Monitoring method of switch unit in converter valve, electronic equipment and storage medium

Abstract

The invention discloses a monitoring method of a switch unit in a converter valve, electronic equipment and a storage medium. The method comprises the steps that an audio signal of a switch unit in the converter valve is collected, the converter valve comprises a plurality of sub-modules, and the switch unit is used for isolating a fault module in the plurality of sub-modules when the fault module occurs in the plurality of sub-modules; performing feature extraction on the audio signal to obtain voiceprint features in the audio signal; the voiceprint features are input into an audio recognition model, the audio recognition model is used for conducting state monitoring on the switch unit, state information of the switch unit is obtained, and the state information is used for representing whether the switch unit is in an abnormal state or not. The technical problem that the monitoring effect on the switch unit is poor in the related technology is solved.

Description

Technical Field

[0001] The present invention relates to the field of power electronic equipment status monitoring. Specifically, it relates to a monitoring method, an electronic device, and a storage medium for a switching unit in a converter valve. Background Technique

[0002] With the wide application of high-voltage direct current (HVDC) transmission technology and the increasing demand for reliability in power systems, the converter valve, as a key device in the DC transmission system, the health status of its sub-module bypass switch is directly related to the stable operation of the system. The bypass switch closes when a sub-module fails, bypassing the faulty module to ensure the continuous flow of current. However, the switch may encounter problems such as electromagnetic interference and poor contact during operation, resulting in changes in its operating sound, thereby affecting its normal function.

[0003] Traditional fault diagnosis methods, such as those based on manual inspection, vibration, and temperature monitoring, have limitations of detection lag, high false alarm rate, and high cost. Especially in large-scale power systems with complex noise environments and numerous monitoring points, related technologies are difficult to accurately diagnose the status of the bypass switch, and thus the monitoring effect of the switching unit in related technologies is poor.

[0004] In view of the above problems, no effective solution has been proposed yet. Summary of the Invention

[0005] Embodiments of the present invention provide a monitoring method, an electronic device, and a storage medium for a switching unit in a converter valve to at least solve the technical problem of poor monitoring effect of the switching unit in related technologies.

[0006] According to one aspect of the embodiments of the present invention, a monitoring method for a switching unit in a converter valve is provided, including: collecting an audio signal of the switching unit in the converter valve, where the converter valve includes multiple sub-modules, and the switching unit is used to isolate a faulty module among the multiple sub-modules when a faulty module appears; extracting features from the audio signal to obtain the voiceprint features in the audio signal; inputting the voiceprint features into an audio recognition model, and using the audio recognition model to monitor the status of the switching unit to obtain the status information of the switching unit, where the status information is used to indicate whether the switching unit is in an abnormal state.

[0007] Further, extracting features from the audio signal to obtain the voiceprint features in the audio signal includes: enhancing the high-frequency part of the audio signal to obtain an enhanced audio signal; extracting features from the enhanced audio signal to obtain the voiceprint features.

[0008] Further, feature extraction is performed on the enhanced audio signal to obtain voiceprint features, including: segmenting the enhanced audio signal to obtain multiple audio signal frames, where there is an overlap between multiple audio signal frames; converting the multiple audio signal frames to the frequency domain to obtain a target frequency spectrum; filtering the target frequency spectrum to obtain the signal energy of the target frequency spectrum; and determining the voiceprint features in the enhanced audio signal based on the signal energy.

[0009] Further, determining the voiceprint features in the enhanced audio signal based on the signal energy includes: processing the logarithm of the signal energy using discrete cosine transform to obtain mel-frequency cepstral coefficients; and performing mean square error normalization on the mel-frequency cepstral coefficients to obtain the voiceprint features.

[0010] Further, collecting the audio signal of the switching unit in the converter valve includes: capturing the initial audio signal of the switching unit based on the acoustic wave sensor array in the converter valve; and performing spatial filtering on the initial audio signal based on the position information of the switching unit to obtain the audio signal.

[0011] Further, performing spatial filtering on the initial audio signal based on the position information of the switching unit to obtain the audio signal includes: performing spatial filtering on the initial audio signal based on the position information to obtain a filtered audio signal; collecting the filtered audio signal based on a preset sampling frequency to obtain a collected audio signal; and intercepting the collected audio signal based on a preset duration to obtain the audio signal.

[0012] Further, the method further includes: obtaining the sample audio signal of the sample switching unit and the sample status information corresponding to the sample audio signal; using the initial audio recognition model to recognize the sample audio signal to obtain the predicted sample status information; constructing a loss function based on the sample status information and the predicted sample status information; and adjusting the model parameters of the initial audio recognition model using the loss function to obtain the audio recognition model.

[0013] Further, the method further includes: in response to the status information indicating that the switching unit is in an abnormal state, outputting a prompt message based on a preset prompt method, where the prompt message is used to prompt that the switching unit is in an abnormal state.

[0014] According to another aspect of the embodiments of the present invention, there is also provided a monitoring device for a switching unit in a converter valve, including: an acquisition module, configured to acquire an audio signal of the switching unit in the converter valve, wherein the converter valve includes a plurality of sub-modules, and the switching unit is configured to isolate a faulty module among the plurality of sub-modules when a faulty module appears in the plurality of sub-modules; a feature extraction module, configured to extract features from the audio signal to obtain a voiceprint feature in the audio signal; a monitoring module, configured to input the voiceprint feature into an audio recognition model, and use the audio recognition model to monitor the state of the switching unit to obtain state information of the switching unit, wherein the state information is used to indicate whether the switching unit is in an abnormal state.

[0015] According to another aspect of the embodiments of the present invention, there is also provided an electronic device, including: a memory storing an executable program; a processor configured to run the program, wherein when the program runs, it executes the above-mentioned monitoring method for the switching unit in the converter valve.

[0016] According to another aspect of the embodiments of the present invention, there is also provided a computer-readable storage medium, which includes a stored executable program, wherein when the executable program runs, it controls the device where the computer-readable storage medium is located to execute the above-mentioned monitoring method for the switching unit in the converter valve.

