Power equipment intelligent fault diagnosis system and method based on voiceprint image, and background processing system

Abstract

The invention relates to the technical field of intelligent diagnosis of power equipment, in particular to an intelligent fault diagnosis system and method for power equipment based on voiceprint images and a background processing system, and the intelligent fault diagnosis system comprises the background processing system and a plurality of pieces of sound signal monitoring equipment. Analyzing the sound basic attribute information received from the plurality of sound signal monitoring devices through a background processing system, determining a faulty power device, and obtaining a voiceprint signal of the faulty power device in a specified time period; performing pre-emphasis processing, framing processing, windowing processing and continuous wavelet transform on the voiceprint signal, and drawing a time-wavelet coefficient graph as a voiceprint image; and performing HOG feature extraction on the voiceprint image, and inputting the voiceprint image into the trained fault multi-classification model to obtain the fault type of the fault power equipment. According to the invention, the early warning of the fault can be realized through non-contact detection, the fault type is intelligently and accurately classified, and the fault downtime is reduced.

Description

Technical Field

[0001] The present disclosure relates to the technical field of intelligent diagnosis of electric power equipment, and in particular to an intelligent fault diagnosis system and method for electric power equipment based on voiceprint images, and a background processing system. Background Art

[0002] As all kinds of power equipment in the power system develop towards large-scale, integrated, highly automated and intelligent, this trend brings rapid economic benefits, but also increases the risk of huge losses caused by sudden equipment failure. In the face of this situation, it is particularly important to monitor the condition of power equipment. Timely, accurate and comprehensive condition monitoring can not only effectively prevent the occurrence of equipment failures and ensure the stable operation of the power system; more importantly, the application of condition monitoring technology also has a high investment return ratio. On the one hand, by reducing unnecessary downtime and maintenance costs, enterprises can save a lot of operating expenses; on the other hand, it can also help enterprises discover equipment hidden dangers in advance and avoid major accidents, thereby ensuring the safe production and sustainable development of enterprises. Summary of the invention

[0003] In order to solve the problems in the related art, the embodiments of the present disclosure provide an intelligent fault diagnosis system and method for electric power equipment based on voiceprint images, and a background processing system.

[0004] In a first aspect, an embodiment of the present disclosure provides an intelligent fault diagnosis system for electric power equipment based on voiceprint images, characterized in that it is applied to an electric power system, the electric power system includes a plurality of electric power equipment, and the intelligent fault diagnosis system includes: a background processing system and a plurality of sound signal monitoring devices, the sound signal monitoring devices corresponding to the electric power equipment;

[0005] The sound signal monitoring device is used to monitor the basic attribute information of the sound of the corresponding power equipment and transmit the basic attribute information of the sound to the background processing system;

[0006] The background processing system is used to analyze the basic attribute information of the sounds received from the multiple sound signal monitoring devices respectively, determine the faulty power equipment, and obtain the voiceprint signal of the faulty power equipment within a specified time period; pre-emphasize the voiceprint signal to obtain a pre-emphasized voiceprint signal; perform frame processing and windowing processing on the pre-emphasized voiceprint signal to obtain a specified number of voiceprint sample signals; perform continuous wavelet transform on the specified number of voiceprint sample signals, extract continuous wavelet transform coefficients, and thus draw a time-wavelet coefficient graph, and use the time-wavelet coefficient graph as the voiceprint image of the voiceprint signal; perform HOG feature extraction on the voiceprint image to obtain a HOG feature vector group; input the HOG feature vector group into a trained fault multi-classification model to obtain the fault type of the faulty power equipment.

[0007] According to an embodiment of the present disclosure, the basic sound attribute information includes any one or more of the following: sound amplitude, sound frequency, sound phase difference;

[0008] The analyzing the basic sound attribute information received from the multiple sound signal monitoring devices respectively to determine the faulty power equipment includes: setting a preset threshold for each of the multiple power equipment, wherein the preset threshold includes any one or more of the following: a preset sound amplitude threshold, a preset sound frequency threshold, and a preset sound phase difference threshold; determining that the power equipment is a faulty power equipment when the basic sound attribute information meets any one or more of the following conditions: the sound amplitude exceeds the preset sound amplitude threshold, the sound frequency exceeds the preset sound frequency threshold, and the sound phase difference exceeds the preset sound phase difference threshold.

[0009] According to an embodiment of the present disclosure, the pre-emphasis processing of the voiceprint signal includes:

[0010] The voiceprint signal is pre-emphasized using a high-pass filter.

[0011] According to an embodiment of the present disclosure, the transfer function of the high-pass filter is shown in the following formula:

[0012] H(z)=1-μz -1 ;

[0013] Wherein, H(z) is the transfer function, μ is the pre-emphasis coefficient, and z is a complex variable.

[0014] According to an embodiment of the present disclosure, the performing frame processing and window processing on the pre-emphasized voiceprint signal includes:

[0015] The frame segmentation process is implemented by using a set frame length and a set frame shift, wherein the set frame length is smaller than the time length of the specified time period, and the set frame shift is smaller than the set frame length.

[0016] According to an embodiment of the present disclosure, the performing frame processing and window processing on the pre-emphasized voiceprint signal includes:

[0017] The windowing process is implemented by using a Hamming window function, wherein the window size of the Hamming window function is the set frame length.

[0018] According to an embodiment of the present disclosure, the Hamming window function is shown in the following formula:

[0019]

[0020] Wherein, ω(n) is the Hamming window function, N is the set frame length, and n is the sampling point.

[0021] According to an embodiment of the present disclosure, performing continuous wavelet transform on the specified number of voiceprint sample signals includes:

[0022] A wavelet basis function and a decomposition layer number are selected, and a continuous wavelet transform is performed on each voiceprint sample signal in the specified number of voiceprint sample signals, so as to extract the continuous wavelet transform coefficients.

