Power equipment voiceprint diagnosis method, device and equipment based on edge calculation

Abstract

The invention relates to a power equipment voiceprint diagnosis method, device and equipment based on edge calculation. The method comprises the following steps: preprocessing collected voiceprint data of power equipment to obtain preprocessed voiceprint data; performing voiceprint feature extraction on the preprocessed voiceprint data to obtain voiceprint features of the voiceprint data; inputting voiceprint features of the voiceprint data into a voiceprint diagnosis model to obtain a target fault type and a target severity of the power equipment; and generating a diagnosis result of the power equipment according to the target fault type and the target severity of the power equipment. By adopting the method, the accuracy of the generated diagnosis result of the power equipment can be improved.

Description

Technical Field

[0001] This application relates to the field of equipment diagnostic technology, and in particular to a method, apparatus and equipment for voiceprint diagnostics of power equipment based on edge computing. Background Technology

[0002] Power equipment is the core of a stable power system, and its operating status determines the reliability of power supply. Due to factors such as mechanical wear and electrical aging, power equipment is prone to failure, and these failures can cause changes in acoustic characteristics. Therefore, analyzing the operating status of power equipment through its acoustic characteristics has become a key technology for power equipment operation and maintenance.

[0003] In related technologies, voiceprint data is often collected on-site at power equipment, and the voiceprint data is compared with preset thresholds to diagnose whether the power equipment is faulty, thus obtaining the diagnostic results of the power equipment.

[0004] However, the method of comparing voiceprint data with preset thresholds to diagnose whether power equipment is faulty has the problem of inaccurate diagnostic results. Summary of the Invention

[0005] Based on this, this application provides a method, apparatus, and device for voiceprint diagnosis of power equipment based on edge computing, which can improve the accuracy of the generated diagnostic results of power equipment.

[0006] In a first aspect, this application provides a method for voiceprint diagnosis of power equipment based on edge computing, the method comprising:

[0007] The collected voiceprint data of the power equipment is preprocessed to obtain preprocessed voiceprint data;

[0008] Voiceprint features are extracted from the preprocessed voiceprint data to obtain the voiceprint features of the voiceprint data.

[0009] By inputting the voiceprint features of the voiceprint data into the voiceprint diagnostic model, the target fault type and target severity of the power equipment can be obtained.

[0010] Based on the target fault type and severity of the power equipment, generate diagnostic results for the power equipment.

[0011] In some embodiments, the collected voiceprint data of power equipment is preprocessed to obtain preprocessed voiceprint data, including:

[0012] The voiceprint data of power equipment is denoised to obtain the denoised voiceprint data.

[0013] The denoised speaker data is framed and windowed to obtain the temporal frame signal.

[0014] Perform a Fourier transform on the time-domain frame signal to obtain the frequency-domain features;

[0015] The frequency domain features are standardized to obtain preprocessed speaker data.

[0016] In some embodiments, voiceprint feature extraction is performed on the preprocessed voiceprint data to obtain the voiceprint features of the voiceprint data, including:

[0017] Extract Mel frequency cepstral coefficient features from preprocessed speaker data;

[0018] Extract the spectral centroid features, spectral roll-off point features, and spectral flux features from the voiceprint data;

[0019] The voiceprint features of the voiceprint data are obtained by fusing the cepstral coefficient features of the Mel frequency, the centroid features of the spectrum, the roll-off point features, and the flux features of the spectrum of the voiceprint data.

[0020] In some embodiments, the voiceprint features of the voiceprint data are input into the voiceprint diagnostic model to obtain the target fault type and target severity of the power equipment, including:

[0021] The voiceprint features of the voiceprint data are input into the convolutional neural network model in the voiceprint diagnosis model to obtain the deep features of the voiceprint data.

[0022] The deep features of the voiceprint data are input into the fault type classification model and the severity classification model in the voiceprint diagnosis model to obtain the target fault type and target severity of the power equipment.

[0023] In some embodiments, the method further includes:

[0024] When the diagnostic results of the power equipment indicate that there is a fault in the power equipment, an early warning signal for the power equipment is generated;

[0025] In response to early warning signals from power equipment, the maintenance priority of the power equipment is determined based on the target fault type and the target severity.

[0026] Based on the maintenance priority of power equipment, generate operation and maintenance guidance information for power equipment.

[0027] In some embodiments, determining the maintenance priority of power equipment based on the target fault type and target severity includes:

[0028] Determine the current level of operation and maintenance resources for the power equipment based on the type of fault.

[0029] The maintenance priority of power equipment is determined based on the severity of the power equipment, its criticality level in the power grid system, the rate of failure development, and the current availability of maintenance resources.

[0030] In some embodiments, before preprocessing the collected voiceprint data of the power equipment to obtain the preprocessed voiceprint data, the method further includes:

[0031] Based on the spatial distribution of power equipment, equipment ledgers, and risk assessment levels, determine the acoustic fingerprint collection strategy for power equipment.

[0032] Based on the voiceprint acquisition strategy of power equipment, the corresponding voiceprint acquisition device of the power equipment is scheduled to collect voiceprint data to obtain the voiceprint data of the power equipment.

[0033] In some embodiments, the method further includes:

[0034] Receive diagnostic query requests from the equipment management platform and send diagnostic results of the power equipment to the equipment management platform;

[0035] The receiving device management platform sends model update data on the update channel, and updates the voiceprint diagnostic model according to the model update data. The update channel is established based on dedicated update information, which is obtained by appending and compiling communication verification information with model update data. The communication verification information is obtained by hash mapping the communication code generated by the communication address of the edge computing device.

[0036] Secondly, this application provides a power equipment voiceprint diagnostic device based on edge computing, comprising:

[0037] The preprocessing module is used to preprocess the collected voiceprint data of the power equipment to obtain preprocessed voiceprint data.

[0038] The feature extraction module is used to extract voiceprint features from the preprocessed voiceprint data to obtain the voiceprint features of the voiceprint data.

[0039] The inference module is used to input the voiceprint features of the voiceprint data into the voiceprint diagnosis model to obtain the target fault type and target severity of the power equipment.

[0040] The diagnostic module is used to generate diagnostic results for power equipment based on the target fault type and severity.

[0041] Thirdly, this application provides an edge computing device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the method of any one of the first aspects.

