Transformer abnormal sound detection method, system and device in complex environment and storage medium

Abstract

The invention discloses a method, a system and equipment for detecting abnormal sound of a transformer in a complex environment and a storage medium. The method comprises the following steps: acquiring a sound signal of the transformer in an operation process; preprocessing the collected original sound signal, wherein the preprocessing comprises pre-emphasis, framing, windowing and fast Fourier transform; a filter bank feature extraction technology is adopted to capture frequency distribution and energy change of the sound signals, and Fbank features are generated; a multi-head attention mechanism and a ResNet-SE model are combined to construct a voiceprint recognition model; sound signals of the transformer in the operation process are collected in real time, the collected sound signals are subjected to preprocessing and feature extraction and then input into the voiceprint recognition model, and fault detection and classification of the transformer are carried out through the voiceprint recognition model. The method can solve the problem that the abnormal sound of the transformer is difficult to accurately detect by a traditional method in a complex noise environment, and improves the detection precision and sensitivity.

Description

Technical Field

[0001] The present invention relates to transformer fault detection technology, and in particular to a transformer abnormal sound detection method, system, equipment and storage medium under complex environment. Background Art

[0002] Power transformers are key equipment in power systems, which can realize the conversion of AC power between different voltage levels to reduce transmission losses and improve transmission efficiency. With the widespread use of transformers, their failure frequency is also increasing. Common failures include loose core, winding failure, insulation aging, etc. If these failures are not discovered and handled in time, they may cause equipment damage or even cause large-scale power outages in the power system. Therefore, timely and effective detection and diagnosis of transformer failures are of great significance to ensure the safe and stable operation of the power system.

[0003] At present, transformer fault diagnosis technology mainly includes oil quality analysis (DGA), infrared thermal imaging and electrical testing. Oil quality analysis determines the internal fault type of the transformer by detecting the gas composition and concentration dissolved in the transformer oil. The technology is mature and highly sensitive, but it requires professional equipment and operation, and cannot achieve online monitoring; infrared thermal imaging uses infrared camera equipment to detect the temperature distribution on the surface of the transformer, identify the hot spot area, and determine the fault location. This method is intuitive and non-contact, suitable for detecting overheating faults, but it is greatly affected by the ambient temperature and has limited resolution; electrical testing determines the health status of the transformer by measuring the electrical parameters of the transformer (such as winding resistance, transformation ratio, dielectric loss, etc.). The test is highly accurate, but it usually requires shutdown and cannot achieve real-time monitoring. Although the above methods have certain application value in transformer fault diagnosis, they still have shortcomings such as insufficient real-time performance, high equipment and operation requirements, and limited online monitoring capabilities.

[0004] To overcome these shortcomings, voiceprint monitoring technology has gradually attracted attention as an emerging fault diagnosis method. Sound waves and vibrations are important information carriers of the operating status of transformer mechanical equipment. Traditional methods have limitations in detecting abnormal sounds in complex noise environments, such as wind, rain, human voices, thunder, and animal calls, which are irregular and random noises that are difficult to predict and filter. Summary of the invention

[0005] Purpose of the invention: The purpose of the present invention is to provide a method, system, device and storage medium for detecting abnormal sound of transformer in complex environment, which can solve the problem that traditional methods are difficult to accurately detect abnormal sound of transformer in complex noise environment, and improve the accuracy and sensitivity of detection.

[0006] Technical solution: A method for detecting abnormal sound of a transformer in a complex environment of the present invention comprises:

[0007] Data acquisition: Collect the sound signals of the transformer during operation, including the normal operation sound and fault sound of the transformer. Organize and filter the collected original sound signals to form a data set. Mark the original sound signals in the data set according to different fault types and normal states. Divide the marked data set into basic training set, basic verification set and basic test set.

[0008] Data preprocessing: The original sound signals in the basic training set, basic validation set and basic test set are preprocessed respectively, including pre-emphasis, framing, windowing, fast Fourier transform and calculation of spectral line energy;

[0009] Feature extraction: Filter bank feature extraction technology is used to capture the frequency distribution and energy changes of the preprocessed sound signals from the preprocessed basic training set, basic validation set, and basic test set, generate Fbank features, and finally obtain the training set, validation set, and test set;

[0010] Model construction and training: The original voiceprint recognition model is constructed by combining the multi-head attention mechanism with the ResNet-SE model. The multi-head attention mechanism captures information from different feature subspaces by executing multiple independent attention operations in parallel. The original voiceprint recognition model is trained using the training set, and the validation set is used to tune the original voiceprint recognition model during the training process. The performance of the original voiceprint recognition model is then tested using the test set to obtain a voiceprint recognition model for transformer abnormal sound detection.

