Power supply line intelligent inspection method based on inspection unmanned aerial vehicle

Abstract

The invention discloses a power supply line intelligent inspection method based on an inspection unmanned aerial vehicle, and belongs to the technical field of power supply line inspection. Historical sample images and truth value labels corresponding to the historical sample images are taken as training data, loss function minimization is taken as a target, and a power supply line anomaly detection model is trained; and carrying the trained power supply line anomaly detection model on the inspection unmanned aerial vehicle, so that the inspection unmanned aerial vehicle performs inspection according to the power supply line and determines an abnormal target position of the power supply line, and finally, an intelligent inspection map of the power supply line can be generated according to the abnormal target position of the power supply line and the inspected power supply line. According to the method, the inspection efficiency of the power supply line can be effectively improved, the inspection accuracy is also improved, missing inspection caused by manual inspection is avoided, and the intelligent inspection map of the power supply line is generated, so that the inspection process becomes traceable.

Description

Technical Field

[0001] The present invention belongs to the technical field of power supply line inspection, and specifically relates to an intelligent inspection method for power supply lines based on an inspection drone. Background Art

[0002] Power supply lines are an important part of the power system, responsible for transmitting the electricity generated by power plants to substations and users at all levels. They are mainly composed of transmission lines, distribution lines and related equipment, and are responsible for the transmission and distribution of high-voltage and low-voltage electricity. The lines are usually in the form of overhead or underground cables and need to traverse various complex terrains and environments. Their safe and stable operation is crucial to ensuring power supply and maintaining the normal operation of the social economy. Therefore, regular inspections and maintenance are key measures to ensure the reliability of power supply lines. In the existing technology, power supply line inspections are mainly carried out by manual inspections to eliminate abnormalities or abnormal conditions in the power supply lines. However, the inspection of power supply lines by manual inspections is not only inefficient, but may also result in missed inspections. Summary of the Invention

[0003] The present invention provides an intelligent inspection method for power supply lines based on an inspection drone, which is used to solve the technical problem of low efficiency caused by manual inspection in the existing technology.

[0004] A smart inspection method for power supply lines based on inspection drones, comprising: Using a deep learning model to build a power supply line anomaly detection model and a loss function corresponding to the power supply line anomaly detection model; Obtaining historical sample images that are pre-stored or input by staff through human-computer interaction and the true value labels corresponding to the historical sample images; Using historical sample images and true value labels corresponding to the historical sample images as training data and minimizing the loss function as a goal, the power supply line anomaly detection model is trained to obtain the trained power supply line anomaly detection model; The trained power line anomaly detection model is loaded onto a patrol drone, allowing the drone to inspect the power lines and determine the target location of power line anomalies. According to the abnormal target location of the power supply line and the inspected power supply line, a smart inspection map of the power supply line is generated, and the smart inspection of the power supply line based on the inspection drone is completed.

[0005] Furthermore, a deep learning model is used to build a power supply line anomaly detection model, including: using a YOLOv7 model to build a power supply line anomaly detection model.

[0006] Furthermore, the loss function corresponding to constructing the power supply line anomaly detection model is:

[0007] in, represents the loss function, Represents the weight of positioning loss, S represents the number of grids in the feature map, and B represents the number of prediction boxes for each grid in the feature map. represents the positioning loss; represents the first coefficient, Represents the second coefficient; for the first i The first j When the IoU value between the predicted box and the corresponding actual box in the expected output image is greater than a given threshold, the predicted box with the largest IoU value is taken as a positive sample, and its corresponding first coefficient is 1, the second coefficient is 0, the first coefficient corresponding to other prediction boxes and the second coefficient are all 0; when the IoU values ​​between all predicted boxes and the corresponding actual boxes in the expected output image are less than a given threshold, the first coefficients corresponding to all predicted boxes are and the second coefficient All are 0; Indicates the prediction confidence corresponding to the prediction box, Indicates the actual confidence corresponding to the actual box, represents the first penalty coefficient, Represents the second penalty coefficient, classes represents the category set, Indicates the first i The first j The prediction box output belongs to the category c The probability of Indicates the probability that the image actually belongs to category c.