[0042] In the technical solution provided in this application embodiment, by collecting voiceprint data of power equipment through edge computing devices and generating diagnostic results for the power equipment, the traditional approach of transmitting voiceprint data to the device management platform to obtain diagnostic results for the power equipment is avoided. This approach suffers from significant delays in voiceprint data acquisition and processing, as well as data loss during transmission, leading to inaccurate diagnostic results. By preprocessing and extracting features from the voiceprint data before inputting it into the voiceprint diagnostic model, the operating status of the power equipment can be identified through the deep features of the voiceprint data, thereby improving the accuracy of the generated diagnostic results for the power equipment. Attached Figure Description

[0043] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0044] Figure 1 A flowchart illustrating a power equipment voiceprint diagnostic method based on edge computing provided for some embodiments;

[0045] Figure 2 A flowchart illustrating a method for preprocessing voiceprint data collected from power equipment to obtain preprocessed voiceprint data, provided in some embodiments.

[0046] Figure 3 A flowchart illustrating a method for extracting voiceprint features from preprocessed voiceprint data to obtain voiceprint features of the voiceprint data, provided in some embodiments.

[0047] Figure 4 A flowchart illustrating a method for inputting voiceprint features from voiceprint data into a voiceprint diagnostic model to obtain the target fault type and target severity of power equipment, provided in some embodiments;

[0048] Figure 5 A flowchart illustrating a method for generating operation and maintenance guidance information for power equipment, provided in some embodiments;

[0049] Figure 6 A schematic diagram of the structure of a power equipment voiceprint diagnostic device based on edge computing provided in some embodiments;

[0050] Figure 7 A schematic diagram of the structure of an edge computing device provided for some embodiments. Detailed Implementation

[0051] The embodiments of the technical solution of this application will now be described in detail with reference to the accompanying drawings. These embodiments are only used to more clearly illustrate the technical solution of this application and are therefore merely examples, and should not be used to limit the scope of protection of this application.

[0052] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the application; the terms “comprising” and “having”, and any variations thereof, in the specification, claims, and foregoing description of the drawings are intended to cover non-exclusive inclusion.

[0053] In the description of the embodiments of this application, technical terms such as "first" and "second" are used only to distinguish different objects and should not be construed as indicating or implying relative importance or implicitly specifying the number, specific order, or primary and secondary relationship of the indicated technical features. In the description of the embodiments of this application, "multiple" means two or more, "multiple groups" means two or more, and "each" means each of the multiple, unless otherwise explicitly defined.

[0054] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0055] In the description of the embodiments in this application, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.

[0056] Unless otherwise specified, the order of execution steps in the embodiments of this application is not limited. It should also be noted that any step in the embodiments of this application can be executed independently, that is, the execution of any step in the above embodiments does not depend on the execution of other steps.

[0057] To address the issue of inaccurate diagnostic results for power equipment in related technologies, this application provides a power equipment voiceprint diagnostic method based on edge computing, applied to an edge computing device. The method includes: preprocessing the collected voiceprint data of the power equipment to obtain preprocessed voiceprint data; extracting voiceprint features from the preprocessed voiceprint data to obtain voiceprint features; inputting the voiceprint features into a voiceprint diagnostic model to obtain the target fault type and target severity of the power equipment; and generating a diagnostic result for the power equipment based on the target fault type and target severity. This method, by collecting voiceprint data from the power equipment on an edge computing device and generating diagnostic results, avoids the previous approach of transmitting voiceprint data to an equipment management platform for diagnostic results. In this approach, the large latency from data collection to processing and the potential loss of voiceprint data during transmission lead to inaccurate diagnostic results. By preprocessing and extracting features from the voiceprint data before inputting it into the voiceprint diagnostic model, the operating status of the power equipment can be identified through the deep features of the voiceprint data, thereby improving the accuracy of the generated diagnostic results.

[0058] The method in this embodiment is applied to an edge computing device (also known as an edge node or an edge computing-based power equipment voiceprint diagnostic device). The edge computing device may include a server, host computer, industrial control computer, tablet computer, personal computer, mobile phone, or other devices with acoustic sensor access, data acquisition, local computing, and communication capabilities. This embodiment does not specifically limit this.

[0059] The edge computing device in this embodiment is a computing device deployed at or near the power equipment site, and has acoustic sensor access, data acquisition, local computing and communication capabilities.

[0060] All actions involving the acquisition of signals, information, or data in this application embodiment are carried out in compliance with the data protection laws and policies of the country where the location is situated, and with authorization from the owner of the corresponding device.

[0061] Figure 1 A flowchart illustrating a power equipment voiceprint diagnostic method based on edge computing is provided for some embodiments, such as... Figure 1 As shown, the method includes the following steps:

[0062] S101. Preprocess the collected voiceprint data of the power equipment to obtain preprocessed voiceprint data.

[0063] Electrical equipment includes, but is not limited to, transformers, circuit breakers, instrument transformers, or reactors, which generate characteristic acoustic signals during operation.

[0064] For example, one or more voiceprint acquisition devices can be set up near the power equipment. These devices can include acoustic sensors, such as multi-channel acoustic sensors. In this embodiment, an array of voiceprint acquisition devices can be set up near the power equipment, or at least one voiceprint acquisition device can be set up at each key component of the power equipment. For instance, an edge computing device synchronously collects voiceprint data from multiple monitoring points of the power equipment using a multi-channel acoustic sensor array, where each monitoring point corresponds to a power device or at least one key component of a power device.

[0065] Voiceprint data refers to the audio signals collected during device operation using a high-precision microphone or acoustic sensor. For example, the sampling rate of the high-precision microphone or acoustic sensor is no less than 16kHz to ensure that high-frequency vibrations and abnormal sound characteristics during device operation can be captured.

[0066] In some embodiments, the edge computing device can acquire the raw acoustic fingerprint signal of the power equipment, perform automatic gain control (AGC) and analog-to-digital conversion (ADC) on the raw acoustic fingerprint signal, and add timestamps, device identifiers, and sensor location metadata to the acquired data. To ensure data validity, the edge computing device also monitors the signal-to-noise ratio in real time, and automatically triggers re-acquisition or issues a sensor status alarm when the signal-to-noise ratio is lower than a preset threshold (e.g., 15dB).

[0067] In some embodiments, edge computing devices can adaptively adjust the sampling frequency and duration according to the operating conditions of power equipment (such as load changes and start / stop status) to achieve an intelligent sampling strategy with optimal energy efficiency.

[0068] In some embodiments, a voiceprint acquisition strategy can be generated based on the spatial distribution of power equipment and monitoring needs; based on the voiceprint acquisition strategy, a voiceprint acquisition device can be scheduled to acquire voiceprint data at a preset time and frequency.

[0069] Voiceprint acquisition strategy is a data acquisition plan that is comprehensively formulated based on the importance of the equipment, historical fault records, operating conditions and environmental factors. It includes at least one parameter such as acquisition time window, sampling frequency, data duration and triggering conditions.

