A method and early warning system for auxiliary identification of transmission line faults

Abstract

This invention discloses an auxiliary identification method and early warning system for transmission line faults, belonging to the field of fault identification technology. The identification method of this invention includes the following steps: acquiring an image to be detected; extracting multiple candidate boxes from the image to be detected using a selective search algorithm; inputting the image to be detected into a convolutional neural network for feature extraction to obtain a feature map; matching the feature map with multiple candidate boxes to obtain multiple candidate box regions; pooling the multiple candidate box regions using a pyramid space to extract multiple fixed-length feature vectors; using a matrix capsule network as a classifier, inputting the multiple feature vectors into the classifier for deeper feature extraction, and identifying the extracted features. The fault identification method of this invention is fast and has stable transmission, avoiding the slow feature extraction speed and time-consuming binary classifiers of other detection methods, thus improving the stability and real-time performance of the identification.

Description

Technical Field

[0001] This invention relates to the field of fault identification technology, and in particular to an auxiliary identification method and early warning system for power transmission line faults. Background Technology

[0002] With the rapid development of society and economy, circuit inspection has gradually been freed from manual inspection. Effective inspection can ensure the normal operation of power transmission lines in snowy and icy weather, maintain the normal life in the area, and greatly reduce the maintenance cost of power transmission lines.

[0003] However, existing inspection methods still suffer from limitations such as limited technical means and the inability to synchronize line defects in real time. Furthermore, in heavy snow, the image quality acquired under icy and snowy conditions is often reduced to varying degrees compared to normal environments due to the influence of atmospheric particles and icing of transmission lines. Figure 1 As shown, atmospheric absorption and scattering caused by snow and ice cover in icy and snowy weather can make images blurry, affecting both the accuracy of image recognition and the workload of maintenance personnel.

[0004] Existing image recognition methods suffer from slow feature extraction speed, low accuracy, and the use of binary classifiers for image classification, which is time-consuming. Summary of the Invention

[0005] This invention provides a method and early warning system for auxiliary identification of power transmission line faults, which can solve the problems existing in the prior art.

[0006] This invention provides a method for auxiliary identification of transmission line faults, comprising the following steps:

[0007] Acquire the image to be detected;

[0008] Multiple candidate boxes are extracted from the image to be detected using a selective search algorithm;

[0009] The image to be detected is input into a convolutional neural network for feature extraction to obtain a feature map;

[0010] The feature map is matched with multiple candidate boxes to obtain multiple candidate box regions;

[0011] Multiple candidate box regions are pooled using pyramid space to extract multiple fixed-length feature vectors;

[0012] Using a matrix capsule network as a classifier, multiple feature vectors are input into the classifier to perform deeper feature extraction. The extracted features are then identified to obtain a fault identification image.

[0013] Preferably, the image to be detected needs to be preprocessed before being extracted by the selective search algorithm. The preprocessing method includes radiometric transformation, atmospheric correction and data augmentation.

[0014] Preferably, the conversion formula between the image to be detected and the feature map is:

[0015] (x,y)=(S*x',S*y')

[0016] In the formula, (x',y') represents the coordinate point on the feature map, (x,y) represents the coordinate point on the image to be detected, and S represents the product of all the step sizes in the convolutional neural network.

[0017] Preferably, the feature map is matched with multiple candidate boxes to obtain multiple candidate box regions using the following formula: Left, Top: x' = x / S+1 Right, Bottom: y' = y / S+1.

[0018] Preferably, the image to be detected is acquired through a camera.

[0019] Preferably, an embedded neural network acceleration chip is used to identify the image to be detected.

[0020] Preferably, the method further includes:

[0021] The fault identification image is transmitted and evaluated, and the fault is confirmed based on the evaluation result.

[0022] Once the fault is confirmed, the linkage alarm mechanism is activated to send the warning information to the operation and maintenance APP.