[0017] In the embodiments of the present invention, first, an audio signal of the switching unit in the converter valve is acquired, then features are extracted from the audio signal to obtain a voiceprint feature in the audio signal, and finally the voiceprint feature is input into an audio recognition model, and the audio recognition model is used to monitor the state of the switching unit to obtain state information of the switching unit. It is easy to note that the converter valve includes a plurality of sub-modules, and the switching unit is configured to isolate a faulty module among the plurality of sub-modules in the converter valve when a faulty module appears. The present application identifies the audio signal generated by the switching unit when isolating the faulty module through the audio recognition model to determine whether the switching unit is in an abnormal state when performing the faulty module isolation operation, achieving the purpose of real-time monitoring of the switching unit in a non-contact manner, improving the real-time performance and accuracy of the monitoring of the switching unit, and thus solving the technical problem of poor monitoring effect of the switching unit in the related art. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The drawings described herein are used to provide a further understanding of the present invention, form a part of this application, and the schematic embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation to the present invention. In the drawings:

[0019] Figure 1 is a flowchart of a monitoring method for a switching unit in a converter valve according to an embodiment of the present invention;

[0020] Figure 2It is a flowchart of a monitoring method for a switching unit in a commutation valve according to an embodiment of the present invention;

[0021] Figure 3 It is a schematic diagram of a monitoring device for a switching unit in a commutation valve according to an embodiment of the present invention. Detailed implementation manners

[0022] In order to enable those skilled in the art to better understand the solution of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative work shall fall within the protection scope of the present invention.

[0023] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily need to be used to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments of the present invention described here can be implemented in an order different from those illustrated or described here. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device including a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0024] According to an embodiment of the present invention, an embodiment of a monitoring method for a switching unit in a commutation valve is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order from that here.

[0025] Figure 1 It is a flowchart of a monitoring method for a switching unit in a commutation valve according to an embodiment of the present invention. As Figure 1 shown, the method includes the following steps:

[0026] Step S102, collect the audio signal of the switching unit in the commutation valve, where the commutation valve includes a plurality of sub-modules, and the switching unit is used to isolate the faulty module among the plurality of sub-modules when a faulty module appears in the plurality of sub-modules.

[0027] The above-mentioned converter valve can refer to the core equipment in a high-voltage direct current (HVDC) transmission system, which is used to realize the conversion between direct current and alternating current. The converter valve consists of a large number of sub-modules. By controlling the on-off of these sub-modules, efficient conversion of electrical energy is achieved. The types of converter valves can include, but are not limited to, thyristor converter valves and voltage-source converter valves. The specific type of converter valve can be determined according to the actual system design and is not limited here.

[0028] The above-mentioned switching unit can refer to the switch in the sub-module of the converter valve. The switching unit can refer to a bypass switch. When a certain component in the sub-module of the converter valve fails, the switching unit can isolate the faulty component from the circuit, allowing the current to bypass the faulty component, thereby improving the fault tolerance of the converter valve, ensuring the continuous transmission of direct current, reducing the system downtime, and enhancing the overall reliability and efficiency of the system.

[0029] The above-mentioned audio signal can refer to the electrical signal version of the sound signal. Specifically, it can refer to the electrical signal converted from the sound generated during the operation of the switching unit by a microphone. The acquisition of the audio signal can be completed by high-sensitivity microphones installed near the switching unit. These microphones can capture the minute sound changes during the switch operation. When acquiring, an appropriate sampling frequency needs to be set to ensure the integrity of the signal and the capture of sound details. At the same time, digital signal processing technology is used to convert the sound signal into an audio signal for subsequent analysis and processing.

[0030] The above-mentioned sub-module can refer to the basic unit that constitutes the converter valve. Each sub-module contains one or more capacitors, one or more power semiconductor switches, and other auxiliary components such as resistors and inductors. The types of sub-modules can include, but are not limited to, half-bridge sub-modules, full-bridge sub-modules, and multi-level sub-modules, etc. The specific sub-module needs to be determined according to the switching elements and topological structure inside the converter valve and is not limited here. The design of the sub-module enables the converter valve to work in high-voltage and high-current environments and at the same time has certain self-protection and fault self-healing capabilities.

[0031] In an alternative embodiment, the converter valve includes a plurality of sub-modules, and each sub-module is equipped with a bypass switch unit. When a fault is detected in a sub-module, the switch unit, i.e., the bypass switch, can quickly close to isolate the faulty module and ensure that the current continues to flow through other healthy sub-modules. At the same time, the audio signals of the switch units in the converter valve are collected by an acoustic sensor array deployed around the switch units. The acoustic sensor array can capture the audio information during the switching operation in all directions and reduce the interference of background noise through spatial filtering technology, thereby ensuring the clarity and integrity of the signals. In the above audio signal collection process, the non-contact and real-time nature of the sound signals are utilized to capture abnormal voiceprints at the first moment of a fault occurrence, providing immediate fault information for the operation and maintenance personnel, reducing the delay in fault detection, and improving the response speed of the system.

[0032] Step S104: Extract features from the audio signals to obtain the voiceprint features in the audio signals.

[0033] The above voiceprint features may refer to feature vectors that can uniquely represent or significantly distinguish a sound signal or a speech sample. The voiceprint features are extracted from the original audio signals through a series of mathematical and signal processing steps, and can reflect the physical attributes and semantic characteristics of the sound. The types of voiceprint features may include, but are not limited to, Mel Frequency Cepstral Coefficients (MFCC for short), Zero Crossing Rate (ZCR for short), Short-time Energy (STE for short), spectral features, zero crossing rate, and energy entropy, etc. The specific voiceprint features need to be determined according to actual requirements and are not limited here. The voiceprint features can be used to identify the abnormal state of the bypass switch and detect potential faults in a timely manner, thereby ensuring the safe operation of the power system.