[0070] For example, firstly, the spatial distribution map, equipment ledger, and risk assessment level of power equipment are obtained. Combined with operation and maintenance experience rules, such as increasing the collection frequency of critical equipment and strengthening monitoring during periods of high failure incidence (e.g., increasing the sampling frequency), at least one of these is used to generate an initial collection strategy. Subsequently, load balancing optimization is performed based on the computing resources, network bandwidth, and energy storage status of edge computing devices to form the final executable voiceprint collection strategy. Collection instructions can be issued to each voiceprint collection device according to a preset timing sequence. The voiceprint collection device starts or puts the sensor module into sleep mode according to the strategy requirements and completes voiceprint data collection according to specified parameters. For example, voiceprint data collection is completed by collecting for 30 seconds every 15 minutes at a sampling rate of 32kHz. In some embodiments, it is also possible to support dynamically triggering temporary collection tasks based on the real-time status of power equipment (such as sudden load increases and / or temperature anomalies).

[0071] In addition, edge computing devices can also have fault tolerance mechanisms such as retrying for failed data acquisition and / or data integrity verification to ensure the comprehensiveness and reliability of voiceprint monitoring.

[0072] Preprocessing refers to a series of signal processing operations performed on the voiceprint data of power equipment to improve signal quality, eliminate interference, and standardize the data format, laying the foundation for subsequent feature extraction and diagnosis. For example, preprocessing operations mainly include at least one step such as noise reduction, framing, windowing, filtering, and normalization to obtain preprocessed voiceprint data.

[0073] S102. Extract voiceprint features from the preprocessed voiceprint data to obtain the voiceprint features of the voiceprint data.

[0074] Voiceprint features refer to the characteristic parameters extracted from preprocessed voiceprint data that can characterize the operating status of power equipment, including at least one of time-domain features, frequency-domain features, and time-frequency-domain features.

[0075] For example, the temporal features may include at least one statistical feature of the preprocessed voiceprint data, such as short-time energy, zero-crossing rate, and amplitude envelope, reflecting the temporal energy distribution and fluctuation characteristics of the voiceprint data.

[0076] For example, frequency domain features may include at least one feature such as Mel-Frequency Cepstral Coefficients (MFCCs), spectral centroid, spectral roll-off point, and spectral flux calculated based on the Fast Fourier Transform (FFT) spectrum. The MFCC feature is obtained through a Mel filter bank and a Discrete Cosine Transform (DCT), effectively characterizing the spectral envelope properties of the voiceprint. Its extraction process includes: pre-emphasis, windowing and framing, FFT transformation, Mel filter bank filtering, logarithmic operation, and DCT transformation.

[0077] For example, time-frequency domain feature extraction obtains the time-frequency distribution characteristics of preprocessed speaker data through wavelet transform or empirical mode decomposition, including at least one of the following: revealing the time-varying characteristics of non-stationary speaker data and advanced acoustic features. Revealing the time-varying characteristics of non-stationary speaker data includes at least one of the following: wavelet energy entropy, intrinsic mode function (IMF) component energy ratio, etc. Advanced acoustic features include at least one of the following: harmonic-to-noise ratio, formant frequency, jitter, flicker, etc., of the preprocessed speaker data. Advanced acoustic features can enhance the ability to identify early, subtle faults in power equipment.

[0078] S103. Input the voiceprint features of the voiceprint data into the voiceprint diagnosis model to obtain the target fault type and target severity of the power equipment.

[0079] For example, the voiceprint diagnostic model can be a classification model built on a deep learning architecture, used to identify the operating status of power equipment based on the input voiceprint feature vector. This operating status can include the confidence scores for each fault type and the confidence scores for each severity level among multiple fault types. For instance, if the confidence scores for all fault types are less than or equal to a preset confidence threshold, the operating status of the power equipment is considered normal. Conversely, if the confidence score for at least one fault type is greater than or equal to a preset confidence threshold, the operating status of the power equipment is considered faulty, and this at least one fault type is identified as the target fault type. In the case of a faulty operating status, the severity level corresponding to the highest confidence score among the severity levels can be identified as the target severity level.

[0080] In some embodiments, the voiceprint diagnostic model can employ a hybrid architecture combining deep convolutional neural networks and attention mechanisms. The model input consists of extracted voiceprint features, which undergo multi-scale feature extraction through three convolutional layers. A SE attention module enhances the response to key features, and finally, a fully connected layer and a Softmax output layer generate the diagnostic results. During training, a historical dataset containing hundreds of thousands of labeled voiceprint samples is used, employing a cross-entropy loss function and the Adam optimizer. Five-fold cross-validation ensures the model's generalization ability. The voiceprint features are then input into the voiceprint diagnostic model to obtain the diagnostic results for power equipment.

[0081] S104. Generate diagnostic results for the power equipment based on the target fault type and severity of the power equipment.

[0082] In some embodiments, the target fault type and target severity of the power equipment can be determined as the diagnostic results of the power equipment.

[0083] In other embodiments, a diagnostic report for the power equipment is generated based on the target fault type and target severity, and the diagnostic report is determined as the diagnostic result of the power equipment.

[0084] In the technical solution provided in this application embodiment, by collecting voiceprint data of power equipment through edge computing devices and generating diagnostic results for the power equipment, the traditional approach of transmitting voiceprint data to the device management platform to obtain diagnostic results for the power equipment is avoided. This approach suffers from significant delays in voiceprint data acquisition and processing, as well as data loss during transmission, leading to inaccurate diagnostic results. By preprocessing and extracting features from the voiceprint data before inputting it into the voiceprint diagnostic model, the operating status of the power equipment can be identified through the deep features of the voiceprint data, thereby improving the accuracy of the generated diagnostic results for the power equipment.

[0085] Figure 2 This is a flowchart illustrating a method for preprocessing voiceprint data collected from power equipment to obtain preprocessed voiceprint data, provided in some embodiments. This method, which explains step S101, includes the following steps:

[0086] S201. Noise reduction processing is performed on the voiceprint data of the power equipment to obtain the noise-reduced voiceprint data.

[0087] Noise reduction processing refers to the use of digital signal processing technology to suppress or eliminate non-device sound source components in voiceprint data, mainly including environmental background noise and electromagnetic interference signals.

[0088] For example, noise reduction processing can employ a deep learning-based denoising autoencoder (DAE). This model is trained using a paired dataset consisting of typical substation environmental noise and clean equipment acoustic prints, which can effectively separate equipment acoustic prints from environmental noise, improving the signal-to-noise ratio by more than 15dB.

