is less than or equal to the corresponding probability threshold, a normality is determined; all traversal results are summarized to obtain the final detection result.

[0090] In summary, by retrieving the probability threshold for each anomaly type from a preset association table, a clear criterion is provided for determining whether a device is malfunctioning. Determining whether the anomaly probability is greater than the corresponding probability threshold accurately distinguishes between normal and abnormal states. If the anomaly probability is greater than the threshold, an anomaly is identified and the anomaly type is output, allowing for timely identification of the specific problem. If the probability is less than or equal to the threshold, the device is considered normal, avoiding misjudgments. Summarizing all traversal results yields the final detection result, providing a comprehensive and systematic view of the overall equipment operation status. This provides maintenance personnel with clear and accurate decision-making support, ensuring the safe and stable operation of power equipment.

[0091] This invention integrates multimodal data from time-series current and voltage signals, infrared thermal imaging images, and equipment vibration spectra. It employs segmented overlapping sampling and Z-score standardization to eliminate time-series signal noise and dimensional differences. Histogram equalization and bilateral filtering enhance infrared image contrast while preserving thermal radiation edge features. Short-time Fourier transform and logarithmic transform standardize the dynamic range of the vibration spectrum, ensuring high fidelity in data preprocessing. Furthermore, it extracts a composite feature set encompassing statistical moments (mean, skewness, kurtosis), energy indicators (RMS, frequency band energy), and spatial distribution characteristics (proportion of high-temperature areas, temperature gradient). This set is then spliced ​​and fused to form a multidimensional correlation feature matrix, comprehensively characterizing the multi-physical field coupling state of the equipment. Finally, by combining a dynamic threshold adjustment mechanism with a pre-trained model to output anomaly probabilities, it achieves accurate identification of composite faults and adaptive diagnosis of operating conditions, significantly improving detection reliability and reducing false negatives and false negatives, providing a highly robust solution for intelligent operation and maintenance of power equipment.

[0092] like Figure 2 As shown, one embodiment of the present invention proposes a power equipment anomaly detection system based on deep learning, the system comprising:

[0093] The data acquisition module 10 is used to acquire the time-series current and voltage signals, infrared thermal imaging images and equipment vibration spectrum of the power equipment at first preset time intervals, and to perform standard processing on the time-series current and voltage signals, infrared thermal imaging images and equipment vibration spectrum.

[0094] The feature extraction module 20 is used to extract a first feature set, a second feature set, and a third feature set from the time-series current and voltage signal, the infrared thermal imaging image, and the equipment vibration spectrum after standard processing, respectively, and to fuse the first feature set, the second feature set, and the third feature set to obtain fused features;

[0095] The probability prediction module 30 is used to input the fused features into a pre-trained anomaly detection model to obtain the anomaly probability of each anomaly type.

[0096] The abnormal result output module 40 is used to traverse all abnormal probabilities according to the abnormal type and output the final detection result according to the traversal result.

[0097] In another aspect, the present invention also proposes a storage medium on which one or more programs are stored, which, when executed by a processor, implement the above-described deep learning-based power equipment anomaly detection method.

[0098] In another aspect, the present invention also proposes an electronic device, including a memory and a processor, wherein the memory is used to store a computer program, and the processor is used to execute the computer program stored in the memory to implement the above-mentioned deep learning-based power equipment anomaly detection method.

[0099] Those skilled in the art will understand that the logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can mean any means that can contain stored, communicated, propagated, or transmitted programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0100] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0101] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0102] While embodiments of the present invention have been described in detail above, it will be apparent to those skilled in the art that various modifications and variations can be made to these embodiments. However, it should be understood that such modifications and variations fall within the scope and spirit of the invention as set forth in the claims. Furthermore, the invention described herein may have other embodiments and can be implemented or carried out in various ways.

Claims

1. A method for detecting anomalies in power equipment based on deep learning, characterized in that, The method includes: The system acquires the time-series current and voltage signals, infrared thermal imaging images, and equipment vibration spectrum of the power equipment at first preset time intervals, and performs standard processing on the time-series current and voltage signals, infrared thermal imaging images, and equipment vibration spectrum. The first feature set, the second feature set, and the third feature set are extracted from the time-series current and voltage signals, infrared thermal imaging images, and equipment vibration spectra after standard processing, respectively. The first feature set, the second feature set, and the third feature set are then fused to obtain the fused features. The fused features are input into a pre-trained anomaly detection model to obtain the anomaly probability for each anomaly type; Iterate through all the anomaly probabilities according to the anomaly type, and output the final detection result based on the iteration results.

