A real-time condition assessment method and system for voltage transformers

Abstract

The application provides a real-time state evaluation method and system for a voltage transformer, and relates to the field of measurement and monitoring of a power system. History operation data of a secondary side of the voltage transformer is collected, and data preprocessing is performed; a KNN-ELM model is used, an evaluation threshold is preset, current operation data is input into the optimal KNN-ELM model, a prediction value of the current three-phase voltage transformer secondary output is obtained, and current evaluation indexes are calculated for comparison and judgment; the health state of the voltage transformer equipment is evaluated, an evaluation threshold self-updating strategy of an online model is established, and it is further judged whether a consistency drift fault occurs. The embodiment ensures the accuracy and reliability of the collected history data and current operation data, so as to avoid deviation of model training and prediction, effectively performs real-time state evaluation on the voltage transformer, discovers and processes abnormal states of equipment in time, and improves the safety and reliability of the power system.

Description

Technical Field

[0001] This invention relates to the field of power system measurement and monitoring, and in particular to a method and system for real-time status assessment of voltage transformers. Background Technology

[0002] With the development of smart grids and the continuous expansion of power system scale, voltage transformers are playing an increasingly important role in ensuring the safe operation of power systems and the accuracy of electricity metering. Currently, voltage transformer error detection mainly relies on periodic offline verification. This method is not only time-consuming and labor-intensive, but also struggles to capture error changes in transformers during operation in a timely manner. Furthermore, offline detection methods are difficult to implement centralized monitoring of large-scale transformers, especially in high-voltage power grids where the complexity of the testing equipment and the harsh operating environment make offline verification insufficient to meet the real-time monitoring needs of the current power grid. Therefore, there is an urgent need for an online monitoring method that can assess the status of voltage transformers in real time, thereby identifying operational anomalies and improving equipment stability. Summary of the Invention

[0003] The main objective of this invention is to provide a real-time status assessment method and system for voltage transformers, thereby solving the problem of difficulty in timely detection of transformer malfunctions in the prior art.

[0004] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: a real-time status assessment method for voltage transformers, comprising the following steps:

[0005] S1: Collect historical operating data of voltage transformers, perform data preprocessing, and save it as the raw dataset;

[0006] The historical operating data is data that characterizes the standard operating state of the transformer, including the secondary output of the three-phase voltage transformer under normal equipment conditions.

[0007] S2: Build a KNN-ELM model and train it using the original dataset, then save the optimal KNN-ELM model;

[0008] S3: Input the current operating data of the voltage transformer into the optimal KNN-ELM model to obtain the predicted value of the current secondary output of the three-phase voltage transformer, and calculate the current evaluation index.

[0009] S4: If the current evaluation index is greater than the evaluation threshold, the current voltage transformer is in an abnormal state, and a system warning is triggered and recorded; the system warning needs to be classified into alarm triggering levels.

[0010] In the preferred scheme, if the current evaluation index is not greater than the evaluation threshold, it is considered a normal state, and the evaluation threshold is updated, specifically as follows:

[0011] Continue to collect online operational data in real time, calculate and dynamically adjust evaluation thresholds using a weighted sliding time window, and determine whether a consistency drift fault has occurred based on the defined alarm trigger levels, including:

[0012] The evaluation threshold is calculated in real time using a weighted sliding time window, and the formula is as follows:

[0013]

[0014] In the formula, t represents the current time; Q q (t+1) is the statistical threshold for the next time step; Q' is the confidence quantile within the sliding window of the previous time step; Q q α is the initial threshold; α is the adjusted weight; Q(t) is the evaluation statistic at the current time; P is the cumulative sample size.

[0015] In the preferred scheme, in step S3, determining whether a consistency drift fault has occurred is achieved by monitoring the rate of change of the evaluation index change ratio to help determine whether the current transformer group has deteriorated simultaneously. The formula is as follows:

[0016]

[0017] In the formula, R mean,i T represents the relative change in the mean of the i-th window compared to the (i-1)-th window; i and T i-1 These are the dynamic mean values ​​of the i-th and (i-1)-th windows, respectively, and n is the total number of windows;

[0018]

[0019] In the formula, ΔR mean,i ξ represents the rate of change of the ratio of the changes in the i-th window, max() takes the maximum of the two, ξ is a minimum positive number, and n is the total number of windows;

[0020]

[0021] In the formula, represents the relative change in standard deviation of the i-th window compared to the (i-1)-th window; S i and S i-1 are the dynamic standard deviations of the i-th and (i-1)-th windows, respectively, and n is the total number of windows;

[0022]

[0023] In the formula, ΔR std,i R represents the rate of change of the i-th window relative to the (i-1)-th window. std,i With R std,i-1 The rate of change, where n is the total number of windows.

