Photovoltaic power station transformer fault identification method and system based on multi-scale co-fusion

Abstract

The invention relates to the technical field of transformer fault identification, and discloses a photovoltaic power station transformer fault identification method and system based on multi-scale co-fusion, and the method comprises the steps: carrying out the short-time Fourier transform of an initial operation sound signal, obtaining an initial sound time-frequency spectrum, carrying out the frequency compression of the initial sound time-frequency spectrum, obtaining a target sound time-frequency spectrum, and carrying out the calculation of the target sound time-frequency spectrum. Calculating a first time-frequency similarity between the target sound time-frequency spectrum and the average fault time-frequency spectrum, calculating a second time-frequency similarity between a plurality of groups of target fault index value sets and a plurality of groups of reference fault index value sets to obtain a second time-frequency similarity set, inputting the target sound time-frequency spectrum into a target neural network to obtain a third time-frequency probability value set, and obtaining a second time-frequency probability value set; and according to the multi-scale weight set, the first time-frequency similarity set, the second time-frequency similarity set and the third time-frequency probability value set, calculating a target time-frequency confidence coefficient set, and identifying a target transformer fault type corresponding to the maximum target time-frequency confidence coefficient. According to the invention, the transformer fault determination accuracy can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of transformer fault identification, and in particular to a method and system for identifying transformer faults in a photovoltaic power station based on multi-scale integration. Background Art

[0002] With the continuous growth of global energy demand and the increasing awareness of environmental protection, photovoltaic power generation technology has become an important development direction in the global energy field. However, during the operation of photovoltaic power stations, key equipment such as transformers often face various faults. If these faults cannot be discovered and handled in time, they will seriously affect the stable operation and power generation efficiency of photovoltaic power stations.

[0003] Traditional transformer fault identification methods mainly rely on manual inspections and regular testing, which is not only inefficient but also difficult to detect and handle potential faults in a timely manner. Currently, the type of transformer fault is mainly determined by a single determination method, such as neural networks, expert diagnosis, or fault data comparison. This single transformer determination method cannot combine the advantages of various determination methods, resulting in low transformer fault determination accuracy. Summary of the invention

[0004] The present invention provides a photovoltaic power station transformer fault identification method and system based on multi-scale integration, the main purpose of which is to improve the accuracy of transformer fault determination.

[0005] To achieve the above objectives, the present invention provides a photovoltaic power station transformer fault identification method based on multi-scale integration, comprising:

[0006] The original operating sound signal of the transformer is collected using a pre-built microphone array, wherein the collection process of the original operating sound signal includes: sound source localization, noise suppression, sound source enhancement and blind source separation, and the duration of the original operating sound signal is the signal fault identification duration;

[0007] Preprocessing the original running sound signal to obtain an initial running sound signal, wherein the preprocessing includes: pre-emphasis, framing and windowing;

[0008] Performing short-time Fourier transform on the initial running sound signal to obtain an initial sound time-frequency spectrum, and performing frequency compression on the initial sound time-frequency spectrum according to preset low-frequency band, medium-frequency band and high-frequency band to obtain a target sound time-frequency spectrum;

[0009] Extracting transformer fault types in sequence from a preset transformer fault type set, obtaining an average fault time-frequency spectrum of the transformer fault type, calculating a first time-frequency similarity between the target sound time-frequency spectrum and the average fault time-frequency spectrum according to a preset matching formula, and obtaining a first time-frequency similarity set;

[0010] Receiving multiple groups of reference fault indicator value sets input by a user according to the transformer fault type, identifying multiple groups of target fault indicator value sets of the target sound time-frequency spectrum, calculating the second time-frequency similarity between the multiple groups of target fault indicator value sets and the multiple groups of reference fault indicator value sets, and obtaining a second time-frequency similarity set, wherein the reference fault indicator set includes: reference fault waveform wavelength, reference fault waveform amplitude and reference fault waveform kurtosis;

[0011] Inputting the target sound time-frequency spectrum into a pre-constructed target neural network to obtain a third time-frequency probability value set, wherein the target neural network is a BP neural network;

[0012] According to the preset multi-scale weight set, the first time-frequency similarity set, the second time-frequency similarity set and the third time-frequency probability value set, the target time-frequency confidence set is calculated using the pre-constructed time-frequency similarity formula, and the time-frequency similarity formula is as follows:

[0013]

[0014] Among them, s i Indicates the target time-frequency confidence level of the fault type of the target sound time-frequency spectrum as the i-th transformer fault type, b 1 represents the first communion weight, represents the time-frequency similarity between the target sound time-frequency spectrum in the first time-frequency similarity set and the average fault time-frequency spectrum of the i-th transformer fault type, b 2 represents the second communion weight, represents the time-frequency similarity between the target fault index value set and the fault index value set of the i-th transformer fault type in the second time-frequency similarity set, b 3 represents the third communion weight, Indicates the time-frequency probability value of the fault type of the target sound time-frequency spectrum in the third time-frequency probability value set as the i-th transformer fault type;

[0015] Identify the maximum target time-frequency confidence in the target time-frequency confidence set, and identify the target transformer fault type corresponding to the maximum target time-frequency confidence

[0016] Optionally, the frequency compression of the initial sound time-frequency spectrum according to the preset low frequency band, middle frequency band and high frequency band to obtain the target sound time-frequency spectrum includes:

[0017] Segment the initial sound time spectrum according to the low frequency band, the middle frequency band and the high frequency band to obtain a low frequency band time spectrum, a middle frequency band time spectrum and a high frequency band time spectrum;

[0018] According to the preset unit time zone, the low-frequency band time-frequency spectrum, the medium-frequency band time-frequency spectrum and the high-frequency band time-frequency spectrum are respectively segmented twice to obtain multiple groups of unit low-frequency band time-frequency sequences, multiple groups of unit medium-frequency band time-frequency sequences and multiple groups of unit high-frequency band time-frequency sequences;

[0019] Performing maximum amplitude extraction in the multiple groups of low-frequency time-frequency sequences, the multiple groups of mid-frequency time-frequency sequences, and the multiple groups of high-frequency time-frequency sequences, respectively, to obtain multiple groups of low-frequency time-frequency point sequences, multiple groups of mid-frequency time-frequency point sequences, and multiple groups of high-frequency time-frequency point sequences;

[0020] Performing time-frequency point fitting on the multiple groups of low-frequency time-frequency point sequences, the multiple groups of mid-frequency time-frequency point sequences, and the multiple groups of high-frequency time-frequency point sequences respectively to obtain a target low-frequency time-frequency spectrum, a target mid-frequency time-frequency spectrum, and a target high-frequency time-frequency spectrum;

[0021] The target low-frequency band time-frequency spectrum, the target mid-frequency band time-frequency spectrum and the target high-frequency band time-frequency spectrum are frequency-connected to obtain a target sound time-frequency spectrum.

[0022] Optionally, obtaining an average fault time spectrum of the transformer fault type includes:

[0023] Acquire a fault time-frequency spectrum set of the transformer fault type;

[0024] Extracting fault time spectra in the fault time spectrum set in sequence, receiving the clipping of the fault time spectrum by a user according to the signal fault identification duration and the signal fault identification frequency segment, and obtaining a time spectrum set to be averaged;

[0025] Determine the signal fault identification time sequence and the signal fault identification frequency sequence according to the signal fault identification time length and the signal fault identification frequency segment, and construct a fault calibration dot matrix according to the signal fault identification time sequence and the signal fault identification frequency sequence;

[0026] Extracting fault calibration points in the fault calibration point matrix in sequence, and identifying the corresponding to-be-averaged amplitude set in the to-be-averaged time spectrum set according to the fault calibration points, wherein the fault calibration points include: fault calibration frequency and fault calibration time;

[0027] Calculating the average amplitude of the amplitude set to be averaged, and determining the three-dimensional fault time-frequency point according to the average amplitude and the fault calibration point;

[0028] Summarize the three-dimensional fault time-frequency points of each fault calibration point to obtain a three-dimensional fault time-frequency point set;

[0029] The three-dimensional fault time-frequency point set is fitted to obtain an average fault time-frequency spectrum.

[0030] Optionally, the matching formula is as follows:

[0031]

[0032] in, represents the first time-frequency similarity between the average fault time-frequency spectrum of the i-th transformer fault type and the target sound time-frequency spectrum, P represents the number of fault calibration points in the fault calibration point matrix, and f p Represents the signal amplitude of the pth fault calibration point in the target sound time spectrum, It represents the signal amplitude of the pth fault calibration point in the average fault time spectrum of the i-th transformer fault type, and |*| represents the absolute value sign.

[0033] Optionally, the multiple sets of target fault indicator value sets of the frequency spectrum of the target sound are identified, including:

[0034] Extracting signal fault calibration frequencies in sequence from the signal fault identification frequency sequence, and identifying amplitude change signals of the signal fault calibration frequencies in the target sound time spectrum;

[0035] Identifying a target waveform wavelength set, a target waveform amplitude set, and a target waveform kurtosis set of the amplitude variation signal;

[0036] According to the target waveform wavelength set, the target waveform amplitude set and the target waveform kurtosis set, respectively calculate the target average waveform wavelength, the target average waveform amplitude and the target average waveform kurtosis to obtain the target fault index value set;

[0037] The target fault index value sets of the fault calibration frequencies of each signal are collected to obtain multiple groups of target fault index value sets.