[0023] Preferably, the evaluation of the identified image includes the following steps:

[0024] The client performs intelligent evaluation on the identified images;

[0025] If the intelligent evaluation result is normal, the image is imported into the database; if the intelligent evaluation result is abnormal, experts will conduct a final evaluation of the image after intelligent evaluation.

[0026] Preferably, a hybrid networking system is formed by combining composite optical fiber overhead ground wire cable with wireless Wi-Fi, and combined with fiber optic industrial Ethernet switch technology to form a self-healing ring communication system for transmitting fault identification images.

[0027] A transmission line fault auxiliary early warning system, comprising:

[0028] The acquisition module is used to acquire images of the transmission line as images to be detected;

[0029] The processing module, electrically connected to the acquisition module, identifies the image to be detected to obtain a fault identification image;

[0030] The transmission module is used to transmit fault identification images;

[0031] The evaluation module, which is electrically connected to the processing module through the transmission module, is used to evaluate the fault identification image and confirm whether a fault has occurred based on the evaluation result.

[0032] The operation and maintenance module is electrically connected to the judgment module. After confirming the occurrence of a fault, it activates the linkage alarm mechanism to send the warning information to the operation and maintenance APP.

[0033] The specific processing procedure of the processing module includes the following steps:

[0034] Multiple candidate boxes are extracted from the image to be detected using a selective search algorithm;

[0035] The image to be detected is input into a convolutional neural network for feature extraction to obtain a feature map;

[0036] The feature map is matched with multiple candidate boxes to obtain multiple candidate box regions;

[0037] Multiple candidate box regions are pooled using pyramid space to extract multiple fixed-length feature vectors;

[0038] Using a matrix capsule network as a classifier, multiple feature vectors are input into the classifier to perform deeper feature extraction. The extracted features are then identified to obtain a fault identification image.

[0039] Compared with the prior art, the beneficial effects of the present invention are:

[0040] The present invention provides a method for auxiliary identification of transmission line faults, which employs a matrix capsule network multivariate classifier and a combined data communication network. This method offers fast identification speed and stable transmission, avoiding the slow feature extraction speed and time-consuming binary classifiers of other detection methods, thereby improving the stability and real-time performance of the identification. Attached Figure Description

[0041] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0042] Figure 1 This is a diagram of a power transmission line under extreme conditions according to the present invention;

[0043] Figure 2This is a flowchart of a transmission line fault auxiliary identification method according to the present invention;

[0044] Figure 3 This is an algorithm structure diagram of a transmission line fault auxiliary identification method according to the present invention;

[0045] Figure 4 This is a flowchart illustrating the operation of a transmission line fault auxiliary early warning system according to the present invention.

[0046] Figure 5 This is a simplified diagram of the operation of a power transmission line fault auxiliary early warning system according to the present invention. Detailed Implementation

[0047] 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 embodiments of the present invention, and not all embodiments. 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.

[0048] Reference Figure 2-5 This invention provides a method for auxiliary identification of transmission line faults, comprising the following steps:

[0049] Step 1: Acquire the image to be detected through the camera and preprocess the image to be detected acquired by the camera. The preprocessing includes radiometric transformation, atmospheric correction and data augmentation. Input the preprocessed image to be detected into the processing module.

[0050] Using affine transformation, images captured by a high-definition camera in the acquisition module from a fixed perspective can be expanded into images with different perspectives. By training with these expanded perspective images, the accuracy of power transmission line fault identification can be improved.

[0051] When performing a rotation transformation, the relationship between the output and the input is as follows:

[0052]

[0053] When performing a translation transformation, the relationship between the output and the input is as follows:

[0054]

[0055] When performing a scaling transformation, the relationship between the output and the input is as follows:

[0056]

[0057] In the formula, t x , t y s is the translation distance, θ is the rotation angle, and sx s y This is the scaling factor.

[0058] Atmospheric correction is applied to images acquired in icy and snowy environments to reduce the effects of absorption and scattering caused by water vapor and particles in the atmosphere, thereby improving image quality and the accuracy of fault identification.

