On-load tap-changer oil chamber pressure monitoring method for extra-high voltage converter transformer

Abstract

The invention discloses an on-load tap-changer oil chamber pressure monitoring method for an extra-high voltage converter transformer. Comprising the steps of collecting pressure, environment and operation condition data, performing data cleaning and normalization processing, establishing a pressure prediction model, extracting features, selecting an algorithm optimization model, performing pressure monitoring and fault early warning, calculating deviation, judging abnormity, analyzing fault types and performing early warning. And during acquisition, various sensors are adopted and are preferably arranged. Data processing has a plurality of specific operations. The model adopts an optimization means of a bidirectional LSTM network. Fault analysis adopts a decision tree method and is combined with oil chromatography data verification, and a three-dimensional fault traceability technology is also added. Compared with the prior art, the method has the advantages that the prediction accuracy is improved by collecting the multi-dimensional data, the abnormity is timely found through real-time monitoring, the model performance is improved through data preprocessing, the phenomena of false alarm and missing alarm are reduced, and the accuracy and reliability of the monitoring result are improved.

Description

Technical Field

[0001] The present invention relates to the field of power equipment monitoring applications, and in particular to a method for monitoring the oil chamber pressure of an on-load tap changer for an ultra-high voltage converter transformer. Background Art

[0002] An on-load tap-changer is a critical component in a transformer used to adjust the voltage by changing the winding turns ratio under load. Its oil compartment provides insulation and arc extinguishing during switching operations. Stable oil compartment pressure is crucial for the proper operation of the on-load tap-changer.

[0003] Common pressure monitoring often simply sets a fixed pressure threshold for alarm, which fails to fully consider the impact of different operating conditions, environmental factors and equipment aging on the oil chamber pressure. It is prone to false alarms or missed alarms, and cannot accurately and timely reflect the actual pressure status and potential fault risks of the oil chamber, posing a hidden danger to the safe and stable operation of the transformer, and cannot meet the working requirements of power equipment monitoring applications. Therefore, a method for monitoring the oil chamber pressure of an on-load tap-changer for ultra-high voltage converter transformers is proposed. Summary of the Invention

[0004] The present invention provides the following technical solution: a method for monitoring the oil compartment pressure of an on-load tap changer for an ultra-high voltage converter transformer, comprising: S1 data collection: S11 pressure data acquisition: First, pressure sensors are installed at the top, middle, and bottom of the on-load tap-changer oil compartment, and the oil compartment pressure data is collected in real time through the pressure sensors. S12 environmental data collection: At the same time, a temperature sensor, a humidity sensor and an atmospheric pressure sensor are installed outside the oil chamber to obtain data on the temperature, humidity and atmospheric pressure of the surrounding environment of the oil chamber; S13 operating condition data collection: Obtain operating condition data such as tap position switching times, switching time intervals, and transformer load current from the on-load tap-changer control system and transformer monitoring system; S2 data processing: Based on the pressure, environment and operating condition data collected in step S1, a sliding average filter algorithm is used to clean the data to remove noise and outliers in the data, and then the cleaned data is normalized; S3 establishes a pressure prediction model: Extract features from the data processed in step S2, and select a machine learning algorithm to establish an oil chamber pressure prediction model. During the training process, a cross-validation method is used to optimize the model and adjust the model parameters. S4 pressure monitoring and fault warning: The data collected in real time in step S1 and pre-processed in step S2 is input into the pressure prediction model trained in step S3 to obtain the predicted value of the oil chamber pressure, and the deviation between the actually collected oil chamber pressure value and the predicted value is calculated. When the deviation exceeds the set threshold, it is judged that there is an abnormality in the oil chamber pressure. Based on the size and change trend of the deviation, the type of fault is further analyzed, and a corresponding fault warning signal is issued simultaneously.

[0005] Preferably, the pressure sensor in step S11 is a MEMS resonant sensor, the measuring range of the pressure sensor is -0.1MPa to 0.5MPa, the number of pressure sensors installed at each measuring point is 2-4 groups, and the surface of the pressure sensor is coated with a polytetrafluoroethylene coating.