[0023] According to an embodiment of the present disclosure, the wavelet basis function is shown in the following formula:

[0024]

[0025] Among them, ψ a，τ (t) is the wavelet basis function, a is the wavelet scaling factor, τ is the wavelet translation factor, and t is the time variable.

[0026] According to an embodiment of the present disclosure, the continuous wavelet transform coefficients are extracted using the following formula:

[0027]

[0028] Among them, WT f (a, τ) is the continuous wavelet transform coefficient, f(t) is the specified number of voiceprint sample signals, is the wavelet basis function ψ a，τ is the conjugate complex number of (t), · is the inner product operation, and dt is the differentiation operation.

[0029] According to an embodiment of the present disclosure, the step of extracting HOG features from the voiceprint image includes:

[0030] Dividing the voiceprint image into a plurality of cell units, and calculating the gradient value and gradient direction of each pixel point in the cell unit;

[0031] Calculating a first directional gradient histogram of the cell unit according to the gradient value and gradient direction of each pixel point in the cell unit, wherein the first directional gradient histogram has a first set vector dimension;

[0032] Using a sliding window to extract image blocks on a voiceprint image divided into a plurality of cell units, to obtain a plurality of image blocks, wherein the sliding window has a set window size and a set sliding step, and the image block has a set number of cell units;

[0033] Calculating a second directional gradient histogram of the image block, wherein the second directional gradient histogram has a second set vector dimension, and the second set vector dimension is the sum of the set number of first set vector dimensions;

[0034] Normalizing the plurality of second directional gradient histograms of the plurality of image blocks to obtain the HOG feature vector group.

[0035] According to an embodiment of the present disclosure, the step of calculating the gradient value and gradient direction of each pixel in the cell unit includes:

[0036] For the cell unit, the gradient value and gradient direction of each pixel point in the cell unit are calculated according to the gradient value in the horizontal coordinate direction and the gradient value in the vertical coordinate direction of each pixel point.

[0037] According to an embodiment of the present disclosure, the gradient value and gradient direction of each pixel point in the cell unit are calculated according to the gradient value in the horizontal coordinate direction and the gradient value in the vertical coordinate direction of each pixel point, including using the following formula for calculation:

[0038]

[0039] Among them, G x (x, y) is the gradient value of each pixel in the horizontal direction, G y (x, y) is the gradient value of each pixel in the ordinate direction, G(x, y) is the gradient value of each pixel, and α(x, y) is the gradient direction of each pixel.

[0040] According to an embodiment of the present disclosure, the gradient value in the horizontal coordinate direction and the gradient value in the vertical coordinate direction of each pixel point are calculated according to the grayscale value of each pixel point in the cell unit.

[0041] According to an embodiment of the present disclosure, the step of calculating the gradient value in the horizontal coordinate direction and the gradient value in the vertical coordinate direction of each pixel point according to the grayscale value of each pixel point in the cell unit includes calculating using the following formula:

[0042]

[0043] Among them, H(x+1,y) is the gradient value of the pixel point (x+1,y), H(x-1,y) is the gradient value of the pixel point (x-1,y), H(x,y+1) is the gradient value of the pixel point (x,y+1), H(x,y-1) is the gradient value of the pixel point (x,y-1), x is the horizontal coordinate of each pixel point, and y is the vertical coordinate of each pixel point.

[0044] According to an embodiment of the present disclosure, the vector dimension of the HOG feature vector group is calculated by the following formula:

[0045]

[0046] Wherein, P is the vector dimension of the HOG feature vector group, l p is the number of pixels in the horizontal coordinate direction of the image block, h p is the number of pixels in the ordinate direction of the image block, l c is the number of pixels in the horizontal coordinate direction of the cell unit, and M is the dimension of the second set vector.

[0047] According to an embodiment of the present disclosure, the trained fault multi-classification model is obtained by training the SVM model.

[0048] In a second aspect, an embodiment of the present disclosure provides an intelligent fault diagnosis method for electric power equipment based on voiceprint images, comprising:

[0049] Obtain the voiceprint signal of the faulty power equipment within a specified time period;

[0050] Performing pre-emphasis processing on the voiceprint signal to obtain a pre-emphasized voiceprint signal;

[0051] Performing frame processing and window processing on the pre-emphasized voiceprint signal to obtain a specified number of voiceprint sample signals;

[0052] Performing continuous wavelet transform on the specified number of voiceprint sample signals, extracting continuous wavelet transform coefficients, thereby drawing a time-wavelet coefficient graph, and using the time-wavelet coefficient graph as a voiceprint image of the voiceprint signal;

[0053] Performing HOG feature extraction on the voiceprint image to obtain a HOG feature vector group;

[0054] The HOG feature vector group is input into a trained fault multi-classification model to obtain the fault type of the faulty power equipment.

[0055] According to an embodiment of the present disclosure, before acquiring the voiceprint signal of the faulty power equipment within a specified time period, the method further includes:

[0056] Acquire basic attribute information of a plurality of sounds corresponding to a plurality of electric power equipment in the electric power system;

[0057] The basic attribute information of the multiple sounds is analyzed to determine the faulty electrical equipment.

[0058] In a third aspect, an embodiment of the present disclosure provides a background processing system, including:

[0059] An acquisition module is configured to acquire a voiceprint signal of a faulty power device within a specified time period;

[0060] A first preprocessing module is configured to perform pre-emphasis processing on the voiceprint signal to obtain a pre-emphasized voiceprint signal;

[0061] A second preprocessing module is configured to perform frame processing and window processing on the pre-emphasized voiceprint signal to obtain a specified number of voiceprint sample signals;

[0062] a third preprocessing module, configured to perform continuous wavelet transform on the specified number of voiceprint sample signals, extract continuous wavelet transform coefficients, thereby drawing a time-wavelet coefficient graph, and using the time-wavelet coefficient graph as a voiceprint image of the voiceprint signal;

[0063] A feature extraction module is configured to perform HOG feature extraction on the voiceprint image to obtain a HOG feature vector group;

[0064] The fault classification module is configured to input the HOG feature vector group into a trained fault multi-classification model to obtain the fault type of the faulty power equipment.

