Deep learning-based vacuum breakdown type discrimination method
By using deep learning technology and vacuum breakdown test data and convolutional neural networks, the vacuum breakdown type can be quickly and accurately identified, solving the problems of computational complexity and result lag in existing technologies and improving the identification efficiency of vacuum circuit breakers.
Patent Information
- Application Number
- CN202310569343.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-05-18
- Publication Date
- 2026-02-17
- Estimated Expiration
- 2043-05-18
AI Technical Summary
Existing vacuum circuit breakers require cumbersome mathematical compensation algorithms and FN formula fitting to identify vacuum breakdown types, resulting in complex calculations and results that lag behind experimental results, thus reducing the efficiency of engineering applications.
A deep learning-based approach was adopted. By constructing a vacuum breakdown test circuit, the breakdown voltage and current data were measured. Convolutional neural networks were used for feature extraction and recognition. Combined with breakdown time localization and pre-breakdown process extraction, a vacuum breakdown type identification model was constructed. The image recognition capability of neural networks was used for fast and accurate identification.
It enables simple, fast and accurate identification of vacuum breakdown types with an accuracy rate of over 85%, completing the identification within seconds, improving efficiency, reducing the workload of technicians, and has broad engineering application prospects.
Smart Images
Figure CN116776242B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to vacuum breakdown technology, specifically disclosing a method for identifying vacuum breakdown types based on deep learning, belonging to the technical field of calculation, estimation, or counting. Background Technology
[0002] Currently, vacuum circuit breakers, as economical, environmentally friendly, compact, and reliable power equipment, have occupied a major share of the medium and low voltage power switch market. However, due to the limitations of the electrode gap insulation capacity of vacuum circuit breakers, SF6 gas circuit breakers are widely used in high-voltage lines. But SF6 gas has a strong greenhouse effect, and its decomposition products are highly corrosive, causing serious environmental damage. There is an urgent need to develop vacuum circuit breakers for higher voltage levels to replace SF6 gas circuit breakers.
[0003] As the core component of a vacuum circuit breaker, the breakdown characteristics of the electrode gap in the vacuum interrupter determine the voltage level of the vacuum circuit breaker. Rapid and accurate identification of vacuum breakdown types is crucial for studying electrode gap breakdown characteristics and improving the withstand voltage level of vacuum circuit breakers. However, existing methods for identifying vacuum breakdown types require processing voltage and current data using mathematical compensation algorithms and FN formula fitting, which is cumbersome and complex, reducing the efficiency of engineering applications for vacuum breakdown type identification. Currently, with the rapid development of artificial intelligence, the accuracy and speed of image recognition have made a qualitative leap, enabling the application of deep learning in various fields. Therefore, introducing technologies such as deep learning to achieve intelligent identification of vacuum breakdown types is of great significance. Summary of the Invention
[0004] The purpose of this invention is to overcome the shortcomings of the prior art and provide a deep learning-based method for identifying vacuum breakdown types. This method aims to achieve the invention objective of simple, fast, and accurate identification of vacuum breakdown types, and solves the technical problems of existing methods for identifying vacuum breakdown types, such as the large amount of data required, complex calculations, and the lag in obtaining breakdown type identification results compared to experiments.
[0005] To achieve the above-mentioned objectives, the present invention employs the following technical solution:
[0006] A deep learning-based method for identifying vacuum breakdown types includes the following steps:
[0007] S1. Set up a vacuum breakdown test circuit, conduct a vacuum breakdown test, measure the breakdown voltage data and breakdown current data, and obtain the breakdown voltage waveform and breakdown current waveform of each vacuum breakdown test. The breakdown voltage waveform and breakdown current waveform of each vacuum breakdown test are obtained by fitting the breakdown voltage data and breakdown current data on the time axis.
[0008] S2. Classify and label the breakdown voltage waveform and breakdown current waveform of each vacuum breakdown test according to the vacuum breakdown type to obtain vacuum breakdown samples. The vacuum breakdown types include, but are not limited to: pulse current induced vacuum breakdown, field emission induced vacuum breakdown and particle induced vacuum breakdown.
