An AI-based power equipment voiceprint monitoring system and method

Abstract

The application provides an AI-based power equipment voiceprint monitoring system and method, wherein the system comprises a voiceprint collection module for collecting voiceprint signals; a data processing module for obtaining and preprocessing the voiceprint signals; a voiceprint recognition module for recognizing target power equipment and voiceprint signal data corresponding to the target power equipment; a fault recognition module for judging whether the target power equipment has faults and recognizing fault types; and a fault early warning module for issuing an alarm to a manager. The application identifies various fault types and main fault types of the target power equipment through wavelet transform, controls the moving direction and collection frequency of the sound sensor through the voiceprint collection module, reduces the influence of interference signals, and improves the accuracy of recognizing fault types.

Description

Technical Field

[0001] This invention relates to the field of computer data processing technology, and in particular to an AI-based power equipment voiceprint monitoring system and method. Background Technology

[0002] Currently, voiceprints are sound wave signals with specific information content collected by electroacoustic instruments. The voiceprints of electrical equipment exhibit a certain periodicity and stability, containing unique information about the equipment. For example, transformers and generators emit different sounds, which can be identified by collecting their voiceprints. The sounds emitted by a normally operating electrical device and those that have malfunctioned are also different. Therefore, collecting the voiceprint signals of electrical equipment can determine whether it has malfunctioned. Different types of malfunctions result in different voiceprint signals. For instance, a transformer emitting a 50Hz "humming" voiceprint signal indicates normal operation; a "crackling" sound indicates an internal short circuit.

[0003] Existing technology CN114371353A discloses a method and system for monitoring power equipment anomalies based on voiceprint recognition. It collects a first sound signal indicating normal operation of the power equipment, converts it into first voiceprint data, performs audio spectrum analysis on the first voiceprint data, extracts first audio features, and stores them in a first audio database. Then, it collects a second sound signal indicating the power equipment is operating, converts it into second voiceprint data, performs audio spectrum analysis on the second voiceprint data, and extracts second audio features. A neural network is used to determine whether the second audio features exist in the first audio database; if not, an anomaly alarm signal is issued. Existing technology CN114371353A determines whether a fault exists in known equipment and matches it against known faults stored in the database to determine the fault type. However, faulty power equipment often contains multiple faults, requiring further identification of multiple fault types and the main fault type. Furthermore, the existing technology does not specify the movement and frequency of the voiceprint acquisition.

[0004] Therefore, further analysis of the faults in the power equipment is needed, and a priority order for future maintenance of the power equipment should be established. Summary of the Invention

[0005] One of the objectives of this invention is to provide an AI-based power equipment voiceprint monitoring system that acquires voiceprint signals, processes them to obtain voiceprint signal data, identifies power equipment based on the voiceprint signal data, determines whether a fault exists based on the voiceprint signal data, and finally identifies the fault type and main fault based on the corresponding voiceprint signal data. At the same time, it determines the movement route and acquisition frequency of the voiceprint acquisition module, reduces the influence of interference signals, and improves the accuracy of fault type identification.

[0006] This invention provides an AI-based power equipment voiceprint monitoring system, comprising:

[0007] The voiceprint acquisition module is used to acquire voiceprint signals from target power equipment.

[0008] The data processing module is used to acquire the voiceprint signal and perform preprocessing to obtain voiceprint signal data;

[0009] The voiceprint recognition module is used to identify target power equipment based on voiceprint signal data;

[0010] The fault identification module is used to acquire the acoustic signature signal data corresponding to the target power equipment and perform wavelet transform, and identify the fault type and main fault of the target power equipment based on the wavelet transform.

[0011] The fault warning module is used to send alarms to the administrator based on the judgment results.

[0012] Preferably, the fault identification module includes:

[0013] The fault diagnosis submodule is used to calculate the overlap ratio based on the acoustic fingerprint signal data corresponding to the target power equipment and the predicted acoustic fingerprint signal data of the target power equipment during normal operation, and to determine whether the target power equipment has a fault based on the overlap ratio.

[0014] The type identification submodule is used to acquire the voiceprint signal data corresponding to the target power equipment and perform wavelet transform if the target power equipment malfunctions, and identify the fault type of the target power equipment based on the wavelet transform.

[0015] Preferably, the fault diagnosis submodule performs the following operations:

[0016] Obtain the attribute data of the target power equipment at the current moment. The attribute data includes equipment type, number of days the equipment has been running continuously, historical maintenance records, and future maintenance records.

[0017] Predict the attribute data of the target power equipment at a future preset time point based on the current attribute data of the target power equipment;

[0018] Based on the attribute data of the target power equipment at a time point after a preset future time, retrieve the historical normal operation voiceprint data and the historical abnormal voiceprint data corresponding to the attribute data of the target power equipment at a time point after a preset future time.

[0019] When the target power equipment reaches a time point after a preset future time, acquire the corresponding voiceprint signal data of the target power equipment, calculate the overlap ratio between the voiceprint signal data of the target power equipment and the historical normal operation voiceprint data corresponding to the attribute data of the target power equipment at the time point after the preset future time, and calculate the overlap ratio between the voiceprint signal data of the target power equipment and the historical abnormal voiceprint data corresponding to the attribute data of the target power equipment at the time point after the preset future time, and determine whether the target power equipment has a fault at the time point after the preset future time.