[0065] In a fourth aspect, an embodiment of the present disclosure provides a computer-readable storage medium on which computer instructions are stored. When the computer instructions are executed by a processor, the method described in any one of the second aspects is implemented.

[0066] In a fifth aspect, a computer program product is provided in an embodiment of the present disclosure, which includes computer instructions, and when the computer instructions are executed by a processor, the method described in any one of the second aspects is implemented.

[0067] The technical solution provided in accordance with the embodiment of the present disclosure includes: a background processing system and multiple sound signal monitoring devices; the sound signal monitoring devices monitor the basic sound attribute information of the corresponding power equipment and transmit the basic sound attribute information to the background processing system; the background processing system analyzes the basic sound attribute information received from the multiple sound signal monitoring devices respectively, determines the faulty power equipment, and obtains the voiceprint signal of the faulty power equipment within a specified time period; then the voiceprint signal is subjected to pre-emphasis processing, frame processing, windowing processing, and continuous wavelet transformation, and continuous wavelet transformation coefficients are extracted, thereby drawing a time-wavelet coefficient graph as a voiceprint image of the voiceprint signal; after HOG feature extraction of the voiceprint image, the obtained HOG feature vector group is input into a trained fault multi-classification model to obtain the fault type of the faulty power equipment.

[0068] The present invention realizes rapid and comprehensive detection and early fault identification of power equipment by performing non-contact detection settings for power equipment, and on this basis, analyzes the voiceprint image obtained by converting the voiceprint signal from one-dimensional information to a two-dimensional image, thereby improving the recognition accuracy, being able to more accurately capture the subtle differences in sound signals, and combining with artificial intelligence algorithms for intelligent analysis and diagnosis, thereby achieving accurate classification of fault types, thereby enabling targeted fault processing according to the fault type, avoiding changing the equipment operating status when not necessary, greatly reducing fault downtime, and improving the operating efficiency of the power system. It has a high degree of intelligence and strong accuracy and reliability in fault diagnosis.

[0069] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Other features, objectives and advantages of the present disclosure will become more apparent through the following detailed description of non-limiting embodiments in conjunction with the accompanying drawings. In the accompanying drawings:

[0071] Figure 1 A schematic diagram of the structure of an intelligent fault diagnosis system for electric power equipment based on voiceprint images according to an embodiment of the present disclosure is shown;

[0072] Figure 2 A schematic diagram showing frame processing and window processing of a pre-emphasized voiceprint signal in an embodiment of the present disclosure is shown;

[0073] Figure 3 A schematic diagram showing a method of calculating a first directional gradient histogram of a cell unit in an embodiment of the present disclosure is shown;

[0074] Figure 4 A schematic diagram showing the division of cell units and image blocks when performing HOG feature extraction on a voiceprint image in an embodiment of the present disclosure is shown;

[0075] Figure 5 A flow chart showing a method for intelligent fault diagnosis of electric power equipment based on voiceprint images according to an embodiment of the present disclosure is shown;

[0076] Figure 6 A schematic diagram showing the structure of a background processing system according to an embodiment of the present disclosure is shown; DETAILED DESCRIPTION

[0077] Hereinafter, exemplary embodiments of the present disclosure will be described in detail with reference to the accompanying drawings so that those skilled in the art can easily implement them. In addition, for the sake of clarity, parts not related to the description of the exemplary embodiments are omitted in the accompanying drawings.

[0078] In the present disclosure, it should be understood that terms such as "include" or "have" are intended to indicate the presence of features, numbers, steps, behaviors, components, parts, or a combination thereof disclosed in the present specification, and are not intended to exclude the possibility that one or more other features, numbers, steps, behaviors, components, parts, or a combination thereof exist or are added.

[0079] It should also be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present disclosure may be combined with each other. The present disclosure will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0080] As mentioned above, as various types of power equipment in the power system are developing towards large-scale, integrated, highly automated and intelligent, the risk of huge losses caused by sudden equipment failures has also increased. In the face of this situation, timely, accurate and comprehensive status monitoring of power equipment can not only effectively prevent equipment failures and ensure the stable operation of the power system, but also help enterprises discover equipment hidden dangers in advance and avoid major accidents.

[0081] After investigation, the inventor found that in the power system, various power equipment will make different sounds in normal operation or abnormal conditions, such as transformers, reactors, fans (such as induced draft fans, forced draft fans, etc.), capacitors, voltage transformers, etc. These sounds can not only help operation and maintenance personnel determine the operating status of the equipment, but also serve as an important basis for equipment failure warning. However, the current method of judging the operating status of equipment based on the sound of the equipment is relatively limited, and mainly relies on the basic properties of the sound such as loudness and frequency to make simple fault judgments. This method is difficult to achieve accurate distinction and classification of equipment failure types.

[0082] The present invention provides an intelligent fault diagnosis system for electric equipment based on voiceprint images, comprising: a background processing system and multiple sound signal monitoring devices; the sound signal monitoring devices monitor the basic sound attribute information of the corresponding electric equipment and transmit the basic sound attribute information to the background processing system; the background processing system analyzes the basic sound attribute information received from the multiple sound signal monitoring devices respectively, determines the faulty electric equipment, and obtains the voiceprint signal of the faulty electric equipment within a specified time period; then pre-emphasizes, frames, windows, and continuously transforms the voiceprint signal, extracts the continuous wavelet transform coefficients, and draws a time-wavelet coefficient graph as the voiceprint image of the voiceprint signal; after performing HOG feature extraction on the voiceprint image, the obtained HOG feature vector group is input into a trained fault multi-classification model to obtain the fault type of the faulty electric equipment.

