A mechanical performance evaluation method for high-voltage circuit breakers based on deep residual networks

Through the deep residual network, the closing process and current waveform data of the high-voltage circuit breaker are evaluated, which solves the problem of low accuracy in mechanical performance evaluation in the existing technology, and achieves more accurate performance status judgment and use Extended lifespan.

CN113506245BActive Publication Date: 2024-12-03STATE GRID ZHEJIANG ELECTRIC POWER CO LTD JIAXING POWER SUPPLY CO +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202110654887.6
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-06-11
Publication Date
2024-12-03
Estimated Expiration
2041-06-11

AI Technical Summary

Technical Problem

The accuracy of the mechanical performance evaluation of existing high-voltage circuit breakers is low, which affects the safety of the power grid operation and has defects in data acquisition and processing.

Method used

Using a deep residual network method, the image, image of the circuit breaker process and current waveform data are obtained and preprocessed by obtaining and preprocessing the circuit breaker, the deep residual network model is constructed, and the mechanical performance evaluation is carried out, and the circuit breaker is judged by machine intelligence learning. Mechanical failure and performance status of the device.

Benefits of technology

It improves the accuracy and reliability of the mechanical performance evaluation of high-voltage circuit breakers, extends the service life of the circuit breakers, and reduces the investment cost of the power grid to the circuit breakers.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN113506245B_ABST
    Figure CN113506245B_ABST
Patent Text Reader

Abstract

The present invention discloses a method for evaluating the mechanical properties of a high-voltage circuit breaker based on a deep residual network, comprising the following steps: S1: acquiring image data of circuit breaker opening completion, circuit breaker opening and closing process image data, circuit breaker closing completion image data and current waveform data of the circuit breaker opening and closing process of different models as sample data; S2: performing data preprocessing on the sample data of step S1, and dividing the preprocessed sample data into a training data set, a test data set and a verification data set; the present invention performs machine intelligent learning through a deep residual network algorithm, intelligently judges the mechanical performance state of the high-voltage circuit breaker, improves the accuracy and reliability of the evaluation result, enables the power grid to accurately grasp the performance state of the high-voltage circuit breaker, prolongs the service life of the high-voltage circuit breaker, and reduces the investment cost of the power grid for the high-voltage circuit breaker.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of high-voltage circuit breakers, and in particular to a method for evaluating the mechanical properties of high-voltage circuit breakers based on a deep residual network. Background Art

[0002] Circuit breakers are important equipment in high-voltage power grids, and they mainly play the role of controlling and protecting equipment. As the use time increases, some parts of the circuit breaker will continue to age or be damaged, such as wear of the mechanical mechanism, wear of the contacts, failure of the secondary mechanism of opening and closing, etc. These problems may cause the circuit breaker to malfunction, and even affect the safe operation of the entire power grid.

[0003] Therefore, it is very important to regularly check the performance of circuit breakers and detect their operating status. However, there are many problems with the current high-voltage circuit breaker detection methods, and the accuracy of the mechanical performance evaluation of high-voltage circuit breakers is low.

[0004] For example, Chinese patent CN201510433043.3 discloses a method and device for evaluating the state of SF6 high-voltage circuit breakers. The application establishes a benchmark analysis model for equipment components and the whole based on comprehensive consideration of equipment aging, equipment strength, and power grid strength, thereby improving the reliability of the state evaluation results; however, the high-voltage circuit breaker state evaluation method of the application has major defects in data acquisition and data processing, resulting in a decrease in the accuracy of the state evaluation results. Summary of the invention

[0005] The present invention mainly solves the problem of low accuracy in mechanical performance status evaluation of high-voltage circuit breakers in the existing technology; provides a high-voltage circuit breaker mechanical performance evaluation method based on a deep residual network, combines a deep residual learning network to evaluate the mechanical performance of the high-voltage circuit breaker, improves the accuracy and reliability of the evaluation results, enables the power grid to accurately grasp the performance status of the high-voltage circuit breaker, and prolongs the service life of the high-voltage circuit breaker.