[0059] Atmospheric scattering models are based on a fundamental assumption: suspended particles are uniformly distributed in the atmosphere, described as follows:

[0060] I (x) =J (x) t (x) +A(1-t(x)) (1)

[0061] Where I represents the luminous intensity of the image, J represents the radiance of the scene, A represents the global atmospheric luminance, and t represents the proportion of light that reaches the image sensor without being scattered. The purpose of atmospheric correction is to recover J, A, and t from I.

[0062] When the atmosphere is uniform, the transmission rate t can be expressed as:

[0063] t (x) =e -βd(x) (2)

[0064] Where is the atmospheric scattering coefficient, and d is the scene depth.

[0065] The dark channel prior is based on the fundamental assumption that in most non-sky scenes, at least one channel has pixels with very low intensity. Based on this assumption, for an input image J, the dark channel is defined as:

[0066]

[0067] Among them, J c It is the color channel of a local image patch centered at x. dark This is the dark channel of image J. If J is an image unaffected by atmospheric absorption and scattering, the dark channel J... dark The intensity is close to zero.

[0068] The estimated value of the transmission rate t is described as follows:

[0069]

[0070] The depth of field of the image needs to be considered, so the parameter λ is introduced to correct the transmission rate.

[0071] J(x)=(I(x)-A) / t(x)+A (5)

[0072] Combining formulas (1)-(5) yields the final image after atmospheric correction.

[0073] The 0.1% of the brightest pixels in the dark channel image are used to estimate the atmospheric light intensity A. These pixels correspond to the maximum values ​​in the original image, which are the estimated atmospheric light intensity values. Since the value of J becomes too large when t(x) is close to 0, causing the entire image to be biased towards white, we set a threshold for t(x), and the minimum value of t(x) is set to 0.1.

[0074] Step 2: Extract multiple candidate bounding boxes from the image to be detected using the Selective Search algorithm;

[0075] Step 3: Input the image to be detected into a convolutional neural network for feature extraction to obtain a feature map;

[0076] Candidate boxes are obtained by detecting the entire original image, while the size of the feature map is different from the size of the original image. The feature map is obtained after a series of operations such as convolution and downsampling of the original image.

[0077] Using the calculation formula directly: Assuming (x', y') represents the coordinates of a point on the feature map, and (x, y) represents a point on the original input image, then the following transformation relationship exists between them:

[0078] (x,y)=(S*x',S*y')

[0079] Where S is the product of all strides in the CNN. Conversely, by solving for (x', y') using (x, y) coordinates, the calculation formula is as follows:

[0080] x'=x / S+1

[0081] y'=y / S+1

[0082] Step 4: Match the feature map with the multiple candidate boxes from Step 2 to obtain multiple candidate box regions; map the obtained candidate box regions to the corresponding positions on the feature map to obtain the feature vectors of the candidate regions, and then match the regions of multiple candidate boxes in the feature map according to the following formula:

[0083] Left, Top: x'=x / S+1Right, Bottom: y'=y / S+1

[0084] Step 5: Use pyramid space to pool multiple candidate box regions to extract multiple fixed-length feature vectors.

[0085] Reference Figure 3When an image is input, it is divided into sections using different sized scales. The diagram illustrates this, showing three different scale sizes (4*4, 2*2, 1*1) used to divide the input image, resulting in a total of 16 + 4 + 1 = 21 blocks. From each of these 21 blocks, a feature is extracted, forming the desired 21-dimensional feature vector.

[0086] The first image divides a complete image into 16 blocks, meaning the size of each block is (w / 4, h / 4).

[0087] The second image is divided into 4 blocks, and the size of each block is (w / 2, h / 2);

[0088] The third image treats the entire image as a block, meaning the block size is (w, h).

