A method for detecting insulation of high voltage power capacitor

Abstract

The present invention relates to the technical field of capacitor insulation detection, and discloses a high-voltage power capacitor insulation detection method, which constructs a feature data set by collecting insulation parameters, environmental conditions and production parameters of different types of capacitors in historical detection, and constructs an insulation parameter prediction model based on this data set. The model can predict the insulation parameter change trend of the capacitor in the future, and compare and analyze it with the insulation parameters detected in real time. By comparing the model prediction value with the real-time detection value, combined with the preset threshold and evaluation rules, the deviation of the capacitor insulation performance can be discovered in time, providing a scientific basis for the maintenance and replacement of the capacitor. The present invention improves the adaptability and accuracy of capacitor insulation detection, provides a strong guarantee for the stability and safety of the power system, and has broad application prospects and practical value.

Description

Technical Field

[0001] The invention relates to the technical field of capacitor insulation detection, in particular to a high-voltage power capacitor insulation detection method. Background Art

[0002] In the power system, high-voltage power capacitors are important reactive power compensation and filtering equipment, and their insulation performance is directly related to the stability and safety of the power system. Especially in the early or middle stage of capacitor operation, minor insulation degradation is often difficult to be discovered in time. Once the insulation performance is seriously degraded, it may cause capacitor failure and even paralyze the entire power system.

[0003] In the prior art, capacitor insulation detection methods are often designed for capacitors of specific types or specifications, which means that these methods have limited applicability and are difficult to adapt to the insulation detection needs of capacitors of different types and specifications. In practical applications, due to the large differences in capacitor types, specifications, operating environments, and production parameters, it is often difficult to obtain accurate and comprehensive detection results using a single or limited detection method.

[0004] Existing capacitor insulation testing methods are also deficient in providing references to standard test results. Due to differences in insulation performance between different types of capacitors, there is a lack of unified, standard test result references, which makes it difficult for testers to make accurate and reliable assessments when judging the insulation status of capacitors. For example, during capacitor insulation testing, testers found that the insulation resistance value of a certain model of capacitor was significantly lower than normal, but due to the lack of reference to standard test results for this model of capacitor, testers could not accurately determine whether this meant that the capacitor had insulation degradation or failure. In this case, if it is misjudged as normal, it may cause the capacitor to fail in subsequent operation; if it is misjudged as a failure, it may lead to unnecessary capacitor replacement and increased repair costs. Summary of the invention

[0005] The purpose of the present invention is to provide a high-voltage power capacitor insulation detection method to solve the problems raised in the above background technology.

[0006] To achieve the above object, the present invention provides the following technical solution: a high-voltage power capacitor insulation detection method, the method comprising:

[0007] Step 1: Collect the time series data of insulation parameters of different types of high-voltage power capacitors in historical tests, as well as the environmental conditions during the test and the production parameters of the tested capacitors; the insulation parameters include insulation resistance, absorption ratio, polarization index, leakage current and dielectric loss factor; the environmental conditions include the temperature and humidity of the test environment; the production parameters include capacitance value, rated voltage, rated capacity, dielectric material, electrode material, pin structure and manufacturing process;

[0008] Step 2: Preprocess and extract features of the data collected in step 1 to form a feature data set;

[0009] Step 3: Build an insulation parameter prediction model for high-voltage power capacitors. This model is based on the feature data set, combined with environmental conditions and capacitor production parameters, to predict the insulation parameter change trend of the capacitor in the future during testing.

[0010] Step 4: Obtain the current production status and environmental condition data of the high-voltage power capacitor;

[0011] Step 5: Input the data obtained in step 4 into the insulation parameter prediction model, and receive the predicted insulation parameters output by the model;

[0012] Step 6: Perform real-time insulation parameter detection on high-voltage power capacitors to obtain data on actual insulation resistance, absorption ratio, polarization index, leakage current, and dielectric loss factor, and clean the original detection data to remove abnormal values;

[0013] Step 7: Plot the real-time insulation parameter data obtained in step 6 and the predicted insulation parameter data output by the model in step 5 into time-varying curves, align the timestamps of the two curve data, and put the two curves into the same coordinate system;

[0014] Step 8: determine whether each point on the two curves has a deviation exceeding the threshold value according to a preset threshold value, and if so, mark it as a deviation point;

[0015] Step 9: Based on the deviation points found and the preset evaluation rules, the insulation status of the high-voltage power capacitor is evaluated and the evaluation results are sent to relevant personnel.

[0016] Preferably, in step 2, the data preprocessing method includes:

[0017] For missing values, linear interpolation algorithm is used to fill them;

[0018] For outliers, the threshold judgment method is used to regard data points that exceed the preset threshold range as outliers and replace them with the mean of their neighboring points;

[0019] Using the Z-score standardization method, each feature value is subtracted from its mean and divided by its standard deviation, so that the processed data conforms to the standard normal distribution. The formula for standardization is: ,in, is the original data, is the mean of the original data, is the standard deviation of the original data, is the standardized data.