[0034] In an alternative embodiment, first, the audio signal is preprocessed using preprocessing techniques such as noise reduction, pre-emphasis, and framing to obtain an audio signal that can better reflect the voiceprint characteristics. Then, the audio signal is transformed from the time domain to the frequency domain through the fast Fourier transform, and a set of triangular filters are applied to filter the spectrum on the Mel scale. The output energy of each filter is calculated to construct the features output by the Mel filter bank. Finally, the Mel frequency cepstral coefficients are calculated to obtain the required voiceprint features. The above process is the extraction process of the voiceprint feature Mel frequency cepstral coefficients. In addition to the Mel frequency cepstral coefficients, the voiceprint characteristics that can significantly distinguish the bypass switch in the normal and abnormal states are also extracted from the audio signal, including the zero-crossing rate, short-time energy, and spectral features, etc. These voiceprint features can not only accurately characterize the spectral distribution and time-domain characteristics of the sound, but also have good noise robustness, ensuring the stability and discriminability of the features even under complex noise conditions in the converter valve operating environment.

[0035] Step S106, input the voiceprint features into the audio recognition model, and use the audio recognition model to monitor the state of the switch unit to obtain the state information of the switch unit, where the state information is used to indicate whether the switch unit is in an abnormal state.

[0036] The above audio recognition model can refer to a model used to identify specific audio patterns or states from voiceprint features. The types of audio recognition models can include but are not limited to the minimum risk Bayesian classifier, support vector machine, convolutional neural network, and hidden Markov model, etc. The specific audio recognition model needs to be determined according to the actual system design and is not limited here. The audio recognition model is the core of the entire anomaly detection system and can be used to realize the mapping between voiceprint features and the state of the switch unit, analyze the collected audio signal in real time, and timely feedback the operating state of the switch unit.

[0037] The above state information can refer to the data or metrics describing the current operating condition of the switch unit based on the output result of the audio recognition model. The types of state information can include that the switch unit is in the normal state and the switch unit is in the abnormal state. The specific state information needs to be determined according to the actual recognition result and is not limited here. The state information is an important basis for the maintenance personnel to carry out fault diagnosis and maintenance decisions. It can help the maintenance personnel quickly locate the problem, take appropriate measures, and avoid or reduce system downtime or losses caused by switch unit failures.

[0038] The above abnormal state may refer to the abnormal working conditions that occur during the operation of the switch unit, that is, when the bypass switch of the converter valve is in the operating or stationary state, its acoustic fingerprint characteristics deviate from the expected pattern or range during normal operation. This deviation may be caused by internal wear, poor contact, electromagnetic interference or other faults in the switch. During normal operation, the operating sound of the bypass switch usually has consistency and predictability. When the switch is abnormal, such as additional noise, operation delay or incomplete closing during the isolation process of the faulty module, its acoustic fingerprint characteristics will change significantly, and these changes constitute the acoustic representation of the abnormal state. Therefore, identifying the abnormal state of the switch unit is crucial for the safe operation of the power system, which can enable the system to give early warnings of faults, reduce unplanned outages, and improve the reliability and efficiency of power transmission.

[0039] In an alternative embodiment, an audio recognition model is used to learn and memorize the acoustic fingerprint characteristics in the normal state. When the acoustic fingerprint characteristics obtained by real-time acquisition and extraction do not match the normal pattern in the model, that is, when an abnormal state is identified, the system will immediately trigger an alarm mechanism to notify the operation and maintenance personnel to conduct inspections and maintenance. Compared with traditional monitoring methods based on physical parameters (such as temperature, vibration), the acoustic fingerprint recognition technology has higher sensitivity and specificity, and can detect initial subtle abnormalities, so as to take preventive measures before the fault develops into a major problem. This method not only reduces the dependence on manual inspections, but also improves the real-time performance and accuracy of fault detection, thereby improving the overall operation and maintenance level and safety performance of the power system, and providing new technical support for the intelligent operation and maintenance of the power system.

[0040] In the embodiment of the present invention, first, the audio signal of the switch unit in the converter valve is collected, then the acoustic fingerprint characteristics in the audio signal are extracted from the audio signal, and finally the acoustic fingerprint characteristics are input into the audio recognition model. The audio recognition model is used to monitor the state of the switch unit to obtain the state information of the switch unit. It is easy to note that the converter valve contains multiple sub-modules, and the switch unit is used to isolate the faulty module in the multiple sub-modules in the converter valve when a faulty module appears. In this application, the audio recognition model is used to identify the audio signal generated when the switch unit isolates the faulty module, so as to judge whether the switch unit is in an abnormal state when performing the faulty module isolation operation, achieving the purpose of real-time monitoring of the switch unit without contact, improving the real-time performance and accuracy of the switch unit monitoring, and further solving the technical problem of poor monitoring effect of the switch unit in the related art.

[0041] Optionally, extracting the acoustic fingerprint characteristics from the audio signal includes: enhancing the high-frequency part of the audio signal to obtain an enhanced audio signal; extracting the acoustic fingerprint characteristics from the enhanced audio signal.

[0042] The above high-frequency part may refer to the components with relatively high frequencies in the switching operation sound, which can reflect information such as the characteristics and contact status of the switch and are crucial for identifying abnormal states.

[0043] The above enhanced audio signal may refer to the audio signal obtained after enhancing the intensity and clarity of the high-frequency part in the audio signal through preprocessing techniques. In the abnormal identification of the bypass switch soundprint of the converter valve, the enhanced audio signal can ensure that the soundprint feature extraction algorithm obtains clearer and more accurate sound detail information, which is very crucial for distinguishing the soundprint features in normal and abnormal states. The above preprocessing technique may refer to a pre-emphasis filter. The pre-emphasis filter compensates for the natural attenuation of high-frequency components during transmission by increasing the high-frequency gain in the signal spectrum. The formula for pre-emphasis is:

[0044] y(n) = x(n) - αx(n - 1);

[0045] In the formula, y(n) represents the enhanced signal, x(n) represents the original signal, x(n - 1) represents the value of the original signal at the previous moment, n represents the discrete time points in the signal, and α represents the pre-emphasis factor. Usually, the value of α is 0.97. This value can balance the signal enhancement effect and the complexity of the algorithm. The value of α here is only for example, and the specific α can be determined according to the actual situation and is not limited here.

