Hybrid power transmission line fault type identification method based on traveling wave diagnosis and machine vision detection
By using machine vision detection and YOLOv8 object detection algorithm in hybrid transmission lines, the nonlinear processing difficulties and computational complexity problems of fault diagnosis in the prior art are solved, and the accurate identification and diagnosis of transmission line faults is achieved, and the system is improved reliability and stability.
Patent Information
- Application Number
- CN202411101502.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-12
- Publication Date
- 2024-12-03
AI Technical Summary
The existing Kalman filtering, adaptive algorithms and genetic algorithms have difficulty in processing nonlinear systems, challenges in the fault diagnosis of hybrid transmission lines, dependence on initial state estimation, subjectivity of parameter selection, high computational complexity, and local Problems such as optimal solution and slow convergence speed lead to insufficient diagnostic accuracy and efficiency.
Using a hybrid transmission line fault type identification method based on traveling wave diagnosis and machine vision detection, the traveling wave signal is monitored in real time through sensors installed on the transmission line, and using machine vision technology and YOLOv8 object detection algorithm to automatically analyze the traveling wave signal characteristics and identify different Type of failure.
It realizes accurate identification and diagnosis of hybrid transmission line faults, improves the level of automation and intelligence of diagnosis, enhances the reliability and stability of the system, provides intelligent decision-making support, and reduces human error and calculation complexity.
Smart Images

Figure CN119064712A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of transmission line fault identification, specifically a hybrid transmission line fault type identification method based on traveling wave diagnosis and machine vision detection. Background Technology
[0002] In the field of fault diagnosis for hybrid transmission lines, Kalman filtering, adaptive algorithms, and genetic algorithms are currently the main methods used to improve fault diagnosis. In laboratory research, genetic algorithms have also been used to optimize fault diagnosis models and parameter selection. These methods have some potential in improving the accuracy and efficiency of fault diagnosis for hybrid transmission lines, but they still have significant shortcomings.
[0003] Kalman filtering is a state-estimation-based filtering algorithm used to estimate system states and provide optimal estimates. However, in traveling wave diagnostics, the nonlinear and uncertain characteristics of power systems limit the application of Kalman filtering. Complex nonlinear relationships exist in power systems, causing Kalman filtering to fail to accurately estimate state variables, thus affecting the accuracy of the diagnostics. Specifically:
[0004] 1) Difficulty in handling nonlinear systems: Kalman filtering is a filtering algorithm based on the assumption of linear systems, and its ability to handle nonlinear systems is limited. In power systems, there are many nonlinear characteristics, such as electromagnetic saturation and electrical discharge. These characteristics make it impossible for Kalman filtering to accurately model and estimate system state variables, thus affecting the accuracy of traveling wave diagnosis.
[0005] 2) Challenges in handling non-Gaussian noise: Kalman filtering assumes that system noise is Gaussian; however, noise in power systems typically exhibits non-Gaussian distribution characteristics. For example, power systems may contain non-Gaussian noise sources such as impulse noise and spike noise, which adversely affect the performance of Kalman filtering. Kalman filtering cannot effectively handle non-Gaussian noise, potentially leading to distorted estimation results.
[0006] 3) Dependence on Initial State Estimation: Kalman filtering requires accurate initial state estimation as input for filtering and estimation processes. However, in practical applications, the initial state is often difficult to estimate accurately, especially under unknown fault conditions. Inaccuracy in initial state estimation gradually accumulates and leads to deviations in the filtering results, thus affecting the accuracy of fault diagnosis.
[0007] 4) Subjectivity in parameter selection: The selection of model parameters in Kalman filtering is crucial to the filtering performance. However, parameter selection is often based on experience and subjective judgment. For complex power systems, the parameter selection may not be precise enough, leading to inaccurate filtering results.