[0020] Preferably, in step 2, a filtering method is used for feature extraction, and the importance of the feature is evaluated by calculating the correlation between each feature and the target insulation parameter; the linear correlation between each feature and the target insulation parameter is measured using the Pearson correlation coefficient, and the calculation formula of the Pearson correlation coefficient is:

[0021]

[0022] Where r represents the correlation coefficient between the feature and the insulation parameter, n represents the number of data points, x represents the feature value, and y represents the target insulation parameter value;

[0023] According to the calculated correlation coefficient, the features whose absolute values ​​are greater than the set threshold are extracted as useful features to form a feature data set for the subsequent training of the insulation parameter prediction model.

[0024] Preferably, in step 3, the step of constructing an insulation parameter prediction model includes:

[0025] Step 3.1: Use the feature dataset as input and divide the feature dataset into a training set and a test set, where the training set is used for model training and the test set is used for model evaluation;

[0026] Step 3.2: Build a long short-term memory network LSTM structure, set the appropriate number of network layers, number of hidden layer units and activation function;

[0027] Step 3.3: Use the training set to train the LSTM model, and iteratively optimize the model parameters through the back propagation algorithm and gradient descent method, so that the model can learn the mapping relationship between the feature data set and the insulation parameters;

[0028] Step 3.4: Use the test set to evaluate the trained LSTM model and verify the prediction performance of the model;

[0029] Step 3.5: Based on the evaluation results, adjust the hyperparameters of the LSTM model, including the number of network layers, number of hidden layer units, and activation function, until the prediction accuracy of the model reaches a satisfactory level;

[0030] Step 3.6: Save the trained insulation parameter prediction model.

[0031] Preferably, the network structure of the insulation parameter prediction model is:

[0032] Input layer: receives a feature data set as input, which includes insulation parameter history data, environmental condition data, capacitor production parameter data, and time-related features within a time window;

[0033] Network layer: Contains two LSTM hidden layers. The first hidden layer contains 128 LSTM units, and the second hidden layer contains 64 LSTM units. Each LSTM unit is equipped with a forget gate, an input gate, and an output gate to control the updating, transmission, and forgetting of information. The ReLU activation function is used to perform nonlinear transformation on the output of each LSTM unit to enhance the expressiveness of the model.

[0034] Output layer: Receives the output of the second hidden layer and maps the output of the LSTM network to the predicted insulation parameter values ​​through a fully connected layer, including the predicted values ​​of insulation resistance, absorption ratio, polarization index, leakage current, and dielectric loss factor.

[0035] Preferably, the LSTM model is iteratively trained using the training set data. In each iteration, a batch of feature data sets are sent to the LSTM model, and the values ​​of the input layer, hidden layer and output layer are calculated by forward propagation, and the LSTM memory unit and hidden state are updated to obtain the predicted insulation parameter output sequence; during the training process, the performance of the model is evaluated regularly using the validation set data, and the learning rate is adjusted or the training is stopped in advance according to the evaluation results to prevent the model from overfitting; the mean square error MSE algorithm is used as the loss function to evaluate the difference between the model prediction value and the actual measurement value. The MSE value is defined as the mean of the sum of the squares of the difference between the predicted value and the actual value. The smaller the MSE value, the higher the prediction accuracy of the model. The specific algorithm is:

[0036]

[0037] Where n is the number of data points, is the parameter value predicted by the model, is the actual measured parameter value.

[0038] Preferably, in step 8, the step of marking the deviation points includes:

[0039] Step 8.1: Align the timestamps of the two curves so that each time point has two values ​​from the real-time insulation parameter detection data and the model-predicted insulation parameter data;

[0040] Step 8.2: Set the insulation parameter deviation threshold, which is preset according to historical detection data and capacitor type, including the maximum allowable deviation range of insulation resistance, absorption ratio, polarization index, leakage current and dielectric loss factor;

[0041] Step 8.3: Traverse all time points on the two curves, and for each time point, calculate the deviation value between the real-time insulation parameter detection data and the model predicted insulation parameter data;

[0042] Step 8.4: Compare the calculated deviation value with the preset insulation parameter deviation threshold. If the deviation value exceeds the threshold, mark the time point as a deviation point, and record the specific value of the deviation and the corresponding parameter type.