[0046] In an optional embodiment, enhancing the high-frequency part through preprocessing techniques such as pre-emphasis can effectively improve the signal-to-noise ratio and ensure that the subtle but important high-frequency details in the captured switching operation sound are not masked by low-frequency noise; then inputting the enhanced audio signal into the feature extraction algorithm, such as Mel-frequency cepstral coefficient calculation, can generate more discriminative soundprint features. Even in a complex power system operation environment, it can stably identify the normal operation and abnormal states of the bypass switch. The application of this technology not only improves the accuracy and robustness of soundprint abnormal identification but also realizes the real-time monitoring of the bypass switch state, significantly improving the operation and maintenance efficiency and safety performance of the high-voltage direct current transmission system.

[0047] Optionally, feature extraction is performed on the enhanced audio signal to obtain the soundprint features, including: segmenting the enhanced audio signal to obtain multiple audio signal frames, where there is an overlap between multiple audio signal frames; converting multiple audio signal frames to the frequency domain to obtain the target spectrum; filtering the target spectrum to obtain the signal energy of the target spectrum; and determining the soundprint features in the enhanced audio signal based on the signal energy.

[0048] The above-mentioned audio signal frames can refer to dividing continuous audio signals into a series of short-time segments of a fixed length. The frame length of these segments is usually around several tens of milliseconds (such as 20 - 30 ms), and there is partial overlap between adjacent frames. The overlap degree can be set to about 50%. The above parameter settings are only for example, and the specific size and overlap degree of the audio signal frames can be adjusted based on the characteristics of the signal and the analysis requirements, which are not limited here. Frame analysis can capture the characteristics of audio signals within a short time window and is particularly important for analyzing transient sounds, such as switch operation sounds. The use of overlapping frames can smooth the changes between frames, avoid information loss and sudden changes in processing, thereby improving the continuity and robustness of feature extraction.

[0049] The above-mentioned target spectrum can refer to the spectral representation obtained after converting the audio signal frames into the frequency domain, which can be completed through fast Fourier transform. The types of target spectra can include but are not limited to amplitude spectra, power spectra, and logarithmic forms of amplitude spectra and power spectra. The specific target spectrum needs to be determined according to actual needs, which are not limited here. The target spectrum can reveal the energy distribution of different frequency components in the signal and provide a basis for subsequent feature extraction.

[0050] The above-mentioned signal energy can refer to the total energy of the signal at a certain frequency or frequency band in the frequency domain, that is, the energy output by a specific filter in the spectrum. Signal energy is a key parameter in voiceprint analysis, which can reflect the intensity of sound in different frequency bands, thereby distinguishing the voiceprints of normal operation sounds and abnormal operation sounds.

[0051] The above-mentioned segmentation can refer to dividing continuous audio signals into multiple short-time segments, that is, audio signal frames. The segmentation types can include but are not limited to segmentation using a fixed window size and a fixed frame shift, or dynamic segmentation according to the signal characteristics. The specific segmentation method needs to be determined according to the signal type and actual needs, which are not limited here. Segmentation can make feature extraction more focused on the short-time characteristics of the signal, avoid the characteristics of each part in the long-time signal from being averaged, and thereby improve the accuracy of feature expression.

[0052] The above-mentioned filtering can refer to processing the signal in the frequency domain using a filter bank to highlight or eliminate certain frequency components. The filter types can include but are not limited to Mel filters, Gaussian filters, Butterworth filters, etc. The specific filter type needs to be determined according to the filtering requirements and signal type, which are not limited here. Filtering can be used to improve the signal-to-noise ratio of the signal, emphasize the target frequency components, and eliminate irrelevant noise, thereby enhancing the clarity and discrimination ability of voiceprint features.

[0053] In an alternative embodiment, first, the enhanced audio signal is segmented into multiple short-time signal frames with a certain overlap between each frame; then these audio signal frames are transformed into the frequency domain using the fast Fourier transform to obtain their target spectra; subsequently, the target spectra are filtered by applying a Mel filter bank, and the signal energy within each filter band is calculated to highlight the key frequency components related to human ear perception in the audio signal, further enhancing the robustness and distinctiveness of the features; finally, the voiceprint features in the enhanced audio signal are determined based on the extracted signal energy. The feature extraction process here not only includes calculating the Mel frequency cepstral coefficients but also includes calculating the first-order and second-order differences of the signal energy. Through the above process, by converting the enhanced audio signal into a frequency-domain representation and through filtering and energy calculation, the voiceprint features reflecting the bypass switch state can be effectively extracted, providing high-quality input data for subsequent classification and anomaly recognition, not only improving the expressiveness of the features but also enhancing the robustness of the recognition algorithm to noise, thus significantly improving the accuracy, real-time performance, and reliability of the converter valve system state monitoring.

[0054] Optionally, determining the voiceprint features in the enhanced audio signal based on the signal energy includes: processing the logarithm of the signal energy using the discrete cosine transform to obtain the frequency cepstral coefficients; performing mean square error normalization on the frequency cepstral coefficients to obtain the voiceprint features.

[0055] The above discrete cosine transform can refer to a mathematical tool for transforming a signal from the spatial or time domain to the frequency-related domain. The discrete cosine transform can be used to convert the logarithmic energy spectrum into frequency cepstral coefficients, and this conversion helps to remove the redundant information of the signal and retain the features crucial for voiceprint recognition.

[0056] The above frequency cepstral coefficients can refer to a set of coefficients extracted from the logarithmic energy spectrum. The frequency cepstral coefficients can be used to reflect the time-frequency characteristics of the sound signal, thereby distinguishing different voiceprints.