[0006] Preferably, in addition to installing the temperature sensor, humidity sensor and atmospheric pressure sensor in step S12, a wind speed sensor is simultaneously installed at the ventilation position around the oil chamber. The wind speed sensor uses the ultrasonic principle for measurement, that is, the wind speed is calculated by transmitting and receiving ultrasonic signals.

[0007] Preferably, the operating condition data acquisition in step S13 obtains real-time status information of the tap changer from the control system through the IEC61850 protocol, wherein the real-time status information of the tap changer includes the tap position code, the switching action time, and the motor drive current waveform. At the same time, the transformer load current is obtained through the optical fiber current transformer, and the obtained data is marked with a unified time stamp and stored in the time series database.

[0008] Preferably, the data processing in step S2 specifically includes: using a sliding average filter with a window size of 5 to denoise the pressure data, and using the 3σ criterion to eliminate abnormal environmental data, while performing linear interpolation on the operating condition data to fill missing values, and then normalizing all features to the [0,1] interval through the min-max method, and finally storing the processed data in the data lake and establishing a dual index of time and space.

[0009] Preferably, the feature extraction in step S3 includes calculating the time domain features and frequency domain features of the pressure data, the gradient change features of the environmental parameters, and the dimensional feature vectors of the tap switching frequency and the load change rate, and the feature selection uses a random forest algorithm to evaluate the importance.

[0010] Preferably, the machine learning algorithm in step S3 adopts a bidirectional LSTM network, the bidirectional LSTM network structure includes an input layer, a bidirectional LSTM layer, an attention mechanism layer and a fully connected output layer, and the oil chamber pressure prediction model is internally provided with an automatic incremental update mechanism every 24 hours.

[0011] Preferably, the fault type analysis in step S4 is performed by a decision tree method, that is, when the pressure is continuously low, it is determined to be a leakage fault, when the pressure fluctuates at a high frequency, it is determined to be an arc fault, and when the pressure changes in a step, it is determined to be a mechanical jam. At the same time, when analyzing the fault type, the oil chromatography online monitoring data is synchronously combined to verify the diagnostic results.

[0012] Preferably, in step S4, a three-dimensional fault tracing technology is added simultaneously when performing fault analysis. The three-dimensional fault tracing technology first captures the vibration signal when the pressure is abnormal in real time by arranging a distributed fiber optic acoustic wave sensor array in the oil chamber, and then uses the wave arrival direction estimation algorithm to locate the origin of the abnormal pressure wave, and reconstructs the three-dimensional pressure gradient field of the oil chamber in combination with the pressure sensor data to visualize the fault diffusion path.

[0013] Preferably, when processing the pressure data in step S2, the pressure data is synchronously sent to the time-frequency joint analysis module. After receiving the data, the time-frequency joint analysis module performs wavelet transform and short-time Fourier transform on the pressure data, extracts the characteristic energy ratio of the 0.1-10Hz frequency band, and constructs a fingerprint of the pressure pulsation. When an abnormal increase in the characteristic frequency energy is detected, a preventive maintenance reminder is triggered.