[0089] Step 6: Use a matrix capsule network as a classifier. Input multiple feature vectors into the classifier. By changing the output of the neurons from scalars to matrices, the neural network can extract more relevant attributes of the features, such as confidence, direction, and lighting conditions. This allows for better identification of the relationships between features that make up the power transmission components, resulting in more accurate recognition results with a small number of samples.

[0090] In a matrix capsule network, let the pose matrix output by capsule i in the l-th layer be V. i The output of capsule j in layer l+1 is V i First, V i With the transformation matrix W which has viewpoint invariance ij Multiplication yields a vote. Among them W ij Training is performed via back-propagation (BP), representing the local-to-global relationship. Voting is weighted by protocol coefficients, which are obtained through expectation-maximization (EM) clustering based on a Gaussian Mixture Model (GMM). Finally, if the routing agreement of capsule i in layer l converges to a capsule in layer l+1, it means that the features extracted by capsule i are components of capsule j, and are clustered into the same cluster along with other capsules in layer l that also contain features from capsule j.

[0091] In this embodiment, an embedded neural network acceleration chip is used to recognize the image to be detected. The chip of this invention is the Hanguang 800 chip.

[0092] Step 7: Transmit and evaluate the identified image, and confirm whether a fault has occurred based on the evaluation results.

[0093] In this embodiment, a hybrid network is formed by combining a composite optical fiber overhead ground wire cable with wireless Wi-Fi, and incorporating fiber optic industrial Ethernet switch technology to create a self-healing loop communication system for transmitting the identified images. The optical fiber cable, with its electromagnetic interference resistance, enables long-distance data transmission, while the wireless link, composed of a wireless network, facilitates data sharing over short distances, making data transmission more efficient and accurate. The identified data is transmitted with electromagnetic interference resistance.

[0094] In this embodiment, the client performs intelligent recognition on the identified image. If there are no abnormalities, the image is imported into the database. If there are abnormalities, experts make a final judgment on the intelligently recognized image.

[0095] Step 8: After confirming the fault has occurred, activate the linkage alarm mechanism to send the warning information to the operation and maintenance APP.

[0096] In this embodiment, the client expert evaluates the fault that causes the warning due to the anomaly. Based on past experience with transmission line faults under extreme conditions, the expert makes a final judgment on whether the fault can be detected. If the fault is consistent with previous warning signs, the expert can send the warning fault node to the maintenance personnel via mobile device or APP. The maintenance personnel then conduct an inspection of the transmission line according to the instructions of the APP.

[0097] The present invention also provides a transmission line fault auxiliary early warning system, including a data acquisition module, a processing module, a transmission module, an evaluation module, and an operation and maintenance module.

[0098] The acquisition module consists of high-definition cameras at each node within the monitoring range, which collect real-time image data of the power transmission lines and input it as the image to be detected into the processing module.

[0099] The processing module is electrically connected to the acquisition module, and the fault identification image is obtained by recognizing the image to be detected.

[0100] The transmission module is electrically connected to the processing module and is used to evaluate the recognized image and determine whether a fault has occurred based on the evaluation result.

[0101] The judgment module, which is electrically connected to the processing module through the transmission module, is used to evaluate the recognized image and determine whether a fault has occurred based on the evaluation result.

[0102] The operation and maintenance module is electrically connected to the judgment module. After confirming the occurrence of a fault, it activates the linkage alarm mechanism to send the warning information to the operation and maintenance APP.

