Low-current grounding system single-phase grounding fault positioning method based on BP neural network
Through the single-phase grounding fault positioning method of a small current grounding system based on BP neural network, the characteristic values of the zero-sequence current signal are extracted and inputted into the neural network model, which solves the problem of insufficient positioning accuracy and real-time in the prior art, and achieves the fault positioning effect with high accuracy and strong real-time.
Patent Information
- Application Number
- CN202411683208.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-22
- Publication Date
- 2025-03-25
AI Technical Summary
The existing single-phase grounding fault positioning method of small current grounding systems has problems with insufficient positioning accuracy and real-time performance, and the neural network structure design is not optimized and the training algorithm convergence speed is slow.
A single-phase grounding fault location method based on a BP neural network is adopted. By collecting zero-sequence current signal data, the fundamental component, fifth harmonic component and IMF energy are extracted, and inputted into the pre-trained BP neural network model to output the judgment results of the fault line, and the specific location of the fault point is determined through further positioning.
It improves the accuracy and real-time nature of fault identification and positioning, is highly adaptable, and can adapt to different types of fault signals without frequent adjustment of parameters.
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of power system fault detection and relates to a single-phase grounding fault locating method for a small current grounding system based on a BP neural network. Background Art
[0002] my country's medium and low voltage distribution networks generally adopt a low current grounding method, that is, the neutral point is not grounded or grounded through an arc suppression coil. When a single-phase grounding fault occurs in a low current grounding system, due to the small fault current, the three-phase line voltage is still symmetrical, which does not affect the continued power supply to users, and is usually allowed to operate with the fault for 1 to 2 hours. However, if not handled in time, it may develop into a phase-to-phase short circuit, causing a larger range of power outages. Therefore, it is of great significance to quickly and accurately locate the single-phase grounding fault of the low current grounding system.
[0003] At present, the main methods for locating single-phase grounding faults in small current grounding systems include traveling wave method, injection method, zero-sequence current amplitude-to-phase method, group amplitude-to-phase method, fifth harmonic method, wavelet analysis method, neural network method, etc. However, these methods all have some limitations in practical applications. For example, the traveling wave method is affected by factors such as traveling wave attenuation, reflection and refraction, and the positioning accuracy is difficult to guarantee; the injection method requires additional signal sources and detection devices, which increases the complexity and cost of the system; the zero-sequence current amplitude-to-phase method and the group amplitude-to-phase method are greatly affected by the accuracy of the current transformer and the unbalanced current; the fifth harmonic method is susceptible to noise interference due to its small fifth harmonic content; although the wavelet analysis method can effectively extract fault characteristics, the selection of wavelet basis and the determination of decomposition scale are somewhat subjective.
[0004] BP neural network is a multi-layer feedforward neural network with powerful nonlinear mapping and self-learning capabilities. In recent years, BP neural network has been widely used in the field of power system fault location. By training BP neural network, the mapping relationship between fault characteristics and fault location can be established to achieve fast and accurate fault location. However, the existing single-phase grounding fault location method of small current grounding system based on BP neural network still has some problems, such as unreasonable selection of fault characteristics, unoptimized neural network structure design, slow convergence speed of training algorithm, etc., resulting in the need to improve positioning accuracy and real-time performance. Summary of the invention
[0005] The present invention proposes a single-phase grounding fault location method for a small current grounding system based on a BP neural network. The method utilizes the BP neural network to fuse and process fault features, thereby improving the accuracy and speed of fault location.
[0006] The technical solution of the present invention is achieved in this way:
[0007] A method for locating single-phase grounding fault in a small current grounding system based on BP neural network comprises the following steps:
[0008] Collect zero-sequence current signal data of small current grounding system;
[0009] Extracting the fundamental component, the fifth harmonic component and the IMF energy of the zero-sequence current signal;
[0010] Input the extracted feature values into the pre-trained BP neural network model;
[0011] Output the judgment result of the fault line through the BP neural network model;
[0012] According to the output results of the BP neural network, the specific location of the fault point is further located.
[0013] Furthermore, the zero-sequence current signal data includes data during normal operation and data during a fault. Furthermore, the step of extracting characteristic values includes:
[0014] The zero-sequence current signal is Fourier transformed to obtain the fundamental component;
[0015] Perform harmonic analysis on the zero-sequence current signal to obtain the fifth harmonic component;
[0016] Perform empirical mode decomposition on the zero-sequence current signal to obtain the IMF energy.