[0009] S3. Divide the vacuum breakdown samples into training set, validation set and test set;
[0010] S4. Build a breakdown time localization module, a pre-breakdown process extraction module, and an image processing module, and combine them with a convolutional neural network to build a feature extraction and recognition module, thereby constructing a vacuum breakdown type identification model.
[0011] S5. Train the vacuum breakdown type identification model using the training set and validation set, test the trained vacuum breakdown type identification model using the test set, and evaluate the test results using evaluation metrics.
[0012] S6. Input the real-time acquired breakdown voltage waveform and breakdown current waveform into the vacuum breakdown type identification model to identify the vacuum breakdown type.
[0013] As a further optimization of the deep learning-based vacuum breakdown type identification method, a test circuit is used to conduct vacuum breakdown tests. This circuit includes a pulse voltage generator, a vacuum chamber or vacuum interrupter, a voltage measuring device, and a current measuring device. The pulse voltage generator, vacuum chamber, and current measuring device are connected in series to form the test circuit. The voltage measuring device is connected in parallel across the vacuum chamber. The current measuring device is a measuring resistor, a Rogowski coil, or a Hall effect coil. The voltage and current measuring devices are electrically connected to the vacuum breakdown type identification model.
[0014] As a further optimization of the deep learning-based vacuum breakdown type identification method, the breakdown time positioning module locates the moment when the voltage first drops to 0 by traversing the breakdown voltage waveform, thereby achieving accurate breakdown time positioning.
[0015] As a further optimization of the deep learning-based vacuum breakdown type identification method, the pre-breakdown process extraction module extracts the breakdown voltage waveform and breakdown current waveform within 30μs before the breakdown time, thereby realizing the pre-breakdown process extraction.
[0016] As a further optimization of the deep learning-based vacuum breakdown type identification method, the image processing module draws breakdown voltage and breakdown current waveforms containing the pre-breakdown process of different vacuum breakdown types based on the breakdown voltage and breakdown current data extracted within 30μs before the breakdown time of different vacuum breakdown types.
[0017] As a further optimization of the deep learning-based vacuum breakdown type identification method, the feature extraction and recognition module uses a convolutional neural network to extract and recognize the features of the breakdown voltage waveform and breakdown current waveform that contain different vacuum breakdown types of pre-breakdown processes.
[0018] As a further optimization of the deep learning-based vacuum breakdown type identification method, a normalization layer is added after the convolutional neural network. The Softmax classifier is used to process the features of the breakdown voltage waveform and breakdown current waveform containing different vacuum breakdown types extracted and identified by the convolutional neural network, so as to make the classification results of the neural network more accurate.
[0019] As a further optimization of the deep learning-based vacuum breakdown type identification method, the constructed vacuum breakdown type identification model is trained using training and validation sets, including: initializing the parameters of the convolutional neural network, setting an appropriate learning rate, number of training iterations, and batch size; and adjusting the parameter settings during training by observing the changing trend of the loss function.
[0020] As a further optimization of the deep learning-based vacuum breakdown type identification method, the confusion matrix is used to evaluate the vacuum breakdown type identification test results through multiple evaluation indicators such as accuracy, precision, recall, F1 score, Matthews correlation coefficient and Cohen-Kappa coefficient.
[0021] The present invention, employing the above-mentioned technical solution, has the following beneficial effects: The vacuum breakdown type identification method based on deep learning proposed in this invention, based on the probabilistic nature of vacuum breakdown, performs feature extraction and identification after locating the breakdown time and extracting the pre-breakdown process. This achieves accurate extraction of different breakdown voltage and current characteristics and breakdown type identification. Vacuum breakdown type identification can be completed using only breakdown voltage and current data, and only requires the breakdown voltage and current data obtained from the test circuit. Utilizing the excellent image recognition capabilities of neural networks, pre-breakdown feature extraction of the breakdown voltage and current waveform is performed with an accuracy rate greater than 85%, and identification is completed in seconds. This achieves simple, accurate, and rapid identification, overcoming the problems of existing vacuum breakdown identification methods requiring a large amount of data, complex calculations, and a lag in obtaining breakdown type identification results compared to experiments. It reduces the workload of technicians, shortens the identification time, and greatly improves the efficiency of vacuum breakdown identification technology, showing broad prospects for engineering applications. Attached Figure Description
[0022] Figure 1 This is a flowchart of the vacuum breakdown type identification method based on deep learning according to the present invention.