[0020] Preferably, the type recognition submodule performs the following operations:

[0021] A wavelet transform function is constructed, and wavelet decomposition is performed on the acoustic signature signal data corresponding to the target power equipment to obtain the acoustic signature harmonics of the target power equipment; among which,

[0022] The wavelet basis function g(t) is expressed as follows:

[0023]

[0024] In the formula, g(t) is the wavelet basis function, ω0 is the center frequency, a is a constant, t is the time variable, and i is the imaginary number. 2 =-1;

[0025] Where f(t) is the acoustic signature signal data, and [-t0, t0] is the range of values ​​for the acoustic signature signal data f(t). The acoustic signature harmonic Ψf(b, τ) of the target power equipment is obtained by wavelet transform through the wavelet basis function g(t). The specific expression is:

[0026]

[0027] In the formula, Ψf(b,τ) is the acoustic harmonic of the target power equipment, b is the scale factor, and τ is the time displacement.

[0028] Acquire acoustic signature data of the target power equipment during normal operation and perform wavelet transform to obtain the equipment's normal harmonic information; among which,

[0029] Equipment normal harmonic information The specific expression is:

[0030]

[0031] In the formula, This is normal harmonic information from the equipment. It is the voiceprint data of the target power equipment during normal operation.

[0032] By filtering out normal harmonic information from the acoustic waveform of the target power equipment, abnormal harmonic information is obtained; among which...

[0033] The specific expression for the abnormal harmonic information Ψy(b,τ) is:

[0034]

[0035] In the formula, Ψy(b,τ) represents the abnormal harmonic information;

[0036] The abnormal harmonic information is reconstructed using wavelet transform to obtain the fault time-domain signal; the expression for the fault time-domain signal is:

[0037]

[0038] In the formula, y(t) is the fault time-domain signal, and C g These are the wavelet transform convergence coefficients;

[0039] Calculate the similarity between each fault time-domain signal and a known fault signal, and sort them according to the similarity; the specific expression for the similarity is:

[0040]

[0041] In the formula, η is the similarity, and h(t) is a known fault signal of a certain type;

[0042] A similarity threshold is set, and the fault type of the target power equipment is obtained based on the known fault signals corresponding to the similarity exceeding the similarity threshold.

[0043] Preferably, the voiceprint recognition module includes:

[0044] The feature extraction unit is used to construct a feature extraction network model and obtain historical voiceprint signals of power equipment to train the feature extraction network model. After training, the feature extraction network model is used to extract features of the target power equipment to obtain the feature data of the target power equipment.

[0045] The feature recognition unit is used to acquire historical feature data of power equipment. After training, the power equipment feature recognition model is used to identify the target power equipment from the feature data.

[0046] The equipment signal unit is used to acquire the corresponding voiceprint signal data based on the identified target power equipment.

[0047] Preferably, the voiceprint recognition module performs the following operations:

[0048] Construct a feature extraction network model;

[0049] Historical voiceprint signals of power equipment are acquired to train a feature extraction network model. After training, the feature extraction network model is used to extract features from the voiceprint signal data to obtain feature data.

[0050] Construct a feature recognition model for power equipment;

[0051] Historical feature data of power equipment is obtained to train the power equipment feature recognition model. After training, the power equipment feature recognition model is used to identify the target power equipment from the feature data.

[0052] Obtain the corresponding voiceprint signal data based on the identified target power equipment;

[0053] Historical voiceprint signals of power equipment are acquired to train a power equipment feature recognition model, wherein the power equipment feature recognition model is a convolutional neural network;

[0054] After training, the power equipment feature recognition model is used to perform feature recognition on the voiceprint signal data to obtain the target power equipment.

[0055] Obtain the corresponding voiceprint signal data based on the identified target power equipment.

[0056] Preferably, the voiceprint acquisition module includes:

[0057] Several sound sensors are used to collect the acoustic signature signals of the target power equipment;

[0058] Cameras are used for real-time video monitoring of target electrical equipment;

[0059] A lifting unit is used to enable the sound sensor 11 to move up and down;

[0060] A horizontal movement unit, mounted on the lifting unit, is used for the horizontal movement of the sound sensor 11;

[0061] The data acquisition and control unit is connected to the sound sensor, camera, horizontal movement unit, and lifting unit, respectively. It is used to plan the shortest monitoring path according to the location of each power device and to control the acquisition frequency of the sound sensor according to the type and failure rate of the power device.

[0062] Preferably, the data processing module includes:

[0063] The noise reduction unit is used to reduce noise in the voiceprint signal;

[0064] The segmentation unit is used to divide the voiceprint signal into several data interval segments according to the periodicity of the voiceprint signal.

[0065] The fitting unit is used to fit several data interval segments after noise reduction according to the minimum mean square error to obtain the voiceprint signal data.

[0066] The storage unit is used to store the voiceprint signal data.

[0067] Preferably, the AI-based power equipment voiceprint monitoring system further includes:

[0068] The real-time monitoring module is used to perform real-time video monitoring of the target power equipment and process the data in a timely manner.

[0069] The real-time monitoring module includes:

[0070] Temperature sensor, used to measure the temperature of target electrical equipment;

[0071] An ammeter is used to measure the current of a target electrical device.

[0072] The power cut-off unit is used to determine whether to cut off the power supply to the target electrical equipment based on the overlap ratio and the current of the target electrical equipment.

[0073] The equipment cooling unit is used to determine whether to cool the target power equipment based on the overlap ratio and the temperature measured on the target power equipment.

[0074] The fire suppression unit is used to determine whether to extinguish the fire on the target electrical equipment by collecting real-time video footage.

