Intelligent sequencing method for single-phase grounding test pull of small-current grounding system

Abstract

The invention relates to the technical field of intelligent sequencing, in particular to an intelligent sequencing method for single-phase grounding test pull of a small-current grounding system. The method comprises the following steps: S1, collecting real-time operation data of each feeder line; s2, extracting features capable of reflecting system state changes from the real-time operation data; s3, judging whether a single-phase earth fault occurs or not by using the extracted features and setting a threshold recognition rule; s4, performing intelligent sorting on all feeder lines by using a sorting training model, and determining a line which is most likely to have a fault; and S5, according to an intelligent sorting result, performing test pulling on the suspected fault line, and verifying which line is a real fault line. According to the design of the invention, through an intelligent sorting method, a line which is most likely to have a single-phase earth fault can be rapidly determined. According to the method, a traditional way of checking lines one by one is avoided, and unnecessary power failure times and time are reduced, so that the operation efficiency and the service quality of a power system are improved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent sorting technology, and in particular to an intelligent sorting method for single-phase grounding pull-out tests in a small current grounding system. Background Art

[0002] In traditional power distribution networks, when a single-phase grounding fault occurs, it is usually necessary to manually or semi-automatically check each feeder to determine the fault location. This approach is time-consuming, especially in large power grids, and may cause long-term power outages. When detecting faults, traditional methods may lead to fault judgment errors due to insufficient manual experience or low equipment detection accuracy, such as false alarms or missed fault lines, thus affecting the reliability of the power system. Without intelligent sorting, power maintenance personnel may perform unnecessary inspections and maintenance on non-faulty lines, which not only wastes manpower and material resources, but may also delay the repair of actual faulty lines. In summary, a single-phase grounding test pull intelligent sorting method for a small current grounding system is provided. Summary of the invention

[0003] The purpose of the present invention is to provide an intelligent sequencing method for single-phase grounding pull tests in a small current grounding system to solve the problems of slow fault detection speed, high risk of misjudgment and waste of resources proposed in the above background technology.

[0004] To achieve the above object, the present invention provides a method for intelligent sorting of single-phase grounding test pull-outs in a small current grounding system, comprising the following steps: S1. Collect real-time operation data of each feeder; S2, extracting features that can reflect changes in system status from real-time operation data; S3, using the extracted features, determining whether a single-phase grounding fault has occurred by setting a threshold recognition rule; S4. Use the sorting training model to perform intelligent sorting on all feeders, determine the line most likely to fail, and in the intelligent sorting process, consider the situation where the zero-sequence current and zero-sequence voltage are close to zero, and use auxiliary zero-sequence current and zero-sequence voltage prediction to further optimize the intelligent sorting process; S5. According to the result of intelligent sorting, test pull the suspected faulty line to verify which line is the real faulty line.

[0005] As a further improvement of the technical solution, in S1, the real-time operation data includes the current, voltage and frequency of the feeder.

[0006] As a further improvement of the technical solution, in S2, extracting features that can reflect changes in system status from real-time operation data includes the following steps: S2.1, preprocessing the real-time operation data; S2.2, using fast Fourier transform to convert the time domain signal into frequency domain representation, analyzing the frequency components of the operating data, and analyzing the instantaneous values ​​of the current and voltage of the operating data; S2.3. Calculate the zero-sequence current and zero-sequence voltage characteristics in the operating data, introduce the correction factor parameters of the feeder impedance, consider the influence of water vapor concentration in the air on the zero-sequence current and zero-sequence voltage, and further optimize the calculation process of the zero-sequence current and zero-sequence voltage; S2.4. Observe the changing trend of features over time and calculate the statistics of features.

