Transformer substation anomaly detection method, device and equipment and storage medium

Abstract

The invention relates to the technical field of sound detection, in particular to a transformer substation anomaly detection method, device and equipment and a storage medium, and the method comprises the steps: carrying out the preprocessing of a collected transformer substation audio signal, obtaining an abnormal sound data set, and carrying out the labeling of the abnormal sound data set, and obtaining the abnormal labeled sample data; removing human voice signals in the abnormal annotation sample data to obtain audio cleaning data, and adding noise to the audio cleaning data to obtain noise-added audio data; obtaining a Mel spectrum based on the noise-added audio data and preset indoor simulation impact response data, and processing the Mel spectrum to obtain an augmented feature sequence; and processing the augmented feature sequence based on a preset neural network to obtain a target vector, and determining the prediction classification of the substation audio signal corresponding to the target vector based on the target vector and a preset activation function. According to the invention, the detection cost of transformer substation anomaly detection is reduced, and the overall operation state of the transformer substation is comprehensively monitored.

Description

Technical Field

[0001] The present application relates to the field of sound detection technology, and in particular to a substation abnormality detection method, device, equipment and storage medium. Background Art

[0002] The power system refers to the overall energy supply and distribution network including power generation, transmission and distribution, and it continues to evolve and expand with technological progress and demand growth. As an important device in the power system, the operating status of the transformer is directly related to the safety of the social power system and the production and life of the power users. During the uninterrupted operation of the power system, the substation may experience unstable operation due to operational errors or aging. In order to analyze the causes of the unstable operation, it is necessary to perform abnormal detection on the substation.

[0003] At present, the methods of detecting abnormalities in substations include: obtaining sensor monitoring data based on temperature sensors, current sensors, pressure sensors, and flow monitoring equipment installed in the substation, and analyzing the sensor monitoring data to determine whether there is any abnormality in the substation.

[0004] However, the current substation anomaly detection method requires the installation of additional sensors in the substation, which increases the monitoring cost of substation anomaly detection; and, when analyzing the sensor monitoring data obtained by the sensor, specially trained personnel are required to analyze the sensor monitoring data, resulting in high personnel training costs, which further increases the monitoring cost of substation anomaly detection; in addition, the sensor monitoring data of each sensor can only reflect a specific state or parameter, and it is impossible to comprehensively monitor the overall operating status of the substation through a single sensor monitoring data; in summary, the current substation anomaly detection method has the problems of high monitoring cost and difficulty in achieving comprehensive monitoring of the overall operating status of the substation. Summary of the invention

[0005] In order to reduce the detection cost of substation anomaly detection and realize comprehensive monitoring of the overall operating status of the substation, the present application provides a substation anomaly detection method, device, equipment and storage medium.

[0006] In a first aspect, the present application provides a substation abnormality detection method, comprising:

[0007] Preprocessing the collected substation audio signals to obtain an abnormal sound data set, and labeling the abnormal sound data set to obtain abnormal labeled sample data;

[0008] Removing the human voice signal from the abnormally labeled sample data to obtain audio cleaning data, and adding noise to the audio cleaning data to obtain noisy audio data;

[0009] Obtaining a mel-spectrogram based on the noise-added frequency data and preset indoor simulated impulse response data, and processing the mel-spectrogram to obtain an augmented feature sequence;

[0010] The augmented feature sequence is processed based on a preset neural network to obtain a target vector, and a predicted classification of the substation audio signal corresponding to the target vector is determined based on the target vector and a preset activation function.

[0011] In a second aspect, the present application provides a substation abnormality detection device, comprising:

[0012] A processing and labeling module is used to pre-process the collected substation audio signals to obtain an abnormal sound data set, and label the abnormal sound data set to obtain abnormal labeled sample data;

[0013] A cleaning and denoising module, used to remove the human voice signal in the abnormally labeled sample data to obtain audio cleaning data, and to denoise the audio cleaning data to obtain noisy audio data;

[0014] A sequence generation module, used to obtain a mel-spectrogram based on the noisy frequency data and preset indoor simulated impulse response data, and process the mel-spectrogram to obtain an augmented feature sequence;

[0015] A prediction and classification module is used to process the augmented feature sequence based on a preset neural network to obtain a target vector, and determine the prediction classification of the substation audio signal corresponding to the target vector based on the target vector and a preset activation function.

[0016] In a third aspect, the present application provides a computer device, the computer device comprising a memory and a processor, the memory storing a computer program, and the processor implementing the steps in the above method when executing the computer program.

[0017] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which implements the steps in the above method when executed by a processor.

[0018] In a fifth aspect, the present application further provides a computer program product, wherein the computer program product comprises a computer program, and when the computer program is executed by a processor, the steps in any of the above method embodiments are implemented.