[0011] Detection module: collects sound signals of the transformer in real time during operation, pre-processes and extracts features of the collected sound signals, and then inputs them into the voiceprint recognition model, which is used to detect and classify transformer faults.

[0012] Furthermore, the original sound signal data in the basic training set, the basic validation set and the basic test set are preprocessed respectively, including:

[0013] The original sound signal is pre-emphasized and the pre-emphasis is achieved by the following formula:

[0014] y[n]=x[n]-0.97·x[n-1]

[0015] Among them, y[n] is the pre-emphasized signal; x[n] is the original signal;

[0016] The continuous sound signal is divided into multiple overlapping short-time frames, and the Hamming window is applied to each frame signal. The formula is as follows:

[0017]

[0018] The windowed signal is:

[0019] y w [n] = x[n]·w[n]

[0020] Perform fast Fourier transform on each frame of windowed signal to convert the time domain signal into the frequency domain signal. The formula is as follows:

[0021]

[0022] Where X[k] is the kth frequency component; N is the length of each frame;

[0023] Calculate the energy spectrum:

[0024] ∣X[k]∣ 2 =X[k]×X*[k]

[0025] where X*[k] is the complex conjugate of X[k].

[0026] Furthermore, the filter bank feature extraction technology is used to capture the frequency distribution and energy changes of the preprocessed sound signals from the preprocessed basic training set, basic validation set and basic test set, respectively, to generate Fbank features, including:

[0027] The power spectrum is passed through a set of Mel filters, and the filter output formula is as follows:

[0028]

[0029] Where, Em is the output energy of the mth filter; Hm[k] is the response of the mth filter at frequency k;

[0030] Taking the logarithm of the output energy of each filter, we get the Fbank feature, which is as follows:

[0031] F m =log(E m ).

[0032] Furthermore, the multi-head attention mechanism captures information from different feature subspaces by executing multiple independent attention operations in parallel. The formula is:

[0033]

[0034] Where Q, K, and V represent query, key, and value matrices, respectively; dk is the dimension of the key vector.

[0035] Furthermore, the ResNet-SE model introduces a residual structure and a compression and excitation module, and adaptively adjusts the weights of feature channels to enhance the feature expression capability of the model. The residual block formula is:

[0036] y=F(x,{W i})+x

[0037] Among them, F(x,{W i}) represents the convolution operation; x represents the input; y represents the output;

[0038] The compression and excitation module adjusts the weights of the feature channels through global average pooling and two fully connected layers. The formula is as follows:

[0039] compression:

[0040]

[0041] excitation:

[0042] s=σ(W2·ReLU(W1·z))

[0043] Among them, X c (i, j) is the cth channel of the input feature map; H and W are the height and width of the feature map respectively; W1 and W2 are the weights of the fully connected layer; σ is the Sigmoid activation function.

[0044] Furthermore, the original voiceprint recognition model is trained using the training set and optimized using the cross entropy loss function and the Adam optimizer. The cross entropy loss function is defined as follows:

[0045]

[0046] Where N is the number of samples; C is the number of categories; y ij is the true label of sample i in category j; is the probability predicted by the model.

[0047] Furthermore, the multi-head attention mechanism enables the original voiceprint recognition model to enhance the focus on features that have an important impact on fault diagnosis by dynamically adjusting the weights of different features.

[0048] Based on the same inventive concept, the present invention provides a transformer abnormal sound detection system in a complex environment, comprising:

[0049] The data acquisition module is used to collect the sound signals of the transformer during operation. The sound signals include the normal operation sound and fault sound of the transformer, sort and filter the collected original sound signals to form a data set, and mark the original sound signals in the data set according to different fault types and normal states, and divide the marked data set into a basic training set, a basic verification set and a basic test set;

[0050] A data preprocessing module, used to preprocess the original sound signals in the basic training set, the basic validation set and the basic test set, including pre-emphasis, framing, windowing and fast Fourier transform;

[0051] The feature extraction module is used to use the filter bank feature extraction technology to capture the frequency distribution and energy change of the preprocessed sound signal from the preprocessed basic training set, basic validation set and basic test set, generate Fbank features, and finally obtain the training set, validation set and test set;

[0052] The model building and training module is used to combine the multi-head attention mechanism with the ResNet-SE model to build an original voiceprint recognition model. The multi-head attention mechanism captures information from different feature subspaces by executing multiple independent attention operations in parallel. The module is used to train the original voiceprint recognition model using the training set, and to optimize the original voiceprint recognition model using the validation set during the training process. The performance of the original voiceprint recognition model is then tested using the test set to obtain a voiceprint recognition model for transformer abnormal sound detection.