[0008] Furthermore, using historical sample images and true value labels corresponding to the historical sample images as training data and minimizing the loss function as the goal, the power supply line anomaly detection model is trained to obtain the trained power supply line anomaly detection model, including: Initialize the hyperparameters of the power supply line anomaly detection model and obtain multiple particles; Using historical sample images and true value labels corresponding to the historical sample images as training data, obtaining a loss function value corresponding to each particle according to the loss function, and selecting the optimal particle according to the loss function value; For any particle, based on the optimal particle, an adaptive learning method is used to perform comprehensive learning on the particle to obtain the particle after comprehensive learning; For any particle after comprehensive learning, the position fusion learning method is used to perform interval learning on the particle to obtain the particle after interval learning; For particles after learning in any interval, the position random flight method is used to perform global flight on the particles to obtain the particles after global flight; Repeat the adaptive learning method, position fusion learning method and position random flight method until the training end conditions are met and the optimal particle is re-determined; According to the re-determined optimal particles, a trained power supply line anomaly detection model is obtained.

[0009] Furthermore, based on the optimal particles, an adaptive learning method is used to perform comprehensive learning on the particles to obtain particles after comprehensive learning, including:

[0010]

[0011]

[0012]

[0013] in, Indicates the t The first training k particles, Indicates the k Particles after comprehensive learning, Indicates the t +1 training session k The training speed of particles, represents the first proportional weight coefficient, represents the second proportional weight coefficient, represents the self-learning factor, Indicates the k The optimal search frequency corresponding to the particle is represents the optimal particle, Indicates the t The first training k The training speed of particles, represents the first velocity update factor, represents the second speed update factor, Indicates the t The first training k The historical optimal state corresponding to the particle is Indicates the maximum value of the proportional weight coefficient, represents the minimum value of the proportional weight coefficient, represents a natural constant, and T represents the preset maximum number of training times.

[0014] Furthermore, the k The optimal search frequency corresponding to the particle Set to:

[0015] in, represents the initial optimization frequency corresponding to the kth particle, represents the frequency enhancement coefficient, exp represents the exponential function with the natural constant e as the base, t Indicates the current number of training times.

[0016] Furthermore, for any particle after comprehensive learning, the position fusion learning method is used to perform interval learning on the particle to obtain the particle after interval learning, including: Based on the current number of training times, the chaotic adaptive step size is obtained as:

[0017]

[0018] in, represents the chaotic adaptive step size, represents the first random number, represents the second random number between (0, 1), exp represents the exponential function with the natural constant e as the base, and T represents the preset maximum number of training times; For any particle after comprehensive learning, randomly match it with two other particles to obtain the first target particle and the second target particle; According to the chaotic adaptive step size, the first target particle and the second target particle, the particles are subjected to interval learning, and the particles obtained after interval learning are:

[0019] in, Indicates the t During the training m Particles after comprehensive learning, Represents the particle after interval learning , represents the third random number between (0,1), represents the first target particle, Represents the second target particle.

[0020] Furthermore, for particles after learning in any interval, a position random flight method is used to perform global flight on the particles to obtain particles after global flight, including:

[0021]

[0022] in, represents the particle after learning in the nth interval during the tth training process, represents the particle after the nth global flight, represents the Rhine flying particle, Indicates the upper limit of the coefficient and the coefficient lower limit The random coefficients between represents the fourth random number between (0,1), represents pi, represents the cosine function, represents random particles after interval learning, and particles With random particles For different particles, represents the worst particle, represents the first random flight coefficient, represents the second random flight coefficient, represents the fifth random number between (0,2], and and Obey the normal distribution, that is , ; represents the intermediate parameter, and , represents the gamma function, Represents the sine function.

[0023] Furthermore, the inspection drone inspects the power supply lines and determines the target location of abnormal power supply line, including: The inspection drone inspects the power supply lines. When the inspection drone detects an abnormality in the power supply line, it will locate itself as the target location of the abnormality in the power supply line.

[0024] Furthermore, according to the abnormal target location of the power supply line and the inspected power supply line, a smart inspection map of the power supply line is generated, including: An inspection route is generated on an electronic map according to the inspected power supply line, and an abnormal point is generated on the electronic map according to the abnormal target position of the power supply line to obtain a smart inspection map of the power supply line.

