Power transformer online monitoring method and system based on voiceprint analysis

Abstract

The invention discloses a power transformer on-line monitoring method and system based on voiceprint analysis, and relates to the technical field of power transformer monitoring, and the method comprises the steps: S1, building a power transformer on-line monitoring system; s2, sound wave signal real-time data acquisition; s3, carrying out data processing; s4, performing fault diagnosis according to the data; s5, autonomous detection: adopting a sound wave generator to send out a sound wave signal, and enabling a voiceprint sensor to cooperate with data processing and fault diagnosis to perform self-inspection; s6, optimization learning is carried out; and S7, convergence of the station-level voiceprint data is carried out, real-time remote monitoring is carried out, and visual display is implemented. According to the power transformer on-line monitoring method based on voiceprint analysis, the functions of voiceprint collection and coverage of transformer substation high-voltage equipment, typical defect recognition of key equipment, abnormal warning and the like are achieved through the industrial voiceprint multi-means sensing technology, and the problems that transformer substation equipment is subjected to manual inspection and sound listening, and the working efficiency is low are solved. And abnormal voiceprint features of potential faults cannot be found in time.

Description

Technical Field

[0001] The present invention relates to the technical field of power transformer monitoring, and particularly to an on-line monitoring method and system for power transformers based on voiceprint analysis. Background Art

[0002] With the continuous development of intelligent inspection technology, the traditional manual inspection mode is gradually transforming towards less manned / unmanned operation. On-line monitoring technology, robot inspection technology, and UAV inspection technology are widely used in the inspection of power systems. Through the coordination of the above means, the inspection efficiency has been greatly improved. Before a serious accident occurs in substation power equipment, there are often latent signals. Visible light, infrared, and other means are mainly used to judge obvious abnormalities, and sound can characterize more early-stage latent faults of equipment.

[0003] During manual inspection, "listening for faults" on-site and making corresponding records are rigid requirements of the regulations. During the off-peak period outside periodic inspections, abnormal voiceprint characteristics of potential faults cannot be detected in a timely manner. In unattended substations, there is a lack of on-line voiceprint monitoring tools and technologies. Summary of the Invention

[0004] The purpose of the present invention is to provide an on-line monitoring method and system for power transformers based on voiceprint analysis to solve the problems raised in the above background art.

[0005] To achieve the above purpose, the present invention provides the following technical solutions: An on-line monitoring method for power transformers based on voiceprint analysis, including the following specific steps: S1: Establish an on-line monitoring system for power transformers, and perform real-time remote monitoring on power transformers based on voiceprint monitoring technology; S2: Real-time data acquisition of acoustic wave signals, and collect acoustic wave signals of power transformers using a voiceprint sensor; S5: Self-detection, use an acoustic wave generator to emit acoustic wave signals, and make the voiceprint sensor cooperate with data processing and fault diagnosis for self-check; S3: Perform data processing, process the collected acoustic wave signals to improve the signal-to-noise ratio; S4: Perform fault diagnosis based on the above data, use machine learning and pattern recognition algorithms to establish a relationship model between voiceprint characteristics and the operating state of the transformer, input the characteristics of the real-time collected acoustic wave signals into the model, and perform state recognition and fault diagnosis; S6: Optimize learning, combine networked AI technology to improve the performance and resource utilization rate of the on-line monitoring system for power transformers; S7: Aggregate station-level voiceprint data, perform real-time remote monitoring, intelligent analysis, and sound visualization, display the operating state of the equipment, and implement visual display.

[0006] Preferably, the voiceprint sensor for collecting the acoustic wave signal of the power transformer in S2 includes a non-contact industrial-grade sensor and a contact industrial-grade sensor. The frequency coverage range of the non-contact industrial-grade sensor and the contact industrial-grade sensor is 20Hz - 96kHz. The non-contact industrial-grade sensor is installed around the power transformer. The deployment positions of the contact industrial-grade sensor include, but are not limited to, more than one of the internal switches, disconnect switches, and switch contacts of the power transformer.

