Power equipment fault diagnosis and prediction method, device and electronic equipment

Abstract

The present disclosure relates to a method, device, and electronic device for diagnosing and predicting faults in power equipment, and relates to the field of artificial intelligence technology. The method comprises: collecting sound data and electrical data from multiple power equipment, locating noise data from the sound data, and extracting fault features based on the time-varying characteristic curve of the sound data where the noise data is located; locating the fault time node where the fault features are extracted, matching the electrical data corresponding to the fault time node, and obtaining fault electrical data; using a sound separation model to separate the noise data from the sound data to obtain the noise data of each power equipment, and determining the fault of each power equipment based on the matching degree of the noise data, time-varying characteristic curve, and fault electrical data of each power equipment; and predicting the fault trend based on the abnormal data trend of the fault type. By applying the disclosed solution, the accuracy and reliability of partial discharge detection can be improved, and the risk of missed detection or misjudgment can be reduced.

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence technology, and in particular to a method, device, and electronic device for diagnosing and predicting faults in power equipment. Background Art

[0002] With the continuous development of power systems, gas-insulated switchgear (GIS) has become a key high-voltage switchgear in power grids due to its significant advantages, including small footprint, easy installation, high operational reliability, and strong environmental adaptability. The stable operation of GIS equipment is crucial to ensuring the continuity and security of power supply.

[0003] However, GIS equipment itself is complex in structure, with numerous internal components, and requires extremely high manufacturing and maintenance processes. Over long-term operation, due to factors such as electric field stress, mechanical vibration, and temperature fluctuations, the insulation materials of GIS equipment may gradually age, leading to partial discharge (PD). Partial discharge is a key sign of GIS equipment insulation degradation. If not detected and addressed promptly, it can progress to insulation breakdown, causing equipment damage, large-scale power outages, and even accidents, resulting in significant economic losses and social impacts.

[0004] Currently, ultra-high frequency partial discharge (UHF) detection technology has been widely used in power systems as an effective means of assessing GIS operating status. This technology utilizes the ultra-high frequency (UHF) electromagnetic wave signals generated by partial discharge for detection, offering high sensitivity and interference immunity. However, UHF PD detection technology also has limitations. For example, the propagation characteristics of UHF signals within GIS equipment are complex, resulting in severe signal attenuation. It is also susceptible to various external electromagnetic interferences, including corona interference noise generated by corona discharge caused by glitches at busbar ends, communication signal interference from carrier communication equipment (30-500kHz) and high-frequency protection signals, random electromagnetic interference from thermal noise, ground grid noise, and coupling noise from distribution lines. Furthermore, discharge signals from other power equipment overlap with the target signal frequency band, or interference from discharges from adjacent equipment. These interference signals can reduce the accuracy and reliability of UHF PD detection, and may even lead to missed detections or misjudgments. Summary of the Invention

[0005] In view of this, the present disclosure provides a method, device and electronic equipment for diagnosing and predicting faults in power equipment, the main purpose of which is to solve the technical problem that the current partial discharge detection technology is interfered with by interference signals, reducing the accuracy and reliability of partial discharge detection, and even leading to missed detection or misjudgment.

[0006] According to a first aspect of the present disclosure, a method for diagnosing and predicting faults in power equipment is provided, the method comprising:

[0007] collecting sound data and electrical data of a plurality of power devices, locating noise data of the power devices from the sound data, and extracting fault characteristics of the power devices based on a time-varying characteristic curve of the sound data located at the noise data;

[0008] Locating the fault time node where the fault feature is extracted, and matching the electrical data of multiple power devices corresponding to the fault time node to obtain the fault electrical data of each power device;

[0009] Separating noise data from the sound data using a sound separation model to obtain noise data for each of the power devices, and determining a fault of each of the power devices based on a degree of matching between the noise data of each of the power devices, a time-varying characteristic curve of the sound data located to the noise data, and fault electrical data of each of the power devices;

[0010] Based on the abnormal data trend of the fault type of each power equipment fault, the fault trend of the power equipment is predicted, wherein the abnormal data trend is used to characterize the deviation of key parameters or indicators in the operation process of the power equipment from the expected trend, and the fault trend characterizes the development and change of the fault over time.

[0011] According to a second aspect of the present disclosure, a device for diagnosing and predicting faults of electric power equipment is provided, the device comprising:

[0012] an acquisition module, configured to acquire sound data and electrical data of a plurality of power devices, locate noise data of the power devices from the sound data, and extract fault characteristics of the power devices based on a time-varying characteristic curve of the sound data located at the noise data;

[0013] a matching module, configured to locate the fault time node from which the fault feature is extracted, and match the electrical data of multiple power devices corresponding to the fault time node to obtain the fault electrical data of each power device;

[0014] a determination module, configured to separate noise data from the sound data using a sound separation model to obtain noise data of each power device, and determine a fault of each power device based on a degree of matching between the noise data of each power device, a time-varying characteristic curve of the sound data located to the noise data, and fault electrical data of each power device;

[0015] A prediction module is used to predict the failure trend of the power equipment based on the abnormal data trend of the failure type of each power equipment failure, wherein the abnormal data trend is used to characterize the deviation of key parameters or indicators in the operation process of the power equipment from the expected trend, and the failure trend characterizes the development and change of the failure over time.

[0016] According to the third aspect of the present disclosure, an electronic device is provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the method of the aforementioned first aspect.

[0017] According to a fourth aspect of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to enable a computer to execute the method of the aforementioned first aspect.

[0018] The present invention provides a method, device, and electronic device for diagnosing and predicting faults in power equipment. Compared with the prior art, the present invention collects sound data and electrical data of multiple power equipment, locates the noise data of the power equipment from the sound data, and extracts the fault characteristics of the power equipment based on the time-varying characteristic curve of the sound data located in the noise data; locates the fault time node of the fault characteristics extracted, and matches the electrical data of multiple power equipment corresponding to the fault time node to obtain the fault electrical data of each power equipment; uses a sound separation model to separate the noise data from the sound data to obtain the noise data of each power equipment, and determines the fault of each power equipment based on the noise data of each power equipment, the time-varying characteristic curve of the sound data located in the noise data, and the degree of matching of the fault electrical data of each power equipment; predicts the fault trend of the power equipment based on the abnormal trend of the data of the fault type of each power equipment fault, wherein the abnormal trend of the data is used to characterize the deviation of key parameters or indicators from the expected trend during the operation of the power equipment, and the fault trend characterizes the development and change of the fault over time. By applying the scheme of the present invention, the fault time node of the sound data of the fault characteristics extracted is located, and the electrical data of multiple power equipment corresponding to the time node are matched to obtain the fault electrical data of each power equipment. This precise matching at each time point allows for a clearer analysis of the correlation between acoustic and electrical signals during partial discharge (PD) events, eliminating interference signals during non-fault periods and improving the reliability of PD detection. In PD detection scenarios, environmental interference signals often appear as noise in the sound data. Using a sound separation model, the true PD sound signature can be separated from the noise, reducing the impact of interference signals on PD sound signature extraction and improving the accuracy of the sound data used for PD detection. The sound signals generated by PD exhibit specific time-varying characteristics, while interference signals exhibit different time-varying characteristics. By analyzing the time-varying characteristic curves of the sound data, the PD fault signature can be accurately extracted, eliminating interference from interference signals and enabling more accurate PD fault detection. For PD detection, the combined consideration of acoustic and electrical information allows for a more comprehensive assessment of the presence and severity of a PD fault. Even in the presence of interference signals, this comprehensive approach allows for more accurate PD fault diagnosis, avoiding missed detections or misjudgments due to interference signals. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the present disclosure.

