Insulator icing detection and ice melting method based on deep learning

Abstract

The invention belongs to the technical field of insulators, and particularly relates to an insulator icing detection and ice melting method based on deep learning, and the method comprises the following specific steps: S1, constructing an insulator image data set, and carrying out the preprocessing of an image in the data set; s2, constructing an insulator image segmentation model for segmenting an insulator existing in the image; s3, training an insulator image segmentation model; s4, calculating the icing degree of the insulator by using the insulator image segmentation model; and S5, monitoring the icing degree of the insulator, and carrying out ice melting treatment on the insulator which is seriously iced. According to the invention, the efficiency of insulator icing detection is effectively improved, manpower required by detection is greatly reduced, and the influence of environmental factors on detection is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of insulators, and in particular to a method for detecting and melting ice on insulators based on deep learning. Background Art

[0002] my country's power grid covers a vast area, with a large-scale high-voltage transmission network already established. The country is also vigorously developing and constructing ultra-high voltage AC and DC transmission and transformation projects, which represent the world's highest technological level to date. Therefore, icing on transmission lines, as a major cause of ice flashover accidents, holds significant research and practical significance. Numerous statistics, including those from the State Grid Corporation of China, indicate that icing on transmission lines can lead to tower collapses, conductor breakages, and insulator failures.

[0003] The main measures to prevent and control icing on transmission lines are "avoidance, resistance, modification, prevention, and melting." "Avoidance" means avoiding areas of heavy ice during transmission line design; "resistance" means improving the ice resistance of transmission lines; "modification" means modifying lines that currently have weak icing resistance; "prevention" means taking relevant measures to prevent icing on transmission lines; and "melting" means using AC or DC current to melt ice. For newly constructed transmission lines, the main measures used are ice prevention and ice melting; for existing old transmission lines, the main measures used are modification, ice prevention, and ice melting; and for planned new lines, the main measures used are avoidance and ice resistance. Although there are many ways to deal with icing disasters on power grids, they can ultimately be summarized as three technical means: de-icing, ice melting, and ice prevention. Among them, the most commonly used ice melting technology in power grids is DC ice melting technology.

[0004] Insulators play a vital role in high-voltage transmission lines. They primarily provide insulation and support, effectively isolating conductors from towers, preventing current leakage, ensuring directional transmission of power along the conductors, and avoiding faults such as short circuits, thereby ensuring safe and stable power transmission. Insulators offer high insulation performance, high strength, and corrosion resistance. Typically made of ceramic, glass, or composite materials, they can withstand significant mechanical stress and high voltage surges. Due to their complex structure, monitoring is relatively difficult, and currently, relatively few monitoring methods exist. Currently, several main methods exist: leakage current, ultraviolet pulse, and image monitoring.

[0005] Leakage current can reflect operating voltage, insulator contamination and icing levels, and ambient humidity. It can also dynamically monitor environmental changes through waveforms, enabling real-time monitoring. However, using leakage current to determine insulator reliability presents challenges: It has a weak correlation with insulator surface conductivity and icing conditions, requiring analysis of susceptible high-order components. Furthermore, waveforms are subject to interference from multiple factors, making it difficult to establish authoritative standards. Manual judgment is required, reducing the reliability of the results. Furthermore, the transmission network is subject to harsh conditions during icing, making it difficult to maintain a reliable power supply voltage for monitoring leakage current, hindering the development of online monitoring.

[0006] Currently, UV imaging analysis of iced insulators primarily assesses their safe operating status based on changes in UV photon counts during ice deposition and discharge. However, UV imaging discharge detection is affected by factors such as gain, detection distance, viewing angle, contamination, air pressure, and atmospheric humidity. Furthermore, there is a lack of consensus and standards for quantitative analysis of UV signatures under various conditions. Therefore, despite its advantages, UV imaging presents challenges in practical application.

[0007] In summary, the current method has the following main shortcomings:

[0008] 1. Determining the ice coverage of insulators based on leakage current requires manual intervention, reducing the reliability of the results. The transmission network is subject to harsh conditions during ice coverage, making it difficult to maintain a reliable power supply voltage for monitoring leakage current, hindering the development of online monitoring.

[0009] 2. The detection of insulator icing status by ultraviolet imaging discharge is affected by factors such as gain, detection distance, viewing angle, contamination, air pressure and atmospheric humidity. In addition, there is a lack of consensus and standards for the quantitative analysis of ultraviolet characteristic quantities under various conditions, making practical application difficult. Summary of the Invention

[0010] To solve the above technical problems, according to one aspect of the present invention, the present invention provides the following technical solutions:

[0011] A method for detecting and melting ice on insulators based on deep learning includes the following specific steps:

[0012] S1: Construct an insulator image dataset and preprocess the images in the dataset;

[0013] S2: Build an insulator image segmentation model to segment the insulators in the image;

[0014] S3: training the insulator image segmentation model;

[0015] S4: Apply the insulator image segmentation model to calculate the degree of insulator ice coverage;

[0016] S5: Monitor the degree of icing on insulators and melt the severely iced insulators.

[0017] As a preferred solution of the insulator ice detection and ice melting method based on deep learning described in the present invention, the specific steps of S1 are as follows:

[0018] S11: Construct an insulator image dataset, which includes two types of images: images of insulators without ice and images of insulators with ice.

[0019] S12: Use the padding method to fill the specified value around the images in the dataset to unify the image size, and use Labelme annotation software to mark the transmission lines and line ice in the images.