[0014] In summary, compared with the prior art, the present invention provides a method for monitoring the oil compartment pressure of an on-load tap-changer for a UHV converter transformer, which has the following beneficial effects: 1. The present invention installs temperature sensors, humidity sensors, and atmospheric pressure sensors on the outside of the oil chamber, and obtains operating condition data from related systems to comprehensively acquire various information related to the oil chamber pressure. When establishing a pressure prediction model, this multi-dimensional data is used as input. The model can learn the complex relationships between different factors, making the pressure predictions made by the model more consistent with the actual situation during the actual monitoring process, reducing the phenomenon of false alarms and missed alarms, and improving the accuracy and reliability of the monitoring results. 2. The present invention collects oil chamber pressure data and related environmental and operating condition data in real time, and continuously performs deviation analysis using an oil chamber pressure prediction model. At each time point, the actual collected oil chamber pressure value is compared with the predicted value obtained by the pressure prediction model, and the deviation is calculated. When the deviation exceeds a set threshold, abnormal changes in the oil chamber pressure can be detected in a timely manner. This real-time monitoring and analysis mechanism can capture abnormal signs in the early stages of a fault, providing sufficient time for equipment maintenance and repair, reducing the risk of equipment damage, and ensuring the safe and stable operation of the transformer. 3. The present invention performs data cleaning through a sliding average filtering algorithm, which can effectively remove random noise in the data. At the same time, through normalization processing, data of different magnitudes are converted into a unified range, so that the data are comparable in subsequent processing and analysis. After data preprocessing, a feature extraction operation is performed to extract the features that best reflect the law of oil chamber pressure changes from the processed data, which can provide more valuable input for the pressure prediction model, so that the oil chamber pressure prediction model can better learn the inherent laws in these data, thereby improving the performance of the oil chamber pressure prediction model, so that it can more accurately reflect the actual situation when predicting the oil chamber pressure, and improve the prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a schematic flow chart of the method of the present invention.

[0016] Figure 2 This is a flow chart of step S1 of the present invention.