[0025] The present invention provides a smart inspection method for power supply lines based on inspection drones. By using historical sample images and true value labels corresponding to the historical sample images as training data and minimizing the loss function as the goal, a power supply line anomaly detection model is trained, and the trained power supply line anomaly detection model is carried on a direct inspection drone, so that the inspection drone inspects the power supply lines and determines the abnormal target positions of the power supply lines. Finally, a smart inspection map of the power supply lines can be generated according to the abnormal target positions of the power supply lines and the inspected power supply lines. This not only effectively improves the inspection efficiency of the power supply lines, but also improves the inspection accuracy, avoids missed inspections caused by manual inspections, and generates a smart inspection map of the power supply lines, making the inspection process traceable. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0027] Figure 1 A flowchart of a method for intelligent inspection of power supply lines based on an inspection drone is provided in an embodiment of the present invention.

[0028] The above drawings illustrate specific embodiments of the present invention, which will be described in more detail below. These drawings and the accompanying description are not intended to limit the scope of the present invention in any way, but rather to illustrate the concept of the present invention to those skilled in the art by reference to specific embodiments. DETAILED DESCRIPTION

[0029] Exemplary embodiments will be described in detail herein, examples of which are illustrated in the accompanying drawings. In the following description, when referring to the drawings, like numbers in different figures represent like or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments are not intended to represent all possible embodiments consistent with the present invention. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the present invention, as detailed in the appended claims.

[0030] The embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0031] like Figure 1 As shown, an embodiment of the present invention provides a method for intelligent inspection of power supply lines based on an inspection drone, comprising: S1. Using a deep learning model to build a power supply line anomaly detection model and a loss function corresponding to the power supply line anomaly detection model; Deep learning models can be classification models that identify and classify targets. By building a power line anomaly detection model using deep learning models, anomalies in the power line can be identified. A loss function was then constructed for the power line anomaly detection model. This loss function can be used to evaluate the training performance of the power line anomaly detection model during training, enabling the model to accurately detect anomalies.

[0032] S2. Obtain historical sample images that are pre-stored or input by a staff member through human-computer interaction and the true value labels corresponding to the historical sample images; Historical sample images are images of power supply lines, and the true value labels corresponding to the historical sample images are labels added to the historical sample images by staff. For embodiments of the present invention, multi-target recognition is required. Therefore, the true value labels corresponding to the historical sample images can be target boxes and categories set in the historical sample images, thereby achieving target detection and classification. Categories can generally include normal categories and abnormal categories. Abnormal categories are usually abnormal conditions in the power supply line, such as contaminated insulator surfaces, damaged insulators, foreign objects on power supply towers, and other abnormal conditions related to the power supply line.

[0033] S3. Using historical sample images and true value labels corresponding to the historical sample images as training data, and minimizing the loss function as a goal, the power supply line anomaly detection model is trained to obtain a trained power supply line anomaly detection model; For example, a particle swarm algorithm or a gradient descent optimization algorithm can be used to train the power supply line anomaly detection model so that the power supply line anomaly detection model can learn the data relationship between historical sample images and the true value labels corresponding to the historical sample images, so that classification can be achieved after the target is detected.

[0034] S4. The trained power supply line anomaly detection model is mounted on a direct inspection drone, so that the inspection drone inspects the power supply line and determines the target location of the power supply line anomaly; The power supply lines should exist in the electronic map so that the inspection drone can inspect the power supply lines according to the electronic map to determine abnormal conditions. Once the abnormal conditions are determined, the abnormal target location of the power supply line can be determined based on the positioning information of the inspection drone.

[0035] S5. Generate a smart inspection map of the power supply line based on the abnormal target location of the power supply line and the inspected power supply line, and complete the smart inspection of the power supply line based on the inspection drone.

[0036] Optionally, after the inspection drone determines the abnormal target location of the power supply line, it can also take photos for evidence collection and obtain abnormal photos. At the same time, points of interest are generated on the electronic map based on the abnormal target location of the power supply line and the abnormal photos, thereby generating abnormal points on the electronic map. After marking the inspected power supply line on the electronic map, a smart inspection map of the power supply line can be obtained.