[0089] For example, based on the characteristic frequency range of the acoustic signature of power equipment (typically 80Hz-12kHz), a finite impulse response (FIR) band-pass filter can be designed to retain effective frequency band components and suppress power frequency interference and high-frequency noise.

[0090] S202. The denoised voiceprint data is framed and windowed to obtain the time-domain frame signal.

[0091] Framing processing divides continuous audioprint data into short-segment analysis units. Windowing processing applies a window function to each frame of signal to reduce spectral leakage.

[0092] For example, S202 may include performing frame-segmentation processing on the denoised voiceprint data to obtain frame-segmented data; and performing windowing processing on the frame-segmented data to obtain a time-domain frame signal.

[0093] For example, frame segmentation can be performed using a preset frame length and a preset frame shift to achieve quasi-stationary processing of non-stationary voiceprint data. For instance, the preset frame length can be between 10ms and 40ms. The preset frame shift can be less than or equal to the preset frame length. For example, the preset frame shift can be 10ms. For example, the first frame contains voiceprint data from 1 to 20ms, the second frame contains voiceprint data from 11 to 30ms, and so on.

[0094] The formula used for windowing can be: ;in, Indicates the first Window function values ​​for each sampling point Let be the window length. In this formula, the signal is weighted and truncated using a cosine function, meaning the weights of the sampling points at both ends gradually decrease (smooth transition), while the weights of the sampling points in the middle are close to 1, thus avoiding spectral leakage caused by sudden signal truncation.

[0095] S203. Perform Fourier transform on the time-domain frame signal to obtain the frequency-domain features.

[0096] The Fourier transform converts a time-domain signal into a frequency-domain representation.

[0097] For example, the Fourier transform uses the Fast Fourier Transform (FFT) algorithm to convert each frame of time-domain signal into a 512-point frequency-domain spectrum, achieving a frequency resolution of 31.25Hz (at a 16kHz sampling rate), thus fully preserving the spectral details of the device's audio signature.

[0098] S204. Standardize the frequency domain features to obtain preprocessed voiceprint data.

[0099] Standardization is the process of normalizing the amplitude of spectral features.

[0100] For example, the standardization process employs a mean-variance normalization method for each frequency point. spectral amplitude Standardize: ;in, Represents the standardized first The frequency point spectral amplitude is the normalized result used for subsequent processing. Indicates the original number The spectral amplitude of a frequency point (such as the amplitude of a certain frequency after Fourier transform). Indicates: the The mean amplitude of all samples at a frequency point (i.e., the average of multiple samples at that frequency point). Indicates the first The standard deviation of amplitude for all samples at a frequency point (reflecting the degree of dispersion of amplitude at that frequency point). and Used to eliminate the influence of amplitude variations under different acquisition conditions.

[0101] In the technical solution provided in this application embodiment, the voiceprint data of power equipment is processed in a pipeline manner by performing noise reduction, framing, windowing, Fourier transform, and standardization, thereby improving the data quality of the voiceprint data and improving the accuracy of the generated diagnostic results.

[0102] Figure 3 A flowchart illustrating a method for extracting voiceprint features from preprocessed voiceprint data in some embodiments is provided, such as... Figure 3 As shown, this method is an explanation of S102, and the method includes:

[0103] S301. Extract Mel frequency cepstral coefficient features from the preprocessed voiceprint data.

[0104] Mel frequency cepstral coefficients (MFCC) are acoustic features extracted by simulating the characteristics of human hearing. They contain 13-39 dimensions of cepstral coefficients and can effectively characterize the short-time spectral envelope of voiceprint data.

[0105] For example, Mel frequency cepstral coefficient feature extraction employs N-dimensional (e.g., 13-dimensional) MFCC coefficients, including M (e.g., 12) cepstral coefficients and NM (e.g., 1) logarithmic energy values.

[0106] The process of extracting Mel frequency cepstral coefficient features includes: pre-emphasizing the pre-processed voiceprint data, then performing frame-by-frame windowing to obtain frame-by-frame windowed data; performing Fast Fourier Transform (FFT) on the frame-by-frame windowed data, and then processing it through K (e.g., 40) Mel-scale triangular filter banks to obtain filtered data; finally, taking the logarithmic energy of the filtered data and performing Discrete Cosine Transform (DCT) to obtain Mel frequency cepstral coefficient features.

[0107] S302. Extract the spectral centroid features, spectral roll-off point features, and spectral flux features from the voiceprint data.

[0108] The centroid of the spectrum is the location of the center of gravity of the spectral energy distribution, reflecting the brightness of the sound. The roll-off point is the frequency point at which the accumulated spectral energy reaches a preset proportion (e.g., 85%), characterizing the shape of the spectrum. Spectral flux is the amount of spectral change between adjacent frames, used to detect the transient characteristics of the audioprint.

[0109] For example, the spectral centroid Represented as: ;in, Indicates the frequency index, from 1 to... , indicating the first One frequency point; This represents the total number of frequency points, i.e., the total number of frequency points involved in the spectrum analysis; This represents the frequency value of the k-th frequency point, usually in Hz; represents the spectral amplitude at the k-th frequency point, and represents the energy intensity at that frequency point.

[0110] For example, the spectral roll-off point can be represented as: ; This indicates the spectral roll-off point, which is the frequency point at which the accumulated energy reaches 85% of the total energy, reflecting the concentrated range of spectral energy. 0.85 represents the spectral amplitude at the k-th frequency point; 0.85 represents the energy percentage threshold, indicating that the cumulative energy must reach 85% of the total energy.

[0111] For example, spectral flux It can be represented as ; It represents the spectral amplitude of the k-th frequency point in the t-th frame, and represents the energy intensity of that frequency point in the current frame; The spectral amplitude of the k-th frequency point in the (t-1)-th frame represents the energy intensity of that frequency point in the previous frame; t represents the frame index, indicating consecutive frames in time (e.g., the t-th frame and the (t-1)-th frame are the spectra of two adjacent audio signals).

[0112] S303. The characteristics of the Mel frequency cepstral coefficients, the spectral centroid, the spectral roll-off point, and the spectral flux of the voiceprint data are fused to obtain the voiceprint characteristics of the voiceprint data.