[0053] The detection module is used to collect the sound signals of the transformer in real time during operation, pre-process and extract the features of the collected sound signals, and then input them into the voiceprint recognition model, and perform transformer fault detection and classification through the voiceprint recognition model.

[0054] Based on the same inventive concept, a transformer abnormal sound detection device in a complex environment of the present invention includes a processor and a memory, wherein the memory stores computer instructions, and the processor is used to execute the computer instructions stored in the memory. When the computer instructions are executed by the processor, the electronic device implements the steps of the transformer abnormal sound detection method in a complex environment as described above.

[0055] Based on the same inventive concept, a computer-readable storage medium of the present invention stores a computer program thereon, and when the program is executed by a processor, the steps of the transformer abnormal sound detection method under a complex environment as described above are implemented.

[0056] Beneficial effects: Compared with the prior art, the significant technical effects of the present invention are:

[0057] High precision and high sensitivity: The combination of the multi-head attention mechanism and the ResNet-SE model significantly improves the accuracy and sensitivity of transformer fault detection, and can accurately identify faults in complex noisy environments.

[0058] Real-time: Through the construction of a real-time monitoring system, the operating status of the transformer can be monitored in real time, ensuring early detection of potential problems and thus avoiding major failures.

[0059] Automation and intelligence: The automation and intelligence of the system significantly reduce the workload of manual inspections, improve operation and maintenance efficiency, and ensure the stable operation of the power system.

[0060] Robustness: The method of the present invention is highly robust to environmental noise and can operate stably in a variety of complex environments, thereby improving the practicability and reliability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 It is a flow chart of a method for detecting abnormal sound of a transformer in a complex environment disclosed in an embodiment of the present invention;

[0062] Figure 2 It is a distribution diagram of a transformer sound signal collection device disclosed in an embodiment of the present invention;

[0063] Figure 3 It is a schematic diagram of an Fbank feature extraction process disclosed in an embodiment of the present invention;

[0064] Figure 4 It is a structural schematic diagram of a voiceprint recognition model disclosed in an embodiment of the present invention;

[0065] Figure 5 It is a spectrum diagram of a transformer fault detection system in different noise environments disclosed in an embodiment of the present invention;

[0066] Figure 6 is a performance comparison chart of an improved ResNet-SE model and an original ResNet-SE model disclosed in an embodiment of the present invention;

[0067] Figure 7 is a chart showing the effects of different learning rates on model accuracy in the experimental results disclosed in the embodiments of the present invention;

[0068] Figure 8 It is a chart showing the change of loss function of an improved Att-ResNet-SE model and a ResNet-SE model during training according to an embodiment of the present invention;

[0069] Fig. 9 It is a confusion matrix diagram of a model reasoning result disclosed in an embodiment of the present invention;

[0070] Fig.10 It is a structural schematic diagram of a transformer abnormal sound detection system in a complex environment disclosed in an embodiment of the present invention;

[0071] Fig.11 It is a structural schematic diagram of a transformer abnormal sound detection device in a complex environment disclosed in an embodiment of the present invention. DETAILED DESCRIPTION

[0072] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. It will be understood by those skilled in the art that the purposes and advantages that can be achieved with the present invention are not limited to the specific description of the above beneficial effects, and the above and other purposes that can be achieved by the present invention will be more clearly understood based on the following detailed description.

[0073] It should be understood by those skilled in the art that the exemplary components, systems and methods described in conjunction with the embodiments disclosed in the present invention can be implemented in hardware, software or a combination of the two. Whether it is implemented in hardware or software depends on the specific application and design of the technical solution. Professional and technical personnel can use different methods to implement the functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0074] Reference to "embodiments" in the present invention means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present invention. The appearance of the phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0075] Example 1

[0076] See also Figure 1 , Figure 1 1 is a flow chart of a method for detecting abnormal sound of a transformer in a complex environment disclosed in an embodiment of the present invention. Figure 1 The transformer abnormal sound detection method described is applied to power systems, such as transformer fault detection, etc., which is not limited in the embodiments of the present invention. Figure 1 As shown, the transformer abnormal sound detection method in the complex environment may include the following operations:

[0077] S1. Data acquisition: Collect the sound signals of the transformer during operation, including the normal operation sound and fault sound of the transformer, sort and filter the collected original sound signals to form a data set, and mark the original sound signals in the data set according to different fault types and normal states, and divide the marked data set into a basic training set, a basic verification set and a basic test set. The details are as follows:

[0078] S1.1. Collect transformer sound signals: Use transformer sound signal collection device to collect transformer sound signals in real time during operation. These signals include normal operation sound and fault sound to ensure the comprehensiveness and accuracy of data.

