Prediction method for wind disaster risk assessment of ultra-high voltage transmission line

Through the combination of YOLO algorithm and decision tree model, wind disaster risk is evaluated on ultra-high voltage transmission lines, and the problem of inaccurate prediction of foreign body invasion in the existing technology is solved, and accurate risk assessment and early warning of wind disasters is achieved to ensure that The power grid is operating stably.

CN119068221APending Publication Date: 2024-12-03ELECTRIC POWER SCI & RES INST OF STATE GRID TIANJIN ELECTRIC POWER CO +2
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411155716.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-22
Publication Date
2024-12-03

AI Technical Summary

Technical Problem

The prior art cannot accurately predict the probability and degree of foreign object in the assessment of wind disaster risk of ultra-high voltage transmission lines, resulting in early warnings that can only be carried out when foreign object infestation occurs or is about to occur, and cannot be quantified in advance.

Method used

Using the YOLO algorithm combined with meteorological data, by establishing a gallery, training a model, calculating foreign body risk level, intrusion probability and dance level, using decision tree and regression tree model for comprehensive risk assessment, and adding regularization terms and CBAM attention mechanism to improve accuracy .

Benefits of technology

Accurate risk assessment of wind disasters on ultra-high voltage transmission lines is achieved, and the risks under the influence of foreign objects and weather can be predicted in advance, the accuracy and efficiency of early warning are improved, and the stable operation of the power grid is ensured.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119068221A_ABST
    Figure CN119068221A_ABST
Patent Text Reader

Abstract

The invention relates to a prediction method for wind disaster risk assessment of an ultra-high voltage transmission line, which assesses foreign matter invasion disaster risks of the transmission line on the basis of a YOLO algorithm principle, and uses the characteristics of YOLOV5, an added regularization item and a CBAM attention mechanism module to quickly process video image data and simultaneously merge accuracy. According to the feature point matching algorithm, a power transmission line galloping algorithm formula matched with an actual situation is obtained according to the actual situation, a more accurate power transmission line galloping calculation method is achieved, the probability of foreign matter types and the probability of occurrence of the foreign matter types under the influence of weather are predicted in advance according to foreign matter features and geographical environments, and the advance is made. And more comprehensive wind-induced disaster comprehensive risk grade evaluation is carried out.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the technical field of wind disaster prediction, and in particular to a prediction method for wind disaster risk assessment of ultra-high voltage transmission lines. Background Art

[0002] As the main force for improving the transmission capacity of the power grid, UHV needs to pay great attention to line laying and subsequent maintenance. Disasters caused by untimely and insufficient maintenance will have a great impact on the overall operation of the power system. Line swaying and foreign object intrusion caused by wind disasters are the most common meteorological disasters in transmission lines. Since the lines are exposed to the external environment, they are easily eroded by meteorological factors, which damage the outside of the lines and affect the inside of the lines, causing the transmission lines in the channel to be single or faulty. The existing common method is to extract features from images of foreign objects, transmission lines, etc. and monitor the transmission line conditions to issue early warnings.

[0003] The existing technologies mostly use image recognition technology to detect foreign object intrusion problems, manual video monitoring of transmission line dancing problems, or substitute meteorological data into mathematical formulas to approximate the degree of transmission line dancing. However, such methods still have unavoidable shortcomings during use. For example, when manual video monitoring of transmission line dancing is used, the degree of dancing cannot be quantified, and image recognition monitoring of foreign object intrusion cannot grasp the advance amount. It can only issue an early warning when the foreign object intrusion has occurred or is about to occur, and the probability of which foreign object intrusion may occur is unclear. Summary of the invention

[0004] The purpose of the present invention is to overcome the shortcomings of the prior art and propose a prediction method for wind disaster risk assessment of ultra-high voltage transmission lines, which can maintain the stable operation of the power grid, quickly and accurately provide the optimal load transfer plan, greatly improve the work efficiency of dispatchers, and ensure the safe and stable operation of the power grid.

[0005] The present invention solves the technical problem by adopting the following technical solutions:

[0006] A prediction method for wind disaster risk assessment of ultra-high voltage transmission lines comprises the following steps:

[0007] Step 1, obtaining foreign body images, transmission line foreign body intrusion images, meteorological data, and transmission line dancing images;

[0008] Step 2: Create a picture library based on the foreign body pictures and the pictures of foreign body intrusion on the power transmission line obtained in step 1;

[0009] Step 3: Based on the image library established in step 2, train using the YOLO algorithm model;

[0010] Step 4: Obtain the foreign body risk level, foreign body intrusion probability and foreign body type probability according to the model trained in step 3 and the meteorological data in step 1;

[0011] Step 5: Input the transmission line dancing picture into the feature point matching algorithm, and combine it with the meteorological data to obtain the dancing level calculation formula and calculate the dancing level;

[0012] Step 6: Calculate the comprehensive risk level of wind disasters based on the foreign body risk level, foreign body intrusion probability and foreign body type probability in step 4 and the dancing level in step 5.