[0006] The above technical problem of the present invention is mainly solved by the following technical solution: A method for evaluating the mechanical properties of a high-voltage circuit breaker based on a deep residual network, comprising the following steps:

[0007] S1: Acquire different types of circuit breaker opening completion image data, circuit breaker opening and closing process image data, circuit breaker closing completion image data, and circuit breaker opening and closing process current waveform data as sample data;

[0008] S2: preprocessing the sample data in step S1, and dividing the preprocessed sample data into a training data set, a test data set, and a validation data set;

[0009] S3: construct a deep residual network model, train the deep residual network model using the training data set of step S2, verify the deep residual network model using the verification data set, and test the deep residual network model using the test data set;

[0010] S4: Use the deep residual network model trained in step S3 to evaluate the mechanical properties of the high-voltage circuit breaker.

[0011] By collecting images when the high-voltage circuit breaker is opened, during the opening and closing process, and when the closing is completed, it is determined whether the high-voltage circuit breaker has mechanical faults such as defects, deformation, and jamming. The mechanical performance status of the circuit breaker is determined according to the opening and closing status of the high-voltage circuit breaker. Machine intelligent learning is performed through the deep residual network algorithm to intelligently determine the mechanical performance status of the high-voltage circuit breaker. At the same time, during the opening and closing process of the circuit breaker, its current state is also compared and detected, making the judgment of the deep residual network more accurate, improving the accuracy and reliability of the evaluation results, and enabling the power grid to accurately grasp the performance status of the high-voltage circuit breaker, extend the service life of the high-voltage circuit breaker, and reduce the power grid's investment cost for high-voltage circuit breakers.

[0012] Preferably, the method for acquiring the image data of the circuit breaker opening and closing process is:

[0013] S11: Find two adjacent planes of the circuit breaker that are at a certain angle, and emit circular light and dark stripes to the two adjacent planes during the opening and closing process of the circuit breaker;

[0014] S12: Collect video data of at least two circuit breaker opening and closing processes, with the starting point, end point and duration of each circuit breaker opening and closing being consistent;

[0015] S13: Frame extraction is performed on the collected video data at a period T to obtain a number of images of the circuit breaker opening and closing process. The setting method of the period T is: Among them, t is the duration of the circuit breaker opening and closing process, δ is the vibration frequency of the circuit breaker opening and closing process, and v is the angular velocity of the circuit breaker opening and closing process.

[0016] The opening and closing process of the circuit breaker is illuminated through two adjacent planes. If the circuit breaker is in a normal state during the opening and closing process, the distance between the light and dark stripes will not change. However, once the circuit breaker has a large fluctuation during the opening and closing process, resulting in a large vibration amplitude, the distance between the light and dark stripes will change. By collecting data at least twice and comparing the data, it can be clearly found whether the circuit breaker is in a normal state. Video collection is performed first and then frame extraction is performed to prevent the problem of the camera being unable to capture the picture in time. After periodic frame extraction processing, the image of the circuit breaker opening and closing process is more representative.

[0017] Preferably, in step S2, the data preprocessing of the current waveform data of the circuit breaker opening and closing process includes the following steps:

[0018] S21: collecting the current waveform of the circuit breaker during the opening and closing process, filtering the collected current waveform through a low-pass filter and a high-pass filter to obtain a filtered current signal;

[0019] S22: performing truncation processing on the filtered current signal at a period T to obtain a plurality of truncation current signals;

[0020] S23: performing Fourier transformation on a plurality of cut-off current signals to obtain frequency and amplitude information at each moment;

[0021] S24: After normalizing the amplitude, a time-frequency image is obtained according to the frequency at each moment and the normalized amplitude;

[0022] S25: Using the obtained time-frequency image as a current characteristic image of the circuit breaker opening and closing process, and using the current characteristic image of the circuit breaker opening and closing process as preprocessed sample data.

[0023] The collected current signal is truncated with the same period T, so that the current characteristic image of the current waveform corresponds to the circuit breaker opening and closing process image. The current characteristic image can more accurately reflect the mechanical properties of the circuit breaker and improve the accuracy of the evaluation result.

[0024] Preferably, constructing a deep residual network model in step S3 includes the following steps:

[0025] S31: Build the basic architecture of deep residual network;

[0026] S32: training the deep residual network using the training data set, and adjusting basic parameters of the deep residual network;

[0027] S33: Use the validation dataset to adjust the hyperparameters of the deep residual network;

[0028] S34: The deep residual network’s ability to evaluate the mechanical properties of high-voltage circuit breakers is tested using a test dataset.