[0024] In the preferred embodiment, the data preprocessing in step S1 includes:

[0025] A1: Determine if the collected data is missing, and mark the missing values ​​as NaN;

[0026] A2: Data is removed using the 3σ principle, and the removed data is marked as NaN. The formula is:

[0027]

[0028] In the formula, xi is the i-th value in the dataset, and N is the total number of data points;

[0029] A3: Interpolate the labeled NaN using PCHIP, the formula is as follows:

[0030]

[0031] In the formula, z0 and z1 are the known time points before and after the point to be interpolated, and f0 and f1 are the known data values ​​corresponding to these time points; f 0′ and f 1′ This is the corresponding first derivative; the interpolated data is saved to the original dataset.

[0032] A4: Organize the output data of the three-phase voltage transformers in the same group of the interpolated original dataset into a matrix form K to obtain the corresponding input vector:

[0033]

[0034] In the formula, w x (t), w y (t) and w z (t) represents the output of the three-phase voltage transformer at time t; M is the sequence length; next, matrix K is multiplied by the sparse matrix Q. (3×6) Z, as the input to the model:

[0035]

[0036] Calculations show that,

[0037] [Z x |Z y |Z z ] = [W y (t)W z (t)|W x (t)W x (t)|W z (t)W y (t)];

[0038] Further simplification:

[0039] [X x |X y |X z ] = K = [W x (t)|W y (t)|W z (t)];

[0040] In the formula, Z x Z y and Z z These are the input feature matrices of the three mutual inductors; Q converts K into the input form of the prediction model, and W... x (t), W y (t) and W z (t) is the corresponding input vector.

[0041] In the preferred scheme, step S2 involves building a KNN-ELM model that combines KNN and ELM models, where:

[0042] The KNN model calculates the Euclidean distance between samples in the feature space of the input vector, identifies the nearest samples in historical data that are most similar to the features of the current input data, sorts them in order, and constructs the corresponding feature vectors.

[0043] The ELM model is used to perform regression prediction on the feature vector to obtain the predicted value of the secondary output voltage of the three-phase voltage transformer at each time step.

[0044] The ELM model introduced above incorporates a penalty factor, as shown in the following formula:

[0045]

[0046] In the formula: β* is the least squares solution of weight β; I is the diagonal identity matrix; C is the regularization coefficient; h is the hidden node output of ELM; h T Let h be the transpose of h.

[0047] In a preferred embodiment, the calculation of the evaluation threshold includes the following steps:

[0048] During model training, the output of the KNN-ELM model is mapped one-to-one with the original dataset after interpolation, and a subtraction operation is performed to obtain the prediction residuals. The squared prediction residuals are selected as the evaluation metric for the offline model, and the formula is as follows:

[0049]

[0050] In the formula, e i For three-phase voltage transformers at t i Predict the residual vector at each time step; For e iThe transpose of Q; i This represents the corresponding squared prediction residual, i.e., the evaluation index;

[0051] Based on the obtained predicted residuals, the distribution of the residual evaluation index is estimated using the Gamma probability model, and the control chart method is used to select an appropriate high quantile as the evaluation threshold for the evaluation index.

[0052] In the preferred embodiment, in step S4, the system early warning is divided into different alarm trigger levels based on the frequency of deviation, including two states: warning and abnormal. The process is as follows:

[0053] 1) If the current online running data Q(t) is greater than the adaptively updated judgment threshold Q q If (t), then marking begins;

[0054] 2) Continue calculating the evaluation metric Q(t) for the online running data in the next time interval, and Q at the time when updates stop. E Comparison, if continuous n E Each time interval is greater than Q. E If so, it will be marked as an alert and logged.

[0055] 3) After being marked as an alert state, if there are n consecutive... E A time interval, or 3*n consecutive intervals E There are at least 2*n time intervals. E If all the criteria for judging the number of indicators are greater than the judgment threshold, it is marked as an abnormal state, triggering a fault alarm and alerting the operation and maintenance personnel.

[0056] Where, n E The value is set as follows: based on the maximum number of consecutive missing values ​​and consecutive values ​​greater than three times the standard deviation in the historical measurement data collected in step S1;

[0057] The logs include timestamps, judgment thresholds, evaluation metrics, alarm status (warning or abnormal), and duration.