[0017] Figure 3 It is a flowchart of the present invention. DETAILED DESCRIPTION

[0018] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0019] See also Figure 1 and Figure 3 , the present invention provides a technical solution, a method for monitoring the oil compartment pressure of an on-load tap-changer for a UHV converter transformer, comprising the following steps; S1 data collection: See also Figure 2 , S11 pressure data acquisition: First, pressure sensors are installed at the top, middle, and bottom of the on-load tap-changer oil compartment. The pressure sensors collect oil compartment pressure data. The pressure sensors use MEMS resonant sensors with a measurement range of -0.1 MPa to 0.5 MPa. Two to four sets of pressure sensors are installed at each measuring point. The surface of the pressure sensors is coated with polytetrafluoroethylene. S12 environmental data collection: At the same time, a temperature sensor, humidity sensor and atmospheric pressure sensor are installed outside the oil chamber to obtain data on the temperature, humidity and atmospheric pressure of the oil chamber's surrounding environment. At the same time, a wind speed sensor is installed at the ventilation position around the oil chamber. The wind speed sensor uses the ultrasonic principle for measurement, that is, it calculates the wind speed by transmitting and receiving ultrasonic signals; S13 operating condition data collection: The operating condition data, including the number of tap position changes, switching intervals, and transformer load current, are acquired from the on-load tap changer control system and the transformer monitoring system. The operating condition data acquisition uses the IEC61850 protocol to acquire the tap changer's real-time status information from the control system. This information includes the tap position code, switching action time, and motor drive current waveform. The transformer load current is acquired through a fiber-optic current transformer. This data is then time-stamped and stored in a time series database. S2 data processing: Based on the pressure, environment, and operating condition data collected in step S1, a sliding average filter algorithm is used to clean the data to remove noise and outliers. The cleaned data is then normalized. Specifically, a sliding average filter with a window size of 5 is used to denoise the pressure data, and the 3σ criterion is used to eliminate abnormal environmental data. At the same time, linear interpolation is performed on the operating condition data to fill missing values. Then, all features are normalized to the [0, 1] interval using the min-max method. Finally, the processed data is stored in the data lake and a dual index of time and space is established. The specific process of the above data processing is as follows; Pressure data cleaning: Pressure data is collected from pressure sensors (MEMS resonant sensors, measuring range -0.1MPa to 0.5MPa, with 2-4 sets installed at each measuring point and coated with PTFE) installed at the top, middle, and bottom of the on-load tap-changer oil compartment. The pressure data is divided into groups of five consecutive data points, and the average value of each group is calculated. For example, the five data points in the first group are added together and divided by 5 to obtain the average value. This average value is used to replace the data point in the middle of the group. This process is repeated sequentially for all pressure data to remove noise. Environmental data cleaning: Environmental data (temperature, humidity, and atmospheric pressure) are obtained from temperature sensors, humidity sensors, and atmospheric pressure sensors installed outside the oil chamber. Wind speed data is also obtained from wind speed sensors located in ventilation locations around the oil chamber (wind speed is calculated by transmitting and receiving ultrasonic signals). For each type of environmental data (such as temperature data), the average of all data points is first calculated. The degree of dispersion (standard deviation) of these data points relative to the average is then calculated. A reasonable numerical range is determined based on the 3σ criterion, for example, the average plus or minus three standard deviations. Each environmental data point is checked. If a data point is not within this reasonable range, it is marked as an abnormal data point and removed from the environmental data set to obtain the cleaned environmental data. Cleaning operating condition data: Obtain operating condition data such as the number of tap position switching times, switching time intervals, and transformer load current from the on-load tap changer control system (which uses the IEC61850 protocol to obtain real-time status information about the tap changer, including tap position coding, switching action time, and motor drive current waveform) and the transformer monitoring system (which uses a fiber-optic current transformer to obtain transformer load current, and stores the data in a time series database after being time-stamped). Check whether there are missing values ​​in the operating condition data. If there are missing values, perform linear interpolation based on the numerical relationship between the valid data before and after the missing value. For example, if the transformer load current data at a certain moment is missing, calculate the reasonable value at the missing moment based on the numerical change trend of the load current before and after the moment, thereby filling the missing value and obtaining the cleaned operating condition data. Data normalization: The cleaned pressure, environment, and operating condition data are collected. These data contain different features. For each feature, find its minimum and maximum values ​​in the data set. For each data point, normalize the feature value in the following way: subtract the minimum value from the feature value, and then divide it by the difference between the maximum and minimum values. For example, if a pressure data feature value is x, the minimum value of the pressure feature is min, and the maximum value is max, then the normalized value is (x - min) / (max - min). Through this process, the values ​​of all features are normalized to the [0,1] interval, and the normalized data is stored in the data lake. In the data lake, a dual index of time and space is established for this data to facilitate subsequent query, analysis, and use; When processing the pressure data, the pressure data is synchronously sent to the time-frequency joint analysis module. After receiving the data, the time-frequency joint analysis module performs wavelet transform and short-time Fourier transform on the pressure data, extracts the characteristic energy ratio of the 0.1-10Hz frequency band, and constructs a fingerprint of the pressure pulsation. When an abnormal increase in the characteristic frequency energy is detected, a preventive maintenance reminder is triggered. The implementation process of the above method is as follows; Pressure data transmission: While processing the pressure data (e.g., cleaning it using a sliding average filter algorithm), pressure data is obtained from the collected pressure data source. This pressure data is collected by pressure sensors installed at the top, middle, and bottom of the on-load tap-changer oil compartment