A method for online monitoring of voltage transformer operating status

Abstract

The present invention relates to the field of data processing, and more specifically, the present invention relates to an online monitoring method for the operating state of a voltage transformer, the method comprising: obtaining historical operating data of a voltage sequence, a current sequence, and a temperature sequence of a voltage transformer and performing preprocessing, obtaining the period of the historical operating data, screening the common period set in the period sequence of each dimension data at the same time, and identifying the most suitable period; constructing an anomaly detection model for the most suitable period; obtaining the period similarity between the real-time operating data and the historical data segments of each most suitable period, obtaining the abnormal result of the anomaly detection model corresponding to each most suitable period, using the similarity of each period as a weight, obtaining the comprehensive abnormal probability, and judging the operating state of the voltage transformer based on the comprehensive abnormal probability. The present invention helps to distinguish natural fluctuations in normal operation from actual abnormal conditions through periodic analysis, thereby reducing false alarms and missed alarms and improving detection accuracy.

Description

Technical Field

[0001] The present invention relates to the field of data processing, and more specifically, to an online monitoring method for the operating status of a voltage transformer. Background Art

[0002] The voltage transformer is a special transformer that is mainly used in power systems to convert high voltage to low voltage to facilitate the operation of measurement and protection equipment. It plays a vital role in power systems, especially in substations and power transmission systems. With the development of technology, online monitoring and intelligent early warning systems based on artificial intelligence technology have emerged. This system can realize intelligent perception, automatic data collection and transmission, and use artificial intelligence technology for fault diagnosis, obtain the health status parameters of high-voltage voltage transformers, and realize intelligent early warning of the working status of transformers. The real-time monitoring and intelligent early warning system of voltage transformers provides a strong guarantee for the safe and stable operation of power transmission systems. By discovering and handling potential faults and abnormal conditions in advance, it reduces power outage time and economic losses, and improves the reliability and economy of power supply.

[0003] The existing Chinese patent application document with publication number CN117826058A discloses an online monitoring method and device for the operating status of a gateway capacitor voltage transformer. The application document includes: obtaining the voltage, current, power factor, temperature and humidity of the gateway capacitor voltage transformer to be tested; inputting the voltage, current, power factor, temperature and humidity into the gateway capacitor voltage transformer prediction model to obtain the ratio difference and angle difference of the gateway capacitor voltage transformer to be tested; according to the ratio difference and angle difference, determining the score of the gateway capacitor voltage transformer to be tested in multiple quality dimensions, the quality dimensions include error state, batch quality, monitoring abnormality, measurement reputation and operating years; according to the score, determining the inspection cycle of the gateway capacitor voltage transformer to be tested.

[0004] The above application can solve the problem of untimely grasp of the operating status of the gateway capacitor voltage transformer, and realize early warning of the operating status of the gateway capacitor voltage transformer to be tested. At present, the operating status of the voltage transformer is affected by multiple factors such as voltage, current and temperature, each of which may affect the performance and status of the equipment to varying degrees, and the operating data information is complex, thus affecting the voltage transformer to produce operating failures. At the same time, as the equipment ages and wears, the performance will gradually decline. This gradual change may be confused with the fault signal, and it is impossible to accurately predict the operating status results. Summary of the invention

[0005] In order to solve the problem that the operating state of the voltage transformer is affected by multiple factors such as voltage, current and temperature, and at the same time, as the equipment ages and wears, the performance will gradually decline, and the operating state results cannot be accurately predicted, the present invention provides solutions in the following aspects.