[0013] Moreover, the meteorological data in step 1 include wind speed and direction, and wind speed and direction of instantaneous strong winds.

[0014] Moreover, the specific implementation method of step 3 is as follows: the transmission line foreign body intrusion picture and the foreign body picture are used as the input end of the YOLO algorithm model, the transmission line foreign body intrusion picture is enhanced by the Mosaic data at the input end to increase the data set, the initial anchor frame is set manually, and the foreign body picture is trained for feature extraction through the adaptive anchor frame at the input end. During each training, the best anchor frame value in different training sets is adaptively calculated, and then the backbone network is entered. The Focus structure is sliced ​​according to the size of the target picture, and the original picture is sliced ​​through the convolution operation of 48 convolution kernels. The CSP1_2 structure divides the feature map into 48 small feature maps. One of them is first batch normalized through the CBL structure to ensure that the branch has nonlinear capabilities, and then passes through two residual structures to improve training efficiency and accuracy. Convolution is performed again, and the other branch is directly convolved. After that, the two branches are activated by connat, BN, and SiLU in turn, and finally through the CBL structure to extract a series of features of the image; finally, the features are fused through the backbone network, and the output is the bounding box of the target position and category information, and the loss function and confidence are used to optimize the model.

[0015] Moreover, the YOLO algorithm model adopts the YOLOv5 algorithm model and adds the label smoothing technology in YOLOv4; at the same time, a regularization term is added to avoid overfitting of the model to the training data and weakening of predictability.

[0016] Moreover, the specific implementation method of obtaining the probability of foreign body intrusion and the probability of foreign body type in step 4 is: when using the YOLO algorithm model for calculation, the CART regression tree model in the decision tree model is adopted, and the model output results are the probability of foreign body intrusion and the probability of foreign body type.

[0017] Moreover, the foreign object intrusion is divided into three levels, level 0 represents no foreign objects or basically no impact on the transmission line, level 1 represents small foreign objects such as balloons and bird nests, and level 2 represents larger foreign objects such as tarpaulins and trees, or foreign objects with a large coverage area that have a significant impact on the transmission line. The output result of the YOLO algorithm is used as the input of the CART model, and the coverage area and lightness are used as feature points for division, where the coverage area is the covering area of ​​the transmission line, and the lightness is the degree to which it can be blown by the corresponding wind speed.

[0018] Moreover, the foreign body risk level = coverage level × P1 + foreign body level × P2, where 0.7 or less is defined as level 0, 0.7 to 1.3 is defined as level 1, and 1.3 or more is defined as level 2. P1 and P2 are the proportion of coverage level to foreign body risk level and the proportion of foreign body level to foreign body risk level, respectively. The fuzzy analytic hierarchy process is used to determine the proportion. When establishing the fuzzy complementary judgment matrix, the initial proportion is set as coverage level: foreign body level = 0.6: 0.4. A 2×2 matrix R = (r ji )m×m, where r ji +r ij =1, r ij =r ik -r jk +0.5(k≤m), i and j represent the number of rows and columns of the matrix respectively, r ji represents row j and column i, r ij represents row i and column j, and k represents the number of subsets.

[0019] Moreover, the result of the dancing level calculation formula in step 5 is the dancing amplitude of the line. If the result is less than the index value, the level is considered to be 0. Otherwise, the dancing level needs to be determined through a regression tree model.

[0020] The advantages and positive effects of the present invention are:

[0021] The present invention evaluates the disaster risk of foreign body intrusion on power transmission lines based on the principle of YOLO algorithm. The characteristics of YOLOV5 and the added regularization term and CBAM attention mechanism module can be used to quickly process video image data while merging accuracy. The feature point matching algorithm obtains a power transmission line dancing algorithm formula that matches the actual situation according to the actual situation, so as to achieve a more accurate transmission line dancing calculation method and predict the probability of foreign body type and the probability of foreign body type occurring under the influence of weather in advance according to foreign body characteristics and geographical environment, so as to make good preparations in advance and conduct a more comprehensive assessment of the comprehensive risk level of wind-induced disasters. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0023] The present invention is further described in detail below with reference to the accompanying drawings.