[0020] As a preferred solution of the insulator ice detection and ice melting method based on deep learning described in the present invention, the specific steps of S2 are as follows:

[0021] S21: Construct the linear inverted residual network LIRNet and use LIRNet as the backbone network of the insulator image segmentation model;

[0022] S22: Design a multi-scale feature fusion module MFF to balance the feature information in the insulator feature maps of different scales;

[0023] S23: Design an adaptive spatial attention mechanism ASA to make the model pay more attention to important areas;

[0024] S24: Construct an insulator image segmentation model to segment the insulators in the image.

[0025] As a preferred solution of the insulator ice detection and ice melting method based on deep learning described in the present invention, the specific steps of S21 are as follows:

[0026] S211: Design the inverted residual double bottleneck module IRDB to effectively extract feature information while reducing the amount of computation;

[0027] S212: Use the IRDB module to build the linear inverse residual network LIRNet as the backbone network of the insulator image segmentation model.

[0028] As a preferred solution of the insulator ice detection and ice melting method based on deep learning described in the present invention, the specific steps of S211 are: inputting the insulator feature map R2 into the inverted residual double bottleneck module IRDB, the insulator feature map R2 is subjected to a convolution layer Conv with a convolution kernel size of 1×1 and a step size of 1 to obtain the insulator inverted residual feature map B1, the insulator inverted residual feature map B1 is subjected to a depthwise separable convolution layer DW with a convolution kernel size of 3×3 and a step size of 2 to obtain the insulator inverted residual feature map B2, the insulator inverted residual feature map B2 is subjected to the convolution block attention layer CBAM to obtain the insulator inverted residual feature map B3, and the insulator inverted residual feature map B3 is obtained. Figure B3 passes through a depthwise separable convolutional layer DW with a convolution kernel size of 3×3 and a stride of 1 to obtain the insulator inverted residual feature map B4. The insulator inverted residual feature map B4 is reduced in dimension through a convolutional layer Conv with a convolution kernel size of 1×1 and a stride of 1, and then passes through a convolutional block attention layer CBAM layer to obtain the insulator inverted residual feature map B5. The insulator inverted residual feature map B3 is further reduced in dimension through a convolutional layer Conv with a convolution kernel size of 1×1 and a stride of 1 to obtain the insulator inverted residual feature map B6. Finally, the insulator inverted residual feature map B5 and the insulator inverted residual feature map B6 are added to the output insulator feature map R3 as the result of the inverted residual double bottleneck module IRDB.

[0029] As a preferred solution of the insulator icing detection and de-icing method based on deep learning described in the present invention, the specific steps of S212 are: the insulator image R1 first passes through a convolution layer Conv with a convolution kernel size of 3×3 and a step size of 2 to obtain an insulator feature map R2, the insulator feature map R2 passes through a first IRDB module to obtain an insulator feature map R3, the insulator feature map R3 passes through a second IRDB module to obtain an insulator feature map R4, the insulator feature map R4 passes through a third IRDB module to obtain an insulator feature map R5, the insulator feature map R5 passes through a fourth IRDB module to obtain an insulator feature map R6, the insulator feature map R6 passes through a convolution layer Conv with a convolution kernel size of 3×3 and a step size of 2 to output an insulator feature map R7 as the result of the linear inverse residual network LIRNet.

[0030] As a preferred solution of the insulator ice detection and de-icing method based on deep learning described in the present invention, the specific step of S22 is: inputting the insulator feature map R3, the insulator feature map R5 and the insulator feature map R7 into the multi-scale feature fusion module MFF, and then the insulator feature map R3 is down-sampled by 4 times through the average pooling AvgPool and the convolution layer Conv with a convolution kernel size of 1×1 and a step size of 1 is used to adjust the number of channels to obtain the insulator multi-scale feature map F1, and the insulator multi-scale feature map F2 is obtained. The insulator feature map F1 and the insulator feature map R5 are weighted added WAdd to obtain the insulator multi-scale feature map F2. The insulator multi-scale feature map F2 is then downsampled by 4 times through average pooling AvgPool and the convolution layer Conv with a convolution kernel size of 1×1 and a step size of 1 is used to adjust the number of channels to obtain the insulator multi-scale feature map F3. Finally, the insulator multi-scale feature map F3 and the insulator feature map R7 are weighted added WAdd to obtain the insulator feature map R8. The feature fusion process can be expressed as follows:

[0031]

[0032] Where Avg is the average pooling, and a, b, c, and d are the calculated weight factors obtained through training.

[0033] As a preferred solution of the insulator icing detection and de-icing method based on deep learning described in the present invention, the specific step of S23 is: inputting the insulator feature map R8 into the adaptive spatial attention mechanism ASA, performing a global maximum pooling GMP operation and a global average pooling GAP operation in the channel dimension on the insulator feature map R8. The purpose of these two operations is to compress the number of channels, and obtain the insulator spatial attention feature map S1 and the insulator spatial attention feature map S2 respectively. Then, the insulator spatial attention feature map S1 and the insulator spatial attention feature map S2 are weightedly added WAdd to obtain the insulator spatial attention feature map S3. The calculation formula is as follows:

[0034] S3=α·GMP(R8)+(1-α)GAP(R8)

[0035] Where GMP is the global maximum pooling, GAP is the global average pooling, and α is the adaptive weight factor obtained through training; then S3 obtains the spatial attention weight γ through the Sigmoid activation function, and finally multiplies γ with the insulator feature map R8 to give weight and output the insulator feature map R9 as the result of the adaptive spatial attention mechanism ASA. The calculation formula is as follows:

[0036] R9=σ(S3)·R8

[0037] Where σ is the Sigmoid activation function.