[0007] As a further improvement of the technical solution, in S2.3, the zero-sequence current and zero-sequence voltage are calculated as: The zero sequence current is: ; The zero sequence voltage is: ; The correction factor parameter introduced into the feeder impedance is: The zero sequence current is: ; The zero sequence voltage is: ; Considering the influence of water vapor concentration in the air on zero-sequence current and zero-sequence voltage, the calculation process of zero-sequence current and zero-sequence voltage is further optimized as follows: The zero sequence current is: ; The zero sequence voltage is: ; in, Indicates zero-sequence current value; Indicates the zero-sequence voltage value; Indicates the A phase current; Indicates the B phase current; Indicates the C phase current; Indicates the voltage of phase A; Indicates the B phase voltage; Indicates the C phase voltage; represents the feeder impedance correction factor; It represents the impedance of the feeder; Indicates the zero-sequence current value after feeder impedance correction; Indicates the zero-sequence voltage value after feeder impedance correction; Indicates the zero-sequence current value after correction by feeder impedance and water vapor concentration in the air; It indicates the zero-sequence voltage value after correction by feeder impedance and water vapor concentration in the air; Represents the correction factor for water vapor concentration in air.

[0008] As a further improvement of the technical solution, in S3, determining whether a single-phase grounding fault has occurred by setting a threshold recognition rule includes the following steps: S3.1, set the threshold value m of the zero-sequence current mean value, and set the threshold value n of the zero-sequence voltage mean value; S3.2, comparing the extracted feature mean with the set threshold, the feature mean including the zero-sequence current mean and the zero-sequence voltage mean; S3.3, using the fault classification tree to classify the extracted feature means and determine whether the feeder is in a fault state; S3.4. If the mean values ​​of multiple features exceed the threshold at the same time, the confidence in the fault judgment is increased, and the current features are compared with the data during historical normal operation to find abnormal patterns.

[0009] As a further improvement of the technical solution, in S3.3, the extracted features are classified using a fault classification tree model, including the following steps: S3.31. Determine the feature set for training and use the fault classification tree to build an algorithm classification model; S3.32, compare the extracted feature mean with a preset threshold value, if the zero-sequence current mean value exceeds the threshold value m, mark the feeder as a suspected fault line; if the zero-sequence voltage mean value exceeds the threshold value n, mark the feeder as a suspected fault line; S3.33, inputting the feature mean data of the line marked as suspected fault into the trained fault classification tree model to make classification decisions; S3.34. Output a classification label, which indicates whether the feeder is in a single-phase grounding fault state.

[0010] As a further improvement of the technical solution, in S4, all feeders are intelligently sorted using a sorting training model, including the following steps: S4.1. Divide the historical operation data into training set, validation set and test set; S4.2. Initialize the parameters of the sorting training model, define the sorting loss function of the sorting training, introduce the statistics of the features into the sorting loss function, consider the situation that the zero-sequence current and zero-sequence voltage are close to zero, use the auxiliary zero-sequence current and zero-sequence voltage prediction to further optimize the sorting loss function, and use the learning update method to optimize the sorting training model parameters; S4.3. Use the data from the training set to train the sorting training model, evaluate the model performance on the validation set, and adjust the model parameters; S4.4. Use the trained and optimized sorting training model to predict the feeder data in the test set, sort the feeders according to the failure probability, and determine the line most likely to fail.

[0011] As a further improvement of the technical solution, in S4.2, the sorting loss function is: ; The statistics of the features introduced in the ranking loss function are: ; Considering the situation where the zero-sequence current and zero-sequence voltage are close to zero, the sorting loss function is further optimized using auxiliary zero-sequence current and zero-sequence voltage prediction as follows: ; in, Represents the weight vector in the sorting training model; Represents the bias term in the sorting training model; represents the total number of samples in the data set; Indicates the sample index; Indicates the number of features; Represents the feature index; represents the weight index; Indicates The true labels of samples; Indicates The predicted probability of samples; Indicates A vector composed of all features of samples; A weight vector representing the predicted mean; Represents the bias term of the predicted mean; A weight vector representing the prediction variance; A weight vector representing the prediction variance; The model predicts the Zero-sequence current value of samples; Indicates The actual zero-sequence current value of samples; The model predicts the Zero-sequence voltage value of samples; Indicates The actual zero-sequence voltage value of samples; Indicates the importance of controlling the prediction error of the feature statistic; Indicates the importance of controlling zero-sequence current and zero-sequence voltage prediction errors; represents the importance of controlling the regularization term; represents weight; Indicates the average value of zero-sequence current; Represents the average value of zero-sequence voltage.