[0007] Preferably, the non-contact industrial-grade sensor transmits the sound signal through the sound wave in the air. The non-contact industrial-grade sensor is one of a dynamic microphone, a capacitive microphone, and an optical microphone. The dynamic microphone uses the principle of electromagnetic induction. The sound wave causes the diaphragm to vibrate, driving the coil to move in the magnetic field, thereby generating an electric current. The capacitive microphone uses the principle of capacitance change. The sound wave causes the diaphragm to vibrate, changing the capacitance value of the capacitor, and then generating an electrical signal. The optical microphone detects the sound wave through the change of laser or light beam, and is suitable for use in extreme environments. The contact microphone detects the sound wave signal by directly contacting the object. The type of the contact microphone includes one of a piezoelectric microphone and a strain gauge microphone. The piezoelectric microphone uses the piezoelectric effect to convert mechanical strain into an electrical signal, and is suitable for the detection of high-frequency sound waves. The strain gauge microphone captures the sound wave by detecting the strain change of the material.

[0008] Preferably, the self-detection in S5 includes the following steps: S5.1: Issue an audio simulation command to self-check the voiceprint sensor and shield the fault alarm; S5.2: Use a sound wave generator to simulate external detection audio and power transformer fault audio. The power transformer fault audio includes, but is not limited to, DC bias fault audio, partial discharge fault audio, heavy overload fault audio, cooler abnormal sound fault audio, clamp piece looseness fault audio, and short-circuit impact fault audio, and record the data information of the emitted audio; S5.3: The voiceprint sensor detects the audio emitted by the sound wave generator and uploads the collected audio data; S5.4: After the data processing in step S3 and the fault diagnosis in S4, compare whether the audio information collected by the voiceprint sensor is consistent with the audio data information emitted by the voiceprint sensor, and determine whether the voiceprint sensor is damaged. That is, when the sound wave generator simulates and emits external detection audio and the voiceprint sensor does not detect the external detection audio, the voiceprint sensor is damaged. On the contrary, if the voiceprint sensor detects the external detection audio, the voiceprint sensor is not damaged; S5.5: Error evaluation. The sound wave generator simulates the power transformer fault audio and calculates the accuracy of the voiceprint sensor in detecting the power transformer fault audio.

[0009] Preferably, the error assessment in S5.5 is that the sound wave generator simulates the DC bias magnetic fault audio, partial discharge fault audio, heavy overload fault audio, cooler abnormal noise fault audio, clamp loose fault audio and short circuit impact fault audio respectively, and records the audio signal emitted by the sound wave generator, the soundprint sensor detects the audio emitted by the sound wave generator, and the data of the audio detected by the soundprint sensor is obtained by the data processing of step S3 and the fault diagnosis of S4, and compared with the audio signal emitted by the sound wave generator, and the accuracy is obtained by the following formula: ;in For a real example, is a true negative example, For a false positive example, is a false negative example.

[0010] Preferably, the data processing in S3 includes data collection, data cleaning, feature extraction and model building. Data collection collects data by monitoring the performance indicators during the operation of the software. Data cleaning cleans and preprocesses the collected data to remove noise and invalid data. Feature extraction extracts performance-related features from the original data for subsequent modeling, and performs feature selection to select features that have a significant impact on performance optimization, thereby reducing the complexity of the model and providing data for subsequent fault diagnosis in S4.

[0011] Preferably, the fault diagnosis in S4 includes model building, online monitoring, fault analysis and warning mechanism. Model building is to establish a relationship model between voiceprint features and transformer operating status through a neural network, and train the model. The model is trained using historical data so that the model understands the relationship between software performance and features, and the parameters of the model are adjusted through methods such as cross-validation. The acoustic wave signal features collected in real time are input into the model to perform state recognition and fault diagnosis. It is determined whether the power transformer has a fault based on the voiceprint features. Fault analysis is to determine the fault type of the power transformer based on the voiceprint features. The fault types of the power transformer include but are not limited to DC bias magnetism, partial discharge, heavy overload, abnormal sound of the cooler, loose clamps and short-circuit impact. The alarm mechanism is that once an abnormal voiceprint is detected, the system will sound an alarm and record relevant data.

[0012] Preferably, the optimized learning in S6 includes the application of deep learning, multi-source data fusion, edge computing and intelligent operation and maintenance. The application of deep learning is to use deep learning models to process complex voiceprint features. Multi-source data fusion is to combine voiceprint data with multiple sensor data such as temperature, vibration and current for comprehensive analysis. Edge computing is to perform data processing and model reasoning on the device side to reduce latency and improve response speed. Intelligent operation and maintenance is to combine Internet of Things technology to achieve intelligent operation and maintenance.