[0058] A real-time condition assessment system for voltage transformers, comprising:

[0059] The data acquisition module is used to collect historical operating data of the voltage transformer, perform data preprocessing, and save it as a raw dataset; the historical operating data is data that characterizes the standard operating state of the transformer, including the secondary output of the three-phase voltage transformer under normal equipment conditions;

[0060] The model training module is used to build the KNN-ELM model, train it using the original dataset, and save the optimal KNN-ELM model.

[0061] The model evaluation module is used to input the current operating data of the voltage transformer into the optimal KNN-ELM model, obtain the predicted value of the current secondary output of the three-phase voltage transformer, and calculate the current evaluation index.

[0062] The early warning trigger module is used to determine if the current voltage transformer is in an abnormal state if the current evaluation index is greater than the evaluation threshold, and to trigger and record the system early warning; the system early warning needs to be divided into alarm trigger levels.

[0063] This invention provides a real-time status assessment method for voltage transformers. It constructs a virtual standard based on historical transformer data, and assesses the voltage transformer's status by comparing the virtual standard with the current operating data. Furthermore, it constructs a regression model by fusing a KNN-ELM model, and uses the model's output to build evaluation indicators as benchmark standards. The technical advantages of this invention are as follows:

[0064] 1) This invention, based on historical data from the secondary side output of three-phase voltage transformers, establishes a virtual standard using a KNN-ELM model and evaluation metrics by leveraging the output coordination correlation among the three-phase voltage transformers. This method transforms the performance degradation of three-phase voltage transformers into an anomaly detection problem. It compares the real-time collected voltage transformer output data with the established virtual standard metrics to determine whether the voltage transformer is in an abnormal state at the current moment. Compared to traditional offline verification methods, this invention extracts statistics from existing voltage transformer output data and analyzes their changing patterns to determine whether the transformer is currently operating abnormally, reducing maintenance costs and lowering system operational risks.

[0065] 2) The present invention adopts a sliding time window update method to adaptively adjust the judgment threshold, which can reduce the negative impact of imbalance fluctuations in the power grid and improve the sensitivity of anomaly detection. At the same time, this adaptive update mechanism ensures the stability and accuracy of monitoring, enabling relatively accurate error detection even in complex power grid environments.

[0066] 3) This invention provides an alarm triggering and discrimination process, which classifies alarm levels based on the frequency of deviations, prompts maintenance personnel whether further inspection is needed, and records relevant data in real time to form a monitoring log.

[0067] 4) This invention uses the rate of change of the standard deviation and the mean threshold to help determine whether the current transformer has a consistency fault.

[0068] 5) This invention performs anomaly detection on operating voltage transformers online, which can detect early anomalies in a timely manner, reduce reliance on manual offline verification, and thus significantly reduce maintenance frequency and cost.

[0069] 6) This invention uses the historical operating data of the current transformer itself as a benchmark, without relying on complex hardware equipment. Therefore, the method described in this invention is applicable to voltage transformers of different voltage types, exhibiting high versatility. This characteristic supports the scalability of the system, enabling it to be applied to various power grid monitoring scenarios, further enhancing the practical value of this technology. Attached Figure Description

[0070] The present invention will be further described below with reference to the accompanying drawings and embodiments:

[0071] Figure 1 This is a schematic diagram of the model process framework of the present invention;

[0072] Figure 2 This is a flowchart of the state assessment method of the present invention;

[0073] Figure 3 This is a comparison diagram showing the output voltage of the present invention before and after PCHIP interpolation;

[0074] Figure 4 This is a diagram illustrating the online monitoring effect of the present invention under adaptive threshold;

[0075] Figure 5 This is a diagram demonstrating the online monitoring effect of the present invention under performance degradation;

[0076] Figure 6 This invention employs Figure 5 The corresponding auxiliary index diagram of the current transformer. Detailed Implementation

[0077] Example 1

[0078] like Figure 1-6 As shown, a real-time status assessment method for voltage transformers includes the following steps:

[0079] S1: Collect historical operating data of voltage transformers, perform data preprocessing, and save it as the raw dataset;

[0080] Historical operating data represents the standard operating status of the instrument transformer, including the secondary output of the three-phase voltage transformer under normal operating conditions.

[0081] S2: Build a KNN-ELM model and train it using the original dataset, then save the optimal KNN-ELM model.

[0082] S3: Input the current operating data of the voltage transformer into the optimal KNN-ELM model to obtain the predicted value of the secondary output of the current three-phase voltage transformer, and calculate the current evaluation index.

[0083] S4: If the current evaluation index is greater than the evaluation threshold, the current voltage transformer is in an abnormal state, and the system will issue an early warning and record it; the system early warning needs to classify the alarm triggering level.