[0038] Optionally, the calculating the second time-frequency similarities between the multiple groups of target fault indicator value sets and the multiple groups of reference fault indicator value sets to obtain the second time-frequency similarity sets includes:

[0039] The second time-frequency similarity is calculated according to the multiple groups of target fault indicator value sets and the multiple groups of reference fault indicator value sets using the following formula:

[0040]

[0041] in, represents the second time-frequency similarity between multiple target fault index value sets and multiple reference fault index value sets of the i-th transformer fault type, k 1 represents the wavelength coefficient, k 2 represents the amplitude coefficient, k 3 represents the kurtosis coefficient, Q represents the number of signal fault calibration frequencies in the signal fault identification frequency sequence, and λ q represents the reference fault waveform wavelength in the qth group of reference fault index value sets, represents the target average waveform wavelength in the qth group of target fault index values, αq represents the reference fault waveform amplitude in the qth group of reference fault index value sets, represents the target average waveform amplitude in the qth group of target fault index values, β q represents the kurtosis of the reference fault waveform in the qth group of reference fault index values, represents the target average waveform kurtosis in the qth group of target fault index value sets;

[0042] The second time-frequency similarities corresponding to each transformer fault type are collected to obtain a second time-frequency similarity set.

[0043] Optionally, inputting the target sound time-frequency spectrum into a pre-constructed target neural network to obtain a third time-frequency probability value set includes:

[0044] Extracting an amplitude array from the amplitude variation signal of the signal fault calibration frequency according to a preset amplitude sampling interval to obtain an amplitude array set corresponding to different signal calibration frequencies;

[0045] The amplitude array set is input into the target neural network to obtain a third frequency similarity set, wherein the input layer of the target neural network includes: a first amplitude array input node, a second amplitude array input node, ..., an nth amplitude array input node, and the output layer of the target neural network includes: a first fault type probability value output node, a second fault type probability value output node, ..., an nth fault type probability value output node.

[0046] Optionally, the multi-scale weight set is obtained by:

[0047] Extracting training sound time-frequency spectrum sets in sequence from a plurality of preset training sound time-frequency spectrum sets;

[0048] Sequentially extracting the time spectra of training sounds from the time spectra set of training sounds;

[0049] Calculate a first training similarity set of the training sound frequency spectrum according to the average fault frequency spectrum and the matching formula, identify a maximum first training similarity in the first training similarity set, identify a first predicted fault type corresponding to the maximum first training similarity, and obtain a first predicted fault type set;

[0050] Identify multiple sets of training fault indicator value sets of the frequency spectrum of the training sound, calculate a second training similarity set according to the multiple sets of reference fault indicator value sets and the multiple sets of training fault indicator value sets, identify a maximum second training similarity in the second training similarity set, identify a second predicted fault type corresponding to the maximum second training similarity, and obtain a second predicted fault type set;

[0051] Predicting a third training probability value set of the frequency spectrum of the training sound using the target neural network, identifying a maximum third training probability value in the third training probability value set, identifying a third predicted fault type corresponding to the maximum third training probability value, and obtaining a third predicted fault type set;

[0052] Acquire the real fault type of the frequency spectrum of the training sound to obtain a real fault type set;

[0053] Respectively calculating the sequence coincidence of the first predicted fault type set, the second predicted fault type set and the third predicted fault type set with the real fault type set to obtain a first sequence coincidence, a second sequence coincidence and a third sequence coincidence;

[0054] Determining whether the first sequence coincidence degree, the second sequence coincidence degree, and the third sequence coincidence degree tend to be stable;

[0055] If the first sequence overlap, the second sequence overlap and the third sequence overlap do not tend to be stable, returning to the above step of sequentially extracting training sound time spectrum sets from the preset multiple sets of training sound time spectrum sets;

[0056] If the first sequence coincidence degree, the second sequence coincidence degree and the third sequence coincidence degree tend to be stable, the first sequence coincidence degree, the second sequence coincidence degree and the third sequence coincidence degree are normalized to obtain a multi-scale weight set.

[0057] Optionally, the determining whether the first sequence coincidence degree, the second sequence coincidence degree, and the third sequence coincidence degree tend to be stable includes:

[0058] Obtaining a first sequence coincidence degree sequence, a second sequence coincidence degree sequence, and a third sequence coincidence degree sequence;

[0059] Draw a first sequence coincidence curve, a second sequence coincidence curve and a third sequence coincidence curve according to the first sequence coincidence sequence, the second sequence coincidence sequence and the third sequence coincidence sequence respectively;

[0060] Identifying a first end slope, a second end slope and a third end slope of the first sequence coincidence curve, the second sequence coincidence curve and the third sequence coincidence curve;

[0061] Determining whether the first terminal slope, the second terminal slope, and the third terminal slope are all less than a preset terminal slope threshold;

[0062] If the first terminal slope, the second terminal slope and the third terminal slope are not all less than the terminal slope threshold, the first sequence coincidence degree, the second sequence coincidence degree and the third sequence coincidence degree are not stable;

[0063] If the first terminal slope, the second terminal slope and the third terminal slope are all less than the terminal slope threshold, the first sequence coincidence, the second sequence coincidence and the third sequence coincidence tend to be stable. To achieve the above purpose, the present invention also provides a photovoltaic power station transformer fault identification system based on multi-scale integration, including:

[0064] A target sound time-frequency spectrum acquisition module is used to collect the original operating sound signal of the transformer using a pre-built microphone array, wherein the collection process of the original operating sound signal includes: sound source positioning, noise suppression, sound source enhancement and blind source separation, wherein the duration of the original operating sound signal is the signal fault identification duration; pre-processing the original operating sound signal to obtain an initial operating sound signal, wherein the pre-processing includes: pre-emphasis, framing and windowing; performing short-time Fourier transform on the initial operating sound signal to obtain an initial sound time-frequency spectrum, and frequency compressing the initial sound time-frequency spectrum according to preset low-frequency bands, medium-frequency bands and high-frequency bands to obtain a target sound time-frequency spectrum;

[0065] A time-frequency classification calculation module is used to extract transformer fault types in a preset transformer fault type set in sequence, obtain an average fault time-frequency spectrum of the transformer fault type, calculate a first time-frequency similarity between the target sound time-frequency spectrum and the average fault time-frequency spectrum according to a preset matching formula, and obtain a first time-frequency similarity set; receive multiple groups of reference fault indicator value sets input by a user according to the transformer fault type, identify multiple groups of target fault indicator value sets of the target sound time-frequency spectrum, calculate a second time-frequency similarity between the multiple groups of target fault indicator value sets and multiple groups of reference fault indicator value sets, and obtain a second time-frequency similarity set, wherein the reference fault indicator set includes: reference fault waveform wavelength, reference fault waveform amplitude and reference fault waveform kurtosis; input the target sound time-frequency spectrum into a pre-constructed target neural network to obtain a third time-frequency probability value set, wherein the target neural network is a BP neural network;

[0066] The target time-frequency confidence set calculation module is used to calculate the target time-frequency confidence set using the pre-constructed time-frequency similarity formula according to the preset multi-scale weight set, the first time-frequency similarity set, the second time-frequency similarity set and the third time-frequency probability value set:

[0067]

[0068] Among them, s i Indicates the target time-frequency confidence level of the fault type of the target sound time-frequency spectrum as the i-th transformer fault type, b 1 represents the first communion weight, represents the time-frequency similarity between the target sound time-frequency spectrum in the first time-frequency similarity set and the average fault time-frequency spectrum of the i-th transformer fault type, b 2represents the second communion weight, represents the time-frequency similarity between the target fault index value set and the fault index value set of the i-th transformer fault type in the second time-frequency similarity set, b 3 represents the third communion weight, Indicates the time-frequency probability value of the fault type of the target sound time-frequency spectrum in the third time-frequency probability value set as the i-th transformer fault type;

[0069] The target transformer fault type identification module is used to identify the maximum target time-frequency confidence in the target time-frequency confidence set, and identify the target transformer fault type corresponding to the maximum target time-frequency confidence.

[0070] In order to solve the above problem, the present invention further provides an electronic device, the electronic device comprising:

[0071] A memory storing at least one instruction; and a processor executing the instruction stored in the memory to implement the photovoltaic power station transformer fault identification method based on multi-scale integration as described above.

[0072] In order to solve the above problems, the present invention also provides a computer-readable storage medium, in which at least one instruction is stored. The at least one instruction is executed by a processor in an electronic device to implement the above-mentioned photovoltaic power station transformer fault identification method based on multi-scale integration.