[0057] The above mean square error normalization can refer to a data preprocessing technique. The types of mean square error normalization can include but are not limited to frame-level mean square error normalization, sample-level normalization, feature-level normalization, etc. The specific mean square error normalization needs to be determined according to the characteristics of the signal to be processed, which is not limited here. Mean square error normalization can eliminate the scale differences of the feature vectors in different audio samples, improve the stability of model training, reduce the risk of overfitting, and enable the classifier to learn and recognize voiceprint features more accurately.

[0058] In an alternative embodiment, the logarithm of the signal energy is processed by using the discrete cosine transform to obtain the cepstral coefficients, and then the mean square error normalization is performed on these coefficients. Finally, the normalized voiceprint features are obtained. The normalized voiceprint features can ensure that the model is more stable during training and operation, reduce false alarms, and accurately identify the abnormal state of the bypass switch in a timely manner, thereby effectively ensuring the operation safety of power electronic equipment, reducing system failures, and improving the overall operation and maintenance efficiency and the reliability of the power system.

[0059] In an alternative embodiment, based on the signal energy after Mel filtering, the Mel-frequency cepstral coefficients are calculated. This step digitizes the key voiceprint features to form a series of numerical values. Then, the calculated Mel-frequency cepstral coefficients are combined into a feature vector, and the formed feature vector is directly input into the classifier for training or real-time monitoring. Finally, the constructed feature vector is normalized to improve the consistency of its statistical characteristics, thereby enhancing the training effect of the model and the accuracy of recognition.

[0060] Optionally, the audio signals of the switching units in the converter valve are collected, including: capturing the initial audio signals of the switching units based on the acoustic wave sensor array in the converter valve; and performing spatial filtering on the initial audio signals based on the position information of the switching units to obtain the audio signals.

[0061] The above-mentioned acoustic wave sensor array may refer to a set formed by arranging multiple acoustic wave sensors in a specific geometric layout, which can be used to capture acoustic wave signals at different positions and directions in space. In the monitoring of the converter valve, the sensor array can be used to comprehensively capture the sounds during the operation of the switching units. The layout types of the acoustic wave sensor array may include, but are not limited to, linear arrays, circular arrays, grid arrays, etc. The specific layout of the acoustic wave sensor array needs to be determined according to the size, shape of the monitoring area and the expected position of the sound source, which is not limited here. The acoustic wave sensor array can be used to enhance the target signal while reducing background noise through the collaborative work of multiple sensors, thereby improving the signal capture quality.

[0062] The above-mentioned initial audio signals may refer to the unprocessed sound signals directly captured from the acoustic wave sensor array, which contain the original sound information during the operation of the switching units in the converter valve. The initial audio signals are the starting point of acoustic monitoring and analysis, providing basic information for subsequent signal processing and feature extraction, and are the key data source for identifying the state of the switching units.

[0063] The above position information may refer to the specific position of each sensor in the acoustic wave sensor array and its relative position with respect to the switching unit in the converter valve. The position information can be directly obtained through the design of the sensor array or can be dynamically determined by additional positioning algorithms such as triangulation, multi-point positioning, etc. to roughly locate the sound source. Performing spatial filtering based on the position information can specifically enhance the target signal while suppressing noise from other directions, thereby improving the quality and signal-to-noise ratio of the collected audio signal.

[0064] The above spatial filtering may refer to a signal processing technique. Spatial filtering utilizes the layout and position information of the acoustic wave sensor array, and through algorithms, enhances the signals emitted from specific directions or positions while suppressing noise interference from other directions. The types of spatial filtering can include but are not limited to beamforming, adaptive filtering, delay and sum filtering, etc. The specific spatial filtering method needs to be determined according to actual requirements and is not limited here. The spatial filtering technology can significantly improve the signal capture quality in a complex noise environment, making the audio signal collected from the converter valve switching unit clearer and more representative, and further improving the accuracy and reliability of subsequent voiceprint feature extraction.

[0065] In an optional embodiment, an initial audio signal is obtained through the acoustic wave sensor array deployed in the converter valve, and then the initial audio signal is processed by spatial filtering to obtain an audio signal. This effectively reduces the influence of background noise during the audio signal acquisition process of the switching unit in the converter valve, ensuring that the obtained audio signal has high quality and good signal-to-noise ratio, providing a solid data basis for subsequent voiceprint feature extraction and abnormal state recognition, improving the accuracy and real-time performance of power electronic device condition monitoring, and thus effectively enhancing the operation and maintenance efficiency and operation reliability of the power system.

[0066] Optionally, performing spatial filtering on the initial audio signal based on the position information of the switching unit to obtain an audio signal includes: performing spatial filtering on the initial audio signal based on the position information to obtain a filtered audio signal; collecting the filtered audio signal based on a preset sampling frequency to obtain a collected audio signal; and intercepting the collected audio signal based on a preset duration to obtain an audio signal.

[0067] The above filtered audio signal may refer to the signal obtained after performing spatial filtering processing on the initial audio signal.

[0068] The above-mentioned preset sampling frequency may refer to the sampling frequency preset according to the highest frequency component of the signal and the sampling theorem during the audio signal acquisition process. The sampling frequency may include but is not limited to 44.1 kHz, 48 kHz, etc. The specific preset sampling frequency needs to be determined according to the frequency range of the signal and the application scenario, which is not limited here. The preset sampling frequency ensures the integrity of the audio signal during the digitization process, avoids the frequency aliasing phenomenon, and provides a high-quality digital signal basis for subsequent signal processing and analysis.

[0069] The above-mentioned acquired audio signal may refer to the digital audio stream continuously or periodically acquired from the acoustic wave sensor array at the preset sampling frequency, which reflects the real-time sound state of the switching unit in the converter valve.

[0070] The above-mentioned preset duration may refer to the signal duration set during the acquisition of the audio signal, which is used to intercept the audio segment containing sufficient acoustic information for analysis. The preset duration may include but is not limited to several seconds, dozens of seconds, etc. The specific preset duration needs to be determined based on the stability of the signal characteristics and the analysis requirements, which is not limited here. The interception of the preset duration ensures that each analyzed signal segment contains complete and coherent acoustic information, facilitating the stable operation of the subsequent feature extraction algorithm and the accurate extraction of voiceprint features.