[0012] As a further improvement of the technical solution, in S4.2, the learning update method is: ; ; in, represents the learning rate; Express The gradient of Express gradient.

[0013] As a further improvement of the technical solution, in S5, according to the result of intelligent sorting, the suspected faulty line is tested to verify which line is the real faulty line, including the following steps: S5.1. According to the results of intelligent sorting, formulate a test pulling plan and start the test pulling from the line most likely to fail; S5.2. Disconnect the line most likely to fail in the sorting sequence and monitor whether the zero-sequence current and zero-sequence voltage return to the normal range; S5.3. Record the changes of zero-sequence current and zero-sequence voltage after disconnecting the line; S5.4. If the zero-sequence current and zero-sequence voltage return to normal, the line can be confirmed as a faulty line and the power supply of other lines can be restored; S5.5. If the zero-sequence current and zero-sequence voltage are still abnormal, restore the power supply of this line and continue to test the next line that is most likely to fail; S5.6. According to steps S5.2-S5.5, test the lines in the sorted list in turn until the real faulty line is found.

[0014] Compared with the prior art, the present invention has the following beneficial effects: 1. In the intelligent sorting method for single-phase grounding test of the small current grounding system, the line most likely to have a single-phase grounding fault can be quickly determined through the intelligent sorting method. This method avoids the traditional way of checking lines one by one, reduces the number and time of unnecessary power outages, and thus improves the operating efficiency and service quality of the power system. By giving priority to checking suspicious lines, the fault point can be quickly located, reducing the time cost of troubleshooting and ensuring the safe and stable operation of the power grid.

[0015] 2. The intelligent sorting method for single-phase grounding test pull of the small current grounding system not only relies on the instant reading of real-time data, but also combines the learning and analysis of historical data. By introducing characteristic statistics, correction factors of zero-sequence current and voltage, and environmental factor correction, the fault judgment is made more accurate and reliable. This helps to reduce the misjudgment rate and ensure that only lines with real problems are isolated for maintenance, avoiding additional losses caused by misoperation. In addition, by optimizing the sorting loss function and adjusting the model parameters using the learning update method, the model is more adapted to the actual operation of the power grid and the accuracy of fault diagnosis is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 The figure is a flow chart of the overall method of the present invention. DETAILED DESCRIPTION

[0017] 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.

[0018] Example: See Figure 1 As shown, this embodiment provides an intelligent sorting method for single-phase grounding test pull-out in a small current grounding system, comprising the following steps: S1. Collect real-time operation data of each feeder; In this embodiment, the real-time operation data includes the current, voltage and frequency of the feeder.