[0037] The present invention provides a smart inspection method for power supply lines based on inspection drones. By using historical sample images and true value labels corresponding to the historical sample images as training data and minimizing the loss function as the goal, a power supply line anomaly detection model is trained, and the trained power supply line anomaly detection model is carried on a direct inspection drone, so that the inspection drone inspects the power supply lines and determines the abnormal target positions of the power supply lines. Finally, a smart inspection map of the power supply lines can be generated according to the abnormal target positions of the power supply lines and the inspected power supply lines. This not only effectively improves the inspection efficiency of the power supply lines, but also improves the inspection accuracy, avoids missed inspections caused by manual inspections, and generates a smart inspection map of the power supply lines, making the inspection process traceable.

[0038] In an embodiment of the present invention, a power line anomaly detection model is constructed using a deep learning model, including: constructing the power line anomaly detection model using the YOLOv7 model. However, it is worth noting that other deep learning models can also be used to construct the power line anomaly detection model, which can also achieve data classification and recognition.

[0039] In an embodiment of the present invention, the loss function corresponding to constructing the power supply line anomaly detection model is:

[0040] in, represents the loss function, Represents the weight of positioning loss, S represents the number of grids in the feature map, and B represents the number of prediction boxes for each grid in the feature map. represents the positioning loss; represents the first coefficient, Represents the second coefficient; for the first i The first j When the IoU value between the predicted box and the corresponding actual box in the expected output image is greater than a given threshold, the predicted box with the largest IoU value is taken as a positive sample, and its corresponding first coefficient is 1, the second coefficient is 0, the first coefficient corresponding to other prediction boxes and the second coefficient are all 0; when the IoU values ​​between all predicted boxes and the corresponding actual boxes in the expected output image are less than a given threshold, the first coefficients corresponding to all predicted boxes are and the second coefficient All are 0; Indicates the prediction confidence corresponding to the prediction box, Indicates the actual confidence corresponding to the actual box, represents the first penalty coefficient, Represents the second penalty coefficient, classes represents the category set, Indicates the first i The first j The prediction box output belongs to the category c The probability of Indicates the probability that the image actually belongs to category c.

[0041] The loss function provided by the embodiment of the present invention can effectively evaluate the power supply line anomaly detection model, and the smaller the loss function value, the better the training effect of the power supply line anomaly detection model.

[0042] In an embodiment of the present invention, a power supply line anomaly detection model is trained using historical sample images and true value labels corresponding to the historical sample images as training data, with the goal of minimizing a loss function, to obtain a trained power supply line anomaly detection model, including: S3.1. Initialize the hyperparameters of the power supply line anomaly detection model and obtain multiple particles; For power line anomaly detection models, the hyperparameters to be trained are generally the connection weights between network layers. These connection weights generally have corresponding upper and lower limits. Therefore, the hyperparameters can be randomly initialized between the upper and lower limits to achieve particle acquisition. Alternatively, a chaotic map initialization method can be used to initialize particles, thereby improving the uniformity of the distribution of initial particles within the solution space, ultimately improving the algorithm's global search capability and training speed.

[0043] S3.2. Using historical sample images and the true value labels corresponding to the historical sample images as training data, obtaining the loss function value corresponding to each particle according to the loss function, and selecting the optimal particle according to the loss function value; The historical sample image can be used as the actual input, and the true value label corresponding to the historical sample image can be used as the expected output to obtain the loss function value corresponding to each particle, and then the particle with the smallest loss function value can be determined as the optimal particle.

[0044] S3.3. For any particle, based on the optimal particle, use an adaptive learning method to perform comprehensive learning on the particle to obtain the particle after comprehensive learning; S3.4. For any particle after comprehensive learning, use the position fusion learning method to perform interval learning on the particle to obtain the particle after interval learning; S3.5. For particles after learning in any interval, use the position random flight method to perform global flight on the particles to obtain the particles after global flight; S3.6. Repeat the adaptive learning method, the position fusion learning method, and the position random flight method until the training end condition is met and the optimal particle is re-determined; The training end condition may be set as follows: when the current number of training times reaches a preset maximum number of training times, the training end condition is met.

[0045] S3.7. Based on the re-determined optimal particles, a trained power supply line anomaly detection model is obtained.

[0046] Optionally, after each search, out-of-bounds processing can be performed on the particle to ensure that the hyperparameters in the particle are always between the corresponding upper and lower limits.

