Long-endurance unmanned aerial vehicle intelligent inspection method applied to power transmission and distribution lines

Abstract

A long-endurance unmanned aerial vehicle intelligent inspection method applied to a power transmission and distribution line comprises path planning, fault identification, model updating and data transmission, and specifically comprises the following steps: S1, considering position information, number information and endurance of an unmanned aerial vehicle, taking the inspection time and the total inspection distance of the unmanned aerial vehicle as objective functions, constructing a path planning model, solving the path planning model by adopting an MOPFA algorithm to obtain a path planning scheme of each unmanned aerial vehicle; and S2, constructing an image data set according to the transmission and distribution line images shot in the inspection process of the unmanned aerial vehicle, enabling the image data set to keep balance through manual intervention in order to ensure the accuracy of the model, dividing the data set into a training set and a test set, and constructing a fault recognition model according to a faster CNN algorithm, the model being capable of recognizing faults of various transmission and distribution lines. The technical problem to be solved by the invention is to provide a long-endurance unmanned aerial vehicle intelligent inspection method applied to power transmission and distribution lines, an unmanned aerial vehicle inspection plan is obtained by solving a mathematical model of path planning, power transmission and distribution line images collected by an unmanned aerial vehicle are further identified through a Faster RCNN training image data set, and the unmanned aerial vehicle inspection plan is obtained based on new image data. Continuous learning is kept, and generalization and accuracy of the fault recognition model are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent inspection of unmanned aerial vehicles (UAVs), and particularly to a long-endurance intelligent inspection method for UAVs applied to transmission and distribution lines. Background Art

[0002] With the rapid development of the power system, the scale of China's power grid is increasing day by day. Among them, transmission and distribution lines play an important role in power transmission and are the hubs connecting the power generation side and the user side. However, the distribution of transmission and distribution lines is relatively extensive, mostly located in hilly areas, mountainous areas, water areas, etc. Due to the influence of the natural environment, faults such as fitting shedding and insulator self-explosion may occur, which have a huge impact on power transmission. To master the operation status of transmission and distribution lines, it is necessary to regularly inspect the transmission and distribution lines, discover and eliminate line faults in time, and improve power supply reliability. Traditional manual inspection has the disadvantages of low efficiency, high risk, small inspection range, etc., and cannot meet the safe and efficient inspection needs of transmission and distribution lines.

[0003] In recent years, with the rapid development of UAV technology, UAVs have gradually attracted the high attention of power enterprises due to their characteristics such as high efficiency, low cost, high safety, and easy operation. In the field of power detection, as a new type of detection platform, UAVs have been widely used in line detection. UAVs can conduct refined inspections on transmission and distribution lines and accurately identify defects such as fitting shedding and insulator self-explosion that are difficult to detect in manual inspections. Moreover, UAV inspections are not restricted by terrain and are particularly suitable for inspection operations in dangerous mountainous and river areas. At the same time, UAV inspections are also applicable to power inspections in some dangerous situations.

[0004] At present, there are problems in the intelligent inspection of UAV transmission lines such as single-point flight, disordered inspection, and low fault recognition accuracy. Therefore, in order to overcome the deficiencies of the existing technology, the present invention proposes a long-endurance intelligent inspection method for UAVs applied to transmission and distribution lines. Based on the inspection task, the optimal inspection path is planned, and during the inspection process, image data is collected in real time, and continuous learning is maintained to strengthen the monitoring and control of inspection resources. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a long-endurance intelligent inspection method for UAVs applied to transmission and distribution lines. By solving the mathematical model of path planning, the UAV inspection plan is obtained. Further, by training the image data set, the images of transmission and distribution lines collected by the UAV are recognized, and based on the new image data, continuous learning is maintained to improve the generalization and accuracy of the fault recognition model.

[0006] To solve the above technical problems, the technical solution adopted by the present invention is: a long-endurance intelligent inspection method for UAVs applied to transmission and distribution lines, including the following steps: S1, Path planning; S2, Fault identification; S3, Model update; S4, Data transmission.

[0007] Preferably, the step S1 is specifically as follows: Considering the position information, quantity information, and endurance of the unmanned aerial vehicle (UAV), taking the UAV inspection time and the total inspection distance as the objective function, constructing a path planning model, and using the MOPFA algorithm to solve the above path planning model to obtain the path planning scheme for each UAV.