[0019] S2. Extracting features that can reflect changes in system status from real-time operation data, the features include zero-sequence current and zero-sequence voltage; In this embodiment, extracting features that can reflect changes in system status from real-time operation data includes the following steps: S2.1. Preprocess the real-time operation data, including data cleaning and data standardization; S2.2, using fast Fourier transform to convert the time domain signal into frequency domain representation, analyzing the frequency components of the operating data, and analyzing the instantaneous values ​​of the current and voltage of the operating data; Among them, fast Fourier transform can be used to analyze the frequency components of signals, such as detecting harmonics and analyzing grid frequency fluctuations. By converting the current or voltage signal in the time domain into a frequency domain representation, specific frequency components can be more easily identified, which is very useful for diagnosing single-phase grounding faults. For example, the zero-sequence current and zero-sequence voltage will change significantly during a single-phase grounding fault. These changes can be clearly seen through FFT conversion, and fault detection and location can be performed accordingly. S2.3. Calculate the zero-sequence current and zero-sequence voltage characteristics in the operating data. Because the zero-sequence component will change significantly during a single-phase grounding fault, introduce the correction factor parameter of the feeder impedance, consider the influence of water vapor concentration in the air on the zero-sequence current and zero-sequence voltage, and further optimize the calculation process of the zero-sequence current and zero-sequence voltage. Among them, the zero-sequence current and zero-sequence voltage are calculated as: The zero sequence current is: ; The zero sequence voltage is: ; Different feeder impedances will lead to differences in the zero-sequence current and voltage measured under the same fault condition. By introducing the feeder impedance correction factor, this difference can be corrected to ensure consistent measurement results even on feeders with different impedances. The correction factor parameters of the feeder impedance make the zero-sequence current and voltage values ​​measured under different feeder conditions more standardized, which is convenient for the consistent application of the fault detection algorithm. The correction factor parameters of the feeder impedance are: The zero sequence current is: ; The zero sequence voltage is: ; Water vapor in the air will change the dielectric constant of the medium, thereby affecting the measured values ​​of current and voltage. By introducing a water vapor correction factor, the influence of humidity on the measurement results can be compensated. Considering the water vapor concentration in the air ensures that the measured values ​​of zero-sequence current and voltage can reflect the true fault state under different humidity environments and are not affected by changes in environmental water vapor. Considering the influence of water vapor concentration in the air on zero-sequence current and zero-sequence voltage, the calculation process of zero-sequence current and zero-sequence voltage is further optimized as follows: The zero sequence current is: ; The zero sequence voltage is: ; The relationship between water vapor concentration and water vapor correction factor is: ; in, Indicates zero-sequence current value; Indicates the zero-sequence voltage value; Indicates the A phase current; Indicates the B phase current; Indicates the C phase current; Indicates the voltage of phase A; Indicates the B phase voltage; Indicates the C phase voltage; Indicates the feeder impedance correction factor, which is used to adjust the calculation result of zero-sequence current according to the impedance characteristics of the feeder; It represents the impedance of the feeder; Indicates the zero-sequence current value after feeder impedance correction; Indicates the zero-sequence voltage value after feeder impedance correction; Indicates the zero-sequence current value after correction by feeder impedance and water vapor concentration in the air; It indicates the zero-sequence voltage value after correction by feeder impedance and water vapor concentration in the air; Indicates the correction factor of water vapor concentration in the air, which is used to compensate for the influence of water vapor on the measured value; Indicates water vapor concentration; It indicates the change in correction factor for each unit increase in water vapor concentration; Indicates the correction factor value when the water vapor concentration is zero; S2.4. Observe the time-varying trend of the feature, such as the time-varying curve of the zero-sequence current or voltage, and calculate the statistics of the feature. The statistics of the feature include the mean or variance of the zero-sequence current and zero-sequence voltage.