[0017] Furthermore, the training steps of the BP neural network model include:
[0018] Collect a large amount of zero-sequence current signal data and mark whether each piece of data is fault data;
[0019] Divide the dataset into training and testing sets;
[0020] Use the training set to train the BP neural network model and adjust the model parameters to minimize the loss function;
[0021] Use the test set to verify the trained model to ensure the generalization ability of the model.
[0022] Furthermore, the structure of the BP neural network model includes an input layer, a hidden layer and an output layer, wherein:
[0023] The number of nodes in the input layer is equal to the number of eigenvalues;
[0024] The number of nodes in the hidden layer is determined according to actual needs;
[0025] The number of nodes in the output layer is 1, indicating the binary classification result of whether it is a faulty line.
[0026] Furthermore, the activation function of the BP neural network model adopts a sigmoid function or a ReLU function.
[0027] Furthermore, the method further includes the step of optimizing the BP neural network model, specifically including:
[0028] Adjust model parameters using gradient descent or Adam optimizer;
[0029] The best hyperparameters are selected through cross-validation.
[0030] Furthermore, the output result of the BP neural network model includes the number of the fault line and the fault probability. Furthermore, the step of further locating the specific position of the fault point includes:
[0031] Use the fault traveling wave method or impedance method to accurately locate the fault line;
[0032] Calculate the distance between the fault point and the power supply side.
[0033] Furthermore, the method further includes the step of updating the model in real time, specifically including:
[0034] Regularly collect new zero-sequence current signal data;
[0035] Retrain the model with new data to adapt to changes in the system.
[0036] Furthermore, the method further includes the step of providing an early warning for the faulty line, specifically including:
[0037] Set the thresholds of fundamental component, fifth harmonic component and IMF energy;
[0038] When the characteristic value of the zero-sequence current signal exceeds the threshold, the early warning mechanism is triggered to notify relevant personnel to conduct inspection and processing.
[0039] Furthermore, the method further includes the step of analyzing historical data of the fault line, specifically including:
[0040] Record the time, location and characteristic value of each fault;
[0041] Analyze the patterns of failure occurrence and predict future failure risks.
[0042] Furthermore, the method also includes a performance evaluation step for the BP neural network model, specifically including:
[0043] Evaluate the performance of the model using metrics such as precision, recall, and F1 score;
[0044] Regularly evaluate the performance of the model to ensure its effectiveness in practical applications.
[0045] Furthermore, the method also includes steps for enhancing the interpretability of the BP neural network model, specifically including:
[0046] Use methods such as LIME (Local Interpretable Model-agnostic Explanations) or SHAP (SHapley Additive exPlanations) to explain the model's decision-making process; enhance the model's transparency and explainability, and improve user trust
[0047] Furthermore, the method is implemented by a computer program, and the program is stored in a computer-readable storage medium.
[0048] The working principle and beneficial effects of the present invention are:
[0049] The single-phase grounding fault location method of a small current grounding system based on a BP neural network provided by the present invention has the following advantages:
[0050] 1. High accuracy: The accuracy of fault identification and location is improved through feature extraction and BP neural network model training.
[0051] 2. Strong real-time performance: The BP neural network model has fast reasoning speed and can realize real-time fault detection.
[0052] 3. Strong adaptability: BP neural network has self-learning ability and can adapt to different types of fault signals without frequent parameter adjustment. DETAILED DESCRIPTION
[0053] The following will be combined with 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.
[0054] Example 1
[0055] 1. Data Collection
[0056] 1.1 Data Source
[0057] The data comes from the operation records of the small current grounding system, including data during normal operation and data when a fault occurs. The data acquisition equipment should have high precision and high sampling rate to ensure the accuracy and integrity of the data.
[0058] 1.2 Data Format
[0059] The collected data shall include but not be limited to the following information:
[0060] Timestamp: The time when the data was collected.
[0061] Zero-sequence current: the zero-sequence current value of each line.
[0062] Line number: The unique identifier of each line.
[0063] Fault flag: whether a fault occurs (0 means normal, 1 means fault).
[0064] 2. Feature extraction
[0065] 2.1 Fundamental component extraction
[0066] The fundamental component is the main component of the zero-sequence current signal, reflecting the normal operation status of the system. The method for extracting the fundamental component is as follows:
[0067] Fourier transform: Perform Fourier transform on the zero-sequence current signal to obtain the frequency domain signal.
