Data-Driven Method and Device for Identifying Ultra-High Frequency Partial Discharge of Hydro Generators

Abstract

This application provides a data-driven method and apparatus for identifying ultra-high frequency partial discharge (UHF) signals from hydro-generators, belonging to the field of partial discharge monitoring technology. The method includes: acquiring monitoring signals collected by a coupling device; discretizing the monitoring signals to obtain an amplitude sequence; inputting the amplitude sequence into a pre-built signal diagnostic model to generate a diagnostic result indicating whether the hydro-generator is operating normally; if the diagnostic result indicates an abnormal operating state of the hydro-generator, extracting a first feature of the monitoring signal; if no interference signal corresponding to the first feature is found in a pre-built background noise database, extracting a second feature of the monitoring signal; and using the second feature to match the UHF event corresponding to the second feature from a pre-built UHF case database. The first feature includes energy characteristics, and the second feature includes time-domain and / or frequency-domain characteristics. This method combines data-driven approaches, a background noise database, and a UHF case database to achieve rapid and accurate identification of UHF signals.

Description

Technical Field

[0001] This application belongs to the field of partial discharge monitoring technology, specifically relating to a data-driven method for identifying ultra-high frequency partial discharge of hydro-generators, a data-driven device for identifying ultra-high frequency partial discharge of hydro-generators, a computer device, and a machine-readable storage medium. Background Technology

[0002] As the core equipment of a hydropower station, the health of the insulation system of a hydro-generator directly affects the safety and stability of the entire power system. In actual production, the detection of partial discharge (PD) signals is an important means of reflecting the insulation status of large power equipment. With the continuous expansion of hydropower installed capacity, higher requirements are placed on the monitoring and evaluation of the insulation status of hydro-generators. In recent years, compared with low-frequency PD monitoring technology, ultra-high frequency (UHF) PD monitoring technology has been gradually adopted due to its higher sensitivity and better anti-interference performance. However, due to the complex operating conditions of hydro-generators, the partial discharge signals detected by UHF PD monitoring are still severely affected by interference. How to achieve accurate PD signal identification is a technical challenge in the field of UHF PD monitoring of hydro-generators. Summary of the Invention

[0003] In view of this, the purpose of this application is to provide a data-driven method for identifying ultra-high frequency partial discharge (UHF) signals from hydro-generators, a data-driven device for identifying UHF signals from hydro-generators, a computer device, and a machine-readable storage medium, in order to overcome the technical problem in the prior art that it is difficult to accurately identify UHF signals due to the complex operating conditions of hydro-generators.

[0004] To achieve the above objectives, the first aspect of this application provides a data-driven method for identifying ultra-high frequency partial discharge (UHF) discharge from hydro-generators, the method comprising:

[0005] Acquire the monitoring signal collected by the UHF partial discharge coupling device during the current monitoring period;

[0006] The monitoring signal is discretized to obtain an amplitude sequence;

[0007] The amplitude sequence is input into a pre-built signal diagnostic model to generate a diagnostic result on whether the hydro-generator is working properly.

[0008] If the diagnostic results indicate that the hydro-generator is in an abnormal operating state, then the first feature of the monitoring signal is extracted;

[0009] If no interference signal corresponding to the first feature is found in the constructed background noise library, then the second feature of the monitoring signal is extracted;

[0010] Using the second feature, match the partial discharge events corresponding to the second feature from the constructed partial discharge case library;

[0011] Each interference signal has its own corresponding first feature, and each partial discharge event has its own corresponding second feature. The first feature includes energy features, and the second feature includes time-domain and / or frequency-domain features. The signal diagnostic model is obtained by training and optimizing the parameters of the classification network.

[0012] In a specific embodiment of this application, the signal diagnosis model is obtained by training and optimizing the parameters of the classification network, including:

[0013] The classification network is trained in the first stage using the first training set to obtain the first diagnostic model. The first training set consists of amplitude sequences obtained by discretizing historical hydro-generator normal operating state signals.