[0071] In an alternative embodiment, spatial filtering is performed by utilizing the position information of the acoustic wave sensor array; subsequently, the filtered audio signal is acquired based on the preset high sampling frequency; finally, the acquired audio signal is accurately intercepted according to the preset duration, which not only ensures that each analyzed segment contains sufficient acoustic information for feature extraction, but also takes into account the processing efficiency and storage requirements, making the signal processing flow more efficient. This process not only significantly improves the quality of the audio signal, but also provides good data preparation for constructing an accurate and robust voiceprint anomaly recognition system, thereby significantly improving the accuracy and real-time performance of the bypass switch state monitoring in the converter valve and enhancing the stability and security of the power system operation.

[0072] Optionally, the method further includes: obtaining the sample audio signal of the sample switching unit and the sample state information corresponding to the sample audio signal; using the initial audio recognition model to recognize the sample audio signal to obtain the predicted sample state information; constructing a loss function based on the sample state information and the predicted sample state information; and using the loss function to adjust the model parameters of the initial audio recognition model to obtain the audio recognition model.

[0073] The above sample switch units can refer to specific bypass switch instances of converter valve sub - modules used for training and testing audio recognition models. They are recorded and data is collected under different operating states. The sample switch units can be used to provide data in the actual operating environment for training the model to learn voiceprint features in normal and abnormal states, as well as for performance evaluation and parameter optimization of the model.

[0074] The above sample audio signals can refer to the audio data captured by the acoustic wave sensor array and pre - processed when the sample switch unit makes a sound in a specific state. The types of sample audio signals can include, but are not limited to, signals in normal states, slightly abnormal states, or severely abnormal states. The specific sample audio signals need to be determined according to the actual situation and are not limited here. The sample audio signals can be used as the direct data source for training the audio recognition model, and the model learns the features in these signals to distinguish switches in different states.

[0075] The above sample status information can refer to the actual status of the switch unit corresponding to the sample audio signal. The sample status information can include, but is not limited to, normal status, slightly faulty status, severely faulty status, etc. The specific sample status information needs to be determined according to the actual status of the switch unit and is not limited here. The sample status information can be used as a label for supervised learning to evaluate the difference between the prediction result of the model and the actual situation, and to guide the adjustment of model parameters to improve the recognition accuracy.

[0076] The above predicted sample status information can refer to the prediction result of the model about the status of the sample switch unit after analyzing the given sample audio signal through the initial audio recognition model. The types of predicted sample status information can include, but are not limited to, normal status, slightly faulty status, severely faulty status, etc. The specific predicted sample status information needs to be determined according to the actual analysis situation and is not limited here. The predicted sample status information can be used to evaluate the accuracy, recognition rate, and robustness of the model. During the training process, the model adjusts its parameters by continuously reducing the difference (i.e., the loss function) between the prediction result and the actual status to improve the recognition performance.

[0077] The above loss function can refer to a mathematical expression that measures the difference between the model's prediction result and the actual sample status information. The types of loss functions can include, but are not limited to, mean squared error, cross - entropy loss, etc. The specific loss function needs to be determined according to the actual situation and is not limited here. The loss function is the core of model training. By adjusting the model parameters, the value of the loss function is made as small as possible, thereby improving the prediction performance and reliability of the model.

[0078] The above-mentioned initial audio recognition model may refer to the initial state of the model architecture and parameters set before the start of training. The types of the initial audio recognition model may include, but are not limited to, Bayesian classifiers, support vector machines, and convolutional neural networks. The specific initial audio recognition model needs to be determined according to actual requirements and is not limited here. The initial audio recognition model is the starting point of the training process. It gradually adjusts its parameters by learning the features and state information in the sample dataset to achieve more accurate state recognition.

[0079] In an alternative embodiment, by obtaining the audio signals of the sample switch unit in different states and their corresponding true state information, using the initial audio recognition model to identify and analyze these sample audio signals, and then obtaining the predicted switch state information of the model; then, using the loss function to quantify the difference between the predicted state information and the true sample state information to guide the adjustment of the model parameters, ensuring that the model learns the voiceprint features that can best distinguish normal and abnormal states, and obtaining the required audio recognition model. Through the continuous iteration and optimization in the training process, the above process enables the model to recognize the subtle voiceprint changes of the bypass switch, timely detect potential fault states, and thus provide key information support for the maintenance and safe operation of the converter valve system.

[0080] In an alternative embodiment, the audio recognition model may be a Bayesian classifier. The construction process of the audio recognition model is as follows:

[0081] First, collect a sufficient number of normal and abnormal sound samples of the bypass switch and label them. Divide the dataset into a training set, a validation set, and a test set for model training and evaluation.

[0082] Then, define the loss matrix L. The loss matrix is used to determine the cost or loss brought by each possible classification decision. The cost L(i, j) when the predicted class is j and the actual class is i. Usually, L(i, j) = 0, indicating no loss for correct classification.

[0083] Next, estimate the class-conditional probability density function P(x|Ci) of each class from the training dataset through the naive Bayesian classifier, calculate the prior probability P(Ci) of each class, that is, the occurrence frequency of each class in the training dataset, and then according to Bayes' theorem, calculate the posterior probability P(Ci|x) of each sample belonging to each class by combining the prior probability and the class-conditional probability density.

[0084] Secondly, for each sample, calculate the risk R(Cj|x) of each class according to the loss matrix and the posterior probability. The risk function formula is:

[0085]

[0086] Wherein, R(Cj|x) represents the risk or loss of each category, K represents the number of categories, L(i, j) represents the loss function, P(Ci|x) represents the posterior probability of each category, and ∑ represents the summation operation.