[0075] This invention also provides an AI-based method for monitoring the acoustic signature of power equipment, comprising:

[0076] Collect acoustic signature signals from the target power equipment;

[0077] The voiceprint signal is preprocessed to obtain voiceprint signal data;

[0078] The target power equipment and its corresponding voiceprint signal data are identified based on the voiceprint signal data.

[0079] Acquire the acoustic signature signal data corresponding to the target power equipment and perform wavelet transform, and identify the fault type and main fault of the target power equipment based on the wavelet transform;

[0080] An alert is issued to the manager based on the assessment results.

[0081] The beneficial effects of this invention are as follows:

[0082] This invention identifies target power equipment by performing wavelet transform on the acquired voiceprint signal, determines whether a fault has occurred by calculating the overlap ratio, and issues an alarm to the administrator based on the judgment result. This invention eliminates interference signals through wavelet transform, improving the accuracy of fault type identification. Other features and advantages of this invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of this invention can be realized and obtained by means of the structures particularly pointed out in the written description, claims, and drawings.

[0083] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0084] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0085] Figure 1 This is a schematic diagram of an AI-based power equipment voiceprint monitoring system according to an embodiment of the present invention;

[0086] Figure 2 This is a schematic diagram of the fault identification module described in an embodiment of the present invention;

[0087] Figure 3 This is a schematic diagram of the voiceprint recognition module described in an embodiment of the present invention;

[0088] Figure 4 This is a schematic diagram of the voiceprint acquisition module described in an embodiment of the present invention;

[0089] Figure 5 This is a schematic diagram of the data processing module described in an embodiment of the present invention;

[0090] Figure 6 This is a schematic diagram of the real-time monitoring module described in an embodiment of the present invention;

[0091] Figure 7 This is a flowchart of an AI-based power equipment voiceprint monitoring method according to an embodiment of the present invention. Detailed Implementation

[0092] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0093] This invention provides an AI-based power equipment voiceprint monitoring system, such as... Figure 1 As shown, it includes a voiceprint acquisition module 1, a data processing module 2, a voiceprint recognition module 3, a fault recognition module 4, and a fault early warning module 5.

[0094] Voiceprint acquisition module 1 is used to acquire voiceprint signals of the target power equipment.

[0095] Data processing module 2 is used to acquire voiceprint signals and perform preprocessing to obtain voiceprint signal data.

[0096] Voiceprint recognition module 3 is used to identify target power equipment based on voiceprint signal data.

[0097] The fault identification module 4 is used to acquire the acoustic signature signal data corresponding to the target power equipment and perform wavelet transform, and identify the fault type and main fault of the target power equipment based on the wavelet transform.

[0098] Fault warning module 5 is used to send alarms to the manager based on the judgment results.

[0099] The working principle and beneficial effects of the above technical solution are as follows:

[0100] Voiceprint acquisition module 1 acquires the voiceprint signal of the target power equipment. Data processing module 2 acquires the voiceprint signal and preprocesses it to obtain voiceprint signal data. Voiceprint recognition module 3 identifies the target power equipment based on the voiceprint signal data and obtains the corresponding voiceprint signal data. Fault identification module 4 calculates the overlap ratio between the voiceprint signal data corresponding to the target power equipment and the historical data during normal operation of the target power equipment to determine whether the target power equipment has a fault and the type of fault. Fault early warning module 5 is used to issue an alarm to the manager based on the judgment result.

[0101] In this embodiment of the invention, the voiceprint signal collected by the voiceprint acquisition module 1 is preprocessed by the data processing module 2. The voiceprint recognition module 3 identifies the target power equipment, and the fault recognition module 4 calculates the overlap ratio to determine whether the target power equipment has a fault. The fault early warning module 5 issues an alarm to the manager based on the judgment result.

[0102] In one embodiment, such as Figure 2 As shown, the fault identification module 4 includes a fault judgment submodule 41 and a type identification submodule 42.

[0103] The fault judgment submodule 41 is used to calculate the overlap ratio between the voiceprint signal data corresponding to the target power equipment and the attribute data of the target power equipment at a time point after a future preset time, as well as the overlap ratio between the voiceprint signal data corresponding to the target power equipment and the attribute data of the target power equipment at a time point after a future preset time, to determine whether the target power equipment has a fault at a time point after a future preset time.

[0104] The type identification submodule 42 is used to acquire the voiceprint signal data corresponding to the target power equipment and perform wavelet transform if the target power equipment fails, and identify the fault type and main fault of the target power equipment based on the wavelet transform.

[0105] The working principle and beneficial effects of the above technical solution are as follows:

[0106] The system predicts the attribute data of the target power equipment at a predetermined time point in the future. It then retrieves historical normal operation acoustic fingerprint data and historical abnormal acoustic fingerprint data corresponding to the attribute data at that predetermined time point. The system calculates the overlap ratio between the acoustic fingerprint signal data of the target power equipment and the historical normal operation acoustic fingerprint data, as well as the overlap ratio between the acoustic fingerprint signal data of the target power equipment and the historical abnormal acoustic fingerprint data, to determine whether a fault has occurred in the target power equipment at the predetermined time point. If a fault has occurred, the system acquires the corresponding acoustic fingerprint signal data, performs wavelet transform on it, and identifies the fault type and main fault based on the wavelet transform.

[0107] First, fault diagnosis is performed, then the fault type is identified. While monitoring the faults of the target power equipment, the type of fault is further analyzed.

[0108] In one embodiment, the fault diagnosis submodule 41 performs the following operations:

[0109] Obtain the attribute data of the target power equipment at the current moment. The attribute data includes equipment type, number of days the equipment has been running continuously, historical maintenance records, and future maintenance records.