[0019] The above-mentioned substation abnormality detection method, device, equipment and storage medium obtain an abnormal sound data set by preprocessing the collected substation audio signal, label the abnormal sound data set to obtain abnormal labeled sample data; remove the human voice signal in the abnormal labeled sample data to obtain audio cleaning data, and add noise to the audio cleaning data to obtain noisy audio data; obtain a mel spectrogram based on the noisy audio data and preset indoor simulated impulse response data, and process the mel spectrogram to obtain an augmented feature sequence; process the augmented feature sequence based on a preset neural network to obtain a target vector, and determine the predicted classification of the substation audio signal corresponding to the target vector based on the target vector and a preset activation function. Through the above implementation, the use of sensors in the substation abnormality detection process is avoided as much as possible, thereby facilitating the reduction of monitoring costs, and since the collected substation audio signal is an overall audio signal related to the substation, the predicted classification of the substation audio signal is determined based on the substation audio signal, which can facilitate the further determination of the cause of the unstable working phenomenon of the substation based on the predicted classification, so as to facilitate the overall monitoring of the operating status of the substation.

[0020] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present application, nor is it intended to limit the scope of the present application. Other features of the present application will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0022] Figure 1 A flow chart of a substation abnormality detection method provided in an embodiment of the present application;

[0023] Figure 2 A schematic diagram of an augmented spectrum provided in an embodiment of the present application;

[0024] Figure 3 A schematic diagram of the structure of a substation abnormality detection device provided in an embodiment of the present application;

[0025] Figure 4 A schematic diagram of the structure of a computer device provided in an embodiment of the present application;

[0026] Figure 5 This is a diagram of the internal structure of a computer-readable storage medium provided in an embodiment of the present application. DETAILED DESCRIPTION

[0027] In order to make the purpose, technical solution and advantages of the present disclosure more clear, the present disclosure is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present disclosure and are not used to limit the present disclosure.

[0028] It should be noted that the terms "first", "second", etc. in the specification and claims of this article and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of this article described here can be implemented in an order other than those illustrated or described here. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, device, product or equipment that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or equipment.

[0029] In this article, the term "and / or" is only a description of the association relationship between related objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article generally indicates that the related objects before and after are in an "or" relationship.

[0030] Embodiment 1

[0031] Figure 1 A flow chart of a substation abnormality detection method provided in Example 1 of the present application, refer to Figure 1 The method may be performed by a device for performing the method, and the device may be implemented by software and / or hardware. The method includes:

[0032] S110, pre-processing the collected substation audio signals to obtain an abnormal sound data set, and labeling the abnormal sound data set to obtain abnormal labeled sample data.

[0033] Among them, the substation audio signal can be collected autonomously by artificial induction, and the substation audio signal includes: environmental sound signal, loose accessory sound signal, partial discharge sound signal, normal sound signal and overload sound signal; the environmental sound signal is the audio signal of the environment where the substation is located, the loose accessory sound signal is the audio signal generated when the substation accessory is loose, the partial discharge sound signal is the audio signal amplified locally in the substation, the normal sound signal is the audio signal when the substation is operating normally, and the overload audio signal is the audio signal generated when the substation is overloaded; preprocessing is the operation of uniformly processing the above-mentioned various substation audio signals to facilitate subsequent processing of various substation audio signals; the abnormal sound data set is a collection of various substation audio signals that have been preprocessed; label annotation is to use labels to annotate the abnormal feature information in various substation audio signals in the abnormal sound data set, wherein the abnormal feature information includes loose accessory features, partial discharge features and overload features, etc.; the abnormal annotation sample data is the abnormal sound data set after label annotation.

[0034] Specifically, an audio acquisition device is set at the substation, and the ambient sound signal, accessory loose sound signal, local discharge sound signal, normal sound signal and overload sound signal at the substation are collected by the audio acquisition device, so as to obtain five substation audio signals; then, the above five substation audio signals are preprocessed to obtain an abnormal sound data set; taking an abnormal sound data in the abnormal sound data set as an example, it is detected whether there is abnormal feature information in the abnormal sound data, and if so, the abnormal feature information is marked with a label corresponding to the abnormal feature information; the abnormal feature information in various abnormal sound data in the abnormal sound data set is marked in the above manner, so as to obtain abnormal marked sample data.

[0035] S120: removing the human voice signal from the abnormally labeled sample data to obtain audio cleaning data, and adding noise to the audio cleaning data to obtain noisy audio data.

[0036] Among them, there may be human voice signals in the abnormal labeled sample data, and human voice signals are easy to have an adverse effect on the substation abnormality detection process. For example, during the detection process, the human voice signal is misjudged as a signal emitted when the substation is abnormal. For this reason, it is necessary to remove the human voice signal from the abnormal labeled sample data. The audio cleaning data is the data formed after the human voice signal is removed from the abnormal labeled sample data; subsequently, a neural network for distinguishing the corresponding classification of the substation audio signal needs to be used to process the data based on the audio cleaning data. In order to improve the anti-interference ability and robustness of the neural network, the audio cleaning data needs to be denoised. Noising is to add Gaussian noise to the audio cleaning data, so as to improve the ability of the neural network to process data in a noisy environment, thereby improving the accuracy of distinguishing the corresponding classification of the substation audio signal; wherein, Gaussian noise is Gaussian noise with a sound intensity range of -25dB to 25dB and different signal-to-noise ratios with a step size of 5dB; the classification includes environmental sound, loose accessory sound, local discharge sound, normal sound, and overload sound; the noisy audio data is the data obtained after the audio cleaning data is denoised.

