Method for predicting residual electrical life of low-voltage direct-current circuit breaker based on arc extinguish chamber inner wall image analysis
Through the method based on the image analysis of the interior wall of the arc extinguishing chamber, the CNN-SE network model is used to predict the remaining electrical life of the low-voltage DC circuit breaker, which solves the problem of difficulty in evaluating the electrical life of the circuit breaker in the prior art, improves the prediction accuracy and reliability, and supports the stable operation of the new energy power system.
Patent Information
- Application Number
- CN202411612992.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-12
- Publication Date
- 2025-03-07
AI Technical Summary
The existing technology is difficult to effectively evaluate the residual power life of low-voltage DC circuit breakers, which affects the stability and sustainability of new energy power systems.
Using the method based on the image analysis of the interior wall of the arc extinguishing chamber, the image is feature extraction and dimensionality reduction through the CNN-SE network model, and the channel weight is adjusted using the SENet attention mechanism, and finally the full connection layer is input to predict the remaining electrical life.
It improves the accuracy and reliability of the residual power life prediction of low-voltage DC circuit breakers, and meets the stability and sustainability needs of new energy power systems.
Smart Images
Figure CN119579510A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the fields of life prediction and machine learning, and in particular to a method for predicting the remaining electrical life of a low-voltage DC circuit breaker based on arc extinguishing chamber inner wall image analysis. Background Art
[0002] As the global energy mix shifts toward green and renewable energy, new energy power systems are becoming increasingly important in modern power grids. DC power supply technology, due to its high efficiency and low losses, has become a key component in renewable energy generation, energy storage systems, and electric transportation. In these systems, low-voltage DC circuit breakers (LVDCs) serve as core protection devices, fulfilling important functions such as fault isolation and overload protection, ensuring safe system operation. Their operational reliability is directly linked to the stability and sustainability of renewable energy power systems. Therefore, a reliable method for assessing the remaining electrical life of LVDC circuit breakers is needed. Summary of the Invention
[0003] In order to solve the problem of predicting the remaining electrical life of a low-voltage DC circuit breaker, the present invention proposes a method for predicting the remaining electrical life of a low-voltage DC circuit breaker based on arc extinguishing chamber inner wall image analysis.
[0004] The present invention is achieved through the following technical solutions:
[0005] A method for predicting the remaining electrical life of a low-voltage DC circuit breaker based on arc extinguishing chamber inner wall image analysis includes the following steps:
[0006] S1: Perform electrical life test on the circuit breaker and obtain a data set.
[0007] S2: Preprocess the data set and divide it into training set and prediction set.
[0008] S3: The preprocessed data is input into the CNN-SE network model. The features in the image are extracted and reduced in dimension through operations such as convolution and pooling. The SENet attention mechanism is used to adjust the channel weights so that the network focuses on more effective features. Finally, the data is input into the fully connected layer for remaining electrical life prediction.
[0009] S4: Use the optimized model to predict the remaining electrical life of the low-voltage DC circuit breaker.
[0010] Furthermore, in step S1, a constant stress electrical life test is performed at 3 times the rated current and rated voltage. During the test, in order to enhance the reliability of the experimental results, three circuit breakers are used for the test at the same time. The circuit breaker is disassembled every 250 times, and the inner wall of the arc extinguishing chamber of the circuit breaker is recorded using a high-definition camera to obtain a data set.
[0011] Furthermore, in step S2, Adobe Photoshop software is used to pre-process the image to remove background irrelevant to the inner wall of the arc extinguishing chamber. The data set of circuit breaker sample 1 is used as a training set, and the data sets of circuit breaker samples 2 and 3 are used as prediction sets.
[0012] Furthermore, the CNN in step S3 consists of four convolutional layers and four pooling layers. The convolution expression is as follows:
[0013]
[0014] in, is the jth output feature map obtained by the rth layer convolution operation; is the jth convolution kernel operated on the i-th input feature map; is the jth convolution kernel bias of the rth layer network; f is the nonlinear activation function.
[0015] The average pooling output is calculated as:
[0016]
[0017] where y k，i，j Represents the rectangular area R related to the kth feature map i，j The average pooling output value, x k，p,q Represents the rectangular area R i,j The element at (p, q) in |R i，j | represents a rectangular area R i，j The number of elements in .