[0047] While traditional particle swarm algorithms offer rapid training speeds, they suffer from poor training effectiveness and are particularly prone to falling into optimal values. This results in an inability to learn the data relationship between historical sample images and their corresponding true value labels, ultimately preventing them from effectively and accurately completing inspection tasks. Therefore, embodiments of the present invention provide an improved particle swarm algorithm to enhance its data relationship learning capabilities and ensure the accurate completion of inspection tasks.

[0048] In an embodiment of the present invention, based on the optimal particles, an adaptive learning method is used to perform comprehensive learning on the particles to obtain particles after comprehensive learning, including:

[0049]

[0050]

[0051]

[0052] in, Indicates the t The first training k particles, Indicates the k Particles after comprehensive learning, Indicates the t +1 training session k The training speed of particles, represents the first proportional weight coefficient, represents the second proportional weight coefficient, represents the self-learning factor, Indicates the k The optimal search frequency corresponding to the particle is represents the optimal particle, Indicates the t The first training k The training speed of particles, represents the first velocity update factor, represents the second speed update factor, Indicates the t The first training k The historical optimal state corresponding to the particle is Indicates the maximum value of the proportional weight coefficient, represents the minimum value of the proportional weight coefficient, represents a natural constant, and T represents the preset maximum number of training times.

[0053] In the embodiment of the present invention, the k The optimal search frequency corresponding to the particle Set to:

[0054] in, represents the initial optimization frequency corresponding to the kth particle, represents the frequency enhancement coefficient, exp represents the exponential function with the natural constant e as the base, t Indicates the current number of training times.

[0055] Compared with the existing particle swarm algorithm, the adaptive learning method provided by the embodiment of the present invention can not only realize the learning of optimal particles and historical optimal states, but also set an additional search item. The search is adjusted by the proportional weight coefficient, which can make a certain amplitude appear in the search process, ensuring the algorithm training speed while improving the ability to escape the local optimal value.

[0056] In an embodiment of the present invention, for any particle after comprehensive learning, interval learning is performed on the particle using a position fusion learning method to obtain the particle after interval learning, including: Based on the current number of training times, the chaotic adaptive step size is obtained as:

[0057]

[0058] in, represents the chaotic adaptive step size, represents the first random number, represents the second random number between (0, 1), exp represents the exponential function with the natural constant e as the base, and T represents the preset maximum number of training times; For any particle after comprehensive learning, randomly match it with two other particles to obtain the first target particle and the second target particle; According to the chaotic adaptive step size, the first target particle and the second target particle, the particles are subjected to interval learning, and the particles obtained after interval learning are:

[0059] in, Indicates the t During the training m Particles after comprehensive learning, Represents the particle after interval learning , represents the third random number between (0,1), represents the first target particle, Represents the second target particle.

[0060] The position fusion learning method provided by the embodiment of the present invention can effectively fuse the position information of other particles. While avoiding search collisions, it can realize the search of unfamiliar areas between multiple particles, improve the ability to find the global optimal solution, and the setting of the chaotic adaptive step size further improves the ability to escape from the local optimal value, which can also effectively ensure the convergence of the algorithm in the later stage of the algorithm.

[0061] In an embodiment of the present invention, for particles after learning in any interval, a position random flight method is used to perform global flight on the particles to obtain particles after global flight, including:

[0062]

[0063] in, represents the particle after learning in the nth interval during the tth training process, represents the particle after the nth global flight, represents the Rhine flying particle, Indicates the upper limit of the coefficient and the coefficient lower limit The random coefficients between represents the fourth random number between (0,1), represents pi, represents the cosine function, represents random particles after interval learning, and particles With random particles For different particles, represents the worst particle, represents the first random flight coefficient, represents the second random flight coefficient, represents the fifth random number between (0,2], and and Obey the normal distribution, that is , ; represents the intermediate parameter, and , represents the gamma function, Represents the sine function.

[0064] The position random flight method provided in the embodiment of the present invention can effectively improve the global search capability of the algorithm. At the same time, it integrates the cosine search strategy, so that the algorithm not only stays away from the worst position during the search process, but also can make large jumps, greatly improving the ability to jump out of local optimal values, and ultimately improving the ability to find the global optimal solution.