[0037] Specifically, detect whether there is a human voice signal in the abnormally labeled sample data. If so, remove the human voice signal in the abnormally labeled sample data to obtain audio cleaning data; further, add Gaussian noise with a sound intensity range of -25dB to 25dB and different signal-to-noise ratios in steps of 5dB to the audio cleaning data to obtain noisy audio data.

[0038] S130, obtaining a mel-spectrogram based on the noisy frequency data and preset indoor simulated impulse response data, and processing the mel-spectrogram to obtain an augmented feature sequence.

[0039] Among them, the substation has its corresponding indoor coefficients, such as sound reflection coefficient and sound absorption coefficient, etc. The indoor impact signal emitted by the indoor sound source can be processed by the indoor coefficient and the preset impact response simulation method, so as to obtain the indoor room impact response; different indoor room impact responses can be calculated by setting different indoor coefficients, and the indoor simulation impact response data is also a collection of indoor room impact responses corresponding to different indoor coefficients; further, a Mel spectrogram can be obtained by processing the noisy audio data and the preset indoor simulation impact response data. In this embodiment, the Mel spectrogram is specifically a logarithmic Mel spectrogram. In other embodiments, it is not specifically limited; the Mel spectrogram can display the change of the frequency spectrum of the corresponding audio signal over time; the augmented feature sequence is a data sequence obtained by processing the Mel spectrogram, and the augmented feature sequence is used for subsequent input into a preset neural network for processing, so as to determine and distinguish the classification corresponding to the substation audio signal.

[0040] Specifically, the noisy frequency data and the indoor simulated impulse response data are processed to obtain a mel-spectrogram; further, the mel-spectrogram is processed to obtain an augmented feature sequence.

[0041] S140. Process the augmented feature sequence based on a preset neural network to obtain a target vector, and determine a predicted classification of the substation audio signal corresponding to the target vector based on the target vector and a preset activation function.

[0042] Among them, the preset neural network is used to process the augmented feature sequence and output the target vector, which is a five-dimensional vector. The five-dimensional vector represents the vectors of the substation audio signal belonging to the environmental sound class, the loose accessory sound class, the local discharge sound class and the overload sound class respectively; the preset activation function is used to process the target vector to obtain the probability distribution [p1, p2, p3, p4, p5], wherein the probability distribution is used to represent the probability p1 that the substation audio signal belongs to the environmental sound class, the probability p2 that the substation audio signal belongs to the loose accessory sound class, the probability p3 that the substation audio signal belongs to the local discharge sound class, the probability p4 that the substation audio signal belongs to the normal sound class, and the probability p5 that the substation audio signal belongs to the overload sound class; in this embodiment, the preset activation function is specifically a softmax activation function, and in other embodiments, the activation function is not specifically limited; further, by determining the maximum probability in the probability distribution [p1, p2, p3, p4, p5], the maximum probability value p is obtained. max , and the maximum probability value p max The corresponding classification is used as the predicted classification of the substation audio signal; for example, assuming that the probability distribution is [0.1, 0.2, 0.3, 0.2, 0.2], the maximum probability value p max =0.3, p max That is, p3, p3 is the probability that the substation audio signal belongs to the partial discharge sound class, and the partial discharge sound class is used as the predicted classification of the substation audio signal. In this way, it can be known that the abnormal situation in the substation is specifically the partial discharge of the substation.

[0043] It should be noted that the preset neural network in this embodiment includes four convolutional layers, which are respectively recorded as the first convolutional layer, the second convolutional layer, the third convolutional layer and the fourth convolutional layer. The size of the convolution kernel corresponding to each convolutional layer is 3*3, and the step size is 1; wherein, the first convolutional layer is connected to the input layer of the neural network, and the first convolutional layer uses 64 convolutional kernels; the second convolutional layer is connected to the first convolutional layer, and the second convolutional layer uses 128 convolutional kernels; the third convolutional layer is connected to the second convolutional layer, and the third convolutional layer uses 256 convolutional kernels; the fourth convolutional layer is connected to the third convolutional layer, and the fourth convolutional layer uses 512 convolutional kernels; 512 feature maps can be obtained by processing the augmented feature sequence through the above four convolutional layers, and further, the 512 feature maps are normalized to obtain a normalized feature map group, and then, the normalized feature map group is activated by a preset RULE function to obtain a high-dimensional feature vector, which is a 512-dimensional feature vector.