[0113] In some embodiments, S303 may include: combining MFCC features (e.g., 13-dimensional) with spectral features (e.g., 3-dimensional, namely spectral centroid features, spectral roll-off point features, and spectral flux features) into a voiceprint vector (e.g., 16-dimensional) using a concatenation method; performing z-score normalization on the concatenated voiceprint vector to eliminate the influence of dimensions and obtain normalized features; and fusing the normalized features to obtain the voiceprint features of the voiceprint data. In some embodiments, fusing the normalized features may include: a weighted fusion strategy based on feature importance, evaluating the contribution of each feature component to device status recognition using a random forest algorithm, assigning different fusion weights, and using the fusion weights to perform weighted processing on the normalized features to obtain the voiceprint features of the voiceprint data (16-dimensional).

[0114] In the technical solution provided in this application embodiment, the characteristics of Mel frequency cepstral coefficients, the spectral centroid characteristics, the spectral roll-off point characteristics, and the spectral flux characteristics of voiceprint data are fused to obtain the voiceprint characteristics of the voiceprint data, thereby enabling the acquisition of deep information of the voiceprint data, which is beneficial to improving the accuracy of the generated diagnostic results.

[0115] Figure 4 A flowchart illustrating a method for inputting voiceprint features from voiceprint data into a voiceprint diagnostic model to obtain the target fault type and severity of power equipment, as provided in some embodiments, is shown below. Figure 4 As shown, the method includes the following steps:

[0116] S401. Input the voiceprint features of the voiceprint data into the convolutional neural network model in the voiceprint diagnosis model to obtain the deep features of the voiceprint data.

[0117] A convolutional neural network (CNN) model is a deep learning model that includes convolutional layers, pooling layers, and activation functions. It is used to automatically learn more discriminative deep feature representations from voiceprint features. Deep features refer to feature vectors with stronger semantic information obtained after undergoing multiple layers of nonlinear transformations.

[0118] S402. Input the deep features of the voiceprint data into the fault type classification model and severity classification model in the voiceprint diagnosis model to obtain the target fault type and target severity of the power equipment.

[0119] The classification model is a multi-classification model built on fully connected layers and the Softmax function, used to identify specific fault types and assess their severity based on deep features.

[0120] For example, the voiceprint diagnostic model adopts a convolutional neural network (CNN) architecture consisting of three convolutional blocks. Each convolutional block contains a convolutional layer, a batch normalization layer, a ReLU activation function, and a max pooling layer. By stacking convolutional layers, the local patterns and global contextual information of the voiceprint features are extracted step by step, and the input voiceprint features (e.g., 16-dimensional) are converted into a deep feature vector of a preset dimension (128-dimensional). Then, the deep feature is input into a classifier consisting of two fully connected layers. The first fully connected layer outputs the probability distribution of six common fault types (including mechanical loosening, partial discharge, insulation degradation, winding deformation, core abnormality, and cooling system fault) through the sigmoid function. The second fully connected layer outputs the evaluation results of four severity levels (normal, slight, moderate, and severe) through the Softmax function. Finally, the system generates a structured diagnostic result containing specific fault types, severity levels, confidence levels, and maintenance suggestions based on the fault type probability distribution and severity evaluation, combined with a preset diagnostic rule base. For example, when a fault of moderate or greater severity is detected, a real-time alarm is automatically triggered and pushed to the maintenance personnel's terminal. At the same time, all diagnostic results are recorded in the equipment health record for trend analysis and predictive maintenance.

[0121] In some embodiments, after S104, the method may include: determining the current state of the power equipment based on the diagnostic results, and generating corresponding response information based on the current state.

[0122] Current status refers to the classification of the operating condition of power equipment based on the voiceprint diagnostic results, including normal status, early warning status, and alarm status. Response refers to the corresponding processing measures triggered according to the equipment status, including maintenance actions such as status recording, early warning push, alarm generation, and maintenance work order creation.

[0123] Response information refers to guiding information that includes specific maintenance recommendations, processing timelines, and resource allocation plans.

[0124] For example, the diagnostic results can be parsed by a status decision engine, and the equipment status can be mapped to three levels: normal, warning (minor fault), and alarm (moderate and above fault) according to the fault type and its severity. Based on the preset response strategy library, the corresponding processing flow is automatically triggered: For the normal status, the system only records the diagnostic results to the equipment health record and updates the equipment operation statistics; for the warning status, the system generates an equipment status warning notification and pushes it to the mobile terminal of the maintenance personnel, and marks the equipment as a status requiring attention in the equipment management platform and suggests that it be checked in the next inspection; for the alarm status, the system immediately generates an alarm event containing equipment information, fault details and handling suggestions, automatically creates an emergency maintenance work order and assigns it to the corresponding maintenance team, and provides on-site warning through audible and visual alarm devices, and links the relevant control system to perform protective operations (such as starting standby equipment, adjusting operating parameters, etc.) according to the fault type. The execution status of all response actions is monitored and recorded in real time, forming a closed-loop management from status diagnosis to response execution.

[0125] In the technical solution provided in this application embodiment, the deep features of the voiceprint data are respectively input into the fault type classification model and the severity classification model in the voiceprint diagnosis model to obtain the target fault type and target severity of the power equipment, thereby enabling the determination of the accuracy of the target fault type and target severity of the power equipment.

[0126] Figure 5 A flowchart illustrating a method for generating operation and maintenance guidance information for power equipment, as provided in some embodiments, is shown below. Figure 5 As shown, this method can be executed after S104; the method can include:

[0127] S501. When the diagnostic results of the power equipment indicate that there is a fault in the power equipment, generate a warning signal for the power equipment.

[0128] Early warning signals refer to standardized alarm data packets generated when a diagnostic model identifies a fault in a device.

[0129] S502. In response to the warning signal of the power equipment, determine the maintenance priority of the power equipment according to the target fault type and target severity of the power equipment.

[0130] In some embodiments, determining the maintenance priority of power equipment based on the target fault type and target severity includes: determining the current level of maintenance resources for the power equipment based on the fault type; and determining the maintenance priority of the power equipment based on the severity of the power equipment, the criticality level of the power equipment in the power grid system, the fault development speed of the power equipment, and the current level of maintenance resources for the power equipment.

[0131] S503. Generate operation and maintenance guidance information for power equipment based on the maintenance priority of the power equipment.

[0132] In this embodiment, by analyzing diagnostic results in real time, a device warning signal containing device code, fault type, timestamp, and confidence level is generated when fault characteristics are identified. This signal is then processed according to a preset rule base, comprehensively considering the severity of the fault (graded from 1 to 4), the device's criticality level in the power grid (critical / important / general), the fault's development rate (through historical data trend analysis), and the current operational resource status. A weighted scoring model is used to calculate high, medium, and low maintenance priorities. In some embodiments, corresponding operational guidance information can be automatically generated based on the maintenance priority. For example, for high priority, an emergency response work order is immediately generated and the emergency response team is notified to arrive within 2 hours; for medium priority, a planned maintenance work order is generated and scheduled for processing within 24 hours; and for low priority, an observational maintenance suggestion is generated and included in the next scheduled inspection plan. All response information is distributed and tracked through the operational management platform.