[0043] Preferably, in step 9, the step of evaluating the insulation state of the high-voltage power capacitor includes:

[0044] Step 9.1: Count the deviation points marked in step 8, and record the timestamp, insulation parameter type, and deviation between the actual detection value and the predicted value corresponding to each deviation point;

[0045] Step 9.2: Analyze the statistical deviation points according to the preset evaluation rules to evaluate the insulation state of the high-voltage power capacitor; the evaluation rules include: the number of deviation points, the size of the deviation, the duration of the deviation and the frequency of the deviation;

[0046] Step 9.3: Generate an insulation status assessment report based on the assessment results, which includes an insulation status overview of the capacitor, detailed information on deviation points, and recommended maintenance measures;

[0047] Step 9.4: Send insulation condition assessment report to relevant personnel.

[0048] Compared with the prior art, the present invention has the following beneficial effects:

[0049] The present invention constructs a comprehensive feature data set by collecting insulation parameter time series data of high-voltage power capacitors of different types and specifications in historical detection, as well as corresponding environmental conditions and production parameters. This data set enables the insulation parameter prediction model to adapt to the insulation detection needs of capacitors of different types and specifications, greatly improving the adaptability and accuracy of detection.

[0050] By constructing an insulation parameter prediction model and combining the current environmental conditions and capacitor production parameters, this method can predict the insulation parameter change trend of the capacitor in the future during testing. This dynamic prediction capability helps to promptly discover potential insulation degradation problems and provide a scientific basis for the maintenance and replacement of capacitors. The present invention compares the insulation parameters predicted by the model with the insulation parameters detected in real time, plots them as a curve that changes over time, and performs timestamp alignment and deviation analysis. This method provides testers with a unified and intuitive reference for test results, helps to accurately evaluate the insulation status of the capacitor, and reduces the possibility of misjudgment and missed judgment. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 A processing step diagram of a high-voltage power capacitor insulation detection method according to the present invention;

[0052] Figure 2 Schematic diagram of data preprocessing steps;

[0053] Figure 3 Flowchart for the construction of insulation parameter prediction model. DETAILED DESCRIPTION

[0054] 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 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 creative work are within the scope of protection of the present invention.

[0055] See also Figure 1-3 The present invention provides a technical solution: a high-voltage power capacitor insulation detection method, the method comprising:

[0056] Step 1: Collect the time series data of insulation parameters of different types of high-voltage power capacitors in historical tests. These data include but are not limited to insulation resistance, absorption ratio, polarization index, leakage current and dielectric loss factor. At the same time, record the environmental conditions during each test, such as the temperature and humidity of the test environment, and the production parameters of the tested capacitors, including capacitance, rated voltage, rated capacity, dielectric material, electrode material, pin structure and manufacturing process, etc. These data will serve as the basis for subsequent model construction and prediction.

[0057] Step 2: Preprocess the collected data, including data cleaning, data normalization and other steps to eliminate noise and outliers in the data and improve the quality of the data. Next, extract features from the preprocessed data to form a feature data set. These features should be able to reflect the changing trend of the capacitor insulation performance and the relationship between environmental conditions and production parameters.

[0058] Step 3: Based on the feature data set, combined with environmental conditions and capacitor production parameters, an insulation parameter prediction model is constructed. The model can use machine learning algorithms, such as support vector machines and neural networks, to predict the insulation parameters of capacitors. During the model training process, historical test data should be fully utilized, and the prediction accuracy and generalization ability of the model should be improved by continuously adjusting the model parameters.

[0059] Step 4: Obtain the current production status and environmental condition data of high-voltage power capacitors. These data can be obtained through real-time monitoring by sensors or by querying the capacitor management system. Ensure the accuracy and real-time nature of the data to provide a basis for subsequent prediction and evaluation.

[0060] Step 5: Input the real-time acquired data into the insulation parameter prediction model and receive the predicted insulation parameters output by the model. These predicted parameters should include insulation resistance, absorption ratio, polarization index, leakage current and dielectric loss factor, etc., so as to fully understand the insulation performance of the capacitor.

[0061] Step 6: Perform real-time insulation parameter testing on high-voltage power capacitors to obtain data on actual insulation resistance, absorption ratio, polarization index, leakage current, and dielectric loss factor. During the testing process, the accuracy and reliability of the testing equipment should be ensured to avoid data errors caused by equipment problems. At the same time, the original test data should be cleaned to remove outliers and improve data quality.

[0062] Step 7: Plot the real-time insulation parameter data and the predicted insulation parameter data into time-varying curves respectively, align the timestamps of the two curve data, and put the two curves into the same coordinate system, so that the difference between the actual detection value and the predicted value can be intuitively observed.

[0063] Step 8: Determine whether there is a deviation exceeding the threshold for each point on the two curves based on the preset threshold. The threshold setting should be preset based on historical detection data and capacitor type to ensure the accuracy and rationality of the deviation judgment. If there is a deviation exceeding the threshold, it is marked as a deviation point, and the specific value of the deviation and the corresponding parameter type are recorded.