[0068] Fundamental component: Extract the fundamental component (frequency is 50Hz or 60Hz) in the frequency domain signal.
[0069] 2.2 Extraction of the fifth harmonic component
[0070] The fifth harmonic component is a high-order harmonic component in the zero-sequence current signal, reflecting the nonlinear characteristics of the system. The method for extracting the fifth harmonic component is as follows:
[0071] Harmonic analysis: Perform harmonic analysis on the zero-sequence current signal to obtain each harmonic component.
[0072] Fifth harmonic component: Extract the fifth harmonic component (frequency is 250Hz or 300Hz) in the frequency domain signal.
[0073] 2.3 IMF Energy Extraction
[0074] IMF energy reflects the instantaneous characteristics of the zero-sequence current signal and helps capture subtle changes in the fault signal. The method for extracting IMF energy is as follows:
[0075] Empirical Mode Decomposition (EMD): Perform EMD decomposition on the zero-sequence current signal to obtain multiple intrinsic mode functions (IMFs).
[0076] IMF Energy: Calculate the energy of each IMF as an eigenvalue.
[0077] 3. Model training
[0078] 3.1BP Neural Network Structure
[0079] BP neural network is a multi-layer feedforward neural network that adjusts weights through the back propagation algorithm to achieve nonlinear mapping of input data. The BP neural network structure used in this method is as follows:
[0080] Input layer: The number of nodes is equal to the number of eigenvalues (fundamental component, fifth harmonic component, and IMF energy).
[0081] Hidden layer: The number of nodes is determined according to actual needs and is generally set to 1.5 times the number of nodes in the input layer.
[0082] Output layer: The number of nodes is 1, indicating the binary classification result of whether it is a faulty line.
[0083] Activation function: Select the sigmoid function or the ReLU function as the activation function for the hidden layer and the output layer.
[0084] 3.2 Training Steps
[0085] Data preparation: Collect a large amount of zero-sequence current signal data and mark whether each piece of data is fault data.
[0086] Data partitioning: Divide the data set into a training set and a test set, generally in a ratio of 7:3.
[0087] Model initialization: Randomly initialize the weights and biases of the BP neural network.
[0088] Forward propagation: Input the feature values of the training set into the BP neural network and calculate the output value.
[0089] Error calculation: Calculate the error between the output value and the true label.
[0090] Back propagation: Adjust the weights and biases of the BP neural network according to the error to minimize the loss function.
[0091] Optimizer selection: Use gradient descent or Adam optimizer to adjust model parameters.
[0092] Model validation: Use the test set to validate the trained model to ensure the generalization ability of the model.
[0093] Hyperparameter selection: Select the best hyperparameters such as learning rate, batch size, etc. through cross-validation.
[0094] 4. Fault diagnosis
[0095] 4.1 Judgment steps
[0096] Data input: Input the characteristic value of the zero-sequence current signal to be detected into the trained BP neural network model.
[0097] Model output: The BP neural network model outputs the judgment result of the faulty line, including the number of the faulty line and the fault probability.
[0098] Result interpretation: Based on the output results, determine whether there is a faulty line and its fault probability.
[0099] 5. Fault location
[0100] 5.1 Positioning steps
[0101] Fault line confirmation: Confirm the fault line based on the output results of the BP neural network model.
[0102] Fault point location: Use the fault traveling wave method or impedance method to accurately locate the fault line and calculate the distance between the fault point and the power supply side.
[0103] 6. Early Warning
[0104] 6.1 Early Warning Mechanism
[0105] Threshold setting: Set the thresholds for the fundamental component, fifth harmonic component and IMF energy.
[0106] Characteristic value monitoring: real-time monitoring of the characteristic value of the zero-sequence current signal.
[0107] Early warning trigger: When the characteristic value of the zero-sequence current signal exceeds the threshold, the early warning mechanism is triggered to notify relevant personnel to conduct inspection and processing.
[0108] 7. Historical data analysis
[0109] 7.1 Data Recording
[0110] Fault record: record the time, location and characteristic values of each fault.
[0111] Data storage: Store fault records in a database for subsequent analysis.
[0112] 7.2 Regularity Analysis
[0113] Data statistics: Conduct statistical analysis on fault records to find out the patterns of fault occurrence.