[0008] Preferably, the step S2 is specifically as follows: According to the images of the power transmission and distribution lines taken during the UAV inspection, constructing an image data set. To ensure the accuracy of the model, manual intervention is used to balance the image data set, and the data set is divided into a training set and a test set. According to the faster CNN algorithm, a fault identification model is constructed, and the model can identify various faults of the power transmission and distribution lines; Preferably, the step S3 is specifically as follows: Considering that the UAV will capture fault types that do not appear in the data set during continuous inspection, there is a problem of insufficient adaptability in the identification of new faults by the above fault identification model. The orthogonal weight correction algorithm is used to update the fault identification model to maintain its continuous learning ability; Preferably, the step S4 is specifically as follows: Transmitting the identification results during the UAV inspection to the control center in real time for the dispatching personnel to take corresponding countermeasures and formulate subsequent inspection plans.

[0009] Preferably, the step S1 includes the following steps: S11, According to the position information, quantity information, and endurance of the UAV starting point, converting them into the constraint conditions of the path planning model; S12, To ensure the efficient inspection efficiency of the UAV, taking the inspection time and the total inspection distance as the objective function in the path planning model; S13, Based on MOPFA, dividing each path planning scheme into pathfinders and followers, randomly updating the position information of the pathfinders with different weights, and leading the followers to continuously optimize until the optimal path is obtained.

[0010] Preferably, the step S2 includes the following steps: S21, Preprocessing the historical image data set, including maintaining the balance of the samples by manual intervention and dividing the training set and the test set according to a fixed ratio; S22, Inputting the training set images into the deep convolutional network to obtain convolutional feature maps, further using the RPN structure to generate candidate regions, and projecting them into the above feature maps to construct a feature matrix; S23. Fix the size of the feature matrix using the ROI pooling layer and input it into the fully connected layer; S24. Calculate the loss function, specifically including classification loss and regression loss. Adopt the joint training method to perform backpropagation on the total loss and train the Faster RCNN model.

[0011] Preferably, in the step S24, the classification loss and the boundary regression loss are as follows: ; ; In the formula, i is the index of the anchor, p i represents the predicted probability that the anchor is the target. If the anchor is a positive sample, is 1, and if it is a negative sample, it is 0. t i represents the boundary region regression parameter vector for predicting the i th anchor, and represents the regression parameter vector of the real region.

[0012] Preferably, in the step S24, a constraint is constructed for the Faster RCNN fully connected layer based on the orthogonal weight correction algorithm to keep the weight changes in the same feature space, including the initialization stage and the continuous learning stage.

[0013] Preferably, the step S4 is specifically as follows: The unmanned aerial vehicle conducts intelligent inspection according to the pre-planned path, collects images in real time, identifies whether a fault has occurred and the type of the fault, and synchronously transmits the results to the control center. The dispatcher can formulate corresponding countermeasures and subsequent inspection plans according to the results fed back by the unmanned aerial vehicle.

[0014] The present invention provides a long-endurance unmanned aerial vehicle intelligent inspection method applied to transmission and distribution lines, having the following beneficial effects: 1. Plan the inspection path of the long-endurance unmanned aerial vehicle. Taking the inspection time and the total inspection distance as the optimization objectives, considering the flight conditions, endurance ability and other constraint conditions, construct a practical path planning model, and solve the optimal path based on MOPFA.

[0015] 2. Preprocess the historical image data, establish a training set and a test set, continuously learn the training set using Faster RCNN to obtain a fault recognition model, and verify the accuracy of the model in the test set. During the inspection process, the unmanned aerial vehicle can input the continuously collected image data into the fault recognition model and return the recognition results to the control center in real time for the dispatcher to formulate subsequent inspection plans.

[0016] 3. Based on the trained fault recognition model, new images collected during the UAV inspection process are recognized. By maintaining the continuous learning of the model based on the orthogonal weight correction algorithm and continuously updating the model, the accuracy of the fault recognition model can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The present invention will be further described below in conjunction with the drawings and embodiments: Figure 1 is the flowchart of the intelligent inspection of the long-endurance UAV of the present invention; Figure 2 is the flowchart of the path planning for the intelligent inspection of the long-endurance UAV of the present invention; Figure 3 is the flowchart of the fault recognition for the intelligent inspection of the long-endurance UAV of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0018] As Figure 1 shown, the method for intelligent inspection of long-endurance UAVs applied to power transmission and distribution lines disclosed by the present invention includes the following steps: S1. Considering relevant parameter information such as the position information, quantity information, and endurance of the UAVs, a path planning model is constructed with the UAV inspection time and the total inspection distance as the objective function. The MOPFA algorithm is used to solve the above path planning model to obtain the path planning scheme for each UAV; S2. According to the images of the power transmission and distribution lines taken during the UAV inspection, an image data set is constructed. To ensure the accuracy of the model, manual intervention can be used to balance the sample data set, and the data set is divided into a training set and a test set. A fault recognition model is constructed according to the faster CNN algorithm, and the model can identify various faults of the power transmission and distribution lines; S3. Considering that the UAVs will capture fault types that do not appear in the data set during continuous inspection, the above fault recognition model has problems with insufficient adaptability in the recognition of new faults. The orthogonal weight correction algorithm can maintain the original knowledge while improving the plasticity of subsequent learning, continuously learn new mapping relationships, update the fault recognition model, and maintain its continuous learning ability.