[0084] If the current evaluation index is not greater than the evaluation threshold, it is considered a normal state. The evaluation threshold is then updated, specifically by continuing to collect online operational data in real time, calculating the evaluation threshold in real time using a weighted sliding time window, and dynamically adjusting it. Based on the classified alarm trigger levels, it is determined whether a consistency drift fault has occurred, including:

[0085] The evaluation threshold is calculated in real time using a weighted sliding time window, and the formula is as follows:

[0086]

[0087] In the formula, t represents the current time; Q q (t+1) is the statistical threshold for the next time step; Q' is the confidence quantile within the sliding window of the previous time step; Q q α is the initial threshold; α is the adjusted weight; Q(t) is the evaluation statistic at the current time; P is the cumulative sample size.

[0088] like Figure 1-2 As shown, this embodiment collects historical operating data of voltage transformers, preprocesses the data to train a model, and saves the optimal KNN-ELM model. A preset evaluation threshold is established, and the current operating data is input into the optimal KNN-ELM model to obtain the predicted value of the secondary output of the three-phase voltage transformer. The current evaluation indicators are calculated and compared to assess the health status of the voltage transformer equipment. An online model evaluation threshold self-updating strategy is established, dynamically adjusting the evaluation threshold in step two as real-time data is continuously accumulated. Alarm triggering levels are classified, and it is further determined whether a consistency drift fault has occurred. This embodiment ensures the accuracy and reliability of the collected historical and current operating data to avoid deviations in model training and prediction, effectively assesses the real-time status of the voltage transformer, promptly detects and handles abnormal equipment conditions, and improves the safety and reliability of the power system.

[0089] In step S1, historical operating data of the voltage transformer is collected. The historical data collection category is the secondary output of the three-phase voltage transformer under normal equipment conditions, so as to characterize the standard operating state of the transformer.

[0090] In the preferred embodiment, the data preprocessing in step S1 includes:

[0091] A1: Determine if the collected data is missing and mark the missing values ​​as NaN.

[0092] A2: Noise is unavoidable during actual data collection. Therefore, the mean μ and standard deviation σ of the collected data are calculated, and the 3σ principle is used to remove data. The removed data is marked as NaN. The formula is:

[0093]

[0094] In the formula, xi is the i-th value in the dataset, and N is the total number of data points.

[0095] Data falling outside the range [μ-3σ, μ+3σ] are considered noise outliers and deleted, then labeled as NaN. A3: The labeled NaNs are interpolated using PCHIP, with the following formula:

[0096]

[0097] In the formula, z0 and z1 are the known time points before and after the point to be interpolated, and f0 and f1 are the known data values ​​corresponding to these time points; f 0′ and f 1′ It is the corresponding first derivative, and the interpolated data is saved to the original dataset.

[0098] like Figure 3 The figure shown is a comparison of the output voltage before and after PCHIP interpolation. PCHIP constructs a piecewise cubic polynomial over each interval, ensuring that the interpolation result at each node maintains consistency with the derivative, preserving the smoothness and monotonicity of the data and avoiding oscillation problems.

[0099] A4: After processing the data using steps A1-A3, organize the output data of the three-phase voltage transformers in the same group of the interpolated original dataset into a matrix form K to obtain the corresponding input vector:

[0100]

[0101] In the formula, w x (t), w y (t) and w z (t) represents the output of the three-phase voltage transformer at time t; M is the sequence length; the essence of the matrix transformation is to concatenate three one-dimensional data together.

[0102] Next, multiply matrix K by the sparse matrix Q. (3×6) Z is used as the input to the model:

[0103]

[0104] Calculations show that,

[0105] [Z x |Z y |Zz ] = [W y (t)W z (t)|W x (t)W z (t)|W x (t)W y (t)] (7);

[0106] Further simplification:

[0107] [X x |X y |X z ] = K = [W x (t)|W y (t)|W z (t)] (8);

[0108] In the formula, Z x Z y and Z z These are the input feature matrices of the three mutual inductors; Q converts K into the input form of the prediction model, and W... x (t), W y (t) and W z (t) is the corresponding input vector.

[0109] Before using the processed data for model training, it is recommended to perform further data validation and visualization analysis to confirm whether the data meets expectations.

[0110] The above steps effectively preprocess the data, laying a solid foundation for subsequent KNN-ELM model training and real-time state evaluation. This embodiment uses the KNN-ELM model.