and sent in real time to the time-frequency joint analysis module. The time-frequency joint analysis module processes the pressure data sent from the pressure data processing flow and performs wavelet transform and short-time Fourier transform on the received pressure data. Wavelet transform: Decomposes the pressure data into components of different scales and frequencies using a specific wavelet function. Short-time Fourier transform: Performs a Fourier transform on the pressure data within a time window to obtain information about the pressure data at different times and frequencies. In the results after the wavelet transform and short-time Fourier transform, the module focuses on the information in the 0.1-10Hz frequency band and calculates the characteristic energy ratio within this frequency band. This may involve operations such as statistics and comparisons of the energy of different frequency components within the frequency band to determine the relative relationship between the energy of different frequency components. A fingerprint of the pressure pulsation is constructed based on the extracted characteristic energy ratio of the 0.1-10Hz frequency band. This fingerprint may be a visualization or data structure representation that reflects the characteristic energy distribution of the pressure data within this frequency band. Preventive maintenance reminder triggering: Continuously monitor the characteristic frequency energy in the constructed pressure pulsation fingerprint map to determine whether the characteristic frequency energy shows abnormal growth. This may require comparison with the characteristic frequency energy range or trend under normal conditions. If the current characteristic frequency energy is found to be outside the normal range or showing an abnormal growth trend, it is determined to be abnormal growth. When abnormal growth of characteristic frequency energy is detected, a preventive maintenance reminder is triggered. This reminder can be sent to relevant maintenance personnel or displayed in the monitoring system. Warning information, etc. S3 establishes a pressure prediction model: Extract features from the data processed in step S2, and select a machine learning algorithm to establish an oil chamber pressure prediction model. During the training process, a cross-validation method is used to optimize the model and adjust the model parameters. The feature extraction includes calculating the time domain features and frequency domain features of the pressure data, the gradient change features of the environmental parameters, and the dimensional feature vectors of the tap switching frequency and the load change rate. The feature selection uses a random forest algorithm to evaluate the importance of the features. The machine learning algorithm uses a bidirectional LSTM network, which consists of an input layer, a bidirectional LSTM layer, an attention mechanism layer, and a fully connected output layer. The oil chamber pressure prediction model is internally equipped with an automatic incremental update mechanism every 24 hours. S4 pressure monitoring and fault warning: The data collected in real time in step S1 and pre-processed in step S2 are input into the pressure prediction model trained in step S3 to obtain a predicted value of the oil chamber pressure. The deviation between the actual collected oil chamber pressure value and the predicted value is calculated. When the deviation exceeds a set threshold, it is determined that there is an abnormality in the oil chamber pressure. Based on the size and change trend of the deviation, the type of fault is further analyzed, and a corresponding fault warning signal is simultaneously issued. Fault type analysis is performed using a decision tree approach. A persistently low pressure is identified as a leakage fault, while high-frequency pressure fluctuations indicate an arc fault, and a step-change in pressure indicates a mechanical jam. The fault type analysis is also verified using online oil chromatography monitoring data. In addition, three-dimensional fault tracing technology is added simultaneously during fault analysis. This technology first captures the vibration signal of abnormal pressure in real time by arranging a distributed fiber optic acoustic sensor array in the oil chamber. A direction of arrival estimation algorithm is then used to locate the origin of the abnormal pressure wave. Combined with the pressure sensor data, the three-dimensional pressure gradient field in the oil chamber is reconstructed to visualize the fault diffusion path. The specific implementation process of the above method is as follows: Distributed fiber optic acoustic sensor array placement and vibration signal capture: A distributed fiber optic acoustic sensor array is deployed within the oil chamber. These sensors are distributed at various locations within the chamber to comprehensively monitor conditions within the chamber. When a pressure anomaly is detected during fault analysis (e.g., by comparing the actual pressure value with the predicted pressure value and determining if the deviation exceeds a threshold), the distributed fiber optic acoustic sensor array begins capturing vibration signals within the chamber in real time. These vibration signals can reflect various physical phenomena that may arise within the chamber due to the pressure anomaly, such as vibrations that may be associated with the fault. Abnormal pressure wave origin location: The captured vibration signal is input into a direction of arrival estimation algorithm, which determines the origin of the abnormal pressure wave by analyzing the arrival time and intensity of the vibration signal at different sensor locations. For example, based on the order in which the signals are received by different sensors and the differences in their characteristics, the specific location in the oil chamber where the pressure wave originated can be calculated. Reconstructing the 3D pressure gradient field in the oil chamber: This involves acquiring data from pressure sensors (MEMS resonant sensors with a measurement range of -0.1 MPa to 0.5 MPa, with 2-4 sensors installed at each measurement point and coated with Teflon) previously installed at the top, middle, and bottom of the on-load tap-changer oil chamber. This data is then combined with information about the origin of the abnormal pressure wave, located using a direction of arrival estimation algorithm. Using this data and information, a specific method is employed to reconstruct the 3D pressure gradient field in the oil chamber. This process may involve constructing a 3D model that reflects the pressure distribution and gradient changes within the oil chamber based on the pressure sensor measurements at different locations in the oil chamber and the origin of the abnormal pressure wave. Visualization of the fault diffusion path: Based on the reconstructed 3D pressure gradient field of the oil chamber, corresponding visualization techniques are used to process it. These techniques can visualize the pressure distribution and pressure gradient changes within the oil chamber, clearly showing the possible fault diffusion path. For example, the pressure levels and pressure change trends can be represented in the 3D model through colors and lines, thereby intuitively showing how the fault spreads within the oil chamber.