[0006] A method for online monitoring of the operating state of a voltage transformer comprises: obtaining historical operating data of the voltage transformer, preprocessing the historical operating data, wherein the historical operating data comprises: a voltage sequence, a current sequence and a temperature sequence; obtaining the period of the historical operating data, screening the period set common to the period sequence of each dimension data at the same time, and calculating the most suitable period in the period set of each dimension data at the same time; constructing a preset network model for the most suitable period, taking the data of the same most suitable period as a training set and training, and obtaining an abnormality detection model for each most suitable period; obtaining the period similarity between the real-time operating data and the historical data segments of each most suitable period, obtaining the abnormal result of the abnormality detection model corresponding to each most suitable period, taking the similarity of each period as a weight, obtaining the comprehensive abnormality probability, and judging the operating state of the voltage transformer based on the comprehensive abnormality probability; the comprehensive abnormality probability satisfies the following relationship: , where represents the comprehensive abnormal probability, represents the number of best-fit cycles, Indicates The similarity of the cycles, Indicates The anomaly probability output by the anomaly detection model corresponding to the period.

[0007] The effect is: by comprehensively analyzing the historical operating data of multiple dimensions such as voltage, current and temperature, this method can more comprehensively capture the operating status of the voltage transformer. By identifying the most suitable cycle and calculating the comprehensive abnormality probability, this method helps to formulate a more reasonable maintenance and overhaul plan, reduce unnecessary maintenance work, and reduce maintenance costs.

[0008] Preferably, the most suitable period includes:

[0009] Use the partial autocorrelation function to obtain the periodic sequence of each dimension data and obtain the The common period of the periodic sequence of each dimension data at time is taken as the periodic set;

[0010] Divide each dimension data sequence into data segments according to the period set, and include the The dimensional data at each moment is taken as the target data segment, the average value of the sum of similarities between the target data segment and other data segments is taken as the degree of fit, and the period corresponding to the maximum degree of fit is taken as the most suitable period.

[0011] The effect is: by screening out the common cycle sets in multi-dimensional data sequences, the key cycles of the voltage transformer's operating status can be identified more accurately. By determining the most suitable cycle, resources and attention can be concentrated on those cycles that are most likely to show abnormalities, optimizing the allocation of monitoring resources. By analyzing the similarity of periodic data segments, the future operating status of the voltage transformer can be predicted, potential problems can be discovered in advance, predictive maintenance can be achieved, normal fluctuations and abnormal conditions can be accurately identified, false alarms caused by misjudging normal fluctuations as abnormalities can be reduced, and the reliability of the monitoring system can be improved.

[0012] Preferably, the degree of fit also includes:

[0013] Calculate the target data segment and each other data segment Pearson correlation coefficient between the target data segment With other data segments The average of the Pearson correlation coefficients between the two groups was taken as the goodness of fit.

[0014] The effect is that the average value of the Pearson correlation coefficient is used as the fit, which can more accurately match the historical period similar to the target data segment.

[0015] Preferably, the degree of fit also includes:

[0016] The target data segment With other data segments The average value of the sum of similarities between the target data segment With other data segments The average of the Pearson correlation coefficients between them was taken as the linear similarity;

[0017] The weights of the distance similarity and the linear similarity are preset respectively, and the weighted sum is performed to obtain the degree of fit.

[0018] The effect is that by combining multiple similarity metrics, the deviation that may be caused by a single metric can be reduced, the robustness of the fit calculation can be improved, and the results can be more reliable. By presetting different weights, the importance of distance similarity and linear similarity can be adjusted according to the actual application scenarios and data characteristics, so that the fit calculation can better meet the requirements of the voltage transformer operating status.

[0019] Preferably, the period similarity includes:

[0020] Run data in real time and all historical data segments that best fit the cycle Middle The sum of the products of the data points between the data segments and the real-time operation data and all historical data segments that best fit the cycle Middle The ratio of the modulus lengths of the data segments is taken as the periodic similarity.

[0021] The effect is that by comparing the similarity between real-time data and historical data, real-time data segments that are similar to historical cycles can be more accurately identified, thereby improving the accuracy of anomaly detection.

[0022] Preferably, the period similarity further includes:

[0023] Calculate real-time operation data separately and all historical data segments that best fit the cycle The sum of squared differences between the peak and valley values ​​gives the cycle similarity.

