Station transformer fault diagnosis early warning method based on sensing, communication, calculation and storage integration

Abstract

The invention relates to a station transformer fault diagnosis early warning method based on sensing, communication, calculation and storage integration, and the method comprises the steps: constructing an integrated system of real-time sensing, lightweight communication, local calculation and intelligent storage through the deep fusion of voiceprint collection hardware, edge calculation, an AI algorithm and data storage; multi-scene adaptive voiceprint acquisition hardware is utilized to realize real-time sensing of operation acoustic signals, key fault features are extracted in combination with feature engineering and an artificial intelligence algorithm, and an autonomous learning mechanism of edge computing and cloud collaboration is utilized to continuously optimize the fault recognition capability. A whole process solution integrating signal acquisition, feature processing, intelligent diagnosis and dynamic iteration is formed, a station transformer fault acoustic diagnosis and early warning system based on sensing, communication, calculation and storage integration is finally developed, early recognition and early warning of station transformer faults are achieved, and the industry is promoted to change from rule-driven threshold alarm to data-driven intelligent diagnosis. The method has the advantages of high applicability, high efficiency and convenience.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of transformer fault diagnosis, and particularly relates to a station transformer fault diagnosis and early warning method based on integrated sensing and algorithm storage. BACKGROUND

[0002] As the core equipment for providing continuous and stable power in a substation, the station transformer undertakes the key task of power supply for other equipment in the substation, and its safe operation directly determines the normal operation of the substation. Once an accident occurs, not only the substation will be powered off, but also serious secondary disasters such as equipment explosion, damage and fire may be caused, which will cause great threat to the stable, economic operation of the power system and social safety. At present, the fault diagnosis technology of high-voltage transformers is relatively mature, and methods such as partial discharge monitoring, oil chromatographic analysis and gas chromatographic monitoring are widely used and have achieved remarkable results. However, there is still a significant gap in the monitoring technology for medium and low voltage transformers such as station transformers. Moreover, the existing traditional contact monitoring method is difficult to adapt to the diversified scenarios of station transformers, and the traditional contact monitoring method also has the problems of high cost and great difficulty in deployment. Therefore, in order to improve the monitoring and diagnosis level of power equipment, it is necessary to develop a non-contact, highly applicable and efficient and convenient station transformer fault diagnosis and early warning method based on integrated sensing and algorithm storage. SUMMARY

[0003] The purpose of the present application is to overcome the deficiencies of the prior art and provide a non-contact, highly applicable and efficient and convenient station transformer fault diagnosis and early warning method based on integrated sensing and algorithm storage.

[0004] The purpose of the present application is achieved by a station transformer fault diagnosis and early warning method based on integrated sensing and algorithm storage, comprising the following steps: Step 1: using multiple sensors to collect the voiceprint time sequence data of the transformer, and integrating and filtering and denoising the collected voiceprint time sequence data; Step 2: based on wavelet packet energy spectrum decomposition and mel cepstral coefficient, and using feature extraction technology, performing time domain, frequency domain and nonlinear feature extraction operation on the voiceprint signal after integration and processing in step 1; and then using feature engineering technology to sort and select the extracted features; Step 3: performing zero-crossing rate calculation on the time domain voiceprint data of the station transformer extracted in step 2, so as to diagnose the state of the on-load tap changer of the station transformer; Step 4: performing total energy value calculation on the frequency domain voiceprint data of the station transformer extracted in step 2, so as to diagnose the state of the winding and iron core of the station transformer; Step 5: summarizing the diagnosis results in steps 3 and 4, issuing an alarm signal, and generating a preliminary fault diagnosis report; Step 6: Acquire historical information such as historical fault data, equipment dynamic operating conditions, and operation and maintenance records of station service transformers, and clean and standardize the acquired historical information to achieve time series alignment and unit unification of historical information; then, based on the processed historical information data, use the Pelican optimization algorithm to optimize the convolution kernel parameters of the preset convolutional neural network to build a fault handling model. Step 7: Input the preliminary fault diagnosis report from Step 5 into the fault handling model, analyze and match the fault data, identify the fault source, find the fault propagation path and root cause, and provide maintenance suggestions based on historical operation and maintenance records. After the maintenance suggestions are manually reviewed and approved, generate maintenance plans and maintenance schemes based on the maintenance suggestions, and distribute the maintenance plans and maintenance schemes to nearby maintenance teams to carry out operation and maintenance operations.

[0005] Furthermore, the zero-crossing rate in step 3 is calculated by dividing the number of times the time-domain acoustic data obtained by the accelerometer array crosses the zero level in each time-domain acoustic signal acquisition cycle by the total number of data acquired by the accelerometer array in each time-domain acoustic signal acquisition cycle. If the zero-crossing rate exceeds the set initial threshold of 10% or exceeds 50% of the historical zero-crossing rate benchmark value, the on-load tap changer of the transformer is diagnosed as being in a fault state; otherwise, the on-load tap changer of the transformer is diagnosed as being in a normal state.