[0023] Figure 2This is a circuit diagram of the vacuum breakdown test circuit built using the deep learning-based vacuum breakdown type identification method of this invention.
[0024] Figure 3 The images show typical pulse current-induced vacuum breakdown voltage and current waveforms from the vacuum breakdown type sample image dataset of the deep learning-based vacuum breakdown type identification method of this invention.
[0025] Figure 4 The images show typical field emission-induced vacuum breakdown voltage and current waveforms from the vacuum breakdown type sample image dataset of the deep learning-based vacuum breakdown type identification method of this invention.
[0026] Figure 5 The images show the vacuum breakdown voltage and breakdown current waveforms induced by typical microparticles in the vacuum breakdown type sample image dataset of the deep learning-based vacuum breakdown type identification method of this invention.
[0027] Figure 6 This is a structural diagram of the convolutional neural network built based on the deep learning-based vacuum breakdown type identification method of this invention.
[0028] Figure 7 This is a flowchart illustrating the process by which a convolutional neural network extracts and identifies the breakdown voltage and breakdown current waveforms that represent different vacuum breakdown types during the pre-breakdown process in the deep learning-based vacuum breakdown type identification method of this invention.
[0029] Figure 8 This image shows the training results of the vacuum breakdown type identification model proposed in the deep learning-based vacuum breakdown type identification method of this invention, representing the accuracy of the model.
[0030] Figure 9 This is a flowchart illustrating the process of identifying vacuum breakdown types using the deep learning-based vacuum breakdown type identification method of this invention.
[0031] The labels in the diagram are as follows: 1. Pulse voltage generator; 2. Vacuum chamber / vacuum interrupter; 3. Current measuring device; 4. Voltage measuring device; 5. Vacuum breakdown type identification model. Detailed Implementation
[0032] The present invention will be further described below with reference to the embodiments. However, those skilled in the art should understand that the present invention is not limited to the following embodiments. Any improvements and equivalent changes made based on the specific embodiments of the present invention are within the protection scope of the claims of the present invention.
[0033] like Figure 1 As shown, the present invention provides a deep learning-based method for identifying vacuum breakdown types, comprising the following six steps.
[0034] S1. Set up a vacuum breakdown test circuit, conduct a vacuum breakdown test, measure the breakdown voltage data and breakdown current data, and obtain the breakdown voltage waveform and breakdown current waveform for each vacuum breakdown test.
[0035] In this step, the following method is used: Figure 2 The vacuum breakdown test circuit shown in the figure conducted a vacuum breakdown test and obtained a large amount of vacuum breakdown voltage data and breakdown current data for various types of electrodes and their polarity reversals. The vacuum breakdown test circuit includes: a pulse voltage generator 1, a vacuum chamber / vacuum interrupter 2, a current measuring device 3, and a voltage measuring device 4. The pulse voltage generator 1, the vacuum chamber / vacuum interrupter 2, and the current measuring device 3 are connected in series to form the test circuit. The voltage measuring device 4 is connected in parallel with the two ends of the vacuum chamber or vacuum interrupter 2. The vacuum breakdown type discrimination model 5 is electrically connected to the voltage measuring device 4 and the current measuring device 3. The current measuring device 3 is a measuring resistor, a Rogowski coil, or a Hall coil.
[0036] By fitting breakdown voltage and breakdown current data onto the time axis, the following was obtained: Figures 3 to 5 The breakdown voltage and breakdown current waveforms are shown.