[0064] Step 9: Evaluate the insulation status of the high-voltage power capacitor based on the deviation points found and the preset evaluation rules. The evaluation rules should comprehensively consider factors such as the number of deviation points, the size of the deviation, the duration of the deviation, and the frequency of the deviation to ensure the accuracy and comprehensiveness of the evaluation results. Finally, send the evaluation results to relevant personnel.

[0065] The present invention will be further described below in conjunction with Examples 1 to 3:

[0066] Embodiment 1: For high voltage capacitors, the detection method of each insulation parameter is as follows:

[0067] Insulation resistance detection: Insulation resistance is one of the important indicators reflecting the insulation performance of capacitors. In order to accurately detect the insulation resistance of high-voltage power capacitors, a high resistance meter is used for testing. Before testing, make sure that the capacitor is powered off and fully discharged to avoid the influence of residual charge on the test results. Connect the positive and negative poles of the high resistance meter to the two ends of the capacitor respectively, set the appropriate test voltage and test time, and then start the test. After the test, the high resistance meter will display the insulation resistance value of the capacitor. In order to ensure the accuracy of the test results, multiple tests are performed and the average value is taken as the final result.

[0068] Absorption ratio and polarization index detection: Absorption ratio and polarization index are parameters that reflect the polarization process of the capacitor insulation medium. In order to detect these two parameters, an insulation resistance tester is used for testing. During the test, first set the test voltage and test time, and then start the test. The tester will record the insulation resistance values ​​at different time points and calculate the absorption ratio and polarization index based on these values. The absorption ratio is usually defined as the ratio of the insulation resistance values ​​measured after a specific time interval (such as 1 minute and 10 minutes), while the polarization index is defined as the ratio of the logarithm of the insulation resistance value change to the logarithm of time within a longer time interval (such as 10 minutes). By testing multiple times and taking the average value, a more accurate absorption ratio and polarization index can be obtained.

[0069] Leakage current detection: Leakage current is the current that flows through the insulating medium of a capacitor at rated voltage. To detect leakage current, a leakage current tester is used. Before testing, make sure that the capacitor has been connected to the rated voltage and stabilized for a period of time. Then, connect the positive and negative poles of the leakage current tester to the two ends of the capacitor respectively and start the test. The tester will display the leakage current value of the capacitor. To ensure the accuracy of the test results, perform multiple tests and take the average value, and pay attention to whether the leakage current changes over time.

[0070] Detection of dielectric loss factor: The dielectric loss factor is a parameter that reflects the dielectric loss characteristics of the capacitor. In order to detect the dielectric loss factor, a dielectric loss tester is used for testing. During the test, the positive and negative poles of the tester are connected to the two ends of the capacitor respectively, and the appropriate test frequency and voltage are set. The tester will display the dielectric loss factor value of the capacitor. In order to ensure the accuracy of the test results, multiple tests are performed and the average value is taken. At the same time, it is necessary to observe whether the dielectric loss factor changes with the changes in frequency and voltage.

[0071] Data cleaning and exception handling: After completing the above tests, a series of insulation parameter data can be obtained. In order to ensure the accuracy and reliability of the data, the data needs to be cleaned and exception handling is performed. First, check whether there are obvious errors or abnormal values ​​in the data, such as values ​​that exceed the reasonable range or values ​​that are obviously deviated from other data points. For these abnormal values, remove or correct them. Secondly, smooth the data to eliminate random fluctuations caused by test equipment or environmental factors.

[0072] The preprocessing methods include missing value processing, outlier processing and data standardization processing. The specific embodiments are as follows:

[0073] Missing value processing: In the process of data collection, some data may be missing due to various reasons. In order to maintain the integrity of the data, the linear interpolation algorithm is used to fill the missing values. The basic idea of ​​the linear interpolation algorithm is to infer the value of the missing data point through a linear relationship based on the value of the known data point. The specific implementation steps are as follows:

[0074] i) Traverse the data set and identify the location of missing values;

[0075] ii) For each missing value, find the two known data points before and after it;

[0076] iii) Use linear interpolation formula to calculate missing values:

[0077] Assume the previous known data point is , the next known data point is , the location of the missing value is , then the missing value It can be calculated by the following formula:

[0078]

[0079] iv) Fill the missing values ​​with the calculated values.

[0080] Outlier processing: Outliers refer to data points that significantly deviate from the overall distribution of the data set. They may have adverse effects on subsequent data analysis and model training. In order to eliminate the impact of outliers, the threshold judgment method is used for processing. The specific implementation steps are as follows:

[0081] i) Set a reasonable threshold range for each feature based on historical data and experience;

[0082] ii) Traverse the data set and determine whether each data point exceeds the set threshold range;

[0083] iii) Data points that exceed the threshold range are considered outliers and replaced with the mean of their neighboring points. Specifically, n data points before and after the outlier can be found (the value of n is determined according to the specific situation), the mean of these points can be calculated, and the outlier can be replaced with the mean.