[0110] Predict the attribute data of the target power equipment at a future preset time point based on the current attribute data of the target power equipment.

[0111] Based on the attribute data of the target power equipment at a time point after a preset future time, retrieve the historical normal operation voiceprint data and the historical abnormal voiceprint data corresponding to the attribute data of the target power equipment at a time point after a preset future time.

[0112] When the target power equipment reaches a time point after a preset future time, acquire the corresponding voiceprint signal data of the target power equipment, calculate the overlap ratio between the voiceprint signal data of the target power equipment and the historical normal operation voiceprint data corresponding to the attribute data of the target power equipment at the time point after the preset future time, and calculate the overlap ratio between the voiceprint signal data of the target power equipment and the historical abnormal voiceprint data corresponding to the attribute data of the target power equipment at the time point after the preset future time, and determine whether the target power equipment has a fault at the time point after the preset future time.

[0113] The working principle and beneficial effects of the above technical solution are as follows:

[0114] To obtain the voiceprint signal data corresponding to the target power equipment, since the voiceprint signal data is periodic, in order to reduce the amount of calculation, one period of voiceprint signal data f(t) (-t0≤t≤t0) can be selected, where [-t0,t0] is the time interval of one period of voiceprint signal data f(t), and t is the time variable.

[0115] Based on the attribute data of the target power equipment at a time point after a preset future time, retrieve the historical normal operation voiceprint data corresponding to the attribute data of the target power equipment at that time point after a preset future time. Historical abnormal voiceprint data corresponding to the attribute data of the target power equipment at a future preset time point.

[0116] The overlap ratio between the voiceprint signal data corresponding to the target power equipment and the historical normal operation voiceprint data corresponding to the attribute data of the target power equipment at a time point after a preset future time is calculated. The specific expression is as follows:

[0117]

[0118] In the formula, P is the overlap ratio between the voiceprint signal data corresponding to the target power equipment and the historical normal operation voiceprint data corresponding to the attribute data of the target power equipment at a time point after a preset time in the future.

[0119] The overlap ratio between the voiceprint signal data corresponding to the target power equipment and the historical abnormal voiceprint data corresponding to the attribute data of the target power equipment at a time point after a preset future time is calculated. The specific expression is as follows:

[0120]

[0121] In the formula, It is the overlap ratio between the voiceprint signal data corresponding to the target power equipment and the historical abnormal voiceprint data corresponding to the attribute data of the target power equipment at a time point after a preset time in the future.

[0122] According to P and This process determines whether a target power device will malfunction at a predetermined time. For example, if the current time is 13:00, the target power device is a transformer that has been running continuously for 100 days, has 10 historical maintenance records and 20 future maintenance records, and the predetermined time is 1 day, the process predicts the target power device's attribute data one day from now based on the current attribute data. Then, it retrieves the corresponding historical normal operation voiceprint data from the database based on the target power device's attribute data one day from now. and historical abnormal voiceprint data P was calculated to be 0.2, while It can be determined that the target power equipment will malfunction one day later. This is achieved by calculating the overlap ratio P and... It improves the preparedness for fault diagnosis and makes necessary preparations for faults.

[0123] In this embodiment, the type identification submodule 42 performs the following operations:

[0124] A wavelet transform function is constructed and wavelet decomposition is performed on the acoustic signature signal corresponding to the target power equipment to obtain the acoustic signature harmonics of the target power equipment.

[0125] The wavelet basis function g(t) is expressed as follows:

[0126]

[0127] In the formula, g(t) is the wavelet basis function, ω0 is the center frequency, a is a constant, t is the time variable, and i is the imaginary number. 2 =-1.

[0128] Where f(t) is the acoustic signature signal data, and [-t0, t0] is the range of values ​​for the acoustic signature signal data f(t). The acoustic signature harmonic Ψf(b, τ) of the target power equipment is obtained by wavelet transform using the wavelet basis function g(t). The specific expression is:

[0129]

[0130] In the formula, Ψf(b,τ) is the acoustic harmonic of the target power equipment, b is the scale factor, and τ is the time displacement.

[0131] Acquire acoustic signature data of the target power equipment during normal operation and perform wavelet transform to obtain the equipment's normal harmonic information; among which,

[0132] Equipment normal harmonic information The specific expression is:

[0133]

[0134] In the formula, This is normal harmonic information from the equipment. It is the voiceprint data of the target power equipment during normal operation.

[0135] The acoustic harmonic filtering of the target power equipment removes the normal harmonic information of the equipment and obtains the abnormal harmonic information.

[0136] The specific expression for the abnormal harmonic information Ψy(b,τ) is:

[0137]

[0138] In the formula, Ψy(b,τ) represents the abnormal harmonic information.

[0139] The abnormal harmonic information is reconstructed using wavelet transform to obtain the fault time-domain signal; the expression for the fault time-domain signal is:

[0140]

[0141] In the formula, y(t) is the fault time-domain signal, and C g These are the wavelet transform convergence coefficients.

[0142] Calculate the similarity between each fault time-domain signal and a known fault signal, and sort them according to the similarity; the specific expression for the similarity is:

[0143]

[0144] In the formula, η is the similarity, and h(t) is a known fault signal of a certain type;

[0145] A similarity threshold is set, and the fault type of the target power equipment is obtained based on the known fault signals corresponding to the similarity exceeding the similarity threshold.