[0006] Furthermore, the total energy value of the frequency domain acoustic data in step 4 is obtained by squaring all the frequency domain acoustic data values ​​obtained by Fourier transforming the time domain acoustic data in each acquisition cycle and then summing them up. If the total energy value exceeds the set initial threshold of 20% or exceeds the total energy value reference value of 50%, the transformer winding is diagnosed as loose fault state; otherwise, the transformer winding is diagnosed as normal state.

[0007] Furthermore, the integration operation of the voiceprint time sequence data in step 1 specifically involves: integrating and registering the data and performing structured processing, thereby increasing the compactness and density of the data and reducing the amount of data transmitted.

[0008] Furthermore, the sensor in step 1 is a bone conduction sensor or an ultrasonic sensor, which is fixed to the station transformer by a permanent magnet structure and acquires the sound signal of the station transformer.

[0009] Furthermore, the warning signal in step 5 can be issued through the warning module, and the warning module sends the warning signal at preset intervals. If a confirmation feedback instruction is received from the operation terminal, the sending stops; otherwise, it continues to send.

[0010] The beneficial effects of this invention are as follows: By deeply integrating acoustic fingerprint acquisition hardware, edge computing, AI algorithms, and data storage, this invention constructs an integrated system of "real-time perception, lightweight communication, local computing, and intelligent storage," breaking down technical barriers in fields such as power monitoring, electroacoustics, and machine learning, and providing a reusable technical framework for power IoT devices. Simultaneously, through multi-scenario adaptable acoustic fingerprint acquisition hardware, real-time perception of operating acoustic signals is achieved. Key fault features are extracted by combining feature engineering and artificial intelligence algorithms. The self-learning mechanism of edge computing and cloud collaboration continuously optimizes fault identification capabilities, forming a complete solution integrating signal acquisition, feature processing, intelligent diagnosis, and dynamic iteration. Ultimately, a station transformer fault acoustic diagnosis and early warning system based on integrated sensing, communication, computing, and storage is developed, enabling early identification and warning of station transformer faults. Furthermore, the self-learning function enables dynamic iteration of the AI ​​model, improving the adaptability of fault identification under complex operating conditions, driving the industry from "rule-driven" threshold alarms to "data-driven" intelligent diagnosis, accelerating the upgrade of transformer monitoring technology to core decision-making systems, and providing key support for unmanned substation operation. In summary, this invention has the advantages of strong applicability and high efficiency and convenience. Detailed Implementation

[0011] The present invention will now be further described.

[0012] Example: A fault diagnosis and early warning method for station transformers based on integrated sensing, communication, computing, and storage, comprising the following steps: Step 1: Use multiple sensors to collect the acoustic fingerprint time-series data of the transformer, and integrate and filter the collected acoustic fingerprint time-series data. The integration operation specifically involves: integrating and registering and performing structured processing to increase the compactness and density of the data and reduce the amount of data transmission. In addition, the sensors adopt bone conduction sensors or ultrasonic sensors, which are fixed to the station transformer through a permanent magnet structure to acquire the sound signal of the station transformer.

[0013] Step 2: Based on wavelet packet energy spectrum decomposition and Mel-frequency cepstral coefficients, and using feature extraction techniques, extract time-domain, frequency-domain, and nonlinear features from the integrated and processed voiceprint signal in Step 1; and then use feature engineering techniques to prioritize and select the extracted features. Step 3: Calculate the zero-crossing rate of the time-domain acoustic fingerprint data of the station service transformer extracted in Step 2 to diagnose the status of the on-load tap changer of the station service transformer. The zero-crossing rate is calculated by dividing the number of times the time-domain acoustic fingerprint data obtained by the accelerometer array crosses the zero level in each time-domain acoustic fingerprint signal acquisition cycle by the total number of data acquired by the accelerometer array in each time-domain acoustic fingerprint signal acquisition cycle. If the zero-crossing rate exceeds the set initial threshold of 10% or exceeds 50% of the historical zero-crossing rate benchmark value, the on-load tap changer of the transformer is diagnosed as being in a fault state; otherwise, the on-load tap changer of the transformer is diagnosed as being in a normal state. Step 4: Calculate the total energy value of the frequency domain acoustic fingerprint data of the station service transformer extracted in Step 2, thereby diagnosing the condition of the station service transformer windings and core. The total energy value of the frequency domain acoustic fingerprint data is obtained by squaring all the frequency domain acoustic fingerprint data values ​​obtained by Fourier transforming the time domain acoustic fingerprint data in each acquisition cycle and then summing them. If the total energy value exceeds the set initial threshold of 20% or exceeds the total energy value reference value of 50%, the transformer winding is diagnosed as loose fault; otherwise, the transformer winding is diagnosed as normal. Step 5: Summarize the diagnostic results from Steps 3 and 4, and generate a preliminary fault diagnosis report while issuing a warning signal through the warning module; furthermore, in this invention, the warning module can send warning signals at preset intervals. If a confirmation feedback instruction is received from the operating terminal, the sending will stop; otherwise, it will continue to send. Step 6: Acquire historical information such as historical fault data, equipment dynamic operating conditions, and operation and maintenance records of station service transformers, and clean and standardize the acquired historical information to achieve time series alignment and unit unification of historical information; then, based on the processed historical information data, use the Pelican optimization algorithm to optimize the convolution kernel parameters of the preset convolutional neural network to build a fault handling model. Step 7: Input the preliminary fault diagnosis report from Step 5 into the fault handling model, analyze and match the fault data, identify the fault source, find the fault propagation path and root cause, and provide maintenance suggestions based on historical operation and maintenance records. After the maintenance suggestions are manually reviewed and approved, generate maintenance plans and maintenance schemes based on the maintenance suggestions, and distribute the maintenance plans and maintenance schemes to nearby maintenance teams to carry out operation and maintenance operations.