[0019] The specific steps of Step 1 are as follows: According to parameter information such as the position information, quantity information, and endurance of the UAV starting point, it is transformed into the constraint conditions of the path planning model. To ensure the efficient inspection efficiency of the UAVs, the inspection time and the total inspection distance are used as the objective function in the path planning model to construct the intelligent inspection path planning model for the UAVs; Figure 2 To use MOPFA to solve the UAV inspection path planning process, the specific steps are as follows: S11. Randomly initialize the population so that it is evenly distributed in the search area: ; Among them, x is the position vector of the individual population, U bd and L bd are the upper and lower boundaries of the search area respectively.

[0020] S12. Calculate the fitness of each individual, assign the best one as the pathfinder, and store it in the memory: ; ; Among them, P i is the probability of selecting and deleting non-dominant solutions in the memory by the roulette wheel method, max and min represent the maximum and minimum values of each objective respectively, box _ size represents the capacity of the memory, d in is the number of adjacent solutions less than the distance d .

[0021] S13. Update the leader position to lead the movement of the followers: ; In the formula, r 1 and u 1 are random vectors uniformly generated in the range of [0, 1], k and k max are the current iteration number and the maximum iteration number respectively, x p is the pathfinder position vector.

[0022] S14. According to the scale of the follower individuals, construct the corresponding pathfinding matrix, and the update process of the followers is as follows: ; In the formula, u 2 is a random vector uniformly generated in the range of [0, 1], x i and x j are the position vectors of the i -th and j -th followers respectively, r 2 and r 3 are random variables uniformly generated in the range of [0, 1], ais the interaction coefficient, indicating the amplitude of the movement of any member together with its adjacent members. b is the attraction coefficient, which sets a random distance to keep the follower at a roughly constant distance from the leader.

[0023] S15. Determine whether the maximum number of iterations has been reached k max .

[0024] Figure 3 is the flowchart for intelligent inspection fault identification of long-endurance UAVs. The specific steps of step S2 are as follows: First, input the image data into the conv layers, and the conv layers use VGG to obtain the corresponding feature maps. Second, the base RPN generates candidate boxes for each image data. Considering that the RPN will generate a large number of candidate boxes, in order to reduce the redundancy of the overlapping parts, the non-maximum suppression algorithm is used for filtering. Specifically, a small network slides continuously in the convolutional feature map, traversing all regions of the feature map to generate candidate boxes.

[0025] Project onto the corresponding feature map to construct a feature matrix.

[0026] Based on the ROI pooling layer, scale each feature matrix to a specified size and perform flattening processing. The two parallel outputs are the classification layer and the bounding box regression network respectively. The classification layer outputs the probabilities of each category, and the bounding box regression network outputs the bounding box regression parameters, and the model is continuously refined through backpropagation.

[0027] Based on the defined loss function, including classification loss and bounding box regression loss: ; ; In the formula, i is the index of the anchor, p i represents the predicted probability that the anchor is the target. If the anchor is a positive sample, then is 1, and if it is a negative sample, it is 0. t i represents predicting the i th bounding box regression parameter vector of the anchor, represents the regression parameter vector of the true region.

[0028] Adopt the joint training method to perform backpropagation on the total loss and train the Faster RCNN model.

[0029] Finally, input the image data to identify faults in real time.

[0030] The fault recognition model trained based on the existing image data set has a poor recognition effect in the new image data. It takes a lot of computer resources to build a new data set and train the model again, so continuous learning is used to circumvent this problem. Specifically, based on the orthogonal weight correction algorithm, constraints are constructed on the Faster RCNN fully connected layer to keep the weight changes in the same feature space, including the initialization stage and the continuous learning stage.

[0031] S3. Considering that the drone will capture fault types that do not appear in the data set during the continuous inspection process, the above fault recognition model has the problem of insufficient adaptability in the recognition of new faults. The orthogonal weight correction algorithm is used to update the fault recognition model to maintain its continuous learning ability. S4. The identification results of the drone inspection process are transmitted to the control center in real time, so that the dispatchers can make corresponding countermeasures and formulate subsequent inspection plans. The drone performs intelligent inspections according to the pre-planned path, collects images in real time, and identifies whether a fault has occurred and the type of fault, and transmits the results to the control center synchronously. The dispatchers can formulate corresponding response plans and subsequent inspection plans based on the results fed back by the drone.