[0133] In the technical solution provided in this application embodiment, the maintenance priority of the power equipment is determined according to the target fault type and target severity of the power equipment, thereby improving the accuracy of the determined maintenance priority.

[0134] In some embodiments, the device management platform may perform the following steps: determine the communication address of the edge computing device and generate a communication code based on the communication address; generate a diagnostic query signal based on the communication code and diagnostic query instructions; send the diagnostic query signal to the edge computing device and receive real-time diagnostic data fed back by the edge computing device based on the diagnostic query signal; and project the real-time diagnostic data to the operation and maintenance monitoring screen.

[0135] It should be noted that the communication address refers to the unique identifier of the edge computing device in the network. For example, it can be an Internet Protocol Address (IP) address, a Media Access Control Address (MAC) address, or a device serial number.

[0136] The communication code is a temporary session credential generated based on the communication address using an encryption algorithm. Diagnostic query signals are standardized data packets containing query instructions and authentication information; real-time diagnostic data includes real-time monitoring results such as the device's current status, fault information, characteristic parameters, and diagnostic confidence levels. The maintenance monitoring dashboard refers to a digital visualization display system deployed in the monitoring center.

[0137] For example, the physical and logical addresses of the edge computing devices are obtained through the device management platform. A 16-bit communication code with timeliness is generated using the SHA-256 hash algorithm combined with a timestamp. This communication code and diagnostic query instructions are encapsulated into a data packet conforming to the Message Queuing Telemetry Transport (MQTT) protocol and sent to the edge computing device through an encrypted channel of Transport Layer Security (TLS). After receiving the data, the edge computing device verifies the validity and permissions of the communication code and then packages the latest device diagnostic data (including device operating status, fault type, severity level, characteristic parameter trend graph, and diagnostic confidence) into JSON format and returns it to the query end. After receiving the real-time diagnostic data, the system pushes it to the operation and maintenance monitoring dashboard via the WebSocket protocol. The dashboard displays the device status panel, fault alarm list, characteristic parameter curve, and multi-device status distribution map in a dynamically updated manner. It also supports intuitive display of device health status through color coding (green for normal, yellow for warning, and red for alarm) and provides data drill-down and trend analysis functions, enabling operation and maintenance personnel to grasp the operating status of all devices in the network in real time.

[0138] In some embodiments, before preprocessing the collected voiceprint data of the power equipment to obtain the preprocessed voiceprint data, the method further includes: determining the voiceprint acquisition strategy of the power equipment based on the spatial distribution of the power equipment, the equipment ledger, and the risk assessment level; and scheduling the voiceprint acquisition device corresponding to the power equipment to collect voiceprint data according to the voiceprint acquisition strategy of the power equipment to obtain the voiceprint data of the power equipment.

[0139] In some embodiments, the method further includes: receiving a diagnostic query request sent by the device management platform, and sending the diagnostic results of the power equipment to the device management platform.

[0140] In some embodiments, the method further includes: receiving model update data sent by a device management platform on an update channel, and updating the voiceprint diagnostic model according to the model update data; wherein the update channel is established based on dedicated update information, the dedicated update information is obtained by appending and compiling communication verification information with model update data, and the communication verification information is obtained by hash mapping based on the communication code generated by the communication address of the edge computing device.

[0141] In some embodiments, the device management platform may perform the following steps: determine the communication address of the edge computing device and generate a communication code based on the communication address; send a model update signal to the edge computing device based on the communication code and receive a response message from the edge computing device after successful verification of the model update signal; after receiving the response message, establish an update channel based on the communication code; and transmit data with the edge computing device based on the update channel to update the voiceprint diagnostic model online.

[0142] Model update signals can include control signals that contain model update instructions and verification information. The update channel is a secure communication link dedicated to model data transmission.

[0143] For example, the MAC address and IP address of the edge computing device are obtained through the device management platform. An RSA asymmetric encryption algorithm combined with a timestamp is used to generate a time-sensitive communication code. This communication code, along with the model update instruction, is encapsulated into a digitally signed data packet and sent to the edge computing device. Upon receiving the packet, the edge computing device verifies the validity of the digital signature and the freshness of the timestamp using a pre-set public key. After successful verification, it returns a response containing a session key. Upon receiving the response, the device management platform establishes a dedicated update channel using AES-256 encryption based on the communication code and session key. Through this channel, the new version of the voiceprint diagnostic model parameter file, optimized by quantization and pruning, is transmitted in chunks to the edge computing device. During transmission, Cyclic Redundancy Check (CRC) verification and a breakpoint resumption mechanism are used to ensure data integrity. After receiving the complete model file, the edge computing device first verifies the model's compatibility and performance indicators in a sandbox environment. Upon successful verification, it hot-swaps to the new model while retaining the old model as a rollback backup. Simultaneously, it sends an update completion confirmation to the system. The entire update process does not interrupt the device monitoring function and supports model updates for a single edge computing device within 5 minutes.

[0144] In some embodiments, the step of establishing an update channel based on a communication code after receiving a response message includes: performing a hash mapping on the communication code after receiving the response message to obtain communication verification information; sending the communication verification information to an edge computing device so that the edge computing device can listen to the communication verification information; appending and compiling the communication verification information with model update data to obtain dedicated update information; and establishing an update channel based on the dedicated update information.

[0145] For example, hash mapping converts a communication code into a fixed-length digest value using a hash function (such as SHA-256), serving as communication verification information. This verification information is a credential used for authentication and data integrity verification. Monitoring refers to edge computing devices continuously monitoring and filtering data packets containing verification information on specific network ports. Attachment compilation combines the communication verification information with model update data using data encapsulation and encryption algorithms to create proprietary update information. This proprietary update information is a complete data unit containing a verification header, encrypted payload, and checksum. The update channel is an end-to-end secure transmission link established based on the verification result.

[0146] For example, after receiving the response information from the edge computing device, the device management platform can perform a hash mapping on the communication code using the SHA-256 algorithm to generate 32 bytes of communication verification information. This verification information is then encapsulated into a User Datagram Protocol (UDP) broadcast packet and sent to the designated listening port of the edge computing device. The edge computing device continuously monitors network traffic using deep packet inspection technology. When it identifies matching verification information, it triggers a successful authentication status. Subsequently, the system serializes and concatenates the communication verification information as a data packet header with the model update data (such as quantized neural network parameters). The system then uses the Advanced Encryption Standard - Galois / Counter Mode (AES-GCM) algorithm to encrypt and protect the integrity of the overall data, generating dedicated update information. Finally, based on the session parameters in this information, a dedicated update channel based on the Datagram Transport Layer Security (DTLS) protocol is established between the two parties. This channel has forward security and anti-replay attack capabilities, supports chunked transmission and dynamic bandwidth adaptation, and ensures the secure, reliable, and efficient execution of the model update process.