[0044] It should be noted that the neural network preset in this embodiment also includes two fully connected layers, which are respectively denoted as the first fully connected layer and the second fully connected layer, wherein the first fully connected layer is used to linearly transform the high-dimensional feature vector into a medium-dimensional feature vector, which is a 128-dimensional feature vector, and the second fully connected layer is used to map the medium-dimensional feature vector into a target vector, which is a 5-dimensional feature vector.

[0045] It should also be noted that pooling layers and Dropout layers are set between adjacent convolutional layers, between adjacent convolutional layers and fully connected layers, and between adjacent fully connected layers. The pooling layer is used for average pooling operation, and the Dropout layer is used for Dropout operation. The average pooling operation is used to reduce the size of the feature map output by the convolutional layer, thereby increasing the speed at which the neural network outputs the target vector. The Dropout operation is used to set the output of some neurons in the neural network to 0 with a certain probability, thereby reducing the overfitting of the neural network and improving the generalization ability of the neural network; each average pooling layer The size of the pooling kernel is 2*2. Through the pooling kernel, an average value can be matched for each 2*2 area in the feature map. The average value is also the average value of the original values ​​of the corresponding 2*2 area. In this way, the height and width of the corresponding feature map can be halved, thereby reducing the size of the feature map. The Dropout layer is used to determine the number of filters that need to be turned off in the next convolutional layer or fully connected layer, and control the number of filters that need to be turned off through a preset probability parameter p. In this embodiment, the probability parameter corresponding to the convolutional layer is preferably 0.2, and the probability parameter corresponding to the fully connected layer is preferably 0.5.

[0046] Specifically, the augmented feature sequence is input into a preset neural network for processing, and each convolution layer in the neural network first processes the augmented feature sequence into 512 feature maps, and then the 512 feature maps are processed into a target vector through each fully connected layer in the neural network; further, the target vector is input into a preset activation function for processing to obtain a probability distribution, and then, the classification corresponding to the maximum probability value in the probability distribution is determined, and the classification is used as the prediction label of the corresponding substation audio signal.

[0047] It should be noted that, in this embodiment, the abnormal sound data set is obtained by preprocessing the collected substation audio signal, and the abnormal sound data set is labeled to obtain abnormal labeled sample data; the human voice signal in the abnormal labeled sample data is removed to obtain audio cleaning data, and the audio cleaning data is denoised to obtain noisy audio data; a mel-spectrogram is obtained based on the noisy audio data and the preset indoor simulated impulse response data, and the mel-spectrogram is processed to obtain an augmented feature sequence; the augmented feature sequence is processed based on a preset neural network to obtain a target vector, and the predicted classification of the substation audio signal corresponding to the target vector is determined based on the target vector and a preset activation function. Through the above implementation, the use of sensors in the substation abnormality detection process is avoided as much as possible, thereby facilitating the reduction of monitoring costs, and, since the collected substation audio signal is an overall audio signal related to the substation, the predicted classification of the substation audio signal is subsequently determined based on the substation audio signal, which can facilitate the further determination of the cause of the unstable working phenomenon of the substation based on the predicted classification, so as to facilitate the comprehensive monitoring of the overall operation status of the substation.

[0048] Embodiment 2

[0049] A substation abnormality detection method provided in the second embodiment of the present application optimizes the "preprocessing the collected substation audio signal to obtain an abnormal sound data set" in the first embodiment; it should be noted that for the part not described in detail in this embodiment, reference can be made to the description of other embodiments, and the method includes:

[0050] S211. Process the substation audio signal into a mono signal.

[0051] Among them, the substation audio signal is a stereo signal. Exemplarily, the stereo signal includes a left channel signal and a right channel signal. Each moment in the time series of the stereo signal has a corresponding left channel signal and a right channel signal. Taking one moment as an example, the mean of the left channel signal and the right channel signal corresponding to the moment is calculated. The mean corresponding to each moment in the time series of the stereo signal is calculated in the above manner, so as to obtain a mono signal.

[0052] Specifically, the mean of the left channel signal and the right channel signal corresponding to each moment in the time series of the substation audio signal is calculated to obtain a mono signal.

[0053] S212: Determine whether the mono signal satisfies a downsampling condition, where the downsampling condition includes that a sampling rate of the mono signal is greater than a sampling rate threshold and a sampling depth is greater than a sampling depth threshold.

[0054] Among them, the sampling rate is the number of samples collected per second from the corresponding original audio signal of the mono signal, the mono signal is encoded by a group of binary numbers, and the sampling depth is the number of bits of a group of binary numbers corresponding to the mono signal; the sampling rate threshold is the threshold set for the sampling rate. In this embodiment, the sampling rate threshold is preferably 48kHz, and the sampling depth threshold is the threshold set for the sampling depth. In this embodiment, the sampling depth threshold is preferably 16bit.

[0055] Specifically, taking a mono signal as an example, the sampling rate and sampling depth of the mono signal are obtained, and then it is determined whether the sampling rate is greater than the sampling rate threshold and the sampling depth is greater than the sampling depth threshold, so as to determine whether the mono signal meets the downsampling condition.