[0038] As a preferred solution of the insulator ice detection and de-icing method based on deep learning described in the present invention, the specific steps of S24 are: first, obtaining an insulator image R1 from the insulator image dataset, inputting the insulator image R1 into the linear inverse residual network LIRNet in the insulator image segmentation model to obtain insulator feature map R3, insulator feature map R5 and insulator feature map R7; inputting the insulator feature map R3, insulator feature map R5 and insulator feature map R7 into the multi-scale feature fusion module MFF for feature fusion to obtain insulator feature map R8, subjecting the insulator feature map R3 to the adaptive spatial attention mechanism ASA to obtain insulator feature map R9, subjecting the insulator feature map R5 to the adaptive spatial attention mechanism ASA to obtain insulator feature map R10, subjecting the insulator feature map R8 to the adaptive spatial attention mechanism ASA to obtain insulator feature map R11, and subjecting the insulator feature map R11 to the spatial attention mechanism MFF. After extracting features using the atrous spatial convolution pooling pyramid ASPP, the insulator feature map R12 is obtained. The insulator feature map R12 is passed through a convolution layer Conv with a convolution kernel size of 1×1 and a stride of 1, and then a 4-fold upsampling operation UpSample is performed to obtain the insulator feature map R13. The insulator feature map R10 and the insulator feature map R13 are concatenated and Concat is performed to obtain the insulator feature map R14. The insulator feature map R14 is passed through a convolution layer Conv with a convolution kernel size of 1×1 and a stride of 1, and then a 4-fold upsampling operation UpSample is performed to obtain the insulator feature map R15. The insulator feature map R9 and the insulator feature map R15 are concatenated and Concat is performed to obtain the insulator feature map R16. The insulator feature map R16 is passed through a convolution layer Conv with a convolution kernel size of 1×1 and a stride of 1, and then a 4-fold upsampling operation UpSample is performed to output the segmented insulator image.

[0039] As a preferred solution of the insulator ice detection and ice melting method based on deep learning described in the present invention, the specific steps of S3 are as follows:

[0040] S31: Input a batch of 32 training images into the insulator image segmentation model for training and calculate the training loss value loss;

[0041] S32: After the insulator image segmentation model is trained using all batches of insulator images in the training set, the validation set is input into the insulator image segmentation model in batches to obtain the corresponding batch loss value batch_loss;

[0042] S33: During training and validation, the parameters of the insulator image segmentation model will be automatically learned and adjusted based on each loss and batch_loss situation;

[0043] S34: When the loss and batch_loss values ​​converge and the difference between them is less than a threshold, the training of the insulator image segmentation model ends; the threshold needs to be determined according to the actual training situation;

[0044] S35: After training, use the test set to test the model effect. The trend of the loss value of the test set can determine whether the model has learned effectively and whether overfitting may occur.

[0045] The specific steps of S4 are as follows:

[0046] S41: Take an un-iced insulator image T0, input it into the insulator image segmentation model, obtain a local insulator image T1 segmented from the insulator image T0, and calculate the number N0 of insulator pixels in the insulator image T1;

[0047] S42: Take an insulator image T2 every 30 minutes and input it into the insulator image segmentation model. Segment the insulator image T2 to obtain a local insulator image T3, and calculate the number N1 of insulator pixels in the insulator image T3. Calculate the ratio of the number of insulator pixels in the insulator image T3 to the number of insulator pixels in the insulator image T1 using the following formula:

[0048]

[0049] Where P is the ratio of the number of insulator pixels in T3 to the number of insulator pixels in T1, that is, the degree of insulator ice coverage;

[0050] The specific steps of S5 are as follows:

[0051] S51: Real-time monitoring of the insulator icing degree P;

[0052] S52: When the ice coverage P of the insulator reaches a level that affects the normal operation of the insulator, an alarm is immediately issued, and the power grid staff will melt the iced insulators by applying a short-term high current to the insulator string to melt the ice using the Joule heating effect, or use laser or ultrasonic technology to remotely remove the ice.

[0053] Compared with existing technologies:

[0054] The present invention constructs an insulator image segmentation model that can accurately identify insulators in the segmented image, so as to monitor the degree of insulator icing in real time. When the degree of icing on the insulator reaches a level that affects the normal operation of the insulator, an alarm is immediately issued to the staff, and the iced insulators are melted. Based on this, the present invention effectively improves the efficiency of insulator icing detection, greatly reduces the manpower required for detection, and reduces the impact of environmental factors on detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 It is a schematic diagram of the overall process of the present invention;

[0056] Figure 2 This is a schematic diagram of the inverted residual double bottleneck module IRDB structure of the present invention;

[0057] Figure 3 This is a schematic diagram of the linear inverted residual network LIRNet structure of the present invention;

[0058] Figure 4 This is a schematic diagram of the structure of the multi-scale feature fusion module MFF of the present invention;

[0059] Figure 5 This is a schematic diagram of the structure of the adaptive spatial attention mechanism ASA of the present invention;

[0060] Figure 6 This is a structural diagram of the insulator image segmentation model of the present invention. DETAILED DESCRIPTION

[0061] To make the objectives, technical solutions and advantages of the present invention more clear, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0062] The present invention provides a method for detecting and melting ice on insulators based on deep learning. Figures 1-6 , including the following specific steps:

[0063] S1: Construct an insulator image dataset and preprocess the images in the dataset.

[0064] The specific steps of S1 are as follows:

[0065] S11: Construct an insulator image dataset, which includes two types of images: images of insulators without ice and images of insulators with ice.

[0066] S12: Use the padding method to fill the specified value around the images in the dataset to unify the image size, and use Labelme annotation software to mark the transmission lines and line ice in the images.

[0067] Among them, pad is a common method of modifying image size during the dataset preprocessing of a neural network model. This method does not change the shape of the original data and does not lose the original information of the image.