[0014] The first diagnostic model is trained in the second stage using the second training set to obtain the second diagnostic model. The second training set consists of a specific amplitude sequence, and the historical monitoring signal corresponding to the specific amplitude sequence is the abnormal working state signal of the hydro-generator.

[0015] The second diagnostic model is trained in the third stage using a third training set, which consists of a target amplitude sequence, and the target amplitude sequence is a historical amplitude sequence that meets preset conditions.

[0016] The preset condition is as follows: the diagnostic result obtained by inputting the second diagnostic model is that the hydro-generator is in an abnormal working state, and then the monitoring signal corresponding to the input amplitude sequence is determined by manual judgment to be the hydro-generator in an abnormal working state signal.

[0017] In specific embodiments of this application, the classification network types include XGBoost, Random Forest, and Extreme Learning Machine.

[0018] In a specific embodiment of this application, the energy characteristic is power spectral density.

[0019] In specific embodiments of this application, the time-domain features include at least one of phase, amplitude, waveform features, and amplitude statistical features.

[0020] In specific embodiments of this application, the waveform features include one of rise time, fall time, duration, and pulse width.

[0021] In specific embodiments of this application, the types of interference signals in the background noise library include corona discharge generated outside the hydro generator, electromagnetic interference, pulse interference, interference signals caused by poor grounding during the acquisition of the monitoring signal, electronic noise of the acquisition equipment itself during the acquisition of the monitoring signal, environmental noise, false signals generated by the acquisition equipment due to temperature changes during the acquisition of the monitoring signal, false discharge signals caused by humidity changes, signal transmission interference, and human interference.

[0022] A second aspect of this application provides a data-driven ultra-high frequency partial discharge identification device for hydro-generators, the device comprising:

[0023] The acquisition module is used to acquire the monitoring signal collected by the UHF partial discharge coupling device during the current monitoring period;

[0024] The signal processing module is used to discretize the monitoring signal to obtain an amplitude sequence;

[0025] The diagnostic module is used to input the amplitude sequence into a pre-built signal diagnostic model and generate a diagnostic result on whether the hydro-generator is working properly.

[0026] The first extraction module is used to extract the first feature of the monitoring signal when the diagnostic results indicate that the hydro-generator is in an abnormal operating state.

[0027] The second extraction module is used to search for the interference signal corresponding to the first feature in the constructed background noise library. If no interference signal corresponding to the first feature is found, the second feature of the monitoring signal is extracted.

[0028] The signal recognition module is used to match the partial discharge event corresponding to the second feature from the constructed partial discharge case library using the second feature;

[0029] Each interference signal has its own corresponding first feature, and each partial discharge event has its own corresponding second feature. The first feature includes energy features, and the second feature includes time-domain and / or frequency-domain features. The signal diagnostic model is obtained by training and optimizing the parameters of the classification network.

[0030] A third aspect of this application provides a computer device, comprising:

[0031] The memory is configured to store instructions; and

[0032] The processor is configured to retrieve the instructions from the memory and, when executing the instructions, to implement the data-driven hydro-generator UHF partial discharge identification method according to the first aspect of this application.

[0033] A fourth aspect of this application provides a machine-readable storage medium storing instructions for causing a machine to perform the data-driven hydro-generator UHF partial discharge identification method according to the first aspect of this application.

[0034] In the above technical solution, a data-driven signal diagnostic model is used to diagnose the working status of the hydro-generator. Then, by combining the background noise library and the partial discharge case library, interference signals and partial discharge events are distinguished, thereby achieving rapid and accurate identification of UHF partial discharge signals.

[0035] Other features and advantages of the embodiments of this application will be described in detail in the following detailed description section. Attached Figure Description

[0036] The accompanying drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the following detailed description to explain the embodiments of this application, but do not constitute a limitation on the embodiments of this application. In the drawings:

[0037] Figure 1 A flowchart illustrating a data-driven UHF partial discharge identification method for hydro-generators according to an embodiment of this application is shown in the schematic diagram.

[0038] Figure 2 This schematic diagram illustrates the composition of a data-driven ultra-high frequency partial discharge identification device for hydro-generators according to an embodiment of this application.