[0073] In order to solve the problem described in the background technology, the present invention needs to first obtain the target sound time-frequency spectrum, then use different judgment methods to determine the transformer fault type on the target sound time-frequency spectrum, and then integrate various judgment results to calculate the target time-frequency confidence set, and finally identify the maximum target time-frequency confidence in the target time-frequency confidence set, and identify the target transformer fault type corresponding to the maximum target time-frequency confidence. When obtaining the target sound time-frequency spectrum, it is necessary to first use a pre-built microphone array to collect the original operating sound signal of the transformer, and then pre-process the original operating sound signal to obtain an initial operating sound signal. At this time, it is necessary to perform a short-time Fourier transform on the initial operating sound signal to obtain an initial sound time-frequency spectrum. Since frequency compression of different proportions is required, it is necessary to first segment the initial sound time-frequency spectrum according to preset low-frequency bands, medium-frequency bands and high-frequency bands, and then perform frequency compression to obtain the target sound time-frequency spectrum. At this time, the transformer fault type can be determined. First, it is necessary to extract the transformer fault types in the preset transformer fault type set in sequence. In the first judgment In the method, it is necessary to first obtain the average fault time spectrum of the transformer fault type, and then calculate the first time-frequency similarity between the target sound time spectrum and the average fault time spectrum according to the preset matching formula to obtain the first time-frequency similarity set. In the second judgment method, it is necessary to receive multiple groups of reference fault index value sets input by the user according to the transformer fault type, and then identify multiple groups of target fault index value sets of the target sound time spectrum, and finally calculate the second time-frequency similarity between the multiple groups of target fault index value sets and multiple groups of reference fault index value sets to obtain the second time-frequency similarity set. In the third judgment method, it is necessary to input the target sound time spectrum into the pre-constructed target neural network to obtain the third time-frequency probability value set. After completing the calculation of the three methods, it is necessary to calculate the target time-frequency confidence set according to the preset multi-scale weight set, the first time-frequency similarity set, the second time-frequency similarity set and the third time-frequency probability value set using the pre-constructed time-frequency similarity formula. At this time, the maximum target time-frequency confidence is identified in the target time-frequency confidence set, and the target transformer fault type corresponding to the maximum target time-frequency confidence is identified. Therefore, the present invention can improve the accuracy of transformer fault determination. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] Figure 1 A schematic diagram of a process flow of a photovoltaic power station transformer fault identification method based on multi-scale integration provided by an embodiment of the present invention;

[0075] Figure 2 A time-frequency spectrum diagram of a target sound provided by an embodiment of the present invention;

[0076] Figure 3 A schematic diagram of multi-scale co-integration calculation provided by an embodiment of the present invention;

[0077] Figure 4 A schematic diagram of a smart voiceprint recognition platform provided by an embodiment of the present invention;

[0078] Figure 5 A functional module diagram of a photovoltaic power station transformer fault identification system based on multi-scale integration provided by an embodiment of the present invention;

[0079] Figure 6 A schematic diagram of the structure of an electronic device for implementing the photovoltaic power station transformer fault identification method based on multi-scale integration provided by an embodiment of the present invention.

[0080] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0081] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0082] The embodiment of the present application provides a method for identifying transformer faults in a photovoltaic power station based on multi-scale integration. The execution subject of the method for identifying transformer faults in a photovoltaic power station based on multi-scale integration includes but is not limited to at least one of the electronic devices such as a server and a terminal that can be configured to execute the method provided in the embodiment of the present application. In other words, the method for identifying transformer faults in a photovoltaic power station based on multi-scale integration can be executed by software or hardware installed in a terminal device or a server device, and the software can be a blockchain platform. The server includes but is not limited to: a single server, a server cluster, a cloud server or a cloud server cluster, etc.

[0083] Reference Figure 1 FIG. 1 is a flow chart of a method for identifying a transformer fault in a photovoltaic power station based on multi-scale integration according to an embodiment of the present invention. In this embodiment, the method for identifying a transformer fault in a photovoltaic power station based on multi-scale integration includes:

[0084] S1. Use the pre-built microphone array to collect the original operating sound signal of the transformer.

[0085] It can be explained that the microphone array refers to an array composed of multiple microphones arranged in a regular pattern. The original sound signal refers to the transformer operation sound signal after sound source localization, noise suppression, sound source enhancement and blind source separation.

[0086] In detail, the acquisition process of the original operation sound signal includes: sound source localization, noise suppression, sound source enhancement and blind source separation. In order to make the original operation sound signal contain information of uniform duration, it is necessary to limit the duration of the original operation sound signal. The duration of the original operation sound signal is the signal fault identification duration. The signal fault identification duration is set by the user and can be 1 minute.

[0087] Furthermore, the sound source localization, noise suppression, sound source enhancement and blind source separation are all existing technologies and will not be described in detail here.

[0088] S2. Preprocess the original operation sound signal to obtain an initial operation sound signal.

[0089] It can be explained that the initial running sound signal refers to the original running sound signal that has been pre-emphasized, framed and windowed.

[0090] In detail, the pre-processing includes: pre-emphasis, framing and windowing. In the pre-processing process, it is necessary to perform pre-emphasis and framing on the original running sound signal. Therefore, the frame length in the framing process can be set to 500ms, and the step length can be set to 1 / 2 frame length to reduce the amount of data processing operations. After the framing process, windowing needs to be performed. While reducing the amount of operations, framing will also bring data errors to the original running sound signal (because it is necessary to frame the original running sound signal after pre-emphasis, resulting in a sharp high-frequency signal at the signal boundary after framing, which is generally manifested as an increase in high-frequency harmonic components in the spectrum, thereby causing the Gibbs effect). In order to reduce this data error, it is necessary to perform endpoint smoothing windowing on the original running sound signal after framing, and a Hamming window can be used for windowing. The pre-emphasis, framing and windowing are prior art and will not be repeated here.

[0091] S3. Performing short-time Fourier transform on the initial running sound signal to obtain an initial sound time-frequency spectrum, and performing frequency compression on the initial sound time-frequency spectrum according to preset low frequency band, middle frequency band and high frequency band to obtain a target sound time-frequency spectrum.

[0092] It can be explained that the short time Fourier transform (Short Time Fourier Transorm) refers to dividing the initial running sound signal into multiple segments for Fourier transform processing. The target sound time spectrum refers to a time spectrum diagram containing frequency, point number (time point number) and amplitude information, which can be found in Figure 2 The low frequency band may be 0-100 Hz, the middle frequency band may be 100-200 Hz, and the high frequency band may be 200-300 Hz. The target sound time spectrum refers to the initial sound time spectrum after frequency compression by frequency band.

[0093] In the embodiment of the present invention, frequency compression is performed on the initial sound time-frequency spectrum according to the preset low frequency band, medium frequency band and high frequency band to obtain the target sound time-frequency spectrum, including:

[0094] Segment the initial sound time spectrum according to the low frequency band, the middle frequency band and the high frequency band to obtain a low frequency band time spectrum, a middle frequency band time spectrum and a high frequency band time spectrum;

[0095] According to the preset unit time zone, the low-frequency band time-frequency spectrum, the medium-frequency band time-frequency spectrum and the high-frequency band time-frequency spectrum are respectively segmented twice to obtain multiple groups of unit low-frequency band time-frequency sequences, multiple groups of unit medium-frequency band time-frequency sequences and multiple groups of unit high-frequency band time-frequency sequences;

[0096] Performing maximum amplitude extraction in the multiple groups of low-frequency time-frequency sequences, the multiple groups of mid-frequency time-frequency sequences, and the multiple groups of high-frequency time-frequency sequences, respectively, to obtain multiple groups of low-frequency time-frequency point sequences, multiple groups of mid-frequency time-frequency point sequences, and multiple groups of high-frequency time-frequency point sequences;

[0097] Performing time-frequency point fitting on the multiple groups of low-frequency time-frequency point sequences, the multiple groups of mid-frequency time-frequency point sequences, and the multiple groups of high-frequency time-frequency point sequences respectively to obtain a target low-frequency time-frequency spectrum, a target mid-frequency time-frequency spectrum, and a target high-frequency time-frequency spectrum;

[0098] The target low-frequency band time-frequency spectrum, the target mid-frequency band time-frequency spectrum and the target high-frequency band time-frequency spectrum are frequency-connected to obtain a target sound time-frequency spectrum.

[0099] It can be understood that the low-frequency band time spectrum, the medium-frequency band time spectrum and the high-frequency band time spectrum respectively refer to the sound time spectrum belonging to the low-frequency band, the medium-frequency band and the high-frequency band in the initial sound time spectrum. For example: when the low-frequency band can be 0-100Hz, the medium-frequency band can be 100-200Hz, and the high-frequency band can be 200-300Hz, the low-frequency band time spectrum, the medium-frequency band time spectrum and the high-frequency band time spectrum respectively refer to the sound time spectrum belonging to 0-100Hz, 100-200Hz and 200-300Hz in the initial sound time spectrum.

[0100] It should be understood that the unit time zone refers to the minimum sampling time interval for frequency compression, for example: 10ms. When the signal fault identification time length of the amplitude signal belonging to 50Hz in the low-frequency band time-frequency spectrum is 1min, the unit low-frequency band time-frequency sequence is the unit low-frequency band time-frequency sequence corresponding to the minimum sampling time intervals such as 0-10ms, 10-20ms, 20-30ms within 0-1min. Since there are multiple equally spaced frequencies in the low-frequency band, there are multiple groups of unit low-frequency band time-frequency sequences. For example: when the low-frequency band is 0-100Hz, and the multiple equally spaced frequencies in the low-frequency band are 10Hz, 20Hz, 30Hz,..., 100Hz, respectively, the number of groups of the unit low-frequency band time-frequency sequence is 10. The multiple groups of unit mid-frequency band time-frequency sequences and the multiple groups of unit high-frequency band time-frequency sequences are similar, and will not be repeated here.

[0101] Further, when the unit low-frequency band time-frequency sequence is 0-10ms, 10-20ms, 20-30ms, the low-frequency band time-frequency point sequence refers to the maximum amplitude sequence of each equally spaced frequency corresponding to 0-10ms, 10-20ms, 20-30ms, for example: when the frequency is 10Hz, the maximum amplitude corresponding to 0-10ms, 10-20ms, 20-30ms is 100mV, 110mV, 111mV respectively, then the first three values ​​of the unit low-frequency band time-frequency sequence are 100mV, 110mV, 111mV respectively. The same is true for the mid-frequency band time-frequency point sequence and the high-frequency band time-frequency point sequence, which will not be repeated here.

[0102] In detail, the time-frequency point fitting refers to drawing a curve spectrum for each group of low-frequency time-frequency point sequence, medium-frequency time-frequency point sequence and high-frequency time-frequency point sequence in a point-by-point fitting manner. The frequency connection refers to connecting the target low-frequency time-frequency spectrum, the target medium-frequency time-frequency spectrum and the target high-frequency time-frequency spectrum according to the frequency segmentation point positions of the low-frequency band, the medium-frequency band and the high-frequency band.