[0008] Adaptive algorithms are algorithms that can automatically adjust parameters or models based on real-time data and environmental changes. In fault diagnosis of hybrid transmission lines, adaptive algorithms can be used to improve the accuracy and adaptability of diagnosis to cope with the nonlinearity, uncertainty, and changes in the system. However, the following drawbacks often occur when applying adaptive algorithms in hybrid transmission lines:
[0009] 1) Complex Network Topology: Hybrid transmission lines have a more complex network topology, including high-voltage transmission lines, numerous substations, and various key nodes and branches. Adaptive algorithms need to consider these complex topological relationships when processing hybrid transmission line data to ensure the algorithm's adaptability and accuracy. However, this network topology complexity may increase the computational complexity of the algorithm.
[0010] 2) Large amounts of real-time data: Hybrid transmission line monitoring systems can provide a large amount of real-time data, including current, voltage, power, and frequency. Adaptive algorithms need to process this data and make real-time parameter adjustments. However, the large amount of real-time data may increase the computational burden on the algorithm and increase the demand for data processing and storage.
[0011] 3) Data quality and reliability: Data in hybrid transmission lines may be affected by noise, measurement errors, and communication failures. Adaptive algorithms have high requirements for data quality and reliability, as inaccurate or unreliable data may lead to incorrect algorithm adjustments and instability.
[0012] 4) Diversity of Fault Types: Hybrid transmission lines may face various types of faults, such as short circuits, grounding faults, open circuits, and equipment overloads. Different types of faults may have different characteristics and impacts, requiring adaptive algorithms to adapt to and identify these different types of faults. However, the diversity of different fault types may increase the complexity and difficulty of the algorithm.
[0013] 5) System Dynamism: Load and network conditions in hybrid transmission lines may change over time. Adaptive algorithms need to be able to adapt to these dynamic changes and adjust parameters promptly to maintain accuracy. However, dynamic system changes may lead to instability in the algorithm's adjustment process or require more frequent parameter updates.
[0014] 6) Complexity of parameter tuning: Adaptive algorithms involve adjusting and optimizing parameters to adapt the algorithm to the actual situation. However, parameter tuning can be challenging in hybrid transmission lines because parameter selection can be influenced by multiple factors, including system topology, load characteristics, and fault types.
[0015] Genetic algorithms are heuristic optimization algorithms inspired by the concepts of heredity and natural selection in biological evolution theory. They simulate the process of biological evolution by performing operations such as selection, crossover, and mutation on individuals in the potential solution space to progressively search for the optimal solution. Here are some common drawbacks of genetic algorithms:
[0016] 1) Initialization Problem: In the fault diagnosis problem of hybrid transmission lines, the performance of genetic algorithms largely depends on the selection of the initial population. Inappropriate initialization may cause the algorithm to get stuck in local optima or require a longer time to find the global optimum. Given the complex topology and diverse fault types of hybrid transmission lines, choosing an appropriate initialization strategy is particularly important.
[0017] 2) High computational complexity: Genetic algorithms involve a large number of crossover and mutation operations between individuals and chromosomes, which increases the computational complexity of the algorithm. Especially when applied to large-scale hybrid transmission line problems, the computation time of the algorithm may increase significantly.
[0018] 3) Local Optimum: Genetic algorithms are prone to getting stuck in local optima during the search process. Since genetic algorithms are heuristic search methods, they explore the solution space through selection and crossover / mutation operations. However, this exploration process may cause the algorithm to stagnate near a local optimum, failing to find a better global optimum.
[0019] 4) Slow convergence speed: Genetic algorithms typically require a long time to converge to the optimal solution. Because the algorithm uses an iterative approach, each generation requires selection, crossover, and mutation operations. This iterative process can consume significant computational resources and time, especially in complex hybrid transmission line fault diagnosis problems.
[0020] 5) Difficulty in parameter selection: Genetic algorithms involve many parameters that require adjustment, such as population size, crossover probability, and mutation probability. Parameter selection significantly impacts the algorithm's performance and convergence, but determining suitable parameter values is challenging. In fault diagnosis of hybrid transmission lines, parameter selection can be even more complex because factors such as system topology, load characteristics, and fault type must be considered.