[0079] The transformer sound signal collection device may use a high-sensitivity microphone or sensor. Figure 2 As shown, in this embodiment, a microphone is used to collect sound signals, the frequency response range of the microphone is 20 Hz to 20 kHz, each sampling time is 3 seconds, and the sampling frequency is 16 kHz.

[0080] S1.2. Data collation and screening: Sort and screen the collected sound signals to eliminate noise interference and invalid data to ensure data quality.

[0081] S1.3. Set category labels: Label the sorted data according to different fault types and normal states, and establish a complete labeling system to facilitate subsequent model training and evaluation.

[0082] S1.4. Dataset division: The labeled dataset is divided into a basic training set, a basic validation set, and a basic test set. Usually, it is divided into 85% basic training set, 10% basic validation set, and 5% basic test set.

[0083] S2, data preprocessing: Figure 3 As shown, the original sound signal data in the basic training set, basic validation set and basic test set are preprocessed respectively, including pre-emphasis, framing, windowing, fast Fourier transform (FFT) and calculation of spectral line energy. The details are as follows:

[0084] S2.1, Pre-emphasis: Pre-emphasize the original sound signal to enhance the high-frequency components and improve the resolution of the spectrum. Pre-emphasis is achieved through the following formula:

[0085] y[n]=x[n]-0.97·x[n-1]

[0086] Where y[n] is the pre-emphasized signal; x[n] is the original signal, and 0.97 is the pre-emphasis coefficient;

[0087] S2.2, Framing and windowing: The continuous sound signal is divided into multiple overlapping short-time frames, each with a length of 25 ms , frame shift is 10 ms . Apply the Hamming window to each frame signal, the formula is as follows:

[0088]

[0089] The windowed signal is:

[0090] y w [n]=x[n].w[n]

[0091] S2.3, Fast Fourier Transform (FFT): Perform fast Fourier transform on each frame of windowed signal to convert the time domain signal into the frequency domain signal. The formula is as follows:

[0092]

[0093] Wherein, X[k] is the kth frequency component; N is the length of each frame.

[0094] S2.4. Calculate the energy spectrum:

[0095] ∣X[k]∣ 2 =X[k]×X*[k]

[0096] where X*[k] is the complex conjugate of X[k].

[0097] S3, feature extraction: Figure 3 As shown, the filter bank feature extraction technology is used to capture the frequency distribution and energy changes of the preprocessed sound signal from the preprocessed basic training set, basic verification set and basic test set, generate Fbank features, and finally obtain the training set, verification set and test set. The details are as follows:

[0098] S3.1, Mel filter bank: The power spectrum is passed through a set of Mel filters, and the filter output formula is as follows:

[0099]

[0100] Where, Em is the output energy of the mth filter; Hm[k] is the response of the mth filter at frequency k;

[0101] S3.2. Calculate the logarithmic energy value: Take the logarithm of the output energy of each filter to obtain the Fbank feature. The formula is as follows:

[0102] F m =log(E m ).

[0103] S4. Model construction and training: Figure 4 As shown in the figure, the original voiceprint recognition model is constructed by combining the multi-head attention mechanism with the ResNet-SE model; the multi-head attention mechanism captures the information of different feature subspaces by executing multiple independent attention operations in parallel; the original voiceprint recognition model is trained with the training set, and the original voiceprint recognition model is tuned with the validation set during the training process, and then the performance of the original voiceprint recognition model is tested with the test set, so as to obtain the voiceprint recognition model for transformer abnormal sound detection.

[0104] Combining the multi-head attention mechanism and the ResNet-SE (Residual Network-Compression and Excitation) model, a voiceprint recognition model for transformer fault detection is constructed.

[0105] Among them, the multi-head attention mechanism: The multi-head attention mechanism captures information in different feature subspaces by executing multiple independent attention operations in parallel. The formula is:

[0106]

[0107] Where Q, K, and V represent query, key, and value matrices, respectively; dk is the dimension of the key vector.

[0108] ResNet-SE model: The ResNet-SE model introduces a residual structure and a compression and excitation module, and adaptively adjusts the weights of feature channels to enhance the model's feature expression capabilities. The residual block formula is:

[0109] y=F(x,{W i})+x

[0110] Where F(x,{Wi}) represents the convolution operation; x represents the input; y represents the output;

[0111] SE module: The compression and excitation module adjusts the weights of the feature channels through global average pooling and two fully connected layers. The formula is as follows:

[0112] Squeeze:

[0113]

[0114] Excitation:

[0115] s=σ(W2·ReLU(W1·z))

[0116] Among them, Xc(i,j) is the cth channel of the input feature map; H and W are the height and width of the feature map respectively; W1 and W2 are the weights of the fully connected layer; σ is the Sigmoid activation function.