[0083] The present invention can realize rapid and comprehensive detection and early fault identification by performing non-contact detection settings on power equipment in the power system, and can reflect potential faults inside the equipment in real time, which is helpful for operation and maintenance personnel to discover and handle faults in advance. At the same time, by analyzing the images obtained by transforming the voiceprint signal from one-dimensional information to two-dimensional images, the recognition accuracy is improved, and the subtle differences in sound signals can be captured more accurately. In combination with the trained fault multi-classification model, accurate classification of fault types is achieved, so that targeted fault handling can be carried out according to the fault type, avoiding changing the equipment operating status when it is not necessary, greatly reducing fault downtime, and improving the operating efficiency of the power system. It has a high degree of intelligence and strong accuracy and reliability in fault diagnosis.

[0084] Figure 1 FIG. 1 is a schematic diagram showing a structure of an intelligent fault diagnosis system for electric power equipment based on voiceprint images according to an embodiment of the present disclosure. Figure 1 As shown, the intelligent fault diagnosis system is applied to an electric power system, the electric power system includes a plurality of electric power equipment, and the intelligent fault diagnosis system includes: a background processing system and a plurality of sound signal monitoring devices, the sound signal monitoring devices corresponding to the electric power equipment.

[0085] According to an embodiment of the present disclosure, the sound signal monitoring device is used to monitor the basic attribute information of the sound of the corresponding power equipment and transmit the basic attribute information of the sound to the background processing system. Wherein, multiple sound signal monitoring devices are communicatively connected to the background processing system (for example, wirelessly connected and / or wiredly connected).

[0086] In the present disclosure, the sound signal monitoring equipment and the power equipment are matched, and the multiple sound signal monitoring equipment and the multiple power equipment are non-contactly set. For example, the sound signal monitoring equipment can be set at several fixed points in the power system, so as to perform non-contact directional monitoring at the fixed points, or the sound signal monitoring equipment can be set next to some key power equipment (such as generators, transformers, induced draft fans, etc.), so as to reduce the impact of the sound generated by other power equipment in the power system during operation on the target power equipment we monitor, so as to achieve a more accurate analysis of the operating status of the target power equipment.

[0087] In the present disclosure, the basic sound attribute information includes any one or more of the following: sound amplitude, sound frequency, sound phase difference, sound timbre, sound pressure, etc. The basic sound attribute information is sound information that can be directly obtained by a sound signal monitoring device without further processing.

[0088] According to an embodiment of the present disclosure, the background processing system is used to analyze the basic attribute information of the sound received from the multiple sound signal monitoring devices respectively, and determine the faulty power equipment.

[0089] Specifically, for each of the multiple power devices, a preset threshold is set, wherein the preset threshold includes any one or more of the following: a preset sound amplitude threshold, a preset sound frequency threshold, and a preset sound phase difference threshold; when the basic sound attribute information meets any one or more of the following conditions, the power device is determined to be a faulty power device: the sound amplitude exceeds the preset sound amplitude threshold, the sound frequency exceeds the preset sound frequency threshold, and the sound phase difference exceeds the preset sound phase difference threshold.

[0090] That is, for the basic sound attribute information of the corresponding power equipment obtained from each sound signal monitoring device, the difference between the basic sound attribute information and the preset threshold is analyzed, and when it exceeds the preset threshold, the power equipment is determined to be a faulty power equipment. For example, when the sound amplitude obtained for a certain power equipment exceeds the preset sound amplitude threshold, the power equipment is a faulty power equipment.

[0091] Among them, the preset threshold can be customized for each electrical equipment. For example, when the electrical equipment is a transformer, since the transformer is in normal working condition, the current passes through the iron core to generate alternating magnetic flux, which will emit uniform electromagnetic sound. The preset threshold can be set as a sound frequency threshold, so as to judge whether the sound frequency of the transformer is abnormal based on the sound frequency threshold.

[0092] Before further sound signal analysis is performed, the present invention first locks in the electrical equipment that may have faults based on the basic attribute information of the sound, thereby avoiding the computational pressure caused by unnecessary in-depth analysis of the sound signals generated during normal operation of the electrical equipment.

[0093] According to an embodiment of the present disclosure, the background processing system is further used to obtain a voiceprint signal of the faulty power equipment within a specified time period; and perform pre-emphasis processing on the voiceprint signal to obtain a pre-emphasized voiceprint signal.

[0094] For the faulty power equipment, the basic attribute information of the sound is further processed by the background processing system to obtain the voiceprint signal within the specified time period, wherein the specified time period can be any time period after the faulty power equipment is determined, for example, 10S, 20S, 60S, etc. The longer the time length of the specified time period is, the more features of the collected voiceprint signal are, so that the subsequent analysis results are more accurate.

[0095] Further, the voiceprint signal is pre-emphasized, including: using a high-pass filter to pre-emphasize the voiceprint signal. The transfer function of the high-pass filter is shown in the following formula:

[0096] H(z)=1-μz -1 ; Wherein, H(z) is the transfer function, μ is the pre-emphasis coefficient, and z is a complex variable.

[0097] In the transfer function H(z), the complex variable z is the frequency domain representation of the signal in the time domain. Through this representation, the voiceprint signal can be converted from the time domain to the frequency domain to analyze the characteristics of the voiceprint signal in the frequency domain.

[0098] By pre-emphasis processing the voiceprint signal, the intensity of the high-frequency components in the voiceprint signal is increased, making it easier to distinguish useful voiceprint features from background noise, thereby improving the accuracy of subsequent analysis.

[0099] According to an embodiment of the present disclosure, the background processing system is further used to perform frame processing and window processing on the pre-emphasized voiceprint signal to obtain a specified number of voiceprint sample signals.

[0100] Furthermore, the performing frame processing and window processing on the pre-emphasized voiceprint signal includes:

[0101] The frame processing is implemented by setting a frame length and a frame shift, wherein the set frame length is less than the time length of the specified time period, and the set frame shift is less than the set frame length. Assuming that the time length of the specified time period is 10s, the set frame length can be set to 25ms, and the set frame shift can be set to 15ms.