[0084] Data standardization: In order to make the data between different features comparable, the Z-score standardization method is used to process the data. The basic idea of ​​the Z-score standardization method is to subtract the mean of each feature value and divide it by its standard deviation, so that the processed data conforms to the standard normal distribution. The specific implementation steps are as follows:

[0085] i) Calculate the mean and standard deviation of each feature;

[0086] ii) For each data point in the data set, standardization is performed according to the following formula: Let the original data be x, the mean be μ, the standard deviation be σ, and the standardized data be z, then:

[0087]

[0088] iii) Replace the corresponding values ​​in the original data set with the standardized data.

[0089] Example 2: In order to extract features useful for insulation parameter prediction from the original data set, a filtering method is used for feature selection. The filtering method is a method based on statistical testing, which evaluates the importance of features by calculating the correlation between each feature and the target insulation parameter.

[0090] The specific implementation steps are as follows:

[0091] Data preparation: First, prepare a dataset containing multiple features and target insulation parameters. Each row in the dataset represents a data point, and each column represents a feature or target insulation parameter.

[0092] Calculate Pearson correlation coefficient: Use the Pearson correlation coefficient to measure the linear correlation between each feature and the target insulation parameter. The calculation formula of the Pearson correlation coefficient is as follows:

[0093]

[0094] Where r is the correlation coefficient between the feature and the insulation parameter, n is the number of data points, x is the feature value, and y is the target insulation parameter value.

[0095] Set a threshold: Set a correlation coefficient threshold based on actual needs and the characteristics of the data set. The threshold can be set based on methods such as experiments or cross-validation.

[0096] Feature selection: Based on the calculated correlation coefficient, features with absolute values ​​greater than the set threshold are extracted as useful features. These features have a strong linear correlation with the target insulation parameters and are therefore more likely to have a positive impact on the prediction of the insulation parameters.

[0097] Constructing feature data set: The selected useful features are used to construct a new feature data set, which will be used for training the subsequent insulation parameter prediction model.

[0098] Assume that the data set contains 10 features (F1, F2, ..., F10) and a target insulation parameter (Y). First, calculate the Pearson correlation coefficient between each feature and the target insulation parameter to obtain a set of correlation coefficient values ​​(r1, r2, ..., r10). Then, set a correlation coefficient threshold, such as 0.5. Next, select the features corresponding to the correlation coefficients with an absolute value greater than 0.5 as useful features. Assuming that the absolute values ​​of r3, r5, and r7 are greater than 0.5, select F3, F5, and F7 as useful features to form a new feature data set.

[0099] The main function of step 3 is to build an insulation parameter prediction model. The following describes in detail the steps and network structure of the insulation parameter prediction model, including data division, LSTM network setting, model training, evaluation and adjustment.

[0100] Step 3.1, data preparation and partitioning: Use the feature data set as input, which contains the historical data of insulation parameters, environmental condition data, capacitor production parameter data and time-related features within the time window. In order to effectively train and evaluate the model, the feature data set is divided into a training set and a test set. The training set is used for model training to help the model learn the mapping relationship between the feature data set and the insulation parameters; the test set is used for model evaluation to verify the prediction performance of the model. The data partition ratio can be adjusted according to the actual situation. The common ratio is 70% training set and 30% test set.

[0101] Step 3.2, LSTM network structure setting: Build a long short-term memory network (LSTM) structure, set the number of network layers, number of hidden layer units and activation function. The specific network structure is as follows:

[0102] Input layer: receives the feature dataset as input, and the input dimension is consistent with the number of features in the feature dataset.

[0103] Network layer: Contains two LSTM hidden layers. The first hidden layer contains 128 LSTM units, and the second hidden layer contains 64 LSTM units. Each LSTM unit has a forget gate, input gate, and output gate to control the update, transmission, and forgetting of information. The ReLU activation function is used to perform nonlinear transformation on the output of each LSTM unit to enhance the expressiveness of the model.

[0104] Output layer: Receives the output of the second hidden layer and maps the output of the LSTM network to the predicted insulation parameter values ​​through a fully connected layer, including the predicted values ​​of insulation resistance, absorption ratio, polarization index, leakage current, and dielectric loss factor.