[0068] Labelme annotation software is a commonly used tool for annotating image information, but is not an innovation of the present invention.

[0069] S2: Build an insulator image segmentation model to segment the insulators in the image;

[0070] This allows, when an insulator image R1 is obtained from the insulator image dataset, the insulator image R1 to be input into the insulator image segmentation model to segment the insulators in the image.

[0071] The specific steps of S2 are as follows:

[0072] S21: Construct a linear inverted residual network (LIRNet) and use LIRNet as the backbone network for the insulator image segmentation model.

[0073] R1 first passes through a convolution layer Conv with a convolution kernel size of 3×3 and a stride of 2 to obtain the insulator feature map R2;

[0074] The specific steps of S21 are as follows:

[0075] S211: Design an inverted residual double bottleneck module IRDB (Inverted residual double bottleneck module) to effectively extract feature information while reducing the amount of computation.

[0076] The specific steps of S211 are as follows: inputting the insulator feature map R2 into the inverted residual double bottleneck module IRDB, performing dimension increase on the insulator feature map R2 through a convolution layer Conv with a convolution kernel size of 1×1 and a step size of 1 to obtain an insulator inverted residual feature map B1, performing dimension increase on the insulator feature map B1 through a depthwise separable convolution layer DW with a convolution kernel size of 3×3 and a step size of 2 to obtain an insulator inverted residual feature map B2, and performing dimension increase on the insulator feature map B2 through a convolution block attention layer CBAM (Convolutional Block Attention Module) obtains the insulator inverted residual feature map B3, and the insulator inverted residual feature map B3 passes through a depth-separable convolution layer DW with a convolution kernel size of 3×3 and a step size of 1 to obtain the insulator inverted residual feature map B4. The insulator inverted residual feature map B4 passes through a convolution layer Conv with a convolution kernel size of 1×1 and a step size of 1 for dimensionality reduction and then passes through a convolution block attention layer CBAM layer to obtain the insulator inverted residual feature map B5. The insulator inverted residual feature map B3 passes through a convolution layer Conv with a convolution kernel size of 1×1 and a step size of 1 for dimensionality reduction to obtain the insulator inverted residual feature map B6. Finally, the insulator inverted residual feature map B5 and the insulator inverted residual feature map B6 are added together to output the insulator feature map R3 as the result of the inverted residual double bottleneck module IRDB; the IRDB structure is as follows Figure 2 shown.

[0077] The core idea of ​​the Convolutional Block Attention Module (CBAM) is to refine the input feature map in two stages: first, using the channel attention module to focus on "which channels are important," and then using the spatial attention module to focus on "where" the informative parts are. This dual attention mechanism enables CBAM to fully capture the key information in the features.

[0078] S212: Use the IRDB module to build the linear inverse residual network LIRNet as the backbone network of the insulator image segmentation model.

[0079] The IRDB module designed in S211 is used to construct a linear inverted residual network LIRNet as the backbone network of the insulator image segmentation model. The specific steps of S212 are as follows: the insulator image R1 first passes through a convolution layer Conv with a convolution kernel size of 3×3 and a step size of 2 to obtain an insulator feature map R2, the insulator feature map R2 passes through the first IRDB module to obtain an insulator feature map R3, the insulator feature map R3 passes through the second IRDB module to obtain an insulator feature map R4, the insulator feature map R4 passes through the third IRDB module to obtain an insulator feature map R5, the insulator feature map R5 passes through the fourth IRDB module to obtain an insulator feature map R6, the insulator feature map R6 passes through a convolution layer Conv with a convolution kernel size of 3×3 and a step size of 2 to output an insulator feature map R7 as the result of the linear inverted residual network LIRNet; the LIRNet structure is as follows Figure 3 The first IRDB module, the second IRDB module, the third IRDB module, and the fourth IRDB module have the same structure.

[0080] Example: An insulator image R1 is obtained from the insulator image dataset. The size of R1 is 512×512×3 (corresponding to the length, width and number of channels of the feature map). R1 ​​first passes through a convolution layer Conv with a convolution kernel size of 3×3 and a step size of 2 to obtain an insulator feature map R2 with a size of 256×256×32. R2 is input into the first IRDB module. R2 passes through a Conv layer with a convolution kernel size of 1×1 and a step size of 1 to obtain an insulator inverted residual feature map B1 with a size of 256×256×128. B1 passes through a depthwise separable convolution layer DW with a convolution kernel size of 3×3 and a step size of 2 to obtain an insulator inverted residual feature map B2 with a size of 128×128×128. B2 is obtained by CBAM. To the insulator inverted residual feature map B3 of size 128×128×128, B3 passes through a DW layer with a convolution kernel size of 3×3 and a step size of 1 to obtain the insulator inverted residual feature map B4 of size 128×128×128, B4 passes through a Conv layer with a convolution kernel size of 1×1 and a step size of 1 for dimensionality reduction, and then passes through a CBAM layer to obtain the insulator inverted residual feature map B5 of size 128×128×64, B3 passes through a Conv layer with a convolution kernel size of 1×1 and a step size of 1 for dimensionality reduction to obtain the insulator inverted residual feature map B6 of size 128×128×64, and finally B5 and B6 are added to output the insulator feature map R3 of size 128×128×64 as the result of the first IRDB module. R3 passes through the second IRDB module to obtain an insulator feature map R4 of size 64×64×128, R4 passes through the third IRDB module to obtain an insulator feature map R5 of size 64×64×256, R5 passes through the fourth IRDB module to obtain an insulator feature map R6 of size 32×32×512, R6 passes through a Conv layer with a convolution kernel size of 3×3 and a stride of 2 to output an insulator feature map R7 of size 16×16×1024 as the result of LIRNet.