[0039] Figure 3 A schematic block diagram of a computer device according to an embodiment of this application is shown. Detailed Implementation

[0040] The specific embodiments of this application will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit the embodiments of this application.

[0041] If the embodiments of this application involve descriptions such as "first" or "second," these descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of those features. Furthermore, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. If the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed in this application.

[0042] In current power systems, due to the attenuation of UHF partial discharge signals, it is difficult to obtain the true value of the partial discharge signal through quantitative analysis. Currently, in engineering applications, UHF partial discharge monitoring usually adopts the method of combining single-unit trend comparison and partial discharge case library for judgment. Its essence is a qualitative method for determining the value of partial discharge. This qualitative analysis method requires high capabilities from on-site technicians. Therefore, the existing online monitoring methods for partial discharge of hydro-generators have certain limitations.

[0043] To overcome the above-mentioned shortcomings, this application proposes a data-driven method for identifying ultra-high frequency partial discharge of hydro-generators, the method comprising:

[0044] Acquire the monitoring signal collected by the UHF partial discharge coupling device during the current monitoring period;

[0045] The acquired monitoring signals are preprocessed and discretized to obtain an amplitude sequence;

[0046] The amplitude sequence is input into a pre-built signal diagnostic model to generate a diagnostic result on whether the hydro-generator is working properly. The signal diagnostic model is obtained by training and optimizing the parameters of a classification network.

[0047] If the diagnostic results indicate that the hydro-generator is in an abnormal operating state, then the first feature of the monitoring signal is extracted, and the first feature includes at least the energy feature;

[0048] If no interference signal corresponding to the first feature is found in the constructed background noise library, the second feature of the monitoring signal is extracted, wherein each interference signal has its own corresponding first feature, and the second feature includes at least time domain features and / or frequency domain features.

[0049] The second feature of the monitored signal is used to match the partial discharge event corresponding to the second feature from the constructed partial discharge case library. Each partial discharge event has its own corresponding second feature.

[0050] It should be understood that in this application, the collected monitoring signal is discretized to obtain a pulse sequence, and the amplitude sequence is a sequence composed of the amplitudes of each pulse within the pulse sequence. Furthermore, the partial discharge signals corresponding to different types of partial discharge events differ in the time domain and / or frequency domain. Specifically, the amplitude, frequency, and waveform of the partial discharge signal change with the location and intensity of the partial discharge. Therefore, the second feature mentioned above refers to these features that can reflect the differences in the time or frequency domain of the partial discharge signal. The above embodiments do not limit the specific time-domain and / or frequency-domain features used; feasible time-domain and / or frequency-domain features need to be determined in conjunction with the specific type of turbine generator unit in the hydropower station.

[0051] This application, as described in the above embodiments, combines a data-driven signal diagnosis model, a background noise library, and a partial discharge case library. First, the data-driven signal diagnosis model identifies whether the monitoring signal collected within the current monitoring period is a signal indicating normal operation or abnormal operation of the hydro-generator. The signal diagnosis model itself possesses the high-precision characteristics of a machine learning model. When an abnormal hydro-generator operation is identified, the background noise library and partial discharge case library are used to determine whether the current monitoring signal is an interference signal or a partial discharge signal. If the signal diagnosis model's result indicates that the hydro-generator is operating normally, then the monitoring signal corresponding to the amplitude sequence input to the signal diagnosis model is the signal indicating normal operation of the hydro-generator, because the monitoring signal... The signal is preprocessed through filtering, denoising, and cleaning before data analysis, but there are still interferences that cannot be filtered out, such as periodic pulse interference. If the diagnostic result of the signal diagnostic model is that the hydro-generator is in an abnormal operating state, then the monitoring signal corresponding to the amplitude sequence of the input signal diagnostic model is the abnormal operating state signal of the hydro-generator. The abnormal operating state signal of the hydro-generator can be a power frequency frequency carrying interference signals such as periodic pulse interference, or a power frequency frequency carrying partial discharge pulses. There are significant differences in energy characteristics between interference signals and partial discharge pulses. Therefore, interference signals and partial discharge signals are distinguished by a first feature that includes at least energy characteristics. After identifying the partial discharge signal, the type of partial discharge event corresponding to the partial discharge signal is identified by a second feature.