[0020] In order to more clearly illustrate the embodiments of the present disclosure or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0021] Figure 1 A schematic diagram of a flow chart of a method for diagnosing and predicting faults of power equipment provided by an embodiment of the present disclosure;

[0022] Figure 2 A schematic diagram of the overall process of a method for diagnosing and predicting faults of power equipment provided by an embodiment of the present disclosure;

[0023] Figure 3 A schematic diagram of the structure of a device for diagnosing and predicting faults in electric power equipment provided in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0024] In order to solve the technical problem that the current partial discharge detection technology is interfered with by interference signals, which reduces the accuracy and reliability of partial discharge detection and even leads to missed detection or misjudgment, the embodiments of the present disclosure provide a method for fault diagnosis and prediction of power equipment, such as Figure 1 As shown, the method includes:

[0025] Step 101: Collect sound data and electrical data of multiple power equipment, locate noise data of the power equipment from the sound data, and extract fault characteristics of the power equipment based on a time-varying characteristic curve of the sound data located to the noise data.

[0026] Sound data can include various sounds emitted by power equipment during operation. Sound data can be audio signals collected by sound sensors (such as microphones) installed near the equipment. For example, a transformer produces a "buzzing" sound due to the vibration of its core during operation. The frequency, pitch, and loudness of this sound vary with the equipment's operating state. Electric motors also produce specific mechanical vibration sounds during operation, with the sound varying under different loads and speeds. Sound data typically exists as an audio file or a digitized audio signal sequence, recording how the operating sound of the equipment changes over time.

[0027] Electrical data refers to data related to the electrical characteristics of power equipment. For generators, this data may include output voltage, current, power, and power factor. For transformers, in addition to input and output voltage and current, it may also include data such as winding temperature and oil temperature. For transmission lines, this data includes parameters such as line current, voltage, resistance, and reactance. This data is typically collected in real time using various electrical measuring instruments (such as voltmeters, ammeters, and power meters) or sensors, and transmitted and stored as digital or analog signals.

[0028] Noise data can be components of the collected sound data that are not related to the normal operation of power equipment. Noise data can come from a variety of sources, such as ambient noise, abnormal noise from the equipment itself, and sound caused by electromagnetic interference. In this embodiment, the power equipment noise data located from the sound data can be mixed noise data from multiple power equipment.

[0029] The time-varying characteristic curve can be used to describe the characteristics of sound data that change over time.

[0030] In the embodiment of the present disclosure, before collecting the sound data and electrical data of multiple power devices, as shown in FIG. Figure 2 As shown, the method further includes:

[0031] Construct a spatial location map containing multiple power devices. The spatial location map can be used to show the distribution of multiple power devices in space.

[0032] According to the spatial location diagram, multiple information collection points are arranged. The information collection points can be used to represent the data collection positions of the sound data and electrical data of multiple power equipment.

[0033] In this embodiment, a spatial model of the combination electrical device to be tested is constructed by constructing a spatial location map of multiple electrical devices. When the model is input, the spatial distribution map of the electrical devices is labeled with the location coordinates of each device. Simultaneously, when designing the actual detection system, multiple collection points with information collection units can be deployed based on the distribution of the electrical devices, and the placement of the collection units can be determined. The provision of sound collection devices ensures that the sound signals of each device can be effectively collected. Electrical collection devices that measure electrical information can be deployed at the collection points to collect electrical data from different devices in close proximity, such as voltage, current, and power. A collection point can be set at least at each device to be tested, each including at least a sound collection device. Multiple collection points can be deployed with only one set of electrical collection devices, or each collection point can be deployed with an electrical collection device. The electrical collection device can be a single device or a comprehensive measurement and control device that integrates a single measurement and control plug-in with a sound collection plug-in and other functional plug-ins.

[0034] After multiple collection points with information collection functions are deployed according to the spatial distribution of power equipment, the multiple information collection points can be used to synchronously collect sound data (such as sound signals generated by equipment operation) and electrical data (such as voltage, current, power, etc., which reflect the electrical status of the equipment) of multiple power equipment.

[0035] By constructing spatial location maps and spatial models and marking the equipment location coordinates, the location of power equipment can be accurately determined, and then the collection points can be reasonably arranged to ensure the comprehensive and accurate collection of sound data (such as sound signals generated by equipment operation) and electrical data (such as voltage, current, power, etc. that reflect the electrical status of the equipment) of multiple power equipment, providing a reliable data basis for subsequent analysis.

[0036] In the embodiment of the present disclosure, before synchronously collecting sound data and electrical data of multiple power devices based on multiple information collection points, the method further includes:

[0037] Each power equipment in the spatial position diagram can be divided into regions according to the equipment size, noise intensity and / or noise influence range of each power equipment to obtain the sound collection space area of ​​each power equipment. The sound collection space area may include the area of ​​each power equipment divided according to the equipment size, noise intensity and / or noise influence range with the equipment center as the coordinate center, as well as the typical noise source area of ​​each power equipment.

[0038] Based on multiple information collection points and sound collection space areas, the sound data and electrical data of multiple power equipment can be collected synchronously.

[0039] In this embodiment, within the GIS switchgear site where the power equipment is located, a measurement tool (such as a total station or laser rangefinder) can be used to accurately measure the location of each power device and record the center coordinates (x, y, z) of each device. For example, for a transformer, circuit breaker, disconnector, or other device within a GIS substation, the location of each device's center point within the site coordinate system can be measured.

[0040] If the site has a ready-made equipment layout diagram (such as a CAD drawing), you can directly obtain the center coordinates and approximate spatial distribution of the equipment from the drawing. Then, based on the scale of the drawing, convert the equipment location information into actual coordinate values.

[0041] Areas likely to generate typical noise can be identified based on the type and operating principle of the power equipment, combined with past experience. For example, transformers generate electromagnetic noise during operation, and strong noise fields may be present around them. High-voltage switchgear also generates transient noise during operation, and the surrounding areas also need to be considered.

[0042] During equipment operation, conduct preliminary noise monitoring of the site using a portable noise meter. Measure noise intensity at different locations to identify areas with high and representative noise intensities. These areas can be used as typical noise source areas.

[0043] For each electrical device, a circular or rectangular area can be created based on the device's size and noise propagation range, with the device's center as the coordinate center. For example, a large transformer can be divided into a sound collection area with a radius of 5-10 meters (adjustable based on actual conditions) around its center. If the device is more regularly shaped, a rectangular area can also be used, with the length and width determined based on the device's dimensions and noise propagation direction.

[0044] For the typical noise source areas identified, separate regions can be divided based on the noise source's intensity and impact range. If the noise source is localized, such as the noise generated by a cooling fan on a piece of equipment, a smaller region can be used; if the noise source is widespread, such as the noise generated by the entire high-voltage busbar area, a larger region can be used.