[0018] Fully connected layer expression:
[0019] O(x)=f(wx+b)
[0020] Where x is the fully connected layer input; w is the weight matrix; b is the bias vector
[0021] Furthermore, the SENet attention mechanism described in step S3 has the function of adaptively adjusting the importance of feature channels, allowing the network to focus on more effective features, thereby improving the accuracy of remaining electrical life prediction. SENet mainly includes two operations: squeeze and excitation.
[0022] The squeeze operation compresses the W×H×C feature map containing global information into a 1×1×C feature vector through global average pooling. This feature vector has a global receptive field. Its calculation formula is:
[0023]
[0024] Where H and W are the feature map sizes; Xc ∈R H×W×C is the input feature map, C is the number of channels; Z is the weight generated by the squeeze operation.
[0025] The purpose of the excitation operation is to extract the correlation between feature channels. This is achieved using two fully connected layers. The first fully connected layer reduces the dimensionality, reducing network parameters and enhancing the generalization ability of the model. The second fully connected layer restores the original dimensionality. The calculation process can be expressed as:
[0026] S c =F ex (Z,W)=σ[W2δ(W1Z)]
[0027] Where δ is the Relu function; σ is the Sigmoid function; is the generated attention weight.
[0028] Multiply the original features by the learned channel weights, and the calculation formula is:
[0029]
[0030] in Represents element-wise multiplication; is the output feature map.
[0031] Furthermore, in the CNN-SE network model described in step S3, the number of convolution kernels in layers 1 to 4 is set to 8, 16, 32, and 32, respectively. The number of channels in the two fully connected layers in the SENet module is 4 and 32, respectively.
[0032] When training the CNN-SE network model, the optimizer Adam is used to backpropagate the parameters of each layer. The batch size is set to 7, the epochs are set to 200, and the optimization learning rate is set to 0.0015. After each 150 iterations of training, the learning rate drops to 10% of the original. The size of the arc extinguishing chamber inner wall image is 700×1279×3.
[0033] Furthermore, in step S4, the optimized CNN-SE network is tested, and the test indicators are the mean absolute error (MAE) and the root mean square error (RMSE). The calculation formula is:
[0034]
[0035]
[0036] in is the predicted value, y i is the true value. BRIEF DESCRIPTION OF THE DRAWINGS
[0037] Figure 1 It is a flow chart of the CNN-SE prediction method;
[0038] Figure 2 This is the residual electrical life prediction result diagram of circuit breaker sample 2;
[0039] Figure 3 This is the residual electrical life prediction result diagram of circuit breaker sample 3; DETAILED DESCRIPTION
[0040] The technical solution is described below in conjunction with the accompanying drawings to make the objects and advantages of the present invention clearer and easier for those skilled in the art to understand. However, the present invention is not limited to the scope of the specific implementation methods. For those of ordinary skill in the art, as long as the changes are obvious within the scope and spirit of the claims, they are all protected by the present invention.
[0041] A method for predicting the remaining electrical life of a low-voltage DC circuit breaker based on arc extinguishing chamber inner wall image analysis, the method specifically comprising the following steps:
[0042] In view of the shortcomings of existing low-voltage DC circuit breaker residual electrical life prediction methods, the present invention proposes a low-voltage DC circuit breaker residual electrical life prediction method based on arc extinguishing chamber inner wall image analysis. This method uses the image of the arc extinguishing chamber inner wall as a feature parameter to predict the residual electrical life of the low-voltage DC circuit breaker.
[0043] The present invention is achieved through the following technical solutions:
[0044] A method for predicting the remaining electrical life of a low-voltage DC circuit breaker based on arc extinguishing chamber inner wall image analysis includes the following steps:
[0045] S1: Perform an electrical life test on the circuit breaker to obtain a data set. The specific steps of the circuit breaker electrical life test are as follows:
[0046] Step 1: Build the electrical life test device:
[0047] The device consists of a test circuit, a low-voltage DC circuit breaker tripping mechanism, a data acquisition system, and a host computer. The test circuit comprises a DC power supply, a test circuit breaker, and an adjustable load. The operating handle, driven by a pneumatic cylinder, opens and closes the circuit breaker. The host computer software system is designed for Windows, using Visual Basic for test development and control of the operating mechanism and data acquisition system.