[0065] Optionally, a greedy strategy or annealing simulation algorithm can be used to control the global flight process to ensure the training speed of the algorithm.

[0066] To sum up, the improved particle swarm algorithm provided by the embodiment of the present invention can effectively solve the technical problems existing in the original particle swarm algorithm, improve the learning ability of the data relationship between historical sample images and the true value labels corresponding to the historical sample images, and ultimately improve the target recognition accuracy during the inspection process, ensure the smooth completion of the power supply line inspection, and effectively help staff improve the inspection efficiency of the power supply line.

[0067] In an embodiment of the present invention, the inspection drone performs inspections on power supply lines and determines the target location of abnormalities in the power supply lines, including: The inspection drone inspects the power supply lines. When the inspection drone detects an abnormality in the power supply line, it will locate itself as the target location of the abnormality in the power supply line.

[0068] In an embodiment of the present invention, generating a smart inspection map of the power supply line according to the abnormal target location of the power supply line and the inspected power supply line includes: An inspection route is generated on an electronic map according to the inspected power supply line, and an abnormal point is generated on the electronic map according to the abnormal target position of the power supply line to obtain a smart inspection map of the power supply line.

[0069] Those skilled in the art will readily appreciate other embodiments of the present invention after considering the specification and practicing the invention disclosed herein. The present invention is intended to cover any variations, uses, or adaptations of the present invention that follow the general principles of the present invention and include common knowledge or customary techniques in the art not disclosed herein. It should be understood that the present invention is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and variations can be made without departing from the scope thereof. The scope of the present invention is limited only by the appended claims.

Claims

1. A smart inspection method for power supply lines based on inspection drones, characterized in that: include: Using a deep learning model to build a power supply line anomaly detection model and a loss function corresponding to the power supply line anomaly detection model; Obtaining historical sample images that are pre-stored or input by staff through human-computer interaction and the true value labels corresponding to the historical sample images; Using historical sample images and true value labels corresponding to the historical sample images as training data and minimizing the loss function as a goal, the power supply line anomaly detection model is trained to obtain the trained power supply line anomaly detection model; The trained power line anomaly detection model is loaded onto a patrol drone, allowing the drone to inspect the power lines and determine the target location of power line anomalies. According to the abnormal target location of the power supply line and the inspected power supply line, a smart inspection map of the power supply line is generated, and the smart inspection of the power supply line based on the inspection drone is completed.

2. The intelligent inspection method for power supply lines based on inspection drones according to claim 1 is characterized in that: A deep learning model is used to build a power supply line anomaly detection model, including: using the YOLOv7 model to build a power supply line anomaly detection model.

3. The intelligent inspection method for power supply lines based on inspection drones according to claim 1 is characterized in that: The loss function corresponding to constructing the power supply line anomaly detection model is: in, represents the loss function, Represents the weight of positioning loss, S represents the number of grids in the feature map, and B represents the number of prediction boxes for each grid in the feature map. represents the positioning loss; represents the first coefficient, Represents the second coefficient; for the first i The first j When the IoU value between the predicted box and the corresponding actual box in the expected output image is greater than a given threshold, the predicted box with the largest IoU value is taken as a positive sample, and its corresponding first coefficient is 1, the second coefficient is 0, the first coefficient corresponding to other prediction boxes and the second coefficient are all 0; when the IoU values ​​between all predicted boxes and the corresponding actual boxes in the expected output image are less than a given threshold, the first coefficients corresponding to all predicted boxes are and the second coefficient All are 0; Indicates the prediction confidence corresponding to the prediction box, Indicates the actual confidence corresponding to the actual box, represents the first penalty coefficient, Represents the second penalty coefficient, classes represents the category set, Indicates the first i The first j The prediction box output belongs to the category c The probability of Indicates the probability that the image actually belongs to category c.