[0013] Preferably, the visual display in S7 includes a power model, a voiceprint feature map, a fault detection result, and a comparison with historical data. The voiceprint feature map is a feature map drawn by extracting voiceprint features, and the feature map includes but is not limited to spectrograms and time-domain waveforms. Figure 1 More than one type, the fault detection result is to display the prediction result of the fault diagnosis model in a graphical way, using bar charts and pie charts to display the probability distribution of different fault types, helping users quickly identify potential faults. The historical data comparison is to display the comparison between historical voiceprint data and current data through line charts and heat maps, assisting in analyzing the performance changes and fault development trends of transformers.

[0014] The present invention also provides an on-line monitoring system for power transformers based on voiceprint analysis, including: A data acquisition module for real-time detecting the voiceprint data of the power transformer and uploading the data. A data processing module for optimizing the data collected by the data acquisition module to provide accurate data for subsequent processing. A fault analysis module for performing fault analysis and judgment on the data optimized by the data processing module. A visualization module for displaying the data of on-line monitoring of the power transformer to assist personnel in understanding the information of on-line monitoring of the power transformer. A data storage module for storing the data generated by on-line monitoring of the power transformer and establishing a database required for comparative analysis by the fault analysis module. A network module for providing network data services for the on-line monitoring system of the power transformer. A self-check module for performing fault self-check on the data acquisition module.

[0015] The technical effects and advantages of the present invention: (1) The present invention uses an on-line monitoring method for power transformers based on voiceprint analysis. Through multi-means perception technology of industrial voiceprints, it realizes the coverage of voiceprint collection of high-voltage equipment in substations, ensures the 7*24H remote and secure collection and monitoring of voiceprint data, and functions such as typical defect identification and abnormal alarm of key equipment, solving the problem that abnormal voiceprint features of potential faults cannot be detected in time during manual inspection of substation equipment by listening. (2) The present invention uses an on-line monitoring method for power transformers based on voiceprint analysis. By using the combination of real-time data acquisition and autonomous detection of acoustic wave signals, the voiceprint sensor can detect the audio emitted by the acoustic wave generator. Thus, by comparing the information of the acoustic wave generator and the information detected by the voiceprint sensor, not only can it be determined whether the voiceprint sensor is damaged, but also the detection accuracy of the voiceprint sensor can be calculated. Moreover, the combination of the voiceprint sensor and the acoustic wave generator can also optimize the fault analysis of the power transformer. Description of the Drawings

[0016] Figure 1 This is the flowchart of the on-line monitoring method for power transformers based on voiceprint analysis of the present invention. Specific embodiments

[0017] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0018] The present invention provides an on-line monitoring method for power transformers based on voiceprint analysis as Figure 1 shown.

[0019] Embodiment 1: It includes the following specific steps: S1: Establish an on-line monitoring system for power transformers, and perform real-time remote monitoring on power transformers based on voiceprint monitoring technology; S2: Real-time data acquisition of acoustic wave signals, and use a voiceprint sensor to collect the acoustic wave signals of power transformers; S5: Self-detection, use an acoustic wave generator to emit acoustic wave signals, and make the voiceprint sensor cooperate with data processing and fault diagnosis for self-check; S3: Perform data processing, process the collected acoustic wave signals, improve the signal-to-noise ratio of the signals, ensure the accuracy of subsequent analysis, and reduce the influence of noise on the on-line monitoring of power transformers; S4: Perform fault diagnosis based on the above data, use machine learning and pattern recognition algorithms to establish a relationship model between voiceprint features and the operating state of transformers, input the features of the real-time collected acoustic wave signals into the model, and perform state recognition and fault diagnosis; S6: Optimize learning, combine networked AI technology to improve the performance and resource utilization rate of the on-line monitoring system for power transformers, thereby improving the performance of the on-line monitoring of power transformers; S7: Aggregate the voiceprint data at the station level, perform real-time remote monitoring, intelligent analysis and sound visualization, display the operating state of the equipment, and implement visual display.