[0021] Traveling wave (TW) inspection is a method for fault diagnosis in hybrid transmission lines. Based on waveform analysis, it aims to determine the type of fault occurring in a hybrid power transmission line system. The propagation and reflection of the traveling wave signal within the system are affected by the fault type, thus providing information about the fault location and nature. However, current TW methods have some limitations. They typically require manual operation and analysis by specialized personnel, which is time-consuming and requires specialized knowledge. Secondly, the application of traditional methods in large-scale hybrid transmission lines is somewhat limited because manually inspecting and analyzing the entire system is difficult and uneconomical. There may also be limitations in diagnosing complex fault types, as manual analysis can be affected by subjective factors and human error. Summary of the Invention
[0022] This invention provides a hybrid transmission line fault type identification method based on traveling wave diagnosis and machine vision detection. This method is based on traveling wave input and machine vision detection. By using sensors installed at various locations, the propagation and reflection of traveling wave signals can be monitored in real time. Through machine vision technology and specially designed algorithms, the characteristics and changes of traveling wave signals can be automatically analyzed, thereby accurately identifying different types of faults.
[0023] A hybrid transmission line fault type identification method based on traveling wave diagnosis and machine vision detection includes the following steps:
[0024] Step 1: Obtain traveling wave data of the three-phase circuits A, B, and C from the hybrid transmission line system, and convert the traveling wave data into corresponding waveform images;
[0025] Step 2: Train the machine vision detector using known fault sample data to obtain the trained machine vision detector;
[0026] Step 3: Input the waveform image obtained in Step 1 into the machine vision detector trained in Step 2. The machine vision detector identifies the fault type in the waveform image and outputs the corresponding fault label.
[0027] Step 4: Diagnose the fault type based on the fault labels output by the machine vision detector;
[0028] Step 5: Output the fault type diagnosis results to the corresponding display interface or device to provide data support for intelligent decision-making in the diagnosis of hybrid transmission lines.
[0029] Furthermore, step 1 is performed using a data acquisition and processing device, which includes:
[0030] Sensor array: Installed at key locations on hybrid transmission lines to monitor the current and voltage data of the three-phase circuits A, B, and C in real time;
[0031] Data acquisition module: used to read current and voltage data from the sensor array and convert the current and voltage data into digital signals for processing;
[0032] Signal processor: Used to preprocess the acquired digital signals and convert the raw data into waveform data;
[0033] Waveform image generation module: This module converts the processed traveling wave data into corresponding waveform images. When generating waveform images, the horizontal axis is classified according to time or the order of data sampling points, and the vertical axis is classified according to the magnitude of current and voltage values.
[0034] Furthermore, the preprocessing of the acquired digital signals specifically includes:
[0035] 1) Noise removal: Using filtering techniques to remove noise from the data;
[0036] 2) Filtering: High-frequency noise is eliminated using a low-pass filter, as shown in the following formula:
[0037]
[0038] Where h[k] are the filter coefficients and M is the order of the filter;
[0039] 3) Normalization: The traveling wave data is normalized to a uniform range, specifically using minimum-maximum normalization, as shown in the following formula:
[0040]
[0041] Where x′ is the normalized data, x min and x max These are the minimum and maximum values of the data, respectively.
[0042] Furthermore, the filtering techniques include mean filtering, median filtering, and wavelet denoising, as detailed below:
[0043] Mean filtering: Smooths data by applying a window average to the traveling wave signal, as shown in the following formula:
[0044]
[0045] Where y[n] is the filtered signal, x[n] is the original signal, and N is the window size.
[0046] Median filtering: This method filters out sharp noise by selecting the median value of the data within a window. The formula is as follows:
[0047] y[n]=median(x[nk],...,x[n],...,x[n+k]) (2)
[0048] Here, median represents the median operation.
[0049] Wavelet denoising: This method uses wavelet transform to decompose the signal into different frequency levels, and then performs thresholding on the high-frequency components to remove noise. The formula is as follows:
[0050]
[0051] Among them, W x (t) are wavelet transform coefficients, ψ is the wavelet function, and s is the scaling factor.