[0032] In summary, the long-endurance UAV intelligent inspection method for power transmission and distribution lines provided by the present invention plans the path based on the existing UAV intelligent inspection method to improve the inspection efficiency; at the same time, the orthogonal weight algorithm is used to construct an orthogonal projector, which overcomes the catastrophic forgetting of the existing fault recognition model, enhances the recognition accuracy of the model, and has broad application prospects.

[0033] The above embodiments are only preferred technical solutions of the present invention and should not be regarded as limiting the present invention. The protection scope of the present invention shall be the technical solutions recorded in the claims, including equivalent replacement solutions of the technical features in the technical solutions recorded in the claims. That is, equivalent replacement improvements within this scope are also within the protection scope of the present invention.

Claims

1. A long-endurance drone intelligent inspection method for power transmission and distribution lines, characterized in that: The following steps are involved: S1, path planning; S2, fault identification; S3, model update; S4. Data transmission.

2. According to claim 1, a long-endurance drone intelligent inspection method for power transmission and distribution lines is characterized in that: The step S1 is specifically as follows: considering the location information, quantity information, and endurance of the drones, taking the drone inspection time and total inspection distance as the objective function, constructing a path planning model, and using the MOPFA algorithm to solve the above path planning model to obtain the path planning solution for each drone.

3. According to claim 1, a long-endurance drone intelligent inspection method for power transmission and distribution lines is characterized in that: The specific steps of step S2 are as follows: an image data set is constructed based on the images of power transmission and distribution lines taken during the drone inspection process. To ensure the accuracy of the model, manual intervention is performed to balance the image data set, and the data set is divided into a training set and a test set. A fault recognition model is constructed based on the faster CNN algorithm, and the model can identify faults of various types of power transmission and distribution lines.

4. According to claim 1, a long-endurance drone intelligent inspection method for power transmission and distribution lines is characterized in that: The specific details of step S3 are as follows: Considering that the drone will capture fault types that do not appear in the data set during the continuous inspection process, the above-mentioned fault recognition model has the problem of insufficient adaptability in the identification of new faults. An orthogonal weight correction algorithm is used to update the fault recognition model to maintain its continuous learning ability.

5. According to claim 1, a long-endurance drone intelligent inspection method for power transmission and distribution lines is characterized in that: The step S4 is specifically as follows: the recognition results during the drone inspection process are transmitted to the control center in real time, so that the dispatching personnel can make corresponding response measures and formulate subsequent inspection plans.

6. According to claim 2, a long-endurance drone intelligent inspection method for power transmission and distribution lines is characterized in that: The step S1 comprises the following steps: S11, converting the location information, quantity information, and endurance of the starting point of the UAVs into constraint conditions of the path planning model; S12. In order to ensure the high inspection efficiency of the UAV, the inspection time and the total inspection distance are used as the objective function in the path planning model; S13. Based on MOPFA, each path planning scheme is divided into pathfinders and followers. The position information of the pathfinder is randomly updated according to different weights, and the leader follower continuously searches for the best until the optimal path is obtained.

7. According to claim 3, a long-endurance drone intelligent inspection method for power transmission and distribution lines is characterized in that: The step S2 comprises the following steps: S21. Preprocess the historical image dataset, including maintaining the balance of samples by manual intervention, and dividing the training set and the test set into a fixed ratio; S22, input the training set image into the deep convolutional network to obtain the convolution feature map, further use the RPN structure to generate the candidate region, and project it into the above feature map to construct the feature matrix; S23, use the ROI pooling layer to fix the size of the feature matrix and input it to the fully connected layer; S24. Calculate the loss function, including classification loss and regression loss, use the joint training method, backpropagate the total loss, and train the Faster RCNN model.

8. According to claim 7, a long-endurance drone intelligent inspection method for power transmission and distribution lines is characterized in that: In step S24, the classification loss and the boundary regression loss are: ； ； In the formula, i is the index of the anchor, p i Represents the predicted probability that the anchor is the target, and the anchor is a positive sample. is 1, and is 0 for negative samples. t i Indicates the prediction i The regression parameter vector of the boundary region of anchors, A vector of regression parameters representing the true region.

9. The method for intelligent inspection of power transmission and distribution lines by using a long-endurance unmanned aerial vehicle according to claim 7 is characterized in that: In step S24, constraints are constructed on the Faster RCNN fully connected layer based on the orthogonal weight correction algorithm to keep the weight changes in the same feature space, including an initialization phase and a continuous learning phase.

10. According to claim 5, a long-endurance drone intelligent inspection method for power transmission and distribution lines is characterized in that: The specific details of step S4 are as follows: the drone performs intelligent inspections along a pre-planned path, collects images in real time, identifies whether a fault has occurred and the type of fault, and transmits the results to the control center synchronously. The dispatcher can formulate corresponding response plans and subsequent inspection plans based on the results fed back by the drone.

Images