[0146] The working principle and beneficial effects of the above technical solution are as follows:

[0147] In this embodiment of the invention, the type identification submodule 42 obtains the acoustic harmonics of the target power equipment through wavelet transform, filters out the normal harmonics of the target power equipment, obtains abnormal harmonic information, reconstructs the fault time domain signal according to the wavelet transform, and calculates the similarity with the known fault type signals one by one to determine the included fault type and the main fault type, eliminates interference signals, and improves the accuracy of fault type identification.

[0148] In this embodiment, a similarity threshold of 0.5 is set. Fault types with a similarity exceeding 0.5 are those present in the target power equipment. The fault type with the highest similarity is the primary fault type.

[0149] For example, the similarity scores of the target power equipment transformer with known fault signals of insulation aging, winding short circuit, and inrush current are calculated to be 0.6, 0.8, and 0.1, respectively. This indicates that the fault types of the target power equipment are winding short circuit and insulation aging, and that winding short circuit is the main fault. Therefore, in future maintenance, winding short circuit repair should be prioritized.

[0150] This invention obtains the acoustic harmonics of the target power equipment through wavelet transform, filters out the normal harmonics of the target power equipment, obtains abnormal harmonic information, reconstructs the fault time-domain signal according to wavelet transform, and sorts it according to the calculated similarity to determine that the target power equipment contains all fault types and the main fault types, eliminates interference signals, improves the accuracy of fault type identification, and provides a priority reference for subsequent maintenance.

[0151] In one embodiment, such as Figure 3 As shown, the voiceprint recognition module 3 includes:

[0152] The feature extraction unit 31 is used to construct a feature extraction network model and obtain historical voiceprint signals of power equipment to train the feature extraction network model. After training, the feature extraction network model is used to extract features from the voiceprint signal data of the target power equipment to obtain the feature data of the target power equipment.

[0153] The feature recognition unit 32 is used to collect historical feature data of power equipment to train the power equipment feature recognition model, and to use the trained power equipment feature recognition model to identify the target power equipment from the feature data.

[0154] The device signal unit 33 is used to output corresponding voiceprint signal data according to the identified target power equipment.

[0155] The working principle and beneficial effects of the above technical solution are as follows:

[0156] Feature extraction unit 31 constructs a feature extraction network model and acquires historical voiceprint signals from power equipment to train the feature extraction network model. After training, the feature extraction network model is used to extract features from the target power equipment, obtaining the feature data of the target power equipment. Feature recognition unit 32 trains a power equipment feature recognition model based on the collected historical feature data of the power equipment, and uses the trained power equipment feature recognition model to identify the target power equipment from the feature data. Equipment signal unit 33 acquires the corresponding voiceprint signal data based on the identified target power equipment.

[0157] In one embodiment, the voiceprint recognition module 3 performs the following operations:

[0158] Construct a feature extraction network model.

[0159] Historical voiceprint signals from power equipment are used to train a feature extraction network model. After training, the feature extraction network model is used to extract features from the voiceprint signal data to obtain feature data.

[0160] Construct a feature recognition model for power equipment.

[0161] Historical feature data of power equipment is obtained to train the power equipment feature recognition model. After training, the power equipment feature recognition model is used to identify the target power equipment from the feature data.

[0162] Obtain the corresponding voiceprint signal data based on the identified target power equipment.

[0163] Historical voiceprint signals of power equipment are acquired to train a power equipment feature recognition model, wherein the power equipment feature recognition model is a convolutional neural network.

[0164] After training, the power equipment feature recognition model is used to perform feature recognition on the voiceprint signal data to obtain the target power equipment.

[0165] Obtain the corresponding voiceprint signal data based on the identified target power equipment.

[0166] The working principle and beneficial effects of the above technical solution are as follows:

[0167] A feature extraction network model is constructed, which includes windowing and framing, FFT units, pass-through filters, logarithmic energy units, and cosine transform units. In this embodiment, Fast Fourier Transform (FFT) is used to extract the characteristic parameters of the sound, calculating spectral parameters, amplitude, power, zero-crossing rate, and Mel-frequency cepstral coefficients (MFCCs) using FFT. The feature extraction network model is first trained using historical acoustic signature signals from power equipment. After training, the feature extraction network model is used to extract features from the acoustic signature signal data to obtain feature data. Then, a power equipment feature recognition model is constructed. Historical feature data of power equipment is obtained and used to train the power equipment feature recognition model. After training, the power equipment feature recognition model is used to identify the target power equipment using the feature data. The equipment signal unit 33 outputs the acoustic signature signal data corresponding to the identified target power equipment.

[0168] This embodiment identifies the target power equipment by performing feature extraction and recognition on the voiceprint signal data of the power equipment. Then, it outputs the corresponding voiceprint signal data based on the identified target power equipment.

[0169] In one embodiment, such as Figure 4 As shown, the voiceprint acquisition module 1 includes:

[0170] Voiceprint acquisition module 1 includes:

[0171] Several sound sensors 11 are used to collect the acoustic signature signals of the target power equipment.

[0172] Camera 12 is used for real-time video monitoring of the target power equipment.

[0173] The lifting unit 13 is used to move the sound sensor 11 up and down.

[0174] The horizontal movement unit 14 is mounted on the lifting unit and is used for the horizontal movement of the sound sensor 11.

[0175] The acquisition control unit 15 is connected to the sound sensor 11, the camera 12, the lifting unit 13 and the horizontal movement unit 14 respectively, and is used to plan the movement route and acquisition frequency of the sound sensor 11 according to the type and outline of the target power equipment and the distance between the target power equipment and the sound sensor.