[0020] Further, the voiceprint sensors for collecting acoustic wave signals of power transformers in S2 include non-contact industrial-grade sensors and contact industrial-grade sensors. The frequency coverage range of the non-contact industrial-grade sensors and contact industrial-grade sensors is 20 Hz - 96 kHz. The non-contact industrial-grade sensors are installed around the power transformer, and the deployment positions of the contact industrial-grade sensors include, but are not limited to, more than one of the internal switches, disconnect switches, and switch contacts of the power transformer. The non-contact industrial-grade sensors transmit sound signals through sound waves in the air. The non-contact industrial-grade sensors are one of dynamic microphones, capacitive microphones, and optical microphones. The dynamic microphone uses the principle of electromagnetic induction. Sound waves cause the diaphragm to vibrate, driving the coil to move in the magnetic field, thereby generating an electric current. The capacitive microphone uses the principle of capacitance change. Sound waves cause the diaphragm to vibrate, changing the capacitance value of the capacitor, and then generating an electrical signal. The optical microphone detects sound waves through changes in laser or light beams and is suitable for use in extreme environments. The contact microphone detects sound wave signals by directly contacting the object. The types of contact microphones include one of piezoelectric microphones and strain gauge microphones. The piezoelectric microphone uses the piezoelectric effect to convert mechanical strain into an electrical signal and is suitable for detecting high-frequency sound waves. The strain gauge microphone captures sound waves by detecting changes in the strain of the material.

[0021] In particular, the self-detection in S5 includes the following steps: S5.1: Issue an audio simulation command to self-check the voiceprint sensor and shield the fault alarm to avoid false alarms of power transformer faults when self-checking the voiceprint sensor. S5.2: Use a sound wave generator to simulate external detection audio and power transformer fault audio. The power transformer fault audio includes, but is not limited to, DC bias fault audio, partial discharge fault audio, heavy overload fault audio, cooler abnormal sound fault audio, clamp loosening fault audio, and short-circuit impact fault audio, and record the data information of the emitted audio. S5.3: The voiceprint sensor detects the audio emitted by the sound wave generator and uploads the collected audio data. S5.4: After the data processing in step S3 and the fault diagnosis in S4, compare whether the audio information collected by the voiceprint sensor is consistent with the audio data information emitted by the voiceprint sensor, and determine whether the voiceprint sensor is damaged. That is, when the sound wave generator simulates and emits external detection audio and the voiceprint sensor does not detect the external detection audio, the voiceprint sensor is damaged. On the contrary, if the voiceprint sensor detects the external detection audio, the voiceprint sensor is not damaged. Regularly performing self-detection on the voiceprint sensor can avoid the problem that the voiceprint sensor fails and is damaged and cannot be detected in time. S5.5: Error evaluation. The sound wave generator simulates the power transformer fault audio and calculates the accuracy of the voiceprint sensor in detecting the power transformer fault audio.

[0022] Furthermore, the error assessment in S5.5 is that the sound wave generator simulates the DC bias fault audio, partial discharge fault audio, heavy overload fault audio, cooler abnormal noise fault audio, clamp loose fault audio and short circuit impact fault audio respectively, and records the audio signal emitted by the sound wave generator, the soundprint sensor detects the audio emitted by the sound wave generator, and the data of the audio detected by the soundprint sensor is obtained by the data processing of step S3 and the fault diagnosis of S4, and compared with the audio signal emitted by the sound wave generator, and the accuracy is obtained by the following formula: ;in For a real example, is a true negative example, For a false positive example, For false negative examples, the accuracy of voiceprint sensor detection can be used to give the accuracy of voiceprint sensor detection of power transformer fault detection data, and the accuracy of voiceprint sensor online fault monitoring of power transformer can also be trained through the autonomous detection step to improve the accuracy of voiceprint sensor detection of power transformer fault detection data, that is, the DC bias magnetic fault audio, partial discharge fault audio, heavy overload fault audio, cooler abnormal noise fault audio, loose clamp fault audio and short circuit impact fault audio are simulated by the sound wave generator respectively, so that the voiceprint sensor detects the fault audio, and then a large amount of data training is performed on data processing and fault diagnosis, and the training data is stored in the database, which provides important data basis for subsequent data processing and fault diagnosis training and diagnosis, thereby improving the accuracy of voiceprint sensor detection of power transformer fault detection data.