[0147] Based on the same inventive concept, this application also provides an edge computing-based power equipment voiceprint diagnosis device for implementing the aforementioned edge computing-based power equipment voiceprint diagnosis method. The solution provided by this device is similar to the implementation described in the above method. Therefore, the specific limitations of one or more edge computing-based power equipment voiceprint diagnosis device embodiments provided below can be found in the limitations of the edge computing-based power equipment voiceprint diagnosis method described above, and will not be repeated here.

[0148] In one exemplary embodiment, Figure 6 A schematic diagram of the structure of a power equipment voiceprint diagnostic device based on edge computing provided in some embodiments, such as Figure 6As shown, the edge computing-based power equipment voiceprint diagnostic device 600 includes:

[0149] The preprocessing module 601 is used to preprocess the collected voiceprint data of the power equipment to obtain preprocessed voiceprint data.

[0150] Feature extraction module 602 is used to extract voiceprint features from preprocessed voiceprint data to obtain voiceprint features of the voiceprint data;

[0151] The inference module 603 is used to input the voiceprint features of the voiceprint data into the voiceprint diagnosis model to obtain the target fault type and target severity of the power equipment.

[0152] The diagnostic module 604 is used to generate diagnostic results for the power equipment based on the target fault type and target severity of the power equipment.

[0153] In some embodiments, the preprocessing module 601 includes a noise reduction unit, a time-domain frame determination unit, a Fourier transform unit, and a standardization unit. The noise reduction unit is used to perform noise reduction processing on the voiceprint data of the power equipment to obtain noise-reduced voiceprint data. The time-domain frame determination unit is used to perform frame segmentation and windowing processing on the noise-reduced voiceprint data to obtain a time-domain frame signal. The Fourier transform unit is used to perform Fourier transform on the time-domain frame signal to obtain frequency domain features. The standardization unit is used to perform standardization processing on the frequency domain features to obtain preprocessed voiceprint data.

[0154] In some embodiments, the feature extraction module 602 includes a first extraction unit, a second extraction unit, and a fusion unit; the first extraction unit is used to extract Mel frequency cepstral coefficient features from the preprocessed voiceprint data; the second extraction unit is used to extract spectral centroid features, spectral roll-off point features, and spectral flux features from the voiceprint data; the fusion unit is used to fuse the Mel frequency cepstral coefficient features, spectral centroid features, spectral roll-off point features, and spectral flux features of the voiceprint data to obtain the voiceprint features of the voiceprint data.

[0155] In some embodiments, the inference module 603 is further configured to input the voiceprint features of the voiceprint data into the convolutional neural network model in the voiceprint diagnosis model to obtain the deep features of the voiceprint data; and input the deep features of the voiceprint data into the fault type classification model and the severity classification model in the voiceprint diagnosis model to obtain the target fault type and target severity of the power equipment.

[0156] In some embodiments, the edge computing-based power equipment voiceprint diagnostic device 600 further includes a warning signal generation module for generating a warning signal for the power equipment when the diagnostic result of the power equipment indicates that the power equipment has a fault; a maintenance priority determination module for determining the maintenance priority of the power equipment according to the target fault type and target severity in response to the warning signal of the power equipment; and an operation and maintenance guidance information generation module for generating operation and maintenance guidance information for the power equipment according to the maintenance priority of the power equipment.

[0157] In some embodiments, the maintenance priority determination module is further configured to determine the current level of maintenance resources for the power equipment based on the fault type of the power equipment; and to determine the maintenance priority of the power equipment based on the severity of the power equipment, the criticality level of the power equipment in the power grid system, the fault development speed of the power equipment, and the current level of maintenance resources for the power equipment.

[0158] In some embodiments, the edge computing-based power equipment voiceprint diagnostic device 600 further includes a voiceprint data acquisition module, which is used to determine the voiceprint acquisition strategy of the power equipment based on the acquired spatial distribution of the power equipment, equipment ledger and risk assessment level; and to schedule the voiceprint acquisition device corresponding to the power equipment to acquire voiceprint data according to the voiceprint acquisition strategy of the power equipment, so as to obtain the voiceprint data of the power equipment.

[0159] In some embodiments, the edge computing-based power equipment voiceprint diagnostic device 600 further includes a communication module for receiving diagnostic query requests sent by the equipment management platform and sending diagnostic results of the power equipment to the equipment management platform.

[0160] In some embodiments, the edge computing-based power equipment voiceprint diagnostic device 600 further includes a communication module and an update module; the communication module is used to receive model update data sent by the device management platform on the update channel; the update module is used to update the voiceprint diagnostic model according to the model update data; wherein, the update channel is established based on dedicated update information, the dedicated update information is obtained by appending and compiling communication verification information with model update data, and the communication verification information is obtained by hash mapping based on the communication code generated by the communication address of the edge computing device.

[0161] The descriptions of the above device embodiments are similar to those of the above method embodiments, and have similar beneficial effects. For technical details not disclosed in the device embodiments of this application, please refer to the descriptions of the method embodiments of this application for understanding.

[0162] Each module in the aforementioned edge computing-based power equipment voiceprint diagnostic device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of the edge computing device in hardware form or independent of it, or stored in the memory of the edge computing device in software form, so that the processor can call and execute the corresponding operations of each module.

[0163] In one exemplary embodiment, Figure 7 This is a schematic diagram of the structure of an edge computing device provided in some embodiments. The edge computing device includes a processor, memory, input / output interface, communication interface, display unit, and input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interface. The processor of the edge computing device provides computing and control capabilities. The memory of the edge computing device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The input / output interface of the edge computing device is used for exchanging information between the processor and external devices. The communication interface of the edge computing device is used for wired or wireless communication with external terminals. Wireless communication can be implemented through Wireless Fidelity (WIFI), mobile cellular networks, Near Field Communication (NFC), or other technologies. When the computer program is executed by the processor, it implements a power equipment voiceprint diagnosis method based on edge computing. The display unit of this edge computing device is used to form a visually visible image and can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of this edge computing device can be a touch layer covering the display screen, or buttons, a trackball, or a touchpad set on the outer casing of the edge computing device, or an external keyboard, touchpad, or mouse, etc.