[0014] In use, this invention first collects the acoustic signature time-series data of the transformer using multiple sensors, and then integrates and filters the collected data for noise reduction. During this process, bone conduction sensors or ultrasonic sensors are used, fixed to the station transformer via a permanent magnet structure to acquire the transformer's sound signal. The acoustic signature time-series data integration operation specifically involves: integration registration and structured processing to increase data compactness and density while reducing data transmission volume. Then, based on wavelet packet energy spectral decomposition and Mel-frequency cepstral coefficients, and using feature extraction techniques, time-domain, frequency-domain, and nonlinear features are extracted from the integrated and processed acoustic signature signal. Finally, the data is further processed using… Feature engineering techniques are used to prioritize and select extracted features. Then, zero-crossing rate calculations are performed on the extracted time-domain acoustic signature data of the station service transformer to diagnose the status of the on-load tap changer. The zero-crossing rate is calculated by dividing the number of times the time-domain acoustic signature data obtained from the accelerometer array crosses zero level in each time-domain acoustic signature signal acquisition cycle by the total number of data acquired by the accelerometer array in each time-domain acoustic signature signal acquisition cycle. If the zero-crossing rate exceeds a set initial threshold of 10% or exceeds 50% of the historical zero-crossing rate benchmark, the on-load tap changer of the transformer is diagnosed as faulty; otherwise, it is diagnosed as normal. Simultaneously, the extracted frequency-domain acoustic signature data of the station service transformer is also analyzed. The total energy value of the acoustic fingerprint data is calculated to diagnose the condition of the station service transformer windings and core. The total energy value of the frequency domain acoustic fingerprint data is obtained by squaring all frequency domain acoustic fingerprint data values ​​obtained from the Fourier transform of the time domain acoustic fingerprint data in each acquisition cycle, and then summing them. If the total energy value exceeds a set initial threshold of 20% or exceeds a reference value of 50%, the transformer winding is diagnosed as having a loose fault condition; otherwise, the transformer winding is diagnosed as being in a normal condition. Finally, the diagnostic results are summarized, and a preliminary fault diagnosis report is generated while issuing a warning signal through the warning module. After completing the above operations, historical fault data, equipment dynamic operating conditions, and maintenance records of the station service transformer are reviewed. The system acquires historical information and cleans and standardizes it to align the time series and unify the units. Then, based on the processed historical data, the system uses the Pelican optimization algorithm to optimize the kernel parameters of a pre-defined convolutional neural network, thus constructing a fault handling model. Finally, the preliminary fault diagnosis report is input into the fault handling model to analyze and match the fault data, identify the fault source, find the fault propagation path and root cause, and provide maintenance suggestions based on historical maintenance records. After the maintenance suggestions are manually reviewed and approved, a maintenance plan and solution are generated based on the suggestions and distributed to nearby maintenance teams for maintenance operations.This invention employs this structure, deeply integrating acoustic signature acquisition hardware, edge computing, AI algorithms, and data storage to construct an integrated system of "real-time perception, lightweight communication, local computing, and intelligent storage." This breaks down technical barriers in fields such as power monitoring, electroacoustics, and machine learning, providing a reusable technical framework for power IoT devices. Simultaneously, it achieves real-time perception of operating acoustic signals through multi-scenario adaptable acoustic signature acquisition hardware, extracts key fault features using feature engineering and artificial intelligence algorithms, and continuously optimizes fault identification capabilities through an autonomous learning mechanism combining edge computing and cloud collaboration. This forms a complete solution integrating signal acquisition, feature processing, intelligent diagnosis, and dynamic iteration, ultimately developing a station transformer fault acoustic diagnosis and early warning system based on integrated sensing, communication, computing, and storage, enabling early identification and warning of station transformer faults. Furthermore, the autonomous learning function enables dynamic iteration of the AI ​​model, improving fault identification adaptability under complex operating conditions, driving the industry from "rule-driven" threshold alarms to "data-driven" intelligent diagnosis, accelerating the upgrade of transformer monitoring technology to core decision-making systems, and providing key support for unmanned substation operation. In summary, this invention has the advantages of strong applicability and high efficiency and convenience.