[0052] In a comparative example, an adaptive threshold is achieved using an unsupervised clustering method. The adaptive threshold is then used to distinguish between interference and partial discharge signals. Additionally, the unsupervised clustering method is used to classify partial discharge signals. Because the types of interference and partial discharge signals are quite large, the number of cluster centers when using the unsupervised clustering method is large, resulting in a large computational load. The recognition efficiency is lower than that of the recognition method in this application, which combines a data-driven signal diagnostic model, a background noise library, and a partial discharge case library.

[0053] Based on this, the embodiments of this application described above achieve rapid and accurate identification of ultra-high frequency partial discharge signals.

[0054] Figure 1 This illustration schematically depicts a data-driven UHF partial discharge identification method for hydro-generators according to an embodiment of this application. For example... Figure 1 As shown, the data-driven UHF partial discharge identification method for hydro-generators may include steps 102 to 112. It is understood that the above-described data-driven UHF partial discharge identification method for hydro-generators may include all of steps 102 to 112, or may include only some of them.

[0055] Step 102: Obtain the monitoring signal collected by the UHF partial discharge coupling device during the current monitoring period.

[0056] In this application, the UHF partial discharge coupling device can be an UHF partial discharge sensor, an UHF partial discharge antenna, or a coupler composed of a transformer and a coupling capacitor, etc.

[0057] Step 104 involves preprocessing and discretizing the acquired monitoring signals to obtain an amplitude sequence. Preprocessing may include bandpass filtering, normalization, etc.

[0058] For example, normalization is performed using the following formula:

[0059] ;

[0060] in, x(t) This represents the monitoring signal after bandpass filtering, where max(|x(t)|) is the signal... x(t) Maximum absolute amplitude, normalized signal x b (t) The range is [-1, 1].

[0061] Step 106: Input the amplitude sequence obtained in step 104 into the pre-built signal diagnostic model to generate a diagnostic result on whether the hydro-generator is working properly.

[0062] Specifically, in this application, the signal diagnostic model is obtained by training and optimizing the parameters of a classification network.

[0063] As an example, classification networks can be of the following types: XGBoost, Random Forest, Extreme Learning Machine, RBF Network, etc.

[0064] As an example, the signal diagnostic model uses the cross-entropy loss function during training and regularization to prevent overfitting.

[0065] As is known, in actual production, the amount of signal data collected by the UHF partial discharge monitoring method is enormous, and most of it consists of signals from the normal operating state of hydroelectric generators. Therefore, there is a lack of abundant partial discharge signal samples that can be used to train the classification network. Consequently, based on the training methods in the common embodiments, the model is prone to overfitting and poor generalization ability. Therefore, as an improvement to the above embodiments of this application, this application proposes a signal diagnostic model construction method based on multi-stage training. Accordingly, the classification network is trained and its parameters are optimized to obtain the signal diagnostic model, specifically including the following steps:

[0066] Step A1: The classification network is trained in the first stage using the first training set to obtain the first diagnostic model. The first training set consists of the amplitude sequence obtained by discretizing the normal operating state signal of the hydro-generator.

[0067] In this application, the samples used in the first stage of training can be referred to as positive samples.

[0068] Step A2: Use the second training set to train the first diagnostic model in the second stage to obtain the second diagnostic model. The second training set consists of a specific amplitude sequence, and the historical monitoring signal corresponding to the specific amplitude sequence is the abnormal working state signal of the hydro-generator.

[0069] In this application, the samples used in the second stage of training can be referred to as negative samples.

[0070] Step A3: Use the third training set to train the second diagnostic model in the third stage. The third training set consists of a target amplitude sequence, which is a historical amplitude sequence that meets preset conditions.