[0081] S22: Design a multi-scale feature fusion module MFF (Multi-scale feature fusion module) to balance the feature information in insulator feature maps of different scales.

[0082] The specific steps of S22 are: inputting the insulator feature map R3, the insulator feature map R5 and the insulator feature map R7 into the multi-scale feature fusion module MFF, then the insulator feature map R3 is downsampled by 4 times through average pooling AvgPool and the convolution layer Conv with a convolution kernel size of 1×1 and a step size of 1 is used to adjust the number of channels to obtain the insulator multi-scale feature map F1, the insulator multi-scale feature map F1 and the insulator feature map R5 are weighted added WAdd (Weighted addition) to obtain the insulator multi-scale feature map F2, and then the insulator multi-scale feature map F2 is downsampled by 4 times through average pooling AvgPool and the convolution layer Conv with a convolution kernel size of 1×1 and a step size of 1 is used to adjust the number of channels to obtain the insulator multi-scale feature map F3, finally the insulator multi-scale feature map F3 and the insulator feature map R7 are weighted added WAdd to obtain the insulator feature map R8. The feature fusion process can be expressed by the following formula:

[0083]

[0084] Where Avg is the average pooling, a, b, c, d are the calculated weight factors obtained through training; the MFF structure is as follows Figure 4 shown.

[0085] Example: R3, R5 and R7 are input into MFF, and then R3 is downsampled by 4 times through AvgPool and the convolution kernel size is 1×1 and the step size is 1. The Conv layer adjusts the number of channels to obtain an insulator multi-scale feature map F1 of size 64×64×256. F1 and R5 are WAdded to obtain an insulator multi-scale feature map F2 of size 64×64×256. F2 is then downsampled by 4 times through AvgPool and the convolution kernel size is 1×1 and the step size is 1. The Conv layer adjusts the number of channels to obtain an insulator multi-scale feature map F3 of size 16×16×1024. Finally, F3 and R7 are WAdded to obtain an insulator feature map R8 of size 16×16×1024, which is output as the result of MFF.

[0086] S23: Design an adaptive spatial attention mechanism ASA (Adaptive Spatial Attention module) to make the model pay more attention to important areas.

[0087] The specific steps of S23 are: inputting the insulator feature map R8 into the adaptive spatial attention mechanism ASA, performing the global maximum pooling GMP (Global Max Pooling) operation and the global average pooling GAP (Global Average Pooling) operation in the channel dimension on the insulator feature map R8. The purpose of these two operations is to compress the number of channels, and obtain the insulator spatial attention feature map S1 and the insulator spatial attention feature map S2 respectively. Then, the insulator spatial attention feature map S1 and the insulator spatial attention feature map S2 are weightedly added WAdd to obtain the insulator spatial attention feature map S3. The calculation formula is as follows:

[0088] S3=α·GMP(R8)+(1-α)GAP(R8)

[0089] Where GMP is the global maximum pooling, GAP is the global average pooling, and α is the adaptive weight factor obtained through training; then S3 obtains the spatial attention weight γ through the Sigmoid activation function, and finally multiplies γ with the insulator feature map R8 to give weight and output the insulator feature map R9 as the result of the adaptive spatial attention mechanism ASA. The calculation formula is as follows:

[0090] R9=σ(S3)·R8

[0091] Where σ is the Sigmoid activation function; the ASA structure diagram is as follows Figure 5 shown.

[0092] Example: Input the insulator feature map R8 into ASA, perform GMP and GAP on R8 in the channel dimension, compress the number of channels, and obtain the insulator spatial attention feature maps S1 and S2 respectively. The sizes of S1 and S2 are both 16×16×1. Then perform WAdd on S1 and S2 to obtain the insulator spatial attention feature map S3 with a size of 16×16×1. Then S3 is activated by the Sigmoid function to obtain the spatial attention weight γ. Finally, γ is multiplied by R8 for weighting and the output is the insulator feature map R9 with a size of 16×16×1024 as the result of ASA.

[0093] S24: Construct an insulator image segmentation model to segment the insulators in the image.

[0094] The insulator image segmentation model is constructed using the backbone network LIRNet built by S21, the multi-scale feature fusion module MFF designed by S22, and the adaptive spatial attention mechanism ASA designed by S23. The model structure is as follows Figure 6 shown.

[0095] The specific steps of S24 are as follows: first, an insulator image R1 is obtained from the insulator image dataset, and the insulator image R1 is input into the linear inverse residual network LIRNet in the insulator image segmentation model to obtain insulator feature maps R3, R5 and R7; the insulator feature maps R3, R5 and R7 are input into the multi-scale feature fusion module MFF for feature fusion to obtain an insulator feature map R8, the insulator feature map R3 is subjected to the adaptive spatial attention mechanism ASA to obtain an insulator feature map R9, the insulator feature map R5 is subjected to the adaptive spatial attention mechanism ASA to obtain an insulator feature map R10, the insulator feature map R8 is subjected to the adaptive spatial attention mechanism ASA to obtain an insulator feature map R11, and the insulator feature map R11 is subjected to the atrous spatial convolution pooling pyramid ASPP (Atrous Spatial Pyramid After extracting features through Pooling, an insulator feature map R12 is obtained. The insulator feature map R12 passes through a convolution layer Conv with a convolution kernel size of 1×1 and a stride of 1, and then performs a 4-fold upsampling operation UpSample to obtain an insulator feature map R13. The insulator feature map R10 and the insulator feature map R13 are concatenated and Concat is performed to obtain an insulator feature map R14. The insulator feature map R14 passes through a convolution layer Conv with a convolution kernel size of 1×1 and a stride of 1, and then performs a 4-fold upsampling operation UpSample to obtain an insulator feature map R15. The insulator feature map R9 and the insulator feature map R15 are concatenated and Concat is performed to obtain an insulator feature map R16. The insulator feature map R16 passes through a convolution layer Conv with a convolution kernel size of 1×1 and a stride of 1, and then performs a 4-fold upsampling operation UpSample to output the segmented insulator image.