[0020] S3, using the extracted features, determining whether a single-phase grounding fault has occurred by setting a threshold recognition rule; In this embodiment, the method of setting a threshold allows the system to respond immediately when it detects that a specific indicator (such as zero-sequence current or voltage) exceeds a preset value, which means that once a fault occurs, the system can quickly identify and take measures to reduce the duration of the fault, thereby protecting the power grid from further damage; using the threshold recognition rule can simplify the complex fault detection problem into a binary judgment (i.e., whether it exceeds the threshold), thereby simplifying the decision-making process, which is particularly important for automation systems because it can reduce processing time and computing resource requirements; the threshold setting is based on a large amount of historical data and experience, which can effectively filter out normal fluctuations and noise, and only respond to situations that are truly beyond the normal range, which can reduce the false alarm rate and improve the accuracy of fault detection; determining whether a single-phase grounding fault has occurred by setting a threshold recognition rule includes the following steps: S3.1, set the threshold value m of the zero-sequence current mean value, and set the threshold value n of the zero-sequence voltage mean value; S3.2, comparing the extracted feature mean with the set threshold, if the feature exceeds the corresponding threshold, marking the feeder as a suspected fault line, the feature mean includes the zero-sequence current mean and the zero-sequence voltage mean; S3.3, using the fault classification tree to classify the extracted feature means and determine whether the feeder is in a fault state; Among them, the structure of the fault classification tree is similar to a tree, each internal node represents a test on an attribute, each branch represents a test output, and each leaf node represents a category (here "normal" or "fault"); the classification tree not only provides the classification results, but also shows the classification process, that is, why a certain feeder is judged to be faulty. This helps to diagnose the problem and provide troubleshooting directions; the fault classification tree can handle multiple types of input features (such as zero-sequence current mean, zero-sequence voltage mean, etc.), and can naturally handle the interaction between features, thereby improving the accuracy and reliability of fault detection; the extracted features are classified using the fault classification tree model, including the following steps: S3.31. Determine a feature set for training, such as zero-sequence current mean, zero-sequence current variance, and zero-sequence voltage mean, and use a fault classification tree to build an algorithm classification model. The construction of the fault classification tree involves recursively selecting the best features for segmentation until the termination condition (such as node purity, maximum depth, etc.) is met; S3.32, compare the extracted feature mean with a preset threshold value, if the zero-sequence current mean value exceeds the threshold value m, mark the feeder as a suspected fault line; if the zero-sequence voltage mean value exceeds the threshold value n, mark the feeder as a suspected fault line; S3.33, input the feature mean data of the line marked as suspected fault into the trained fault classification tree model, make classification decisions, and compare the features along the path of the fault classification tree until reaching the leaf node; S3.34, output classification label ("normal" or "fault"), the classification label indicates whether the feeder is in a single-phase ground fault state; S3.4. If multiple feature means (such as zero-sequence current mean and zero-sequence voltage mean) exceed the threshold at the same time, the confidence of the fault judgment is increased (the confidence reflects the degree of certainty of the algorithm as to whether a specific feeder may have a fault). Consider the changing trend of the feature over a period of time, rather than relying solely on the instantaneous reading. For example, if the zero-sequence current rises sharply in a short period of time, it is more likely to be a fault signal. Compare the current feature with the data during historical normal operation to find abnormal patterns.