[0029] The basic architecture of the deep residual network can adopt the existing basic architecture of the deep residual network. On the basis of the existing basic architecture of the deep residual network, the basic parameters of the deep residual network algorithm are modified, which greatly accelerates the establishment of the deep residual network model and makes it more in line with the mechanical performance evaluation of the high-voltage circuit breaker of the present invention.

[0030] Preferably, in step S31, the basic architecture of the deep residual network includes a convolutional layer, a residual module, a pooling layer and a fully connected layer, wherein the number of convolutional layers, residual modules and pooling layers can be set according to actual conditions.

[0031] Preferably, the deep residual network uses a gradient descent method and a binary cross entropy loss function to optimize basic parameters. The gradient descent method and the binary cross entropy loss function can quickly optimize the parameters of the deep residual network, making it easier for the deep residual network to converge globally.

[0032] Preferably, the hyperparameters include the number of sample data, the size of single sample data, the number of neurons, the learning rate and the number of iterations.

[0033] Preferably, in step S2, the data preprocessing of the circuit breaker opening completion image data, the circuit breaker opening and closing process image data, and the circuit breaker closing completion image data includes the following steps:

[0034] S221: grayscale processing is performed on the image data to obtain a grayscale image;

[0035] S222: Setting a golden circle, wherein the interior of the golden circle is filled with a feature of a grayscale value between 0 and 255, and the specific method for setting the grayscale value of the internal feature is as follows: Where K is the grayscale value of the feature inside the golden circle, N is the number of features of the circuit breaker in the grayscale image, and L i is the gray value of the i-th feature of the circuit breaker in the gray image;

[0036] S223: Use the golden circle to perform coverage detection on the grayscale image, retain some features in the grayscale image whose grayscale value is greater than the golden circle, and return the grayscale value of some features in the grayscale image whose grayscale value is less than the golden circle to 0, so as to obtain the circuit breaker opening completion feature image, the circuit breaker opening and closing process feature image, and the circuit breaker closing completion feature image. In general, the circuit breaker opening completion image data, the circuit breaker opening and closing process image data, and the circuit breaker closing completion image data directly obtained by taking pictures will have other features in addition to the structural features of the circuit breaker. The golden circle is used to extract and enhance the features, so that the circuit breaker features are more obvious.

[0037] The beneficial effects of the present invention are as follows: by collecting images when the high-voltage circuit breaker is opened, during the opening and closing process, and when the closing is completed, it is determined whether the high-voltage circuit breaker has mechanical faults such as defects, deformation, and jamming; the mechanical performance state of the circuit breaker is determined according to the opening and closing state of the high-voltage circuit breaker; machine intelligent learning is performed through a deep residual network algorithm to intelligently determine the mechanical performance state of the high-voltage circuit breaker, thereby improving the accuracy and reliability of the evaluation results, so that the power grid can accurately grasp the performance state of the high-voltage circuit breaker, extend the service life of the high-voltage circuit breaker, and reduce the investment cost of the power grid for the high-voltage circuit breaker. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 It is a flowchart of a method for evaluating mechanical properties of a high-voltage circuit breaker according to an embodiment of the present invention.

[0039] Figure 2 It is a flowchart of a method for preprocessing current waveform data according to an embodiment of the present invention. DETAILED DESCRIPTION

[0040] The technical solution of the present invention is further specifically described below through embodiments and in conjunction with the accompanying drawings.

[0041] Embodiment: A method for evaluating mechanical properties of high-voltage circuit breakers based on deep residual networks, such as Figure 1 As shown, the following steps are included:

[0042] S1: Obtain the image data of the circuit breaker opening completion, the image data of the circuit breaker opening and closing process, the image data of the circuit breaker closing completion and the current waveform data of the circuit breaker opening and closing process as sample data; the method for obtaining the image data of the circuit breaker opening and closing process is:

[0043] S11: Find two adjacent planes of the circuit breaker that are at a certain angle, and emit circular light and dark stripes to the two adjacent planes during the opening and closing process of the circuit breaker;

[0044] S12: Collect video data of at least two circuit breaker opening and closing processes, with the starting point, end point and duration of each circuit breaker opening and closing being consistent;

[0045] S13: Frame extraction is performed on the collected video data at a period T to obtain a number of images of the circuit breaker opening and closing process. The setting method of the period T is: Among them, t is the duration of the circuit breaker opening and closing process, δ is the vibration frequency of the circuit breaker opening and closing process, and v is the angular velocity of the circuit breaker opening and closing process.