[0024] A prediction method for wind disaster risk assessment of ultra-high voltage transmission lines comprises the following steps:

[0025] Step 1: Obtain foreign object images, transmission line foreign object intrusion images, meteorological data, and transmission line vibration images.

[0026] Meteorological data include wind speed and direction of gusts, and wind speed and direction of instantaneous strong winds.

[0027] Step 2: Create a picture library based on the foreign body pictures and the pictures of foreign body intrusion on the power transmission line obtained in step 1.

[0028] Step 3: Based on the image library established in step 2, the YOLO algorithm model is used for training.

[0029] Pictures of foreign objects intruding on transmission lines, pictures of foreign objects that are likely to affect transmission lines. Because under normal circumstances, there are few data sources, the data set is small, the model training degree is reduced, and it deviates from the normal expected value, data enhancement processing (color gamut conversion, enlargement, reduction, arrangement, etc.) is performed on the existing pictures to expand the scope of the data set and enhance the degree of model training.

[0030] When extracting features from a gallery, commonly used methods include grayscale conversion, histogram, and extraction of texture features. These methods are prone to unclear features and weak sense of boundaries. Since the objects in power transmission line images are densely packed and the density of foreign matter intrusion is small, the YOLO algorithm is the most appropriate choice. When the YOLO algorithm is used for model training, the model may be over-generalized. Since the algorithm used in this patent is the YOLOV5 model, there is no corresponding treatment measure for the over-generalization of the model. Therefore, the label smoothing technology used in the YOLOV4 model is introduced, and a regularization term is added to prevent the model from over-fitting the training data and reducing predictability.

[0031] The transmission line foreign body intrusion picture and foreign body picture are used as the input of YOLOv5m. The transmission line foreign body intrusion picture is enhanced by the Mosaic data at the input end to increase the data set, and the initial anchor frame is set manually. The feature extraction training of the foreign body picture is performed through the adaptive anchor frame at the input end. Each time training, the optimal anchor frame value in different training sets is adaptively calculated, and then enters the backbone network. The Focus structure can be sliced ​​according to the size of the target picture. Through the convolution operation of 48 convolution kernels, the original picture is sliced ​​into 48 small feature maps. The CSP1_2 structure divides the feature map into two branches, one of which is first batch normalized through the CBL structure (the CBL structure is divided into three parts: conv convolution, BN batch normalization, and activation function) to ensure that the branch has nonlinear ability, and then passes through 2 residual structures to improve training efficiency and accuracy, and then convolution is performed. The other branch is directly convolved. After that, the two branches are activated by connat, BN, and SiLU in turn, and finally through the CBL structure to extract a series of features of the picture. Finally, the features are fused through the backbone network, and the output is the bounding box of the target position and category information. The loss function and confidence are used to optimize the model. In the backbone network structure, the convolution process may make the model too complicated during training and lack generalization ability. Therefore, a regular term is added to the loss function to reduce the risk of overfitting of the model. Since neural networks such as YOLOv5 are prone to absolute modeling due to overconfidence, a smoothing parameter is added to the confidence loss calculation. For example, the original label is (0, 1), and after adding the smoothing parameter, it becomes (0.05, 0.95), turning the absolutely true into a highly likely true.

[0032] After the image enters the input end, it will be converted into a matrix for processing. At this time, due to the large difference in spatial scale between the physical image of the foreign object and the physical image of the foreign object intrusion on the power transmission line, the matrix obtained will be greatly different from the original matrix, and the robustness and accuracy of the adaptive anchor frame are insufficient. Therefore, the CBAM attention mechanism module is added to the model to increase the weight in the matrix and enhance the model's ability to pay attention to image features. When detecting the dancing image of the power transmission line, the line motion features are relatively small and do not conform to the detection principle of the YOLO algorithm, so the feature point matching algorithm is used.

[0033] Step 4: Obtain the foreign body risk level, foreign body intrusion probability and foreign body type probability based on the model trained in step 3 and the meteorological data in step 1.

[0034] The model trained by the YOLO algorithm and the model trained by the feature point matching algorithm are divided into two parts. The model trained by the YOLO algorithm is combined with the geographical environment pictures and the transmission line foreign body intrusion pictures and meteorological data, and then the type and level of foreign body intrusion that the transmission line may suffer can be judged in advance when there is no image data; the model trained by the feature point matching algorithm is combined with the transmission line dancing pictures and meteorological data to obtain a specific formula for the line dancing at a specific point. The result is the line dancing amplitude. If the result is less than the index value, the level is considered to be 0, otherwise the dancing level needs to be determined by the regression tree model.