[0020] This solution comprehensively obtains various information related to the oil chamber pressure by installing temperature sensors, humidity sensors and atmospheric pressure sensors on the outside of the oil chamber and obtaining operating condition data from related systems. When establishing the pressure prediction model, these multi-dimensional data are used as input. The model can learn the complex relationship between different factors, so that in the actual monitoring process, the pressure prediction made by the model is more in line with the actual situation, reducing the phenomenon of false alarms and missed alarms, and improving the accuracy and reliability of the monitoring results.

[0021] This solution collects oil chamber pressure data and related environmental and operating condition data in real time, and continuously performs deviation analysis through the oil chamber pressure prediction model. At each time point, the actual collected oil chamber pressure value is compared with the predicted value obtained by the pressure prediction model and the deviation is calculated. When the deviation exceeds the set threshold, abnormal changes in the oil chamber pressure can be discovered in time. This real-time monitoring and analysis mechanism can capture abnormal signs in the early stages of a fault, providing sufficient time for equipment maintenance and overhaul, reducing the risk of equipment damage, and ensuring the safe and stable operation of the transformer.

[0022] This solution uses a sliding average filtering algorithm to perform data cleaning, which can effectively remove random noise in the data. At the same time, through normalization processing, data of different magnitudes are converted into a unified range, making the data comparable in subsequent processing and analysis. After data preprocessing, a feature extraction operation is performed to extract the features that best reflect the law of oil chamber pressure changes from the processed data, which can provide more valuable input for the pressure prediction model, enabling the oil chamber pressure prediction model to better learn the inherent laws in these data, thereby improving the performance of the oil chamber pressure prediction model, enabling it to more accurately reflect the actual situation when predicting the oil chamber pressure, and improving the prediction accuracy.

[0023] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0024] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for monitoring the oil compartment pressure of an on-load tap-changer for a UHV converter transformer, characterized in that: The following steps are involved: S1 data collection: S11 pressure data acquisition: First, pressure sensors are installed at the top, middle, and bottom of the on-load tap-changer oil compartment, and the oil compartment pressure data is collected in real time through the pressure sensors. S12 environmental data collection: At the same time, a temperature sensor, a humidity sensor and an atmospheric pressure sensor are installed outside the oil chamber to obtain data on the temperature, humidity and atmospheric pressure of the surrounding environment of the oil chamber; S13 operating condition data collection: Obtain operating condition data such as tap position switching times, switching time intervals, and transformer load current from the on-load tap-changer control system and transformer monitoring system; S2 data processing: Based on the pressure, environment and operating condition data collected in step S1, a sliding average filter algorithm is used to clean the data to remove noise and outliers in the data, and then the cleaned data is normalized; S3 establishes a pressure prediction model: Extract features from the data processed in step S2, and select a machine learning algorithm to establish an oil chamber pressure prediction model. During the training process, a cross-validation method is used to optimize the model and adjust the model parameters. S4 pressure monitoring and fault warning: The data collected in real time in step S1 and pre-processed in step S2 is input into the pressure prediction model trained in step S3 to obtain the predicted value of the oil chamber pressure, and the deviation between the actually collected oil chamber pressure value and the predicted value is calculated. When the deviation exceeds the set threshold, it is judged that there is an abnormality in the oil chamber pressure. Based on the size and change trend of the deviation, the type of fault is further analyzed, and a corresponding fault warning signal is issued simultaneously.

2. The method for monitoring oil compartment pressure of an on-load tap-changer for a UHV converter transformer according to claim 1, characterized in that: The pressure sensor in step S11 is a MEMS resonant sensor. The measuring range of the pressure sensor is -0.1 MPa to 0.5 MPa. The number of pressure sensors installed at each measuring point is 2-4 groups. The surface of the pressure sensor is coated with a polytetrafluoroethylene coating.