[0021] S4. Use the sorting training model to intelligently sort all feeders and determine the line most likely to fail; In this embodiment, the intelligent sorting model can predict which feeders are most likely to fail based on historical data and real-time data, thereby guiding maintenance personnel to check these feeders first. This method greatly reduces the time of blind search and improves the speed and efficiency of troubleshooting. By determining the lines most likely to fail in advance through intelligent sorting, the fault point can be found and isolated in the shortest time, reducing the impact on users, shortening power outage time, and improving the reliability of power supply. By quickly locating the fault through intelligent sorting, the fault can be prevented from spreading and the impact on other healthy feeders can be reduced, thereby enhancing the stability and reliability of the entire power system. The sorting model is trained based on a large amount of historical data and can learn the correlation between different features and faults, thereby improving the accuracy of fault detection. Even weak fault signals may be captured in time. All feeders are intelligently sorted using the sorting training model, including the following steps: S4.1. Divide historical operation data (including normal operation data and fault data) into a training set, a validation set and a test set; S4.2. Initialize the parameters of the sorting training model, define the sorting loss function of the sorting training, introduce the statistics of the features into the sorting loss function, consider the significant changes between the zero-sequence current and zero-sequence voltage features in normal operation and fault state, and then consider the situation where the zero-sequence current and zero-sequence voltage are close to zero, use auxiliary zero-sequence current and zero-sequence voltage prediction to further optimize the sorting loss function, and use the learning update method to optimize the sorting training model parameters; Among them, the sorting loss function is: ; The introduction of feature statistics (such as mean, variance, etc.) is to enable the model to better capture the differences between different feeders in normal operation and fault conditions. Through these statistics, the model can more comprehensively understand the state changes of the feeders, thereby improving the accuracy of fault sorting; feature statistics can provide multi-dimensional information about the operating status of the feeders, helping the model not only rely on the measurement value at a single moment, but also comprehensively consider the change trend over a period of time, making the sorting results more reliable; by introducing feature statistics, the model's ability to resist noise can be enhanced, so that the model can still give more accurate sorting results when facing incomplete or inconsistent data; statistics (such as mean, variance, etc.) reflect the overall performance of the feature over a period of time, rather than just the instantaneous value, which helps the model maintain good performance under different operating conditions and enhances the generalization ability of the model; the feature statistics introduced in the sorting loss function are: ; In the case of close to zero, the traditional detection method may not be able to accurately distinguish the normal state from the minor fault state because the signal is too weak. By introducing auxiliary prediction, the sensitivity of detection can be improved in the case of weak signal. When the zero-sequence current and zero-sequence voltage are close to zero, the conventional detection method may have difficulty in identification, especially for minor faults or early fault signs. Auxiliary prediction can make up for the lack of identification in this case; auxiliary prediction can help the model better identify those feeders that are close to normal operation but have minor faults, thereby improving the accuracy of fault detection, especially for those cases where the signal is weak but there is indeed a fault; by optimizing the sorting loss function to consider the situation where the zero-sequence current and zero-sequence voltage are close to zero, the false alarm (false positive) and missed alarm (false negative) caused by weak signal can be reduced, making fault detection more reliable; considering the zero-sequence current and zero-sequence voltage are close to zero, the sorting loss function is further optimized using auxiliary zero-sequence current and zero-sequence voltage prediction as follows: ; in, Represents the weight vector in the sorting training model; Represents the bias term in the sorting training model; Represents the total number of samples in the data set; Indicates the sample index; Indicates the number of features; Represents the feature index; represents the weight index; Indicates The true label of each sample (0 means no fault, 1 means fault); Indicates The predicted probability of samples; Indicates A vector composed of all features of samples; A weight vector representing the predicted mean; Represents the bias term of the predicted mean; A weight vector representing the prediction variance; A weight vector representing the prediction variance; The model predicts the Zero-sequence current value of samples; Indicates The actual zero-sequence current value of samples; The model predicts the Zero-sequence voltage value of samples; Indicates The actual zero-sequence voltage value of samples; It represents the importance of the prediction error of the control characteristic statistic and is a hyperparameter used to consider the significant changes of zero-sequence current and zero-sequence voltage characteristics between normal operation and fault state; It represents the importance of controlling the prediction error of zero-sequence current and zero-sequence voltage and is a hyperparameter used to consider the sparsity of zero-sequence current and zero-sequence voltage; Indicates the importance of controlling the regularization term and is a hyperparameter; represents weight; Indicates the average value of zero-sequence current; Indicates the average value of zero-sequence voltage; Furthermore, the learning update method updates the model parameters iteratively, gradually reducing the value of the loss function so that the model parameters gradually approach the optimal solution. This method can effectively find the parameter combination that minimizes the loss; the learning update method can be flexibly applied to different types of loss functions and model architectures. Whether it is a simple linear model or a complex nonlinear model, the parameters can be optimized by adjusting the learning rate and other hyperparameters; the implementation of the learning update method is relatively simple, and it can be achieved by adjusting the learning rate and gradient calculation. In addition, its debugging process is relatively direct, and the learning status of the model can be intuitively understood by monitoring the changes in the loss function; the learning update method is: ; ; in, represents the learning rate; Express The gradient of Express The gradient of S4.3. Use the data from the training set to train the sorting training model, evaluate the model performance on the validation set, and adjust the model parameters; S4.4. Use the trained and optimized sorting training model to predict the feeder data in the test set, sort the feeders according to the failure probability, and determine the line most likely to fail.

[0022] S5. According to the result of intelligent sorting, test pull the suspected fault line to verify which line is the real fault line; In this embodiment, according to the result of intelligent sorting, a test pull is performed on the suspected faulty line to verify which line is the real faulty line, including the following steps: S5.1. According to the results of intelligent sorting, formulate a test pulling plan and start the test pulling from the line most likely to fail; S5.2. Disconnect the line most likely to fail in the sorting sequence and monitor whether the zero-sequence current and zero-sequence voltage return to the normal range; S5.3. Record the changes of zero-sequence current and zero-sequence voltage after disconnecting the line; S5.4. If the zero-sequence current and zero-sequence voltage return to normal, the line can be confirmed as a faulty line and the power supply of other lines can be restored; S5.5. If the zero-sequence current and zero-sequence voltage are still abnormal, restore the power supply of this line and continue to test the next line that is most likely to fail; S5.6. According to steps S5.2-S5.5, test the lines in the sorted list in turn until the real faulty line is found.