[0071] Specifically, the preset conditions mentioned in this application refer to: the diagnostic result obtained by the input second diagnostic model is that the hydro-generator is in an abnormal operating state, and then the monitoring signal corresponding to the input amplitude sequence is determined by manual inspection to be an abnormal operating state signal of the hydro-generator. The samples used in the third stage of training are the samples obtained after being identified by the second diagnostic model and manually labeled, which are also negative samples.

[0072] In the improved embodiment described above, multi-stage training fully utilizes both positive and negative samples. Furthermore, the third training set consists of expanded negative samples identified by the second diagnostic model and manually labeled. This sample expansion is more efficient than directly manually labeling negative samples from a large number of historical monitoring signals in the ordinary embodiment. Therefore, multi-stage training improves the efficiency of constructing the signal diagnostic model and the recognition accuracy of the constructed model.

[0073] In summary, the above-described improved embodiments enhance the efficiency and accuracy of partial discharge identification.

[0074] Step 108: If the diagnostic result indicates that the hydro-generator is in an abnormal operating state, then the first feature of the monitoring signal is extracted; if the diagnostic result indicates that the hydro-generator is in a normal operating state, then the partial discharge identification ends.

[0075] In this application, the first feature includes at least an energy feature. For example, the energy feature is the power spectral density.

[0076] Step 110: If no interference signal corresponding to the first feature of the extracted monitoring signal is found in the constructed background noise library, then the second feature of the monitoring signal is extracted, wherein each interference signal has its own corresponding first feature; if an interference signal corresponding to the first feature of the extracted monitoring signal is found in the constructed background noise library, then the monitoring signal collected in the current monitoring period is determined to be the interference signal, and the partial discharge identification ends.

[0077] In this application, the second feature includes at least time-domain features and / or frequency-domain features.

[0078] In one optional embodiment of this application, the time-domain features include one or more of phase, amplitude, waveform features, and amplitude statistical features. The waveform features include one or more of rise time, fall time, duration, and pulse width. The amplitude statistical features include one or more of the average amplitude and the standard deviation of amplitude.

[0079] Since different types of partial discharge signals correspond to different spectral distributions, in an optional embodiment of this application, the frequency domain characteristics can be the intensity and frequency distribution of different frequency components in the spectrum.

[0080] As an example, the types of interference signals in the background noise library include corona discharge generated outside the hydro-generator, electromagnetic interference, pulse interference, interference signals caused by poor grounding during monitoring signal acquisition, electronic noise of the acquisition equipment itself during monitoring signal acquisition, environmental noise, erroneous signals generated by the acquisition equipment due to temperature changes during monitoring signal acquisition, false discharge signals caused by humidity changes, signal transmission interference, and human interference. The sources of these interference signals are shown in Table 1 below.

[0081] Table 1

[0082]

[0083] Step 112: Using the extracted second feature of the monitoring signal, match the partial discharge event corresponding to the second feature from the constructed partial discharge case library. Each partial discharge event has its own corresponding second feature. If no partial discharge event corresponding to the second feature is matched from the constructed partial discharge case library using the extracted second feature of the monitoring signal, the monitoring signal is manually marked. If the difference between the first feature of the monitoring signal and the first feature of the interference signal in the background noise library is within a first preset range, the interference signal type to which the monitoring signal belongs is manually marked; otherwise, the partial discharge event type to which the monitoring signal belongs is manually marked. When manually analyzing and determining a new partial discharge event type, offline partial discharge testing can be performed. For example, partial discharge testing can be carried out at various UHF partial discharge coupling device locations on the stator winding of a hydro-generator using offline testing. The partial discharge signals collected at each location are collected, and the feature parameters of these partial discharge signals are extracted. By matching and analyzing the extracted feature parameters with the feature parameters of the monitoring signal to be classified, the partial discharge event type of the monitoring signal to be classified is determined. Preferably, before extracting the characteristic parameters of these partial discharge signals, the partial discharge signals collected at each location are corrected using the maximum attenuation coefficient, which can be determined through the process in the general embodiment.