[0096] ASPP (Atrous Spatial Pyramid Pooling) is a deep learning technique for extracting multi-scale features, commonly used in semantic segmentation tasks. It includes components such as 1×1 convolution, dilated convolution, and pyramid pooling. It uses dilated convolutions with different dilation rates to acquire features with different receptive fields. Global pooling and 1×1 convolutions are then used to obtain global information. Finally, these features are stacked and fused to enhance the model's expressiveness.

[0097] Example: First, an insulator image R1 is obtained from the insulator image dataset. The size of R1 is 512×512×3 (corresponding to the length, width and number of channels of the feature map). R1 ​​is input into the LIRNet module in the insulator image segmentation model to obtain an insulator feature map R3 with a size of 128×128×64, an insulator feature map R5 with a size of 64×64×256, and an insulator feature map R7 with a size of 16×16×1024. R3, R5 and R7 is input into MFF for feature fusion to obtain an insulator feature map R8 with a size of 16×16×1024. R3 is subjected to ASA to obtain an insulator feature map R9 with a size of 128×128×64. R5 is subjected to ASA to obtain an insulator feature map R10 with a size of 64×64×256. R8 is subjected to ASA to obtain an insulator feature map R11 with a size of 16×16×1024. R11 is subjected to Atrous Spatial Pyramid Pooling (ASPP) After extracting features using Pooling, an insulator feature map R12 of size 16×16×256 is obtained. R12 passes through a Conv with a convolution kernel size of 1×1 and a stride of 1, and then performs a 4-fold upsampling operation UpSample to obtain an insulator feature map R13 of size 64×64×256. R10 and R13 are concatenated to obtain an insulator feature map R14 of size 64×64×512. R14 passes through a Conv with a convolution kernel size of 1×1 and a stride of 1, and then performs a 4-fold upsampling operation UpSample to obtain an insulator feature map R15 of size 128×128×64. R9 and R15 are concat-processed to obtain an insulator feature map R16 of size 128×128×128. R16 passes through a Conv with a convolution kernel size of 1×1 and a stride of 1, and then performs a 4-fold upsampling operation UpSample to output the segmented insulator image.

[0098] S3: Train the insulator image segmentation model.

[0099] During the training and validation process of the insulator image segmentation model, the insulator image dataset was divided into training, validation, and test sets in a ratio of 6:3:1. During each training round, the training images in the training set were randomly shuffled and divided into 300 complete batches with a batch_size of 32. Each batch contained 32 consecutive training images and their annotations. The validation set maintained a fixed order and was divided into 94 batches in the same manner.

[0100] The specific steps of S3 are as follows:

[0101] S31: Input a batch of 32 training images into the insulator image segmentation model for training and calculate the training loss value loss;

[0102] S32: After the insulator image segmentation model is trained using all batches of insulator images in the training set, the validation set is input into the insulator image segmentation model in batches to obtain the corresponding batch loss value batch_loss;

[0103] S33: During training and validation, the parameters of the insulator image segmentation model will be automatically learned and adjusted based on each loss and batch_loss situation;

[0104] S34: When the loss and batch_loss values ​​converge and the difference between them is less than a threshold, the training of the insulator image segmentation model ends; the threshold needs to be determined according to the actual training situation;

[0105] S35: After training is completed, the model effect is tested using the test set. The trend of the loss value change of the test set can determine whether the model has learned effectively and whether overfitting may occur.

[0106] S4: Apply the insulator image segmentation model to calculate the degree of insulator ice coverage.

[0107] When a fixed monitoring device such as a camera is installed at the insulator position to be monitored, the specific steps of S4 are as follows:

[0108] S41: Take an un-iced insulator image T0, input it into the insulator image segmentation model, obtain a local insulator image T1 segmented from the insulator image T0, and calculate the number N0 of insulator pixels in the insulator image T1;

[0109] S42: Take an insulator image T2 every 30 minutes and input it into the insulator image segmentation model. Segment the insulator image T2 to obtain a local insulator image T3, and calculate the number N1 of insulator pixels in the insulator image T3. Calculate the ratio of the number of insulator pixels in the insulator image T3 to the number of insulator pixels in the insulator image T1 using the following formula:

[0110]

[0111] Where P is the ratio of the number of insulator pixels in T3 to the number of insulator pixels in T1, that is, the degree of insulator icing.

[0112] S5: Monitor the degree of icing on insulators and melt the severely iced insulators.

[0113] The specific steps of S5 are as follows:

[0114] S51: Real-time monitoring of the insulator icing degree P;

[0115] S52: When the ice coverage P of the insulator reaches a level that affects the normal operation of the insulator, an alarm is immediately issued, and the power grid staff will melt the iced insulators by applying a short-term high current to the insulator string to melt the ice using the Joule heating effect, or use laser or ultrasonic technology to remotely remove the ice.