[0046] S2: preprocess the sample data in step S1, and divide the preprocessed sample data into a training data set, a test data set, and a verification data set; Figure 2As shown, the data preprocessing of the current waveform data of the circuit breaker opening and closing process includes the following steps:

[0047] S21: collecting the current waveform of the circuit breaker during the opening and closing process, filtering the collected current waveform through a low-pass filter and a high-pass filter to obtain a filtered current signal;

[0048] S22: performing truncation processing on the filtered current signal at a period T to obtain a plurality of truncation current signals;

[0049] S23: performing Fourier transformation on a plurality of cut-off current signals to obtain frequency and amplitude information at each moment;

[0050] S24: After normalizing the amplitude, a time-frequency image is obtained according to the frequency at each moment and the normalized amplitude;

[0051] S25: using the obtained time-frequency image as a current characteristic image of the circuit breaker opening and closing process, and using the current characteristic image of the circuit breaker opening and closing process as preprocessed sample data;

[0052] The data preprocessing of the circuit breaker opening completion image data, the circuit breaker opening and closing process image data, and the circuit breaker closing completion image data includes the following steps:

[0053] S221: grayscale processing is performed on the image data to obtain a grayscale image;

[0054] S222: Setting a golden circle, wherein the interior of the golden circle is filled with a feature of a grayscale value between 0 and 255, and the specific method for setting the grayscale value of the internal feature is as follows: Where K is the grayscale value of the feature inside the golden circle, N is the number of features of the circuit breaker in the grayscale image, and L i is the gray value of the i-th feature of the circuit breaker in the gray image;

[0055] S223: Use the golden circle to perform coverage detection on the grayscale image, retain some features in the grayscale image whose grayscale values ​​are greater than the golden circle, and return the grayscale values ​​of some features in the grayscale image whose grayscale values ​​are less than the golden circle to 0, so as to obtain the circuit breaker opening completion feature image, the circuit breaker opening and closing process feature image and the circuit breaker closing completion feature image.

[0056] S3: Build a deep residual network model, including the following steps:

[0057] S31: Build the basic architecture of the deep residual network, set the number of convolutional layers, residual modules, pooling layers, and fully connected layers;

[0058] S32: Use the training data set to train the deep residual network, and use the gradient descent method and binary cross entropy loss function to optimize and adjust the basic parameters;

[0059] S33: Use the validation data set to adjust the hyperparameters of the deep residual network; the hyperparameters include the number of sample data, the size of a single sample data, the number of neurons, the learning rate, and the number of iterations;

[0060] S34: The deep residual network’s ability to evaluate the mechanical properties of high-voltage circuit breakers is tested using a test dataset.

[0061] S4: Use the deep residual network model trained in step S3 to evaluate the mechanical properties of the high-voltage circuit breaker.

[0062] The above-described embodiment is only a preferred solution of the present invention and does not limit the present invention in any form. There are other variations and modifications without exceeding the technical solution described in the claims.

Claims

1. A method for evaluating the mechanical properties of high-voltage circuit breakers based on deep residual networks, characterized in that: The following steps are involved: S1: Obtain the image data of the circuit breaker opening completion, the circuit breaker opening and closing process image data, the circuit breaker closing completion image data and the current waveform data of the circuit breaker opening and closing process as sample data; when the circuit breaker is opening and closing, find the adjacent planes, emit light and dark stripes to the two planes when the circuit breaker is opening and closing, collect video data and extract frames to obtain the opening and closing process image; S2: preprocess the sample data of step S1, and divide the preprocessed sample data into a training data set, a test data set, and a verification data set; filter the circuit breaker opening and closing current waveform to obtain a current signal, truncate the current signal, perform Fourier transformation on the post-processing information to obtain a time-frequency image, and use the time-frequency image as a current feature image, which is further used as sample data; S3: construct a deep residual network model, train the deep residual network model using the training data set of step S2, verify the deep residual network model using the verification data set, and test the deep residual network model using the test data set; S4: Use the deep residual network model trained in step S3 to evaluate the mechanical properties of the high-voltage circuit breaker.