[0037] S2. Classify and label the breakdown voltage waveform and breakdown current waveform of each vacuum breakdown test according to the type of vacuum breakdown to obtain the vacuum breakdown sample;
[0038] In this step, the breakdown voltage and current data are categorized into three types based on the vacuum breakdown type: pulse current-induced vacuum breakdown, field emission-induced vacuum breakdown, and particle-induced vacuum breakdown. Typical pulse current-induced vacuum breakdown voltage and current waveforms are shown below. Figure 3 As shown, the typical field emission induced vacuum breakdown voltage and breakdown current waveforms are as follows: Figure 4 As shown, the typical particle-induced vacuum breakdown voltage and breakdown current waveforms are as follows: Figure 5 As shown.
[0039] S3. Divide the vacuum breakdown samples into training set, validation set and test set; in this step, the ratio of training set to validation set is 9:1.
[0040] S4. Build a breakdown time localization module, a pre-breakdown process extraction module, and an image processing module, and combine them with a convolutional neural network to build a feature extraction and recognition module, thereby constructing a vacuum breakdown type identification model.
[0041] In this step, a breakdown time localization module, a pre-breakdown process extraction module, and an image processing module were built. By traversing the breakdown voltage waveform, the moment when the voltage first drops to 0 is located, thus accurately pinpointing the breakdown time. Then, breakdown voltage and current data within the first 30μs of the breakdown time are extracted, and based on this data, breakdown voltage and current waveforms including the pre-breakdown process are plotted. The feature extraction and recognition module employs a convolutional neural network, such as... Figure 6 As shown, the convolutional neural network includes 13 convolutional layers, 5 max-pooling layers, and 3 fully connected layers. The convolutional layers use 3x3 convolutional blocks with a stride of 2, and each convolutional layer is followed by a rectified linear unit (CLU). After each max-pooling layer, the image size is reduced by a factor of 4, the number of input channels is increased by a factor of 2, and the output is a 512-channel feature map. Then, it is activated by two 1x1x4096 layers and one 1x1x3 fully connected layer with a CLU. Finally, a normalization layer is passed through a Softmax classifier, outputting the prediction probabilities and results for each type. The process of extracting and recognizing breakdown voltage and breakdown current waveform features containing different vacuum breakdown types during the pre-breakdown process using a convolutional neural network is as follows: Figure 7 As shown.
[0042] S5. Train the vacuum breakdown type identification model using the training set and validation set, test the trained vacuum breakdown type identification model using the test set, and evaluate the test results using evaluation metrics.
[0043] In this step, the vacuum breakdown type discrimination model is trained using a training set and a validation set. This includes initializing parameters, setting the learning rate to 0.0001, the number of training epochs to 100, and the batch size to 128. During training, the parameter settings are adjusted based on the trend of the loss function to stabilize the vacuum breakdown type discrimination model. Using the confusion matrix, the vacuum breakdown type discrimination test results are evaluated using multiple metrics, including accuracy, precision, recall, F1 score, Matthews Correlation Coefficient, and Cohen's Kappa Coefficient. The test results are shown in Table 1. The accuracy of the vacuum breakdown type discrimination model gradually increases with the number of training epochs. Figure 8 As shown.
[0044] Accuracy 0.887 Recall rate 0.898 F1 score 0.904 Accuracy 0.898 Matthews Correlation Coefficient 0.714 Cohen's Kappa Coefficient 0.712
[0045] Table 1
[0046] S6. Input the breakdown voltage-current data into the vacuum breakdown type identification model to identify the vacuum breakdown type. The process is as follows: Figure 9 As shown.
[0047] The above embodiments are merely illustrative of the technical concept of the present invention and should not be construed as limiting the scope of protection of the present invention. Any modifications made to the technical solutions based on the technical concept proposed in this invention shall fall within the scope of protection of this invention.