[0117] The original voiceprint recognition model is trained using the training set and optimized using the cross entropy loss function and the Adam optimizer. The cross entropy loss function is defined as follows:

[0118]

[0119] Where N is the number of samples; C is the number of categories; y ij is the true label of sample i in category j; is the probability predicted by the model.

[0120] S5. Fault detection: Real-time collection of sound signals of the transformer during operation, pre-processing and feature extraction of the collected sound signals, and then input into the voiceprint recognition model, and the voiceprint recognition model is used to detect and classify transformer faults. Specifically, it includes:

[0121] S5.1. Real-time collection of transformer voiceprint data by microphone: In practical applications, a microphone is used to collect transformer operation sound signals in real time to ensure the real-time and continuity of data. The collected signal preprocessing steps are the same as those in the training stage, including pre-emphasis, framing, windowing, and fast Fourier transform (FFT); the feature extraction steps are also the same as those in the training stage.

[0122] S5.2. Input into the voiceprint recognition model: Input the voiceprint data collected in real time into the trained voiceprint recognition model for fault detection and classification.

[0123] S5.3. Output transformer voiceprint classification: Use the trained model to classify the extracted Fbank features. The model outputs the transformer operating status classification results, including normal status and various fault states (such as loose core, etc.). The classification results can be expressed as:

[0124]

[0125] in, is the predicted classification result; f(F) is the predicted output of the model for the input feature F; the argmax function returns the category index with the highest probability.

[0126] like Figure 5 The data in the model is trained by mixing 80% of the core loosening data and normal data with other noise data. The model structure is as follows: Figure 4 As shown, after the model of this structure is trained and exported, it is evaluated. The results are as follows:

[0127] from Fig. 9 As can be seen, the model has achieved very good results in 12 sound classifications.

[0128] The 12 sounds correspond to different working conditions and record the sample situations accompanied by different background noises (human voice, wind noise, thunder, thunderstorm, bird song and no background noise). The sample distribution describes the real application scenarios of outdoor transformers in more detail. Each working condition and its description has a unique label, with a total of 12 different descriptions, including Fig. 9 What is displayed is a confusion matrix, where normal represents normal and eighty represents 80 percent of the core is loose.

[0129] Overall, the model performs well on most categories, however, the model has obvious misclassification problems on some categories. For example, the proportion of the normal category being misclassified as bird_normal and wind_normal is 0.20 and 0.12 respectively, indicating that the model has certain challenges in distinguishing these categories. In addition, the misclassification problem of the wind_normal category is particularly prominent, with a proportion of 0.28 being misclassified as eighty, resulting in an overall accuracy of only 0.80 for this category. Similarly, the prediction accuracy of the eighty category is also low, at only 0.72, showing the model's shortcomings in handling this category. In summary, although the model performs well on most categories, the misclassification problems of the normal, wind_normal, and eighty categories require further optimization of the model or improvement of the dataset to improve the overall classification performance.

[0130] like Figure 6 As shown in the figure, different models and input data types have a significant impact on performance. Overall, the models using Fbank data perform better than the models using MFCC data, especially the improved Resnet-Se model with a larger number of parameters, which achieves nearly 100% performance on Fbank data. In contrast, the models with smaller parameters perform poorly on both data types, especially when using MFCC data, with a performance of only 51.31%. This shows that the number of parameters, input data type, and model architecture of the model have an important impact on its performance, especially when Fbank data is combined with the improved Resnet-Se model with a high number of parameters, the performance is the best.

[0131] like Figure 7 As shown, Figure 7 The trend of the improved model accuracy changing with the learning rate is shown. As the learning rate gradually increases from 0.0001 to 0.001, the accuracy of the model is significantly improved, reaching a maximum value of 99.61%. However, when the learning rate continues to increase to 0.01 and 0.1, the accuracy of the model begins to decline, dropping to 98.16% and 97.02% respectively. This shows that a moderate learning rate (0.001) can effectively improve model performance, while a too high learning rate may lead to a significant decrease in model accuracy. Therefore, during model training, choosing an appropriate learning rate is crucial to achieving optimal performance.