[0035] When using the YOLO algorithm model for calculation, the CART regression tree model in the decision tree model is used. The model output results are the probability of foreign body intrusion and the probability of foreign body type. Foreign body intrusion is divided into three levels. Level 0 represents no foreign body or basically no impact on the transmission line. Level 1 represents small foreign bodies such as balloons and bird nests. Level 2 represents larger foreign bodies such as tarpaulins and trees, or foreign bodies with a large coverage area that have a significant impact on the transmission line. The output results of the YOLO algorithm are used as the input of the CART model, and the coverage area and lightness are used as feature points for division. The coverage area is the covering area of ​​the transmission line, and the lightness is the degree to which it can be blown by the corresponding wind speed.

[0036]

[0037]

[0038] The corresponding lightness of the foreign body is known.

[0039]

[0040] Foreign body risk level = coverage level × P1 + foreign body level × P2, where 0.7 or less is defined as level 0, 0.7 to 1.3 is defined as level 1, and 1.3 or more is defined as level 2. P1 and P2 are the proportion of coverage level to foreign body risk level and the proportion of foreign body level to foreign body risk level, respectively. The fuzzy analytic hierarchy process is used to determine the ratio. When establishing the fuzzy complementary judgment matrix, the initial proportion is set as coverage level: foreign body level = 0.6: 0.4. A 2 × 2 matrix R = (rji)m × m is established, where rji + rij = 1, rij = rik - rjk + 0.5 (k ≤ m), i and j represent the number of rows and columns in the matrix, respectively, and r ji represents row j and column i, r ij represents row i and column j, and k represents the number of subsets.

[0041] The output results of the YOLO algorithm for the images of foreign body intrusion and the images of no foreign body intrusion are used as part of the input of the regression tree, the gust wind speed and direction, the wind speed and direction of the instantaneous maximum wind are used as the second part of the input of the regression tree, and the grayscale transformation matrix of the geographical environment image at that time is used as the third part of the input. Since the amount of data of foreign body intrusion events is small, it is best to use the k-fold cross-validation method. The three parts of the input in the same time period are taken as a sample, and then divided into k subsets according to the time difference, of which (k-1) subsets are used as training sets and 1 subset is used as a validation set. Training is performed once, and the index: mean square error is calculated. Repeat k times, and finally the average value of the index is taken as the final performance index of this regression tree. When the regression tree splits, it is easy to cause the regression tree to overfit the data and not adapt to it because the split is too fine. When the wind speed of the instantaneous maximum wind is selected as the pre-pruning feature condition, such problems can be effectively avoided, and the probability of foreign body intrusion and the probability of foreign body type under the same meteorological conditions can be obtained.

[0042] Step 5: Input the transmission line dancing picture into the feature point matching algorithm, and combine it with the meteorological data to obtain the dancing level calculation formula and calculate the dancing level.

[0043] When using the feature point matching algorithm for calculation, unlike the YOLO algorithm, the feature point matching algorithm is based on direct image operations, eliminating the need for a lot of training, and the acquired image information can be unified in spatial scale. The operation uses the SIFT algorithm to output a feature descriptor used to describe the image features, with the feature descriptor as part of the input, the gust wind speed and direction, and the instantaneous maximum wind speed and direction as another part of the input, and the rest is similar.

[0044] Step 6: Calculate the comprehensive risk level of wind disasters based on the foreign body risk level, foreign body intrusion probability and foreign body type probability in step 4 and the dancing level in step 5.

[0045] The comprehensive risk level of wind disasters is calculated by combining the foreign body risk level and the dancing level of transmission lines. Similar to the foreign body risk level, the initial weight is 1:1, and then it is determined by the fuzzy hierarchical analysis method.

[0046] It should be emphasized that the embodiments described in the present invention are illustrative rather than restrictive. Therefore, the present invention includes but is not limited to the embodiments described in the specific implementation manner. Any other implementation manners derived by those skilled in the art based on the technical solution of the present invention also fall within the scope of protection of the present invention.