[0024] The effect is that the difference in periodicity between real-time data and historical data can be captured more accurately through the difference between peak and valley values.

[0025] Preferably, the structure of the preset network model is a bp neural network model structure, wherein the network structure of the bp neural network model includes: input layer: input real-time voltage transformer operation data; hidden layer; output layer: probability of abnormal operation of the voltage transformer.

[0026] Preferably, the training process of the anomaly detection model is:

[0027] Set labels for each operation data that best fits the cycle, build an abnormal operation data set, and use 70% of the abnormal operation data set as a training set and 30% as a validation set. Input the data of the training set into the preset network model to obtain the probability of the corresponding abnormality, and verify it based on the validation set.

[0028] The cross entropy loss function is used to calculate the difference between the network prediction output and the actual target, and the network parameters of the model are iteratively updated until the loss value is less than the set loss value, the update is stopped, or the preset number of iterations is reached to obtain the anomaly detection model.

[0029] Preferably, judging the operating state of the voltage transformer based on the comprehensive abnormal probability includes:

[0030] In response to the comprehensive abnormality probability being greater than a preset threshold, there is an abnormality in the operating state of the voltage transformer, otherwise the operating state is normal.

[0031] The present invention has the following effects:

[0032] 1. The present invention helps to distinguish natural fluctuations in normal operation from actual abnormal conditions through periodic analysis, thereby reducing false alarms and missed alarms. By analyzing the current operating cycle data of the voltage transformer with the corresponding data cycle in history, the performance change trend of the equipment can be predicted, so that maintenance can be performed before a fault occurs, reducing the need for emergency repairs.

[0033] 2. The present invention helps to capture the operating status of the voltage transformer by simultaneously considering data of multiple dimensions such as voltage, current and temperature, thereby improving the accuracy of fault prediction. At the same time, the operating data analysis is obtained in real time with the corresponding data cycle in history. Through the corresponding prediction model, the comprehensive abnormality probability is calculated, potential abnormalities can be discovered in time, early warning of faults can be achieved, unnecessary frequent inspections of equipment can be reduced, and operating costs can be reduced. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] By reading the following detailed description with reference to the accompanying drawings, the above and other objects, features and advantages of the exemplary embodiments of the present invention will become readily understood. In the accompanying drawings, several embodiments of the present invention are shown in an exemplary and non-limiting manner, and the same or corresponding reference numerals represent the same or corresponding parts, wherein:

[0035] Figure 1 It is a method flow chart of steps S1 to S4 in a method for online monitoring of the operating status of a voltage transformer according to an embodiment of the present invention. DETAILED DESCRIPTION

[0036] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are 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 those skilled in the art without creative work are within the scope of protection of the present invention.

[0037] The specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0038] Reference Figure 1 A method for online monitoring of the operating status of a voltage transformer includes steps S1 to S4, which are specifically as follows:

[0039] S1: Acquire historical operation data of the voltage transformer and pre-process the historical operation data, wherein the historical operation data includes: a voltage sequence, a current sequence and a temperature sequence.

[0040] Exemplarily, the historical operation data includes but is not limited to: voltage sequence, current sequence and temperature sequence, wherein the historical operation data is acquired according to the time series, ensuring that the data points of each acquired sequence correspond in the time series. Further explanation, it is necessary to check the missing values ​​in the data set of the historical operation data and use the interpolation method to supplement them to ensure the integrity of the data, wherein the interpolation method is an existing well-known technology and will not be described in detail.

[0041] S2: Obtain the cycle of historical operation data, filter out the common cycle set in the cycle sequence of each dimension data at the same time, and calculate the most suitable cycle in the cycle set of each dimension data at the same time.