[0105] Step 3.3, model training: Use the training set to train the LSTM model. During the training process, the model parameters are iteratively optimized through the back-propagation algorithm and the gradient descent method. In each iteration, a batch of feature data sets are sent to the LSTM model, and the values ​​of the input layer, hidden layer, and output layer are calculated through forward propagation. The LSTM memory unit and hidden state are updated to obtain the predicted insulation parameter output sequence. During the training process, the mean square error (MSE) algorithm is used as the loss function to evaluate the difference between the model prediction value and the actual measurement value. The MSE value is defined as the mean of the sum of the squares of the difference between the predicted value and the actual value. The smaller the MSE value, the higher the prediction accuracy of the model. The specific algorithm is:

[0106]

[0107] Where n is the number of data points, is the parameter value predicted by the model, is the actual measured parameter value.

[0108] Step 3.4, model evaluation: Use the test set to evaluate the trained LSTM model. Verify the model's prediction performance by calculating the MSE value or other evaluation indicators on the test set. If the model performance does not meet expectations, further adjustments are required.

[0109] Step 3.5, model adjustment and optimization: According to the evaluation results, adjust the hyperparameters of the LSTM model, including the number of network layers, the number of hidden layer units, and the activation function. By constantly trying different hyperparameter combinations and retraining the model, the prediction accuracy of the model reaches a satisfactory level. In addition, during the training process, the performance of the model can be regularly evaluated using the validation set data, and the learning rate can be adjusted or the training can be stopped early according to the evaluation results to prevent the model from overfitting.

[0110] Step 3.6, model saving: save the trained insulation parameter prediction model.

[0111] The LSTM model has a strong nonlinear modeling capability and can capture complex patterns and relationships in the data. In the prediction of insulation parameters, insulation performance may be affected by a variety of factors, and there may be nonlinear relationships between these factors and insulation parameters. LSTM can effectively learn and represent these nonlinear relationships through its nonlinear activation function and multi-layer network structure. The following is a Python code example for building the model:

[0112]

[0113] The code uses StandardScaler to standardize the data, and uses the train_test_split function to divide the data set into a training set and a test set, with a ratio of 70% training set and 30% test set.

[0114] The code uses the Sequential class of Keras to build an LSTM model. The model contains two LSTM hidden layers, the first hidden layer has 128 LSTM units, and the second hidden layer has 64 LSTM units. The output layer is a fully connected layer with a linear activation function because this is a regression problem. The model is compiled using the Adam optimizer and the mean squared error (MSE) as the loss function. The mean absolute error (MAE) is also added as an evaluation metric. The model is trained using the training set data, and 20% of the data is used as a validation set during training to monitor the performance of the model. 50 epochs are trained, and each batch contains 32 samples. Model evaluation: The trained model is evaluated using the test set data, and the loss and MAE of the test set are output. Finally, the trained model is saved as an HDF5 file for subsequent use.

[0115] Embodiment 3: In order to accurately mark the deviation points between the real-time insulation parameter detection data and the model-predicted insulation parameter data, this embodiment is used to describe the implementation of marking the deviation points in step 8, and the specific steps include:

[0116] Step 8.1, timestamp alignment: align the timestamps of the real-time insulation parameter detection data and the model-predicted insulation parameter data to ensure that at each time point, both values ​​from the real-time detection and the model prediction can be obtained simultaneously for subsequent deviation calculations. This can be achieved by matching and sorting the timestamps of the two curves.

[0117] Step 8.2, set the insulation parameter deviation threshold: According to the historical detection data and capacitor type, preset the insulation parameter deviation threshold. These thresholds include the maximum allowable deviation range of insulation resistance, absorption ratio, polarization index, leakage current and dielectric loss factor. The setting of these thresholds should be based on the actual application scenarios and the performance requirements of the capacitor to ensure the accuracy and rationality of the deviation judgment.

[0118] Step 8.3, deviation value calculation: traverse all time points on the two aligned curves, and for each time point, calculate the deviation value between the real-time insulation parameter detection data and the model predicted insulation parameter data. The deviation value can be calculated by simple difference calculation, that is, the real-time detection value minus the model predicted value, or by relative deviation or percentage deviation.

[0119] Step 8.4, deviation point marking: Compare the calculated deviation value with the preset insulation parameter deviation threshold. If the deviation value exceeds the threshold, mark the time point as a deviation point, and record the specific value of the deviation and the corresponding parameter type.

[0120] Taking insulation resistance as an example, suppose that at a certain point in time, the real-time insulation parameter detection data is 100MΩ, the model predicts the insulation parameter data to be 80MΩ, and the preset insulation resistance deviation threshold is ±10MΩ. First, calculate the deviation value, that is, 100MΩ - 80MΩ = 20MΩ. Then, compare the deviation value 20MΩ with the threshold value ±10MΩ. Since 20MΩ is greater than 10MΩ, this time point is marked as a deviation point, and the deviation value 20MΩ and the corresponding parameter type "insulation resistance" are recorded.