[0023] Specifically, the data processing in S3 includes data collection, data cleaning, feature extraction and model building. Data collection collects data by monitoring the performance indicators during the operation of the software. Data cleaning cleans and preprocesses the collected data to remove noise and invalid data. Feature extraction extracts performance-related features from the original data for subsequent modeling and feature selection to select features that have a significant impact on performance optimization, reduce the complexity of the model, and provide data for subsequent fault diagnosis in S4. Data processing belongs to the existing technology and will not be introduced in detail here.

[0024] Furthermore, the fault diagnosis in S4 includes model establishment, online monitoring, fault analysis, and warning mechanism. Model establishment is to establish a relationship model between voiceprint features and the operating state of the transformer through a neural network and train the model. Historical data is used to train the model to enable the model to understand the relationship between software performance and features, and the parameters of the model are adjusted through methods such as cross-validation to improve the prediction accuracy of the model. The characteristics of the acoustic wave signals collected in real time are input into the model for state recognition and fault diagnosis. Whether there is a fault in the power transformer is judged according to the voiceprint features. Fault analysis is to judge the fault type of the power transformer according to the voiceprint features. The fault types of the power transformer include but are not limited to more than one of DC bias, partial discharge, heavy overload, abnormal sound of the cooler, clamp loosening, and short-circuit impact, so as to classify the faults of the power transformer, facilitate the analysis and confirmation of the types of power transformer faults, and facilitate the subsequent maintenance of the power transformer. The warning mechanism is that once abnormal voiceprint is detected, the system will issue an alarm and record relevant data for subsequent analysis and processing.

[0025] Embodiment 2: On the basis of Embodiment 1, the optimization learning in S6 includes the application of deep learning, multi-source data fusion, edge computing, and intelligent operation and maintenance. The application of deep learning is to use a deep learning model to process complex voiceprint features to improve the accuracy of fault detection. Multi-source data fusion is to comprehensively analyze by combining voiceprint data with data from multiple sensors such as temperature, vibration, and current to enhance the fault prediction ability. Edge computing is to perform data processing and model inference at the device end to reduce latency and improve the response speed. Intelligent operation and maintenance is to combine Internet of Things technology to achieve intelligent operation and maintenance and improve the automation level of the power system.

[0026] Furthermore, the visual display in S7 includes a power model, a voiceprint feature map, a fault detection result, and a comparison with historical data. The voiceprint feature map is a feature map drawn through the extracted voiceprint features. The feature map includes but is not limited to more than one of a spectrogram and a time-domain waveform Figure 1 to help users understand the operating state of the transformer. The fault detection result is to display the prediction result of the fault diagnosis model in a graphical way, and use bar charts and pie charts to display the probability distribution of different fault types to help users quickly identify potential faults. Through real-time data visualization, the operation and maintenance personnel can timely discover the abnormal state of the transformer and quickly take measures. The comparison with historical data is to display the comparison between historical voiceprint data and current data through line charts and heat maps to assist in analyzing the performance change and fault development trend of the transformer. By visualizing complex data, it helps the management make data-based decisions and improve the scientific nature of transformer maintenance and management. The visual display can also be used as a tool for training and demonstration to help new employees or customers understand the principle of voiceprint analysis and the functions of the monitoring system.

[0027] The present invention also provides an on-line monitoring system for power transformers based on voiceprint analysis, including a data acquisition module, a data processing module, a fault analysis module, a visualization module, a data storage module, a network module and a self-check module. The data acquisition module is used to detect the voiceprint data of the power transformer in real time and upload the data. The data processing module is used to optimize the data collected by the data acquisition module to provide accurate data for the follow-up. The fault analysis module is used to analyze and judge the faults of the data optimized by the data processing module. The visualization module is used to display the data of the on-line monitoring of the power transformer to assist personnel in understanding the information of the on-line monitoring of the power transformer. The data storage module is used to store the data generated by the on-line monitoring of the power transformer and establish a database required for comparative analysis by the fault analysis module, providing a positive basis for the fault analysis module to analyze the faults of the power transformer. The network module is used to provide network data services for the on-line monitoring system of the power transformer. The self-check module is used to perform fault self-check on the data acquisition module.