[0056] S213: If yes, downsample the monaural signal to obtain a downsampled signal.

[0057] Among them, if the sampling rate of a mono signal is greater than the sampling rate threshold and the sampling depth is greater than the sampling depth threshold, it means that the data volume of the mono signal is large. In order to reduce the data volume of the mono signal while ensuring that the main sound characteristics of the mono signal remain unchanged, thereby improving the efficiency of substation abnormality detection, the mono signal needs to be downsampled. The downsampled signal is the sound signal obtained after downsampling the mono signal.

[0058] Specifically, if it is determined that the sampling rate of the mono signal is greater than the sampling rate threshold and the sampling depth is greater than the sampling depth threshold, the mono signal is downsampled to obtain a downsampled signal; it should be noted that some mono signals meet the downsampling conditions and are downsampled, and the remaining mono signals do not meet the downsampling conditions and do not need to be downsampled. The downsampled signal specifically includes a downsampled mono signal and a not downsampled mono signal.

[0059] S214: Normalize the downsampled signal to obtain an abnormal sound data set.

[0060] By normalizing the downsampled signals, each downsampled signal can be cut into an audio segment of 10 seconds in length, and each abnormal sound data set is a collection of the audio segments.

[0061] Specifically, each down-sampled signal is normalized to obtain an abnormal sound data set.

[0062] S215: Label the abnormal sound data set to obtain abnormal labeled sample data.

[0063] S220: removing the human voice signal from the abnormally labeled sample data to obtain audio cleaning data, and adding noise to the audio cleaning data to obtain noisy audio data.

[0064] S230, obtaining a mel-spectrogram based on the noisy frequency data and preset indoor simulated impulse response data, and processing the mel-spectrogram to obtain an augmented feature sequence.

[0065] S240, obtaining a target vector by processing the augmented feature sequence based on a preset neural network, and determining a predicted classification of the substation audio signal corresponding to the target vector based on the target vector and a preset activation function.

[0066] Embodiment 3

[0067] A substation abnormality detection method provided in Embodiment 3 of the present application optimizes the "removing the human voice signal in the abnormal labeled sample data to obtain audio cleaning data" in Embodiment 1; it should be noted that for the parts not described in detail in this embodiment, reference may be made to the descriptions of other embodiments, and the method includes:

[0068] S310: Preprocess the collected substation audio signals to obtain an abnormal sound data set, and label the abnormal sound data set to obtain abnormal labeled sample data.

[0069] S321. Process the abnormal labeled sample data based on a preset voice labeling model to obtain voice labeling data.

[0070] Among them, the generation process of the preset human voice labeling model is as follows: first, the SED (event detection) model based on the PANNS model is used as the labeling model to be trained, and then the labeling model to be trained is trained through the AudioSet data set (the AudioSet data set focuses on the recognition of various audio events and contains a large number of audio clips. Each audio clip is accompanied by corresponding labels, which are used for classification and training of classification algorithms) to obtain the human voice labeling model; the human voice labeling model is used to label the human voice signal in the abnormal labeled sample data; the human voice labeling data is also the abnormal labeled sample data that completes the labeling of the human voice signal.

[0071] Specifically, the abnormal labeled sample data is input into a preset vocal labeling model for processing, and the vocal labeling model labels the vocal signals in the abnormal labeled sample data, thereby obtaining vocal labeling data.

[0072] S322: Process the voice tag data based on the voice probability of each frame of audio clip in the voice tag data and a preset voice probability threshold to obtain audio cleaning data.

[0073] Among them, the human voice marking data is composed of multiple frames of audio clips, each frame of the audio clip has a corresponding human voice probability, and a corresponding human voice probability threshold is set for the human voice probability. The human voice marking data can be processed according to the human voice probability corresponding to each frame of the audio clip and the corresponding human voice probability threshold, that is, deleting several frames of audio clips related to the human voice signal in the human voice marking data, and the audio cleaning data is the human voice marking data after deleting several frames of audio clips.

[0074] Specifically, the human voice probability corresponding to each frame of audio clips in the human voice marking data is first obtained, and then several frames of audio clips related to human voice signals in the human voice marking data are deleted according to the human voice probability and a preset human voice probability threshold, thereby obtaining audio cleaning data.

[0075] It should be noted that by deleting several frames of audio segments related to the human voice signal in the human voice mark data, it is possible to prevent the human voice signal in the human voice mark data from having an adverse effect on the abnormality detection of the substation.

[0076] S323: Noise the audio cleaning data to obtain noisy audio data.

[0077] S330, obtaining a mel-spectrogram based on the noisy frequency data and preset indoor simulated impulse response data, and processing the mel-spectrogram to obtain an augmented feature sequence.

[0078] S340, processing the augmented feature sequence based on a preset neural network to obtain a target vector, and determining a predicted classification of the substation audio signal corresponding to the target vector based on the target vector and a preset activation function.