2. A method for evaluating mechanical properties of high-voltage circuit breakers based on deep residual networks according to claim 1, characterized in that: The method for acquiring the image data of the circuit breaker opening and closing process is: S11: Find two adjacent planes of the circuit breaker that are at a certain angle, and emit circular light and dark stripes to the two adjacent planes during the opening and closing process of the circuit breaker; S12: Collect video data of at least two circuit breaker opening and closing processes, with the starting point, end point and duration of each circuit breaker opening and closing being consistent; S13: Frame extraction is performed on the collected video data at a period T to obtain a number of images of the circuit breaker opening and closing process. The setting method of the period T is: Among them, t is the duration of the circuit breaker opening and closing process, δ is the vibration frequency of the circuit breaker opening and closing process, and v is the angular velocity of the circuit breaker opening and closing process.

3. A method for evaluating mechanical properties of high-voltage circuit breakers based on deep residual networks according to claim 2, characterized in that: In step S2, the data preprocessing of the current waveform data of the circuit breaker opening and closing process includes the following steps: S21: collecting the current waveform of the circuit breaker during the opening and closing process, filtering the collected current waveform through a low-pass filter and a high-pass filter to obtain a filtered current signal; S22: performing truncation processing on the filtered current signal at a period T to obtain a plurality of truncation current signals; S23: performing Fourier transformation on a plurality of cut-off current signals to obtain frequency and amplitude information at each moment; S24: After normalizing the amplitude, a time-frequency image is obtained according to the frequency at each moment and the normalized amplitude; S25: Using the obtained time-frequency image as a current characteristic image of the circuit breaker opening and closing process, and using the current characteristic image of the circuit breaker opening and closing process as preprocessed sample data.

4. A method for evaluating mechanical properties of a high-voltage circuit breaker based on a deep residual network according to claim 1, 2 or 3, characterized in that: In step S3, a deep residual network model is constructed, including the following steps: S31: Build the basic architecture of deep residual network; S32: training the deep residual network using the training data set, and adjusting basic parameters of the deep residual network; S33: Use the validation dataset to adjust the hyperparameters of the deep residual network; S34: The deep residual network’s ability to evaluate the mechanical properties of high-voltage circuit breakers is tested using a test dataset.

5. A method for evaluating mechanical properties of high-voltage circuit breakers based on deep residual networks according to claim 4, characterized in that: In step S31, the basic architecture of the deep residual network includes a convolutional layer, a residual module, a pooling layer and a fully connected layer.

6. A method for evaluating mechanical properties of high-voltage circuit breakers based on deep residual networks according to claim 5, characterized in that: The deep residual network uses gradient descent and binary cross entropy loss function to optimize basic parameters.

7. A method for evaluating mechanical properties of a high-voltage circuit breaker based on a deep residual network according to claim 4, characterized in that: The hyperparameters include the number of sample data, the size of a single sample data, the number of neurons, the learning rate, and the number of iterations.

8. The method for evaluating mechanical properties of a high-voltage circuit breaker based on a deep residual network according to claim 1, characterized in that: In step S2, the data preprocessing of the circuit breaker opening completion image data, the circuit breaker opening and closing process image data, and the circuit breaker closing completion image data includes the following steps: S221: grayscale processing is performed on the image data to obtain a grayscale image; S222: Setting a golden circle, wherein the interior of the golden circle is filled with a feature of a grayscale value between 0 and 255, and the specific method for setting the grayscale value of the internal feature is as follows: Where K is the grayscale value of the feature inside the golden circle, N is the number of features of the circuit breaker in the grayscale image, and L i is the gray value of the i-th feature of the circuit breaker in the gray image; S223: Use the golden circle to perform coverage detection on the grayscale image, retain some features in the grayscale image whose grayscale values ​​are greater than the golden circle, and return the grayscale values ​​of some features in the grayscale image whose grayscale values ​​are less than the golden circle to 0, so as to obtain the circuit breaker opening completion feature image, the circuit breaker opening and closing process feature image and the circuit breaker closing completion feature image.