[0121] The steps to accurately assess the insulation condition of high-voltage power capacitors include:

[0122] Step 9.1: Count the deviation points marked in step 8. The purpose of this step is to record the timestamp corresponding to each deviation point, the type of insulation parameter (such as insulation resistance, absorption ratio, polarization index, leakage current and dielectric loss factor, etc.), and the deviation between the actual detection value and the predicted value.

[0123] Step 9.2: According to the preset evaluation rules, the statistical deviation points are analyzed to evaluate the insulation status of the high-voltage power capacitor. The evaluation rules include the following aspects:

[0124] Number of deviation points: Count the total number of deviation points and the number of deviation points corresponding to each insulation parameter type. The more deviation points there are, the more unstable the insulation state of the capacitor may be.

[0125] Deviation: Analyze the deviation of each deviation point, as well as the maximum deviation and average deviation corresponding to each insulation parameter type. The larger the deviation, the more obvious the difference between the actual detection value and the predicted value, and the worse the insulation state of the capacitor may be.

[0126] Deviation duration: The duration of the statistical deviation point, that is, the time period during which the deviation occurs continuously. The longer the deviation lasts, the more likely the insulation condition of the capacitor may continue to deteriorate.

[0127] Frequency of deviation: Analyze the frequency of deviation points, that is, the number of deviation points that appear per unit time. The higher the frequency of deviation, the more frequently the insulation state of the capacitor changes, and closer attention may be needed.

[0128] Based on the above evaluation rules, the insulation status of the high-voltage power capacitor can be comprehensively evaluated to determine whether it is in a normal state, a slightly abnormal state, or a seriously abnormal state.

[0129] Step 9.3: Generate an insulation status assessment report based on the assessment results. The report contains the following:

[0130] Overview of the insulation status of the capacitor: Provide an overall description of the insulation status of the capacitor, including normal, slightly abnormal, or seriously abnormal.

[0131] Deviation point details: Lists all marked deviation points, including timestamp, insulation parameter type, actual detection value, predicted value, deviation amount, and deviation duration.

[0132] Recommended maintenance measures: Based on the evaluation results, corresponding maintenance measures are recommended, such as strengthening monitoring, performing maintenance or replacing capacitors.

[0133] Step 9.4: Send the generated insulation status assessment report to relevant personnel, such as operation and maintenance personnel, management personnel or technical experts, etc. Through the report, relevant personnel can timely understand the insulation status of the capacitor and take corresponding measures according to the recommended maintenance measures.

[0134] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0135] Although 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 the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A high-voltage power capacitor insulation detection method, characterized in that: The method comprises: Step 1: Collect the time series data of insulation parameters of different types of high-voltage power capacitors in historical tests, as well as the environmental conditions during the test and the production parameters of the tested capacitors; the insulation parameters include insulation resistance, absorption ratio, polarization index, leakage current and dielectric loss factor; the environmental conditions include the temperature and humidity of the test environment; the production parameters include capacitance value, rated voltage, rated capacity, dielectric material, electrode material, pin structure and manufacturing process; Step 2: Preprocess and extract features of the data collected in step 1 to form a feature data set; Step 3: Build an insulation parameter prediction model for high-voltage power capacitors. This model is based on the feature data set, combined with environmental conditions and capacitor production parameters, to predict the insulation parameter change trend of the capacitor in the future during testing. Step 4: Obtain the current production status and environmental condition data of the high-voltage power capacitor; Step 5: Input the data obtained in step 4 into the insulation parameter prediction model, and receive the predicted insulation parameters output by the model; Step 6: Perform real-time insulation parameter detection on high-voltage power capacitors to obtain data on actual insulation resistance, absorption ratio, polarization index, leakage current, and dielectric loss factor, and clean the original detection data to remove abnormal values; Step 7: Plot the real-time insulation parameter data obtained in step 6 and the predicted insulation parameter data output by the model in step 5 into time-varying curves, align the timestamps of the two curves, and place the two curves in the same coordinate system; Step 8: determine whether each point on the two curves has a deviation exceeding the threshold value according to a preset threshold value, and if so, mark it as a deviation point; Step 9: Evaluate the insulation status of the high-voltage power capacitor based on the deviation points found and the preset evaluation rules, and send the evaluation results to relevant personnel; In step 3, the step of constructing an insulation parameter prediction model includes: Step 3.1: Use the feature dataset as input and divide the feature dataset into a training set and a test set, where the training set is used for model training and the test set is used for model evaluation; Step 3.2: Build a long short-term memory network LSTM structure, set the appropriate number of network layers, number of hidden layer units and activation function; Step 3.3: Use the training set to train the LSTM model, and iteratively optimize the model parameters through the back propagation algorithm and gradient descent method, so that the model can learn the mapping relationship between the feature data set and the insulation parameters; Step 3.4: Use the test set to evaluate the trained LSTM model and verify the prediction performance of the model; Step 3.5: Based on the evaluation results, adjust the hyperparameters of the LSTM model, including the number of network layers, number of hidden layer units, and activation function, until the prediction accuracy of the model reaches a satisfactory level; Step 3.6: Save the trained insulation parameter prediction model; The network structure of the insulation parameter prediction model is: Input layer: receives a feature data set as input, which includes insulation parameter history data, environmental condition data, capacitor production parameter data, and time-related features within a time window; Network layer: Contains two LSTM hidden layers. The first hidden layer contains 128 LSTM units, and the second hidden layer contains 64 LSTM units. Each LSTM unit is equipped with a forget gate, an input gate, and an output gate to control the updating, transmission, and forgetting of information. The ReLU activation function is used to perform nonlinear transformation on the output of each LSTM unit to enhance the expressiveness of the model. Output layer: receives the output of the second hidden layer and maps the output of the LSTM network to the predicted insulation parameter values ​​through a fully connected layer, including the predicted values ​​of insulation resistance, absorption ratio, polarization index, leakage current and dielectric loss factor; In step 8, the steps of marking the deviation points include: Step 8.1: Align the timestamps of the two curves so that each time point has two values ​​from the real-time insulation parameter detection data and the model-predicted insulation parameter data; Step 8.2: Set the insulation parameter deviation threshold, which is preset according to historical detection data and capacitor type, including the maximum allowable deviation range of insulation resistance, absorption ratio, polarization index, leakage current and dielectric loss factor; Step 8.3: Traverse all time points on the two curves, and for each time point, calculate the deviation value between the real-time insulation parameter detection data and the model predicted insulation parameter data; Step 8.4: Compare the calculated deviation value with the preset insulation parameter deviation threshold. If the deviation value exceeds the threshold, mark the time point as a deviation point, and record the specific value of the deviation and the corresponding parameter type.