[0102] Since the sound is short-term stable, the voiceprint signal is framed, that is, the continuous voiceprint signal is processed according to the set frame length (for example, 25ms, 30ms, etc.) and the set frame shift (for example, 15ms, 20ms, etc.), so as to convert the continuous voiceprint signal into a discrete voiceprint sample signal, such as Figure 2 shown.

[0103] The size of the set frame length can be selected according to needs. For example, a shorter frame length can provide finer time domain details, but may sacrifice signal stability and frequency domain details; a longer frame length can improve signal stability and frequency domain details, but may reduce time domain details, resulting in the inability to capture rapidly changing sound characteristics.

[0104] When performing the frame processing, the frame shift refers to the size of the overlap between two adjacent frames. The size of the frame shift can also be selected according to needs. A smaller frame shift can provide higher temporal details because there is less overlap between adjacent frames, but it may lead to an increase in the amount of calculation; a larger frame shift can reduce the amount of calculation, but it may reduce the temporal details, causing some rapidly changing sound features to be ignored.

[0105] The specified number of voiceprint sample signals is related to the specified time period, the set frame length and the set frame shift. Figure 2 It can be seen that after the set frame length and the set frame shift are set, the number of voiceprint sample signals within the specified time period can be determined.

[0106] According to an embodiment of the present disclosure, the windowing process is implemented by using a Hamming window function, wherein the window size of the Hamming window function is the set frame length; the Hamming window function is shown in the following formula:

[0107]

[0108] Wherein, ω(n) is the Hamming window function, N is the set frame length, and n is the sampling point.

[0109] The windowing process is to achieve a smooth transition of the edge of the voiceprint signal, prevent spectrum leakage, and avoid the collected voiceprint signal from being truncated within a limited time, so as to make the obtained voiceprint sample signal more accurate.

[0110] According to an embodiment of the present disclosure, the background processing system is also used to perform continuous wavelet transform on the specified number of voiceprint sample signals, extract continuous wavelet transform coefficients, and thus draw a time-wavelet coefficient graph, and use the time-wavelet coefficient graph as the voiceprint image of the voiceprint signal.

[0111] Furthermore, performing continuous wavelet transform on the specified number of voiceprint sample signals includes:

[0112] Select a wavelet basis function and a decomposition layer number, and perform a continuous wavelet transform on each voiceprint sample signal in the specified number of voiceprint sample signals, so as to extract the continuous wavelet transform coefficient. The wavelet basis function is shown in the following formula:

[0113]

[0114] Among them, ψ a，τ (t) is the wavelet basis function, a is the wavelet scaling factor, τ is the wavelet translation factor, and t is the time variable.

[0115] The continuous wavelet transform coefficients are extracted using the following formula:

[0116]

[0117] Among them, WT f (a, τ) is the continuous wavelet transform coefficient, f(t) is the specified number of voiceprint sample signals, is the wavelet basis function ψ a，τ is the conjugate complex number of (t), · is the inner product operation, and dt is the differentiation operation.

[0118] In a specific implementation, a specified number of voiceprint sample signals can be input into a wavelet transform program, and then the wavelet basis function and the number of decomposition layers are selected, and continuous wavelet transform is performed on each voiceprint sample signal to extract the wavelet coefficients, thereby drawing a wavelet coefficient graph. In the wavelet coefficient graph, the horizontal axis is time and the vertical axis is frequency, and the size of the wavelet coefficient is reflected by the depth of color in the graph.

[0119] According to an embodiment of the present disclosure, the background processing system is further used to perform Histogram of Oriented Gradients (HOG) feature extraction on the voiceprint image to obtain a HOG feature vector group.

[0120] Performing HOG feature extraction on the voiceprint image includes:

[0121] The voiceprint image is divided into a plurality of cell units, and the gradient value and gradient direction of each pixel point in the cell unit are calculated; according to the gradient value and gradient direction of each pixel point in the cell unit, a first directional gradient histogram of the cell unit is calculated, and the first directional gradient histogram has a first set vector dimension.

[0122] Figure 3 A schematic diagram showing the calculation of the first directional gradient histogram of a cell unit in an embodiment of the present disclosure.

[0123] like Figure 3 As shown, assuming that the size of a cell unit is 8*8 (pixels), each pixel point includes two values: a gradient value and a gradient direction, and then 180 degrees are divided into 9 (first set vector dimension) groups: 0°-20°, 20°-40°, 40°-60°, 60°-80°, 80°-100°, 100°-120°, 120°-140°, 140°-160°, 160°-180°, that is, the 64 pixels in each cell unit are divided into the above 9 groups of angle ranges according to the gradient direction, and the gradient values ​​of one or more pixels in each angle range are accumulated to obtain the first directional gradient histogram, and the vector dimension of the first directional gradient histogram is 9.

[0124] exist Figure 3 The same color is used to indicate that the pixel points of the cell unit are within the same gradient direction range, and the gradient value of the first pixel point is a, the gradient value of the second pixel point is b, the gradient value of the third pixel point is c, and the gradient value of the fourth pixel point is d. Assuming that the gradient direction range of the first pixel point and the third pixel point is between 0°-20°, and the gradient direction range of the second pixel point and the fourth pixel point is between 40°-60°, the gradient values ​​a and c are accumulated between 0°-20°, and the gradient values ​​b and d are accumulated between 40°-60°; and so on, the gradient value of each pixel point in a cell unit is accumulated according to the gradient direction, so as to obtain the first directional gradient histogram of the cell unit.

[0125] Figure 4 A schematic diagram showing the division of cell units and image blocks when performing HOG feature extraction on a voiceprint image in an embodiment of the present disclosure is shown.

[0126] like Figure 4 As shown, a sliding window is used to extract image blocks on a voiceprint image divided into a plurality of cell units to obtain a plurality of image blocks, wherein the sliding window has a set window size and a set sliding step, and the image block has a set number of cell units; a second directional gradient histogram of the image block is calculated, the second directional gradient histogram has a second set vector dimension, and the second set vector dimension is the sum of the set number of first set vector dimensions.