[0028] Finally, it should be noted that the above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, for those skilled in the art, they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for online monitoring of power transformers based on voiceprint analysis, characterized in that: The specific steps include: S1: Establish an online monitoring system for power transformers and conduct real-time remote monitoring of power transformers based on voiceprint monitoring technology; S2: Real-time data collection of acoustic wave signals, using voiceprint sensors to collect acoustic wave signals of power transformers; S3: Perform data processing to process the collected sound wave signals and improve the signal-to-noise ratio of the signals; S4: Perform fault diagnosis based on the above data, use machine learning and pattern recognition algorithms to establish a relationship model between the soundprint characteristics and the transformer operating status, input the real-time collected sound wave signal characteristics into the model, and perform status identification and fault diagnosis; S5: Autonomous detection, using the sound wave generator to send out sound wave signals, so that the voiceprint sensor can perform self-detection in conjunction with data processing and fault diagnosis; S6: Optimize learning and combine networked AI technology to improve the performance and resource utilization of the power transformer online monitoring system; S7: Aggregate station-level voiceprint data to conduct real-time remote monitoring, intelligent analysis and sound visualization, display equipment operation status, and implement visual display.

2. The method for online monitoring of power transformers based on voiceprint analysis according to claim 1 is characterized in that: The voiceprint sensor for collecting the acoustic wave signal of the power transformer in S2 includes a non-contact industrial-grade sensor and a contact industrial-grade sensor. The frequency coverage range of the non-contact industrial-grade sensor and the contact industrial-grade sensor is 20Hz-96kHz. The non-contact industrial-grade sensor is installed around the power transformer. The deployment location of the contact industrial-grade sensor includes but is not limited to one or more internal switches, knife switches and switch contacts of the power transformer.

3. The method for online monitoring of power transformers based on voiceprint analysis according to claim 2 is characterized in that: The non-contact industrial-grade sensor transmits sound signals through sound waves in the air. The non-contact industrial-grade sensor is one of a dynamic microphone, a condenser microphone and an optical microphone. The dynamic microphone uses the principle of electromagnetic induction. Sound waves make the diaphragm vibrate, driving the coil to move in the magnetic field, thereby generating current. The condenser microphone uses the principle of capacitance change. Sound waves make the diaphragm vibrate, changing the capacitance value of the capacitor, and then generating an electrical signal. The optical microphone detects sound waves through changes in lasers or light beams and is suitable for use in extreme environments. The contact microphone detects sound wave signals by directly contacting an object. The types of the contact microphone include one of a piezoelectric microphone and a strain gauge microphone. The piezoelectric microphone uses the piezoelectric effect to convert mechanical strain into electrical signals and is suitable for detecting high-frequency sound waves. The strain gauge microphone captures sound waves by detecting strain changes in materials.

4. The method for online monitoring of power transformers based on voiceprint analysis according to claim 1 is characterized in that: The autonomous detection in S5 includes the following steps: S5.1: Issue an audio simulation command to perform a self-check on the voiceprint sensor and shield the fault alarm; S5.2: Use a sound wave generator to simulate external detection audio and power transformer fault audio. The power transformer fault audio includes but is not limited to DC bias fault audio, partial discharge fault audio, heavy overload fault audio, cooler abnormal noise fault audio, clamp loose fault audio and short circuit impact fault audio, and record the data information of the emitted audio; S5.3: The voiceprint sensor detects the audio emitted by the sound wave generator and uploads the collected audio data; S5.4: After the data processing in step S3 and the fault diagnosis in step S4, the audio information collected by the voiceprint sensor is compared with the audio data information sent by the voiceprint sensor to see whether they are consistent, and whether the voiceprint sensor is damaged is determined, that is, if the sound wave generator simulates the external detection audio, and the voiceprint sensor does not detect the external detection audio, then the voiceprint sensor is damaged; otherwise, if the voiceprint sensor detects the external detection audio, then the voiceprint sensor is not damaged; S5.5: Error evaluation, the sound wave generator simulates the power transformer fault audio, and calculates the accuracy of the voiceprint sensor in detecting the power transformer fault audio.