[0052] Furthermore, step 2 specifically includes:
[0053] Initialize the parameters of the YOLOv8 model using random initialization or pre-trained weights;
[0054] Input the waveform image into the model and calculate the feature map of the input waveform for each layer:
[0055] Z (l) =f(W (l) ·Z (l-1) +b (l) ) (6)
[0056] Among them, Z (l) It is the output of the l-th layer, W (l) and b (l) These are the weights and biases of the l-th layer, respectively, and f is the activation function;
[0057] The total loss of the model is calculated using a loss function, which includes classification loss, coordinate loss, and confidence loss.
[0058] L = L cls +λ coord L coord +λ conf L conf (7)
[0059] Among them, L cls It is classification loss, L coord It is coordinate loss, L conf It is the confidence loss, λ coord and λ conf These are weighting coefficients;
[0060] The gradient of the loss function with respect to the model parameters is calculated using the backpropagation algorithm, and the model parameters are then updated.
[0061]
[0062] Where η is the learning rate;
[0063] Based on the performance on the validation set, the hyperparameters of the model are adjusted, including the learning rate and batch size.
[0064] Regularly evaluate the model's performance on the validation set, analyze the model's error detection results on the validation set, identify the causes of the errors, and improve the model's feature extraction layer or adjust the model architecture based on the causes of the errors.
[0065] This invention provides data support for intelligent decision-making in hybrid transmission line systems by diagnosing fault types and outputting diagnostic results, thereby improving the reliability and stability of these systems. The proposed method is expected to automate and intelligently diagnose faults, offering an efficient and reliable solution for the maintenance and management of hybrid power transmission line systems. Attached Figure Description
[0066] Figure 1 This is a flowchart illustrating the hybrid transmission line fault type identification method based on traveling wave diagnosis and machine vision detection of the present invention.
[0067] Figure 2 This is a schematic diagram of the data acquisition and processing device of the present invention;
[0068] Figure 3 This is a schematic diagram of a noisy image;
[0069] Figure 4 This is a diagram illustrating the denoising process.
[0070] Figure 5 This is a diagram of the YOLOv8 network structure;
[0071] Figure 6 This is a schematic diagram of the forward propagation layer of the C2f module;
[0072] Figure 7 This is a diagram of the computation process of the YOLOv8 network. Detailed Implementation
[0073] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0074] Please see Figure 1This invention provides a hybrid transmission line fault type identification method based on traveling wave diagnosis and machine vision detection. By acquiring traveling wave data and converting it into waveform images, and combining machine vision detector training and target detection, it achieves accurate identification and classification of different fault types.
[0075] The present invention specifically includes the following steps:
[0076] Step 1: Traveling Wave Input. Traveling wave data for the three phases A, B, and C of the hybrid transmission line system are acquired and converted into corresponding waveform images. Each phase's waveform image is represented using a different color group for subsequent machine vision inspection.
[0077] To achieve traveling wave input, this embodiment of the invention designs an intelligent data acquisition and processing device, such as... Figure 2 As shown, it includes the following main components:
[0078] 1) Sensor Array: Sensor arrays installed at key locations on hybrid transmission lines are used to monitor the current and voltage data of the three-phase circuits (A, B, and C) in real time. The selected sensors should possess high precision, high sensitivity, and anti-interference capabilities to ensure accurate acquisition of traveling wave data.
[0079] 2) Data Acquisition Module: This module is responsible for reading current and voltage data from the sensor array and converting them into digital signals for processing. It features high-speed data acquisition and real-time transmission capabilities to ensure rapid response to the status of power transmission lines.
[0080] 3) Signal Processor: This processor performs preprocessing on the acquired digital signals, including denoising, filtering, and normalization, to improve data quality and accuracy. It also converts the raw data into waveform data and performs coordinate-based classification for subsequent machine vision inspection.
[0081] 4) Waveform Image Generation Module: This module converts the processed traveling wave data into corresponding waveform images. When generating waveform images, the horizontal axis is categorized according to time or the order of data sampling points, and the vertical axis is categorized according to the magnitude of current and voltage values to ensure a consistent arrangement and effective visualization of the waveforms.
[0082] Through the above devices and steps, accurate acquisition and processing of traveling wave data of transmission lines can be achieved, generating corresponding waveform images, which provides a reliable data foundation for subsequent fault diagnosis.