[0042] The most suitable cycle includes:

[0043] Use the partial autocorrelation function to obtain the periodic sequence of each dimension data and obtain the The common period of the periodic sequence of each dimension data at time is taken as the periodic set;

[0044] Divide each dimension data sequence into data segments according to the period set, and include the The dimensional data at each moment is taken as the target data segment, the average value of the sum of similarities between the target data segment and other data segments is taken as the degree of fit, and the period corresponding to the maximum degree of fit is taken as the most suitable period.

[0045] It should be noted that the partial autocorrelation function (PACF) is a well-known technology and will not be described in detail.

[0046] Specifically, the fit satisfies the following relationship:

[0047] ;

[0048] In the formula, The cycle is The degree of fit, represents the number of adjacent data segments in each dimension, represents the Euclidean distance, Indicates the target data segment, Indicates the other data segments data segment.

[0049] It should be noted that if A smaller value indicates that the target data segment and other data segments are in the same period. This indicates that the data has a high degree of periodic consistency in this cycle, thereby identifying the periodic features that best represent the data, and obtaining what factors affect the features in this cycle in the historical operating data, so as to facilitate the formulation of maintenance plans.

[0050] It should also be explained that, in this embodiment, the Euclidean distance measures the straight-line distance between two data points, and focuses on the absolute difference of the data. The Euclidean distance is affected by the scale of the data. If the magnitude of the data is different, even if the trend is the same, the Euclidean distance may be large. Therefore, two other embodiments are provided. The specific steps are as follows:

[0051] In addition, in the second embodiment, the target data segment can also be calculated and each other data segment Pearson correlation coefficient between the target data segment With other data segments The average of the Pearson correlation coefficients between the two groups was taken as the goodness of fit.

[0052] Specifically, the target data segment and other data segments The Pearson correlation coefficient between them satisfies the following relationship:

[0053] ;

[0054] In the formula, Indicates the target data segment and Data segment The Pearson correlation coefficient, Indicates the target data segment Middle data points, represents the mean of the target data, Indicates Data segment Middle data points, Indicates Data segment The mean of Represents the total number of data points in the segment.

[0055] Specifically, the fit satisfies the following relationship:

[0056] ;

[0057] In the formula, The cycle is The degree of fit, represents the number of adjacent data segments in each dimension, Indicates the target data segment and Data segment The Pearson correlation coefficient.

[0058] That is, in this embodiment, the Pearson correlation coefficient measures the linear correlation between two data sets, that is, whether they are related to each other in a straight line. In other words, the Pearson correlation coefficient is a dimensionless measure that is not affected by the scale of the data, that is, it focuses on the relative trend of data changes rather than the absolute difference.

[0059] Both of the above two embodiments can analyze the periodic fit, and the considerations are different, and the specific selection can be made according to the implementation situation.

[0060] In the third embodiment, the degree of fit also includes:

[0061] The target data segment With other data segments The average value of the sum of similarities between the target data segment With other data segments The average of the Pearson correlation coefficients between them was taken as the linear similarity;

[0062] The weights of the distance similarity and the linear similarity are preset respectively, and the weighted sum is performed to obtain the degree of fit.

[0063] Specifically, the fit satisfies the following relationship:

[0064] ;

[0065] In the formula, The cycle is The degree of fit, , Represent the weights of Euclidean distance and Pearson correlation coefficient respectively, represents the number of adjacent data segments in each dimension, represents the Euclidean distance, Indicates the target data segment, Indicates the other data segments data segments, Indicates the target data segment and Data segment The Pearson correlation coefficient.

[0066] For example, , , which can be adjusted according to specific circumstances.

[0067] To further illustrate, combining the Pearson correlation coefficient and the Euclidean distance can analyze the similarity of data from different perspectives. The Pearson correlation coefficient focuses on the linear relationship between the data, while the Euclidean distance focuses on the absolute difference between the data points. This multi-angle analysis can provide a more comprehensive similarity measure. For example, if there are outliers in the data, the Pearson correlation coefficient may not be affected much, while the Euclidean distance may be greatly affected. By weighted averaging, the impact of outliers on the results can be reduced to a certain extent. The combined use of the two metrics can reduce the limitations of a single metric and improve the reliability of the analysis results.