[0114] Trend prediction: Based on historical data, predict the risk of future failures.
[0115] 8. Model maintenance and update
[0116] Data collection: Regularly collect new zero-sequence current signal data to ensure the timeliness and accuracy of the data.
[0117] Data preprocessing: Preprocess the newly collected data, including cleaning, standardization and other operations.
[0118] Model retraining: Use new data to retrain the existing BP neural network model to adapt to system changes and new data distribution.
[0119] Model evaluation: Evaluate the retrained model to ensure that its performance and generalization ability still meet the requirements.
[0120] Model deployment: Redeploy the updated model to the actual application, replacing the old model.
[0121] 9. Performance Evaluation
[0122] 9.1 Evaluation Metrics
[0123] Accuracy: The proportion of faulty lines that the model correctly identifies.
[0124] Recall: The proportion of all faulty lines that the model correctly identifies.
[0125] F1 score: An evaluation indicator that comprehensively considers precision and recall.
[0126] 9.2 Evaluation Steps
[0127] Data preparation: Use an independent validation set for performance evaluation.
[0128] Model testing: Input the feature values of the validation set into the model and calculate the output results.
[0129] Indicator calculation: Calculate the accuracy, recall and F1 score based on the output results and the true labels.
[0130] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.
Claims
1. A method for locating single-phase grounding fault in a small current grounding system based on BP neural network, characterized in that: The steps include: Collect zero-sequence current signal data of small current grounding system; Extracting the fundamental component, the fifth harmonic component and the IMF energy of the zero-sequence current signal; Input the extracted feature values into the pre-trained BP neural network model; Output the judgment result of the fault line through the BP neural network model; According to the output results of the BP neural network, the specific location of the fault point is located.
2. According to the method for locating single-phase grounding fault in a small current grounding system based on BP neural network in claim 1, it is characterized in that: The zero-sequence current signal data includes data during normal operation and data during a fault.
3. The method for locating single-phase grounding fault in a small current grounding system based on BP neural network according to claim 1 is characterized in that: The steps to extract eigenvalues include: The zero-sequence current signal is Fourier transformed to obtain the fundamental component; Perform harmonic analysis on the zero-sequence current signal to obtain the fifth harmonic component; Perform empirical mode decomposition on the zero-sequence current signal to obtain the IMF energy.
4. The method for locating single-phase grounding fault in a small current grounding system based on BP neural network according to claim 1 is characterized in that: The training steps of the BP neural network model include: Collect a large amount of zero-sequence current signal data and mark whether each piece of data is fault data; Divide the dataset into training and testing sets; Use the training set to train the BP neural network model and adjust the model parameters to minimize the loss function; Use the test set to verify the trained model to ensure the generalization ability of the model.
5. The method for locating single-phase grounding fault in a small current grounding system based on BP neural network according to claim 1 is characterized in that: The structure of the BP neural network model includes an input layer, a hidden layer, and an output layer, where: The number of nodes in the input layer is equal to the number of eigenvalues; The number of nodes in the hidden layer is determined according to actual needs; The number of nodes in the output layer is 1, indicating the binary classification result of whether it is a faulty line.
6. The method for locating single-phase grounding fault in a small current grounding system based on BP neural network according to claim 1 is characterized in that: The activation function of the BP neural network model uses the sigmoid function or the ReLU function.
7. The method for locating single-phase grounding fault in a small current grounding system based on BP neural network according to claim 1 is characterized in that: It also includes the steps of optimizing the BP neural network model, specifically including: Adjust model parameters using gradient descent or Adam optimizer; The best hyperparameters are selected through cross-validation.
8. The method for locating single-phase grounding fault in a small current grounding system based on BP neural network according to claim 1 is characterized in that: The output result of the BP neural network model includes the number of the fault line and the fault probability.
9. The method for locating single-phase grounding fault in a small current grounding system based on BP neural network according to claim 1 is characterized in that: The step of further locating the specific position of the fault point includes: Use the fault traveling wave method or impedance method to accurately locate the fault line; Calculate the distance between the fault point and the power supply side.
10. The method for locating single-phase grounding fault in a small current grounding system based on BP neural network according to claim 1, characterized in that: It also includes steps for real-time updating of the model, including: Regularly collect new zero-sequence current signal data; Retrain the model with new data to adapt to changes in the system.