[0083] The preprocessing of the acquired digital signals, including denoising, filtering, and normalization, is as follows:
[0084] 1) Noise Reduction: Noise is removed from the data using filtering techniques. Common filtering methods include: Mean Filtering: This smooths the data by applying a window average to the traveling wave signal, as shown in the following formula:
[0085]
[0086] Where y[n] is the filtered signal, x[n] is the original signal, and N is the window size.
[0087] Median filtering: This method filters out sharp noise by selecting the median value of the data within a window. The formula is as follows:
[0088] y[n]=median(x[nk],...,x[n],...,x[n+k]) (2)
[0089] Here, median represents the median operation.
[0090] Wavelet denoising: This method uses wavelet transform to decompose the signal into different frequency levels, and then performs thresholding on the high-frequency components to remove noise. The formula is as follows:
[0091]
[0092] Among them, W x (t) are wavelet transform coefficients, ψ is the wavelet function, and s is the scaling factor.
[0093] The images before and after denoising are shown below. Figure 3 and Figure 4 As shown.
[0094] 2) Filtering: High-frequency noise is eliminated using a low-pass filter, as shown in the following formula:
[0095]
[0096] Where h[k] are the filter coefficients and M is the order of the filter.
[0097] 3) Normalization: The traveling wave data is normalized to a uniform range, usually using min-max normalization, as shown in the following formula:
[0098]
[0099] Where x′ is the normalized data, x min and x max These are the minimum and maximum values of the data, respectively.
[0100] Step 2: Machine Vision Detector Training. The machine vision detector is trained using known fault sample data to detect and classify different fault types. This fault sample data serves as the training dataset, which is generated by the mainnet diagnostic and maintenance personnel during the diagnostic process by labeling and organizing the fault data. The training process can employ the YOLOv8 object detection algorithm or other suitable algorithms. Step 2 specifically includes (e.g., Figure 7 As shown):
[0101] Step 2.1: Model Selection and Configuration
[0102] Traditional waveform monitoring methods typically employ techniques such as Kalman filtering and wavelet transform. However, these methods have limitations in handling complex scenarios. This invention proposes a method based on YOLOv8 (e.g.,...). Figure 5 The waveform monitoring method of target detection technology (shown) has the following advantages:
[0103] Comprehensive Utilization of Multiple Waveform Information for Monitoring: This invention employs YOLOv8 target detection technology, enabling comprehensive monitoring of waveforms using multiple waveform information sources. Compared to traditional methods, this AI-based monitoring technology provides richer and more comprehensive waveform state information, thereby improving the accuracy and reliability of monitoring.
[0104] Lightweight Neural Network Architecture: This invention designs a lightweight neural network architecture to improve the model's inference speed. This enables the monitoring system to achieve real-time monitoring on resource-constrained embedded devices. Simultaneously, methods such as model compression and quantization are employed to further reduce the model's computational complexity, meeting the requirements of real-time monitoring.
[0105] Loss function design adapted to complex scenarios: This invention designs a loss function more suitable for extracting waveform information, adapting to waveform monitoring tasks in complex scenarios. This allows for more accurate monitoring of state changes, improving the robustness and reliability of the monitoring system.
[0106] like Figure 6 As shown, the C2f module is a key component of YOLOv8, and its design is mainly inspired by the C3 module and the ideas of ELAN (Lightweight Attention Network). The main goal of the C2f module is to provide richer gradient flow information while keeping the model lightweight, thereby enhancing the model's performance in the field of waveform diagnostics.
[0107] Step 2.2: Training and Optimization
[0108] Model initialization: Initialize the parameters of the YOLOv8 model. Model parameters can be initialized randomly or using pre-trained weights.
[0109] Forward propagation: Input the waveform image into the model and calculate the feature map of the input waveform for each layer.
[0110] Z (l) =f(W (l) ·Z (l-1) +b (l) ) (6)
[0111] Among them, Z (l) It is the output of the l-th layer, W (l) and b (l) ...
[0112] Loss calculation: The total loss of the model is calculated using the loss function, including classification loss, coordinate loss, and confidence loss.