[0079] Embodiment 4

[0080] A substation abnormality detection method provided in a fourth embodiment of the present application optimizes the "processing the human voice mark data to obtain audio cleaning data based on the human voice probability of each frame audio segment in the human voice mark data and a preset human voice probability threshold" in the third embodiment; it should be noted that for the parts not described in detail in this embodiment, reference can be made to the descriptions of other embodiments, and the method includes:

[0081] S410: Preprocess the collected substation audio signals to obtain an abnormal sound data set, and label the abnormal sound data set to obtain abnormal labeled sample data.

[0082] S421. Process the abnormal labeled sample data based on a preset voice labeling model to obtain voice labeling data.

[0083] S422A: Determine whether there are at least a preset number of consecutive frames of audio segments in the human voice tag data whose human voice probabilities are greater than a preset human voice probability threshold.

[0084] The preset number of frames in this embodiment is preferably 3 frames, and the preset human voice probability threshold is preferably 0.3 in this embodiment.

[0085] Specifically, it is determined whether there are at least three consecutive audio segments in the human voice tag data and the probability of human voice being present is greater than 0.3.

[0086] S422B: If yes, delete the continuous preset number of frames of audio segments from the vocal tag data to obtain audio cleaning data.

[0087] Among them, if it is determined that there are at least 3 consecutive frames of audio clips in the voice mark data and the probability of human voice is greater than 0.3, it means that the human voice signal of the voice mark data is too obvious, which will have an adverse effect on the substation abnormality detection. Therefore, it is necessary to delete at least 3 consecutive frames of audio clips from the voice mark data, and recombine the audio clips of the remaining frames in the voice mark data to obtain audio cleaning data.

[0088] Specifically, if it is determined that there are at least a preset number of consecutive frames of audio clips in the voice mark data whose voice probabilities are greater than a preset voice probability threshold, the at least a preset number of consecutive frames of audio clips are deleted from the voice mark data, and the audio clips of the remaining frames in the voice mark data are recombined to obtain audio cleaning data.

[0089] S423: Noise the audio cleaning data to obtain noisy audio data.

[0090] S430, obtaining a mel-spectrogram based on the noisy frequency data and preset indoor simulated impulse response data, and processing the mel-spectrogram to obtain an augmented feature sequence.

[0091] S440, obtaining a target vector by processing the augmented feature sequence based on a preset neural network, and determining a predicted classification of the substation audio signal corresponding to the target vector based on the target vector and a preset activation function.

[0092] Embodiment 5

[0093] A substation abnormality detection method provided in the fifth embodiment of the present application optimizes the "obtaining a mel-spectrogram based on the noise-added frequency data and the preset indoor simulated impulse response data" in the first embodiment; it should be noted that for the parts not described in detail in this embodiment, reference may be made to the descriptions of other embodiments, and the method includes:

[0094] S510: Preprocess the collected substation audio signals to obtain an abnormal sound data set, and label the abnormal sound data set to obtain abnormal labeled sample data.

[0095] S520: remove the human voice signal in the abnormally marked sample data to obtain audio cleaning data, and add noise to the audio cleaning data to obtain noisy audio data.

[0096] S531 . Combining the noisy audio data with the indoor simulated impulse response data to obtain audio combined data.

[0097] The audio combined data is data obtained by combining the noise-added audio data and the indoor simulated impulse response data.

[0098] S532: Segment the audio merged data to obtain audio segmentation data.

[0099] Among them, this embodiment uses a Hanning window to segment the audio merged data. The window length of the Hanning window is 512 sampling points and the step length is 160 sampling points. By segmenting the audio merged data through the Hanning window, audio segmentation data corresponding to multiple time periods can be obtained.

[0100] Specifically, the audio merged data is segmented using a preset Hanning window, thereby obtaining audio segmentation data corresponding to multiple time periods one by one.

[0101] S533: Perform Fourier transform on the audio segmentation data to obtain a time period spectrum diagram.

[0102] The Fourier transform in this embodiment is specifically a short-time Fourier transform (STFT), which is used to convert the audio segmentation data into a corresponding spectrum diagram, which is recorded as a time period spectrum diagram.

[0103] Specifically, short-time Fourier transform is used to transform the audio segmentation data, thereby obtaining a time period spectrum diagram.

[0104] S534, filtering the time period spectrum to obtain a mel-spectrogram.

[0105] Among them, in this embodiment, a 128-order Mel filter is preset, and the 128-order Mel filter is used to filter the time period spectrum diagram to obtain a Mel spectrogram. In this embodiment, the Mel spectrogram is specifically a logarithmic Mel spectrogram.

[0106] Specifically, the time period spectrum is filtered by a preset 128-order Mel filter to obtain a Mel spectrum.

[0107] S535, processing the mel-spectrogram to obtain an augmented feature sequence.

[0108] S540, obtaining a target vector by processing the augmented feature sequence based on a preset neural network, and determining a predicted classification of the substation audio signal corresponding to the target vector based on the target vector and a preset activation function.