3. The method for monitoring oil compartment pressure of an on-load tap-changer for a UHV converter transformer according to claim 1, characterized in that: In step S12, in addition to installing the temperature sensor, humidity sensor and atmospheric pressure sensor, a wind speed sensor is also installed at the ventilation position around the oil chamber. The wind speed sensor uses the ultrasonic principle for measurement, that is, it calculates the wind speed by transmitting and receiving ultrasonic signals.

4. The method for monitoring oil compartment pressure of an on-load tap-changer for a UHV converter transformer according to claim 1, characterized in that: The operating condition data acquisition in step S13 acquires real-time status information of the tap changer from the control system via the IEC61850 protocol. The real-time status information of the tap changer includes the tap position code, switching action time, and motor drive current waveform. Simultaneously, the transformer load current is acquired via a fiber optic current transformer. The acquired data is then time-stamped and stored in a time series database.

5. The method for monitoring oil compartment pressure of an on-load tap-changer for a UHV converter transformer according to claim 1, characterized in that: The data processing in step S2 specifically includes: using a sliding average filter with a window size of 5 to denoise the pressure data, and using the 3σ criterion to eliminate abnormal environmental data. At the same time, linear interpolation is performed on the operating condition data to fill missing values. Then, all features are normalized to the [0, 1] interval using the min-max method. Finally, the processed data is stored in the data lake and a dual index of time and space is established.

6. The method for monitoring oil compartment pressure of an on-load tap-changer for a UHV converter transformer according to claim 1, characterized in that: The feature extraction in step S3 includes calculating the time domain features and frequency domain features of the pressure data, the gradient change features of the environmental parameters, and the dimensional feature vectors of the tap switching frequency and the load change rate, and the feature selection uses the random forest algorithm to evaluate the importance.

7. The method for monitoring oil compartment pressure of an on-load tap-changer for a UHV converter transformer according to claim 1, characterized in that: The machine learning algorithm in step S3 adopts a bidirectional LSTM network. The bidirectional LSTM network structure includes an input layer, a bidirectional LSTM layer, an attention mechanism layer and a fully connected output layer. The oil chamber pressure prediction model is internally provided with an automatic incremental update mechanism every 24 hours.

8. The method for monitoring oil compartment pressure of an on-load tap-changer for a UHV converter transformer according to claim 1, characterized in that: The fault type analysis in step S4 is performed using a decision tree method, that is, when the pressure is continuously low, it is determined to be a leakage fault, when the pressure fluctuates at a high frequency, it is determined to be an arc fault, and when the pressure changes in a step, it is determined to be a mechanical jam. At the same time, the fault type analysis is synchronously combined with the oil chromatography online monitoring data to verify the diagnostic results.

9. The method for monitoring oil compartment pressure of an on-load tap-changer for a UHV converter transformer according to claim 1, characterized in that: In step S4, a three-dimensional fault tracing technology is simultaneously added when performing fault analysis. The three-dimensional fault tracing technology first captures the vibration signal when the pressure is abnormal in real time by arranging a distributed fiber optic acoustic wave sensor array in the oil chamber, and then uses a direction of arrival estimation algorithm to locate the origin of the abnormal pressure wave. The three-dimensional pressure gradient field of the oil chamber is reconstructed in combination with the pressure sensor data to visualize the fault diffusion path.

10. The method for monitoring oil compartment pressure of an on-load tap-changer for a UHV converter transformer according to claim 1, characterized in that: When processing the pressure data in step S2, the pressure data is synchronously sent to the time-frequency joint analysis module. After receiving the data, the time-frequency joint analysis module performs wavelet transform and short-time Fourier transform on the pressure data, extracts the characteristic energy ratio of the 0.1-10 Hz frequency band, and constructs a fingerprint of the pressure pulsation. When an abnormal increase in the characteristic frequency energy is detected, a preventive maintenance reminder is triggered.

Images