[0023] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments, and the above embodiments and descriptions are only preferred examples of the present invention, and are not intended to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, and these changes and improvements all fall within the scope of the present invention to be protected.

Claims

1. An intelligent sorting method for single-phase grounding test in a small current grounding system, characterized by: The following steps are involved: S1. Collect real-time operation data of each feeder; S2, extracting features that can reflect changes in system status from real-time operation data; S3, using the extracted features, determining whether a single-phase grounding fault has occurred by setting a threshold recognition rule; S4. Use the sorting training model to perform intelligent sorting on all feeders, determine the line most likely to fail, and in the intelligent sorting process, consider the situation where the zero-sequence current and zero-sequence voltage are close to zero, and use auxiliary zero-sequence current and zero-sequence voltage prediction to further optimize the intelligent sorting process; S5. According to the result of intelligent sorting, test pull the suspected faulty line to verify which line is the real faulty line.

2. The intelligent sorting method for single-phase grounding test in a small current grounding system according to claim 1 is characterized in that: In S1, the real-time operation data includes the current, voltage and frequency of the feeder.

3. The intelligent sorting method for single-phase grounding test in a small current grounding system according to claim 2 is characterized in that: In S2, extracting features that can reflect changes in system status from real-time operation data includes the following steps: S2.1, preprocessing the real-time operation data; S2.2, using fast Fourier transform to convert the time domain signal into frequency domain representation, analyzing the frequency components of the operating data, and analyzing the instantaneous values ​​of the current and voltage of the operating data; S2.

3. Calculate the zero-sequence current and zero-sequence voltage characteristics in the operating data, introduce the correction factor parameters of the feeder impedance, consider the influence of water vapor concentration in the air on the zero-sequence current and zero-sequence voltage, and further optimize the calculation process of the zero-sequence current and zero-sequence voltage; S2.

4. Observe the changing trend of features over time and calculate the statistics of features.

4. The intelligent sorting method for single-phase grounding test in a small current grounding system according to claim 3 is characterized in that: In S2.3, the zero-sequence current and zero-sequence voltage are calculated as: The zero sequence current is: ； The zero sequence voltage is: ； The correction factor parameter introduced into the feeder impedance is: The zero sequence current is: ； The zero sequence voltage is: ； Considering the influence of water vapor concentration in the air on zero-sequence current and zero-sequence voltage, the calculation process of zero-sequence current and zero-sequence voltage is further optimized as follows: The zero sequence current is: ； The zero sequence voltage is: ； in, Indicates the zero-sequence current value; Indicates the zero-sequence voltage value; Indicates the A phase current; Indicates the B phase current; Indicates the C phase current; Indicates the voltage of phase A; Indicates the B phase voltage; Indicates the C phase voltage; represents the feeder impedance correction factor; It represents the impedance of the feeder; Indicates the zero-sequence current value after feeder impedance correction; Indicates the zero-sequence voltage value after feeder impedance correction; Indicates the zero-sequence current value after correction by feeder impedance and water vapor concentration in the air; It indicates the zero-sequence voltage value after correction by feeder impedance and water vapor concentration in the air; Represents the correction factor for water vapor concentration in the air.

5. The intelligent sorting method for single-phase grounding test in a small current grounding system according to claim 4 is characterized in that: In S3, determining whether a single-phase grounding fault has occurred by setting a threshold recognition rule includes the following steps: S3.1, set the threshold value m of the zero-sequence current mean value, and set the threshold value n of the zero-sequence voltage mean value; S3.2, comparing the extracted feature mean with the set threshold, the feature mean including the zero-sequence current mean and the zero-sequence voltage mean; S3.3, using the fault classification tree to classify the extracted feature means and determine whether the feeder is in a fault state; S3.

4. If the mean values ​​of multiple features exceed the threshold at the same time, the confidence in the fault judgment is increased, and the current features are compared with the data during historical normal operation to find abnormal patterns.