[0109] Embodiment 6

[0110] A substation abnormality detection method provided in Example 6 of the present application optimizes the "processing the Mel-spectrogram to obtain an augmented feature sequence" in Example 1; it should be noted that for the parts not described in detail in this embodiment, reference can be made to the descriptions of other embodiments, and the method includes:

[0111] S610: Preprocess the collected substation audio signals to obtain an abnormal sound data set, and label the abnormal sound data set to obtain abnormal labeled sample data.

[0112] S620: remove the human voice signal in the abnormally marked sample data to obtain audio cleaning data, and add noise to the audio cleaning data to obtain noisy audio data.

[0113] S631. Obtain a mel-spectrogram based on the noisy frequency data and preset indoor simulated impulse response data.

[0114] S632: Based on the preset time domain mask parameters and frequency domain mask parameters, a time domain mask and a frequency domain mask are added to the mel-spectrogram to obtain an augmented spectrogram.

[0115] Among them, the preset time domain mask parameters include: the maximum continuous mask length T in the time domain; the real mask length t in the time domain, where t obeys the uniform distribution of [0, T]; the starting point t0 of the time domain mask, where the range of t0 is (0, τ-t0), and τ is the time domain length of the Mel-spectrogram; the preset frequency domain mask parameters include: the maximum continuous mask length F in the frequency domain; the real mask length f in the time domain, where f obeys the uniform distribution of [0, F]; the starting point f0 of the time domain mask, where the range of f0 is (0, v-f0), and v is the frequency domain length of the Mel-spectrogram; the time domain mask is a mask generated based on the time domain mask parameters, and the frequency domain mask is a mask generated based on the frequency domain mask parameters. The augmented spectrogram is the spectrogram obtained after setting the time domain mask and the frequency domain mask on the Mel-spectrogram, as shown in FIG. Figure 2 As shown, the horizontal black mask is a time domain mask, and the vertical black mask is a frequency domain mask.

[0116] Specifically, a time domain mask and a frequency domain mask are generated respectively based on preset time domain mask parameters and frequency domain mask parameters. Furthermore, the time domain mask and the frequency domain mask are set on the mel-spectrogram to obtain an augmented spectrogram.

[0117] S633, normalizing the augmented spectrum to obtain the augmented feature sequence.

[0118] Among them, normalization is to process the augmented spectrogram into an augmented feature sequence, and the augmented feature sequence is a feature sequence used for subsequent input into the neural network.

[0119] Specifically, the augmented spectrogram is normalized to obtain an augmented feature sequence.

[0120] S640: Process the augmented feature sequence based on a preset neural network to obtain a target vector, and determine a predicted classification of the substation audio signal corresponding to the target vector based on the target vector and a preset activation function.

[0121] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0122] Embodiment 7

[0123] Based on the same inventive concept, this embodiment also provides a substation anomaly detection device for implementing the substation anomaly detection method involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more substation anomaly detection device embodiments provided below can refer to the above limitations on the substation anomaly detection method, and will not be repeated here.

[0124] In this embodiment, Figure 3 As shown, a substation abnormality detection device is provided, comprising:

[0125] The processing and labeling module is used to pre-process the collected substation audio signals to obtain an abnormal sound data set, and label the abnormal sound data set to obtain abnormal labeled sample data.

[0126] The cleaning and denoising module is used to remove the human voice signal in the abnormally labeled sample data to obtain audio cleaning data, and to denoise the audio cleaning data to obtain noisy audio data.

[0127] The sequence generation module is used to obtain a mel-spectrogram based on the noisy frequency data and preset indoor simulated impulse response data, and process the mel-spectrogram to obtain an augmented feature sequence.

[0128] A prediction and classification module is used to process the augmented feature sequence based on a preset neural network to obtain a target vector, and determine the prediction classification of the substation audio signal corresponding to the target vector based on the target vector and a preset activation function.

[0129] Each module in the above-mentioned substation abnormality detection device can be implemented in whole or in part by software, hardware and a combination thereof. Each of the above-mentioned modules can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute the operations corresponding to each of the above modules.

[0130] It should be noted that, in this embodiment, the abnormal sound data set is obtained by preprocessing the collected substation audio signal, and the abnormal sound data set is labeled to obtain abnormal labeled sample data; the human voice signal in the abnormal labeled sample data is removed to obtain audio cleaning data, and the audio cleaning data is denoised to obtain noisy audio data; a mel-spectrogram is obtained based on the noisy audio data and the preset indoor simulated impulse response data, and the mel-spectrogram is processed to obtain an augmented feature sequence; the augmented feature sequence is processed based on a preset neural network to obtain a target vector, and the predicted classification of the substation audio signal corresponding to the target vector is determined based on the target vector and a preset activation function. Through the above implementation, the use of sensors in the substation abnormality detection process is avoided as much as possible, thereby facilitating the reduction of monitoring costs, and, since the collected substation audio signal is an overall audio signal related to the substation, the predicted classification of the substation audio signal is subsequently determined based on the substation audio signal, which can facilitate the further determination of the cause of the unstable working phenomenon of the substation based on the predicted classification, so as to facilitate the comprehensive monitoring of the overall operation status of the substation.