[0127] Assuming that the sliding window is the size of 4 cell units, and assuming that the sliding step is 8 (pixels), if the size of each cell unit is 8*8 (pixels), the size of the sliding window is 16*16 (pixels), and by extracting the image block of the voiceprint image through the sliding window, multiple 16*16 (pixel) image blocks can be obtained, each of which includes 4 cell units. By dividing 180 degrees into 9 groups (as mentioned above), since the vector dimension of the first directional gradient histogram of a cell unit is 9, the second set vector dimension of an image block is 9*4=36.

[0128] Those skilled in the art should understand that the size of the cell unit, the division method of the gradient direction (dividing 0-180 degrees into 9 groups), and the setting of the window size and sliding step of the sliding window are not technical means to limit the protection scope of the present disclosure. Those skilled in the art can customize the settings as needed.

[0129] According to an embodiment of the present disclosure, the plurality of second directional gradient histograms of the plurality of image blocks are normalized to obtain the HOG feature vector group.

[0130] Wherein, the gradient value and gradient direction of each pixel point in the cell unit include:

[0131] For the cell unit, the gradient value and gradient direction of each pixel point in the cell unit are calculated according to the gradient value in the horizontal coordinate direction and the gradient value in the vertical coordinate direction of each pixel point.

[0132] Specifically, the calculating the gradient value and gradient direction of each pixel point in the cell unit according to the gradient value in the horizontal coordinate direction and the gradient value in the vertical coordinate direction of each pixel point includes calculating using the following formula:

[0133]

[0134] Among them, G x (x, y) is the gradient value of each pixel in the horizontal direction, G y (x, y) is the gradient value of each pixel in the ordinate direction, G(x, y) is the gradient value of each pixel, and α(x, y) is the gradient direction of each pixel.

[0135] According to an embodiment of the present disclosure, the gradient value in the horizontal coordinate direction and the gradient value in the vertical coordinate direction of each pixel point are calculated according to the grayscale value of each pixel point in the cell unit.

[0136] Specifically, the calculating the gradient value in the horizontal coordinate direction and the gradient value in the vertical coordinate direction of each pixel point according to the gray value of each pixel point in the cell unit includes calculating using the following formula:

[0137]

[0138] Among them, H(x+1,y) is the gradient value of the pixel point (x+1,y), H(x-1,y) is the gradient value of the pixel point (x-1,y), H(x,y+1) is the gradient value of the pixel point (x,y+1), H(x,y-1) is the gradient value of the pixel point (x,y-1), x is the horizontal coordinate of each pixel point, and y is the vertical coordinate of each pixel point.

[0139] According to an embodiment of the present disclosure, the vector dimension of the HOG feature vector group is calculated by the following formula:

[0140]

[0141] Wherein, P is the vector dimension of the HOG feature vector group, l p is the number of pixels in the horizontal coordinate direction of the image block, h p is the number of pixels in the ordinate direction of the image block, l c is the number of pixels in the horizontal coordinate direction of the cell unit, and M is the dimension of the second set vector.

[0142] According to an embodiment of the present disclosure, the background processing system is further used to input the HOG feature vector group into a trained fault multi-classification model to obtain the fault type of the faulty power equipment.

[0143] In the present disclosure, the trained fault multi-classification model is obtained by training a support vector machine (SVM) model.

[0144] The specific training process is as follows: for any power equipment, obtain voiceprint signals of different fault types (within a specified time period), and add labels to the voiceprint signals of each fault type (fault 1, fault 2, fault 3...); then perform pre-emphasis processing, frame processing, windowing processing, and continuous wavelet transform on the voiceprint signals of different fault types, extract the continuous wavelet transform coefficients, and draw a time-wavelet coefficient graph as a voiceprint image, and then perform HOG feature extraction to obtain a HOG feature vector group; divide the HOG feature vector group into a training set (generally 70%) and a test set (generally 30%), input the training set into the SVM model for training, adjust hyperparameters (such as C parameters, gamma parameters, etc.) to obtain a trained fault multi-classification model, and use the test set to perform performance testing on the trained fault multi-classification model.

[0145] The present invention realizes rapid and comprehensive detection and early fault identification of power equipment by performing non-contact detection settings for power equipment, and on this basis, analyzes images obtained by converting voiceprint signals from one-dimensional information to two-dimensional images, thereby improving recognition accuracy, being able to more accurately capture subtle differences in sound signals, and combining artificial intelligence algorithms for intelligent analysis and diagnosis, thereby achieving accurate classification of fault types, thereby enabling targeted fault processing based on fault types, avoiding unnecessary changes in equipment operating status, greatly reducing fault downtime, and improving the operating efficiency of the power system. It has a high degree of intelligence and strong accuracy and reliability in fault diagnosis.

[0146] Figure 5 FIG. 1 is a flow chart showing a method for intelligent fault diagnosis of electric power equipment based on voiceprint images according to an embodiment of the present disclosure. Figure 5 As shown, the method includes steps S501 to S506:

[0147] In step S501, the voiceprint signal of the faulty power equipment within a specified time period is obtained.

[0148] In the step S502, the voiceprint signal is pre-emphasized to obtain a pre-emphasized voiceprint signal.

[0149] In the step S503, the pre-emphasized voiceprint signal is subjected to frame processing and window processing to obtain a specified number of voiceprint sample signals.

[0150] In the step S504, continuous wavelet transform is performed on the specified number of voiceprint sample signals to extract continuous wavelet transform coefficients, thereby drawing a time-wavelet coefficient graph, and the time-wavelet coefficient graph is used as the voiceprint image of the voiceprint signal.

[0151] In the step S505, HOG feature extraction is performed on the voiceprint image to obtain a HOG feature vector group.

[0152] In the step S506, the HOG feature vector group is input into a trained fault multi-classification model to obtain the fault type of the faulty power equipment.

[0153] According to an embodiment of the present disclosure, before obtaining the voiceprint signal of the faulty power equipment within a specified time period, it also includes: obtaining multiple basic sound attribute information corresponding to multiple power equipment in the power system; analyzing the multiple basic sound attribute information to determine the faulty power equipment.