[0045] If the zones overlap or are too close together, consider merging or adjusting their boundaries. For example, the zones of two adjacent power equipment may partially overlap. In this case, based on the actual noise distribution, the zone boundaries can be readjusted to ensure that each zone covers the primary noise source while minimizing interference between zones.

[0046] Based on multiple information collection points and a clear sound collection space, the sound and electrical data of multiple power equipment can be collected synchronously and efficiently. This organized collection method avoids blind collection and improves the efficiency and accuracy of data collection.

[0047] like Figure 2 As shown, within the time interval [T0, T1], all sound collection devices can be started to ensure that they collect sound data synchronously within the same time range. Synchronization of multiple collection devices can be achieved through external synchronization signals (such as GPS clock signals), or using the synchronization function built into the collection devices.

[0048] Match the center coordinates of the divided sound collection area with the center coordinates of the power equipment. If the center coordinates of a sound collection area are close to the center coordinates of a power equipment (within a certain error range), then the area is matched with the equipment. For example, the center coordinates of a sound collection area are (x1, y1), which are close to the center coordinates of a transformer equipment. match( 、 is the allowable error range), then this area corresponds to the transformer equipment.

[0049] For typical noise source areas, the corresponding device or device combination can be determined based on the source of the noise. For example, if the noise in an area primarily comes from the cooling fan of a transformer, then this area corresponds to that transformer. If the noise in an area is generated by multiple devices, such as the noise in the high-voltage bus area, which may be generated by multiple connected devices, then this area corresponds to a combination of these devices.

[0050] like Figure 2 As shown, multiple power equipment can be collected Sound data and electrical data at the moment; in the time interval The sound data and electrical data of multiple power equipment are collected synchronously. The sound data sampling rate is set to The electrical data sampling rate is set according to the characteristics of the equipment. After processing, the collected data format of each power equipment can be obtained by separation:

[0051] Sound data: , where N is the number of spatial regions where sound data were collected;

[0052] Electrical data: , where M is the number of electrical equipment for which electrical data is collected.

[0053] It should be pointed out that N is not necessarily equal to M. Sound data is more focused on regional division. That is, in the spatial location diagram of multiple power equipment, multiple power equipment are divided by region, and the center coordinates and regional range are marked. Each power equipment has an independent regional range. The sound collection space area not only includes the area corresponding to each power equipment with the center of the equipment as the coordinate center, but also includes other areas where typical noise sources exist.

[0054] Since acoustic and electrical data are collected synchronously, there may be slight time offsets, necessitating time alignment. This can be adjusted by analyzing characteristic points in the data (such as sudden changes in acoustic signals and electrical parameters during power equipment operation) to ensure perfect temporal alignment. This approach allows for focused time-domain analysis of fault points in subsequent model inputs.

[0055] The sound data S(t) corresponding to each sound collection area is associated with the electrical data E(t) of the power equipment corresponding to the area. For example, for a sound collection area corresponding to a transformer device, its sound data (assuming there is only one microphone in the area) and the electrical data of the transformer (Assuming that two electrical parameters, voltage and current, are collected) are associated to form a complete data set for subsequent analysis and processing.

[0056] For the embodiments of the present disclosure, Figure 2 As shown, after collecting the sound data and electrical data of multiple power equipment, the noise data of the power equipment can be located from the sound data, which may include:

[0057] Calculating the energy of each channel sound signal of the sound data, detecting energy mutation points according to the energy of each channel sound signal, and locating noise data in the sound data according to the energy mutation points; or,

[0058] Frequency domain analysis is performed on the sound signal of each channel of the sound data to obtain the spectrum distribution of the sound signal of each channel, and abnormal frequency components in the spectrum distribution are identified to locate the noise data in the sound data based on the abnormal frequency components. The abnormal frequency components are frequency components that are greater than or less than a preset frequency component range.

[0059] In this embodiment, the energy of the sound signal of each channel is calculated, and the specific formula is as follows:

[0060]

[0061] in, It can represent the energy of the i-th signal at time t, It can represent the amplitude of the i-th signal at time t, the square of the amplitude of each sampling point (modulus square), that is, the square operation is to ensure that the energy value is always non-negative, N can represent the number of sampling points, and n can represent the index variable of the summation, from 1 to N, which can be used to traverse all sampling points.

[0062] By analyzing the energy changes of the sound signal, we can find the points where the energy suddenly increases or decreases. These points may be the beginning or end of the noise. By detecting these energy mutation points, we can locate the time period when the noise data appears.

[0063] The sound signal is analyzed in the frequency domain through the short-time Fourier transform. Among them, the short-time Fourier transform can be used to convert the time domain signal into the time-frequency domain. Its basic idea is to divide the signal into multiple short-time segments and perform the Fourier transform on each segment to obtain the distribution information of the signal in time and frequency. The specific formula is as follows:

[0064]

[0065] in, It can represent the short-time Fourier transform result, which represents the signal strength or spectrum density at time t and frequency f. It can represent the original sound signal, which changes with time t. It can be represented by the short-time Fourier transform operation, which is to transform the signal The process of converting to time and frequency domain representation.

[0066] One possible method is to use the results of a short-time Fourier transform to determine the spectral distribution of each channel's sound signal. Specifically, for each time point t, the signal strength or spectral density corresponding to different frequencies f at that time point can be obtained, thereby plotting a spectrogram at that time point. By analyzing spectrograms at multiple time points, the spectral distribution of the entire sound signal can be determined.

[0067] Preset frequency component range can be set , which can be determined based on the normal frequency characteristics of the sound signal. For example, for human voice signals, the frequency range is usually If the frequency component range is between , then the preset frequency component range can be set to this interval. In the spectrum distribution, frequency components that are greater than the maximum frequency or less than the minimum frequency in the preset frequency component range are identified and regarded as abnormal frequency components. Abnormal frequency components can be caused by noise, because the frequency components of normal sound signals are generally concentrated in the preset frequency range.

[0068] Based on the identifiable abnormal frequency components, the noise data within the sound data is located. Specifically, in the time-frequency domain, the time and frequency locations corresponding to the abnormal frequency components are found. For each abnormal frequency component, the time period and frequency range in which it occurs are recorded. The sound data corresponding to these time periods and frequency ranges are the noise data.

[0069] For the embodiments of the present disclosure, Figure 2 As shown, after the noise data of the power equipment is located, the fault characteristics of the power equipment can be extracted according to the time-varying characteristic curve of the sound data located in the noise data, which may specifically include:

[0070] Calculate the short-time energy and short-time zero-crossing rate of the sound data. The short-time energy is used to reflect the energy of the sound signal of each channel of the sound data within a short time window. The short-time zero-crossing rate is used to reflect the frequency change characteristics of the sound signal of each channel of the sound data within a short time window.

[0071] Determine the time-varying characteristic curve of the sound data according to the short-time energy and the short-time zero-crossing rate;

[0072] Based on the Mel-frequency cepstral coefficients and combined with the time-varying characteristic curve of sound data, the fault characteristics of the corresponding power equipment are extracted.

[0073] In this embodiment, the time-varying characteristic curve of the sound data is extracted. Specifically, the time-varying characteristic curve can be obtained by calculating the short-time energy and the short-time zero-crossing rate. The calculation formula of the short-time energy is as follows:

[0074]

[0075] in, It can represent the short-term energy in the window starting at time point t, It can represent the amplitude of the signal at time point n, t can represent the starting point of the current time window, and W can represent the length of the time window. It can be expressed from time point t to time point All discrete time points n between Perform accumulation operation.