Claims
1. A vacuum breakdown type discrimination method based on deep learning, characterized by, The method comprises the following steps: Step 1, vacuum breakdown test is performed on the vacuum interrupter, breakdown voltage data and breakdown current data are measured, and breakdown voltage waveform and breakdown current waveform of each vacuum breakdown test are obtained by fitting the breakdown voltage data and the breakdown current data on a time axis; Step 2, the breakdown voltage waveform and the breakdown current waveform of each vacuum breakdown test are classified and marked according to the vacuum breakdown type, and vacuum breakdown samples are obtained; Step 3, the vacuum breakdown samples are divided into a training set, a verification set and a test set; Step 4, a breakdown time positioning module, a pre-breakdown process extraction module, an image processing module and a feature extraction and recognition module based on a convolutional neural network are built, a vacuum breakdown type discrimination model is constructed, the breakdown time positioning module is used to traverse the breakdown voltage waveform, capture the time when the breakdown voltage is first reduced to 0 as the breakdown time, the pre-breakdown process extraction module is used to extract the breakdown voltage data and the breakdown current data within 30 microseconds before the breakdown time, the image processing module is used to obtain breakdown voltage waveform graphs and breakdown current waveform graphs containing pre-breakdown processes of different vacuum breakdown types according to the extraction results of the pre-breakdown process, and the feature extraction and recognition module based on the convolutional neural network is used to extract and recognize the features of the breakdown voltage waveform graphs and the breakdown current waveform graphs containing the pre-breakdown processes of the different vacuum breakdown types; the feature extraction and recognition module based on the convolutional neural network comprises 5 convolutional blocks, 5 maximum pooling layers and 3 fully connected layers, the breakdown voltage waveform graphs and the breakdown current waveform graphs of the pre-breakdown processes of the different vacuum breakdown types are input into the first convolutional block, the images output by each convolutional block are reduced in size by 4 times and expanded in channel number by 2 times after passing through a maximum pooling layer, the images output by the first to fourth maximum pooling layers are transmitted to the second to fifth convolutional blocks respectively, the image output by the fifth maximum pooling layer is transmitted to the three fully connected layers, the images output by the three fully connected layers are activated by a rectified linear unit, and each convolutional block comprises at least two convolutional layers, and the output of each convolutional layer is connected to a rectified linear unit; Step 5, the vacuum breakdown type discrimination model is trained by using the training set and the verification set, the trained vacuum breakdown type discrimination model is tested by using the test set, and the test results are evaluated by using evaluation indexes; Step 6, the breakdown voltage waveform and the breakdown current waveform obtained in real time are input into the trained vacuum breakdown type discrimination model, and a vacuum breakdown type discrimination result is obtained. 2.The deep learning-based vacuum breakdown type discrimination method according to claim 1, wherein, The step 1 performs the vacuum breakdown test on the vacuum interrupter by building a vacuum breakdown test circuit, the vacuum breakdown test circuit comprises a pulse voltage generating device, a vacuum interrupter, a voltage measuring device and a current measuring device, the pulse voltage generating device, the vacuum interrupter and the current measuring device are connected in series to form a test circuit, and the voltage measuring device is connected in parallel across the two ends of the vacuum interrupter. 3.The deep learning-based vacuum breakdown type discrimination method of claim 1, wherein The vacuum breakdown types include but are not limited to pulse current induced vacuum breakdown, field emission induced vacuum breakdown and particle induced vacuum breakdown. 4.The deep learning-based vacuum breakdown type discrimination method of claim 1, wherein The convolutional neural network further comprises a normalization layer for classifying the extracted and identified features, and the normalization layer is a Softmax classifier. 5.The deep learning-based vacuum breakdown type discrimination method according to claim 1, wherein, The specific method for training the vacuum breakdown type discrimination model by using the training set and the validation set in the step 5 is as follows: initializing the convolutional neural network parameters including but not limited to a learning rate, a training number and a batch size, and adjusting the parameter settings through a loss function change trend during the training process. 6.The deep learning-based vacuum breakdown type discrimination method according to claim 1, wherein, The evaluation indexes in the step 5 include but are not limited to a confusion matrix, an accuracy, a precision, a recall, an F1 score, a Matthews correlation coefficient and a Cohen kappa coefficient.