[0131] Embodiment 8

[0132] In this embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Figure 4As shown. The computer device includes a processor, a memory and a network interface connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a substation abnormality detection method is implemented.

[0133] Those skilled in the art will understand that Figure 4 The structure shown in the figure is only a block diagram of a part of the structure related to the scheme of the present disclosure, and does not constitute a limitation on the computer device to which the scheme of the present disclosure is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0134] Embodiment 9

[0135] In this embodiment, a computer readable storage medium is provided. Figure 5 As shown, a computer program is stored thereon, and when the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0136] Embodiment 10

[0137] In this embodiment, a computer program product is provided, including a computer program. When the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0138] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this disclosure are all information and data authorized by the user or fully authorized by all parties.

[0139] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided by the present disclosure can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided by the present disclosure may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited thereto. The processor involved in each embodiment provided by the present disclosure may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited thereto.

[0140] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0141] The above-described embodiments only express several implementation methods of the present disclosure, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present invention. It should be pointed out that, for a person of ordinary skill in the art, several modifications and improvements can be made without departing from the concept of the present disclosure, and these all belong to the protection scope of the present disclosure. Therefore, the protection scope of the present disclosure shall be subject to the attached claims.

Claims

1. A substation abnormality detection method, characterized in that: include: Preprocessing the collected substation audio signals to obtain an abnormal sound data set, and labeling the abnormal sound data set to obtain abnormal labeled sample data; Removing the human voice signal from the abnormally labeled sample data to obtain audio cleaning data, and adding noise to the audio cleaning data to obtain noisy audio data; Obtaining a mel-spectrogram based on the noise-added frequency data and preset indoor simulated impulse response data, and processing the mel-spectrogram to obtain an augmented feature sequence; The augmented feature sequence is processed based on a preset neural network to obtain a target vector, and a predicted classification of the substation audio signal corresponding to the target vector is determined based on the target vector and a preset activation function.

2. The method according to claim 1, characterized in that The preprocessing of the collected substation audio signals to obtain an abnormal sound data set includes: Processing the substation audio signal into a mono signal; Determine whether the mono signal satisfies a downsampling condition, wherein the downsampling condition includes that a sampling rate of the mono signal is greater than a sampling rate threshold and a sampling depth is greater than a sampling depth threshold; If yes, down-sampling the mono signal to obtain a down-sampled signal; The downsampled signal is normalized to obtain an abnormal sound data set.

3. The method according to claim 1, characterized in that The step of removing the human voice signal from the abnormally labeled sample data to obtain audio cleaning data includes: Processing the abnormal labeled sample data based on a preset voice labeling model to obtain voice labeling data; The human voice marking data is processed based on the human voice probability of each frame audio segment in the human voice marking data and a preset human voice probability threshold to obtain audio cleaning data.

4. The method according to claim 3, characterized in that The step of processing the vocal marker data based on the vocal probability of each audio segment in the vocal marker data and a preset vocal probability threshold to obtain the audio cleaning data includes: Determine whether there are at least a preset number of consecutive frames of audio segments in the human voice tag data, and the human voice probability is greater than a preset human voice probability threshold; If so, the audio segments of the preset number of consecutive frames are deleted from the vocal mark data to obtain audio cleaning data.

5. The method according to claim 1, characterized in that The step of obtaining a mel-spectrogram based on the noise-added frequency data and preset indoor simulated impulse response data includes: Combining the noise-added frequency data with the indoor simulated impulse response data to obtain audio combined data; Segmenting the audio merged data to obtain audio segmentation data; Performing Fourier transform on the audio segmentation data to obtain a time period spectrum diagram; The time period spectrum graph is filtered to obtain a mel-spectrogram.

6. The method according to claim 1, characterized in that The step of processing the mel-spectrogram to obtain an augmented feature sequence includes: Based on the preset time domain mask parameters and frequency domain mask parameters, a time domain mask and a frequency domain mask are added to the mel spectrogram to obtain an augmented spectrogram; The augmented spectrum is normalized to obtain the augmented feature sequence.

7. A substation abnormality detection device, characterized in that: The device comprises: A processing and labeling module is used to pre-process the collected substation audio signals to obtain an abnormal sound data set, and label the abnormal sound data set to obtain abnormal labeled sample data; A cleaning and denoising module, used to remove the human voice signal in the abnormally labeled sample data to obtain audio cleaning data, and to denoise the audio cleaning data to obtain noisy audio data; A sequence generation module, used to obtain a mel-spectrogram based on the noisy frequency data and preset indoor simulated impulse response data, and process the mel-spectrogram to obtain an augmented feature sequence; A prediction and classification module is used to process the augmented feature sequence based on a preset neural network to obtain a target vector, and determine the prediction classification of the substation audio signal corresponding to the target vector based on the target vector and a preset activation function.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images