[0154] Figure 6 FIG. 1 is a schematic diagram showing a structure of a background processing system according to an embodiment of the present disclosure. Figure 6 As shown, the background processing system includes: an acquisition module 601 , a first preprocessing module 602 , a second preprocessing module 603 , a third preprocessing module 604 , a feature extraction module 605 and a fault classification module 606 .

[0155] The acquisition module 601 is configured to acquire the voiceprint signal of the faulty power equipment within a specified time period.

[0156] The first pre-processing module 602 is configured to perform pre-emphasis processing on the voiceprint signal to obtain a pre-emphasized voiceprint signal.

[0157] The second pre-processing module 603 is configured to perform frame processing and window processing on the pre-emphasized voiceprint signal to obtain a specified number of voiceprint sample signals.

[0158] The third preprocessing module 604 is configured to perform continuous wavelet transform on the specified number of voiceprint sample signals, extract continuous wavelet transform coefficients, and draw a time-wavelet coefficient graph, and use the time-wavelet coefficient graph as the voiceprint image of the voiceprint signal.

[0159] The feature extraction module 605 is configured to perform HOG feature extraction on the voiceprint image to obtain a HOG feature vector group.

[0160] The fault classification module 606 is configured to input the HOG feature vector group into a trained fault multi-classification model to obtain the fault type of the faulty power equipment.

[0161] In particular, according to an embodiment of the present disclosure, the method described above can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program tangibly contained on a machine-readable medium, and the computer program includes a program code for executing the above method. In such an embodiment, the computer program can be downloaded and installed from a network through a communication part, and / or installed from a removable medium.

[0162] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of a code, and the module, a program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0163] The units or modules involved in the embodiments described in the present disclosure may be implemented by software or programmable hardware. The units or modules described may also be set in a processor, and the names of these units or modules do not constitute limitations on the units or modules themselves in some cases.

[0164] As another aspect, the present disclosure further provides a computer-readable storage medium, which may be a computer-readable storage medium included in the electronic device or computer system in the above embodiment; or a computer-readable storage medium that exists independently and is not assembled into a device. The computer-readable storage medium stores one or more programs, and the programs are used by one or more processors to execute the method described in the present disclosure.

[0165] The above description is only a preferred embodiment of the present disclosure and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the present disclosure is not limited to the technical solution formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the inventive concept. For example, the above features are replaced with the technical features with similar functions disclosed in the present disclosure (but not limited to) by each other.

Claims

1. An intelligent fault diagnosis system for power equipment based on voiceprint image, characterized in that: Applied to an electric power system, the electric power system includes a plurality of electric power equipment, the intelligent fault diagnosis system includes: a background processing system and a plurality of sound signal monitoring devices, the sound signal monitoring devices corresponding to the electric power equipment; The sound signal monitoring device is used to monitor the basic attribute information of the sound of the corresponding power equipment and transmit the basic attribute information of the sound to the background processing system; The background processing system is used to analyze the basic attribute information of the sound received from the multiple sound signal monitoring devices respectively, determine the faulty power equipment, and obtain the voiceprint signal of the faulty power equipment within a specified time period; pre-emphasize the voiceprint signal to obtain a pre-emphasized voiceprint signal; perform frame processing and windowing processing on the pre-emphasized voiceprint signal to obtain a specified number of voiceprint sample signals; perform continuous wavelet transform on the specified number of voiceprint sample signals, extract continuous wavelet transform coefficients, and thus draw a time-wavelet coefficient graph, and use the time-wavelet coefficient graph as the voiceprint image of the voiceprint signal; perform directional gradient histogram HOG feature extraction on the voiceprint image to obtain a HOG feature vector group; input the HOG feature vector group into a trained fault multi-classification model to obtain the fault type of the faulty power equipment.

2. The system according to claim 1, characterized in that: The basic sound attribute information includes any one or more of the following: sound amplitude, sound frequency, and sound phase difference; The analyzing the basic sound attribute information received from the multiple sound signal monitoring devices respectively to determine the faulty power equipment includes: setting a preset threshold for each of the multiple power equipment, wherein the preset threshold includes any one or more of the following: a preset sound amplitude threshold, a preset sound frequency threshold, and a preset sound phase difference threshold; determining that the power equipment is a faulty power equipment when the basic sound attribute information meets any one or more of the following conditions: the sound amplitude exceeds the preset sound amplitude threshold, the sound frequency exceeds the preset sound frequency threshold, and the sound phase difference exceeds the preset sound phase difference threshold.

3. The system according to claim 1, characterized in that The pre-emphasis processing of the voiceprint signal includes: The voiceprint signal is pre-emphasized using a high-pass filter.

4. The system according to claim 3, characterized in that The transfer function of the high-pass filter is shown in the following formula: H(z)=1-μz -1 ; Wherein, H(z) is the transfer function, μ is the pre-emphasis coefficient, and z is a complex variable.

5. The system according to claim 1, characterized in that The performing frame processing and window processing on the pre-emphasized voiceprint signal includes: The frame segmentation process is implemented by using a set frame length and a set frame shift, wherein the set frame length is smaller than the time length of the specified time period, and the set frame shift is smaller than the set frame length.

6. The system according to claim 5, characterized in that The performing frame processing and window processing on the pre-emphasized voiceprint signal includes: The windowing process is implemented by using a Hamming window function, wherein the window size of the Hamming window function is the set frame length.

7. The system according to claim 6, characterized in that The Hamming window function is shown in the following formula: Wherein, ω(n) is the Hamming window function, N is the set frame length, and n is the sampling point.

8. The system according to claim 1, characterized in that The performing continuous wavelet transform on the specified number of voiceprint sample signals comprises: A wavelet basis function and a decomposition layer number are selected, and a continuous wavelet transform is performed on each voiceprint sample signal in the specified number of voiceprint sample signals, so as to extract the continuous wavelet transform coefficients.