[0076] By framing the sound signal and calculating the short-time energy of each frame, a time-varying characteristic curve is generated, showing how the short-time energy changes over time. In fault diagnosis, equipment failures can generate sudden noise bursts that can cause significant changes in short-time energy. Therefore, the short-time energy time-varying characteristic curve can be used to detect abnormalities in equipment operation.

[0077] The calculation formula of short-time zero-crossing rate is as follows:

[0078]

[0079] in, It can be expressed as the short-term zero-crossing rate within the window starting at time point t, It can represent symbolic functions, It can represent the amplitude of the signal at time point n, t can represent the start of the current time window, and W can represent the length of the time window. It can be expressed from time point t to time point All discrete time points n between Perform accumulation operation.

[0080] By framing the sound signal and calculating the short-term zero-crossing rate for each frame, a time-varying characteristic curve is generated, showing how the short-term zero-crossing rate changes over time. Changes in the short-term zero-crossing rate can also reflect abnormalities in the equipment's operating status. For example, a device failure may cause the frequency components of the sound signal to change, resulting in a change in the short-term zero-crossing rate.

[0081] Among them, the symbol function Defined as:

[0082]

[0083] The zero-crossing rate of each window can be calculated by sliding the time window to obtain the short-term zero-crossing rate time-varying characteristic curve.

[0084] Mel-frequency cepstral coefficients are a commonly used method for extracting sound features, simulating the nonlinear perception of the human ear to sound frequencies. The sound signal is first preprocessed with operations such as pre-emphasis, framing, and windowing. A fast Fourier transform (FFT) is then performed to obtain the spectrum, which is then filtered through a Mel filter bank to obtain the Mel spectrum. The Mel spectrum is then subjected to logarithmic operations and discrete cosine transforms (DCTs) to ultimately obtain the Mel-frequency cepstral coefficients. MFCCs can reflect the cepstral characteristics of sound signals on the Mel-frequency scale and are effective for distinguishing different types of fault sounds. The Mel-frequency cepstral coefficients are expressed as follows:

[0085]

[0086] in, It can represent the i-th Mel frequency cepstral coefficient at time point t, The Mel-frequency cepstrum coefficient representing the amplitude of the i-th signal at time t.

[0087] Time-frequency domain features can be extracted by obtaining the local maxima and minima of the time-spectrum. The time-spectrum is generated by performing a short-time Fourier transform on the sound signal. Local maxima and minima are then searched for within the time-spectrum. These local extreme points contain important time-frequency domain characteristics of the sound signal, such as the energy concentration of the fault sound at specific times and frequencies.

[0088] Step 102: locate the fault time node of the extracted fault feature, and match the electrical data of multiple power devices corresponding to the fault time node to obtain the fault electrical data of each power device.

[0089] For the embodiments of the present disclosure, Figure 2 As shown, when the extracted fault feature parameter exceeds the preset threshold, it is determined that the fault feature is detected, and the time node at this time is recorded, that is, the time node when the fault feature is located and extracted.

[0090] Each electrical equipment in the power system is equipped with a corresponding electrical monitoring device, which can collect the electrical data of the equipment in real time, such as current, voltage, power, frequency, etc., and store this data in the database. The data is recorded in time series.

[0091] like Figure 2 As shown, according to the located fault time node, the electrical data of all power equipment corresponding to the fault time node can be queried and extracted from the database. Specifically, for each power equipment, the electrical data record closest in time is queried in the database to obtain the fault electrical data corresponding to each power equipment at the fault time node, wherein the fault electrical data represents the fault electrical data at the fault time node. The fault electrical data of the power equipment is a collection of fault electrical data. The fault electrical data is shown in the following formula:

[0092]

[0093] in, Can represent a collection of fault electrical data, It can represent the failure time node, may represent the i-th faulty electrical data sample in the faulty electrical data set.

[0094] Step 103: Use the sound separation model to separate the noise data from the sound data to obtain the noise data of each power device, and determine the fault of each power device based on the noise data of each power device, the time-varying characteristic curve of the sound data located to the noise data, and the degree of matching of the fault electrical data of each power device.

[0095] The noise data of each electric device may be single noise data separated from mixed noise data in the sound data.

[0096] For the embodiments of the present disclosure, Figure 2 As shown, before using the sound separation model to separate the noise data from the sound data to obtain the noise data of each power device, the method further includes:

[0097] Through deep learning models, a sound separation model is constructed by combining convolutional neural networks and generative adversarial networks;

[0098] Data features are extracted from the noise data, and the data features are input into a trained sound separation model to separate the noise data from the sound data using the sound separation model to obtain the noise data of each power device.

[0099] In this embodiment, a sound separation model of multiple power equipment in the spatial position map is obtained by a deep learning model, and the noise data of each power equipment is separated by the sound separation model; specifically, a deep learning model is used to construct a sound separation model using a convolutional neural network (CNN) and a generative adversarial network (GAN), and the input is mixed sound data. , output as separated single device sound data ;

[0100] The model training goal is to minimize the separation error:

[0101]

[0102] in, It can represent the objective function of model training, namely separation error, It can represent the real sound data of a single device at time t for the i-th sample, It can represent the sound data of a single device at time t for the ith sample after separation.

[0103] A sound separation model is constructed using a convolutional neural network (CNN) and a generative adversarial network (GAN). By combining the feature extraction capabilities of CNN with the generative adversarial mechanism of GAN, a more stable sound separation model is constructed. CNN extracts high-quality features, and GAN generates sounds that are closer to real signals through adversarial training. This fusion method can improve the sound separation effect.

[0104] In the feature extraction stage, CNN can be used to extract the spectral features (i.e., data features) of the sound signal. The convolutional layer of CNN captures local features in the audio spectrogram, and the pooling layer reduces the dimension of the features. The extracted spectral features are used as the input of the subsequent GAN to enhance the generator's understanding of the sound features. In the generative adversarial stage, the generator receives the spectral features and random noise extracted by CNN to generate a separated sound signal. The discriminator distinguishes the generated sound signal from the real sound signal and provides feedback to the generator. During the training process, the generator and discriminator are optimized alternately. The generator tries to generate sounds that are closer to the real signal, while the discriminator tries to distinguish between real and fake signals.

[0105] Through a joint training strategy, the CNN component is trained independently at the beginning of training to extract high-quality features. The CNN is then combined with the GAN for end-to-end training. A joint loss function is used to constrain the sound signal generated by the generator to be closer in features to the real signal. Multi-task learning is used to simultaneously optimize both the accuracy of sound separation and the quality of the generated signal. The joint loss function is designed to first perform a reconstruction loss, using either mean squared error (MSE) or L1 loss to measure the difference between the generated signal and the target signal. The discriminator's output is used as the loss to encourage the generator to produce more realistic signals to counteract the loss. A feature matching loss is used to further constrain the generator by comparing the outputs of the generated and real signals at the CNN feature layer.

[0106] Furthermore, as a necessary step, we pre-trained a CNN on a large-scale sound dataset to extract features. We alternately trained the generator and discriminator, using techniques such as gradient penalties to stabilize the training process. We used techniques such as dropout and batch normalization in both the generator and discriminator to prevent overfitting and achieve regularization. We also adjusted hyperparameters such as the learning rate and the generator and discriminator update frequency to optimize model performance.