[0087] Then, for a given sample, select the category that minimizes the risk as the classification result, that is, select C ′ such that R(C ′ |x) ≤ R(Cj|x) holds for all j.

[0088] Finally, use the training set to train the classifier, adjust the model parameters through an iterative algorithm (such as gradient descent), use the validation set to evaluate and optimize the performance of the model during the training process, and determine the audio recognition model.

[0089] Optionally, the method further includes: in response to the status information indicating that the switch unit is in an abnormal state, outputting a prompt message based on a preset prompt method, where the prompt message is used to prompt that the switch unit is in an abnormal state.

[0090] The above abnormal state may refer to the situation where the bypass switch of the converter valve sub-module deviates from its normal working state during operation. The types of abnormal states may include, but are not limited to, electromagnetic noise abnormalities, poor contact, wear, and fault states caused by design or manufacturing defects. The specific abnormal state needs to be determined according to the actual state of the switch unit and is not limited here.

[0091] The above preset prompt method may be a mechanism preset during the system design stage for sending an alarm to the operation and maintenance personnel or the control center when an abnormal state is recognized. The preset prompt method may include, but is not limited to, audible and visual alarms, SMS notifications, email notifications, system interface pop-ups, activation of remote alarm systems, etc. The specific preset prompt method needs to be determined according to the specific application scenario, system response time, and personnel availability and is not limited here. The preset prompt method ensures that the system can timely and effectively convey the fault information to the relevant personnel, enabling them to take immediate action, reducing the impact of the fault, and improving the system maintenance efficiency.

[0092] The above prompt message may refer to the detailed content of the alarm output by the system through the preset prompt method when the recognition system determines that the bypass switch is in an abnormal state, which is used to specifically describe the nature, location, and possible impacts of the abnormality. The prompt message may include, but is not limited to, the type of abnormality (electromagnetic noise, poor contact, etc.), the identification of the faulty switch unit (location, number), recommended handling measures, potential impact range, etc. The specific prompt message needs to be determined according to the system design and fault response requirements and is not limited here. The prompt message provides the operation and maintenance personnel with the specific information of the fault, helps them quickly locate the fault source, take targeted repair measures, reduce the fault troubleshooting time, and improve the fault handling efficiency.

[0093] In an alternative embodiment, when the audio recognition model determines that the bypass switch is in an abnormal state, the system will immediately activate a preset prompt method, output detailed prompt information, alert relevant operation and maintenance personnel or automatically start a fault handling process. The implementation of this process can not only significantly shorten the time from fault occurrence to discovery, but also reduce the blindness of troubleshooting and handling by providing specific fault information, thereby improving the operation efficiency and safety of the entire power system and providing real-time and accurate decision-making support for the operation and maintenance management of power electronic equipment.

[0094] In an alternative embodiment, Figure 2 is a flowchart of a monitoring method for a switch unit in a converter valve according to an embodiment of the present invention, as Figure 2 shown, the method includes the following steps:

[0095] Step S202, deployment and data collection of the acoustic wave sensor array; Step S204, extraction of the acoustic fingerprint feature of the bypass switch; Step S206, construction and training of the minimum Bayesian classifier; Step S208, identification of abnormal acoustic fingerprints of the bypass switch; Step S210, system testing and adjustment; Step S212, device design and implementation.

[0096] According to another aspect of the embodiments of the present invention, there is also provided a monitoring device for a switch unit in a converter valve. The device can execute the monitoring method for the switch unit in the converter valve provided in the above embodiments. The specific implementation manner and preferred application scenarios are the same as those in the above embodiments and will not be elaborated herein.

[0097] Figure 3 is a schematic diagram of a monitoring device for a switch unit in a converter valve according to an embodiment of the present invention, as Figure 3 shown, the device includes: an acquisition module 302, configured to acquire the audio signal of the switch unit in the converter valve, where the converter valve includes a plurality of sub-modules, and the switch unit is configured to isolate the faulty module among the plurality of sub-modules when a faulty module appears in the plurality of sub-modules; a feature extraction module 304, configured to extract features from the audio signal to obtain the acoustic fingerprint feature in the audio signal; a monitoring module 306, configured to input the acoustic fingerprint feature into an audio recognition model, and use the audio recognition model to monitor the state of the switch unit to obtain the state information of the switch unit, where the state information is used to indicate whether the switch unit is in an abnormal state.

[0098] Optionally, the feature extraction module includes: an enhancement unit, configured to enhance the high-frequency part of the audio signal to obtain an enhanced audio signal; a feature extraction unit, configured to extract features from the enhanced audio signal to obtain the acoustic fingerprint feature.

[0099] Optionally, the feature extraction unit includes: a segmentation subunit configured to segment the enhanced audio signal to obtain a plurality of audio signal frames, where there is an overlap between the plurality of audio signal frames; a conversion subunit configured to convert the plurality of audio signal frames into the frequency domain to obtain a target spectrum; a filtering subunit configured to filter the target spectrum to obtain the signal energy of the target spectrum; and a determination subunit configured to determine the voiceprint feature in the enhanced audio signal based on the signal energy.

[0100] Optionally, the determination subunit includes: processing the logarithm of the signal energy by using a discrete cosine transform to obtain mel-frequency cepstral coefficients; and performing mean square error normalization on the mel-frequency cepstral coefficients to obtain the voiceprint feature.

[0101] Optionally, the acquisition module includes: a capture unit configured to capture an initial audio signal of the switching unit based on a sound wave sensor array in the converter valve; and a spatial filtering unit configured to perform spatial filtering on the initial audio signal based on the position information of the switching unit to obtain an audio signal.

[0102] Optionally, the spatial filtering unit includes: a spatial filtering subunit configured to perform spatial filtering on the initial audio signal based on the position information to obtain a filtered audio signal; an acquisition subunit configured to acquire the filtered audio signal based on a preset sampling frequency to obtain an acquired audio signal; and a truncation subunit configured to truncate the acquired audio signal based on a preset duration to obtain an audio signal.