[0132] like Figure 8 As shown, Figure 8The figure shows the changing trend of the loss function (Train Loss) of two different models (Att-Resnet-Se and improved Resnet-Se) during the training process with the number of training rounds (Epoch). As can be seen from the figure, with the increase of the number of training rounds, the training loss of both models shows a rapid downward trend, and the loss decreases most significantly in the early stage of training (the first 10 Epochs). The loss value of the improved Resnet-Se model (orange line) is relatively high in the early stage, but it shows a relatively stable downward trend in subsequent training and stabilizes after about 20 Epochs. The loss value of the Att-Resnet-Se model (blue line) fluctuates greatly during the entire training process, especially in the first 30 Epochs, but it also reaches a lower loss value in the end. Overall, the training loss of the improved Resnet-Se model decreases more steadily, indicating that the model may have better convergence and stability. The loss values ​​of the two models at the end of training are close, indicating that the performance of the two models after training may not be much different, but the improved Resnet-Se model shows better stability and consistency during training. (Note: The improved Resnet-Se is optimized in terms of parameters, while the Att-Resnet-Se model is an improved Resnet-Se with an attention mechanism added to increase feature clustering.)

[0133] The solution of the present invention uses a multi-head attention mechanism, which dynamically adjusts the weights of different features, so that the model can pay more attention to the features that have an important impact on fault diagnosis, thereby improving the diagnostic performance. In addition, the ResNet-SE model, which combines the residual network (ResNet) and the compression and excitation (SE) module, can adaptively adjust the weights of each channel by introducing the SE module, further enhancing the model's ability to capture important features.

[0134] Based on the voiceprint monitoring principle, the present invention combines the innovative method of multi-head attention mechanism and ResNet-SE deep learning model to establish a voiceprint sample library and model library of transformer operating status, aiming to realize real-time monitoring of the operating status of large transformer units, thereby overcoming the shortcomings of the existing technology and improving the accuracy and real-time performance of transformer fault detection.

[0135] The present invention aims to solve the technical problem that it is difficult for traditional methods to accurately detect abnormal sounds of transformers in complex noise environments, and the existing technologies have problems such as insufficient real-time performance, high equipment and operation requirements, and limited online monitoring capabilities. The present invention proposes a method for detecting abnormal sounds of transformers in complex environments, which can achieve high-precision and high-sensitivity transformer fault detection in complex noise environments.

[0136] Example 2

[0137] See also Fig.10 , Fig.10 : is a schematic diagram of a structure of a transformer abnormal sound detection system in a complex environment disclosed in an embodiment of the present invention. The system can realize transformer fault detection, specifically including:

[0138] The data acquisition module is used to collect the sound signals of the transformer during operation. The sound signals include the normal operation sound and fault sound of the transformer, sort and filter the collected original sound signals to form a data set, and mark the original sound signals in the data set according to different fault types and normal states, and divide the marked data set into a basic training set, a basic verification set and a basic test set;

[0139] A data preprocessing module, used to preprocess the original sound signals in the basic training set, the basic validation set and the basic test set, including pre-emphasis, framing, windowing and fast Fourier transform;

[0140] The feature extraction module is used to use the filter bank feature extraction technology to capture the frequency distribution and energy change of the preprocessed sound signal from the preprocessed basic training set, basic validation set and basic test set, generate Fbank features, and finally obtain the training set, validation set and test set;

[0141] The model building and training module is used to combine the multi-head attention mechanism with the ResNet-SE model to build an original voiceprint recognition model. The multi-head attention mechanism captures information from different feature subspaces by executing multiple independent attention operations in parallel. The module is used to train the original voiceprint recognition model using the training set, and to optimize the original voiceprint recognition model using the validation set during the training process. The performance of the original voiceprint recognition model is then tested using the test set to obtain a voiceprint recognition model for transformer abnormal sound detection.

[0142] The fault detection module is used to collect the sound signals of the transformer in real time during operation, pre-process and extract the features of the collected sound signals, and then input them into the voiceprint recognition model to detect and classify the transformer faults through the voiceprint recognition model.

[0143] In an optional embodiment, the transformer abnormal sound detection system in a complex environment includes: a) collecting sound signals of the transformer during operation; b) preprocessing the collected original sound signals, including pre-emphasis, framing, windowing and fast Fourier transform; c) using filter bank (Fbank) feature extraction technology to capture the frequency distribution and energy changes of the sound signal and generate Fbank features; c) combining the multi-head attention mechanism with the ResNet-SE model to construct a voiceprint recognition model; d) collecting the sound signals of the transformer during operation in real time, preprocessing and feature extracting the collected sound signals, and then inputting them into the voiceprint recognition model, and performing transformer fault detection and classification through the voiceprint recognition model.