[0084] For example, test partial discharge signals are injected into each preset injection point of the stator winding bar, and then the partial discharge signals are collected at the locations of the ultra-high frequency partial discharge coupling device. The transfer function and attenuation coefficient of the stator winding bar are then determined by a network vector analyzer, thereby determining the maximum attenuation coefficient. The partial discharge signals collected at each location are reconstructed through the transfer function, and the reconstructed partial discharge signals are multiplied by the maximum attenuation coefficient to complete the correction of the partial discharge signals.

[0085] It is important to understand that the types of interference signals in the background noise database are gradually being improved, as are the types of partial discharge events in the partial discharge case database. When the current background noise database and partial discharge case database are insufficient to classify the monitored signals, manual analysis is required to classify the monitored signals. The new interference signal types obtained from the classification are then added to the background noise database, or the new partial discharge events obtained from the classification are added to the partial discharge case database.

[0086] As an example, the partial discharge (PD) case library includes various typical PD events and various ordinary PD events. Typical PD events refer to PD events occurring under extreme conditions in hydro-generators, while ordinary PD events refer to PD events occurring under normal conditions in hydro-generators. One typical PD event is a PD case under extreme humidity conditions, specifically simulating partial discharge in a high-humidity environment. Moisture can significantly alter the properties of insulating materials, increasing the probability of PD. This type of PD event is obtained through simulation under these conditions. Another typical PD event is a PD case under high-temperature conditions, specifically simulating PD at temperatures higher than normal operating temperatures. High temperatures may accelerate the aging of insulating materials, thus triggering PD. This type of PD event is obtained through simulation under these conditions. A third typical PD event is a PD case under strong electromagnetic interference environments, specifically simulating PD in environments with strong electromagnetic interference. In specific applications, the complex electromagnetic interference environment of actual power systems is simulated to obtain PD events under these conditions. Another typical partial discharge (PD) event is the PD case resulting from insulation material aging. Specifically, this involves using aged insulation materials or accelerating aging tests to simulate the long-term operation of a hydro-generator, examining the impact of insulation aging on PD, and obtaining PD events under these conditions. Another typical PD event is the PD case caused by mechanical stress on the hydro-generator. Specifically, during PD testing of the hydro-generator, mechanical vibration or impact is applied to the generator. Since mechanical stress or impact may cause damage to the insulation structure, resulting in PD, the PD event under these conditions is obtained by examining the PD caused by this insulation damage. A third typical PD event is the PD case caused by different types of insulation defects. Specifically, this involves constructing insulation defect models with different types and sizes, such as bubbles, cracks, and impurities, and examining the impact of these different insulation defects on PD, obtaining PD events under these conditions. Another typical partial discharge (PD) event is the PD case under extreme operating conditions of a hydro-generator. Specifically, this involves simulating extreme operating conditions of the hydro-generator, such as rapid voltage increase, rapid voltage decrease, and overload, as these extreme conditions can exacerbate PD phenomena. The PD event under these extreme conditions is obtained through PD simulation. Another typical PD event is the PD case under combined fault conditions. Specifically, this involves conducting PD simulation tests on a hydro-generator under multiple extreme conditions simultaneously, such as high temperature, high humidity, and mechanical stress. The combined impact of these combined fault conditions on PD is examined, and the PD event under these conditions is obtained.

[0087] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0088] Corresponding to the data-driven UHF partial discharge identification method for hydro-generators in the above embodiments, Figure 2 The diagram illustrates the components of a data-driven hydro-generator UHF partial discharge identification device 400 provided in an embodiment of this application. For ease of explanation, only the parts relevant to the embodiment of this application are shown.

[0089] like Figure 2 As shown, the data-driven UHF partial discharge identification device 400 for hydro-generators includes:

[0090] The acquisition module 410 is used to acquire the monitoring signal collected by the UHF partial discharge coupling device during the current monitoring period;

[0091] Signal processing module 420 is used to discretize the monitoring signal to obtain an amplitude sequence;

[0092] Diagnostic module 430 is used to input the amplitude sequence into a pre-built signal diagnostic model to generate a diagnostic result on whether the hydro-generator is working properly;

[0093] The first extraction module 440 is used to extract the first feature of the monitoring signal when the diagnostic result indicates that the working state of the hydro-generator is abnormal;

[0094] The second extraction module 450 is used to search for an interference signal corresponding to the first feature in the constructed background noise library. If no interference signal corresponding to the first feature is found, the second feature of the monitoring signal is extracted.