2. A high-voltage power capacitor insulation detection method according to claim 1, characterized in that: In step 2, the data preprocessing method includes: For missing values, linear interpolation algorithm is used to fill them; For outliers, the threshold judgment method is used to regard data points that exceed the preset threshold range as outliers and replace them with the mean of their neighboring points; Using the Z-score standardization method, each feature value is subtracted from its mean and divided by its standard deviation, so that the processed data conforms to the standard normal distribution. The formula for standardization is: ,in, is the original data, is the mean of the original data, is the standard deviation of the original data, is the standardized data.

3. A high-voltage power capacitor insulation detection method according to claim 1, characterized in that: In step 2, a filtering method is used to extract features, and the importance of features is evaluated by calculating the correlation between each feature and the target insulation parameter; the linear correlation between each feature and the target insulation parameter is measured using the Pearson correlation coefficient, and the calculation formula of the Pearson correlation coefficient is: Where r represents the correlation coefficient between the feature and the insulation parameter, n represents the number of data points, x represents the feature value, and y represents the target insulation parameter value; According to the calculated correlation coefficient, the features whose absolute values ​​are greater than the set threshold are extracted as useful features to form a feature data set for the subsequent training of the insulation parameter prediction model.

4. A high-voltage power capacitor insulation detection method according to claim 1, characterized in that: The LSTM model is iteratively trained using the training set data. In each iteration, a batch of feature data sets are fed into the LSTM model, and the values ​​of the input layer, hidden layer, and output layer are calculated through forward propagation. The LSTM memory unit and hidden state are updated to obtain the predicted insulation parameter output sequence. During the training process, the validation set data is regularly used to evaluate the performance of the model. The learning rate is adjusted or the training is stopped early based on the evaluation results to prevent the model from overfitting. The mean square error (MSE) algorithm is used as the loss function to evaluate the difference between the model prediction value and the actual measurement value. The MSE value is defined as the mean of the sum of the squares of the difference between the predicted value and the actual value. The smaller the MSE value, the higher the prediction accuracy of the model. The specific algorithm is as follows: Where n is the number of data points, is the parameter value predicted by the model, is the actual measured parameter value.

5. A high-voltage power capacitor insulation detection method according to claim 1, characterized in that: In step 9, the step of evaluating the insulation state of the high-voltage power capacitor includes: Step 9.1: Count the deviation points marked in step 8, and record the timestamp, insulation parameter type, and deviation between the actual detection value and the predicted value corresponding to each deviation point; Step 9.2: Analyze the statistical deviation points according to the preset evaluation rules to evaluate the insulation state of the high-voltage power capacitor; the evaluation rules include: the number of deviation points, the size of the deviation, the duration of the deviation and the frequency of the deviation; Step 9.3: Generate an insulation status assessment report based on the assessment results, which includes an insulation status overview of the capacitor, detailed information on deviation points, and recommended maintenance measures; Step 9.4: Send insulation condition assessment report to relevant personnel.

Images