[0048] Step 2: Conduct circuit breaker electrical life test:
[0049] The low-voltage DC circuit breaker undergoing the electrical life test has a rated voltage of 500V and a rated current of 50A. The electrical life test is carried out under the condition of an overload current of 3 times the rated current. To avoid the uniformity of the experimental results, the electrical life test is carried out simultaneously using three circuit breakers from the same batch.
[0050] During the test, a high-definition camera photographed the interior of the arc extinguishing chamber every 250 circuit breaker cycles. To ensure constant environmental variables during each shot, the photography process was conducted in a small studio. The light source, the placement of the test specimen, and the distance from the test specimen to the camera were fixed in the studio. This ensured consistency of irrelevant variables during each shot, enhancing the reliability of the experimental results.
[0051] S2: Preprocess the data set and divide it into training set and prediction set. The specific steps are:
[0052] Adobe Photoshop software was used to preprocess the images and remove background irrelevant to the inner wall of the arc extinguishing chamber. The data set of circuit breaker specimen 1 was used as the training set, and the data sets of circuit breaker specimens 2 and 3 were used as the prediction sets.
[0053] By training the data set of circuit breaker test piece 1, the remaining electrical life prediction is performed on the data sets of circuit breaker test piece 2 and circuit breaker test piece 3 respectively.
[0054] Circuit breaker 1 failed to open after 3485 trips. Circuit breaker 2 failed to open after 3720 trips. Circuit breaker 3 failed to close after 3486 trips. After determining the total number of electrical life operations through electrical life testing, the remaining life referred to in this article is the number of remaining operations corresponding to the current number of operations for the circuit breaker—that is, the difference between the total number of operations and the current number of operations.
[0055] S3: The pre-processed data is input into the CNN-SE network model. Through operations such as convolution and pooling, the features in the image are extracted and reduced in dimension. The SENet attention mechanism is used to adjust the channel weights so that the network focuses on more effective features. Finally, the data is input into the fully connected layer for remaining electrical life prediction. The specific structure of the CNN-SE network model is as follows:
[0056] CNN consists of four convolutional layers and four pooling layers. The convolution expression is as follows:
[0057]
[0058] in, is the jth output feature map obtained by the rth layer convolution operation; is the jth convolution kernel operated on the i-th input feature map; is the jth convolution kernel bias of the rth layer network; f is the nonlinear activation function.
[0059] The average pooling output is calculated as:
[0060]
[0061] where y k,i,j Represents the rectangular area R related to the kth feature map i,j The average pooling output value, x k，p，q Represents the rectangular area R i,j The element at (p, q) in |R i，j | represents a rectangular area R i,j The number of elements in .
[0062] Fully connected layer expression:
[0063] O(x)=f(wx+b)
[0064] Where x is the fully connected layer input; w is the weight matrix; b is the bias vector
[0065] The SENet attention mechanism adaptively adjusts the importance of feature channels, allowing the network to focus on more effective features, thereby improving the accuracy of remaining life prediction. SENet mainly includes two operations: squeeze and excitation.
[0066] The squeeze operation compresses the W×H×C feature map containing global information into a 1×1×C feature vector through global average pooling. This feature vector has a global receptive field. Its calculation formula is:
[0067]
[0068] Where H and W are the feature map sizes; X c ∈R H×W×C is the input feature map, C is the number of channels; Z is the weight generated by the squeeze operation.
[0069] The purpose of the excitation operation is to extract the correlation between feature channels. This is achieved using two fully connected layers. The first fully connected layer reduces the dimensionality, reducing network parameters and enhancing the generalization ability of the model. The second fully connected layer restores the original dimensionality. The calculation process can be expressed as:
[0070] S c =F ex (Z,W)=σ[W2δ(W1Z)]
[0071] Where δ is the Relu function; σ is the Sigmoid function; S c ∈R 1×1×Cis the generated attention weight.
[0072] Multiply the original features by the learned channel weights, and the calculation formula is:
[0073]
[0074] in Represents element-wise multiplication; is the output feature map.
[0075] In the CNN-SE network model, the number of convolution kernels in layers 1 to 4 is set to 8, 16, 32, and 32, respectively. The number of channels in the two fully connected layers in the SENet module is 4 and 32, respectively.