4. The intelligent inspection method for power supply lines based on inspection drones according to claim 1 is characterized in that: Using historical sample images and true value labels corresponding to the historical sample images as training data and minimizing the loss function as the goal, the power supply line anomaly detection model is trained to obtain the trained power supply line anomaly detection model, including: Initialize the hyperparameters of the power supply line anomaly detection model and obtain multiple particles; Using historical sample images and true value labels corresponding to the historical sample images as training data, obtaining a loss function value corresponding to each particle according to the loss function, and selecting the optimal particle according to the loss function value; For any particle, based on the optimal particle, an adaptive learning method is used to perform comprehensive learning on the particle to obtain the particle after comprehensive learning; For any particle after comprehensive learning, the position fusion learning method is used to perform interval learning on the particle to obtain the particle after interval learning; For particles after learning in any interval, the position random flight method is used to perform global flight on the particles to obtain the particles after global flight; Repeat the adaptive learning method, position fusion learning method and position random flight method until the training end condition is met and the optimal particle is re-determined; According to the re-determined optimal particles, a trained power supply line anomaly detection model is obtained.

5. The intelligent inspection method for power supply lines based on inspection drones according to claim 4 is characterized in that: Based on the optimal particles, an adaptive learning method is used to perform comprehensive learning on the particles to obtain particles after comprehensive learning, including: in, Indicates the t The first training k particles, Indicates the k Particles after comprehensive learning, Indicates the t +1 training session k The training speed of particles, represents the first proportional weight coefficient, represents the second proportional weight coefficient, represents the self-learning factor, Indicates the k The optimal search frequency corresponding to each particle is represents the optimal particle, Indicates the t The first training k The training speed of particles, represents the first velocity update factor, represents the second speed update factor, Indicates the t The first training k The historical optimal state corresponding to the particle is Indicates the maximum value of the proportional weight coefficient, represents the minimum value of the proportional weight coefficient, represents a natural constant, and T represents the preset maximum number of training times.

6. The intelligent inspection method for power supply lines based on inspection drones according to claim 5 is characterized in that: The said k The optimal search frequency corresponding to the particle Set to: in, represents the initial optimization frequency corresponding to the kth particle, represents the frequency enhancement coefficient, exp represents the exponential function with the natural constant e as the base, t Indicates the current number of training times.

7. The intelligent inspection method for power supply lines based on inspection drones according to claim 6 is characterized in that: For any particle after comprehensive learning, the position fusion learning method is used to perform interval learning on the particle to obtain the particle after interval learning, including: Based on the current number of training times, the chaotic adaptive step size is obtained as: in, represents the chaotic adaptive step size, represents the first random number, represents the second random number between (0, 1), exp represents the exponential function with the natural constant e as the base, and T represents the preset maximum number of training times; For any particle after comprehensive learning, randomly match it with two other particles to obtain the first target particle and the second target particle; According to the chaotic adaptive step size, the first target particle and the second target particle, the particles are subjected to interval learning, and the particles obtained after interval learning are: in, Indicates the t During the training m Particles after comprehensive learning, Represents the particle after interval learning , represents the third random number between (0,1), represents the first target particle, Represents the second target particle.

8. The intelligent inspection method for power supply lines based on inspection drones according to claim 7 is characterized in that: For particles after learning in any interval, the position random flight method is used to perform global flight on the particles to obtain the particles after global flight, including: in, represents the particle after learning in the nth interval during the tth training process, represents the particle after the nth global flight, represents the Rhine flying particle, Indicates the upper limit of the coefficient and the coefficient lower limit The random coefficients between represents the fourth random number between (0,1), represents pi, represents the cosine function, represents random particles after interval learning, and particles With random particles For different particles, represents the worst particle, represents the first random flight coefficient, represents the second random flight coefficient, represents the fifth random number between (0,2], and and Obey the normal distribution, that is , ; represents the intermediate parameter, and , represents the gamma function, Represents the sine function.

9. The intelligent inspection method for power supply lines based on inspection drones according to claim 1 is characterized in that: The inspection drone inspects the power supply lines and determines the target location of abnormal power supply lines, including: The inspection drone inspects the power supply lines. When the inspection drone detects an abnormality in the power supply line, it will locate itself as the target location of the abnormality in the power supply line.

10. The intelligent inspection method for power supply lines based on inspection drones according to claim 9 is characterized in that: Generate a smart inspection map of the power supply line based on the abnormal target location of the power supply line and the inspected power supply line, including: An inspection route is generated on an electronic map according to the inspected power supply line, and an abnormal point is generated on the electronic map according to the abnormal target position of the power supply line to obtain a smart inspection map of the power supply line.

Images