[0113] L = L cls +λ coord L coord +λ conf L conf (7)
[0114] Among them, L cls It is classification loss, L coord It is coordinate loss, L conf It is the confidence loss, λ coord and λ conf It is the weighting coefficient.
[0115] Backpropagation: The backpropagation algorithm is used to calculate the gradient of the loss function with respect to the model parameters and update the model parameters.
[0116]
[0117] Where η is the learning rate.
[0118] Hyperparameter tuning: Adjust the model's hyperparameters, such as learning rate and batch size, based on the performance on the validation set to achieve better training results.
[0119] Step 2.3: Model Evaluation
[0120] During training, periodically evaluate the model's performance on the validation set. This helps understand the model's generalization ability and prevents overfitting. Analyze the model's error detection results on the validation set to identify its shortcomings and make targeted improvements.
[0121] 1) Type Analysis: Statistically analyze the specific types of error samples (e.g., false positives, false negatives) to understand which fault types are prone to misjudgment or false negatives. Analyze the characteristics of error samples to identify the causes of errors, such as signal noise or unclear fault characteristics.
[0122] 2) Improvement: Based on the error analysis results, improve the model's feature extraction layer or adjust the model architecture. For example, increase the number of convolutional layers to enhance feature extraction capabilities, or introduce an attention mechanism to focus on important features. Further enrich the training data through data augmentation (such as rotation, scaling, cropping, etc.) to improve the model's robustness.
[0123] Based on the performance on the validation set, adjust the model's hyperparameters, such as the learning rate and batch size, to achieve better training results.
[0124] The learning rate scheduling strategy (cosine annealing) is used to dynamically adjust the learning rate to improve training effectiveness.
[0125]
[0126] Where, η t η is the learning rate in round t, η0 is the initial learning rate, and T is the total number of training rounds.
[0127] Regularization techniques (such as L2 regularization and Dropout) are used to prevent overfitting and improve the model's generalization ability.
[0128]
[0129] Where λ is the regularization coefficient.
[0130] To further enrich the training data, data augmentation methods such as image rotation, scaling, cropping, and flipping are used to improve the robustness of the model.
[0131] x aug =augment(x) (11)
[0132] Where, x aug This is the enhanced image.
[0133] Step 3: Object Detection. Input the waveform image obtained in Step 1 into the machine vision detector trained in Step 2 for object detection. The machine vision detector will identify the fault type in the waveform image and output the corresponding fault label.
[0134] Specifically, the preprocessed waveform images are input into the trained YOLOv8 model. The YOLOv8 model analyzes the features in the waveform images, such as the shape, size, and location of the fault, and classifies the fault into predefined categories, such as short circuits and ground faults. The system assesses the scope and severity of the fault's impact and provides intelligent decision support to the operations and maintenance team, such as prioritization and repair strategy recommendations.
[0135] Step 4: Fault Type Diagnosis. Based on the fault labels output by the machine vision detector, fault type diagnosis is performed. A fault type database can be established, mapping fault labels to specific fault types to achieve accurate fault type diagnosis. Simultaneously, the diagnostic results can be used as data samples for retraining to continuously improve and optimize the system's accuracy.
[0136] Step 5: Output Diagnostic Results. Output the fault type diagnostic results to the corresponding display interface or device, providing data support for intelligent decision-making in hybrid transmission line diagnosis. This allows the system to promptly detect faults and provide accurate fault type information, facilitating rapid response and repair, and improving the reliability and stability of the hybrid transmission line system.
[0137] After receiving feedback from the operations and maintenance team, the target detection and fault diagnosis results were added to the fault sample library. The YOLOv8 model was retrained using the enhanced dataset to improve its accuracy and generalization ability. During retraining, the model's parameters and structure were continuously optimized to adapt to fault detection needs in different environments.
[0138] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.