9. The system according to claim 8, characterized in that The wavelet basis function is shown in the following formula: Among them, ψ a，τ (t) is the wavelet basis function, a is the wavelet scaling factor, τ is the wavelet translation factor, and t is the time variable.

10. The system according to claim 9, characterized in that The continuous wavelet transform coefficients are extracted using the following formula: Among them, WT f (a, τ) is the continuous wavelet transform coefficient, f(t) is the specified number of voiceprint sample signals, is the wavelet basis function ψ a，τ is the conjugate complex number of (t), · is the inner product operation, and dt is the differentiation operation.

11. The system according to claim 1, characterized in that The extracting HOG features from the voiceprint image includes: Dividing the voiceprint image into a plurality of cell units, and calculating the gradient value and gradient direction of each pixel point in the cell unit; Calculating a first directional gradient histogram of the cell unit according to the gradient value and gradient direction of each pixel point in the cell unit, wherein the first directional gradient histogram has a first set vector dimension; Using a sliding window to extract image blocks on a voiceprint image divided into a plurality of cell units, to obtain a plurality of image blocks, wherein the sliding window has a set window size and a set sliding step, and the image block has a set number of cell units; Calculating a second directional gradient histogram of the image block, wherein the second directional gradient histogram has a second set vector dimension, and the second set vector dimension is the sum of the set number of first set vector dimensions; Normalizing the plurality of second directional gradient histograms of the plurality of image blocks to obtain the HOG feature vector group.

12. The system according to claim 11, characterized in that The step of calculating the gradient value and gradient direction of each pixel in the cell unit includes: For the cell unit, the gradient value and gradient direction of each pixel point in the cell unit are calculated according to the gradient value in the horizontal coordinate direction and the gradient value in the vertical coordinate direction of each pixel point.

13. The system according to claim 12, characterized in that The step of calculating the gradient value and the gradient direction of each pixel point in the cell unit according to the gradient value in the horizontal coordinate direction and the gradient value in the vertical coordinate direction of each pixel point comprises calculating using the following formula: Among them, G x (x, y) is the gradient value of each pixel in the horizontal direction, G y (x, y) is the gradient value of each pixel in the ordinate direction, G(x, y) is the gradient value of each pixel, and α(x, y) is the gradient direction of each pixel.

14. The system according to claim 13, characterized in that The gradient value in the horizontal coordinate direction and the gradient value in the vertical coordinate direction of each pixel point are calculated according to the gray value of each pixel point in the cell unit.

15. The system according to claim 14, characterized in that Calculating the gradient value in the horizontal coordinate direction and the gradient value in the vertical coordinate direction of each pixel point according to the gray value of each pixel point in the cell unit includes calculating using the following formula: Among them, H(x+1,y) is the gradient value of the pixel point (x+1,y), H(x-1,y) is the gradient value of the pixel point (x-1,y), H(x,y+1) is the gradient value of the pixel point (x,y+1), H(x,y-1) is the gradient value of the pixel point (x,y-1), x is the horizontal coordinate of each pixel point, and y is the vertical coordinate of each pixel point.

16. The system according to claim 11, characterized in that The vector dimension of the HOG feature vector group is calculated by the following formula: Wherein, P is the vector dimension of the HOG feature vector group, l p is the number of pixels in the horizontal coordinate direction of the image block, h p is the number of pixels in the ordinate direction of the image block, l c is the number of pixels in the horizontal coordinate direction of the cell unit, and M is the dimension of the second set vector.

17. The system according to claim 1, characterized in that The trained fault multi-classification model is obtained by training the support vector machine (SVM) model.

18. An intelligent fault diagnosis method for power equipment based on voiceprint image, characterized in that: include: Obtain the voiceprint signal of the faulty power equipment within a specified time period; Performing pre-emphasis processing on the voiceprint signal to obtain a pre-emphasized voiceprint signal; Performing frame processing and window processing on the pre-emphasized voiceprint signal to obtain a specified number of voiceprint sample signals; Performing continuous wavelet transform on the specified number of voiceprint sample signals, extracting continuous wavelet transform coefficients, thereby drawing a time-wavelet coefficient graph, and using the time-wavelet coefficient graph as a voiceprint image of the voiceprint signal; Performing HOG feature extraction on the voiceprint image to obtain a HOG feature vector group; The HOG feature vector group is input into a trained fault multi-classification model to obtain the fault type of the faulty power equipment.

19. The method according to claim 18, characterized in that Before obtaining the voiceprint signal of the faulty power equipment within a specified time period, the method further includes: Acquire basic attribute information of a plurality of sounds corresponding to a plurality of electric power equipment in the electric power system; The basic attribute information of the multiple sounds is analyzed to determine the faulty electrical equipment.

20. A background processing system, characterized in that: include: An acquisition module is configured to acquire a voiceprint signal of a faulty power device within a specified time period; A first preprocessing module is configured to perform pre-emphasis processing on the voiceprint signal to obtain a pre-emphasized voiceprint signal; A second preprocessing module is configured to perform frame processing and window processing on the pre-emphasized voiceprint signal to obtain a specified number of voiceprint sample signals; a third preprocessing module, configured to perform continuous wavelet transform on the specified number of voiceprint sample signals, extract continuous wavelet transform coefficients, thereby drawing a time-wavelet coefficient graph, and using the time-wavelet coefficient graph as a voiceprint image of the voiceprint signal; A feature extraction module is configured to perform HOG feature extraction on the voiceprint image to obtain a HOG feature vector group; The fault classification module is configured to input the HOG feature vector group into a trained fault multi-classification model to obtain the fault type of the faulty power equipment.

21. A computer-readable storage medium having computer instructions stored thereon, characterized in that: When the computer instructions are executed by a processor, the method according to any one of claims 18 to 19 is implemented.

22. A computer program product, comprising computer instructions, which implement the method according to any one of claims 18 to 19 when executed by a processor.

Images