[0076] When training the CNN-SE network model, the optimizer Adam is used to backpropagate the parameters of each layer. The batch size is set to 7, the epochs are set to 200, and the optimization learning rate is set to 0.0015. After each 150 iterations of training, the learning rate drops to 10% of the original. The size of the arc extinguishing chamber inner wall image is 700×1279×3.
[0077] S4: Calculate the mean absolute error (MAE) and root mean square error (RMSE) for the optimized CNN-SE network model. The calculation formulas for the mean absolute error (MAE) and root mean square error (RMSE) are as follows:
[0078]
[0079]
[0080] in is the predicted value, y i is the true value.
[0081] Use the optimized model to predict the remaining electrical life of low-voltage DC circuit breakers:
[0082] Using the circuit breaker #1 dataset as the training set, predictions were made for circuit breaker #2 and circuit breaker #3. The RMSEs for circuit breaker #2 and circuit breaker #3 were 130.02 and 116.07, respectively, and the MAEs were 117.58 and 116.07, respectively. This prediction accuracy met engineering requirements, allowing us to leverage the performance degradation data of other products from the same batch to predict the remaining electrical life of this product.
[0083] In terms of algorithm model construction, this application uses the CNN-SE network model as a prediction. When processing image data, this method does not require complex preprocessing or manual feature extraction. It directly uses the original image as input, allowing the network to learn the optimal representation features, reducing the influence of human subjective factors on feature selection, and is more efficient and robust.
Claims
1. A method for predicting the remaining electrical life of a low-voltage DC circuit breaker based on arc extinguishing chamber inner wall image analysis, characterized in that: The following steps are involved: S1: Conduct an electrical life test on the circuit breaker under constant current stress conditions, take photos of the inner wall of the arc extinguishing chamber of the circuit breaker at regular intervals, and establish a corresponding database. S2: Preprocess the images, remove the parts irrelevant to the arc extinguishing chamber, and divide them into training set and test set. S3: Build the CNN-SE network. S4: Use self-attention mechanism to optimize network output. S5: Use the optimized model to predict the life of the circuit breaker.
2. The electrical life test according to claim 1, characterized in that: The electrical life test was carried out on three circuit breakers at the same time at three times the rated current. The circuit breakers were dismantled every 250 times and the inner wall of the arc extinguishing chamber was photographed and recorded using a high-definition camera.
3. The pretreatment method according to claim 1, characterized in that: The background of the picture taken by the high-definition camera is removed to keep only the inner wall of the arc extinguishing chamber. The data of circuit breaker sample 1 is used as the training set, and the data of circuit breaker samples 2 and 3 are used as the test set.
4. The method for predicting the remaining electrical life of a low-voltage DC circuit breaker based on the image analysis of the inner wall of the arc extinguishing chamber according to claim 1 is characterized in that: The CNN-SE model in step S3 includes a CNN network and an attention mechanism. The principle of the attention mechanism can be understood as follows: when the model processes input data, in order to achieve a specific task, it needs to pay attention to some important information from the input data and ignore some unimportant information. Therefore, the attention mechanism can adjust the model's attention to each input data by weighting different input data, thereby improving the performance of the model.
5. The method for predicting the remaining electrical life of a low-voltage DC circuit breaker based on the image analysis of the inner wall of the arc extinguishing chamber according to claim 1 is characterized in that: The Adam algorithm is used to train the CNN-SE model with the ReLU function as the activation function.
6. The method for predicting the remaining electrical life of a low-voltage DC circuit breaker based on the image analysis of the inner wall of the arc extinguishing chamber according to claim 1 is characterized in that: According to the actual electrical life test, the total number of circuit breaker operations is determined. The remaining life is the remaining number of operations corresponding to the current number of operations of the circuit breaker, that is, the difference between the total number of operations and the current number of operations. By establishing a mathematical model between the current image and the remaining number of operations, the remaining electrical life under a certain operation can be obtained.
7. A method for predicting the remaining electrical life of a low-voltage DC circuit breaker based on arc extinguishing chamber inner wall image analysis according to claim 1, characterized in that: The prediction results obtained in step S4 are evaluated using two evaluation indicators, mean absolute error (MAE) and root mean square error (RMSE), as shown in the formula: in, represents the predicted value; y i Represents the true value.