[0103] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0104] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for auxiliary identification of transmission line faults, characterized in that, Includes the following steps: Acquire the image to be detected; Multiple candidate boxes are extracted from the image to be detected using a selective search algorithm; The image to be detected is input into a convolutional neural network for feature extraction to obtain a feature map; The feature map is matched with multiple candidate boxes to obtain multiple candidate box regions; Multiple candidate box regions are pooled using pyramid space to extract multiple fixed-length feature vectors; Using a matrix capsule network as a classifier, multiple feature vectors are input into the classifier to perform deeper feature extraction. The extracted features are then identified to obtain a fault identification image. The conversion formula between the image to be detected and the feature map is as follows: (x,y)=(S*x',S*y') In the formula, (x',y') represents the coordinate point on the feature map, (x,y) represents the coordinate point on the image to be detected, and S represents the product of all the strides in the convolutional neural network; The method of pooling multiple candidate box regions using pyramid space specifically includes the following steps: The candidate box region is divided into multiple blocks using different sized scales; among which the multiple scales include 4*4, 2*2, and 1*1. Extract one feature from each block to generate a feature vector; In the matrix capsule network, let the first... Layered capsules The output attitude matrix is The capsule in the ℓ+1 layer The output is ,Will With a transformation matrix that is invariant to viewpoint Multiplication is voted Voting needs to be weighted by the agreement coefficient; if the capsule of the ℓth layer If the routing protocol converges to a capsule at layer ℓ+1, then the capsule... The extracted feature is capsule It is one of the components, and belongs to the capsule along with the ℓth layer. Other capsules, which are characteristic components of the capsule, are clustered together into the same cluster.

2. The method for auxiliary identification of transmission line faults as described in claim 1, characterized in that, Before extracting the image to be detected using a selective search algorithm, the image needs to be preprocessed. The preprocessing methods include radiometric transformation, atmospheric correction, and data augmentation.

3. The method for auxiliary identification of transmission line faults as described in claim 1, characterized in that, Multiple candidate box regions are obtained by matching the feature map with multiple candidate boxes using the following formula: Left, Top: x' = x / S + 1 Right, Bottom: y' = y / S + 1.

4. The method for auxiliary identification of transmission line faults as described in claim 1, wherein the image to be detected is acquired by a camera.

5. The method for auxiliary identification of transmission line faults as described in claim 1, characterized in that, An embedded neural network acceleration chip is used to recognize the image to be detected.

6. The method for auxiliary identification of transmission line faults as described in claim 1, characterized in that, The method further includes: The fault identification image is transmitted and evaluated, and the fault is confirmed based on the evaluation result. Once the fault is confirmed, the linkage alarm mechanism is activated to send the warning information to the operation and maintenance APP.

7. The method for auxiliary identification of transmission line faults as described in claim 6, characterized in that, The evaluation of the identified images includes the following steps: The client performs intelligent evaluation on the identified images; If the intelligent evaluation result is normal, the image is imported into the database; if the intelligent evaluation result is abnormal, experts will conduct a final evaluation of the image after intelligent evaluation.

8. The method for auxiliary identification of transmission line faults as described in claim 6, characterized in that, A hybrid network is formed by combining composite optical fiber overhead ground wire cable with wireless Wi-Fi, and combined with fiber optic industrial Ethernet switch technology to form a self-healing ring communication system for transmitting fault identification images.

9. A transmission line fault auxiliary early warning system based on the transmission line fault auxiliary identification method according to any one of claims 1-8, characterized in that, include: The acquisition module is used to acquire images of the transmission line as images to be detected; The processing module, electrically connected to the acquisition module, identifies the image to be detected to obtain a fault identification image; The transmission module is used to transmit fault identification images; The evaluation module, which is electrically connected to the processing module through the transmission module, is used to evaluate the fault identification image and confirm whether a fault has occurred based on the evaluation result. The operation and maintenance module is electrically connected to the judgment module. After confirming the occurrence of a fault, it activates the linkage alarm mechanism to send the warning information to the operation and maintenance APP. The specific processing procedure of the processing module includes the following steps: Multiple candidate boxes are extracted from the image to be detected using a selective search algorithm; The image to be detected is input into a convolutional neural network for feature extraction to obtain a feature map; The feature map is matched with multiple candidate boxes to obtain multiple candidate box regions; Multiple candidate box regions are pooled using pyramid space to extract multiple fixed-length feature vectors; Using a matrix capsule network as a classifier, multiple feature vectors are input into the classifier to perform deeper feature extraction. The extracted features are then identified to obtain a fault identification image.

Images