[0103] Optionally, the apparatus further includes: an acquisition module configured to acquire a sample audio signal of a sample switching unit and sample status information corresponding to the sample audio signal; an identification module configured to identify the sample audio signal by using an initial audio identification model to obtain predicted sample status information; a construction module configured to construct a loss function based on the sample status information and the predicted sample status information; and an adjustment module configured to adjust model parameters of the initial audio identification model by using the loss function to obtain an audio identification model.

[0104] Optionally, the apparatus further includes: an output module configured to output a prompt message based on a preset prompt manner in response to the status information indicating that the switching unit is in an abnormal state, where the prompt message is used to prompt that the switching unit is in an abnormal state.

[0105] According to another aspect of the embodiments of the present invention, there is also provided an electronic device, including: a memory storing an executable program; and a processor configured to run the program, where when the program runs, the methods in the embodiments of the present invention are executed.

[0106] According to another aspect of the embodiments of the present invention, there is also provided a computer-readable storage medium, where the computer-readable storage medium includes a stored program, and when the program runs, the methods in the embodiments of the present invention are executed by a processor in a device where the computer-readable storage medium is located.

[0107] The computer storage medium in the above steps may be a medium in a computer memory for storing certain discontinuous physical quantities. The main computer storage media include semiconductors, magnetic cores, magnetic drums, magnetic tapes, laser discs, etc. The stored program included in the computer-readable storage medium can be a set of instructions that can be recognized and executed by a computer, running on an electronic computer, and is an information-based tool to meet certain needs of people.

[0108] In the above embodiments of the present invention, the descriptions of the respective embodiments have their own emphases. For parts not detailed in a certain embodiment, reference may be made to the relevant descriptions of other embodiments.

[0109] In several embodiments provided in the present application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only illustrative. For example, the division of the units can be a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be through some interfaces. The indirect couplings or communication connections of the units or modules can be in electrical or other forms.

[0110] The units described as separate components may or may not be physically separated. The components displayed as units may or may not be physical units, that is, they can be located in one place, or can be distributed to multiple units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0111] In addition, the functional units in the various embodiments of the present invention can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above integrated units can be implemented in the form of hardware or in the form of software functional units.

[0112] When the integrated unit 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. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The foregoing storage medium includes: various media that can store program codes, such as USB flash drives, read-only memories (ROMs), random access memories (RAMs), mobile hard disks, magnetic disks, or optical discs.

[0113] The foregoing are only the preferred embodiments of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present invention.

Claims

1. A method for monitoring a switch unit in a converter valve, characterized in that: include: Collecting an audio signal of a switch unit in a converter valve, wherein the converter valve comprises a plurality of submodules, and the switch unit is used to isolate a faulty module among the plurality of submodules when a faulty module occurs among the plurality of submodules; Extracting features from the audio signal to obtain voiceprint features in the audio signal; The voiceprint feature is input into an audio recognition model, and the state of the switch unit is monitored using the audio recognition model to obtain state information of the switch unit, wherein the state information is used to indicate whether the switch unit is in an abnormal state.

2. The method for monitoring the switch unit in the converter valve according to claim 1, characterized in that: Extracting features from the audio signal to obtain voiceprint features in the audio signal includes: enhancing the high frequency part of the audio signal to obtain an enhanced audio signal; Feature extraction is performed on the enhanced audio signal to obtain the voiceprint feature.

3. The method for monitoring the switch unit in the converter valve according to claim 2, characterized in that: Extracting features from the enhanced audio signal to obtain the voiceprint features includes: Segmenting the enhanced audio signal to obtain a plurality of audio signal frames, wherein the plurality of audio signal frames overlap; Convert the multiple audio signal frames into a frequency domain to obtain a target frequency spectrum; Filtering the target spectrum to obtain signal energy of the target spectrum; The voiceprint feature in the enhanced audio signal is determined based on the signal energy.

4. The method for monitoring the switch unit in the converter valve according to claim 3, characterized in that: Determining the voiceprint feature in the enhanced audio signal based on the signal energy includes: Processing the logarithm of the signal energy by discrete cosine transform to obtain frequency cepstrum coefficients; The frequency cepstrum coefficients are normalized by mean square error to obtain the voiceprint feature.

5. The method for monitoring a switch unit in a converter valve according to claim 1, characterized in that: Collect the audio signals of the switch unit in the converter valve, including: capturing an initial audio signal of the switch unit based on an acoustic wave sensor array in the converter valve; The initial audio signal is spatially filtered based on the position information of the switch unit to obtain the audio signal.

6. The method for monitoring the switch unit in the converter valve according to claim 5, characterized in that: Performing spatial filtering on the initial audio signal based on the position information of the switch unit to obtain the audio signal includes: Performing spatial filtering on the initial audio signal based on the position information to obtain a filtered audio signal; Sampling the filtered audio signal based on a preset sampling frequency to obtain a sampled audio signal; The collected audio signal is intercepted based on a preset time length to obtain the audio signal.

7. The method for monitoring a switch unit in a converter valve according to any one of claims 1 to 6, characterized in that: The method further comprises: Acquire a sample audio signal of a sample switch unit and sample state information corresponding to the sample audio signal; Using an initial audio recognition model to recognize the sample audio signal, and obtain predicted sample state information; Constructing a loss function based on the sample state information and the predicted sample state information; The loss function is used to adjust the model parameters of the initial audio recognition model to obtain the audio recognition model.

8. The method for monitoring a switch unit in a converter valve according to any one of claims 1 to 6, characterized in that: The method further comprises: In response to the state information indicating that the switch unit is in the abnormal state, prompt information is output based on a preset prompt mode, wherein the prompt information is used to prompt that the switch unit is in the abnormal state.

9. An electronic device, characterized in that: include: A memory storing an executable program; A processor, configured to run the program, wherein the program executes the method according to any one of claims 1 to 8 when running.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored executable program, wherein when the executable program is executed, the device where the storage medium is located is controlled to execute the method according to any one of claims 1 to 8.

Images