[0103] S4. Extract transformer fault types in sequence from a preset transformer fault type set, obtain an average fault time-frequency spectrum of the transformer fault type, calculate a first time-frequency similarity between the target sound time-frequency spectrum and the average fault time-frequency spectrum according to a preset matching formula, and obtain a first time-frequency similarity set.

[0104] It is understandable that the transformer fault type set may include: DC bias magnetic fault, no-load open circuit fault, phase-to-phase short circuit fault, three-term unbalance fault, component loose fault, etc. The average fault time-frequency spectrum refers to the fault time-frequency spectrum after the mean value is calculated for the fault time-frequency spectrum set of a certain transformer fault type. The first time-frequency similarity refers to the similarity between the target sound time-frequency spectrum and the average fault time-frequency spectrum, and the first time-frequency similarity set refers to the similarity set between the target sound time-frequency spectrum and the average fault time-frequency spectrum of various transformer fault types.

[0105] In the embodiment of the present invention, the step of obtaining the average fault time spectrum of the transformer fault type includes:

[0106] Acquire a fault time-frequency spectrum set of the transformer fault type;

[0107] Extracting fault time spectra in the fault time spectrum set in sequence, receiving the clipping of the fault time spectrum by a user according to the signal fault identification duration and the signal fault identification frequency segment, and obtaining a time spectrum set to be averaged;

[0108] Determine the signal fault identification time sequence and the signal fault identification frequency sequence according to the signal fault identification time length and the signal fault identification frequency segment, and construct a fault calibration dot matrix according to the signal fault identification time sequence and the signal fault identification frequency sequence;

[0109] Extracting fault calibration points in the fault calibration point matrix in sequence, and identifying the corresponding to-be-averaged amplitude set in the to-be-averaged time spectrum set according to the fault calibration points, wherein the fault calibration points include: fault calibration frequency and fault calibration time;

[0110] Calculating the average amplitude of the amplitude set to be averaged, and determining the three-dimensional fault time-frequency point according to the average amplitude and the fault calibration point;

[0111] Summarize the three-dimensional fault time-frequency points of each fault calibration point to obtain a three-dimensional fault time-frequency point set;

[0112] The three-dimensional fault time-frequency point set is fitted to obtain an average fault time-frequency spectrum.

[0113] Furthermore, the fault time-frequency spectrum set refers to a set of time-frequency spectrums corresponding to multiple transformer fault types. The fault identification frequency segment refers to a frequency segment composed of a low frequency segment, a medium frequency segment and a high frequency segment. The trimming process can be determined by the user to determine the time starting position and frequency starting position of the trimming. Since the duration of the fault time-frequency spectrum is different, in order to ensure that the duration and frequency interval of the fault time-frequency spectrum are respectively equal to the signal fault identification duration and the signal fault identification frequency segment, trimming is required.

[0114] In detail, the signal fault identification time sequence refers to the fault identification time sequence selected at equal intervals within the signal fault identification time length, the signal fault identification frequency sequence refers to the fault identification frequency sequence selected at equal intervals within the signal fault identification frequency range, and the fault calibration point array refers to the array composed of the signal fault identification time sequence and the signal fault identification frequency sequence. The fault calibration point refers to a fault identification time within the fault identification time length and a fault identification frequency calibration point within the signal fault identification frequency range, which can be referred to Figure 2 As shown, the fault calibration point can be (200ms, 100Hz). The amplitude set to be averaged refers to the amplitude set corresponding to the fault calibration point in the time-frequency spectrum to be averaged. The three-dimensional fault time-frequency point refers to the three-dimensional point coordinates determined by the fault calibration frequency, fault calibration time and average amplitude.

[0115] In the embodiment of the present invention, the matching formula is as follows:

[0116]

[0117] in, represents the first time-frequency similarity between the average fault time-frequency spectrum of the i-th transformer fault type and the target sound time-frequency spectrum, P represents the number of fault calibration points in the fault calibration point matrix, and f p Represents the signal amplitude of the pth fault calibration point in the target sound time spectrum, It represents the signal amplitude of the pth fault calibration point in the average fault time spectrum of the i-th transformer fault type, and |*| represents the absolute value sign.

[0118] S5. Receive multiple groups of reference fault indicator value sets input by the user according to the transformer fault type, identify multiple groups of target fault indicator value sets of the target sound time-frequency spectrum, calculate the second time-frequency similarity between the multiple groups of target fault indicator value sets and the multiple groups of reference fault indicator value sets, and obtain a second time-frequency similarity set.

[0119] In detail, the reference fault indicator set includes: a reference fault waveform wavelength, a reference fault waveform amplitude, and a reference fault waveform kurtosis.

[0120] It can be understood that the reference fault indicator value set refers to the reference fault waveform wavelength, reference fault waveform amplitude and reference fault waveform kurtosis corresponding to a certain signal fault identification frequency. Since there are multiple groups of signal fault identification frequencies, there are multiple groups of reference fault indicator value sets. The target fault indicator value set refers to the target fault waveform wavelength, target fault waveform amplitude and target fault waveform kurtosis corresponding to the signal fault identification frequency in the target sound time-frequency spectrum. The second time-frequency similarity refers to the degree of similarity between multiple groups of target fault indicator value sets and multiple groups of reference fault indicator value sets. The second time-frequency similarity set refers to a set composed of second time-frequency similarities. Since there are multiple types of transformer faults, there are multiple sets composed of second time-frequency similarities.

[0121] In the embodiment of the present invention, the multiple sets of target fault indicator value sets of the target sound time spectrum include:

[0122] Extracting signal fault calibration frequencies in sequence from the signal fault identification frequency sequence, and identifying amplitude change signals of the signal fault calibration frequencies in the target sound time spectrum;

[0123] Identifying a target waveform wavelength set, a target waveform amplitude set, and a target waveform kurtosis set of the amplitude variation signal;

[0124] According to the target waveform wavelength set, the target waveform amplitude set and the target waveform kurtosis set, respectively calculate the target average waveform wavelength, the target average waveform amplitude and the target average waveform kurtosis to obtain the target fault index value set;

[0125] The target fault index value sets of the fault calibration frequencies of each signal are collected to obtain multiple groups of target fault index value sets.

[0126] Understandable, see Figure 2As shown, the signal fault identification frequency sequence can be 10Hz, 20Hz, 30Hz, etc. The amplitude change signal refers to the change signal corresponding to the signal fault calibration frequency in the target sound time-frequency spectrum. The target waveform wavelength set, target waveform amplitude set and target waveform kurtosis set respectively refer to the waveform wavelength set, waveform amplitude set and waveform kurtosis set of each peak and trough in the amplitude change signal. The waveform wavelength can be the wavelength of the peak or trough. The target average waveform wavelength, target average waveform amplitude and target average waveform kurtosis respectively refer to the average values ​​of the target waveform wavelength set, target waveform amplitude set and target waveform kurtosis set.

[0127] In the embodiment of the present invention, the step of calculating the second time-frequency similarities between the multiple groups of target fault indicator value sets and the multiple groups of reference fault indicator value sets to obtain the second time-frequency similarity sets includes:

[0128] The second time-frequency similarity is calculated according to the multiple groups of target fault indicator value sets and the multiple groups of reference fault indicator value sets using the following formula:

[0129]

[0130] in, represents the second time-frequency similarity between multiple target fault index value sets and multiple reference fault index value sets of the i-th transformer fault type, k 1 represents the wavelength coefficient, k 2 represents the amplitude coefficient, k 3 represents the kurtosis coefficient, Q represents the number of signal fault calibration frequencies in the signal fault identification frequency sequence, and λ q represents the reference fault waveform wavelength in the qth group of reference fault index value sets, represents the target average waveform wavelength in the qth group of target fault index values, α q represents the reference fault waveform amplitude in the qth group of reference fault index value sets, represents the target average waveform amplitude in the qth group of target fault index values, β q represents the kurtosis of the reference fault waveform in the qth group of reference fault index values, represents the target average waveform kurtosis in the qth group of target fault index value sets;

[0131] The second time-frequency similarities corresponding to each transformer fault type are collected to obtain a second time-frequency similarity set.

[0132] S6. Input the target sound time-frequency spectrum into a pre-constructed target neural network to obtain a third time-frequency probability value set.

[0133] In detail, the target neural network is a BP neural network. The third time-frequency probability value set refers to the probability value set output by each output node of the output layer of the target neural network, each output node corresponds to a transformer fault type, and the probability value output by the output node indicates the probability that the transformer fault type corresponding to the target sound time-frequency spectrum is the transformer fault type corresponding to the output node, for example: when the output node is a loose component output node, the probability value output by the output node is 30%, which means that the target neural network determines that the transformer fault type corresponding to the target sound time-frequency spectrum is 30% of the probability of loose components.

[0134] In the embodiment of the present invention, the step of inputting the target sound time-frequency spectrum into a pre-constructed target neural network to obtain a third time-frequency probability value set includes:

[0135] Extracting an amplitude array from the amplitude variation signal of the signal fault calibration frequency according to a preset amplitude sampling interval to obtain an amplitude array set corresponding to different signal calibration frequencies;

[0136] The amplitude array set is input into the target neural network to obtain a third frequency similarity set, wherein the input layer of the target neural network includes: a first amplitude array input node, a second amplitude array input node, ..., an nth amplitude array input node, and the output layer of the target neural network includes: a first fault type probability value output node, a second fault type probability value output node, ..., an nth fault type probability value output node.