[0107] An audio dataset containing a mixture of multiple sounds is used for training and testing. The separation effect is then evaluated using indicators such as signal-to-noise ratio (SNR) and signal-to-distortion ratio (SDR). The results are compared with methods that use CNN or GAN alone to verify the performance of the fusion model and correct the model.

[0108] For the embodiments of the present disclosure, Figure 2 As shown, the fault of each power device can be determined based on the noise data of each power device, the time-varying characteristic curve of the sound data located to the noise data, and the matching degree of the fault electrical data of each power device, which may specifically include:

[0109] Calculate the matching degree between the noise data of each power device and the fault electrical data of each power device and the time-varying characteristic curve by using the correlation coefficient method and the Euclidean distance method;

[0110] If the matching degree exceeds a preset matching degree threshold, it is determined that a corresponding fault exists in the power equipment, wherein the preset matching degree threshold may be pre-set according to matching degree values ​​of a normal state and a fault state.

[0111] In this embodiment, a relatively accurate fault diagnosis can be achieved solely based on the time-varying characteristic curves of fault electrical and acoustic data. For early-stage faults, acoustic feature separation can detect subtle anomalies in equipment operation, enabling earlier detection than traditional electrical monitoring methods. Furthermore, combined with comprehensive consideration of the amount of electrical information, this approach can provide a more sensitive early warning reference for operations departments or automated monitoring systems. Specifically, the acoustic separation model can be used to isolate the noise data of each electrical device, serving both as input for fault diagnosis and as a reference for fault verification.

[0112] The correlation coefficient method is used to calculate the matching degree to evaluate the similarity between the fault electrical data and the separated noise data. Specifically, the matching degree is calculated by calculating the correlation coefficient and the Euclidean distance.

[0113] The correlation coefficient is a statistic that measures the degree of linear correlation between two variables. In sound separation and fault diagnosis, the correlation coefficient is used to evaluate the similarity between separated noise data and fault electrical data.

[0114] The formula for the Pearson Correlation Coefficient is:

[0115]

[0116] in, It can be expressed as the Pearson correlation coefficient, can represent the i-th observation value of the first variable, can represent the i-th observation value of the second variable, represents the mean of all observed values ​​of the first variable, represents the mean of all observations of the second variable, and n represents the number of observations.

[0117] Euclidean distance measures the straight-line distance between two points in Euclidean space. In sound separation and fault diagnosis, Euclidean distance can be used to evaluate the difference between separated noise data and fault electrical data.

[0118] The formula for Euclidean distance is:

[0119]

[0120] in, It can represent the Euclidean distance between two points. It can represent the i-th coordinate of the first point, It represents the i-th coordinate of the second point, and n represents the number of coordinates.

[0121] For the embodiment of the present disclosure, before calculating the degree of matching between the noise data of each power device and the fault electrical data of each power device and the time-varying characteristic curve using the correlation coefficient method and the Euclidean distance method, the method further includes:

[0122] Adopting heterogeneous data fusion method, the noise data of each power equipment and the fault electrical data of each power equipment are processed separately, and then the noise data of each power equipment and the fault electrical data of each power equipment are fused through the defined association rules or models;

[0123] The matching degree between the fused noise data of each power device and the fault electrical data of each power device and the time-varying characteristic curve is calculated by using the correlation coefficient and the Euclidean distance.

[0124] In this embodiment, in the calculation of the matching degree, electrical data and noise data are two different types of data, which may have great differences in physical meaning, data structure and dimension. Therefore, directly calculating the matching degree of electrical data and noise data may lead to misleading results. Therefore, when processing electrical data and noise data, a heterogeneous data fusion method is adopted to process electrical data and noise data separately and fuse them through defined association rules or models. The association rules between electrical data and noise data are defined. When the current in the electrical data suddenly increases, the short-term energy in the noise data also increases significantly, which solves the problem of misleading results that may be caused by direct matching and improves the availability and accuracy of the data. The rule is expressed as shown in the following formula:

[0125]

[0126] in, It can represent the current change in electrical data. It can represent the preset current change threshold, It can represent the preset noise short-time energy change threshold. It can represent the short-term energy variation in noise data.

[0127] The electrical data and noise data are jointly analyzed to find the correlation between them. The principal component analysis (PCA) method is used to project the two types of data into a common space for analysis.

[0128] PCA is a dimensionality reduction technique that projects data onto a new coordinate system through linear transformation, maximizing the variance along the new coordinate axes (principal components). PCA can be used to extract common features from both electrical and noise data. PCA not only reduces the data's dimensionality but also preserves key trends, simplifying subsequent analysis and improving efficiency.

[0129] Specifically, the data is first standardized. The electrical data and noise data are standardized separately to make their mean 0 and standard deviation 1, so as to eliminate the dimensional differences between the data and improve the consistency of the data, as shown in the following formula:

[0130]

[0131] in, It can represent the data matrix after normalization. can represent the original data matrix, It can be represented as the mean vector, represents a standard deviation vector.

[0132] Next, we construct a data matrix and concatenate the electrical data and noise data into a matrix X, as shown in the following formula:

[0133]

[0134] in, can represent a data matrix, Can represent electrical data matrix, can represent a noisy data matrix.

[0135] Then calculate the covariance matrix and calculate the data matrix The covariance matrix of , as shown in the following formula:

[0136]

[0137] in, It can be expressed as the covariance matrix, can represent a data matrix, Representable data matrix The transpose of It can represent the number of samples, It can represent the normalization coefficient.

[0138] Solve the eigenvalues ​​and eigenvectors of the variance matrix as shown in the following formula:

[0139]

[0140] in, It can represent the eigenvalue of the covariance matrix, It can represent the i-th eigenvalue of the covariance matrix, It can be represented as the i-th eigenvector of the covariance matrix.

[0141] Select the eigenvectors corresponding to the first k largest eigenvalues ​​to form the projection matrix , as shown in the following formula:

[0142]

[0143] The original data matrix Projecting into the principal component space to obtain the reduced dimensionality data :

[0144]

[0145] in, Can represent the data after dimensionality reduction, can represent the original data matrix, It can be represented as a projection matrix consisting of the first k eigenvectors of the covariance matrix.

[0146] Data after dimensionality reduction by PCA It can be used for subsequent fault diagnosis analysis. The principal components extracted by PCA can capture the main change trends in electrical data and noise data.

[0147] By applying the curve correspondence analysis, the matching degree of the fault electrical data, the separated noise data and the time-varying characteristic curve is obtained. The occurrence of a fault is determined based on the matching degree, as shown in the following formula:

[0148]

[0149] in, It can represent the matching degree of the i-th group of data and can be used to measure the matching degree between the fault electrical data and the separated noise data in terms of the time-varying characteristic curve. The matching value can reflect the strength of the similarity or correlation between the data. It can represent the correlation coefficient calculation function, which is used to calculate the correlation between two data series. It can represent the i-th data in the fault electrical data sequence, It can represent the failure time node, It can represent the fault time node in the separated noise data Related i-th data.