[0111] In the preferred scheme, step S2 involves building a KNN-ELM model that combines KNN and ELM models, where:

[0112] The KNN model calculates the Euclidean distance between samples in the feature space of the input vector, identifies the nearest samples in historical data that are most similar to the features of the current input data, sorts them in order, and constructs the corresponding feature vectors.

[0113] The feature vector expression is:

[0114] x i =g([d i,1 ,θ i,1 ,x i,1 ],[d i,2 ,θ i,2 ,x i,2 ],…,[d i,k ,θ i,k ,xi,k ]) (9);

[0115] In the formula, d and θ are the relative polar coordinates of the training sample and the new sample, respectively; x is the label value of the training sample.

[0116] Then, the ELM model is used to perform regression prediction on the eigenvectors to obtain the predicted value of the secondary output voltage of the three-phase voltage transformer at each time step.

[0117] The ELM model introduces a penalty factor, as shown in the following formula:

[0118]

[0119] In the formula, β* is the least squares solution of weight β; I is the diagonal identity matrix; C is the regularization coefficient; and h is the hidden node output of ELM.

[0120] When building a KNN-ELM model, it is necessary to carefully select hyperparameters such as the value of K (the number of neighboring samples), the number of hidden nodes in the ELM, and the regularization coefficient C to optimize the model's performance.

[0121] After obtaining the outputs of the KNN and ELM models, they are used to calculate the evaluation threshold. In real-time status evaluation, the evaluation threshold is used to determine whether the current device status is normal. If the current evaluation index exceeds the threshold, a warning or alarm may be triggered.

[0122] In this embodiment, the ELM model receives the feature vector extracted by the KNN model as input and performs regression prediction to obtain the predicted secondary output voltage value of the three-phase voltage transformer at each time step.

[0123] In the preferred embodiment, the calculation of the evaluation threshold includes the following steps:

[0124] During model training, the output of the KNN-ELM model is mapped one-to-one with the original dataset after interpolation, and a subtraction operation is performed to obtain the prediction residuals. The squared prediction residuals are selected as the evaluation metric for the offline model, and the formula is as follows:

[0125]

[0126] In the formula, e i For three-phase voltage transformers at t i Predict the residual vector at each time step; For e i The transpose of Q; i This represents the corresponding squared prediction residual, i.e., the evaluation index.

[0127] Based on the obtained predicted residuals, the distribution of the residual evaluation index is estimated using a Gamma probability model. A suitable high quantile is selected as the evaluation threshold for the evaluation index Q using the control chart method. This evaluation threshold is denoted as Q0. q .

[0128] like Figure 4 As shown, Figure a is a graph of evaluation indicators and evaluation thresholds for the KNN-ELM model in this embodiment when it is offline, and Figure b is a graph of real-time online test results of the KNN-ELM model obtained using the thresholds in Figure a.

[0129] A self-updating strategy for the evaluation threshold is proposed to reduce the impact of three-phase imbalance fluctuations in the power grid on the transformer evaluation statistic Q. This strategy achieves adaptive adjustment by calculating the evaluation threshold in real time using a weighted sliding time window. It updates the current threshold by assigning higher weights to newer data within the sliding window, while simultaneously and very slowly expanding the influence of normal data on Q through the accumulation of P. q The weighting effect of (t+1).

[0130] In the preferred embodiment, the system early warning in step S4 is divided into different alarm trigger levels based on the frequency of deviation, including two states: warning and abnormal. The process is as follows:

[0131] 1) If the current online running data Q(t) is greater than the adaptively updated judgment threshold Q q (t), then start marking, and set Q q (t) is labeled as Q E .

[0132] 2) Continue calculating the evaluation index Q(t) of the online operation data for the next time interval. If n consecutive... E Each time interval is greater than Q. E If it does, it will be marked as an alert and logged.

[0133] 3) After being marked as an alert state, if there are n consecutive... E A time interval, or 3*n consecutive intervals E There are at least 2*n time intervals. E The number of discriminant indicators are all greater than Q. E If the condition is abnormal, it will be marked as an abnormal state, triggering a fault alarm and alerting maintenance personnel.

[0134] Where, n E The value is set as follows: based on the maximum number of consecutive missing values ​​and consecutive values ​​greater than three times the standard deviation in the historical measurement data collected in step S1.

[0135] The logs include timestamps, judgment thresholds, evaluation metrics, alert status (warning or abnormal), and duration.

[0136] This embodiment provides an alarm triggering and discrimination process, classifies alarm levels based on the frequency of deviations, prompts maintenance personnel whether further inspection is needed, and records relevant data in real time to form a monitoring log.