[0015] The above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of the claims of the present invention.

Claims

1. A method for fault diagnosis and early warning of station transformers based on integrated sensing, communication, computing, and storage, characterized in that, Includes the following steps: Step 1: Use multiple sensors to collect the acoustic fingerprint timing data of the transformer, and integrate and filter the collected acoustic fingerprint timing data for noise reduction. Step 2: Based on wavelet packet energy spectrum decomposition and Mel-frequency cepstral coefficients, and using feature extraction techniques, extract time-domain, frequency-domain, and nonlinear features from the integrated and processed voiceprint signal in Step 1. Subsequently, feature engineering techniques were used to rank and select the extracted features based on their importance. Step 3: Calculate the zero-crossing rate of the time-domain acoustic fingerprint data of the station service transformer extracted in Step 2, thereby diagnosing the status of the on-load tap changer of the station service transformer. Step 4: Calculate the total energy value of the frequency domain acoustic signature data of the station service transformer extracted in Step 2, so as to diagnose the condition of the station service transformer windings and core. Step 5: Summarize the diagnostic results from Steps 3 and 4, issue a warning signal, and generate a preliminary fault diagnosis report. Step 6: Acquire historical information such as historical fault data, equipment dynamic operating conditions, and operation and maintenance records of station service transformers, and clean and standardize the acquired historical information to achieve time series alignment and unit unification of historical information; then, based on the processed historical information data, use the Pelican optimization algorithm to optimize the convolution kernel parameters of the preset convolutional neural network to build a fault handling model. Step 7: Input the preliminary fault diagnosis report from Step 5 into the fault handling model, analyze and match the fault data, identify the fault source, find the fault propagation path and root cause, and provide maintenance suggestions based on historical operation and maintenance records. After the maintenance suggestions are manually reviewed and approved, generate maintenance plans and maintenance schemes based on the maintenance suggestions, and distribute the maintenance plans and maintenance schemes to nearby maintenance teams to carry out operation and maintenance operations.

2. The fault diagnosis and early warning method for station transformers based on integrated sensing, communication, computing, and storage as described in claim 1, characterized in that: The zero-crossing rate in step 3 is calculated by dividing the number of times the time-domain acoustic data obtained by the accelerometer array crosses the zero level in each time-domain acoustic signal acquisition cycle by the total number of data acquired by the accelerometer array in each time-domain acoustic signal acquisition cycle. If the zero-crossing rate exceeds the set initial threshold of 10% or exceeds 50% of the historical zero-crossing rate benchmark value, the on-load tap changer of the diagnostic transformer is in a fault state; otherwise, the on-load tap changer of the diagnostic transformer is in a normal state.

3. The fault diagnosis and early warning method for station transformers based on integrated sensing, communication, computing, and storage as described in claim 1, characterized in that: The total energy value of the frequency domain acoustic fingerprint data in step 4 is obtained by squaring all the frequency domain acoustic fingerprint data values ​​obtained by Fourier transform of the time domain acoustic fingerprint data in each acquisition cycle and then summing them up. If the total energy value exceeds the set initial threshold of 20% or exceeds the total energy value reference value of 50%, the transformer winding is diagnosed as loose fault state; otherwise, the transformer winding is diagnosed as normal state.

4. The fault diagnosis and early warning method for station transformers based on integrated sensing, communication, computing, and storage as described in claim 1, characterized in that: The integration operation of voiceprint time-series data in step 1 specifically involves: integrating and registering the data and performing structured processing to increase the compactness and density of the data and reduce the amount of data transmitted.

5. The fault diagnosis and early warning method for station transformers based on integrated sensing, communication, computing, and storage as described in claim 1, characterized in that: The sensor in step 1 is a bone conduction sensor or an ultrasonic sensor, which is fixed to the station transformer by a permanent magnet structure and acquires the sound signal of the station transformer.

6. The method for fault diagnosis and early warning of station transformers based on integrated sensing, communication, computing, and storage as described in claim 1, characterized in that: The warning signal in step 5 can be sent through the warning module, and the warning module sends the warning signal at preset intervals. If a confirmation feedback instruction is received from the operation terminal, the sending will stop; otherwise, it will continue to send.