[0150] In a multi-device environment, fault location is a challenge. Using a sound separation model and spatial location maps, faulty devices can be precisely located. Furthermore, a relatively accurate fault diagnosis can be achieved solely based on the time-varying characteristic curves of the fault's electrical and acoustic data. For early-stage faults, sound feature separation can capture subtle anomalies in device operation, enabling earlier detection than traditional electrical monitoring methods. Furthermore, combined with comprehensive consideration of the amount of electrical information, this approach can provide a more sensitive early warning reference for operations departments or automated monitoring systems. Specifically, the sound separation model can be used to separate the noise data of each electrical device, serving both as input for fault diagnosis and as a reference for fault verification.

[0151] This matching calculation method provides a quantitative evaluation metric, making fault diagnosis more objective and accurate. Furthermore, by considering the matching degree between the time-varying characteristic curves of the fault electrical data and the separated noise data, it can more comprehensively reflect the operating status of the equipment and improve the reliability of fault diagnosis.

[0152] Step 104: predict the failure trend of the power equipment based on the abnormal data trend of the fault type of each power equipment fault, wherein the abnormal data trend is used to characterize the deviation of key parameters or indicators in the operation process of the power equipment from the expected trend, and the failure trend characterizes the development and change of the failure over time.

[0153] For the embodiments of the present disclosure, Figure 2 As shown, the fault characteristics of each power device can be classified into different fault types based on preset summary values, and different fault type thresholds can be defined for each fault type. The preset summary values ​​can be characteristic patterns or standards derived from analyzing and summarizing power device data of known fault types, such as fault electrical data (e.g., specific current and voltage change patterns), noise data (e.g., the presence of specific frequency noise), and time-varying characteristic curves (e.g., curve shape and slope change characteristics). Fault types can be classified into different categories, such as short circuit faults, open circuit faults, and insulation faults. Different fault types correspond to different fault characteristic manifestations, which are defined and differentiated by the preset summary values.

[0154] The fault type threshold determines whether the fault characteristics are significant enough to qualify as a fault type. Different fault types have different fault type thresholds due to their different fault mechanisms and manifestations. For example, a short-circuit fault might have a relatively high fault type threshold due to its large current fluctuations; whereas an insulation fault might have a relatively low fault type threshold due to its relatively subtle characteristic changes.

[0155] If the fault characteristic value exceeds the fault type threshold of the corresponding fault type, it is determined that the corresponding fault type has occurred in the power equipment, as shown in the following formula, where the fault characteristic value can be extracted from the sound data, electrical data, etc. actually collected from the power equipment, and is used to reflect the fault characteristics of the equipment.

[0156]

[0157] in, Can indicate the fault type of power equipment, 、 ... Can indicate different fault types, for example, It may be a short circuit fault. It may be an insulation fault, It can be other types of faults, etc. The definition of specific fault types can be determined according to the actual fault conditions and classification standards of the power equipment. Can represent fault characteristic value, It can indicate the fault type threshold corresponding to different fault types.

[0158] In the disclosed embodiments, different fault types can manifest differently in power equipment operating data (such as electrical data and acoustic data). For example, a short circuit fault may cause a sharp increase in current, while an insulation fault may increase leakage current or decrease insulation resistance. The changes in these abnormal data over time constitute abnormal data trends. By analyzing these trends, we can understand the characteristics of the fault at different time points, providing a basis for determining fault trends.

[0159] Time series analysis can be used to analyze abnormal trends in fault data. Time series analysis is a statistical method for analyzing data arranged in chronological order. In power equipment fault analysis, fault characteristics are arranged in chronological order to form time series data. Using time series analysis, the changes in fault characteristics at different time points can be calculated to obtain the temporal trend of fault characteristics, as shown in the following formula:

[0160]

[0161] in, Can represent fault characteristics In time The change in the current moment With the previous moment Compared with the fault characteristics The increment of Can represent the time Moment, The value of the fault characteristic, Can be expressed in time Moment, The value of the fault feature is the fault feature value at the previous moment.

[0162] if Continuous increase means that the fault characteristic is changing more and more with time, indicating that the fault may be getting worse and the equipment is getting worse. For example, if the fault characteristic is the current change, A continued increase may mean that the short circuit fault is getting worse and the current is getting larger. Stability indicates that the change in the fault characteristic over time gradually decreases and remains within a certain range. This means that the fault characteristic value no longer changes significantly, indicating that the fault has entered a stable phase. Although the equipment is faulty, the fault's development has temporarily stopped or is progressing very slowly. For example, a stable change in the leakage current during an insulation fault may indicate that the extent of insulation damage has not yet expanded. In summary, according to the power equipment fault diagnosis and prediction method provided by the present disclosure, compared with the prior art, the present disclosure collects sound data and electrical data of multiple power equipment, locates the noise data of the power equipment from the sound data, and extracts the fault characteristics of the power equipment according to the time-varying characteristic curve of the sound data located to the noise data; locates the fault time node of the fault feature extracted, and matches the electrical data of multiple power equipment corresponding to the fault time node to obtain the fault electrical data of each power equipment; uses the sound separation model to separate the noise data from the sound data to obtain the noise data of each power equipment, and determines the fault of each power equipment according to the noise data of each power equipment, the time-varying characteristic curve of the sound data located to the noise data, and the matching degree of the fault electrical data of each power equipment; predicts the fault trend of the power equipment according to the abnormal trend of the data of the fault type of each power equipment fault, wherein the abnormal trend of the data is used to characterize the deviation of the key parameters or indicators of the operation process of the power equipment from the expected trend, and the fault trend characterizes the development and change of the fault over time. By applying the scheme of the present disclosure, the fault time node of the sound data of the fault feature extracted is located, and the electrical data of multiple power equipment corresponding to the time node are matched to obtain the fault electrical data of each power equipment. This precise matching at each time point allows for a clearer analysis of the correlation between acoustic and electrical signals during partial discharge (PD) events, eliminating interference signals during non-fault periods and improving the reliability of PD detection. In PD detection scenarios, environmental interference signals often appear as noise in the sound data. Using a sound separation model, the true PD sound signature can be separated from the noise, reducing the impact of interference signals on PD sound signature extraction and improving the accuracy of the sound data used for PD detection. The sound signals generated by PD exhibit specific time-varying characteristics, while interference signals exhibit different time-varying characteristics. By analyzing the time-varying characteristic curves of the sound data, the PD fault signature can be accurately extracted, eliminating interference from interference signals and enabling more accurate PD fault detection. For PD detection, the combined consideration of acoustic and electrical information allows for a more comprehensive assessment of the presence and severity of a PD fault. Even in the presence of interference signals, this comprehensive approach allows for more accurate PD fault diagnosis, avoiding missed detections or misjudgments due to interference signals.