Claims
1. A hybrid transmission line fault type identification method based on traveling wave diagnosis and machine vision detection, It is characterized in that The steps include: Step 1: Obtain the traveling wave data of the three-phase circuits A, B, and C from the hybrid transmission line system, and convert the traveling wave data into corresponding waveform images; Step 2: Use known fault sample data to train the machine vision detector to obtain a trained machine vision detector; Step 3: Input the waveform image processed in step 1 into the machine vision detector trained in step 2. The machine vision detector identifies the fault type in the waveform image and outputs the corresponding fault label. Step 4: Diagnose the fault type according to the fault label output by the machine vision detector; Step 5: Output the fault type diagnosis results to the corresponding display interface or device to provide data support for intelligent decision-making in hybrid transmission line diagnosis.
2. The hybrid power transmission line fault type identification method based on traveling wave diagnosis and machine vision detection as claimed in claim 1, Features: Step 1 is performed using a data acquisition and processing device, the data acquisition and processing device comprising: Sensor array: installed at key locations of hybrid transmission lines to monitor the current and voltage data of the A, B, and C three-phase circuits in real time; Data acquisition module: used to read current and voltage data from the sensor array and convert the current and voltage data into digital signals for processing; Signal processor: used to pre-process the collected digital signals and convert the raw data into waveform data; Waveform image generation module: used to convert the processed traveling wave data into the corresponding waveform image. When generating the waveform image, the horizontal axis is classified according to the order of time or data sampling points, and the vertical axis is classified according to the size of the current and voltage values.
3. The hybrid power transmission line fault type identification method based on traveling wave diagnosis and machine vision detection as claimed in claim 2, Features: The preprocessing of the collected digital signal specifically includes: 1) Denoising: Use filtering technology to remove noise from data; 2) Filtering: Eliminate high-frequency noise through a low-pass filter. The formula is as follows: Where h[k] is the filter coefficient and M is the filter order; 3) Normalization: Normalize the traveling wave data to a uniform range, specifically using minimum-maximum normalization, the formula is as follows: Among them, x′ is the normalized data, x min and x max are the minimum and maximum values of the data respectively.
4. The hybrid power transmission line fault type identification method based on traveling wave diagnosis and machine vision detection as claimed in claim 3, Features: The filtering techniques include mean filtering, median filtering and wavelet denoising, as follows: Mean filtering: Smooth the data by window averaging the traveling wave signal. The formula is as follows: Where y[n] is the filtered signal, x[n] is the original signal, and N is the window size. Median filtering: Filter out sharp noise by selecting the median of the data in the window. The formula is as follows: y[n]=median(x[nk],...,x[n],...,x[n+k]) (2) Among them, median represents the median operation. Wavelet denoising: Use wavelet transform to decompose the signal into different frequency levels, and perform threshold processing on the high-frequency components to remove noise. The formula is as follows: Among them, W x (t) is the wavelet transform coefficient, ψ is the wavelet function, and s is the scale factor.
5. The hybrid transmission line fault type identification method based on traveling wave diagnosis and machine vision detection as claimed in claim 1, Features: Step 2 specifically includes: Initialize the parameters of the YOLOv8 model. The model parameters are randomly initialized or pre-trained weights. Input the input waveform image into the model and calculate the input feature map of each layer of waveform: Z (l) =f(W (l) ·Z (l-1) +b (l) ) (6) Among them, Z (l) is the output of layer l, W (l) and b (l) are the weight and bias of the lth layer respectively, and f is the activation function; The total loss of the model is calculated through the loss function, including classification loss, coordinate loss and confidence loss: L=L cls +λ coord L coord +λ conf L conf (7) Among them, L cls is the classification loss, L coord is the coordinate loss, L conf is the confidence loss, λ coord and λ conf is the weight coefficient; The gradient of the loss function with respect to the model parameters is calculated through the back-propagation algorithm, and the model parameters are updated: Where η is the learning rate; According to the performance on the validation set, adjust the hyperparameters of the model, wherein the hyperparameters of the model include a learning rate and a batch size; Regularly evaluate the performance of the model on the validation set, analyze the model's error detection results on the validation set, find out the causes of the errors, and improve the model's feature extraction layer or adjust the model architecture based on the causes of the errors.
Citation Information
Cited By
Fault arc detection model training method, fault arc detection method and device
CN119250165A