[0095] The signal recognition module 460 is used to match the partial discharge event corresponding to the second feature from the constructed partial discharge case library using the second feature;

[0096] Each interference signal has its own corresponding first feature, and each partial discharge event has its own corresponding second feature. The first feature includes energy features, and the second feature includes time-domain and / or frequency-domain features. The signal diagnostic model is obtained by training and optimizing the parameters of the classification network.

[0097] The process by which each module in the data-driven ultra-high frequency partial discharge identification device 400 for hydro-generators provided in this application implements its respective function can be specifically referred to the foregoing. Figure 1 The descriptions of the embodiments shown and other related method embodiments are not repeated here.

[0098] It should be noted that the information interaction and execution process between the above modules are based on the same concept as the method embodiments of this application. Their specific functions and technical effects can be found in the method embodiments section, and will not be repeated here.

[0099] Figure 3 A schematic block diagram of a computer device according to an embodiment of this application is shown.

[0100] In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 3 As shown in the figure, the computer device includes a processor A01, a network interface A02, a display screen A04, an input device A05, and a memory (not shown) connected via a system bus. The processor A01 provides computing and control capabilities. The memory includes internal memory A03 and a non-volatile storage medium A06. The non-volatile storage medium A06 stores an operating system B01 and a computer program B02. The internal memory A03 provides an environment for the operation of the operating system B01 and the computer program B02 stored in the non-volatile storage medium A06. The network interface A02 is used for communication with external terminals via a network connection. When the computer program is executed by the processor A01, it implements a data-driven method for identifying ultra-high frequency partial discharge of a hydro-generator. The display screen A04 can be a liquid crystal display (LCD) or an e-ink display. The input device A05 can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the computer device casing, or an external keyboard, touchpad, or mouse.

[0101] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0102] In one embodiment, the data-driven ultra-high frequency partial discharge identification device 400 for hydro-generators provided in this application can be implemented as a computer program, which can be implemented in the form of, for example... Figure 3 The computer device shown operates on this device. The memory of the computer device can store various program modules that constitute the data-driven hydro-generator UHF partial discharge identification device 400. The computer program, composed of these program modules, causes the processor to execute the steps in the data-driven hydro-generator UHF partial discharge identification method of the various embodiments of this application described in this specification.

[0103] In one embodiment, this application also provides a machine-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the data-driven hydro-generator UHF partial discharge identification method described in the above embodiments.

[0104] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0105] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0106] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0107] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A data-driven based ultra-high frequency partial discharge identification method for a hydro-generator, characterized in that, The method comprises the following steps: acquiring a monitoring signal collected by an ultrahigh frequency partial discharge coupling device in a current monitoring period; discretizing the monitoring signal to obtain an amplitude sequence; inputting the amplitude sequence into a pre-constructed signal diagnosis model to generate a diagnosis result of whether the hydro-generator is working normally; if the diagnosis result indicates that the working state of the hydro-generator is abnormal, extracting a first feature of the monitoring signal; if an interference signal corresponding to the first feature is not found in a constructed background noise library, extracting a second feature of the monitoring signal; matching a partial discharge event corresponding to the second feature from a constructed partial discharge case library by using the second feature; wherein each interference signal has a first feature corresponding thereto, each partial discharge event has a second feature corresponding thereto, the first feature comprises an energy feature, and the second feature comprises a time domain and / or frequency domain feature; the signal diagnosis model is obtained by training and parameter optimization of a classification network, specifically: a first training set is used to perform first stage training on the classification network to obtain a first diagnosis model, the first training set is composed of amplitude sequences obtained by discretizing historical hydro-generator normal working state signals; a second training set is used to perform second stage training on the first diagnosis model to obtain a second diagnosis model, the second training set is composed of specific amplitude sequences, and the historical monitoring signals corresponding to the specific amplitude sequences are hydro-generator abnormal working state signals; a third training set is used to perform third stage training on the second diagnosis model, the third training set is composed of target amplitude sequences, and the target amplitude sequences are historical amplitude sequences satisfying a preset condition; wherein the preset condition is that a diagnosis result obtained by inputting the second diagnosis model is that the working state of the hydro-generator is abnormal, and then it is artificially determined that the monitoring signal corresponding to the input amplitude sequence is a hydro-generator abnormal working state signal.