[0163] Based on the above Figure 1The specific implementation of the method shown in this embodiment provides a power equipment fault diagnosis and prediction device, such as Figure 3 As shown, the device includes: a collection module 31, a matching module 32, a determination module 33, and a prediction module 34;

[0164] An acquisition module 31 is configured to acquire sound data and electrical data of a plurality of power devices, locate noise data of the power devices from the sound data, and extract fault characteristics of the power devices based on a time-varying characteristic curve of the sound data located at the noise data;

[0165] a matching module 32 for locating the fault time node from which the fault feature is extracted, and matching the electrical data of multiple power devices corresponding to the fault time node to obtain the fault electrical data of each power device;

[0166] a determination module 33 configured to separate noise data from the sound data using a sound separation model to obtain noise data of each power device, and determine a fault of each power device based on a matching degree between the noise data of each power device, a time-varying characteristic curve of the sound data located to the noise data, and fault electrical data of each power device;

[0167] The prediction module 34 is used to predict the failure trend of the power equipment based on the abnormal data trend of the failure type of each power equipment failure, wherein the abnormal data trend is used to characterize the deviation of key parameters or indicators in the operation process of the power equipment from the expected trend, and the failure trend characterizes the development and change of the failure over time.

[0168] In specific application scenarios, such as Figure 3 As shown, the device further includes: a first construction module 35, a layout module 36;

[0169] A first constructing module 35 is configured to construct a spatial location diagram including the plurality of power devices, wherein the spatial location diagram is configured to display the distribution of the plurality of power devices in space;

[0170] The layout module 36 is configured to layout a plurality of information collection points according to the spatial location diagram, wherein the information collection points are used to represent data collection locations of the sound data and electrical data of the plurality of power devices.

[0171] In a specific application scenario, the acquisition module 31 can be used to synchronously acquire the sound data and electrical data of the multiple power devices based on the multiple information acquisition points.

[0172] In specific application scenarios, such as Figure 3 As shown, the device further includes: a division module 37;

[0173] The division module 37 is used to divide each power device in the spatial position diagram into regions to obtain a sound collection spatial region of each power device, wherein the sound collection spatial region includes an area of ​​each power device divided with the device center as the coordinate center and a typical noise source area of ​​each power device.

[0174] In a specific application scenario, the acquisition module 31 may be used to synchronously acquire the sound data and electrical data of the plurality of power devices based on the plurality of information acquisition points and the sound acquisition spatial area.

[0175] In a specific application scenario, the acquisition module 31 may be used to calculate the energy of the sound signal of each channel of the sound data, detect energy mutation points according to the energy of the sound signal of each channel, and locate noise data in the sound data according to the energy mutation points; or,

[0176] Perform frequency domain analysis on the sound signal of each channel of the sound data to obtain a spectrum distribution of the sound signal of each channel, and identify abnormal frequency components in the spectrum distribution to locate noise data in the sound data based on the abnormal frequency components, where the abnormal frequency components are frequency components greater than or less than a preset frequency component range.

[0177] In a specific application scenario, the acquisition module 31 can be used to calculate the short-time energy and short-time zero-crossing rate of the sound data, where the short-time energy is used to reflect the energy of the sound signal of each channel of the sound data within a short time window, and the short-time zero-crossing rate is used to reflect the frequency change characteristics of the sound signal of each channel of the sound data within a short time window;

[0178] determining a time-varying characteristic curve of the sound data according to the short-time energy and the short-time zero-crossing rate;

[0179] Based on the Mel-frequency cepstral coefficients and in combination with the time-varying characteristic curve of the sound data, the fault characteristics of the corresponding power equipment are extracted.

[0180] In specific application scenarios, such as Figure 3 As shown, the device further includes: a second building module 38;

[0181] The second building module 38 is used to build the sound separation model through a deep learning model combined with a convolutional neural network and a generative adversarial network.

[0182] In a specific application scenario, the determination module 33 can be used to extract data features from the noise data, input the data features into the trained sound separation model, and obtain the noise data of each power device.

[0183] In a specific application scenario, the determination module 33 may be configured to calculate the degree of matching between the noise data of each power device and the fault electrical data of each power device and the time-varying characteristic curve by using a correlation coefficient method and an Euclidean distance method;

[0184] If the matching degree exceeds a preset matching degree threshold, it is determined that a corresponding fault exists in the power equipment.

[0185] In specific application scenarios, such as Figure 3 As shown, the device further includes: a processing module 39;

[0186] The processing module 39 is used to adopt a heterogeneous data fusion method to process the noise data of each power device and the fault electrical data of each power device separately, and to fuse the noise data of each power device and the fault electrical data of each power device through a defined association rule or model.

[0187] In a specific application scenario, the determination module 33 may be used to calculate the matching degree between the fused noise data of each power device and the fault electrical data of each power device and the time-varying characteristic curve through correlation coefficient and Euclidean distance.

[0188] It should be noted that for other corresponding descriptions of the functional units involved in the power equipment fault diagnosis and prediction device provided in this embodiment, please refer to Figure 1 The corresponding description in will not be repeated here.

[0189] Based on the above Figure 1 The method shown in FIG. 1 is a method for performing the above-mentioned steps. Accordingly, this embodiment further provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the computer program can realize the above-mentioned steps. Figure 1 The method shown.

[0190] Based on this understanding, the technical solution of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, USB flash drive, mobile hard disk, etc.), and includes a number of instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute the methods of various implementation scenarios of the present disclosure.

[0191] Based on the above Figure 1 The method shown, and Figure 3 In order to achieve the above-mentioned purpose, the embodiment of the present disclosure further provides an electronic device, which includes a storage medium and a processor; the storage medium is used to store a computer program; the processor is used to execute the computer program to achieve the above-mentioned Figure 1 The method shown.

[0192] Optionally, the physical device may also include a user interface, a network interface, a camera, a radio frequency (RF) circuit, a sensor, an audio circuit, a Wi-Fi module, and the like. The user interface may include a display screen and an input unit such as a keyboard. Optional user interfaces may also include a USB interface and a card reader interface. Optionally, the network interface may include a standard wired interface or a wireless interface (such as a Wi-Fi interface).

[0193] Those skilled in the art will understand that the above-mentioned physical device structure provided in this embodiment does not constitute a limitation on the physical device, and may include more or fewer components, or a combination of certain components, or different component arrangements.

[0194] The storage medium may also include an operating system and a network communication module. The operating system is a program that manages the hardware and software resources of the physical device, supporting the execution of information processing programs and other software and / or programs. The network communication module is used to enable communication between components within the storage medium, as well as with other hardware and software within the physical information processing device.

[0195] Through the description of the above implementation methods, those skilled in the art can clearly understand that the present disclosure can be implemented by means of software plus a necessary general hardware platform, or by hardware. Compared with the prior art, the technical solution in the present disclosure collects sound data and electrical data of multiple power devices, locates the noise data of the power device from the sound data, and extracts the fault characteristics of the power device according to the time-varying characteristic curve of the sound data located at the noise data; locates the fault time node of the fault characteristic extracted, and matches the electrical data of multiple power devices corresponding to the fault time node to obtain the fault electrical data of each power device; uses the sound separation model to separate the noise data from the sound data to obtain the noise data of each power device, and determines the fault of each power device according to the noise data of each power device, the time-varying characteristic curve of the sound data located at the noise data, and the matching degree of the fault electrical data of each power device; predicts the fault trend of the power device according to the abnormal data trend of the fault type of each power device fault, wherein the abnormal data trend is used to characterize the deviation of key parameters or indicators from the expected trend during the operation of the power device, and the fault trend characterizes the development and change of the fault over time. By applying the solution disclosed herein, the fault time node of the sound data from which fault characteristics are extracted is located, and the electrical data corresponding to the time node for multiple power devices is matched to obtain the fault electrical data for each power device. This precise matching at the time node allows for a clearer analysis of the correlation between the sound and electrical signals during partial discharge (PD) events, eliminating interference signals during non-fault periods and improving the reliability of PD detection. In PD detection scenarios, environmental interference signals often appear as noise in the sound data. Using a sound separation model, the true PD sound characteristics can be separated from the noise, reducing the impact of interference signals on PD sound feature extraction and improving the accuracy of the sound data used for PD detection. The sound signals generated by PD have specific time-varying characteristics, while the time-varying characteristics of interference signals are different. By analyzing the time-varying characteristic curves of the sound data, the PD fault characteristics can be accurately extracted, eliminating the interference of interference signals, and thus more accurately detecting PD faults. For PD detection, comprehensively considering the amount of sound and electrical information allows for a more comprehensive assessment of whether a device has a PD fault and the severity of the fault. Even if there are interference signals, this comprehensive consideration method can more accurately judge partial discharge faults and avoid missed detection or misjudgment due to interference signals.