[0164] Those skilled in the art will understand that Figure 7 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 edge computing device to which the present application is applied. A specific edge computing device may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0165] For example, an edge computing device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of the method of any of the above embodiments.

[0166] In one embodiment, a computer-readable storage medium is provided, wherein a computer program, when executed by a processor, implements the steps of the method provided in any of the above embodiments.

[0167] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps of the method provided in any of the above embodiments.

[0168] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the methods described above.

[0169] The processor, functional modules, or functional units in any embodiment of this application may include an integration of one or more of the following: a general-purpose processor, an application-specific integrated circuit (ASIC), a digital signal processor (DSP), a digital signal processing device (DSPD), a programmable logic device (PLD), a field-programmable gate array (FPGA), a central processing unit (CPU), a graphics processing unit (GPU), an embedded neural network processing unit (NPU), a controller, a microcontroller, a microprocessor, a programmable logic device, a discrete gate or transistor logic device, a discrete hardware component, a quantum computing-based data processing logic unit, an artificial intelligence (AI) processor, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0170] The memory or computer-readable storage medium in any embodiment of this application may include at least one of non-volatile memory and volatile memory. Non-volatile memory includes integration of one or more of the following: Read Only Memory (ROM), Programmable Read-Only Memory (PROM), Erasable Programmable Read-Only Memory (EPROM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Ferromagnetic Random Access Memory (FRAM), Flash Memory, Magnetic Surface Memory, Optical Disc, Compact Disc Read-Only Memory (CD-ROM), Magnetic Tape, Floppy Disk, Flash Memory, Optical Memory, High-Density Embedded Non-Volatile Memory, Resistive Random Access Memory (ReRAM), Magnetoresistive Random Access Memory (MRAM), Ferroelectric Random Access Memory (FRAM), Phase Change Memory (PCM), Graphene Memory, Volatile Memory, etc. Volatile memory includes one or more of the following: Random Access Memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0171] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0172] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for voiceprint diagnosis of power equipment based on edge computing, characterized in that, Applications in edge computing devices, including: The collected voiceprint data of the power equipment is preprocessed to obtain preprocessed voiceprint data; The preprocessed voiceprint data is subjected to voiceprint feature extraction to obtain the voiceprint features of the voiceprint data; The voiceprint features of the voiceprint data are input into the voiceprint diagnostic model to obtain the target fault type and target severity of the power equipment. Based on the target fault type and target severity of the power equipment, a diagnostic result for the power equipment is generated.

2. The method according to claim 1, characterized in that, The preprocessing of the collected voiceprint data from the power equipment to obtain preprocessed voiceprint data includes: The voiceprint data of the power equipment is subjected to noise reduction processing to obtain noise-reduced voiceprint data; The denoised speaker data is subjected to frame segmentation and windowing to obtain a time-domain frame signal; Perform a Fourier transform on the time-domain frame signal to obtain the frequency-domain features; The frequency domain features are standardized to obtain the preprocessed voiceprint data.

3. The method according to claim 1 or 2, characterized in that, The step of extracting voiceprint features from the preprocessed voiceprint data to obtain the voiceprint features of the voiceprint data includes: Extract Mel frequency cepstral coefficient features from the preprocessed voiceprint data; Extract the spectral centroid features, spectral roll-off point features, and spectral flux features from the voiceprint data; The voiceprint features of the voiceprint data are obtained by fusing the Mel frequency cepstral coefficient features, the spectral centroid features, the spectral roll-off point features, and the spectral flux features of the voiceprint data.

4. The method according to claim 1 or 2, characterized in that, The step of inputting the voiceprint features of the voiceprint data into the voiceprint diagnostic model to obtain the target fault type and target severity of the power equipment includes: The voiceprint features of the voiceprint data are input into the convolutional neural network model in the voiceprint diagnosis model to obtain the deep features of the voiceprint data. The deep features of the voiceprint data are input into the fault type classification model and the severity classification model in the voiceprint diagnosis model to obtain the target fault type and target severity of the power equipment.

5. The method according to claim 1 or 2, characterized in that, The method further includes: If the diagnostic results of the power equipment indicate that the power equipment is faulty, a warning signal for the power equipment is generated. In response to the warning signal of the power equipment, the maintenance priority of the power equipment is determined according to the target fault type and target severity of the power equipment; Based on the maintenance priority of the power equipment, operation and maintenance guidance information for the power equipment is generated.

6. The method according to claim 5, characterized in that, The step of determining the maintenance priority of the power equipment based on the target fault type and target severity includes: Based on the fault type of the power equipment, determine the current level of operation and maintenance resources for the power equipment; The maintenance priority of the power equipment is determined based on the severity of the power equipment, the criticality level of the power equipment in the power grid system, the failure development speed of the power equipment, and the current level of operation and maintenance resources for the power equipment.

7. The method according to claim 1 or 2, characterized in that, Before preprocessing the collected voiceprint data from the power equipment to obtain the preprocessed voiceprint data, the method further includes: Based on the spatial distribution, equipment ledger, and risk assessment level of the power equipment, a voiceprint acquisition strategy for the power equipment is determined. According to the voiceprint acquisition strategy of the power equipment, the voiceprint acquisition device corresponding to the power equipment is scheduled to collect voiceprint data to obtain the voiceprint data of the power equipment.

8. The method according to claim 1 or 2, characterized in that, The method further includes: Receive diagnostic query requests sent by the equipment management platform, and send the diagnostic results of the power equipment to the equipment management platform; The device receives model update data sent by the device management platform on the update channel, and updates the voiceprint diagnostic model according to the model update data; wherein, the update channel is established based on dedicated update information, the dedicated update information is obtained by appending and compiling communication verification information with the model update data, and the communication verification information is obtained by hash mapping based on the communication code generated by the communication address of the edge computing device.

9. A power equipment voiceprint diagnostic device based on edge computing, characterized in that, include: The preprocessing module is used to preprocess the collected voiceprint data of the power equipment to obtain preprocessed voiceprint data. The feature extraction module is used to extract voiceprint features from the preprocessed voiceprint data to obtain the voiceprint features of the voiceprint data. The inference module is used to input the voiceprint features of the voiceprint data into the voiceprint diagnosis model to obtain the target fault type and target severity of the power equipment. The diagnostic module is used to generate diagnostic results for the power equipment based on the target fault type and target severity.

10. An edge computing device, characterized in that, The method includes a memory and a processor, the memory storing a computer program, characterized in that the processor executes the computer program to implement the steps of the method according to any one of claims 1 to 8.

Images