[0137] It can be understood that the amplitude sampling interval refers to the time interval for selecting the amplitude in the amplitude variation signal. For example, if the amplitude sampling interval is 10 points (10ms), the amplitude can be collected in the amplitude variation signal at intervals of 10ms to obtain an amplitude array. Since each signal calibration frequency corresponds to an amplitude variation signal, different signal calibration frequencies correspond to an amplitude array set composed of multiple amplitude arrays.

[0138] S7. Calculate a target time-frequency confidence set using a pre-constructed time-frequency similarity formula according to a preset multi-scale weight set, the first time-frequency similarity set, the second time-frequency similarity set and the third time-frequency probability value set.

[0139] Furthermore, the multi-scale weight set represents a set of calculation weights for the first time-frequency similarity, the second time-frequency similarity and the third time-frequency probability value. The target time-frequency confidence refers to the confidence that the target sound time-frequency spectrum belongs to a certain transformer fault type, obtained by weighted calculation of the first time-frequency similarity, the second time-frequency similarity and the third time-frequency probability value. The confidence refers to the degree of credibility of belonging to a certain transformer fault type. The process of calculating the target time-frequency confidence set using the pre-constructed time-frequency similarity formula according to the multi-scale weight set, the first time-frequency similarity set, the second time-frequency similarity set and the third time-frequency probability value set can be found in Figure 3 shown.

[0140] In detail, the time-frequency similarity formula is as follows:

[0141]

[0142] Among them, s i Indicates the target time-frequency confidence level of the fault type of the target sound time-frequency spectrum as the i-th transformer fault type, b 1 represents the first communion weight, represents the time-frequency similarity between the target sound time-frequency spectrum in the first time-frequency similarity set and the average fault time-frequency spectrum of the i-th transformer fault type, b 2 represents the second communion weight, represents the time-frequency similarity between the target fault index value set and the fault index value set of the i-th transformer fault type in the second time-frequency similarity set, b 3 represents the third communion weight, The time-frequency probability value indicating that the fault type of the target sound time-frequency spectrum in the third time-frequency probability value set is the i-th transformer fault type.

[0143] In an embodiment of the present invention, the method for acquiring the multi-scale weight set includes:

[0144] Extracting training sound time-frequency spectrum sets in sequence from a plurality of preset training sound time-frequency spectrum sets;

[0145] Sequentially extracting the time spectra of training sounds from the time spectra set of training sounds;

[0146] Calculate a first training similarity set of the training sound frequency spectrum according to the average fault frequency spectrum and the matching formula, identify a maximum first training similarity in the first training similarity set, identify a first predicted fault type corresponding to the maximum first training similarity, and obtain a first predicted fault type set;

[0147] Identify multiple sets of training fault indicator value sets of the frequency spectrum of the training sound, calculate a second training similarity set according to the multiple sets of reference fault indicator value sets and the multiple sets of training fault indicator value sets, identify a maximum second training similarity in the second training similarity set, identify a second predicted fault type corresponding to the maximum second training similarity, and obtain a second predicted fault type set;

[0148] Predicting a third training probability value set of the frequency spectrum of the training sound using the target neural network, identifying a maximum third training probability value in the third training probability value set, identifying a third predicted fault type corresponding to the maximum third training probability value, and obtaining a third predicted fault type set;

[0149] Acquire the real fault type of the frequency spectrum of the training sound to obtain a real fault type set;

[0150] Respectively calculating the sequence coincidence of the first predicted fault type set, the second predicted fault type set and the third predicted fault type set with the real fault type set to obtain a first sequence coincidence, a second sequence coincidence and a third sequence coincidence;

[0151] Determining whether the first sequence coincidence degree, the second sequence coincidence degree, and the third sequence coincidence degree tend to be stable;

[0152] If the first sequence overlap, the second sequence overlap and the third sequence overlap do not tend to be stable, returning to the above step of sequentially extracting training sound time spectrum sets from the preset multiple sets of training sound time spectrum sets;

[0153] If the first sequence coincidence degree, the second sequence coincidence degree and the third sequence coincidence degree tend to be stable, the first sequence coincidence degree, the second sequence coincidence degree and the third sequence coincidence degree are normalized to obtain a multi-scale weight set.

[0154] It is understandable that the multiple groups of training sound time-frequency spectra sets refer to a set of multiple groups of sound time-frequency spectra when transformer faults occurred in the past. The calculation method of the first training similarity set is the same as the calculation method of the first time-frequency similarity set, which will not be repeated here. The multiple groups of training fault indicator value sets refer to the multiple groups of fault indicator value sets corresponding to the training sound time-frequency spectra (since there are multiple groups of signal fault identification frequencies, there are multiple groups of fault indicator value sets), and the calculation method of the second training similarity set is consistent with that of the second time-frequency similarity set, which will not be repeated here. The calculation method of the third training probability value set is consistent with that of the third time-frequency probability value set, which will not be repeated here. The first sequence overlap, the second sequence overlap and the third sequence overlap respectively refer to the serialized overlap of the first predicted fault type set, the second predicted fault type set and the third predicted fault type set with the real fault type set. For example, when the first predicted fault type set is DC bias fault, no-load open circuit fault, DC bias fault, three-item unbalanced fault, and DC bias fault, and the real fault type set is no-load open circuit fault, no-load open circuit fault, three-item unbalanced fault, three-item unbalanced fault, and DC bias fault, since the first and third items are different, the first sequence overlap is 3 / 5.

[0155] In the embodiment of the present invention, the determining whether the first sequence coincidence degree, the second sequence coincidence degree, and the third sequence coincidence degree tend to be stable includes:

[0156] Obtaining a first sequence coincidence degree sequence, a second sequence coincidence degree sequence, and a third sequence coincidence degree sequence;

[0157] Draw a first sequence coincidence curve, a second sequence coincidence curve and a third sequence coincidence curve according to the first sequence coincidence sequence, the second sequence coincidence sequence and the third sequence coincidence sequence respectively;

[0158] Identifying a first end slope, a second end slope and a third end slope of the first sequence coincidence curve, the second sequence coincidence curve and the third sequence coincidence curve;

[0159] Determining whether the first terminal slope, the second terminal slope, and the third terminal slope are all less than a preset terminal slope threshold;

[0160] If the first terminal slope, the second terminal slope and the third terminal slope are not all less than the terminal slope threshold, the first sequence coincidence degree, the second sequence coincidence degree and the third sequence coincidence degree are not stable;

[0161] If the first terminal slope, the second terminal slope and the third terminal slope are all smaller than the terminal slope threshold, the first sequence coincidence degree, the second sequence coincidence degree and the third sequence coincidence degree tend to be stable.

[0162] It can be understood that the first terminal slope, the second terminal slope and the third terminal slope refer to the end point slopes of the first sequence coincidence curve, the second sequence coincidence curve and the third sequence coincidence curve respectively. The smaller the terminal slope, the more stable the corresponding sequence coincidence is.

[0163] S8. Identify a maximum target time-frequency confidence in the target time-frequency confidence set, and identify a target transformer fault type corresponding to the maximum target time-frequency confidence.

[0164] In an embodiment of the present invention, when the target time-frequency confidence set is a DC bias fault confidence of 2, a no-load open circuit fault confidence of 1, a phase-to-phase short circuit fault confidence of 3, a three-term unbalance fault confidence of 5, and a component loose fault confidence of 7, then the target transformer fault type is a component loose fault.

[0165] Furthermore, in the embodiment of the present invention, the transformer fault type can be identified through the intelligent voiceprint recognition platform. The intelligent voiceprint recognition platform can be found in Figure 4 shown.

[0166] In order to solve the problem described in the background technology, the present invention needs to first obtain the target sound time-frequency spectrum, then use different judgment methods to determine the transformer fault type on the target sound time-frequency spectrum, and then integrate various judgment results to calculate the target time-frequency confidence set, and finally identify the maximum target time-frequency confidence in the target time-frequency confidence set, and identify the target transformer fault type corresponding to the maximum target time-frequency confidence. When obtaining the target sound time-frequency spectrum, it is necessary to first use a pre-built microphone array to collect the original operating sound signal of the transformer, and then pre-process the original operating sound signal to obtain an initial operating sound signal. At this time, it is necessary to perform a short-time Fourier transform on the initial operating sound signal to obtain an initial sound time-frequency spectrum. Since frequency compression of different proportions is required, it is necessary to first segment the initial sound time-frequency spectrum according to preset low-frequency bands, medium-frequency bands and high-frequency bands, and then perform frequency compression to obtain the target sound time-frequency spectrum. At this time, the transformer fault type can be determined. First, it is necessary to extract the transformer fault types in the preset transformer fault type set in sequence. In the first judgment In the method, it is necessary to first obtain the average fault time spectrum of the transformer fault type, and then calculate the first time-frequency similarity between the target sound time spectrum and the average fault time spectrum according to the preset matching formula to obtain the first time-frequency similarity set. In the second judgment method, it is necessary to receive multiple groups of reference fault index value sets input by the user according to the transformer fault type, and then identify multiple groups of target fault index value sets of the target sound time spectrum, and finally calculate the second time-frequency similarity between the multiple groups of target fault index value sets and multiple groups of reference fault index value sets to obtain the second time-frequency similarity set. In the third judgment method, it is necessary to input the target sound time spectrum into the pre-constructed target neural network to obtain the third time-frequency probability value set. After completing the calculation of the three methods, it is necessary to calculate the target time-frequency confidence set according to the preset multi-scale weight set, the first time-frequency similarity set, the second time-frequency similarity set and the third time-frequency probability value set using the pre-constructed time-frequency similarity formula. At this time, the maximum target time-frequency confidence is identified in the target time-frequency confidence set, and the target transformer fault type corresponding to the maximum target time-frequency confidence is identified. Therefore, the present invention can improve the accuracy of transformer fault determination.