[0196] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprises" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device that includes a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article or device. In the absence of further limitations, an element defined by the sentence "comprises a ..." does not exclude the presence of other identical elements in the process, method, article or device that includes the element.

[0197] The foregoing are merely specific embodiments of the present disclosure, intended to enable those skilled in the art to understand and implement the present disclosure. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present disclosure. Therefore, the present disclosure is not to be limited to these embodiments, but rather to the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A method for diagnosing and predicting faults in power equipment, characterized in that: The method comprises: collecting sound data and electrical data of a plurality of power devices, locating noise data of the power devices from the sound data, and extracting fault characteristics of the power devices based on a time-varying characteristic curve of the sound data located at the noise data; Locating the fault time node where the fault feature is extracted, and matching the electrical data of multiple power devices corresponding to the fault time node to obtain the fault electrical data of each power device; Separating noise data from the sound data using a sound separation model to obtain noise data for each of the power devices, and determining a fault of each of the power devices based on a degree of matching between the noise data of each of the power devices, a time-varying characteristic curve of the sound data located to the noise data, and fault electrical data of each of the power devices; Based on the abnormal data trend of the fault type of each power equipment fault, the fault trend of the power equipment is predicted, wherein the abnormal data trend is used to characterize the deviation of key parameters or indicators in the operation process of the power equipment from the expected trend, and the fault trend characterizes the development and change of the fault over time.

2. The method according to claim 1, characterized in that Before collecting the sound data and electrical data of the plurality of electrical devices, the method further includes: Constructing a spatial location map including the plurality of power devices, wherein the spatial location map is used to display the distribution of the plurality of power devices in space; Arrange a plurality of information collection points according to the spatial location diagram, wherein the information collection points are used to represent data collection locations of the sound data and electrical data of the plurality of power devices; The collecting of sound data and electrical data of a plurality of electrical equipment includes: Based on the multiple information collection points, the sound data and electrical data of the multiple power devices are synchronously collected.

3. The method according to claim 2, characterized in that The method further comprises: Divide each power device in the spatial position diagram into regions to obtain a sound collection spatial region for each power device, wherein the sound collection spatial region includes a region of each power device divided with the center of the device as a coordinate center and a typical noise source region of each power device; The synchronously collecting the sound data and electrical data of the plurality of power devices based on the plurality of information collection points includes: Based on the multiple information collection points and the sound collection spatial area, the sound data and electrical data of the multiple power devices are synchronously collected.

4. The method according to claim 1, wherein The locating noise data of the electric power equipment from the sound data comprises: Calculating the energy of the sound signal of each channel of the sound data, detecting energy mutation points according to the energy of the sound signal of each channel, and locating noise data in the sound data according to the energy mutation points; or Perform frequency domain analysis on the sound signal of each channel of the sound data to obtain a spectrum distribution of the sound signal of each channel, and identify abnormal frequency components in the spectrum distribution to locate noise data in the sound data based on the abnormal frequency components, where the abnormal frequency components are frequency components greater than or less than a preset frequency component range.

5. The method according to claim 4, characterized in that The extracting the fault feature of the electric power equipment according to the time-varying characteristic curve of the sound data located at the noise data includes: Calculating the short-time energy and short-time zero-crossing rate of the sound data, wherein the short-time energy is used to reflect the energy of the sound signal of each channel of the sound data within a short time window, and the short-time zero-crossing rate is used to reflect the frequency change characteristics of the sound signal of each channel of the sound data within the short time window; determining a time-varying characteristic curve of the sound data according to the short-time energy and the short-time zero-crossing rate; Based on the Mel-frequency cepstral coefficients and in combination with the time-varying characteristic curve of the sound data, the fault characteristics of the corresponding power equipment are extracted.

6. The method according to claim 4, characterized in that Before using the sound separation model to separate the noise data from the sound data to obtain the noise data of each power device, the method further includes: The sound separation model is constructed by combining a deep learning model with a convolutional neural network and a generative adversarial network; The method of separating noise data from the sound data using a sound separation model to obtain noise data of each power device includes: Data features are extracted from the noise data, and the data features are input into the trained sound separation model to obtain the noise data of each electric power device.

7. The method according to claim 1, characterized in that Determining the fault of each power device according to the noise data of each power device, the time-varying characteristic curve of the sound data located to the noise data, and the matching degree of the fault electrical data of each power device includes: Calculating the matching degree between the noise data of each power device and the fault electrical data of each power device and the time-varying characteristic curve by using a correlation coefficient method and a Euclidean distance method; If the matching degree exceeds a preset matching degree threshold, it is determined that a corresponding fault exists in the power equipment.

8. The method according to claim 7, characterized in that Before calculating the degree of matching between the noise data of each power device and the fault electrical data of each power device and the time-varying characteristic curve by using the correlation coefficient method and the Euclidean distance method, the method further includes: Using a heterogeneous data fusion method, the noise data of each power device and the fault electrical data of each power device are processed separately, and the noise data of each power device and the fault electrical data of each power device are fused using a defined association rule or model; The calculating the matching degree between the noise data of each power device and the fault electrical data of each power device by using the correlation coefficient and the Euclidean distance includes: The matching degree between the fused noise data of each electric device and the fault electrical data of each electric device and the time-varying characteristic curve is calculated by using the correlation coefficient and the Euclidean distance.

9. A device for diagnosing and predicting faults in electric power equipment, characterized in that: include: an acquisition module, configured to acquire sound data and electrical data of a plurality of power devices, locate noise data of the power devices from the sound data, and extract fault characteristics of the power devices based on a time-varying characteristic curve of the sound data located at the noise data; a matching module, configured to locate the fault time node from which the fault feature is extracted, and match the electrical data of multiple power devices corresponding to the fault time node to obtain the fault electrical data of each power device; a determination module, configured to separate noise data from the sound data using a sound separation model to obtain noise data of each power device, and determine a fault of each power device based on a degree of matching between the noise data of each power device, a time-varying characteristic curve of the sound data located to the noise data, and fault electrical data of each power device; A prediction module is used to predict the failure trend of the power equipment based on the abnormal data trend of the failure type of each power equipment failure, wherein the abnormal data trend is used to characterize the deviation of key parameters or indicators in the operation process of the power equipment from the expected trend, and the failure trend characterizes the development and change of the failure over time.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 8 is implemented.

Images