[0068] S3: Build a preset network model for the most suitable period, use the data of the same most suitable period as a training set and perform training to obtain an anomaly detection model for each most suitable period.

[0069] The preset network model structure is a bp neural network model structure, wherein the network structure of the bp neural network model includes: input layer: input real-time voltage transformer operation data; hidden layer; output layer: probability of abnormal operation of the voltage transformer.

[0070] The training process of the anomaly detection model is:

[0071] Set labels for each operation data that best fits the cycle, where the labels are normal and abnormal, and construct an abnormal operation data set. Use 70% of the abnormal operation data set as a training set and 30% as a validation set. Input the data of the training set into the preset network model to obtain the probability of the corresponding abnormality, and verify it based on the validation set.

[0072] The cross entropy loss function is used to calculate the difference between the network prediction output and the actual target, and the network parameters of the model are iteratively updated until the loss value is less than the set loss value, the update is stopped, or the preset number of iterations is reached to obtain the anomaly detection model.

[0073] S4: Obtain the cycle similarity between the real-time operation data and the historical data segments of each most suitable cycle, obtain the abnormal results of the abnormality detection model corresponding to each most suitable cycle, use the similarity of each cycle as a weight, obtain the comprehensive abnormal probability, and judge the operating status of the voltage transformer based on the comprehensive abnormal probability.

[0074] Run data in real time and all historical data segments that best fit the cycle Middle The sum of the products of the data points between the data segments and the real-time operation data and all historical data segments that best fit the cycle Middle The ratio of the modulus lengths of the data segments is taken as the periodic similarity.

[0075] Further explanation, the data will be run in real time According to the historical data segment that best fits the cycle The length of the corresponding period is divided into two periods, which makes it easier to compare real-time data with historical data on the same time scale. Time alignment ensures that anomalies are detected under the same conditions, improving the accuracy of anomaly detection. At the same time, it is helpful to predict future data changes.

[0076] Specifically, the period similarity satisfies the following relationship:

[0077] ;

[0078] In the formula, Indicates The similarity of the cycles, Indicates real-time operation data Middle data points, Represents all historical data segments that best fit the cycle Middle The first data segment data points, Represents the total number of data points in the segment.

[0079] In addition, in another embodiment, real-time operation data is calculated respectively and all historical data segments that best fit the cycle The sum of squared differences between the peak and valley values ​​gives the cycle similarity.

[0080] Specifically, the period similarity satisfies the following relationship:

[0081] ;

[0082] In the formula, Indicates real-time operation data The historical data segment that best fits the cycle The cycle similarity between represents the total number of peaks and valleys, Indicates real-time operation data Middle Peak value, Indicates the historical data segment that best fits the cycle Middle Peak value, Indicates real-time operation data Middle A valley value, Indicates the historical data segment that best fits the cycle Middle A valley value.

[0083] That is to say, Indicates the total amount of difference between real-time data and historical data in terms of peaks and valleys. It directly reflects the degree of deviation between the two data sets at these key points. A larger period similarity value means that the real-time data has a larger deviation from the historical data, which may indicate anomalies or changes.

[0084] Specifically, the comprehensive abnormal probability satisfies the following relationship:

[0085] ;

[0086] In the formula, represents the comprehensive abnormal probability, represents the number of best-fit cycles, Indicates The similarity of the cycles, Indicates The anomaly probability output by the anomaly detection model corresponding to the period.

[0087] In response to the comprehensive abnormality probability being greater than a preset threshold, there is an abnormality in the operating state of the voltage transformer, otherwise the operating state is normal.

[0088] Exemplarily, the preset threshold is 0.8, which can be adjusted according to actual conditions.