[0176] The working principle and beneficial effects of the above technical solution are as follows:

[0177] First, determine the starting point O coordinate and the ending point E coordinate of the voiceprint acquisition module 1.

[0178] Obtain the location coordinates of each power device and use them as path points, and form a set U from the location coordinates of each power device.

[0179] Define the set of known shortest path nodes V and the set of known shortest path nodes W.

[0180] Calculate the distance from the starting point to the location coordinates of each power device.

[0181] First, set V contains only the starting point O, while set W contains U and E. The distances between the starting point O and all points in set W are calculated, where... It represents the distance between the starting point O and the i-th point in the set W.

[0182] from Select the point S1 with the shortest distance, and move point S1 from set W to set V. Then calculate the distance from point S1 to the remaining points in set W. in It represents the distance between the starting point O and the i-th point in the set W.

[0183] if Compare If the distance is small, update the distance to that point; otherwise, leave it unchanged.

[0184] After the update is complete, select the point S2 with the shortest distance, move S2 from set W to set V, and then calculate the distance from S2 to the remaining points in set W again, and so on, until set V contains set U.

[0185] By connecting the distances calculated and stored from the starting point O and all points within set V, a shortest monitoring path is obtained, starting from the starting point O, passing through S1, S2, ..., and finally reaching the endpoint E. The data acquisition and control unit 15 controls the horizontal movement unit 14 to move, and sequentially collects acoustic signature signals from each power device using the sound sensor 11 starting from the starting point O. Collecting acoustic signature signals from numerous devices saves time by minimizing the shortest monitoring path for the horizontal movement unit 14.

[0186] Simultaneously, based on the type of target power equipment and the frequency of its failure, the sampling frequency F of the sound sensor 11 is determined, specifically expressed as:

[0187]

[0188] In the formula, f d is the operating frequency of the target power equipment, μ is a constant related to the type of the target power equipment, σ is the equipment failure rate of the target power equipment, and 2t0 is the period length of the acoustic signature signal of the target power equipment.

[0189] The acquisition control unit 15 controls the movement of the sound sensor 11 through the lifting unit 13 and the horizontal moving unit 14, and at the same time, the acquisition control unit 15 controls the sound sensor 11 to acquire the frequency.

[0190] In this embodiment of the invention, the voiceprint acquisition module 1 calculates a shortest monitoring path based on the location of each target power device and the location of the starting and ending points, saving monitoring time. At the same time, based on the type of target power device and the frequency of its failure, the sound sensor 11 determines the acquisition frequency F, and strengthens the monitoring frequency for devices with high failure rates, ensuring the accuracy of the monitoring data and improving the monitoring effectiveness of the power devices.

[0191] In one embodiment, such as Figure 5 As shown, the data processing module 2 includes a segmentation unit 21, a noise reduction unit 22, a fitting unit 23, and a storage unit 24.

[0192] The noise reduction unit 21 is used to reduce noise in the voiceprint signal.

[0193] Segmentation unit 22 is used to divide the voiceprint signal into several data interval segments according to the periodicity of the voiceprint signal.

[0194] Fitting unit 23 is used to fit several data interval segments after noise reduction according to the minimum mean square error to obtain voiceprint signal data.

[0195] Storage unit 24 is used to store voiceprint signal data.

[0196] The working principle and beneficial effects of the above technical solution are as follows:

[0197] The noise reduction unit 21 reduces the noise of the voiceprint signal z(t) to remove background noise, such as the chirping of insects, the conversations of nearby employees, etc. Since the voiceprint signal z(t) of the target power equipment is periodic, the segmentation unit 22 divides the voiceprint signal z(t) into n periodic segments according to the periodicity, obtaining The fitting unit 23 fits a number of periodic segments after noise reduction according to the minimum mean square error, obtaining the voiceprint signal data f(t) (-t0 < t < t0), and the storage unit 24 stores the voiceprint signal data f(t) (-t0 < t < t0).

[0198] In the embodiment of the present invention, by dividing the voiceprint signal into several data interval segments according to its periodicity, and at the same time performing noise reduction and curve fitting on the data interval segments, the extracted voiceprint signal data ensures the authenticity and representativeness of the signal, reduces the amount of data operation, and improves the speed of data operation.

[0199] In one embodiment, it further includes a real-time monitoring module 6 connected to the fault identification module 4 for performing real-time camera monitoring on the target power equipment and processing it in a timely manner. As Figure 6 shown, the real-time monitoring module 6 includes:

[0200] The temperature sensor 61 measures the temperature of the target power equipment.

[0201] The ammeter 62 measures the current of the target power equipment.

[0202] The power cut-off unit 63 decides whether to cut off the power supply to the target power equipment according to the reclosing ratio and the current of the target power equipment.

[0203] The equipment cooling unit 64 is used to decide whether to cool down the target power equipment according to the reclosing ratio and the temperature measured by the target power equipment.

[0204] The fire extinguishing unit 65 is used to judge whether to extinguish the fire on the target power equipment by collecting real-time camera images.

[0205] The working principle and beneficial effects of the above technical solutions are as follows:

[0206] The power cut-off unit 63 determines whether to cut off power supply based on the overlap ratio and the current measured by the ammeter 62 on the target electrical equipment. If the overlap ratio is small and the current is large, the power cut-off unit 63 immediately cuts off the power supply. The equipment cooling unit 64 determines whether to cool down the target electrical equipment based on the overlap ratio and the temperature measured by the temperature sensor 61. If the overlap ratio is small and the temperature is too high, the equipment cooling unit 64 immediately cools down the target electrical equipment. The fire extinguishing unit 65 determines whether there is a fire on the target electrical equipment through the camera 12. If a fire occurs, the fire extinguishing unit 65 extinguishes the fire.