2. The deep learning-based power equipment anomaly detection method according to claim 1, characterized in that, The steps for standard processing of time-series current and voltage signals, infrared thermal imaging images, and equipment vibration spectra include: Segmented overlapping sampling is used for time-series current and voltage signals, and Z-Score normalization is applied to eliminate dimensional differences; Histogram equalization is performed on infrared thermal imaging images to enhance contrast, while bilateral filtering is used to preserve edge thermal radiation features; The vibration spectrum of the equipment is generated into a time-frequency spectrum using short-time Fourier transform, and the dynamic range is compressed to the [-5, 5] dB interval using logarithmic transform.

3. The deep learning-based power equipment anomaly detection method according to claim 2, characterized in that, The first feature set is extracted using the following formula: The first feature set includes mean, standard deviation, skewness, kurtosis, RMS value, and peak factor. Calculate the skewness using the following formula: Calculate kurtosis using the following formula: Calculate the effective value using the following formula: The peak factor is calculated using the following formula: Among them, S i Let μ be the skewness of the i-th electrical signal, X(t,i) be the sampled value of the i-th electrical signal at time t, and μ be the skewness of the signal. i Let σ be the mean of the i-th electrical signal. i K is the standard deviation of the i-th electrical signal. i R is the kurtosis of the i-th electrical signal. i Let C be the effective value of the i-th electrical signal, T be the total number of electrical signals, and C be the effective value of the i-th electrical signal. i Let be the peak factor of the i-th electrical signal.

4. The deep learning-based power equipment anomaly detection method according to claim 3, characterized in that, The second feature set includes the highest temperature, lowest temperature, maximum temperature difference, percentage of high-temperature region, and maximum temperature gradient; The percentage of high-temperature areas is calculated using the following formula: Ω={(x,y)|X(x,y)>T max -5%ΔT}; The maximum temperature gradient is calculated using the following formula: Where H represents the proportion of high-temperature regions, Ω represents the set of high-temperature regions, h and w are the length and width of the infrared thermal image, respectively, X(x,y) is the temperature value at coordinates (x,y) in the infrared thermal image, and T... max The highest temperature is ΔT, the maximum temperature difference is G. max This represents the maximum temperature gradient.

5. The deep learning-based power equipment anomaly detection method according to claim 4, characterized in that, The third feature set includes total energy, first frequency energy, second frequency energy, third frequency energy, and frequency center; The frequencies in the vibration spectrum of the equipment are sorted, and the total energy of the highest third is taken as the first frequency energy, the total energy of the next third is taken as the second frequency energy, and the total energy of the lowest third is taken as the third frequency energy.

6. The deep learning-based power equipment anomaly detection method according to claim 5, characterized in that, The step of fusing the first feature set, the second feature set, and the third feature set to obtain the fused features includes: The features in the first feature set, the second feature set, and the third feature set are concatenated to obtain a fused feature matrix, which is a one-dimensional feature matrix.

7. The deep learning-based power equipment anomaly detection method according to claim 5, characterized in that, The step of traversing all anomaly probabilities according to the anomaly type and outputting the final detection result based on the traversal results includes: Based on the anomaly type, obtain the probability threshold for each anomaly type from a preset association table; And determine whether the anomaly probability is greater than the corresponding probability threshold; If the abnormal probability is greater than the corresponding probability threshold, an abnormality is determined and the abnormality type is output. If the abnormal probability is less than or equal to the corresponding probability threshold, it is judged as normal; All traversal results are summarized to obtain the final detection result.

8. A deep learning-based power equipment anomaly detection system, characterized in that, The system includes: The data acquisition module is used to acquire the time-series current and voltage signals, infrared thermal imaging images, and equipment vibration spectrum of the power equipment at first preset time intervals, and to perform standard processing on the time-series current and voltage signals, infrared thermal imaging images, and equipment vibration spectrum. The feature extraction module is used to extract a first feature set, a second feature set, and a third feature set from the time-series current and voltage signals, infrared thermal imaging images, and equipment vibration spectra that have undergone standard processing, and to fuse the first feature set, the second feature set, and the third feature set to obtain fused features; The probability prediction module is used to input the fused features into a pre-trained anomaly detection model to obtain the anomaly probability for each anomaly type. The abnormal result output module is used to traverse all abnormal probabilities according to the abnormal type and output the final detection result based on the traversal result.

9. A storage medium, characterized in that, The storage medium stores one or more programs that, when executed by a processor, implement the deep learning-based power equipment anomaly detection method as described in any one of claims 1-7.

10. An electronic device comprising a memory and a processor, wherein: The memory is used to store computer programs; When the processor executes the computer program stored in the memory, it implements the deep learning-based power equipment anomaly detection method as described in any one of claims 1-7.

Images