6. The intelligent sorting method for single-phase grounding test in a small current grounding system according to claim 5 is characterized in that: In S3.3, the extracted features are classified using a fault classification tree model, including the following steps: S3.

31. Determine the feature set for training and use the fault classification tree to build an algorithm classification model; S3.32, compare the extracted feature mean with a preset threshold value, if the zero-sequence current mean value exceeds the threshold value m, mark the feeder as a suspected fault line; if the zero-sequence voltage mean value exceeds the threshold value n, mark the feeder as a suspected fault line; S3.33, inputting the feature mean data of the line marked as suspected fault into the trained fault classification tree model to make classification decisions; S3.

34. Output a classification label, which indicates whether the feeder is in a single-phase grounding fault state.

7. The intelligent sorting method for single-phase grounding test in a small current grounding system according to claim 6 is characterized in that: In S4, all feeders are intelligently sorted using a sorting training model, including the following steps: S4.

1. Divide the historical operation data into training set, validation set and test set; S4.

2. Initialize the parameters of the sorting training model, define the sorting loss function of the sorting training, introduce the statistics of the features into the sorting loss function, consider the situation that the zero-sequence current and zero-sequence voltage are close to zero, use the auxiliary zero-sequence current and zero-sequence voltage prediction to further optimize the sorting loss function, and use the learning update method to optimize the sorting training model parameters; S4.

3. Use the data from the training set to train the sorting training model, evaluate the model performance on the validation set, and adjust the model parameters; S4.

4. Use the trained and optimized sorting training model to predict the feeder data in the test set, sort the feeders according to the failure probability, and determine the line most likely to fail.

8. The intelligent sorting method for single-phase grounding test in a small current grounding system according to claim 7 is characterized in that: In S4.2, the sorting loss function is: ； The statistics of the features introduced in the ranking loss function are: ； Considering the situation where the zero-sequence current and zero-sequence voltage are close to zero, the sorting loss function is further optimized using auxiliary zero-sequence current and zero-sequence voltage prediction as follows: ； in, Represents the weight vector in the sorting training model; Represents the bias term in the sorting training model; represents the total number of samples in the data set; Indicates the sample index; Indicates the number of features; Represents the feature index; represents the weight index; Indicates The true labels of samples; Indicates The predicted probability of samples; Indicates A vector composed of all features of samples; A weight vector representing the predicted mean; Represents the bias term of the predicted mean; A weight vector representing the prediction variance; A weight vector representing the prediction variance; The model predicts the Zero-sequence current value of samples; Indicates The actual zero-sequence current value of samples; The model predicts the Zero-sequence voltage value of samples; Indicates The actual zero-sequence voltage value of samples; Indicates the importance of controlling the prediction error of the feature statistic; Indicates the importance of controlling zero-sequence current and zero-sequence voltage prediction errors; represents the importance of controlling the regularization term; represents weight; Indicates the average value of zero-sequence current; Represents the average value of zero-sequence voltage.

9. The intelligent sorting method for single-phase grounding test in a small current grounding system according to claim 8 is characterized in that: In S4.2, the learning update method is: ； ； in, represents the learning rate; Express The gradient of Express gradient.

10. The intelligent sorting method for single-phase grounding test in a small current grounding system according to claim 9 is characterized in that: In S5, according to the result of intelligent sorting, the suspected faulty line is tested to verify which line is the real faulty line, including the following steps: S5.

1. According to the results of intelligent sorting, formulate a test pulling plan and start the test pulling from the line most likely to fail; S5.

2. Disconnect the line most likely to fail in the sorting sequence and monitor whether the zero-sequence current and zero-sequence voltage return to the normal range; S5.

3. Record the changes of zero-sequence current and zero-sequence voltage after disconnecting the line; S5.

4. If the zero-sequence current and zero-sequence voltage return to normal, the line can be confirmed as a faulty line and the power supply of other lines can be restored; S5.

5. If the zero-sequence current and zero-sequence voltage are still abnormal, restore the power supply of this line and continue to test the next line that is most likely to fail; S5.

6. According to steps S5.2-S5.5, test the lines in the sorted list in turn until the real faulty line is found.

Images