Claims

1. A prediction method for wind disaster risk assessment of ultra-high voltage transmission lines, characterized by: The following steps are involved: Step 1, obtaining foreign body images, transmission line foreign body intrusion images, meteorological data, and transmission line dancing images; Step 2: Create a picture library based on the foreign body pictures and the pictures of foreign body intrusion on the power transmission line obtained in step 1; Step 3: Based on the image library established in step 2, train using the YOLO algorithm model; Step 4: Obtain the foreign body risk level, foreign body intrusion probability and foreign body type probability according to the model trained in step 3 and the meteorological data in step 1; Step 5: Input the transmission line dancing picture into the feature point matching algorithm, and combine it with the meteorological data to obtain the dancing level calculation formula and calculate the dancing level; Step 6: Calculate the comprehensive risk level of wind disasters based on the foreign body risk level, foreign body intrusion probability and foreign body type probability in step 4 and the dancing level in step 5.

2. The prediction method for wind disaster risk assessment of UHV transmission lines according to claim 1 is characterized by: The meteorological data in step 1 include wind speed and direction, and wind speed and direction of instantaneous gale.

3. The prediction method for wind disaster risk assessment of UHV transmission lines according to claim 1 is characterized in that: The specific implementation method of step 3 is as follows: the transmission line foreign body intrusion picture and the foreign body picture are used as the input end of the YOLO algorithm model, the transmission line foreign body intrusion picture is enhanced by the Mosaic data at the input end to increase the data set, the initial anchor frame is set manually, and the foreign body picture is trained for feature extraction through the adaptive anchor frame at the input end. During each training, the optimal anchor frame value in different training sets is adaptively calculated, and then the backbone network is entered. The Focus structure is sliced ​​according to the size of the target picture, and the original picture is sliced ​​into 48 small feature maps through the convolution operation of 48 convolution kernels. The CSP1_2 structure divides the feature map into two branches, one of which is first batch normalized through the CBL structure to ensure that the branch has nonlinear ability, and then passes through 2 residual structures to improve training efficiency and accuracy, and then convolution is performed, and the other branch is directly convolved. After that, the two branches are activated by connat, BN, and SiLU in turn, and finally a series of features of the picture are extracted through the CBL structure; Finally, the features are fused through the backbone network, and the output is the bounding box of the target location and category information. The loss function and confidence level are used to optimize the model.

4. The prediction method for wind disaster risk assessment of UHV transmission lines according to claim 2 is characterized in that: The YOLO algorithm model adopts the YOLOv5 algorithm model and adds the label smoothing technology in YOLOv4; at the same time, a regularization term is added to avoid overfitting of the model to the training data and weakening of predictability.

5. The prediction method for wind disaster risk assessment of UHV transmission lines according to claim 1 is characterized in that: The specific implementation method for obtaining the probability of foreign body intrusion and the probability of foreign body type in step 4 is: when using the YOLO algorithm model for calculation, the CART regression tree model in the decision tree model is adopted, and the model output results are the probability of foreign body intrusion and the probability of foreign body type.

6. The prediction method for wind disaster risk assessment of UHV transmission lines according to claim 5 is characterized by: The foreign object intrusion is divided into three levels, level 0 represents no foreign objects or basically no impact on the transmission line, level 1 represents small foreign objects such as balloons and bird nests, and level 2 represents larger foreign objects such as tarpaulins and trees, or foreign objects with a large coverage area that have a significant impact on the transmission line. The output result of the YOLO algorithm is used as the input of the CART model, and the coverage area and lightness are used as feature points for division, where the coverage area is the covering area of ​​the transmission line, and the lightness is the degree to which it can be blown by the corresponding wind speed.

7. A prediction method for wind disaster risk assessment of UHV transmission lines according to claim 6, characterized in that: The foreign body risk level = coverage level × P1 + foreign body level × P2, where 0.7 or less is defined as level 0, 0.7 to 1.3 is defined as level 1, and 1.3 or more is defined as level 2. P1 and P2 are the proportion of coverage level to foreign body risk level and the proportion of foreign body level to foreign body risk level, respectively. The fuzzy analytic hierarchy process is used to determine the proportion. When establishing the fuzzy complementary judgment matrix, the initial proportion is set as coverage level: foreign body level = 0.6: 0.

4. A 2×2 matrix R = (r ji )m×m, where r ji +r ij =1, r ij =r ik -r jk +0.5(k≤m), i and j represent the number of rows and columns of the matrix respectively, r ji represents row j and column i, r ij represents row i and column j, and k represents the number of subsets.

8. The prediction method for wind disaster risk assessment of UHV transmission lines according to claim 1 is characterized by: The result of the dancing level calculation formula in step 5 is the dancing amplitude of the line. If the result is less than the index value, the level is considered to be 0. Otherwise, the dancing level needs to be determined through a regression tree model.