[0167] like Figure 5 , which is a functional module diagram of a photovoltaic power station transformer fault identification system based on multi-scale integration provided by an embodiment of the present invention.

[0168] The photovoltaic power station transformer fault identification system 100 based on multi-scale integration of the present invention can be installed in an electronic device. According to the functions implemented, the photovoltaic power station transformer fault identification system 100 based on multi-scale integration can include a titanium rod acquisition module 101 to be tested, a friction titanium rod map acquisition module 102, a high-temperature titanium rod map acquisition module 103 and an optimal titanium rod screening module 104. The module of the present invention can also be called a unit, which refers to a series of computer program segments that can be executed by an electronic device processor and can complete fixed functions, which are stored in the memory of the electronic device.

[0169] The target sound time-frequency spectrum acquisition module 101 is used to collect the original operation sound signal of the transformer by using a pre-built microphone array, wherein the collection process of the original operation sound signal includes: sound source localization, noise suppression, sound source enhancement and blind source separation, wherein the duration of the original operation sound signal is the signal fault identification duration; pre-processing the original operation sound signal to obtain an initial operation sound signal, wherein the pre-processing includes: pre-emphasis, framing and windowing; performing short-time Fourier transform on the initial operation sound signal to obtain an initial sound time-frequency spectrum, and frequency compressing the initial sound time-frequency spectrum according to preset low frequency band, medium frequency band and high frequency band to obtain a target sound time-frequency spectrum;

[0170] The time-frequency classification calculation module 102 is used to extract transformer fault types in a preset transformer fault type set in sequence, obtain an average fault time-frequency spectrum of the transformer fault type, calculate a first time-frequency similarity between the target sound time-frequency spectrum and the average fault time-frequency spectrum according to a preset matching formula, and obtain a first time-frequency similarity set; receive multiple groups of reference fault indicator value sets input by the user according to the transformer fault type, identify multiple groups of target fault indicator value sets of the target sound time-frequency spectrum, calculate a second time-frequency similarity between the multiple groups of target fault indicator value sets and multiple groups of reference fault indicator value sets, and obtain a second time-frequency similarity set, wherein the reference fault indicator set includes: reference fault waveform wavelength, reference fault waveform amplitude and reference fault waveform kurtosis; input the target sound time-frequency spectrum into a pre-constructed target neural network to obtain a third time-frequency probability value set, wherein the target neural network is a BP neural network;

[0171] The target time-frequency confidence set calculation module 103 is used to calculate the target time-frequency confidence set using a pre-constructed time-frequency similarity formula according to a preset multi-scale weight set, the first time-frequency similarity set, the second time-frequency similarity set and the third time-frequency probability value set:

[0172]

[0173] Among them, s iIndicates the target time-frequency confidence level of the fault type of the target sound time-frequency spectrum as the i-th transformer fault type, b 1 represents the first communion weight, represents the time-frequency similarity between the target sound time-frequency spectrum in the first time-frequency similarity set and the average fault time-frequency spectrum of the i-th transformer fault type, b 2 represents the second communion weight, represents the time-frequency similarity between the target fault index value set and the fault index value set of the i-th transformer fault type in the second time-frequency similarity set, b 3 represents the third communion weight, Indicates the time-frequency probability value of the fault type of the target sound time-frequency spectrum in the third time-frequency probability value set as the i-th transformer fault type;

[0174] The target transformer fault type identification module 104 is used to identify the maximum target time-frequency confidence in the target time-frequency confidence set, and identify the target transformer fault type corresponding to the maximum target time-frequency confidence.

[0175] In detail, each module in the photovoltaic power station transformer fault identification system 100 based on multi-scale integration in the embodiment of the present invention is used in the same manner as above. Figure 1 The photovoltaic power station transformer fault identification method based on multi-scale integration described in the present invention has the same technical means and can produce the same technical effects, so it will not be repeated here.

[0176] like Figure 6 , is a schematic diagram of the structure of an electronic device for implementing a photovoltaic power station transformer fault identification method based on multi-scale integration provided by an embodiment of the present invention.

[0177] The electronic device 1 may include a processor 10, a memory 11 and a bus 12, and may also include a computer program stored in the memory 11 and executable on the processor 10, such as a photovoltaic power station transformer fault identification method program based on multi-scale integration.

[0178] Among them, the memory 11 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, mobile hard disk, multimedia card, card-type memory (for example: SD or DX memory, etc.), magnetic memory, disk, optical disk, etc. In some embodiments, the memory 11 can be an internal storage unit of the electronic device 1, such as a mobile hard disk of the electronic device 1. In other embodiments, the memory 11 can also be an external storage device of the electronic device 1, such as a plug-in mobile hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (SecureDigital, SD) card, a flash card (Flash Card), etc. equipped on the electronic device 1. Further, the memory 11 also includes an internal storage unit of the electronic device 1 and an external storage device. The memory 11 can not only be used to store application software and various types of data installed in the electronic device 1, such as the code of the photovoltaic power station transformer fault identification method program based on multi-scale integration, but also can be used to temporarily store data that has been output or is to be output.

[0179] The processor 10 may be composed of an integrated circuit in some embodiments, for example, a single packaged integrated circuit, or a plurality of integrated circuits with the same or different functions, including one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and combinations of various control chips, etc. The processor 10 is the control core (Control Unit) of the electronic device, and uses various interfaces and lines to connect various components of the entire electronic device, and executes or executes programs or modules stored in the memory 11 (such as a photovoltaic power station transformer fault identification method program based on multi-scale integration, etc.), and calls data stored in the memory 11 to execute various functions of the electronic device 1 and process data.

[0180] The bus 12 may be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The bus 12 may be divided into an address bus, a data bus, a control bus, etc. The bus 12 is configured to realize connection and communication between the memory 11 and at least one processor 10, etc.

[0181] Figure 6 Only an electronic device with components is shown, and those skilled in the art will understand that Figure 6The structure shown does not constitute a limitation on the electronic device 1, and may include fewer or more components than shown in the figure, or combine certain components, or arrange the components differently.

[0182] For example, although not shown, the electronic device 1 may also include a power source (such as a battery) for supplying power to each component. Preferably, the power source may be logically connected to the at least one processor 10 through a power management device, so that the power management device can realize functions such as charging management, discharging management, and power consumption management. The power source may also include any components such as one or more DC or AC power sources, recharging devices, power failure detection circuits, power converters or inverters, power status indicators, etc. The electronic device 1 may also include a variety of sensors, Bluetooth modules, Wi-Fi modules, etc., which will not be repeated here.

[0183] Furthermore, the electronic device 1 may also include a network interface. Optionally, the network interface may include a wired interface and / or a wireless interface (such as a WI-FI interface, a Bluetooth interface, etc.), which is generally used to establish a communication connection between the electronic device 1 and other electronic devices.

[0184] Optionally, the electronic device 1 may further include a user interface, which may be a display, an input unit (such as a keyboard), or a standard wired interface or a wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, and an OLED (Organic Light-Emitting Diode) touch device. The display may also be appropriately referred to as a display screen or a display unit, which is used to display information processed in the electronic device 1 and to display a visual user interface.

[0185] The photovoltaic power station transformer fault identification method program based on multi-scale integration stored in the memory 11 of the electronic device 1 is a combination of multiple instructions. When running in the processor 10, it can achieve:

[0186] The original operating sound signal of the transformer is collected using a pre-built microphone array, wherein the collection process of the original operating sound signal includes: sound source localization, noise suppression, sound source enhancement and blind source separation, and the duration of the original operating sound signal is the signal fault identification duration;

[0187] Preprocessing the original running sound signal to obtain an initial running sound signal, wherein the preprocessing includes: pre-emphasis, framing and windowing;

[0188] Performing short-time Fourier transform on the initial running sound signal to obtain an initial sound time-frequency spectrum, and performing frequency compression on the initial sound time-frequency spectrum according to preset low-frequency band, medium-frequency band and high-frequency band to obtain a target sound time-frequency spectrum;

[0189] Extracting transformer fault types in sequence from a preset transformer fault type set, obtaining an average fault time-frequency spectrum of the transformer fault type, calculating a first time-frequency similarity between the target sound time-frequency spectrum and the average fault time-frequency spectrum according to a preset matching formula, and obtaining a first time-frequency similarity set;

[0190] Receiving multiple groups of reference fault indicator value sets input by a user according to the transformer fault type, identifying multiple groups of target fault indicator value sets of the target sound time-frequency spectrum, calculating the second time-frequency similarity between the multiple groups of target fault indicator value sets and the multiple groups of reference fault indicator value sets, and obtaining a second time-frequency similarity set, wherein the reference fault indicator set includes: reference fault waveform wavelength, reference fault waveform amplitude and reference fault waveform kurtosis;

[0191] Inputting the target sound time-frequency spectrum into a pre-constructed target neural network to obtain a third time-frequency probability value set, wherein the target neural network is a BP neural network;

[0192] According to the preset multi-scale weight set, the first time-frequency similarity set, the second time-frequency similarity set and the third time-frequency probability value set, the target time-frequency confidence set is calculated using the pre-constructed time-frequency similarity formula, and the time-frequency similarity formula is as follows:

[0193]

[0194] Among them, s i Indicates the target time-frequency confidence level of the fault type of the target sound time-frequency spectrum as the i-th transformer fault type, b 1 represents the first communion weight, represents the time-frequency similarity between the target sound time-frequency spectrum in the first time-frequency similarity set and the average fault time-frequency spectrum of the i-th transformer fault type, b 2 represents the second communion weight, represents the time-frequency similarity between the target fault index value set and the fault index value set of the i-th transformer fault type in the second time-frequency similarity set, b 3 represents the third communion weight, Indicates the time-frequency probability value of the fault type of the target sound time-frequency spectrum in the third time-frequency probability value set as the i-th transformer fault type;

[0195] A maximum target time-frequency confidence is identified in the target time-frequency confidence set, and a target transformer fault type corresponding to the maximum target time-frequency confidence is identified.