[0207] This embodiment of the invention ensures the safety of the target power equipment and eliminates fire hazards by using the real-time monitoring module 6 to monitor temperature and current in real time.

[0208] This invention also provides an AI-based method for monitoring the voiceprint of power equipment, such as... Figure 7 As shown, it includes the following steps:

[0209] Step 1: Collect the acoustic signature signal of the target power equipment.

[0210] Step 2: Preprocess the voiceprint signal to obtain voiceprint signal data.

[0211] Step 3: Identify the target power equipment and the corresponding voiceprint signal data based on the voiceprint signal data.

[0212] Step 4: Obtain the acoustic signature signal data corresponding to the target power equipment and perform wavelet transform, and identify the fault type and main fault of the target power equipment based on the wavelet transform.

[0213] Step 5: Issue an alert to the manager based on the judgment result.

[0214] The working principle and beneficial effects of the above technical solution are as follows:

[0215] Voiceprint acquisition module 1 acquires voiceprint signals. Data processing module 2 performs noise reduction and fitting on the voiceprint signals to obtain voiceprint signal data. Voiceprint recognition module 3 identifies the target power equipment and its corresponding voiceprint signal data based on the voiceprint signal data. Fault identification module 4 acquires the voiceprint signal data corresponding to the target power equipment, performs wavelet transform, and identifies the fault type and main fault of the target power equipment based on the wavelet transform. Fault early warning module 5 receives warning signals and sends alarms to the administrator.

[0216] In this embodiment of the invention, the acoustic fingerprint data of power equipment is collected and identified, and a fault is detected in the target power equipment, triggering a warning signal. This improves fault prediction accuracy.

[0217] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. An AI-based voiceprint monitoring system for power equipment, characterized in that, include: The voiceprint acquisition module is used to acquire voiceprint signals from target power equipment. The data processing module is used to acquire the voiceprint signal and perform preprocessing to obtain voiceprint signal data; The voiceprint recognition module is used to identify target power equipment based on voiceprint signal data; The fault identification module is used to acquire the acoustic signature signal data corresponding to the target power equipment and perform wavelet transform, and identify the fault type and main fault of the target power equipment based on the wavelet transform. The fault early warning module is used to send alarms to the administrator based on the judgment results; The fault identification module includes: The fault diagnosis submodule is used to calculate the overlap ratio based on the acoustic fingerprint signal data corresponding to the target power equipment and the predicted acoustic fingerprint signal data of the target power equipment during normal operation, and to determine whether the target power equipment has a fault based on the overlap ratio. The type identification submodule is used to acquire the voiceprint signal data corresponding to the target power equipment and perform wavelet transform if the target power equipment malfunctions, and identify the fault type of the target power equipment based on the wavelet transform. The type identification submodule performs the following operations: A wavelet transform function is constructed and wavelet decomposition is performed on the acoustic signature signal data corresponding to the target power equipment to obtain the acoustic signature harmonics of the target power equipment. Acquire acoustic signature data of the target power equipment during normal operation and perform wavelet transform to obtain the normal harmonic information of the equipment; The acoustic harmonic filtering of the target power equipment removes the normal harmonic information of the equipment and obtains the abnormal harmonic information. The abnormal harmonic information is reconstructed by wavelet transform to obtain the fault time-domain signal; Calculate the similarity between each fault time-domain signal and a known fault signal, and sort them according to the similarity. A similarity threshold is set, and the fault type of the target power equipment is obtained based on the known fault signals corresponding to the similarity exceeding the similarity threshold.

2. The AI-based power equipment voiceprint monitoring system as described in claim 1, characterized in that, The fault diagnosis submodule performs the following operations: Obtain the attribute data of the target power equipment at the current moment. The attribute data includes equipment type, number of days the equipment has been running continuously, historical maintenance records, and future maintenance records. Predict the attribute data of the target power equipment at a future preset time point based on the current attribute data of the target power equipment; Based on the attribute data of the target power equipment at a time point after a preset future time, retrieve the historical normal operation voiceprint data and the historical abnormal voiceprint data corresponding to the attribute data of the target power equipment at a time point after a preset future time. When the target power equipment reaches a time point after a preset future time, acquire the corresponding voiceprint signal data of the target power equipment, calculate the overlap ratio between the voiceprint signal data of the target power equipment and the historical normal operation voiceprint data corresponding to the attribute data of the target power equipment at the time point after the preset future time, and calculate the overlap ratio between the voiceprint signal data of the target power equipment and the historical abnormal voiceprint data corresponding to the attribute data of the target power equipment at the time point after the preset future time, and determine whether the target power equipment has a fault at the time point after the preset future time.

3. The AI-based power equipment voiceprint monitoring system as described in claim 1, characterized in that, The wavelet basis function g(t) is expressed as follows: In the formula, g(t) is the wavelet basis function, ω0 is the center frequency, a is a constant, t is the time variable, and i is the imaginary number. 2 =-1; where, f(t) is the acoustic signature signal data, and [-t0, t0] is the range of values ​​for the acoustic signature signal data f(t). The acoustic signature harmonic Ψf(b, τ) of the target power equipment is obtained by wavelet transform using the wavelet basis function g(t). The specific expression is: In the formula, Ψf(b,τ) is the acoustic harmonic of the target power equipment, b is the scale factor, and τ is the time displacement. Equipment normal harmonic information The specific expression is: In the formula, This is normal harmonic information from the equipment. It is the voiceprint data of the target power equipment during normal operation. in, The specific expression for the abnormal harmonic information Ψy(b,τ) is: In the formula, Ψy(b,τ) represents the abnormal harmonic information; The expression for the fault time-domain signal is: In the formula, y(t) is the fault time-domain signal, and C g These are the wavelet transform convergence coefficients; The specific expression for similarity is: In the formula, η is the similarity, and h(t) is a known fault signal of a certain type.