2. The data-driven based hydro-generator ultra-high frequency partial discharge identification method according to claim 1, characterized in that, The classification network is one of XGBoost, random forest and extreme learning machine.

3. The data-driven based ultra-high frequency partial discharge identification method for hydro-generators according to claim 1, characterized in that, The energy feature is a power spectral density.

4. The data-driven based ultra-high frequency partial discharge recognition method for hydroelectric generators according to claim 1, characterized in that, The time domain feature comprises at least one of a phase, an amplitude, a waveform feature and an amplitude statistical feature.

5. The data-driven based ultra-high frequency partial discharge identification method of a hydro-generator according to claim 4, characterized in that, The waveform feature comprises one of a rise time, a fall time, a duration and a pulse width.

6. The data-driven based hydro-generator ultra-high frequency partial discharge identification method according to claim 1, characterized in that, The types of interference signals in the background noise library comprise corona discharge generated from outside the hydro-generator, electromagnetic interference, pulse interference, interference signals caused by poor grounding during collection of the monitoring signal, electronic noise of the collection device itself during collection of the monitoring signal, environmental noise, false signals generated by the collection device caused by temperature change during collection of the monitoring signal, false discharge signals caused by humidity change, signal transmission interference and human interference.

7. A data-driven based ultra-high frequency partial discharge identification device for a hydro-generator, characterized in that, The method comprises the following steps: a collection module, configured to acquire a monitoring signal collected by an ultrahigh frequency partial discharge coupling device in a current monitoring period; a signal processing module, configured to discretize the monitoring signal to obtain an amplitude sequence; a diagnosis module, configured to input the amplitude sequence into a pre-constructed signal diagnosis model to generate a diagnosis result of whether the hydro-generator is working normally; The first extraction module is configured to extract a first feature of the monitoring signal when the diagnosis result indicates that the working state of the hydro-generator is abnormal. The second extraction module is configured to search for an interference signal corresponding to the first feature in the constructed background noise library, and extract a second feature of the monitoring signal if no interference signal corresponding to the first feature is found. The signal identification module is configured to match a partial discharge event corresponding to the second feature from the constructed partial discharge case library by using the second feature. Each interference signal has a corresponding first feature, and each partial discharge event has a corresponding second feature. The first feature includes an energy feature, and the second feature includes a time domain and / or frequency domain feature. The signal diagnosis model is obtained by training and parameter optimization of a classification network. Specifically, the classification network is trained in a first stage by using a first training set composed of amplitude sequences obtained by discretizing historical normal working state signals of the hydro-generator, to obtain a first diagnosis model. The first diagnosis model is trained in a second stage by using a second training set composed of specific amplitude sequences, to obtain a second diagnosis model. The specific amplitude sequences correspond to abnormal working state signals of the hydro-generator. The second diagnosis model is trained in a third stage by using a third training set composed of target amplitude sequences, which are historical amplitude sequences satisfying a preset condition. The preset condition is that the diagnosis result obtained by inputting the second diagnosis model is that the working state of the hydro-generator is abnormal, and the monitoring signal corresponding to the input amplitude sequence is determined by artificial judgment to be an abnormal working state signal of the hydro-generator.

8. A computer device, comprising: The memory is configured to store instructions. The processor is configured to call the instructions from the memory and implement the data-driven hydro-generator ultra-high frequency partial discharge identification method according to any one of claims 1 to 6 when the instructions are executed. The machine-readable storage medium stores instructions for causing a machine to execute the data-driven hydro-generator ultra-high frequency partial discharge identification method according to any one of claims 1 to 6. ​ 9. A machine-readable storage medium, characterized in that, ​

Images