[0144] The solution of the present invention is aimed at detecting abnormal sound of transformers under complex environmental noise conditions. Based on the principle of voiceprint monitoring, combined with the multi-head attention mechanism and the ResNet-SE (Residual Network-Compression and Excitation) deep learning model, a voiceprint sample library and model library of the transformer operating status are established. By constructing a voiceprint monitoring system for the vibration state of the transformer, real-time monitoring of the operating status of large transformer units is achieved, and transformer faults such as loose core can be accurately identified in complex noise environments. The system has high precision and high sensitivity, ensuring early detection of potential problems, thereby avoiding major failures and ensuring the stable operation of the power system.

[0145] Example 3

[0146] See also Fig.11 , Fig.11 Schematic diagram of a transformer abnormal sound detection device in a complex environment disclosed in an embodiment of the present invention. Fig.11 The described device can be applied to a power system, such as transformer fault detection, etc., which is not limited in the embodiments of the present invention.

[0147] like Fig.11 As shown, the device may include a processor and a memory, wherein the memory stores computer instructions, and the processor is used to execute the computer instructions stored in the memory. When the computer instructions are executed by the processor, the electronic device implements the steps of the method described in the above embodiment and can achieve the technical effect consistent with the above method.

[0148] The memory may include computer system readable media in the form of volatile memory, such as random access memory (RAM) and / or cache memory. The device may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, the memory may be used to read and write non-removable, non-volatile magnetic media (commonly referred to as a "hard drive"). A program / utility having a set (at least one) of program modules may be stored in, for example, the memory, such program modules including but not limited to an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include the implementation of a network environment. The program modules typically perform the functions and / or methods in the embodiments described herein.

[0149] The processor executes various functional applications and data processing by running the program stored in the memory, such as implementing the method provided in the first embodiment of the present invention.

[0150] Example 4

[0151] Embodiment 4 of the present invention further provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the steps of the method described in the above embodiment are implemented and the technical effect consistent with the above method can be achieved.

[0152] The computer storage medium of the embodiment of the present invention can adopt any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium can be, for example, but not limited to, a system, device or device of electricity, magnetism, light, electromagnetic, infrared, or semiconductor, or any combination of the above. More specific examples (non-exhaustive list) of computer-readable storage media include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this document, a computer-readable storage medium can be any tangible medium containing or storing a program, which can be used by an instruction execution system, a device or a device or used in combination with it.

[0153] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, which carry computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0154] The program code embodied on the computer readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

[0155] Computer program code for performing the operation of the present invention may be written in one or more programming languages ​​or combinations thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer through any type of network including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0156] Of course, the storage medium containing computer executable instructions provided by an embodiment of the present invention, whose computer executable instructions are not limited to the above method operations, can also execute related operations in the method provided by any embodiment of the present invention.

[0157] The specific embodiments described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A transformer abnormal sound detection method in a complex environment, characterized in that: include: Data acquisition: Collect the sound signals of the transformer during operation, including the normal operation sound and fault sound of the transformer. Organize and filter the collected original sound signals to form a data set. Mark the original sound signals in the data set according to different fault types and normal states. Divide the marked data set into basic training set, basic verification set and basic test set. Data preprocessing: The original sound signals in the basic training set, basic validation set and basic test set are preprocessed respectively, including pre-emphasis, framing, windowing, fast Fourier transform and calculation of spectral line energy; Feature extraction: Filter bank feature extraction technology is used to capture the frequency distribution and energy changes of the preprocessed sound signals from the preprocessed basic training set, basic validation set, and basic test set, generate Fbank features, and finally obtain the training set, validation set, and test set; Model construction and training: The original voiceprint recognition model is constructed by combining the multi-head attention mechanism with the ResNet-SE model. The multi-head attention mechanism captures information from different feature subspaces by executing multiple independent attention operations in parallel. The original voiceprint recognition model is trained using the training set, and the validation set is used to tune the original voiceprint recognition model during the training process. The performance of the original voiceprint recognition model is then tested using the test set to obtain a voiceprint recognition model for transformer abnormal sound detection. Fault detection: The sound signals of the transformer during operation are collected in real time. After preprocessing and feature extraction, the collected sound signals are input into the voiceprint recognition model, and the transformer fault detection and classification are performed through the voiceprint recognition model.