[0116] In summary, in the present invention, an insulator image dataset is first constructed, which includes two types of images: un-iced insulator images and iced insulator images, for subsequent model construction and training. Then, an insulator image segmentation model is designed to segment the insulators in the image. In the insulator image segmentation model, the present invention designs a linear inverted residual network LIRNet (Linear Inverted Residual Network) to extract semantic information from insulator feature maps of different scales; an inverted residual double bottleneck module IRDB (Inverted residual double bottleneck module) is designed in LIRNet to effectively extract feature information while reducing the amount of computation; a multi-scale feature fusion module MFF (Multi-scale feature fusion module) is designed to balance the feature information in insulator feature maps of different scales; and an adaptive spatial attention mechanism ASA (Adaptive Spatial Attention module) is designed to make the model pay more attention to important areas. The present invention inputs an insulator image into an insulator image segmentation model, segments the insulators in the image, and determines the degree of insulator icing by calculating the ratio of the number of insulator pixels in the insulator image to the number of insulator pixels in the un-iced insulator image. When the degree of icing on an insulator reaches a level that affects its normal operation, an alert is immediately issued to personnel, and the iced insulator is melted. This invention effectively improves the efficiency of insulator icing detection, significantly reduces the manpower required for detection, and mitigates the impact of environmental factors on detection.

[0117] Although the present invention has been described above with reference to embodiments, various modifications may be made thereto and equivalent components may be substituted without departing from the scope of the present invention. In particular, as long as there are no structural conflicts, the various features of the embodiments disclosed herein may be combined with each other in any manner, and the omission of an exhaustive description of such combinations in this specification is solely for the sake of space and resource conservation. Therefore, the present invention is not limited to the specific embodiments disclosed herein, but includes all technical solutions falling within the scope of the claims.

Claims

1. A method for detecting and melting ice on insulators based on deep learning, characterized in that: The specific steps are as follows: S1: Construct an insulator image dataset and preprocess the images in the dataset; S2: Build an insulator image segmentation model to segment the insulators in the image; S3: training the insulator image segmentation model; S4: Apply the insulator image segmentation model to calculate the degree of insulator ice coverage; S5: Monitor the degree of icing on insulators and melt the severely iced insulators.

2. The method for detecting and melting ice on insulators based on deep learning according to claim 1, characterized in that: The specific steps of S1 are as follows: S11: Construct an insulator image dataset, which includes two types of images: images of insulators without ice and images of insulators with ice. S12: Use the padding method to fill the specified value around the images in the dataset to unify the image size, and use Labelme annotation software to mark the transmission lines and line ice in the images.

3. The method for detecting and melting ice on insulators based on deep learning according to claim 1, characterized in that: The specific steps of S2 are as follows: S21: Construct the linear inverted residual network LIRNet and use LIRNet as the backbone network of the insulator image segmentation model; S22: Design a multi-scale feature fusion module MFF to balance the feature information in the insulator feature maps of different scales; S23: Design an adaptive spatial attention mechanism ASA to make the model pay more attention to important areas; S24: Construct an insulator image segmentation model to segment the insulators in the image.

4. The method for detecting and melting ice on insulators based on deep learning according to claim 3, characterized in that: The specific steps of S21 are as follows: S211: Design the inverted residual double bottleneck module IRDB to effectively extract feature information while reducing the amount of computation; S212: Use the IRDB module to build the linear inverse residual network LIRNet as the backbone network of the insulator image segmentation model.

5. The method for detecting and melting ice on insulators based on deep learning according to claim 4, characterized in that: The specific steps of S211 are as follows: inputting the insulator feature map R2 into the inverted residual double bottleneck module IRDB, the insulator feature map R2 is subjected to a convolution layer Conv with a convolution kernel size of 1×1 and a step size of 1 to obtain the insulator inverted residual feature map B1, the insulator inverted residual feature map B1 is subjected to a depthwise separable convolution layer DW with a convolution kernel size of 3×3 and a step size of 2 to obtain the insulator inverted residual feature map B2, the insulator inverted residual feature map B2 is subjected to a convolution block attention layer CBAM to obtain the insulator inverted residual feature map B3, the insulator inverted residual feature map B3 is subjected to a convolution kernel size of 3×3 and a step size of 1 depth separable convolution layer DW to obtain the insulator inverted residual feature map B4, the insulator inverted residual feature map B4 is reduced in dimension by a convolution layer Conv with a convolution kernel size of 1×1 and a stride of 1, and then passed through a convolution block attention layer CBAM layer to obtain the insulator inverted residual feature map B5, the insulator inverted residual feature map B3 is further reduced in dimension by a convolution layer Conv with a convolution kernel size of 1×1 and a stride of 1 to obtain the insulator inverted residual feature map B6, finally, the insulator inverted residual feature map B5 is added to the insulator inverted residual feature map B6 to output the insulator feature map R3 as the result of the inverted residual double bottleneck module IRDB.

6. The method for detecting and melting ice on insulators based on deep learning according to claim 4, characterized in that: The specific steps of S212 are: the insulator image R1 first passes through a convolution layer Conv with a convolution kernel size of 3×3 and a step size of 2 to obtain an insulator feature map R2, the insulator feature map R2 passes through a first IRDB module to obtain an insulator feature map R3, the insulator feature map R3 passes through a second IRDB module to obtain an insulator feature map R4, the insulator feature map R4 passes through a third IRDB module to obtain an insulator feature map R5, the insulator feature map R5 passes through a fourth IRDB module to obtain an insulator feature map R6, the insulator feature map R6 passes through a convolution layer Conv with a convolution kernel size of 3×3 and a step size of 2 to output an insulator feature map R7 as the result of the linear inverse residual network LIRNet.