[0137] In the preferred scheme, in step S3, determining whether a consistency drift fault has occurred is achieved by monitoring the rate of change of the evaluation index change ratio to help determine whether the current transformer group has deteriorated simultaneously. The formula is as follows:

[0138]

[0139] In the formula, R mean,i T represents the relative change in the mean of the i-th window compared to the (i-1)-th window; i and T i-1 These are the dynamic mean values ​​of the i-th and (i-1)-th windows, respectively, and n is the total number of windows;

[0140]

[0141] In the formula, ΔR mean,i ξ represents the rate of change of the change ratio of the i-th window, max() takes the maximum of the two, ξ is the minimum positive number, and n is the total number of windows;

[0142]

[0143] In the formula, R std,i S represents the relative change in the standard deviation of the i-th window compared to the (i-1)-th window; i and S i-1 are the dynamic standard deviations of the i-th and (i-1)-th windows, respectively, and n is the total number of windows;

[0144]

[0145] In the formula, ΔR std,i R represents the rate of change of the i-th window relative to the (i-1)-th window. std,i With R std,i-1 The rate of change, n is the total number of windows, and ξ is a very small positive number to prevent the denominator from being 0.

[0146] The degradation drift of instrument transformers increases over time, manifesting as the accumulation of systematic errors and an increase in the measurement ratio error caused by the slow changes in specific parameters. It is an extremely low probability event that two-phase and three-phase transformers simultaneously begin degradation drift and maintain the same degradation rate when operating independently. Therefore, when the model initially determines an instrument transformer fault, it sets the fault as a single-phase fault, records the values ​​and trends of the rate of change and acceleration indices, and then uses the change ratio R... mean,i R std,i The rate of change of the ratio ΔRmean,i ΔR std,i The change ratio index and the rate of change of the change ratio index are compared with those at the time of the first fault to help determine whether it is a two-phase or three-phase fault.

[0147] To assess the effectiveness of the real-time status evaluation, a fault data point that varies linearly over time was added to the normal data set. This model was then used for real-time evaluation, and the evaluation results were input into the database. Figure 5 As shown, the real-time indicators of the fault data are significantly higher than the evaluation threshold, indicating a good judgment result. Figure 5-6 As shown, the change ratio R mean R std and the rate of change of the relative change ratio ΔR mean ΔR std The trend of change. To illustrate the trend more clearly, a local area is cropped from the graph. It can be observed that before the fault occurred, ΔR... mean ΔR std It fluctuates randomly around 0, and after a gradual change due to a fault at point 6000, it exhibits a nearly stable linear trend. and Δk is a constant, n E The number of fault data. The rate of change of the mean of the fault data. This represents the rate of change of the standard deviation of the fault. The ratio of the change in the mean to the standard deviation is R0. mean R std The initial alternating oscillations then transform into a trend where the standard deviation significantly follows the increase of the mean, i.e. and υ is a threshold coefficient greater than 1. The mean of the fault data. The standard deviation of the fault data. This is the mean of the normal values. This represents the standard deviation of the normal values. One helpful reference indicator for judgment is as follows:

[0148] Table 1. Judgment Criteria and Categories of Reference Indicators

[0149]

[0150] As shown in Table 1, Δq is a constant. The mean of the data for the next time window. The standard deviation of the data for the next time window. The rate of change of the mean of the data in the next time window. This represents the rate of change of the standard deviation of the data for the next time window. The above criteria will only hold true if all conditions are met simultaneously.

[0151] This embodiment, based on historical data from the secondary side output of three-phase voltage transformers, establishes a virtual standard using the KNN-ELM model and evaluation indicators through the output coordination correlation between three-phase voltage transformers. This transforms the performance degradation of the three-phase voltage transformers into an anomaly detection problem. The real-time collected voltage transformer output data is compared with the established virtual standard indicators to determine whether the voltage transformer is in an abnormal state at the current moment. First, compared to traditional offline verification methods, by extracting statistics from existing voltage transformer output data and analyzing their changing patterns, the presence of anomalies in the operating transformer at the current moment is determined, reducing maintenance costs and lowering system operation risks. Second, online anomaly detection of operating voltage transformers enables timely detection of early anomalies, reducing reliance on manual offline verification and significantly lowering maintenance frequency and costs. Third, using the transformer's own historical operating data as a benchmark eliminates the need for complex hardware equipment; therefore, the method described in this invention is applicable to voltage transformers of different voltage types, exhibiting high versatility. This characteristic supports system scalability, enabling its application in various power grid monitoring scenarios and further enhancing the practical value of this technology.