2. The transformer abnormal sound detection method in a complex environment according to claim 1 is characterized in that: The original sound signal data in the basic training set, basic validation set and basic test set are preprocessed respectively, including: The original sound signal is pre-emphasized and the pre-emphasis is achieved by the following formula: y[n]=x[n]-0.97·x[n-1] Among them, y[n] is the pre-emphasized signal; x[n] is the original signal; The continuous sound signal is divided into multiple overlapping short-time frames, and the Hamming window is applied to each frame signal. The formula is as follows: The windowed signal is: y W [n]=X[n]·W[n] Perform fast Fourier transform on each frame of windowed signal to convert the time domain signal into the frequency domain signal. The formula is as follows: Where X[k] is the kth frequency component; N is the length of each frame; Calculate the energy spectrum: |X[k]| 2 =X[k]×X * [k] Among them, X * [k] is the complex conjugate of X[k].

3. The method for detecting abnormal sound of a transformer in a complex environment according to claim 2, characterized in that: The filter bank feature extraction technology is used to capture the frequency distribution and energy changes of the preprocessed sound signal from the preprocessed basic training set, basic validation set and basic test set, and generate Fbank features, including: The power spectrum is passed through a set of Mel filters, and the filter output formula is as follows: Where, Em is the output energy of the mth filter; Hm[k] is the response of the mth filter at frequency k; Taking the logarithm of the output energy of each filter, we get the Fbank feature, which is as follows: F m =log(E m )。 4. The transformer abnormal sound detection method in a complex environment according to claim 1 is characterized in that: The multi-head attention mechanism captures information in different feature subspaces by performing multiple independent attention operations in parallel. The formula is: Where Q, K, and V represent query, key, and value matrices, respectively; dk is the dimension of the key vector.

5. The transformer abnormal sound detection method in a complex environment according to claim 1 is characterized in that: The ResNet-SE model introduces a residual structure and a compression and excitation module, and adaptively adjusts the weights of feature channels to enhance the feature expression capability of the model. The residual block formula is: y=F(x,{W i })+X Where F(x, {Wi}) represents the convolution operation; x represents the input; y represents the output; The compression and excitation module adjusts the weights of the feature channels through global average pooling and two fully connected layers. The formula is as follows: compression: excitation: s=σ(W2·ReLU(W1·z)) Among them, Xc(i, j) is the cth channel of the input feature map; H and W are the height and width of the feature map respectively; W1 and W2 are the weights of the fully connected layer; σ is the Sigmoid activation function.

6. The transformer abnormal sound detection method in a complex environment according to claim 1 is characterized in that: The original voiceprint recognition model is trained using the training set and optimized using the cross entropy loss function and the Adam optimizer. The cross entropy loss function is defined as follows: Where N is the number of samples; C is the number of categories; y ij is the true label of sample i in category j; is the probability predicted by the model.

7. The transformer abnormal sound detection method in a complex environment according to claim 1, characterized in that: The multi-head attention mechanism dynamically adjusts the weights of different features, so that the original voiceprint recognition model enhances the attention of features that have an important impact on fault diagnosis.

8. A transformer abnormal sound detection system in a complex environment, characterized in that: include: The data acquisition module is used to collect the sound signals of the transformer during operation. The sound signals include the normal operation sound and fault sound of the transformer, sort and filter the collected original sound signals to form a data set, and mark the original sound signals in the data set according to different fault types and normal states, and divide the marked data set into a basic training set, a basic verification set and a basic test set; The data preprocessing module is used to preprocess the original sound signals in the basic training set, the basic validation set and the basic test set, including pre-emphasis, framing, windowing, fast Fourier transform and calculation of spectral line energy; The feature extraction module is used to use the filter bank feature extraction technology to capture the frequency distribution and energy change of the preprocessed sound signal from the preprocessed basic training set, basic validation set and basic test set, generate Fbank features, and finally obtain the training set, validation set and test set; The model building and training module is used to combine the multi-head attention mechanism with the ResNet-SE model to build an original voiceprint recognition model. The multi-head attention mechanism captures information from different feature subspaces by executing multiple independent attention operations in parallel. The module is used to train the original voiceprint recognition model using the training set, and to optimize the original voiceprint recognition model using the validation set during the training process. The performance of the original voiceprint recognition model is then tested using the test set to obtain a voiceprint recognition model for transformer abnormal sound detection. The fault detection module is used to collect the sound signals of the transformer in real time during operation, pre-process and extract the features of the collected sound signals, and then input them into the voiceprint recognition model to detect and classify the transformer faults through the voiceprint recognition model.

9. A transformer abnormal sound detection device in a complex environment, characterized in that: The electronic device comprises a processor and a memory, wherein the memory stores computer instructions, and the processor is used to execute the computer instructions stored in the memory. When the computer instructions are executed by the processor, the electronic device implements the steps of the method for detecting abnormal sound of a transformer in a complex environment as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the method for detecting abnormal sound of a transformer in a complex environment as claimed in any one of claims 1 to 7.

Images