5. The method for online monitoring of power transformers based on voiceprint analysis according to claim 4 is characterized in that: The error assessment in S5.5 is that the sound wave generator simulates the audio of DC bias magnetic fault, partial discharge fault, heavy overload fault, cooler abnormal noise fault, loose clamp fault and short circuit impact fault respectively, and records the audio signal emitted by the sound wave generator. The soundprint sensor detects the audio emitted by the sound wave generator, and the data of the audio detected by the soundprint sensor is obtained by the data processing of step S3 and the fault diagnosis of S4, and compared with the audio signal emitted by the sound wave generator, and the accuracy is obtained by the following formula: ;in For a real example, is a true negative example, For a false positive example, is a false negative example.

6. The method for online monitoring of power transformers based on voiceprint analysis according to claim 1 is characterized in that: The data processing in S3 includes data collection, data cleaning, feature extraction and model building. Data collection collects data by monitoring the performance indicators during the operation of the software. Data cleaning cleans and preprocesses the collected data to remove noise and invalid data. Feature extraction extracts performance-related features from the original data for subsequent modeling and performs feature selection to select features that have a significant impact on performance optimization, thereby reducing the complexity of the model and providing data for subsequent fault diagnosis in S4.

7. The method for online monitoring of power transformers based on voiceprint analysis according to claim 6 is characterized in that: The fault diagnosis in S4 includes model building, online monitoring, fault analysis and warning mechanism. Model building is to build a relationship model between voiceprint features and transformer operating status through a neural network, and train the model. The model is trained using historical data so that the model understands the relationship between software performance and features, and the parameters of the model are adjusted through methods such as cross-validation. The acoustic wave signal features collected in real time are input into the model to perform state recognition and fault diagnosis. It is determined whether the power transformer has a fault based on the voiceprint features. Fault analysis is to determine the fault type of the power transformer based on the voiceprint features. The fault types of the power transformer include but are not limited to DC bias magnetism, partial discharge, heavy overload, abnormal sound of the cooler, loose clamps and short-circuit impact. The alarm mechanism is that once an abnormal voiceprint is detected, the system will sound an alarm and record relevant data.

8. The method for online monitoring of power transformers based on voiceprint analysis according to claim 1 is characterized in that: The optimized learning in S6 includes the application of deep learning, multi-source data fusion, edge computing and intelligent operation and maintenance. The application of deep learning is to use deep learning models to process complex voiceprint features. Multi-source data fusion is to combine voiceprint data with multiple sensor data such as temperature, vibration and current for comprehensive analysis. Edge computing is to perform data processing and model reasoning on the device side to reduce latency and improve response speed. Intelligent operation and maintenance is to combine Internet of Things technology to achieve intelligent operation and maintenance.

9. The method for online monitoring of power transformers based on voiceprint analysis according to claim 1, characterized in that: The visualization display in S7 includes power model, voiceprint feature graph, fault detection result and historical data comparison. The voiceprint feature graph is a feature graph drawn by extracting voiceprint features. The feature graph includes but is not limited to one or more frequency spectrum graph and time domain waveform graph. The fault detection result is to display the prediction result of the fault diagnosis model in a graphical manner, and use bar graphs and pie charts to display the probability distribution of different fault types to help users quickly identify potential faults. The historical data comparison is to display the comparison between historical voiceprint data and current data through line graphs and heat maps to assist in analyzing the performance changes and fault development trends of the transformer.

10. An online monitoring system for power transformers based on voiceprint analysis, implementing the online monitoring method for power transformers based on voiceprint analysis according to any one of claims 1 to 9, characterized in that: include: Data acquisition module, used to detect power transformer voiceprint data in real time and upload the data; The data processing module is used to optimize the data collected by the data acquisition module and provide accurate data for subsequent use; A fault analysis module is used to analyze and judge the faults of the data after being optimized and processed by the data processing module; Visualization module, used to display the data of online monitoring of power transformers, and assist personnel to understand the information of online monitoring of power transformers; The data storage module is used to store the data generated by the online monitoring of the power transformer and to establish the database required for comparative analysis by the fault analysis module; Network module, used to provide network data services for the online monitoring system of power transformers; The self-check module is used to perform fault self-check on the data acquisition module.

Images