[0196] Specifically, the specific implementation method of the processor 10 for the above instructions can refer to Figures 1 to 6 The description of the relevant steps in the corresponding embodiments will not be repeated here.

[0197] Furthermore, if the module / unit integrated in the electronic device 1 is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. The computer-readable storage medium can be volatile or non-volatile. For example, the computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a mobile hard disk, a magnetic disk, an optical disk, a computer memory, and a read-only memory (ROM).

[0198] The present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor of an electronic device, the computer program can implement:

[0199] The original operating sound signal of the transformer is collected using a pre-built microphone array, wherein the collection process of the original operating sound signal includes: sound source localization, noise suppression, sound source enhancement and blind source separation, and the duration of the original operating sound signal is the signal fault identification duration;

[0200] Preprocessing the original running sound signal to obtain an initial running sound signal, wherein the preprocessing includes: pre-emphasis, framing and windowing;

[0201] Performing short-time Fourier transform on the initial running sound signal to obtain an initial sound time-frequency spectrum, and performing frequency compression on the initial sound time-frequency spectrum according to preset low-frequency band, medium-frequency band and high-frequency band to obtain a target sound time-frequency spectrum;

[0202] Extracting transformer fault types in sequence from a preset transformer fault type set, obtaining an average fault time-frequency spectrum of the transformer fault type, calculating a first time-frequency similarity between the target sound time-frequency spectrum and the average fault time-frequency spectrum according to a preset matching formula, and obtaining a first time-frequency similarity set;

[0203] Receiving multiple groups of reference fault indicator value sets input by a user according to the transformer fault type, identifying multiple groups of target fault indicator value sets of the target sound time-frequency spectrum, calculating the second time-frequency similarity between the multiple groups of target fault indicator value sets and the multiple groups of reference fault indicator value sets, and obtaining a second time-frequency similarity set, wherein the reference fault indicator set includes: reference fault waveform wavelength, reference fault waveform amplitude and reference fault waveform kurtosis;

[0204] Inputting the target sound time-frequency spectrum into a pre-constructed target neural network to obtain a third time-frequency probability value set, wherein the target neural network is a BP neural network;

[0205] According to the preset multi-scale weight set, the first time-frequency similarity set, the second time-frequency similarity set and the third time-frequency probability value set, the target time-frequency confidence set is calculated using the pre-constructed time-frequency similarity formula, and the time-frequency similarity formula is as follows:

[0206]

[0207] Among them, s i Indicates the target time-frequency confidence level of the fault type of the target sound time-frequency spectrum as the i-th transformer fault type, b 1 represents the first communion weight, represents the time-frequency similarity between the target sound time-frequency spectrum in the first time-frequency similarity set and the average fault time-frequency spectrum of the i-th transformer fault type, b 2 represents the second communion weight, represents the time-frequency similarity between the target fault index value set and the fault index value set of the i-th transformer fault type in the second time-frequency similarity set, b 3 represents the third communion weight, Indicates the time-frequency probability value of the fault type of the target sound time-frequency spectrum in the third time-frequency probability value set as the i-th transformer fault type;

[0208] A maximum target time-frequency confidence is identified in the target time-frequency confidence set, and a target transformer fault type corresponding to the maximum target time-frequency confidence is identified.

[0209] In the several embodiments provided by the present invention, it should be understood that the disclosed devices, systems and methods can be implemented in other ways. For example, the system embodiments described above are only illustrative, and actual implementation may have other division methods.

[0210] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0211] In addition, each functional module in each embodiment of the present invention may be integrated into one processing unit, each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of hardware plus software functional modules.

[0212] It is obvious to those skilled in the art that the present invention is not limited to the details of the above exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0213] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention.

Claims

1. A photovoltaic power station transformer fault identification method based on multi-scale integration, characterized in that: The method comprises: The original operating sound signal of the transformer is collected using a pre-built microphone array, wherein the collection process of the original operating sound signal includes: sound source localization, noise suppression, sound source enhancement and blind source separation, and the duration of the original operating sound signal is the signal fault identification duration; Preprocessing the original running sound signal to obtain an initial running sound signal, wherein the preprocessing includes: pre-emphasis, framing and windowing; Performing short-time Fourier transform on the initial running sound signal to obtain an initial sound time-frequency spectrum, and performing frequency compression on the initial sound time-frequency spectrum according to preset low-frequency band, medium-frequency band and high-frequency band to obtain a target sound time-frequency spectrum; Extracting transformer fault types in sequence from a preset transformer fault type set, obtaining an average fault time-frequency spectrum of the transformer fault type, calculating a first time-frequency similarity between the target sound time-frequency spectrum and the average fault time-frequency spectrum according to a preset matching formula, and obtaining a first time-frequency similarity set; Receiving multiple groups of reference fault indicator value sets input by a user according to the transformer fault type, identifying multiple groups of target fault indicator value sets of the target sound time-frequency spectrum, calculating the second time-frequency similarity between the multiple groups of target fault indicator value sets and the multiple groups of reference fault indicator value sets, and obtaining a second time-frequency similarity set, wherein the reference fault indicator set includes: reference fault waveform wavelength, reference fault waveform amplitude and reference fault waveform kurtosis; Inputting the target sound time-frequency spectrum into a pre-constructed target neural network to obtain a third time-frequency probability value set, wherein the target neural network is a BP neural network; According to the preset multi-scale weight set, the first time-frequency similarity set, the second time-frequency similarity set and the third time-frequency probability value set, the target time-frequency confidence set is calculated using the pre-constructed time-frequency similarity formula, and the time-frequency similarity formula is as follows: Among them, s i Indicates that the fault type of the target sound time-frequency spectrum is the target time-frequency confidence of the i-th transformer fault type, b1 represents the first fusion weight, represents the time-frequency similarity between the target sound time-frequency spectrum in the first time-frequency similarity set and the average fault time-frequency spectrum of the i-th transformer fault type, b2 represents the second fusion weight, represents the time-frequency similarity between the target fault index value set and the fault index value set of the i-th transformer fault type in the second time-frequency similarity set, b3 represents the third comprehension weight, Indicates the time-frequency probability value of the fault type of the target sound time-frequency spectrum in the third time-frequency probability value set as the i-th transformer fault type; A maximum target time-frequency confidence is identified in the target time-frequency confidence set, and a target transformer fault type corresponding to the maximum target time-frequency confidence is identified.

2. The photovoltaic power station transformer fault identification method based on multi-scale integration according to claim 1 is characterized in that: The frequency compression of the initial sound time-frequency spectrum according to the preset low frequency band, medium frequency band and high frequency band to obtain the target sound time-frequency spectrum includes: Segment the initial sound time spectrum according to the low frequency band, the middle frequency band and the high frequency band to obtain a low frequency band time spectrum, a middle frequency band time spectrum and a high frequency band time spectrum; According to the preset unit time zone, the low-frequency band time-frequency spectrum, the medium-frequency band time-frequency spectrum and the high-frequency band time-frequency spectrum are respectively segmented twice to obtain multiple groups of unit low-frequency band time-frequency sequences, multiple groups of unit medium-frequency band time-frequency sequences and multiple groups of unit high-frequency band time-frequency sequences; Performing maximum amplitude extraction in the multiple groups of low-frequency time-frequency sequences, the multiple groups of mid-frequency time-frequency sequences, and the multiple groups of high-frequency time-frequency sequences, respectively, to obtain multiple groups of low-frequency time-frequency point sequences, multiple groups of mid-frequency time-frequency point sequences, and multiple groups of high-frequency time-frequency point sequences; Performing time-frequency point fitting on the multiple groups of low-frequency time-frequency point sequences, the multiple groups of mid-frequency time-frequency point sequences, and the multiple groups of high-frequency time-frequency point sequences respectively to obtain a target low-frequency time-frequency spectrum, a target mid-frequency time-frequency spectrum, and a target high-frequency time-frequency spectrum; The target low-frequency band time-frequency spectrum, the target mid-frequency band time-frequency spectrum and the target high-frequency band time-frequency spectrum are frequency-connected to obtain a target sound time-frequency spectrum.

3. The photovoltaic power station transformer fault identification method based on multi-scale integration as claimed in claim 2 is characterized in that: The method of obtaining the average fault time spectrum of the transformer fault type includes: Acquire a fault time-frequency spectrum set of the transformer fault type; Extracting fault time spectra in the fault time spectrum set in sequence, receiving the clipping of the fault time spectrum by a user according to the signal fault identification duration and the signal fault identification frequency segment, and obtaining a time spectrum set to be averaged; Determine the signal fault identification time sequence and the signal fault identification frequency sequence according to the signal fault identification time length and the signal fault identification frequency segment, and construct a fault calibration dot matrix according to the signal fault identification time sequence and the signal fault identification frequency sequence; Extracting fault calibration points in the fault calibration point matrix in sequence, and identifying the corresponding to-be-averaged amplitude set in the to-be-averaged time spectrum set according to the fault calibration points, wherein the fault calibration points include: fault calibration frequency and fault calibration time; Calculating the average amplitude of the amplitude set to be averaged, and determining the three-dimensional fault time-frequency point according to the average amplitude and the fault calibration point; Summarize the three-dimensional fault time-frequency points of each fault calibration point to obtain a three-dimensional fault time-frequency point set; The three-dimensional fault time-frequency point set is fitted to obtain an average fault time-frequency spectrum.