4. The AI-based power equipment voiceprint monitoring system as described in claim 1, characterized in that, The voiceprint recognition module includes: The feature extraction unit is used to construct a feature extraction network model and obtain historical voiceprint signals of power equipment to train the feature extraction network model. After training, the feature extraction network model is used to extract features of the target power equipment to obtain the feature data of the target power equipment. The feature recognition unit is used to acquire historical feature data of power equipment. After training, the power equipment feature recognition model is used to identify the target power equipment from the feature data. The equipment signal unit is used to acquire the corresponding voiceprint signal data based on the identified target power equipment.

5. The AI-based power equipment voiceprint monitoring system as described in claim 1, characterized in that, The voiceprint recognition module performs the following operations: Construct a feature extraction network model; Historical voiceprint signals of power equipment are acquired to train a feature extraction network model. After training, the feature extraction network model is used to extract features from the voiceprint signal data to obtain feature data. Construct a feature recognition model for power equipment; Historical feature data of power equipment is obtained to train the power equipment feature recognition model. After training, the power equipment feature recognition model is used to identify the target power equipment from the feature data. Obtain the corresponding voiceprint signal data based on the identified target power equipment.

6. The AI-based power equipment voiceprint monitoring system as described in claim 1, characterized in that, The voiceprint acquisition module includes: Several sound sensors are used to collect the acoustic signature signals of the target power equipment; Cameras are used for real-time video monitoring of target electrical equipment; A lifting unit is used to enable the sound sensor to move up and down; A horizontal movement unit, mounted on the lifting unit, is used for the horizontal movement of the sound sensor; The data acquisition and control unit is connected to the sound sensor, camera, horizontal movement unit, and lifting unit, respectively. It is used to plan the shortest monitoring path according to the location of each power device and to control the acquisition frequency of the sound sensor according to the type and failure rate of the power device.

7. The AI-based power equipment voiceprint monitoring system as described in claim 6, characterized in that, The data processing module includes: The noise reduction unit is used to reduce noise in the voiceprint signal; The segmentation unit is used to divide the voiceprint signal into several data interval segments according to the periodicity of the voiceprint signal. The fitting unit is used to fit several data interval segments after noise reduction according to the minimum mean square error to obtain the voiceprint signal data. The storage unit is used to store the voiceprint signal data.

8. The AI-based power equipment voiceprint monitoring system as described in claim 7, characterized in that, Also includes: The real-time monitoring module is used to perform real-time video monitoring of the target power equipment and process the data in a timely manner. The real-time monitoring module includes: Temperature sensor, used to measure the temperature of target electrical equipment; An ammeter is used to measure the current of a target electrical device. The power cut-off unit is used to determine whether to cut off the power supply to the target electrical equipment based on the overlap ratio and the current of the target electrical equipment. The equipment cooling unit is used to determine whether to cool the target power equipment based on the overlap ratio and the temperature measured on the target power equipment. The fire suppression unit is used to determine whether to extinguish the fire on the target electrical equipment by collecting real-time video footage.

9. A method for monitoring the voiceprint of power equipment based on AI, characterized in that, include: Collect acoustic signature signals from the target power equipment; The voiceprint signal is preprocessed to obtain voiceprint signal data; The target power equipment and its corresponding voiceprint signal data are identified based on the voiceprint signal data. Acquire the acoustic signature signal data corresponding to the target power equipment and perform wavelet transform, and identify the fault type and main fault of the target power equipment based on the wavelet transform; An alert is issued to the manager based on the assessment results; The process of acquiring the acoustic signature signal data corresponding to the target power equipment, performing wavelet transform, and identifying the fault type and main fault of the target power equipment based on the wavelet transform includes: The overlap ratio is calculated based on the acoustic fingerprint signal data corresponding to the target power equipment and the predicted acoustic fingerprint signal data of the target power equipment during normal operation. The overlap ratio is then used to determine whether the target power equipment has malfunctioned. If the target power equipment malfunctions, acquire the corresponding voiceprint signal data of the target power equipment and perform wavelet transform, and identify the fault type of the target power equipment based on the wavelet transform; The process of acquiring the acoustic signature signal data corresponding to the target power equipment, performing wavelet transform, and identifying the fault type of the target power equipment based on the wavelet transform includes: A wavelet transform function is constructed and wavelet decomposition is performed on the acoustic signature signal data corresponding to the target power equipment to obtain the acoustic signature harmonics of the target power equipment. Acquire acoustic signature data of the target power equipment during normal operation and perform wavelet transform to obtain the normal harmonic information of the equipment; The acoustic harmonic filtering of the target power equipment removes the normal harmonic information of the equipment and obtains the abnormal harmonic information. The abnormal harmonic information is reconstructed by wavelet transform to obtain the fault time-domain signal; Calculate the similarity between each fault time-domain signal and a known fault signal, and sort them according to the similarity. A similarity threshold is set, and the fault type of the target power equipment is obtained based on the known fault signals corresponding to the similarity exceeding the similarity threshold.

Images