[0152] Example 2

[0153] To further illustrate with reference to Embodiment 1, a real-time status assessment system for voltage transformers includes:

[0154] The data acquisition module is used to collect historical operating data of the voltage transformer, perform data preprocessing, and save it as a raw dataset. The historical operating data is data that characterizes the standard operating state of the transformer, including the secondary output of the three-phase voltage transformer under normal equipment conditions.

[0155] The model training module is used to build a KNN-ELM model, train it using the original dataset, and save the optimal KNN-ELM model.

[0156] The model evaluation module is used to input the current operating data of the voltage transformer into the optimal KNN-ELM model, obtain the predicted value of the current secondary output of the three-phase voltage transformer, and calculate the current evaluation index.

[0157] The early warning trigger module is used to determine if the current voltage transformer is in an abnormal state if the current evaluation index is greater than the evaluation threshold, and to trigger and record the system early warning; the system early warning needs to be classified into alarm trigger levels.

[0158] The above embodiments are merely preferred technical solutions of the present invention and should not be considered as limitations on the present invention. The scope of protection of the present invention should be limited to the technical solutions described in the claims, including equivalent substitutions of the technical features described in the claims. That is, equivalent substitutions and improvements within this scope are also within the scope of protection of the present invention.

Claims

1. A method for real-time status assessment of voltage transformers, characterized in that, Includes the following steps: S1: Collect historical operating data of voltage transformers, perform data preprocessing, and save it as the raw dataset; The historical operating data is data that characterizes the standard operating state of the transformer, including the secondary output of the three-phase voltage transformer under normal equipment conditions. S2: Build a KNN-ELM model and train it using the original dataset, then save the optimal KNN-ELM model; S3: Input the current operating data of the voltage transformer into the optimal KNN-ELM model to obtain the predicted value of the current secondary output of the three-phase voltage transformer, and calculate the current evaluation index. In step S3, it is determined whether a consistency drift fault has occurred. This is done by monitoring the rate of change of the evaluation index to help determine whether the current transformer group has deteriorated simultaneously. The formula is as follows: ； In the formula, Indicates the first Window relative to the first The relative change in the mean of the window; and They are the first and the The dynamic mean of the windows, where n is the total number of windows; ； In the formula, Indicates the first The window's change ratio is the rate of change, and max() takes the maximum of the two. Let n be a very small positive number, and n be the total number of windows; ； In the formula, Indicates the first i Window relative to the first The relative change in the standard deviation of the window; and They are the first and the The dynamic standard deviation of the windows, where n is the total number of windows; ； In the formula, Indicates the first i Window relative to the first The rate of change of the window's change ratio, i.e. and The rate of change, where n is the total number of windows; When the model first determines that the current transformer is faulty, it sets the fault of the current transformer group at this time as a single-phase fault, records the values ​​and trends of the rate of change and acceleration indices, and uses the change ratio... , The rate of change of the ratio of change , The change ratio and the rate of change of the change ratio are compared with those at the time of the first fault to help determine whether it is a two-phase or three-phase fault. The specific judgment conditions are as follows: Single-phase fault: Two-phase or three-phase fault: in, It is a constant. The number of fault data. The rate of change of the mean of the fault data. The standard deviation of the fault is the rate of change. For threshold coefficients greater than 1, The mean of the fault data. The standard deviation of the fault data. This is the mean of the normal values. The standard deviation of the normal values; It is a constant. The mean of the data for the next time window. The standard deviation of the data for the next time window. The rate of change of the mean of the data in the next time window. The rate of change of the standard deviation of the data in the next time window; S4: If the current evaluation index is greater than the evaluation threshold, the current voltage transformer is in an abnormal state, and a system warning is triggered and recorded; the system warning needs to be classified into alarm triggering levels.

2. The real-time status assessment method for voltage transformers according to claim 1, characterized in that, If the current evaluation index is not greater than the evaluation threshold, it is considered a normal state. The evaluation threshold is then updated as follows: Continue to collect online operational data in real time, calculate evaluation thresholds in real time using a weighted sliding time window, and dynamically adjust them, including: The evaluation threshold is calculated in real time using a weighted sliding time window, and the formula is as follows: ； In the formula, The current moment; Q q ( t +1) is the statistical threshold for the next time step; Q’ The confidence quantile within the sliding window at the previous time step; Q q This is the initial threshold; To adjust the weights; Q ( t () is the evaluation statistic at the current moment; P This represents the cumulative number of samples.