4. The photovoltaic power station transformer fault identification method based on multi-scale integration as claimed in claim 3 is characterized in that: The matching formula is as follows: in, represents the first time-frequency similarity between the average fault time-frequency spectrum of the i-th transformer fault type and the target sound time-frequency spectrum, P represents the number of fault calibration points in the fault calibration point matrix, and f p Represents the signal amplitude of the pth fault calibration point in the target sound time spectrum, It represents the signal amplitude of the pth fault calibration point in the average fault time spectrum of the i-th transformer fault type, and |*| represents the absolute value sign.

5. The photovoltaic power station transformer fault identification method based on multi-scale integration as claimed in claim 4 is characterized in that: The multiple sets of target fault indicator value sets of the target sound frequency spectrum when identifying the target sound include: Extracting signal fault calibration frequencies in sequence from the signal fault identification frequency sequence, and identifying amplitude change signals of the signal fault calibration frequencies in the target sound time spectrum; Identifying a target waveform wavelength set, a target waveform amplitude set, and a target waveform kurtosis set of the amplitude variation signal; According to the target waveform wavelength set, the target waveform amplitude set and the target waveform kurtosis set, respectively calculate the target average waveform wavelength, the target average waveform amplitude and the target average waveform kurtosis to obtain the target fault index value set; The target fault index value sets of the fault calibration frequencies of each signal are collected to obtain multiple groups of target fault index value sets.

6. The photovoltaic power station transformer fault identification method based on multi-scale integration as claimed in claim 5 is characterized in that: The calculating the second time-frequency similarities of the multiple groups of target fault indicator value sets and the multiple groups of reference fault indicator value sets to obtain a second time-frequency similarity set includes: The second time-frequency similarity is calculated according to the multiple groups of target fault indicator value sets and the multiple groups of reference fault indicator value sets using the following formula: in, represents the second time-frequency similarity between multiple target fault index value sets and multiple reference fault index value sets of the i-th transformer fault type, k1 represents the wavelength coefficient, k2 represents the amplitude coefficient, k3 represents the kurtosis coefficient, Q represents the number of signal fault calibration frequencies in the signal fault identification frequency sequence, λ q represents the reference fault waveform wavelength in the qth group of reference fault index value sets, represents the target average waveform wavelength in the qth group of target fault index values, α q represents the reference fault waveform amplitude in the qth group of reference fault index value sets, represents the target average waveform amplitude in the qth group of target fault index values, β q represents the kurtosis of the reference fault waveform in the qth group of reference fault index values, represents the target average waveform kurtosis in the qth group of target fault index value sets; The second time-frequency similarities corresponding to each transformer fault type are collected to obtain a second time-frequency similarity set.

7. The photovoltaic power station transformer fault identification method based on multi-scale integration as claimed in claim 6 is characterized in that: The step of inputting the target sound time-frequency spectrum into a pre-constructed target neural network to obtain a third time-frequency probability value set includes: Extracting an amplitude array from the amplitude variation signal of the signal fault calibration frequency according to a preset amplitude sampling interval to obtain an amplitude array set corresponding to different signal calibration frequencies; The amplitude array set is input into the target neural network to obtain a third frequency similarity set, wherein the input layer of the target neural network includes: a first amplitude array input node, a second amplitude array input node, ..., an nth amplitude array input node, and the output layer of the target neural network includes: a first fault type probability value output node, a second fault type probability value output node, ..., an nth fault type probability value output node.

8. The photovoltaic power station transformer fault identification method based on multi-scale integration as claimed in claim 7 is characterized in that: The method for obtaining the multi-scale weight set includes: Extracting training sound time-frequency spectrum sets in sequence from a plurality of preset training sound time-frequency spectrum sets; Sequentially extracting the time spectra of training sounds from the time spectra set of training sounds; Calculate a first training similarity set of the training sound frequency spectrum according to the average fault frequency spectrum and the matching formula, identify a maximum first training similarity in the first training similarity set, identify a first predicted fault type corresponding to the maximum first training similarity, and obtain a first predicted fault type set; Identify multiple sets of training fault indicator value sets of the frequency spectrum of the training sound, calculate a second training similarity set according to the multiple sets of reference fault indicator value sets and the multiple sets of training fault indicator value sets, identify a maximum second training similarity in the second training similarity set, identify a second predicted fault type corresponding to the maximum second training similarity, and obtain a second predicted fault type set; Predicting a third training probability value set of the frequency spectrum of the training sound using the target neural network, identifying a maximum third training probability value in the third training probability value set, identifying a third predicted fault type corresponding to the maximum third training probability value, and obtaining a third predicted fault type set; Acquire the real fault type of the frequency spectrum of the training sound to obtain a real fault type set; Respectively calculating the sequence coincidence of the first predicted fault type set, the second predicted fault type set and the third predicted fault type set with the real fault type set to obtain a first sequence coincidence, a second sequence coincidence and a third sequence coincidence; Determining whether the first sequence coincidence degree, the second sequence coincidence degree, and the third sequence coincidence degree tend to be stable; If the first sequence overlap, the second sequence overlap and the third sequence overlap do not tend to be stable, returning to the above step of sequentially extracting training sound time spectrum sets from the preset multiple sets of training sound time spectrum sets; If the first sequence coincidence degree, the second sequence coincidence degree and the third sequence coincidence degree tend to be stable, the first sequence coincidence degree, the second sequence coincidence degree and the third sequence coincidence degree are normalized to obtain a multi-scale weight set.

9. The photovoltaic power station transformer fault identification method based on multi-scale integration as claimed in claim 8, characterized in that: The determining whether the first sequence coincidence degree, the second sequence coincidence degree, and the third sequence coincidence degree tend to be stable includes: Obtaining a first sequence coincidence degree sequence, a second sequence coincidence degree sequence, and a third sequence coincidence degree sequence; Draw a first sequence coincidence curve, a second sequence coincidence curve and a third sequence coincidence curve according to the first sequence coincidence sequence, the second sequence coincidence sequence and the third sequence coincidence sequence respectively; Identifying a first end slope, a second end slope and a third end slope of the first sequence coincidence curve, the second sequence coincidence curve and the third sequence coincidence curve; Determining whether the first terminal slope, the second terminal slope, and the third terminal slope are all less than a preset terminal slope threshold; If the first terminal slope, the second terminal slope and the third terminal slope are not all less than the terminal slope threshold, the first sequence coincidence degree, the second sequence coincidence degree and the third sequence coincidence degree are not stable; If the first terminal slope, the second terminal slope and the third terminal slope are all smaller than the terminal slope threshold, the first sequence coincidence degree, the second sequence coincidence degree and the third sequence coincidence degree tend to be stable.

10. A photovoltaic power station transformer fault identification system based on multi-scale integration, characterized in that: The system comprises: A target sound time-frequency spectrum acquisition module is used to collect the original operating sound signal of the transformer using a pre-built microphone array, wherein the collection process of the original operating sound signal includes: sound source positioning, noise suppression, sound source enhancement and blind source separation, wherein the duration of the original operating sound signal is the signal fault identification duration; pre-processing the original operating sound signal to obtain an initial operating sound signal, wherein the pre-processing includes: pre-emphasis, framing and windowing; performing short-time Fourier transform on the initial operating sound signal to obtain an initial sound time-frequency spectrum, and frequency compressing the initial sound time-frequency spectrum according to preset low-frequency bands, medium-frequency bands and high-frequency bands to obtain a target sound time-frequency spectrum; A time-frequency classification calculation module is used to extract transformer fault types in a preset transformer fault type set in sequence, obtain an average fault time-frequency spectrum of the transformer fault type, calculate a first time-frequency similarity between the target sound time-frequency spectrum and the average fault time-frequency spectrum according to a preset matching formula, and obtain a first time-frequency similarity set; receive multiple groups of reference fault indicator value sets input by a user according to the transformer fault type, identify multiple groups of target fault indicator value sets of the target sound time-frequency spectrum, calculate a second time-frequency similarity between the multiple groups of target fault indicator value sets and multiple groups of reference fault indicator value sets, and obtain a second time-frequency similarity set, wherein the reference fault indicator set includes: reference fault waveform wavelength, reference fault waveform amplitude and reference fault waveform kurtosis; input the target sound time-frequency spectrum into a pre-constructed target neural network to obtain a third time-frequency probability value set, wherein the target neural network is a BP neural network; The target time-frequency confidence set calculation module is used to calculate the target time-frequency confidence set using the pre-constructed time-frequency similarity formula according to the preset multi-scale weight set, the first time-frequency similarity set, the second time-frequency similarity set and the third time-frequency probability value set: Among them, s i Indicates that the fault type of the target sound time-frequency spectrum is the target time-frequency confidence of the i-th transformer fault type, b1 represents the first fusion weight, represents the time-frequency similarity between the target sound time-frequency spectrum in the first time-frequency similarity set and the average fault time-frequency spectrum of the i-th transformer fault type, b2 represents the second fusion weight, represents the time-frequency similarity between the target fault index value set and the fault index value set of the i-th transformer fault type in the second time-frequency similarity set, b3 represents the third comprehension weight, Indicates the time-frequency probability value of the fault type of the target sound time-frequency spectrum in the third time-frequency probability value set as the i-th transformer fault type; The target transformer fault type identification module is used to identify the maximum target time-frequency confidence in the target time-frequency confidence set, and identify the target transformer fault type corresponding to the maximum target time-frequency confidence.

Images