[0089] In the description of this specification, "plurality" or "several" means at least two, such as two, three or more, etc., unless otherwise clearly and specifically defined.

[0090] Although this specification has shown and described a number of embodiments of the present invention, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Those skilled in the art will conceive of many modifications, changes and alternatives without departing from the ideas and spirit of the present invention. It should be understood that in the practice of the present invention, various alternatives to the embodiments of the present invention described herein may be employed.

Claims

1. A method for online monitoring of the operating status of a voltage transformer, characterized in that: include: Acquire historical operation data of the voltage transformer and preprocess the historical operation data, wherein the historical operation data includes: a voltage sequence, a current sequence and a temperature sequence; Get the cycle of historical operation data, filter the common cycle set in the cycle sequence of each dimension data at the same time, and calculate the most suitable cycle in the cycle set of each dimension data at the same time. The most suitable cycle includes: using the partial autocorrelation function to obtain the cycle sequence of each dimension data, and obtaining the first The common period of the periodic sequence of each dimension data at time is taken as the periodic set; the data segment of each dimension data sequence is divided according to the periodic set, and the data segment containing the first The dimension data at the moment is taken as the target data segment, the average value of the sum of similarities between the target data segment and other data segments is taken as the degree of fit, and the period corresponding to the maximum degree of fit is taken as the most suitable period; Constructing a preset network model for the most suitable period, using the data of the same most suitable period as a training set and performing training to obtain an anomaly detection model for each most suitable period; Obtain the cycle similarity between the real-time operation data and the historical data segments of each most suitable cycle, obtain the abnormal results of the abnormality detection model corresponding to each most suitable cycle, use the similarity of each cycle as a weight, obtain the comprehensive abnormal probability, and judge the operation status of the voltage transformer based on the comprehensive abnormal probability; The comprehensive abnormal probability satisfies the following relationship: , where represents the comprehensive abnormal probability, represents the number of best-fit cycles, Indicates The similarity of the cycles, Indicates The anomaly probability output by the anomaly detection model corresponding to the period.

2. A method for online monitoring of voltage transformer operating status according to claim 1, characterized in that: The period similarity includes: Run data in real time and all historical data segments that best fit the cycle Middle The sum of the products of the data points between the data segments and the real-time operation data and all historical data segments that best fit the cycle Middle The ratio of the modulus lengths of the data segments is taken as the periodic similarity.

3. The method for online monitoring of the operating status of a voltage transformer according to claim 1, characterized in that: The period similarity also includes: Calculate real-time operation data separately and all historical data segments that best fit the cycle The sum of squared differences between the peak and valley values ​​gives the cycle similarity.

4. A method for online monitoring of voltage transformer operating status according to claim 1, characterized in that: The structure of the preset network model is a bp neural network model structure, wherein the network structure of the bp neural network model includes: an input layer: inputting real-time voltage transformer operation data; a hidden layer; and an output layer: the probability of abnormal operation of the voltage transformer.

5. The method for online monitoring of the operating status of a voltage transformer according to claim 1, characterized in that: The training process of the anomaly detection model is: Set labels for each operation data that best fits the cycle, build an abnormal operation data set, and use 70% of the abnormal operation data set as a training set and 30% as a validation set. Input the data of the training set into the preset network model to obtain the probability of the corresponding abnormality, and verify it based on the validation set. The cross entropy loss function is used to calculate the difference between the network prediction output and the actual target, and the network parameters of the model are iteratively updated until the loss value is less than the set loss value, the update is stopped, or the preset number of iterations is reached to obtain the anomaly detection model.

6. A method for online monitoring of voltage transformer operating status according to claim 1, characterized in that: The method of judging the operating state of the voltage transformer based on the comprehensive abnormal probability includes: In response to the comprehensive abnormality probability being greater than a preset threshold, there is an abnormality in the operating state of the voltage transformer, otherwise the operating state is normal.

Images