3. The real-time status assessment method for voltage transformers according to claim 1, characterized in that, The data preprocessing in step S1 includes: A1: Determine if the collected data is missing, and mark the missing values ​​as NaN; A2: Use 3 σ The principle is to remove data, and the removed data is marked as NaN. The formula is: ； ； In the formula, xi is the i-th value in the dataset, and N is the total number of data points; A3: Interpolate the labeled NaN using PCHIP, the formula is as follows: ； In the formula, z 0 and z 1 represents the known time points before and after the point to be interpolated. f 0 and f 1 represents the known data values ​​corresponding to these time points; and This is the corresponding first derivative; the interpolated data is saved to the original dataset. A4: Organize the output data of the three-phase voltage transformers in the same group of the interpolated original dataset into a matrix form K to obtain the corresponding input vector: ； In the formula, w x ( t ), w y ( t )and w z ( t ) respectively represent the three-phase voltage transformers at Output at any moment; M The sequence length is given; next, the matrix... K Multiply by a sparse matrix Q (3×6) , as input to the model Z : ； Calculations show that, ； Further simplification: ； In the formula, Z x , Z y and Z z These are the input feature matrices of the three mutual inductors; Q Will K Converted into the input format of the prediction model, , and This is the corresponding input vector.

4. The real-time status assessment method for voltage transformers according to claim 1, characterized in that, In step S2, the KNN-ELM model is built by combining the KNN and ELM models, where: The KNN model calculates the Euclidean distance between samples in the feature space of the input vector, identifies the nearest samples in historical data that are most similar to the features of the current input data, sorts them in order, and constructs the corresponding feature vectors. The ELM model is used to perform regression prediction on the feature vector to obtain the predicted value of the secondary output voltage of the three-phase voltage transformer at each time step. The ELM model introduced above incorporates a penalty factor, as shown in the following formula: ； In the formula: For weight β The least squares solution; I It is a diagonal identity matrix; C The regularization coefficient is . h Output for hidden nodes in ELM; for The transpose of .

5. The real-time status assessment method for voltage transformers according to claim 1, characterized in that, The calculation of the evaluation threshold includes the following steps: During model training, the output of the KNN-ELM model is mapped one-to-one with the original dataset after interpolation, and a subtraction operation is performed to obtain the prediction residuals. The squared prediction residuals are selected as the evaluation metric for the offline model, and the formula is as follows: ； In the formula, For three-phase voltage transformers Predict the residual vector at each time step; for The transpose of the matrix; This represents the corresponding squared prediction residual, i.e., the evaluation index; Based on the obtained predicted residuals, the distribution of the residual evaluation index is estimated using the Gamma probability model, and the control chart method is used to select an appropriate high quantile as the evaluation threshold for the evaluation index.

6. The real-time status assessment method for voltage transformers according to claim 1, characterized in that, In step S4, the system warning is divided into different alarm trigger levels based on the frequency of deviation, including two states: warning and abnormal. The process is as follows: 1) If the currently online running data Q ( t () greater than the adaptively updated judgment threshold Q q ( t If the condition is met, then marking will begin; 2) Continue calculating the evaluation metrics for the online operation data in the next time interval. Q ( t ), and the time when updates stop. Q E Compare, if continuous Each time interval is greater than Q E If so, it will be marked as an alert and logged. 3) After being marked as a warning state, if there are consecutive A time interval, or 3 consecutive times At least 2* in each time interval If all the criteria for judging the number of indicators are greater than the judgment threshold, it is marked as an abnormal state, triggering a fault alarm and alerting the operation and maintenance personnel. in, The value is set as follows: based on the maximum number of consecutive missing values ​​and consecutive values ​​greater than three times the standard deviation in the historical measurement data collected in step S1; The logs include timestamps, judgment thresholds, evaluation metrics, alarm status, and duration.

7. A real-time condition assessment system for voltage transformers, characterized in that, The method for real-time status assessment of a voltage transformer according to claim 1 includes: The data acquisition module is used to collect historical operating data of the voltage transformer, perform data preprocessing, and save it as a raw dataset; the historical operating data is data that characterizes the standard operating state of the transformer, including the secondary output of the three-phase voltage transformer under normal equipment conditions; The model training module is used to build the KNN-ELM model, train it using the original dataset, and save the optimal KNN-ELM model. The model evaluation module is used to input the current operating data of the voltage transformer into the optimal KNN-ELM model, obtain the predicted value of the current secondary output of the three-phase voltage transformer, and calculate the current evaluation index. The early warning trigger module is used to determine if the current voltage transformer is in an abnormal state if the current evaluation index is greater than the evaluation threshold, and to trigger and record the system early warning; the system early warning needs to be divided into alarm trigger levels.

Images