7. The method for detecting and melting ice on insulators based on deep learning according to claim 3, characterized in that: The specific steps of S22 are: inputting the insulator feature map R3, the insulator feature map R5 and the insulator feature map R7 into the multi-scale feature fusion module MFF, then the insulator feature map R3 is downsampled by 4 times through average pooling AvgPool and the convolution layer Conv with a convolution kernel size of 1×1 and a step size of 1 is used to adjust the number of channels to obtain the insulator multi-scale feature map F1, the insulator multi-scale feature map F1 and the insulator feature map R5 are weighted added WAdd to obtain the insulator multi-scale feature map F2, then the insulator multi-scale feature map F2 is downsampled by 4 times through average pooling AvgPool and the convolution layer Conv with a convolution kernel size of 1×1 and a step size of 1 is used to adjust the number of channels to obtain the insulator multi-scale feature map F3, finally the insulator multi-scale feature map F3 and the insulator feature map R7 are weighted added WAdd to obtain the insulator feature map R8. The feature fusion process can be expressed by the following formula: Where Avg is the average pooling, and a, b, c, and d are the calculated weight factors obtained through training.

8. The method for detecting and melting ice on insulators based on deep learning according to claim 3, characterized in that: The specific steps of S23 are: inputting the insulator feature map R8 into the adaptive spatial attention mechanism ASA, performing the global maximum pooling GMP operation and the global average pooling GAP operation in the channel dimension on the insulator feature map R8. The purpose of these two operations is to compress the number of channels, and obtain the insulator spatial attention feature map S1 and the insulator spatial attention feature map S2 respectively. Then, the insulator spatial attention feature map S1 and the insulator spatial attention feature map S2 are weightedly added WAdd to obtain the insulator spatial attention feature map S3. The calculation formula is as follows: S3=α·GMP(R8)+(1-α)GAP(R8) Where GMP is the global maximum pooling, GAP is the global average pooling, and α is the adaptive weight factor obtained through training; then S3 obtains the spatial attention weight γ through the Sigmoid activation function, and finally multiplies γ with the insulator feature map R8 to give weight and output the insulator feature map R9 as the result of the adaptive spatial attention mechanism ASA. The calculation formula is as follows: R9=σ(S3)·R8 Where σ is the Sigmoid activation function.

9. The method for detecting and melting ice on insulators based on deep learning according to claim 3, characterized in that: The specific steps of S24 are as follows: first, an insulator image R1 is obtained from the insulator image dataset, and the insulator image R1 is input into the linear inverse residual network LIRNet in the insulator image segmentation model to obtain an insulator feature map R3, an insulator feature map R5 and an insulator feature map R7; the insulator feature map R3, the insulator feature map R5 and the insulator feature map R7 are input into the multi-scale feature fusion module MFF for feature fusion to obtain an insulator feature map R8, the insulator feature map R3 is subjected to the adaptive spatial attention mechanism ASA to obtain an insulator feature map R9, the insulator feature map R5 is subjected to the adaptive spatial attention mechanism ASA to obtain an insulator feature map R10, the insulator feature map R8 is subjected to the adaptive spatial attention mechanism ASA to obtain an insulator feature map R11, and the insulator feature map R11 is subjected to the void spatial convolution pooling pyramid ASPP to extract features. To the insulator feature map R12, the insulator feature map R12 passes through a convolution layer Conv with a convolution kernel size of 1×1 and a stride of 1, and then performs a 4-fold upsampling operation UpSample to obtain the insulator feature map R13, the insulator feature map R10 and the insulator feature map R13 are concatenated and Concat is performed to obtain the insulator feature map R14, the insulator feature map R14 passes through a convolution layer Conv with a convolution kernel size of 1×1 and a stride of 1, and then performs a 4-fold upsampling operation UpSample to obtain the insulator feature map R15, the insulator feature map R9 and the insulator feature map R15 are concatenated and Concat is performed to obtain the insulator feature map R16, the insulator feature map R16 passes through a convolution layer Conv with a convolution kernel size of 1×1 and a stride of 1, and then performs a 4-fold upsampling operation UpSample to output the segmented insulator image.

10. The method for detecting and melting ice on insulators based on deep learning according to claim 1, characterized in that: The specific steps of S3 are as follows: S31: Input a batch of 32 training images into the insulator image segmentation model for training and calculate the training loss value loss; S32: After the insulator image segmentation model is trained using all batches of insulator images in the training set, the validation set is input into the insulator image segmentation model in batches to obtain the corresponding batch loss value batch_loss; S33: During training and validation, the parameters of the insulator image segmentation model will be automatically learned and adjusted based on each loss and batch_loss situation; S34: When the loss and batch_loss values ​​converge and the difference between them is less than a threshold, the training of the insulator image segmentation model ends; the threshold needs to be determined according to the actual training situation; S35: After training, use the test set to test the model effect. The trend of the loss value of the test set can determine whether the model has learned effectively and whether overfitting may occur. The specific steps of S4 are as follows: S41: Take an un-iced insulator image T0, input it into the insulator image segmentation model, obtain a local insulator image T1 segmented from the insulator image T0, and calculate the number N0 of insulator pixels in the insulator image T1; S42: Take an insulator image T2 every 30 minutes and input it into the insulator image segmentation model. Segment the insulator image T2 to obtain a local insulator image T3, and calculate the number N1 of insulator pixels in the insulator image T3. Calculate the ratio of the number of insulator pixels in the insulator image T3 to the number of insulator pixels in the insulator image T1 using the following formula: Where P is the ratio of the number of insulator pixels in T3 to the number of insulator pixels in T1, that is, the degree of insulator ice coverage; The specific steps of S5 are as follows: S51: Real-time monitoring of the insulator icing degree P; S52: When the ice coverage P of the insulator reaches a level that affects the normal operation of the insulator, an alarm is immediately issued, and the power grid staff will melt the iced insulators by applying a short-term high current to the insulator string to melt the ice using the Joule heating effect, or use laser or ultrasonic technology to remotely remove the ice.

Images